100% found this document useful (1 vote)
261 views641 pages

Sensors Fundamentals Volume 1

This document provides information about a multi-volume book series on sensors published between 1989-1991. Volume 1 focuses on fundamentals and general aspects of sensors. The series was edited by experts in sensor technology and aimed to provide a comprehensive survey of the state of sensor technology and help establish a system for classifying sensors. Each volume covers different types of sensors, including mechanical, thermal, magnetic, optical, and chemical sensors.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
100% found this document useful (1 vote)
261 views641 pages

Sensors Fundamentals Volume 1

This document provides information about a multi-volume book series on sensors published between 1989-1991. Volume 1 focuses on fundamentals and general aspects of sensors. The series was edited by experts in sensor technology and aimed to provide a comprehensive survey of the state of sensor technology and help establish a system for classifying sensors. Each volume covers different types of sensors, including mechanical, thermal, magnetic, optical, and chemical sensors.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 641

Sensors

Volume 1
Fundamentals and
General Aspects
Sensors
A Comprehensive Survey
Edited by
W. Gopel (Universitat Tubingen, FRG)
J. Hesse (Leybold AG, Koln, FRG)
J. N. Zemel (University of Pennsylvania,
Philadelphia, PA, USA)

Published in 1989:
Vol. 1Fundamentals and General Aspects
(Volume Editors: T. Grandke, W. H. KO)
Vol. 5 Magnetic Sensors
(Volume Editors: R. Boll, K. J. Overshott)

Remaining volumes of this closed-end series


will be published by 1991:

Vol. 2/3 Mechanical Sensors (scheduled for 1991)


Vol. 4 Thermal Sensors (scheduled for 1990)
Vol. 6 Optical Sensors (scheduled for 1991)
Vol. 7/8 Chemical and Biochemical Sensors (scheduled for 1990)

0 VCH Verlagsgesellschaft mbH, D-6940 Weinheim (Federal Republic of Germany), 1989

Distribution
VCH Verlagsgesellschaft mbH, P. 0.Box 101161, D-6940 Weinheim (Federal Republic of Germany)
Switzerland: VCH Verlags-AG, P.0.Box, CH-4020 Basel (Switzerland)
United Kingdom and Ireland: VCH Publishers (UK)Ltd., 8 Wellington Court, Wellington Street,
Cambridge CBI 1HW (England)
USA and Canada: VCH Publishers Inc., Suite 909,220 East 23rd Street, New York,
NY 10010-4606 (USA)

ISBN 3-527-26767-0(VCH, Weinheim) ISBN 0-89573-673-X (VCH, New York)


Sensors
A Comprehensive Survey
Edited by
W. Gopel, J. Hesse ,J. N. Zemel

Volume 1
Fundamentals and
General Aspects
Edited by
T. Grandke, W. H. KO
Series Editors:
Prof. Dr. W. Gopel Prof. Dr. J. Hesse Prof. Dr. J. N. Zemel
Institut fiir Physikalische und Leybold AG Center for Sensor Technology
Theoretische Chemie der Universitat Bonner StraBe 498 University of Pennsylvania
Auf der Morgenstelle 8 D-5000 Koln 51, FRG Philadelphia, PA 19104-6390, USA
D-7400 Tiibingen, FRG
Volume Editors:
Dr. T. Grandke Prof. Dr. W. H. KO
Siemens AG Electronics Design Center
Forschungslaboratorien Case Western Reserve University
Paul-Gossen-StraBe 100 Bingham Building
D-8520 Erlangen, FRG Cleveland, O H 44016, USA

This book was carefully produced. Nevertheless, authors, editors and publisher do not warrant the
information contained therein to be free of errors. Readers are advised to keep in mind that statements,
data, illustrations, procedural details or other items may inadvertently be inaccurate.

Published jointly by
VCH Verlagsgesellschaft mbH, Weinheim (Federal Republic of Germany)
VCH Publishers Inc., New York, NY (USA)
Editorial Directors: Dip1.-Phys. W. Greulich, Dipl.-Chem. Dr. M. Weller, N. Banerjea-Schultz
Production Manager: Dipl.-Wirt.-Ing. (FH) H.-J. Schmitt

Library of Congress Card No.: 89-24940

British Library Cataloguing-in-PublicationData:


Sensors: a comprehensive survey.
Vol. 1, Fundamentals and general aspects.
1. Sensors
I. Gopel, W. 11. Hesse, J. 111. Zemel, J. N. IV Grandke, T. V KO, W. H.
62010044
ISBN 3-527-26767-0

Deutsche Bibliothek Cataloguing-in-Publication Data:


Sensors : a comprehensive survey / ed. by W. Gopel . . . -
Weinheim ;Basel (Switzerland); Cambridge : New York, NY :
VCH.
NE: Gopel, Wolfgang [Hrsg.]
Vol. 1. Fundamentals and general aspects / ed. by T. Grandke ;
W. H. KO. - 1989
ISBN 3-527-26767-0 (Weinheim . ..)
ISBN 0-89573-673-X (New York)
NE: Grandke, Thomas [Hrsg.]

OVCH Verlagsgesellschaft mbH, D-6940 Weinheim (Federal Republic of Germany), 1989


Printed on acid-free paper
A11 rights reserved (including those of translation intoother languages). No part of this book may be reproduced
in any form - by photoprinting, microfilm, or any other means -nor transmitted or translated into a machine
language without written permission from the publishers. Registered names, trademarks, etc. used in this book,
even when not specifically marked as such are not to be considered unprotected by law.
+
Composition: Filmsatz Unger Sommer GmbH, D-6940 Weinheim.
Printing: Zechnersche Buchdruckerei, D-6720 Speyer.
Bookbinding: GroRbuchbinderei J. Schaffer, D-6718 Griinstadt.
Printed in the Federal Republic of Germany
Preface to the Series
The economic realities of productivity, quality, and reliability for the industrial societies of
the 21st century are placing major demands on existing manufacturing technologies. To meet
both present and anticipated requirements, new and improved methods are needed. It is now
recognized that these methods must be based on the powerful techniques employing com-
puter-assisted information systems and production methods. To be effective, the measure-
ment, electronics and control components, and sub-systems, in particular sensors and sensor
systems, have to be developed in parallel as part of computer-controlled manufacturing
systems. Full computer compatibility of all components and systems must be aimed for. This
strategy will, however, not be easy to implement, as seen from previous experience. One major
aspect of meeting future requirements will be to systematize sensor research and development.
Intensive efforts to develop sensors with computer-compatible output signals began in the
mid 1970’s; relatively late compared to computer and electronic measurement peripherals. The
rapidity of the development in recent years has been quite remarkable but its dynamism is af-
fected by the many positive and negative aspects of any rapidly emerging technology. The
positive aspect is that the field is advancing as a result of the infusion of inventive and finan-
cial capital. The downside is that these investments are distributed over the broad field of
measurement technology consisting of many individual topics, a wide range of devices, and
a short period of development. As a consequence, it is not surprising that sensor science and
technology still lacks systematics. For these reasons, it is not only the user who has difficulties
in classifying the flood of emerging technological developments and solutions, but also the
research and development scientists and engineers.
The aim of “Sensors” is to give a survey of the latest state of technology and to prepare
the ground for a future systematics of sensor research and technology. For these reasons the
publishers and the editors have decided that the division of the handbook into 8 volumes
should be based on physical and technical principles.
Volume 1 (editors: T. Grandke/Siemens AG (FRG) and W. H. Ko/Case Western Reserve
University (USA)) deals with general aspects and fundamentals: physical principles, basic
technologies, and general applications.
In Volume 2 and 3 mechanical sensors are introduced. With respect to the measured
parameters Volume 2 (editors: K. W. Bonfig/Siegen University (FRG), and D. J. Whitehouse/
Warwick University (UK)) concentrates on geometrical quantities such as length, speed, flow
etc.; Volume 3 (editors: T. KemCny/Miki Budapest (Hungary), and N. de Rooij/Neuchatel
University (Switzerland)) considers dynamometric quantities such as force, pressure, vibration
etc.
Volume 4 (editors : J. Scholz/Degussa AG (FRG) and T. Ricolfi/Consiglio Nazionale Delle
Richerche (Italy)) refers to thermal sensors.
Volume 5 (editors: R. Boll/Vacuumschmelze GmbH (FRG) and K. J. Overshott/Brighton
Polytechnic (UK)) deals with magnetic sensors.
Volume 6 (editors: E. WagnedFraunhofer-Gesellschaft e. V. (FRG) and R. Dandlikerl
Neuchatel University (Switzerland)) treats optical sensors.
Volume 7 and 8 (editors: W. Giipel/Tubingen University (FRG), L. Lundstrdm/Linkoping
University (Sweden), T. A. JonedHealth and Safety Executive (UK) t, M. Kleitz/LIES-
ENSEEG (France) and T. Seiyama/Kyushu University (Japan)) concentrate on chemical and
bio-chemical sensors.
VI Preface to the Series

Each volume is, in general, divided into the following three parts: specific physical and
technological fundamentals and relevant measuring parameters ; types of sensors and their
technologies; most important applications and discussion of emerging trends.
The series editors wish to thank their colleagues who have contributed to this important
enterprise whether in editing or writing articles. Thank is also due to Dip1.-Phys. W. Greulich,
Dr. M. Weller, and Mrs. N. Banerjea-Schultz of VCH for their support in bringing this series
into existence.

W. Gopel, Tiibingen J. Hesse, Koln J. N. Zemel, Philadelphia, PA


August 1989
Preface for Volume 1 of “Sensors”
As microprocessors and VLSI’s (very large scale integrated circuits) become generally
available, the field of measurement and instrumentation is undergoing a drastic change. Sen-
sors are a key element in this rapidly moving evolution, so the demand for sensors has soared
in the last decade. Solid state sensors that combine integrated circuits and micromachining
technologies as well as new materials open an avenue that can lead to many families of sensors
to meet the new demands in performance, size and cost.
Sensor research and development has flourished during the last decade and a wealth of
knowledge has been accumulated. Since sensors are considered “devices to convert a physical
or chemical variable into a signal suitable for measurement” as defined by the IEC, the field
of sensors encompasses an extremely wide area of science and engineering. “Sensors” are
becoming a unique branch of the engineering sciences with activities spreading over research,
development, and methods of applications. The need to organize our knowledge relevant to
sensors and make it readily available is well recognized by scientists, engineers and industries
interested in research, design, development, and applications.
This handbook series was written by experts in the sensor field in an effort to meet the needs
of our community. Volume 1 deals with the fundamentals and common principles of sensors
and covers the wide areas of principles, technologies, signal processing, and applications.
Common principles of sensors are in the focus of the first four chapters. While some of
the common aspects relating to lifetime considerations, design, and modeling are treated with
some emphasis on microsensors, we nevertheless feel that the information presented there will
be of value to everybody interested in sensors. Another topic relevant to the majority of all
sensors is that of signal processing and related problems which are outlined in chapters 10 to
13.
Flexible and powerful technologies are indispensable if sensors are to be a commercial suc-
cess. An assortment of typical sensor technologies is therefore given in chapters 5 to 9, where
emphasis has been placed on modern technologies rather than on the more traditional
workshop tools which have been in use for decades. Finally, we present a variety of sensor
applications which range from the simplest and cheapest temperature sensor for home ap-
pliances to the highly sophisticated, expensive and most rugged devices which have been tried
and tested in aerospace applications.
In view of the extreme variety of the field of sensors it would have been presumptuous to
attempt an exhaustive treatment in this volume, so naturally some readers will miss some of
their favorite themes. We nevertheless hope that anybody who might be at variance with our
selection of topics will forgive us.
Finally, it is a pleasure to express our sincere thanks to all those who have made this work
possible through their endeavours and patience: the publisher, the series editors, the authors,
and - last but not least - our families who once more had to set aside their interests.

Thomas Grandke and Wen H. KO


Erlangen and Cleveland, August 1989
Contents

List of Contributors ...................... XI

Sensor Fundamentals
1 Introduction . . . . . . . . . . . . . . ............ 1
T Grandke, J H m e
2 Sensor Parameters . . . . . . . . . . . ............ 17
M. Tabib-Azar
3 Sensor Modeling . . . . . . . . . . . . ............ 45
H. Baltes, A. Nathan
4 Sensor Design and Packaging . . . . . . . . . . . . . . . . . . . 79
R. Smith, S. Collins

Basic Sensor Technologies


5 Semiconductor Integrated Circuit Technology and Micromachining . . 107
W H. KO,J ;I:Suminto
6 Thin and Thick Films ...................... 169
S. Chang
7 Integrated Magnetic Sensors ................... 195
H. Baltes, A. Nathan
8 Optical Fibres and Integrated Optics ............... 217
E. Moore
9 Ceramics and Oxides ...................... 247
J Arndt

Sensor Interfaces
10 Signal Processing . . . . . . .................. 279
H.R. '7flinkler
11 Multisensor Signal Processing .................. 3 13
R. Miiller
12 Smart Sensors . . . . . . . .................. 331
J. E. Brignell
13 Interface Systems . . . . . . .................. 355
1 Atkinson
X Contents

Sensor Applications
14 Automotive: On-board Sensors . . . . . . . . . . . . . . . . . . 383
I. Igarashi
15 Automotive: Traffic Surveillance and Control . . . . . . . . . . . . 407
I? Drebinger
16 Home Appliances . . . . . . . . . . . . . . . . . . . . . . . . 425
T Kobayashi
17 Automated Manufacturing . . . . . . . . . . . . . . . . . . . . 445
1 Rogos
18 Process Engineering . . . . . . . . . . . . . . . . . . . . . . . 475
E. Nicklaus, IK Noerpel, K. Hartmann
19 Energy Production . . . . . . . . . . . . . . . . . . . . . . . 493
R. Sachedeva
20 Environmental Monitoring .................... 529
R. Brown, E. T Zellers
21 Medical Diagnostics and Patient Monitoring . . . . . . . . . . . . 555
M. R. Neuman
22 Aerospace Instrumentation . . . . . . . . . . . . . . . . . . . . 579
I. Warshawsky

Index.. . . . . . . . . . . . . . . . . . . . . . . . . . . . 603

List of Symbols and Abbreviations . . . . . . . . . . . . . . . . . 63 1


List of Contributors
Prof. Dr. Jorg Arndt Dr. Shih-Chia Chang
Universitiit Tubingen General Motors Research Laboratory
Mineralogisches Petrografisches Department of Electrical and
Institut Electronics Engineering
Wilhelmstr. 56 30500 Mound Road
D-7400 Tubingen, FRG Warren, MI 48090, USA
T: (7071) 296802 T: (3 13) 986-0253
Tfx: (3 13) 986-01 36
John Atkinson
University of Southampton Scott D. Collins
Department of Electronics and Damiens Associates
Information Engineering Davies, CA 95616, USA
Southampton SO9 5NH, U.K.
T: (703) 592616
Tfx: (703) 592865 Ramon Perez DePaula
6304 Wynkoop Boulevard Bethesda,
MD 20817, USA
Prof. Dr. Henry Baltes T: (202) 453-21 53
ETH Honggerberg, HPT Tfx: (202) 755-9234/9235
Institute of Quantum Electronics,
Physical Electronics Laboratory
CH-8093 Zurich, Switzerland Peter Drebinger
T: (1)-3772090 Siemens AG, N SI SVT E 14
Tfx: (1)-372-0974 Hofmannstr. 51
Tx: 823474 ehpz ch D-8000 Munchen, FRG
T: (89) 72222189
Tfx: (89) 72226598
Prof. John Ernest Brignell
University of Southampton
Department of Electronics and Dr. Thomas Grandke
Information Engineering Siemens AG, Forschungszentrum
Southampton SO9 5NH, U.K. Paul-Gossen-Str. 100
T: (703)-559122 ext. 3580 D-8520 Erlangen, FRG
Tfx: (703)-592865 T: (91 31) 72 1077
Tx: 47 661 Tfx: (91 31) 733008
Tx: 62921-310
Dr. Richard Brown
University of Michigan Dr. Klaus Hartmann
Department of Electrical and Bayer AG
Computer Engineering IN-PLT PAT, Geb. H 1
Ann Arbor, MI 48109, USA D-5090 Leverkusen, FRG
T: (3 13) 763-4207 T: (214) 3065902
Tfx: (3 13) 747-1781 Tfx: (214) 3065872
XI1 List of Contributors

Prof. Dr. Joachim Hesse Prof. Dr. Arokia Nathan


Leybold AG University of Waterloo
Bonner Str. 498 Dept. of Electrical Engineering
D-5000 KOln 51, FRG Waterloo, Ontario, Canada N2L 3G1
T: (221) 3471170 T: (519) 8851211 ext. 6014
Tfx: (221) 3471250 Tfx: (5 19) 8 886197
Ik: 8 88481-0 lk d
Dr. Isemi Igarashi Prof. Dr. Michael R. Neuman
Toyota Central Research and Case Western Reserve University
Development Labs Electronics Design Center
41-1, Aza Yokomichi, Oaza Nagakute Bingham Building
Nagakute-cho, Aichi-gun Cleveland, Ohio 44106
Aichi-Ken, 480-11, Japan T: (216) 368-3209
T: (5616) 261 11
Tfx: (5616) 26245
Ik: 781-449-6023 Dr. Eberhard Nicklaus
Bayer AG
Prof. Dr. Wen H. KO IN-Stab, Geb. H I
Case Western Reserve University D-5090Leverkusen, FRG
Electronics Design Center T: (214) 307844
Bingham Building Tfx: (214) 3065839
Cleveland, Ohio 44016, USA
T: (216) 368-2071
Tfx: (216) 368-3209 Dr. Wolfgang Noerpel
Tk: 980676 Bayer AG
IN-PLT PAT ELB, Geb. 105
Dr. Tetsuji Kobayashi D-5600Wuppertal 1, FRG
International Superconductivity T: (202) 36-7806
Technology Center (ISTEC) TfX: (202) 36-2299
34-3-Shimbashi 5-chome
Minato-ku
Tokyo 105, Japan Dr. JUrgen Rogos
T: (3) 431-4002 Innovationsgesellschaft INPRO
Tfx: (3) 431-4044 Niirnberger Str. 68/69
Dr. Emery Lightner Moore D-1000Berlin 30, FRG
900 Calle-Brusca T: (30) 219009-4042
Thousand Oaks, CA 91360, USA Tfx: (30) 2190090
T: (818) 715-3026
Tfx: (8 18) 7 12-7219
Rajiv Sachdeva
Prof. Dr. Rudolf Miiller Siemens India Ltd.,
Technische Universiat Miinchen 4A Ring Road, J. P. Estate
Lehrstuhl fUr Technische Elektronik New Dehli 110002, India
Arcisstr. 21, Geb. N2, Zi. N5317 T: (11) 3 31 8144/3 3171 52
D-8000 Miinchen 2, FRG Tfx: (11) 3 31 41 78
T: (89) 21052930 'Ik: 031-65254
List of Contributors XI11

Dr. Rosemary L. Smith Prof. Dr. Hans-Rolf Trankler


University of California Hochschule der Bundeswehr
Department of Electrical Engineering Miinchen
and Computer Science Institut fur Men- und
Bainer Hall Automatisierungstechnik
Davis, CA 95616, USA D-8014 Neubiberg, FRG
T : (9 16) 7 52-41 40 T: (89) 60043741
Tfx: (89) 6014693
Tx: 05 215800
Dr. James T. Suminto
Endevco/Alled Signal, Inc. Dr. Isidore Warshawsky
355 North Pastoria Avenue NASA Lewis Research Center
Sunnyvale, CA 94086 Mail Stop 301-2
T : (480) 739-3533 Cleveland, OH 44135, USA
T: (216) 433-3731
Dr. Massood Tabib-Azar Tfx: (216) 433-8000
Case Western Reserve University
Department of Electrical Engineering Edward T. Zellers
and Applied Physics University of Michigan
514 a Glennan Building Department of Electrical and Com-
Cleveland, Ohio 44106, USA puter Engineering
T: (216) 368-6431 Ann Arbor, MI 48109, USA
TfX: (216) 368-3209 T: (3 13) 763-4207
Tx: 980676 Tfx: (3 13) 747-1781
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

Introduction
THOMAS GRANDKE.Siemens AG. Erlangen. FRG. and
JOACHIM HESSE.Leybold AG. Koln. FRG

Contents
1.1 Historical Remarks ........................ 2
1.2 Definitions and Standards . . . . . . . . . . . . . . . . . . . . . 2
1.3 Physical and Chemical Transduction Principles . . . . . . . . . . . . 5
1.4 Classification of Sensors . . . . . . . . . . . . . . . . . . . . . 8
1.5 Sensor Markets . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2 1 Introduction

1.1 Historical Remarks

The industrial field has always been dependant on measuring instruments for the registra-
tion of physical parameters. For this reason, the technology of sensors and transducers has
a long tradition. Wilhelm von Siemens, for example, built one of the first sensors in 1860. He
made use of the temperature dependance of a resistor made of copper wire for temperature
measurements. Between 1920 and 1940, the intensive expansion of large-scale processes
created problems in measuring and control technology which proved to be very stimulating
for these fields. The development of semiconductor technology had its beginnings in the
1950s, since then the opportunities for. electronic signal processing and control techniques
have improved enormously. As a result, the design of measuring instruments and systems
changed to such an extent that the aim is now to realize as many functions and components
by electronic means as possible.
This development, which had its beginnings in the 1960s, met with a fundamental problem:
there was a lack of sensors capable of converting the generally non-electrical parameters into
electronic-compatible signals. These sensors and the appropriate electronics were not only ex-
pected to be precise, reliable, and capable of integration, but also had to be acceptable with
regard to their overall dimensions and price. The first intensive efforts to solve these problems
were started during the 1970s. Ever since, the course of development has been dynamic and
reveals both the advantages and disadvantages of a growing, prosperous market. Sensor
technology is now greatly promoted in the international market. The difficulty is that the ac-
tivities have to encompass many details in order to cover the vast range of applications in in-
dustrial measurement technology and research instrumentation.
.Today, about 100 relevant measuring variables need solutions, mostly concerning measuring
ranges and applications in various respects. Each solution often involves various viable
physical effects and technologies. The more than 10000 relevant publications on sensors and
lo00 companies which offer sensors in Western Europe alone easily demonstrate this
multiplicity. The measuring and control industry is strongly involved in this area and pioneer-
ing work has been done there. In the USA and Japan, considerable emphasis has been placed
on sensor technology which is closely tied to electronics (cf. Section 1.5). Today, users profit
from this strong competition.

1.2 Definitions and Standards

There is no doubt that we should begin a major work on sensors, such as this one, with
a definition of the scope, which thus implies that we should give a precise definition of what
a sensor is. This task is, however, by no means as simple as it might seem. Although sensor
is a common technical term that has been in frequent use for only about a decade, instruments
working just like sensors have been in use ever since man first attempted to gather reliable in-
formation concerning his physical, chemical, and biological environment. It is only during the
last few years that by virtue of the rapidly increasing capabilities of microprocessors, the idea
I.2 Definitions and Standards 3

of constructing a technical analog of the human intelligence together with the human senses
has arisen. Since then, a tendency has developed to designate all instruments suitable for the
technical measurement of physical, chemical, or biological quantities as sensors.
Hence one must be aware of the fact that a number of very different disciplines, each with
its own historical background, are lumped together. It is therefore not surprising that no one
unanimous concept of a sensor has yet been agreed upon in this heterogeneous community!
However, at least some meaningful definition is essential for this work and we suggest one
that follows the ANSI MC6.1 1975, “Electrical Transducer Nomenclature and Terminology”
[l]. This standard, which was prepared by the Instrument Society of America, defines a
transducer as “a device which provides a usable output in response to a specified measurand”.
Further, an output is defined as an “electrical quantity”, and a measurand is “a physical
quantity, property, or condition which is measured”.
For the purpose of our work, we have to generalize these definitions slightly. Firstly, we also
allow for chemical and biochemical quantities, properties, or conditions as being measurands.
Secondly, the range of usable outputs needs not be restricted to electrical quantities. Although
at present, all kinds of information processing have been performed almost exclusively by
using electrical signals, it is conceivable that optical information processing will gain impor-
tance in the future, so that optical signals could someday equally well represent a usable
output.
On the other hand, no limits with respect to the complexity of the device are defined. That
means, on the basis of the ANSI standard, one could even consider an MR system for medical
imaging, weighing several tons and costing several million dollars, as being a transducer! In
order to somehow limit the range of topics, we have to exclude such devices from the scope
of this series. Similarly, complex analytical systems such as gas chromatographs and spec-
trometers are not to be considered as sensors in the sense of the above definition. This would,
however, not mean that infegrated gas chromatographs or spectrometers should also not be
considered as sensors. Evidently, the criterion of complexity does not allow a sharp
discrimination, but we hope that it can at least serve as a rough guideline.
Interestingly, it is noted in the ANSI standard of 1975 that “transducer” is usually preferred
to “sensor”. If one evaluates the actual literature, it seems to be more the reverse situation.
Note that, eg, Middelhoek and Noorlag [2], also designated displays as (output) transducers
which convert electrical signals into optical signals, which is a more general definition than
that used here. Similarly, acoustic engineers designate electroacoustic devices which are
suitable for transmitting and receiving (u1tra)sonic waves as transducers. With respect to the
scope of this series, which is focused on “input transducers” or “measuring transducers”, this
generalization is not of concern, so we propose to treat the two nouns (transducer and sensor)
as being equivalent. There are a number of further synonyms, eg, detector, ...meter (where
. ..represents some measurand, as in flowmeter), pickup, gauge, transmitter, cell, end instru-
ment, and others where each one of these has a different historical origin.
The ANSI paper furnishes many more technical terms and definitions relevant to sensors,
some of which will be discussed in detail in Chapter 2. However, no rules are given concerning
the mechanical and electrical interfaces of sensors. In fact, such rules are virtually nonexistent,
eg, there are neither standardized dimensions for sensor housings nor standardized signals for
analog or digital electronic interfaces (except for temperature sensors [3]). One particular
reason for this is that the field of sensors is currently exploring its boundaries, and it is not
wise to impose too many standards before these boundaries are sufficiently well known. On
4 I Introduction

the other hand, a lack of standards hampers the dissemination of sensors, so this problem
should be dealt with as soon as possible (see also Section 1.6, Chapters 4 and 13).
Also, some kind of substructure of a sensor is missing, a point to which we shall return later
in this section. Before that we would like to review some definitions of sensors which can be
found in the literature. Although a thorough discussion of all available definitions would cer-
tainly be boring, we considered it worthwile at least to take a glance at the literature. Norton
[4] adopted the ANSI standard, which is not surprising as he chaired the corresponding com-
mittee. The IEC draft 65/84 [5] defines a sensor as being “The primary element of a measur-
ing chain which converts the input variable into a signal suitable for measurement”. While
this might seem essentially equivalent to the ANSI standard at first sight, it is further specified
that typical examples of sensors are the measuring junction of a thermocouple or the platinum
wire of a resistance thermometer (this elucidation being very similar to the german VDI/VDE
2600 [6]). Thus it seems to correspond to the sensor element [ 3 , 71 which is used in other
definitions (see below). Seippel [8] gave separate definitions for transducers, sensors, and
detectors, although he admitted that a distinction between these devices is “a very thin line”.
Presumably, it is no longer worth elaborating on these differences because they are not
reflected in practical use. Wolber and Wise [9] again essentially adopted the ANSI definition
as they defined a sensor as a “single-parameter measuring instrument which transduces a
physical parameter into a corresponding electrical signal with significant fidelity”. Mid-
delhoek and Noorlag [2] defined a sensor as an “input transducer of an information process-
ing system”. These few examples show that by and large the various definitions are similar,
although the actual phrasing and interpretation can be different. On the other hand, some at-
tempts have been made to define the concept of a sensor in terms of technologies or costs
(eg, “a reliable low-cost device .. .” [lo]). It may be that these considerations apply to certain
types of sensors. They are, however, definitely not applicable to the whole field of sensorology,
and so we prefer not to use these kinds of definitions.
In the German literature, some authors, eg, Hesse and Kuttner [3] and Scholz [lo], have at-
tempted to visualize a sensor as being composed of several elements. According to these ideas,
a sensor element or elementary sensor is, for example, a silicon pressure sensor chip or a strain
gauge. A sensor is a sensor element with housing and electrical connections included. A sensor
system is a sensor which incorporates some kind of signal processing (analog or digital). This
kind of nomenclature, which the German AMA (Arbeitsgemeinschaft fur Messwertauf-
nehmer) is trying to popularize at present, may offer some advantages in the context of market
studies and sales statistics [ll]. However, with respect to the aim of this work, we do not see
strong reasons for adopting it. The distinction between sensor element and sensor, again, is
not reflected in practical use and may even be difficult in many cases, eg, for fiber-optic sen-
sors. On the other hand, a semiconductor sensor with signal processing on the chip would
already be a sensor system, so the differences between sensors and sensor systems could well
vanish. The concept of smart or intelligent sensors, which is related to this trend, will be dealt
with in Chapter 12.
To summarize, we recommend that a sensor be understood in the sense of the ANSI stan-
dard with the proper generalizations and limitations that we have added above. Any more ex-
tensive definitions are not practicable because they are not accepted by the sensor community,
or because they are applicable to only a limited class of sensors.
1.3 Physical and Chemical Transduction Principles 5

1.3 Physical and Chemical Transduction Principles

Recalling the definition we adopted in the preceding section, a sensor is supposed to supply
a “usable output in response to a specified measurand”. In terms of today’s analog or digital
electronic world, a usable output can only be some sort of electrical signal which lends itself
to signal processing, the establishment of control loops, etc. On the other hand, the most in-
teresting measurands (by the number of sensors sold [12]) are temperature, geometrical quan-
tities and fluid mechanical quantities. Even if we allow for pneumatic (eg, for fluidic control
systems) or optical signals (eg, for optical data processing) as being usable outputs, it is evi-
dent that in general the specified measurand is inherently different from the desired output
signal. Thus, in order to convert the former quantity into the latter, at least one transduction
principle has to be invoked to make a sensor work.
Usually, these transduction principles are better known as physical or chemical effects. A
dictionary of such effects has been compiled by Ballentyne and Lovett [13], and more than
350 physical effects are to be found in a dictionary by Schubert [14]. Although not all of them
will be suitable for use in sensing applications, it is clear that some scheme is necessary to
classify this large number of effects. Lion [15] proposed the grouping of various principles ac-
cording to the form of energy in which the signals are received and generated, thus ending
up with a matrix-like arrangement. He distinguished six classes of signals, namely:

- mechanical;
- thermal;
- electrical;
- magnetic;
- radiant;
- chemical.

A number of examples belonging to each class are given in Table 1-1. Note that especially
the chemical signal domain contains an enormous variety of signals, ranging from the pH
value of a liquid to the activity of biological agents. Obviously there are some ambiguities in
this kind of grouping which will be discussed later.

Table 1-1. Signal domains with examples

Mechanical Length, area, volume, all time derivatives such as linear/angular velocity/acceleration,
mass flow, force, torque, pressure, acoustic wavelength and intensity
Thermal Temperature, (specific) heat, entropy, heat flow, state of matter
Electrical Voltage, current, charge, resistance, inductance, capacitance, dielectric constant, polariza-
tion, electric field, frequency, dipole moment
Magnetic Field intensity, flux density, magnetic moment, permeability
Radiant Intensity, phase, wavelength, polarization, reflectance, transmittance, refractive index
Chemical Composition, concentration, reaction rate, pH, oxidation/reduction potential
Table 1-2. Physical and chemical transduction principles m
Secondary
Signal
Mechanical Thermal Electrical Magnetic Radiant Chemical
Primary
Signal
~ ~~~

Mechanical (Fluid) Mechanical and Friction Effects Piezoelectricity Magnetomechanical Ef- Photoelastic Systems
Acoustic Effects: (eg, Friction Piezoresistivity fects: (Stress-induced
eg, Diaphragm, Calorimeter) Resistive, Capacitive, eg, Piezomagnetic Effect Birefringence)
Gravity Balance, Coolings Effects and Inductive Effects Interferometers
Echo Sounder (eg, Thermal Flow Sagnac Effect
Meters) Doppler Effect
Thermal Thermal Expansion Seebeck Effect Thermooptical Effects Reaction Activation
(Bimetalic Strip, Liquid- Thermoresistance (eg, in Liquid Crystals) eg, Thermal Dissociation
in-Glass and Gas Ther- Pyroelectricity Radiant Emission
mometers, Resonant Fre Thermal (Johnsen)
quency) Noise
Radiometer Effect
(Light Mill)
Electrical Electrokinetic and Elec- Joule (Resistive) Heating Charge Collectors Biot-Savart’s Law Electrooptical Effects: Electrolysis
tromechanical Effects: Peltier Effect Langmuir Probe eg, Kerr Effect Electromigration
eg, Piezoelectricity Pockels Effect
Electrometer Electroluminescence
Ampere’s Law
Magnetic Magnetomechanical Ef- Thermomagnetic Effects: Thermomagnetic Effects: Magnetooptical Effects:
fects: eg, Righi-Leduc Effect eg, Ettingshausen-Nernst eg, Faraday Effect
eg, Magnetostriction Galvanomagnetic Effects Effect Cotton-Mouton
Magnetometer eg, Ettingshausen Effect Galvanomagnetic Ef- Effect
fects: eg,
Hall Effect,
Magnetoresistance
Radiant Radiation Pressure Bolometer Photoelectric Effects: Photorefractive Effects Photosynthesis,
Thermopile eg, Photovoltaic Effect Optical Bistability -dissociation
Photoconductive
Effect
Chemical Hygrometer Calorimeter Potentiometry, Conduc- Nuclear Magnetic (Emission and Absorp-
Electrodeposition Cell Thermal Conductivity timetry, Amperometry Resonance tion)
Photoacoustic Effect Cell Flame Ionization Spectroscopy
Volta Effect Chemiluminescence
Gas Sensitive Field Ef-
fect
1.3 Physical and Chemical Tmnsduction Principles 7

Taking it for granted that each transduction principle can be distinguished by well-defined
input and output signals which can be attached to the classes defined above, a 6 x 6 matrix
will be required to list the various principles. In view of the vast number of principles available,
an exhaustive listing is beyond the scope of this work. Instead, we have compiled a number
of principles in Table 1-2 which either have important applications in instrumentation, or are
just instructive so as to visualize the basic systematics of this grouping. (Biological effects
which are just gaining importance with respect to (bio)chemical sensing have not been listed
for the sake of conciseness. In the framework of the present systematics, such effects would
show up in the lower right corner of Table 1-2). Also note that not all of the entries designate
proper physical or chemical effects or laws (as, eg, piezoelectricity, Joule heating, and
Ampere’s law do), but rather represent transducers which realize a certain effect which
does not have a typical designation (eg, electrometer, Langmuir probe). (It has been proposed
[16] to name as transducers only those instruments which are governed by principles from off-
diagonal matrix elements and to call the others modifiers since they d o not change the signal
domain. This kind of notation has, however, not found widespread use, so we prefer not to
adopt it.)
Although most of the entries in Table 1-2 will be self-explanatory, it might be helpful to
discuss at least some of them in more detail. First, it should be noted that some effects are
reciprocal, whereas others are not. Examples of reciprocal effects are the piezoelectric effect
(mechanical stress generates an electrical charge and vice versa), the Seebeck and Peltier ef-
fects (generation of a thermovoltage by temperature gradients and vice versa), and
chemoluminescence and photodissociation (generation of photons by chemical reactions and
vice versa). Most of the listed effects are not reciprocal, ie, they do not have a direct counter-
part which facilitates the reverse transduction by the same physical or chemical principle.
A diaphragm (mechanical/mechanical) converts pressure into deflection and stress. Either
of these output signals can be used to generate an electrical signal: capacitive or inductive ef-
fects are suitable for evaluating the deflection of the diaphragm, and piezoelectric or
piezoresistive effects can be used to convert the stress in the diaphragm into an electrical si-
gnal. This example demonstrates that even a very simple pressure transducer employs two
transduction principles in series, so it does not make sense to attempt a classification of sen-
sors by the physical or chemical effect.
In principle, each box of the matrix could be further subdivided into a submatrix with rows
and columns designated by the signal types within each major domain (see Table 1-1).
Although this could be of value for designing transducers, it would not add anything new to
the discussion of physical and chemical effects to which this section is devoted.
Not all of the effects listed in Table 1-2 have been exploited yet for sensing applications.
Optical bistability in solids is a fairly new effect which might have some potential for future
applications, but before that more basic studies are required [17]. Most of the thermomagnetic
and galvanomagnetic effects (except the Hall and magnetoresistance effects) are too small for
application in transducers. Another important criterion is the possibility of separating the
various effects in a material to avoid cross sensitivities, eg, most sensors are assumed to
operate in a certain temperature range and it is desirable that the effects in question are as
temperature insensitive as possible over the corresponding range, otherwise compensation
methods have to be developed. Finally, the measurand itself should not be affected noticeably
by the measurement, which is by no means trivial since signal conversion in transducers is
always based on energy conversion. Even if quantum mechanical effects and the uncertainty
8 1 Introduction

principle can be neglected, this aim cannot always be achieved simply by miniaturizing the
transducers, eg, flowmeters such as hot wire anemometers or pressure probes cannot be made
so small that they do not interact with the fine structure of the flow. In general, these and
other criteria prohibit the use of a number of physical and chemical effects for sensing ap-
plications.
Clearly, the kind of systematics outlined here have certain limitations. First, the definition
of signal domains contains some ambiguities. Radiant intensity in the infrared region might
also be considered as a thermal signal. On the other hand, radiant signals are essentially high-
frequency electric and magnetic fields. Finally, thermal energy is nothing more than
microscopic mechanical (Brownian) motion. Consequently, the classification of some effects
its not always possible without some ambiguity. In particular, effects with thermal or radiant
input/output signals could be found in either rows or columns of Table 1-2.
Another difficulty arises from the fact that a number of effects have more than one input
or output signal. Consider, for example, the thermomagnetic and galvanomagnetic effects:
they all have two input quantities, viz., a thermal or electrical one and a magnetic one. The
Hall effect has an electric current and a magnetic field as primary signals, so it could equally
well be found in the third or fourth row of Table 1-2. Since this effect is mainly used to
measure magnetic fields, we have preferred to display it in the fourth row. A similar argument
can be applied to the photoacoustic effect, also known as the optoacoustic effect. This effect
establishes the conversion of intensity-modulated optical radiation into audible sound in a gas.
As this conversion depends on the chemical composition of the medium, a chemical quantity
is also an input signal to this effect. Again, we have attached this effect to that input signal
which is primarily measured in practical applications, so it is found in the sixth row of the
matrix.
Finally, it should be noted that even conceptually simple sensors such as a strain-gauge
pressure sensor use more than one transduction principle (see above). This notion has two im-
portant consequences: firstly, sensors cannot be classified by the physical or chemical effect
that they employ unless one defines what the “main” effect in each sensor is - in our view
this would be a hopeless task. Secondly, we cannot discard those effects that do not yield an
electrical output signal - only the “final” effect in a transducer is generally subject to this
condition.
Despite all these complications, we consider the matrix of transduction principles to be an
illustrative means of structuring the vast variety of physical and chemical effects which could
be of value for sensing applications.

1.4 Classification of Sensors

The very general definition of a sensor outlined in Section 1.2 means that an enormous
variety of instruments will be included in this concept. Therefore, some classification of sen-
sors is as essential as it has been for the physical and chemical transduction principles as
discussed in the preceding section. Also, some classification scheme is required for the proper
organization of this book series, so we have to consider this subject in more depth in order
to develop an adequate approach to this problem.
1.4 Classjfication of Sensors 9

It appears that a number of different criteria are in use for classifying sensors, some of
which are listed below:

- physical or chemical effect/transduction principle;


- measurand (primary input variable);
- technology and material;
- application;
- cost;
- accuracy.

Preferences for any of these criteria are generally based on the type of user. For students,
the principle grouping may be adequate, for application engineers, the measurand or applica-
tion group may be desired, and for development engineers, the technology and material group-
ing may be preferred. Cost and accuracy, which are most frequently used in catalog and data
sheets, are important properties which will govern the future dissemination of sensors.
The most rigorous attempt to establish some order in the large number of sensors was
undertaken by Middelhoek and Noorlag [2]. Their approach was based mainly on a classifica-
tion of transduction principles. They proposed a standardized representation of each principle
and considered that a sensor will in general employ more than one such principle. In their
notation, a transduction principle is characterized by the input signal and the output signal,
essentially as has been done in Table 1-2.
In addition, they distinguished between self-generating and modulating principles. Self-
generating principles, also known as active [6], operate without an auxiliary energy source (eg,
the thermoelectric, piezoelectric, and photoelectric effects). Modulating principles, also
known as passive, require an auxiliary energy source, eg, the piezoresistive and
magnetoresistive effects require that an electric current is driven through the material and this
current will be modulated by the mechanical stress or the magnetic field. Consequently, self-
generating principles are sufficiently characterized by two items (input and output signals)
whereas modulating principles require at least three items which make a simple graphical
visualization (ie, twodimensional) of the variety of these principles in the form of a matrix
impossible (it should be noted that sometimes “active” and “passive” are used in an opposite
sense to that explained here [18]).
In order to illustrate this kind of systematics, we have sketched the representation of a
piezoresistive pressure sensor [19] and a magnetostrictive fiber-optic sensor [20] in Figure 1-1.
As explained in the preceding section, the pressure transducer starts with a transduction of
one mechanical quantity (pressure) into another (stress) and ends with the conversion of stress
into an electrical signal, eg, via the piezoresistive effect, the first step being self-generating
whereas the second is modulating. A fiber-optic magnetic-field sensor based on a
magnetostrictive jacket even employs three principles, as can be seen in Figure 1-1 b. Firstly,
the magnetic field is converted to a length change of the jacket and the enveloped fiber.
Secondly, the length change is converted into an optical signal via an interferometric set-up.
Finally, the optical signal is converted into an electrical signal by utilizing a photodiode.
Although this approach is very fundamental in nature, it will not help us to structure the
present book series as each sensor is characterized by a multitude of quantities, ie, the input
and output signals of all transduction principles which are realized in the sensor. We had
already anticipated in Section 1.3 that a classification by the transduction principle would not
10 1 Introduction

m h . mh. el%.

Figure 1-1.
Symbolic representation of
(a) a piezoresistive pressure trans-
ducer and (b) a magnetostrictive
fiber-optic magnetic field sensor.
Open squares represent self-generat-
ing transduction principles, squares
with a diagonal modulating prin-
ciples [2].

be viable. A more practical approach would be to use the measurand for which a sensor is
designed as the main criterion. In fact, the measurand will be identical with the primary input
signal, so we could also say that we classify by the input signal of the first transductionprinci-
ple. By and large, we have adopted this classification scheme for the organization of the pre-
sent series as it seems to be the most straightforward one available.
This means that the volume titles will essentially reflect the left-hand column in Table 1-1.
It would, however, not be wise to apply this scheme too pedantically, partly because some sen-
sors are not expected to be found under the heading of their primary input variable (for
historical reasons), and partly because the availability of appropriate authors and editors had
to be taken into account. The resulting volume titles are as follows:

Mechanical Sensors
Thermal Sensors
Magnetic Sensors
Optical Sensors
Chemical and Biochemical Sensors.

The exact organization of the individual volume contents exhibits a number of exceptions
from the general rule stated above, which deserve some explanations:
a. There will be no volume devoted to the measurement of electrical quantities. This is due,
in part, to the fact that electrical signals often are readily converted into a “usable output”
without invoking elaborate transduction principles. Sensors for recording electrical signals are,
however, covered to some extent in the other volumes.
b. All fiber-optic sensors will be covered in the volume on optical sensors, regardless of
their primary input signal.
c. Magnetic sensors suitable for sensing mechanical quantities (eg, displacement) will be
treated in the volume on magnetic sensors.
1.5 Sensor Markets 11

(In the latter two cases we have classified some by the intermediate rather than by the
primary signal, as it apprears that fiber-optic sensors and magnetic sensors are common con-
cepts and each should be treated in a single context.)
A classification by sensor materials and technologies is also of great relevance because the
availability of materials and technologies governs the availability of sensors. In fact, strong
efforts are being made to mature technologies which will in particular render the possibility
of producing inexpensive sensors. To take this situation into account, we have provided a
number of sections on major sensor technologies in this volume. Some of the topics treated
in these sections will be treated again from different points of view in the subsequent volumes,
yet we felt that a concise treatment of technologies is an essential part of an introductory
volume. However, since sensor technologies make up an extremely wide and heterogeneous
field (like sensors themselves), it is only possible to outline the “mainstream” technologies in
Chapters 5 to 9. It has, eg, been impossible to cover the rapidly emerging technologies of
chemical and biological materials and technology. It is felt that a description of biochemical
sensor materials and technologies are more advantageously placed with chemical and
biochemical sensors in Volumes 7 and 8.
Sensors could also be categorized according to their applications. Actually, this is an ex-
tremely important criterion because sensors for a specific measurand may look very different
in different fields of application, eg, a pressure sensor for use in an automobile will have to
be much cheaper than one for use in industrial process control, and a catheter-tip pressure
sensor for biomedical applications has to be extremely small. Therefore, we have also provided
a series of chapters on all major fields of applications in this volume, and in each chapter the
specific demands for sensors for a particular application will be outlined.

1.5 Sensor Markets

Sensor development has primarily followed the demands of the market. For 15 years, this
market has been stimulated mainly by electronic-aided automation of processes and systems
and the quality control of products. This trend in automation is now world-wide and in all
branches. It is not surprising that the application of sensors has mainly advanced in the
leading and innovative industries. According to one report [21], in 1988 the world market for
sensors (not including COMECON countries and military applications) amounted to $ 24.1
billion; $ 5.0 billion were allotted to process engineering, $2.8 billion to automotive engineer-
ing, $ 1.3 billion to safety, $ 1.2 billion to mechanical engineering, and $0.8 billion to aircraft
engineering. The regional shares in this market also reflect the strong position of the leading
industries there: $ 5.2 billion in the USA, $ 5.2 billion in Western Europe, and $ 3.0 billion
in Japan. In detail, of course, this picture is variable. The regional shares concerning
mechanical sensors are likely to be correct. However, whereas U.S. companies hold the lead
in the field of flow sensors, no such market strengths are evident with respect to temperature
sensors. The various and even application-specific demands, however, result in a fairly
heterogeneous market regarding measuring parameters. According to the cited report [21], in
1988 sensors for flow ($ 1.8 billion), pressure ($ 1.7 billion), speed ($ 1.4 billion), chemical
parameters ($ 1.2 billion), level ($ 1.15 billion), temperature ($ 1.0 billion), and position
($ 0.9 billion) proved to have the highest sales.
12 I Intmduction

The market analysis is even more differentiated with respect to the classification of applied
measuring principles and technologies (cf. Table 1-2). A survey also exists of particular com-
binations which may be of importance either today or in the future. Today's relationships may
change as soon as priorities for particular applications or technologies decrease, disappear,
or develop. Therefore, reliable forecasts are not only of great interest but are essential in the
industry. Adequate forecasts concerning the development of the worldwide sensor market
already existed in 1980, but the various definitions of sensors made reliable representations
and comparisons complicated. Recent analyses should be based on generally more reliable
data and thus should be more precise. A number of previously published estimations are
shown in Figure 1-2. Despite the wide spread of reliable data (which may be expected in a
dynamically developing market), an annual growth rate of 5-10070 for the next 10 years should
be regarded as a reasonable estimate. With respect to applications, automotive technology will
gain great importance. It has been predicted [21] that this demand for sensors will account
for a rise from $ 2.8 billion in 1988 to $ 11.4 billion by the year 2000. Applications of sensors
will place strong emphasis on driver guidance systems, engine and transmission control, safety
systems, and environmental protection. In this context, the application of sensors will concen-
trate more on safety systems (45% by 2000) and less on engine and transmission control
(49% in 1988). High growth rates are also expected in the field of environmental control
($ 0.08 billion in 1988; $ 0.5 billion by 2000). Medical technology will also experience a high
growth rate of ca. 9% per annum. Sensors will remain important in process engineering, in
safety applications, and in mechanical engineering, but with annual growth rates not much
higher than 5%. It is also expected that the regional market shares will change. According to

30
Market
Intelligence
Research

Figure 1-2.
Sensor world market
forecasts of the early 1980s.
1.5 Sensor Markets 13

the forecasts [21], the USA ($ 14.5 billion) will be in the leading position by 2000, ahead of
Western Europe ($ 13.5 billion) and Japan ($ 10.0 billion) (no estimates are available for other
parts of the world such as the COMECON countries, China, or India). This means that Japan
will have the relatively highest growth rate until then, especially with regard to applications
of sensors in automotive and mechanical engineering. Within this framework, the markets for
various measuring parameters are very likely to expand evenly. Previous and recent forecasts
(cf., Figure 1-3) hardly reveal any difference in this respect. During the period covered
(1988-95), absolute values are also similar despite the different definitions and exchange rates.
Pressure, flow, temperature, and level will remain the most important standard parameters for
sensors.

Flow Speed Binary Temperoture Flue gas


sensors sensors sensors sensors sensors
Pressure Chemicol Level Position Sensors for
sensors sensors sensors sensors opticol porameters

Figure 1-3. World market for the most important sensors in 1988 and 1995.

Naturally, such forecasts are bound to estimates regarding the economic growth in general
and in individual branches and regions. The figures for the world market for sensors should
be fairly precise. In detail, however, unpredictable market fluctuations can easily cause
changes. This might be the case, for example, whenever the potentials of modern
biotechnologies are exploited thoroughly. The corresponding demand for modern sensors
would, indeed, have a considerable effect [22]. Breakthroughs in the fields of modern
technologies could also cause structural changes. Owing to the general demand for further
miniaturization, micromechanics based on silicon and thin-film technology will provide im-
portant contributions [12]. Miniaturized optical sensors in fiber technology [23] or in in-
tegrated optical technologies correspond with this trend. It is therefore safe to say that those
technologies which allow the integration of sensor functions and signal processing, preferably
in a monolithic design, will gain in significance [24, 251. Silicon technology offers significant
advantages but problems occur because of the limited applicable temperature range. Accor-
14 I Introduction

ding to the general view, the share of these “miniaturized and intelligent” sensors in Western
Europe will have increased from 20% in 1980 to 45% in 1995. The world market for “smart”
sensors amounted to ca. $ 0.5 billion in 1988 and is expected to increase to ca. $ 1.3 billion
in 1995 [24]. This development will therefore stimulate the market, especially from a long-term
point of view [24]. These trends are illustrated in the form of a quantity vs. price diagram in
Figure 1-4.

Medicol 0
Estoblished
market

aborotories c _
\

\ theropy 1

f Trend

Industrial \ \\Building\
process
techndogy

t
cu It ure

Sensor production volume

Figure 1-4. Market segments and trends in sensor technology (logarithmic scale).

1.6 Trends

In addition to microprocessors, sensors are today the key components in measuring and
control technology. In the future both elements will form an integrated unit and be the basis
of systems. Following that line, sensor technology is already in the transition phase between
pure signal transformation and signal processing. Future tasks will consequently be more
system oriented rather than the supply of single sensors.
Nevertheless, the characteristics of single sensors still deserve improvements in various
respects. New principles, materials, and technologies produced a wealth of possible new sen-
sors in the 1980s which work successfully in the laboratory. This process is continuously going
1.7 References 15

on, eg, biological materials are said to play an important role with respect to future
(bio)chemical sensors. It is, however, much more challenging to develop concepts which prove
to be reliable and robust in the field and also render the possibility of economic production.
A well-known example is that of silicon sensors, which have been around since the mid 1970s,
but only during the last few years have made the breakthrough to real mass production as was
often prognosticated. On the whole, the further development of single sensors will therefore
be evolutionary rather than revolutionary, ie, with steady but not spectacular progress.
Another trend which will continue is that of miniaturization, at least in those areas which
will profit from smaller sensors (eg, medical, automotive, and aerospace applications). On the
other hand, this trend has also brought about some unforeseen problems which are mainly
related to the fact that different structural properties scale very differently with size (eg, sur-
face 1*, volume and mass 13, and moment of inertia Is, where 1 is the linear dimension of a
structure). Development engineers who had been used to thinking in terms of macroscopic
dimensions had to learn that structures and materials can behave very differently if they are
scaled down to micrometer size.
Concerning the systems aspect, sensors also have to acquire further characteristics,
especially with regard to

- sensor signal processing supported by integrated and often modular designed electronics
at the place of measurement,
- arithmetic processing of these signals supported by parametric hierarchically organized
software,
- a higher speed of signal processing,
- the provision of suitable and standardized sensor interfaces,
- the provision of standardized integration concepts for sensor systems in control systems.

Adaptable multi-sensor systems will offer such a solution. Those systems will consist of
several mostly simple sensors and a secondary processing unit which computes the signals in
parallel. It will then be possible to reduce the extensive registration of a complex measuring
and control parameter to the measurement of several individual parameters. Such multi-sensor
solutions create powerful and precise systems with high selectivity. Interfering variables will
be eliminated by computation and built-in self-supervision guarantees reliability. Advanced
methods of signal processing (eg, pattern recognition) have to be developed to achieve these
goals.
The realization of such systems requires an approach based on systems technology, which
needs a theoretical conception in the first place. Modeling and computer-aided design will play
a key role in this context.

1.7 References

[l] "Electrical Transducer Nomenclature and Terminology", A N S I Standard MC6.1-1975(ISA S3 7.1)


Research Triangle Park, NC: Instrument Society of America, 1975.
[2] Middelhoek, S., Noorlag, D. J. W., "Three-dimensional Representation of Input and Output
Transducers", Sens. Actuators 2 (1981/82) 29-41.
16 I Introduction

131 Hesse, D., Kuttner, H., "Entwicklungstendenzen in der Sensorik", Industrie-Elektrik und
Elektronik 28, Nr. 5 (1983) 36-44.
[4] Norton, H. N., Sensor and Analyzer Handbook, Englewood Cliffs, NJ: Prentice Hall, 1982.
[5] Terms and Definitions in Industrial-Process Measurement and Control, (IEC draft 65/84), Interna-
tional Electrotechnical Committee, 1982.
[6] Metrologie, VDIIVDE 2600, Dusseldorf, FRG: VDI, 1973.
[7] Scholz, J., "Sensoren und Normung", ETZ 107 (1986) 340-344.
[8] Seippel, R. G., Eansducers, Sensors and Detectors, Reston: Reston Publishing, 1983.
19) Wolber, W. G., Wise, K. D., "Sensor Development in the Microcomputer Age", IEEE Tramactions
ED-26 (1979) 1864-1874.
"Low Cost Sensors", in: Mackintosh State of the Art Series, Luton, U K Benn Electronics.
Editorial, Sensor Magazin, No. 1, (1988) 3.
Sensors-Miniaturizationand Integration, 1988, Battelle Europe, 6000 Frankfurt, FRG.
Ballentyne, D. W. G., Lovett, D. R., A Dictionary of Named wfects and Laws in Chemistry, Physics
and Mathematics,London: Chapman and Hall, 1980; see also: Scheibner, E. J., "Solid-State Physical
Phenomena and Effects", IRE Trans. Comp. Parts, Part I-IV, 8 (1961), 133-151; 9 (1962), 19-32,
61-74, 119-141.
Schubert, J., Physikalische Effekte, Weinheim, FRG: Physik-Verlag, 1984.
Lion, K. S., "Transducers: Problems and Prospects", IEEE Transactions IECI-16 (1969) 2-5.
Middelhoek, S., Noorlag, D. J. W., "Signal Conversion in Solid-state Transducers", Sens. Actuators
2 (1982) 211 -228.
Gibbs, H. M., Optical Bistability: Controlling Light with Light, New York: Academic Press 1985.
Sensorik, W. Heywang (ed.); Berlin: Springer, 1984.
Tufte, 0. N., et al., "Silicon Diffused-Element Piezoresistive Diaphragms", J. Appl. Phys. 33 (1962)
3 322-3 327.
[20] Pitt, G. D. et al., "Optical-Fibre Sensors", IEE Proceedings I32 (1985) 214-248.
1211 Sensor Technology 2000, Basel: Prognos AG, 1988.
1221 Biosensors 2000, Basel: Prognos AG, 1988.
[23] Fiber Optic Sensor Market III, Market Intelligence Research Company (MIRC), Mountain View,
CA, 1988.
[24] Intelligence Sensor Markets, Market Intelligence Research Company (MIRC), Mountain View, CA,
1988.
[25] Intelligent Sensors: The Merging of Electronics and Sensing, Technical Insights, Inc., Fort Lee, NJ,
1988.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

2 Sensor Parameters
M . Tmm.AzAR. Case Western Reserve University. Cleveland. OH. USA

Contents
2.1 Introduction ........................... 18
2.2 Transducer Characteristics . . . . . . . . . . . . . . . . . . . . . 18
2.2.1 Static Characteristics . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.2 Dynamic Characteristics . . . . . . . . . . . . . . . . . . . . . . 23
2.2.3 Environmental Parameters . . . . . . . . . . . . . . . . . . . . . 25
2.3 Methods for Characterization of Transducers . . . . . . . . . . . . . 27
2.3.1 Electrical Methods . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.2 Optical Methods . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.3.3 Mechanical Methods . . . . . . . . . . . . . . . . . . . . . . . . 31
2.3.4 Thermal Methods . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3.5 Chemical and Biological Methods . . . . . . . . . . . . . . . . . . 32
2.3.6 Calibration Techniques . . . . . . . . . . . . . . . . . . . . . . . 32
2.3.7 Documentation of Results . . . . . . . . . . . . . . . . . . . . . 33
2.4 Reliability and Long-Term Stability Issues . . . . . . . . . . . . . . . 34
2.4.1 Reliability Models and Testing . . . . . . . . . . . . . . . . . . . . 34
2.4.2 Reliability Models . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.4.3 Screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4.4 Accelerated Aging Test . . . . . . . . . . . . . . . . . . . . . . . 37
2.4.5 Failure Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4.5.1 Electrical Failure Mechanisms . . . . . . . . . . . . . . . . . . . . 40
2.4.5.2 Mechanical Failure Mechanisms . . . . . . . . . . . . . . . . . . . 41
2.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
18 2 Sensor Parameters

2.1 Introduction

Data from the outside world are usually acquired through sensors. Depending on how ac-
curately the reality is to be gauged, the sophistication of the sensors increases and the relation-
ship is not usually linear or simple. Thus the sensors are an important part of the acquisition
systems that, in providing an interface between the system and the outside world, have to com-
ply with the requirement of both environments. This puts stringent requirements on the
chemical, mechanical, and electrical characteristics of the sensors, especially in corrosive,
harsh, or delicate environments. These situations are encountered in industrial and biomedical
applications. Hence, a reliable and systematic characterization of the sensor is of utmost im-
portance in today’s rapidly advancing field of sensors.
The actuators that perform the inverse task of altering the outside world also have
parameters similar to those of sensors. The material in this chapter can also be applied to
them. The parameters of the actuators become increasingly important when the economy of
the situation demands high-efficiency generation of, for example, a displacement.
In Section 2.2, all the terminologies that are used in describing various parameters of sen-
sors will be discussed. These include all the parameters that are essential in high-fidelity mapp-
ing of reality, such as accuracy and frequency response, all the parameters that obscure a
desired mapping, such as noise, and all the parameters that are important in transferring the
data to the acquisition system, such as signal levels and impedances. Also discussed in Section
2.2 are environmental parameters such as temperature and vibration that might influence the
performance of a sensor in an undesirable manner.
In Section 2.3, different methods for the characterization of the above parameters with
specific examples taken from solid-state sensors will be discussed. Also described are the com-
monly accepted means of documentation of the experimental results, emphasizing statistical
analysis in reporting and drawing meaningful conclusions.
Reliability and long-term stability of transducers are becoming increasingly important as
they are widely employed in various applications and especially because of their important
usage in biomedical applications. Also related to the reliability issues is environmental and ac-
celerated failure testing, all of which are discussed in Section 2.4.
Most of the examples used to illustrate various concepts and usage of terminologies are
taken from solid-state sensors. Sections 2.3 and 2.4 are primarily focused on solid-state sen-
sors. However, in most cases one may use similar approaches to those described there to study
other kinds of sensors.

2.2 Transducer Characteristics

2.2.1 Static Characteristics [l- 141

I . Accuracy: A measure of how closely the result of the experiment (sensor output) approx-
imates the true value. Since the true value of the unknown (measurand) is not known a priori,
a comparative measurement is needed in specifying the accuracy of a transducer. Inaccuracy
2.2 lhnsducer Chamcterisricr 19

is usually described as follows:

Ea (To) = 100 (X, - XJ/X, (2-1)

where X,is the true value of the unkown X,and X, is its measured value In practice, the
inaccuracy is expressed as a percentage of full-scale output (FSO):

Ef (Vo) = 100 (X, - X J / X , . (2-2)

Clearly, I ef I Q IE , I. X,can be obtained either from other measurements or from the Na-
tional Bureau of Standards (NBS). Since for a repeated number of measurements at a f i d
value of X,different values of X, might be obtained, it is useful to define an error bar
denoting the maximum range of X,. Error bars can be generalized to introduce an error
band in the case of an experimental curve Error band, then, denotes the maximum and
minimum values of X, throughout the range of X,.Error bars and error bands can be ob-
tained statically or dynamically. In dynamic error measurements, the system is subjected to
various real-life disturbances such as shock, vibration, or acceleration. In static error
measurements, the error is measured under ideal conditions.
2. Precision: Describes how exactly and reproducibly an unknown value is measured. It has
nothing to do with how accurately the measured value represents the unknown parameter. For
example, when using a ruler to measure the length of a rod, accuracy refers to how closely
the measured value represents the true length of the rod. The precision refers to how carefully
the number is read from the ruler or how carefully the ruler is set next to the rod. Clearly,
accuracy without precision does not have any meaning and precision does not imply accuracy.
3. Resolution: The smallest increment in the value of the measurand that results in a detec-
table increment in the output. It is expressed as a percentage of the measurand range (To MR).
For example, if a temperature sensor yields an increment of AV output voltage in response
to a AT change in the temperature of an object, then the maximum resolution (R,) is the
smallest AT (denoted by AT-) that yields a detectable AV and it is expressed as To M R

A T-
R,, (VO) = 100. (2-3)
T m - Tlnill

The average resolution (R,,) is then given by the average of R (T) over the range of the
measurand (temperature):
n

where n is the number of AT, in the measurand range that are considered.
4. Sensitivity: Incremental ratio of the output Q) to the input (x);

S = Ay/h.

In the above equation, x denotes the desired measurand.


20 2 Sensor Parameters

5. Selectivity/Specijicity: In non-ideal sensors, the output might change owing to a change


in the environmental parameters or other variables. To be more general, we consider a system
composed of n sensors each having an output y p @ = 1, ...,n). Let us define a partial sen-
sitivity (Sap) as a measure of the sensitivity of the p t h sensor to these other interfering
variables (xu):

The selectivity matrix for this system is defined as a square matrix with S, as the apth
entry. An ideally selctive system is characterized by a diagonal matrix having only aa entries.
An ideally specific system is characterized by a matrix having only a single a a t h entry. A
quantitative measure of selectivity and sensitivity is given by the following relationships [14]:
Selectivity:
A = Min ( Sam
for a = 1, ..., n . (2-7)

Specificity:

na= I Saul -1.


n

c
p= I
ISapl - ISauI

In ideally selective sensors A = 00 and in ideally specific sensors l7, = 01. No selectivity
or specifity is designated by, respectively, A = 0 or nu= 0.
6. Minimum detectable signal (MDS) [6]:Assuming that the signal or the measurand does
not contain any noise, the minimum signal level that yields a readable transducer output is
determined by the noise performance of the transducer. To account for the noise level
generated by the transducer, all the internal noise sources of the transducer can be bunched
together to form a single noise source. This single noise source, which is called the equivalent
input noise source, when connected to the input of the ideal (noiseless) transducer yields the
output noise level of the transducer under study. The minimum signal level that yields a
reliable transducer output signal, the MDS, is usually taken as the rout-mean-square (RMS)
equivalent input noise (signal-to-noise ratio of 0 dB).
7. Threshold: Starting from a measurand of value zero, the smallest initial increment in the
measurand that results in a detectable output is the threshold. Threshold is usually due to
device nonlinearity and it is different from MDS.
8. Nonlinearity: A measure of deviation from linearity [ie, for a linear device if y 1 = f ( x l )
and y , = f (xJ, then y , + y , = f (xI + x J ] of the sensor, which is usually described in
terms of the percentage deviation in FSO at a given value of the measurand. There are two
methods to specify non-linearity: (i) deviation from best-fit straight line and (ii) deviation
from terminal-based straight line. The first method is most commonly used by manufacturers
and expresses the deviation of the transducer output from a best-fit straight line as shown in
Figure 2-1 b. The second method, on the other hand, expresses the deviation of the output
from a straight line that is drawn between the two terminal ends of the output versus input
-
0

4
100

LVJ
D(
Y

3
n
L
# 2.2 Transducer Characteristics

terminal-based
d
3
n
d
3
0
21

0
=I

0
0 00 0 too
measurand ( X range\ measurand (7. range)

(a 1 (b)
repeatabili tv

Figure 2-1.
Output versus input charac-
teristics of a transducer.
(a) Terminal-based line non-
linearity.
(b) Best-fit straight line non-
linearity. 0 100 0 100
measurand ( X range) measurand ( X range)
(c) Hysteresis.
(d) Repeatability. (C) (d)

curve as shown in Figure 2-1 a. The maximum nonlinearity in the first method is always half
the maximum nonlinearity of the second method [ll].
9. Distortion: Deviation from an expected output. In some cases distortion is a measure of
deviation from linearity. However, in general it need not be related to nonlinearity at all. For
example, the distortion at the output of an electronic sinusoidal oscillator, which is inherently
a nonlinear device, is a measure of the deviation of the output from an ideal sinusoidal
waveform. In the case of an amplifier, the distortion at the output is caused by non-linearities
which result in the presence (or nonproportional amplification) of frequency components at
the output that are not present at the input.
10. Conformance (conformity): Closeness of an experimental curve to a theoretical curve
or to curves obtained using least-squares or other fits. It is expressed in 070 FSO at any given
value of the measurand.
11. Hysteresis: Difference in the output of the sensor for a given input value X when X is
reached from two opposite directions, ie, from X - and X + as shown in Figure 2-1c.
Hysteresis in mechanical sensors is usually caused by a lag in the action of the sensing element.
In sensors that operate on the basis of the principle of charge sensing, the hysteresis is usually
caused by injection-type slow traps. In magnetic sensors it is caused by a lag in the alignment
of magnet moments in response to an external magnetic field. There are two other causes that
may result in hysteresis-type behavior and they should not be confused with true hysteresis.
These are friction error, which is usually observed in potentiometric transducers, and backlash
error, which is usually observed in mechanical actuators that employ gears. The friction error
can be reduced and in some cases eliminated by dithering the transducer. The backlash error
may be eliminated by careful design and fabrication of gears.
22 2 Sensor Parameters

12. Repeatability [ l l ] : The difference in the output readings at a given value of the
measurand X , where X is consecutively reached from X - (or ) 'A as shown in Figure 2-1 d.
It is expressed as 070 FSO.
13. Span: Operating range, full-scale range: the range of input variable (Xmu- Xmin) that
produces a meaningful sensor output.
14. Noise [7]:Random fluctuation in the value of the measurand that causes random fluc-
tuation in the output. Noise at the sensor output is due to either internal noise sources, such
as resistors at finite temperatures, or externally generated mechanical and electromagnetic
fluctuations. AC power line interference (50 or 60 Hz) and other external interferences are also
considered as noise, even though they are not random. The external noise will become more
important as the transducer size is made progressively smaller. An exception is discussed in
[6].The external noise in sensors is primarily associated with the random fluctuation of the
particular measurand which usually has several different components. Most of these com-
ponents can be identified as the equivalent of one of the following internal noise mechanisms
[8, 91. Internal noises, which are usually electrical in nature, are of four types: shot noise,
Johnson (or thermal) noise, recombination-generation (r-g) noise and l / f (or flicker) noise.
Shot noise is caused by charge carriers crossing a barrier at random. It is present in Schottky
barrier diodes, p-n junctions (including n - p tunnel diodes) and in thermionic emission.
+ +

Johnson noise is caused by random motion of charge carriers which produce a fluctuating emf
at the output terminals. It is present in all resistive components. Recombination-generation
(r-g) noise in semiconductors is caused by trapping and detrapping of charge carriers causing
a random fluctuation in the number of carriers and resistance. Among commonly encountered
r-g noise is burst noise. l / f noise has a spectral density that varies inversely with frequency,
being very large at very low frequencies. The origin of l / f noise is not yet well understood
and recent experiments suggesting fluctuations in the number of charge carriers (essentially
an r-g type of process) or fluctuation in the carrier mobility are still inconclusive [7].It should
be mentioned that in practice, the noise generated at the transducer is usually not the limiting
factor in the measurement accuracy [8, 91.
IS. Output Impedance: For a discussion of transducer output impedance and the restriction
that it puts on the input impedance of the following amplifier stage, see Section 2.3.
16. Grounding [12]:This is performed to establish a common node among different parts
of the system with the requirement that no potential variation along this common node with
respect to any point inside the node may occur.
17. Isolation [12]: This is performed to reduce undesirable electrical, magnetic, elec-
tromagnetic, and mechanical coupling among various parts of the system and between the
system and the outside.
18. Instability and Drgt: Change of sensitivity or the output level (with zero input) with
time, temperature, and any other parameter that is not considered part of the input.
19. Overall Performance: Tivo methods can be used in determining the overall performance
of a transducer: (i) the worst case approach, which assumes that all errors add in the same
direction and determines the overall error as the linear sum of the performance errors; (ii) the
root-mean-square (RMS) approach, which uses the root-mean-square of all the performance
errors. RMS does not assume all errors are in one direction, which makes it more reasonable
than the worst-case approach [ l l ] .
20. An Example: Figure 2-2 shows the characteristics of a resistive temperature sensor bet-
ween 10 and 15 "C. The input and output are expressed in terms of the percentage of full-scale
2.2 Damducer Characteristics 23

100 -
- -- reference line

2 80 -
4
3
+ -+ calibration cycle '1
w calibration cycle *2
-
0
0 ) .
W

-
U

g60
L
4

v
c

g40
a
-
.

t - *
3

5 20
-
0

0 -
minimum detectable signal
I

0 20 40 60 80 100
MEASURAND (percent range)

Figure 2-2. Output versus input characteristics of a resistive temperature sensor.

input and output. Two consecutive calibration cycles and error bands are shown. In
transducers where charge trapping is important, the amount of time that is spent in extreme
values of the measurand before performing the calibration cycling should be mentioned. The
inaccuracy is around E, = 10.1% (cf = 7% FSO). (It should be emphasized that for the pur-
pose of showing clearly the meaning of various parameters, a particularly poor temperature
sensor was chosen. Today's sensors have much better characteristics than that shown here).
At T = 10°C, the smallest increment in temperature that yielded a readable output, the resolu-
tion, was 0.01 "C. The maximum nonlinearity, the conformance, and the distortion are 20%
FSO. The span is 5 "C. The minimum detectable signal and the threshold are 10% FSO. The
maximum hysteresis is around 25% FSO. The static error band is around +20% FSO and
-20% FSO from the reference line.

2.2.2 Dynamic Characteristics [l, 10, 13, 15, 161

1. Transferfunction: H ( s ) = Y ( s ) / X ( s )where Y ( s )and X ( s ) are the unilateral Laplace


transforms of the input and the output respectively, with the initial conditions all set equal
to zero. The parameter s in the Laplace transform is j w + 0,where o is the frequency and
0 is a factor that it usually chosen to guarantee the convergence of the Laplace integral. If
a real function F ( t ) is defined in the region 0 < t < 00, then the Laplace transform of f ( t ) ,
denoted F(s), is given by the Laplace integral:

m
F(s) = 1 e-"'f(t) dt . (2-9)
0
24 2 Sensor Parameters

2. Frequency response: Change of the amplitude and phase of the output as a function of
the frequency of a unit amplitude sinusoidal input. It is displayed graphically as plots of
10 log I H ( w )I and phase of H ( w ) versus log (w), where H (w) is the Fourier transform of
the impulse response described below. These plots are called Bode plots.
3. Zmpulse response: h ( t )is the response of the system to a unit impulse of the stimulus.
The Fourier transform of h ( t ) gives the frequency response of the system. The Fourier
transform of a real function h ( t ) is defined as

F(w) = 2 h ( t ) eior dt (2-10a)


1/2. --OD

and the inverse Fourier transform gives h ( t ) :

1 ”
h (t) = - 1 F(o)e-ior d o (2-lob)
1/2n -m

In causal systems, h ( t ) is zero from t = - w up to the point in time that the input unit
impulse is applied. Therefore, the Fourier transform of the impulse response can be obtained
from the transfer function of the system by setting 0 = 0 in the equation of the transfer func-
tion. Figure 2-3 shows the impulse response of the temperature transducer that is described
in item 20 in the previous section. The heat pulse that was used to excite the sensor is also
shown in Figure 2-3. It should be noted that this pulse is not an ideal impulse, which is defined
as

d(t) = 0 t 2 0 (2-11)

+m
j d(t)dt = 1. (2-12)
--OD

-- Input pulse
-Impulse response

Figure 2-3.
Input heat pulse (---) applied

0.0,
______-------- to a resistive temperature
sensor and the pulse response
0 1 2 3 4 5 (-) of the temperature
TIME ($1 sensor.

Figure 2-4 shows the amplitude of the Fourier transforms of the impulse response and of the
input pulse.
4. Step response: The response of a transducer to a step change in the input. i t is usually
described in terms of rise time, decay time, and the time constant (Figure 2-5). The time cons-
2.2 Tkansducer Characteristics 25

I I - - - ,

--- Fourier transform of the input pulse


- Fourier transform of the impulse response
'r
'.'., -
n
*
g 20:
'c.
'. -. -
5
8.

-I

10-
'r
--r
-------_----------- ----r-----..
-
4

0;. . . I . . . I . . . I . . . 1 ' .

--
-
0)

In 80' : 98 2 Response Time


I
95 2 Response Time
/
b
+
L
2
I-.*90 2 Response Time
(t 0.9 )
I /
A. ---- . - ---------------I
60'
L
0

-
I --- I-----.I - - - - - -- - - - <

* . I , . .. * I -

tant (r) is defined only if the time response is truly exponential. In this case the following rela-
tionships hold: to,, = 0.104 r, = 0.693 r, to,g = 2.303 r, and tO.g/to,s = 3.32. The above
response times are defined in Figure 2-5. A measurement of t o , g / t O , 5 serves as a quick check
to see whether the response is exponential or not.

2.2.3 Environmental Parameters [I, 1I]


By environmental parameters we mean all the external variables such as temperature,
pressure, humidity, and vibration that affect the performance of the transducer. It should be
emphasized that a parameter is considered as an environmental parameter (EP) only if it is
26 2 Sensor Parameters

not the one to be sensed. For example, in pressure sensors, temperature variation, which in-
troduces undesirable variations in the sensor output, is considered to be an EP. Keeping all
the remainder of the EPs constant, in linear systems, the effect of a particular EP on the per-
formance of the transducer can be studied in terms of its effect on the static and dynamic
parameters described in the previous section.
Since temperature is one of the important EPs in most transducers, it is instructive to con-
sider its effect as a particular example. Other EPs can be treated using similar approaches.
The effect of temperature on a sensor performance is determined by two components which
are expressed as 070 FSO: temperature zero error and temperature span error.
The temperature zero error is the change in the output level of a transducer due to
temperature variation when the input (the measurand) is set equal to zero (Figure 2-6a). The
temperature span error is the change in the output level of a transducer due to temperature
variation when the input (the measurand) is set equal to its full scale (100% MR) (Figure
2-6a).

temper a tur e
span e r r o r

3
Y

-
Input-100 X MR

i//
0 0 100
v)
LL measurand (XVR) (a)

100
inputlzero 3: MR

lx zero error
ature
30
v
150
Y
temperature (OCI 3 .
0
4
3
0

0 100
rneasurand (XVR)

error'band spec.

temperature (OC 1

Figure 2-6. (a) Temperature zero and temperature span errors. (b) Butterfly specification.
2.3 Methods for Characterization of Transducers 27

Temperature performance is usually expressed as an error bar at given values of measurand


and temperature. Putting these error bars next to each other at different temperatures, the but-
terfly specification, as shown in Figure 2-6b, can be obtained. Using the butterfly specifica-
tion the upper and lower bounds of the transducer output error are determined at any given
temperature.
Temperature performance may also be expressed as an error band [ll], as shown in Figure
2-6b. Specification by the error band means that transducer error at a given input could have
any value within the specified band at any given temperature.

2.3 Methods for Characterization of Transducers [l, 5, 14-32]

2.3.1 Electrical Methods

The majority of electrical measurements consist of (1) measuring impedances and voltages
or currents, (2) breakdown fields, (3) leakage current, (4) electrical crosstalk, and (5) industrial
noise.
1. Impedances [S]: In electrical measurement it is always a good practice to have some idea
about the output impedance of the transducer. This information is necessary in choosing the
measuring equipment with proper input impedance. The measuring equipment may be a
digital voltmeter or it may consist of a preamplifier that conditions the transducer output for
subsequent detection. It is well known that in current (voltage)-sensitive schemes the output
impedance of a transducer should be larger (smaller) than the input impedance of the
measurement equipment. A variety of electrometers with input impedances in excess of
10l4... 10l6Q, nanovoltmeters, and picoammeters are now available to detect currents as low
as 10 - I 7 A, voltages as low as 10 -9 V, and charges as low as 8 x 10 -I6 C [17].
2. Breakdown [18-231: The breakdown of an insulating film is usually studied using a
metal-insulator-metal (MIM) or metal-insulator-semiconductor (MIS) structure as shown in
Figure 2-7. The insulator film can be deposited using a variety of techniques, including
chemical vapor deposition (CVD), evaporation, and sputtering. Imagine, eg, that the substrate
is silicon and the insulator is either silicon dioxide or silicon nitrite. The upper electrode is
usually either a metallic film such as aluminium or polysilicon and it can be deposited using
evaporation, CVD or sputtering. The electrical contact with the upper electrode is performed
on the thick layer as shown in Figure 2-7 to avoid damaging the thin insulator film. Depending
on the electrical test with a particular independent parameter, the breakdown phenomenon is
called dielectric strength, wear-out, or current-induced breakdown.
In dielectric strength measurement schemes, the applied voltage across the MIM or MIS
structure is ramped or stepped. At a certain voltage, the current through the structure becomes
unstable, increasing without bound for a short period of time, or the voltage across the struc-
ture drops to zero. In these measurements the ramp rate should be adjusted to yield com-
parable time-to-breakdown for different insulator thicknesses. Very small voltage ramp rates
should be avoided since they yield lower breakdown voltages. The breakdown field is
calculated by dividing the breakdown voltage by the insulator thickness. It should be men-
tioned that in these measurements appropriate precautions should be taken to prevent the
28 2 Sensor Pommeters

/metal gate

yL1
insulator

pJ-J*
semiconductor
o r metal

ground plate
a) HIM or MIS structure

Ilconst. I
I I

t(breakdown) t(breakdown)
b) Dielectrlc strength measurement c) Wear-out measurement

Vbrsrkdown
Vinj

b
t(breakdown1

d) Current-lnduced breakdown.
Figure 2-7. (a) MIM or MIS structure used to measure the breakdown field. (b) Dielectric strength
measurement. (c) Wear-out measurement. (d) Current-induced bnakdown.

destruction of the voltage and current sources when the breakdown occurs. A common prac-
tice is to use an in-line fuse or a current-limiting resistor.
The wear-out is measured as the length of time elapsed until breakdown occurs, under a
constant specified field (Figure 2-7).
The cmnt-induced breakdown is measured by injecting a preset current into the insulator.
Voltage and time are recorded until the voltage suddenly drops. The breakdown field is
calculated from the maximum value of the voltage
In studying breakdwon phenomena, various parameters are used to eliminate alternative
breakdown mechanisms [18]. The breakdown mechanisms are divided into two categories: in-
trinsic and extrinsic breakdowns. Intrinsic breakdown refers to high-field induced imperfec-
tions that consequentlyresult in higher local fields and, hence, in material breakdown. Extrin-
sic breakdown refers to imperfection/defect-induced high local fields in the insulator, which
result in a much lower breakdown field strength than expected. The breakdown mechanism
is identical in both the extrinsic and intrinsic cases. In the intrinsic case, the microvoids and
defects are generated due to the charge injection and space charge build-up whereas in the ex-
2.3 Methods for Characterization of Transducers 29

trinsic case the imperfections exist prior to the charge injection and the space charge build-up.
In some of the measurements, the MIM(S) structure area is treated as a parameter to study
the effect of the weak-spot assisted breakdown. When the MIM(S) area is reduced, it is usually
observed that the breakdown field is increased. Variation in insulator thickness is an impor-
tant parameter that can lead to nonuniform high electric fields causing weak-spot assisted
breakdown.
Temperature is another parameter that is used as an acceleration factor to obtain an activa-
tion energy (see Section 2.4).
3. Leakage [18-281: Leakage current measurements are performed to assess (1) the quality
of p-n junctions, especially in the case of silicon p-n junctions, (ii) insulator quality, and
(iii) the integrity of the packaging scheme.
i. In p-n junctions, the reverse saturation current usually has two components: the diffusion
current and the generation-recombination (r-g) current. In wide band gap semiconductors, the
(r-g) component of the reverse saturation current is usually much larger than the diffusion
component. In particular, in silicon the ratio of the r-g component to the diffusion component
is of the order of 7000. Therefore, the reverse saturation current measurement yields informa-
tion about the concentration of the r-g centers in the space charge region of the p-n diode.
ii. In MIS structures the leakage current through the insulator is monitored to identify and
study the prebreakdown period and also to assess the quality of the insulator.
iii. In the case of metallic packaging materials, the current leakage through the packaging
is a severe problem and the transducer interconnection should be carefully designed to prevent
a close encounter between the energized wires and the package. Also, the insulator used to
isolate the transducer from the package should be carefully chosen to have the desired thermal
and electrical properties under operating conditions.
4. Crosstalk: Crosstalk usually occurs in multi-channel or array transducers. It can be the
result of an overlap between the observation volume of the adjacent transducer elements or
it may be introduced after the transduction inside the transducer system due to capacitive/in-
ductive coupling or coupling through the common voltage sources. In any case it results in
correlated outputs that adversely affect the resolution of the transducer array. It is difficult
to quantify the observation volume. In semiconductors, for example, when excess carriers are
introduced, the observation volume of a current-sensing plane contact is the area of that plane
multiplied by the diffusion length of the excess carriers. However, when only electrical distur-
bances are considered in semiconductors and without introducing excess carriers, the observa-
tion volume is the area of the plane electrode multiplied by the Debye length. In other words,
the current-sensing elements should be placed a few Debye lengths apart in order to be con-
sidered independent. In gases, the mean free path, for example, would determine the observa-
tion volume in addition to the geometry of the transducer. In sensing chemical substances,
the individual sensors are very often sensitive to more than one chemical substance. This gives
rise to an inherent crosstalk between sensors that are designed to be sensitive to a given
chemical substance. See Section 2.2, item 5, and Chapter 11 for more discussion on how to
quantify crosstalk and selectivity. The crosstalk can be measured by measuring the correlation
at the output of two adjacent sensors or the correlation between the output of a chemical sen-
sor in response to various chemicals. Crosstalk usually has two components: the internal and
the external couplings (cross sensitivity of chemical sensors are of this type).
5. Industrial noise [12, 261: One of the major problems in sensitive transducers is their
vulnerability to industrial noise, which may be in the following forms: mechanical (acoustic)
30 2 Sensor Parameters

vibrations, electrical 50 or 60 Hz signal, AC magnetic fluctuations, electromagnetic in-


terference, and photon pollution. Therefore, it is desirable to test the immunity of the
transducers to these noises. A loud speaker or a shaker table with a pseudo-random noise
input are good sources of mechanical and acoustical vibrations. A transformer or a magnetic
tape eraser and fluorescent light are good magnetic and electrical noise sources.
Electromagnetic interference sources are difficult to find because these interferences might
occur over large frequency ranges. Usually an antenna attached to a few kHz or MHz
oscillator may provide a good electromagnetic interference source. Incandescent light and
fluorescent lights are good photon pollution sources, especially if a photon source with
50 (100) or 60 (120) Hz fluctuations is required. Mechanical noise is usually coupled to the
transducer through the air or directly through solid links, Damping materials such as foams
are used for mechanical isolation. The electrical fluctuations are usually coupled capacitively,
inductively, resistively, or through the ground loop. The magnetic fluctuations are coupled by
inducing electrical currents in conductive loops (inductors) or via the eddy current effect.
Faraday shielding, which basically means enclosing the transducer inside a low-resistivity,
thick (compared to the skin depth), electrically grounded package, is usually used for electrical
shielding. High-permeability materials (eg, high mu-metals) are used for magnetic shielding
(see also Vol. 5, Chapter 1). Also, sometimes it is possible to reduce electrical and magnetic
couplings by changing the arrangements of interconnections and grounding nodes. The elec-
tromagnetic interference is usually coupled through pieces of battery or interconnecting wires
that act as an antenna. It can be shielded by the Faraday shielding. The photon interference
is coupled through light-sensitive areas of transducers where the conductivity, for example, is
affected by light. It can be shielded using opaque materials.

2.3.2 Optical Methods [27-311

Most optical measurements consist of optical microscopy, interferometry, spectroscopy, and


ellipsometry. The most important application of optical measurements is in the area of optical
sensors, which has become an intense field of research in recent years. The optical
measurements are attenuation measurements, dielectric thickness measurements, refractive
index measurements, transmission spectroscopy, and reflectance spectroscopy.
The most basic measure of the quality of a waveguide is its absorption coefficient, which
determines the attenuation of the waveguide. The absorption coefficient (a) is defined through
the following relationship between the input light intensity (Ii)and the output light inten-
sity (Io):

Zo = Ii e-ax (2-13)

It is well known that a is a function of wavelength. Therefore, transmission spectroscopy


(I, versus input light wavelength) is first performed and then the absorption coefficient is
calculated at each value of the wavelength. This procedure requires a knowledge of the length
of the waveguide and it also measures an effective absorption coefficient averaged over the
length. Depending on the value of the a, various measurement schemes are developed [28].
Measurement of very small a’s is particularly difficult. In thickness measurements in-
terferometry is used [29]. The stepped area is used as one mirror in a modified Michelson in-
31

Qq
2.3 Methods for Characterization of Transducers

terferometer and the fringe displacement is measured to determine the step height. This
measurement is usually performed in an interferometric mitroscope [29, 301.
The refractive index is usually measured using ellipsometry [30], which requires a precise
knowledge of the substrate's refractive index. The refractive index can also be measured
by finding the Brewster angle for the film (waveguide). In this technique, the reflectivity of
the substrate with and without the film is measured as a function of the incident angle
(Figure 2-8).

I \

:;...
.:.:.
(.:. t
/INCIDENT LIGHT

v) COUPLING
PRISM

I WAVE GUIDE

,.
IQJ
I .
IDt
1

Figure 2-8. Optical measurements of (a) refractive index and (b) apparent refractive index. (a) The reflec-
tivity is measured as a function of incident angle both for the film-coated substrate and
uncoated substrate. The Brewster angle of the film is the angle corresponding to the crossover
of the two reflectivity curves. The refractive index of the film is equal to the tangent of
its Brewster angle. (b) The apparent refractive index is given by N = np sin [Op + sin-'
(sin O i h P )].

The Brewster angle of the film is the angle corresponding to the crossover point of these
two reflectivity curves [31]. A reflectivity curve is obtained by measuring the intensity of the
reflected beam as a function of the incident angle. It is polarization dependent and it is usually
obtained for parallel and perpendicular polarizations of light [31]. In general, the Brewster
angle is given by
tan 8, = n'/n (2-14)

where n' is the refractive index of the film and n is the refractive index of air (Figure 2-8).
Refractive index and thickness can also be determined from measurements of the apparent
refractive index (N) corresponding to a particular mode of propagation:

N = np sin [e, + sin-' (sin ei/n,)] (2-15)


where np is refractive index of a prism and 8, is the angle between the base and the input face
of the prism [29] (Figure 2-8).

2.3.3 Mechanical Methods


Mechanical characterization of transducers can be divided into two categories: mechanical
properties related to the overall reliability and integrity of the transducer and mechanical pro-
perties related to the transduction process itself. The mechanical reliability issues are discussed
32 2 Sensor Parameters

in Section 2.4. Similar techniques to those that are used in determining the mechanical proper-
ties and the integrity of the transducers are also applied to characterize the mechanical proper-
ties related to the transduction process.

2.3.4 Thermal Methods

Thermal characterization of transducers is divided into two parts: thermal properties related
to the reliability of the transducer and thermal properties related to the transduction process.
The first category deals with the proper operating temperature range and the second with the
temperature sensitivity of a transducer. The former is discussed in Section 2.4 and the latter
in Section 2.2.3.

2.3.5 Chemical and Biological Methods

Transducers can be tested for their resistance to chemicals to assess their resistance to corro-
sion in industrial and biological environments. This is the subject of the reliability studies
presented in Section 2.4. There are also chemical and biological transducers which, by design,
are sensitive to various chemicals and they perform the task of either stimulation (eg, inducing
certain reactions) or sensing in those environments. These transducers and their characteriza-
tions are discusses in detail in [l] and [14] and Volumes 7 and 8 of this series. Another issue
of importance in biomedical sensors is safety. Biomedical sensors should not produce toxic
or harmful substances either by themselves or through reaction with the environment or when
damaged or accidentally fractured.

2.3.6 Calibration Techniques [l, 5, 261

In the calibration of a transducer, the objective is to establish a relationship between the


output of the transducer and the true value of the measurand. A calibration factor is usually
determined to relate the apparent value of the measurand, measured by the transducer, and
its true value. The true value of a measurand can either be determined through other
measurements or it can be obtained from the National Bureau of Standards. The calibration
factor may or may not be a function of the measurand. The presence of random error will
result in a distribution of calibration factors at a given measurand value. Random error may
be caused by environmental parameters in addition to the random fluctuations in the
measurand value or they may be caused by internal factors. In any case, the systematic error
(error caused by faulty equipment, for example) should not be present or should be kept below
the random error. Another source that results in the distribution of the calibration factor is
the distribution in the characteristics of the transducers. The calibration curves are most useful
when they can be used with any of the transducers that are supposed to be identical but vary
owing to the manufacturing process.
Calibration curves are constructed by plotting the transducer output (usually percentage of
FSO) as a function of the measurand (usually percentage of full-scale range). A reference
linear line is also plotted and the error band is described. To demonstrate repeatability, at least
2.3 Methods for Characterization of Transducers 33

two consecutively calibration cycles (Figure 2-2) should be plotted. The most important task
in calibrating a transducer is to ensure that the quantity of the measurand that is being applied
to the transducer input is indeed what is being independently measured. In pressure sensors,
for example, the water pressures at the uncalibrated and the calibrated transducers are equal
only if these two transducers are at the same level. To see whether this acceptable situation
exists or not, one may exchange the positions of two transducers to ensure that the measured
values also agree.

2.3.7 Documentation of Results [26, 321

Since mistakes can be made in reading or transferring the data from experimental curves
and data tables, the original data sheet or graph are considered to be a most important docu-
ment. Hence they should be carefully labeled, recorded, and annotated as they are taken. Also,
a short statement at the head of the data sheet should explain the purpose of the experiment
and list all the variables to be measured. The experimental data themselves should be tabulated
and properly identified. The experimental graphs should have properly defined axies with
labels showing the independent (horizontal axis) and dependent (vertical axis) variables. All
experimental parameters such as the gain of amplifiers, bias conditions, and time constants
of filters used in modifying the signal should be clearly recorded. In general, all the informa-
tion needed to duplicate the experiment should be clearly indicated on the original data sheet.
Attached to the original data sheet is usually a report explaining what was done and how
it was accomplished. It should explain the significance of the results. In general, the form of
the report consists of three sections:
1. Abstract, summarizing the results and conclusion.
2. Important details of the procedure, data, analysis, and error estimates.
3. Calculation, supporting information and references.
The abstract usually contains the purpose of the study, a summary of the results and a short
conclusion emphasizing the significance of the study and the accomplishments. The impor-
tant details of the experimental procedure include the data and the analysis of the data, in
addition to some error estimates. The data can be presented in a variety of formats. These
include linear, semi-log, and log-log graphs. In linear graphs the transducer output is plotted
as a function of the measurand (or any other variable) using linear scales (Figures 2-2 and 2-5).
In semi-log graphs the transducer output is plotted as a function of the logarithm of an in-
dependent variable. In log-log graphs the logarithm of the transducer output is plotted as a
function of the logarithm of the independent variable. In addition to the above formats, there
are infinite ways of presenting the data; for example, in linear versus inverse of the indepen-
dent variable graphs the transducer output (or dependent variable) is plotted as a function of
the inverse of the measurand (see Figure 2-10 later). Also, for comparison, it is sometimes
necessary to normalize the entire experimental curve to have a specific value at a particular
measurand value. Normalization is often performed so that the area under the experimental
curve has a specific value. All the calculations and references are usually included as an appen-
dix. For a description of errors in measurement, see [26] and [32]. In data analysis, statistical
methods can be very helpful. It should be noted that statistical evaluations cannot improve
the accuracy of a measurement. Therefore, the system errors, which are usually systematic
34 2 Sensor ~rameters

rather than random, should be kept below the random errors for the results of the statistical
evaluations to be meaningful. Among various quantities used in the statistical analysis of the
data are averages, standard deviations, and variance [26, 321.

2.4 Reliability and Long-Term Stability Issues [33-441

2.4.1 Reliability Models and Testing

There are two approaches to obtain reliable transducers: (a) to identify, by means of testing,
the functional and reliable portion of a batch of transducers, and @) to perform failure
analysis and to attempt to eliminate any mechanism of failure and thus reduce subsequent
failures. The latter approach includes the analysis of failed devices and accelerated testing.
Usually, in practice, the second approach is supplemented by the first one. Many modern
miniaturized transducers are fabricated using integrated circuit and micromachining
technologies on silicon, and therefore their failure analysis is expected to be similar to the
failure analysis of microelectronic circuits. Also, the approaches used in failure analysis of
microelectroniccircuits can be used in analyzing transducers fabricated with other techniques.
Failure can be roughly divided into the following categories;
- catastrophic early life failures;
- short-term drifts in the parameters;
- long-term drifts and failures.
Catastrophic failures refer to the complete failure of the transducer in meeting the required
performance specification. It can occur early in the life of the device or it can occur during
or at the end of its expected normal life. In the first case it is called infant mortality and in
the second wear-out. Wear-out, short-term and long-term drifts, or changes in transducer
parameters, therefore, are the most important characteristics that need to be studied. Short-
and long-term drifts usually result in creep, zero shift, and sensitivity shift of the transducer
[l]. Creep is studied by recording the output of the transducer as a function of time while
keeping the input (measurand) fixed. Zero shift is creep when the measurand value is zero. The
sensitivity shift mults in a change in the slope of the calibration curve (input versus output
curve, Figure 2-2) and is studied by obtaining calibration curves consecutively as a function
of time or after stress cycles.
To quantify reliability, usually an acceptable range for transducer performance is defined.
Any transducer with a performance falling in this range is accepted and the remainder are re-
jected. Failure, therefore, may refer to unacceptable transducer performance in addition to its
catastrophic failure.
Before describing a few useful models that are used in reliability studies, some terminologies
(331 are defined. In the following definitions, x is a random variable such as time-to-failure
and f (x) is a probability density function:

(2-16)
2.4 Reliability and Long-Term Stability Issues 35

1. Expectation values: The mean @) of a probalitity density function f (x), which is also
referred to as the average or the expectation value, is given by

(2-17)

2. Variance (a *): A measure of how compactly the probability is concentrated around p,


the mean:

s
+m
a2 = ( x - ,U)*f(X)dx (2-18)
--m

3. Standard deviation (a): The positive square root of the variance.


4. Probability function or cumulative distribution function, F (x):Gives the probalitity that
a measured value will fall in the interval between - w and x :

(2-19)

5. Reliability function, R (x): The probability that an item will survive for a stated interval
(0 to x). If F (x) is the probalitity of failure, then one may write

number of non-failed component parts at instant x


R (x) = (2-20)
number of component parts at x = 0

or

+m
R (x) = 1 - F(x) = S f ( t ) dt (2-21)
X

6. Hazard function, h (x): The conditional probability of failure in an incremental interval


dx given that there has not been a failure just before du:

h (x) = --
f (XI f (XI
(2-22)
R(x) 1 - F(x)

2.4.2 Reliability Models.[33-401

There are many distribution functions used in reliability statistics, eg, the normal or Gaus-

aZ)
sian distribution, the log-normal distribution, the exponential distribution, the gamma ( y )
distribution, the chi-squared distribution and the Weibull distribution. For a detailed
discussion of these distributions and their applications, see [33-441. Here, we describe two dif-
ferent studies which, using different assumptions and conditions, involve the exponential and
the log-normal distributions.
36 2 Sensorpbmmeters

In the semiconductor industry it has been observed that the failure rate has the time
dependence shown in Figure 2-9. The failure rate as a function of time exhibits a “bathtub”
shape with three distinct regions: (I) infant mortality region, (11) constant failure rate region,
and (111) wearsut region. Screening is usually performed to eliminate the defective trans-
ducers that are responsible for region (I). The constant failure rate region constitutesthe work-
ing life period of the device. Since under normal conditions the life of a transducer might be
very long, a so-called accelerated aging test is usually performed to estimate failure rates. The
accelerated aging test is performed in determining and estimating the failure rates and ac-
celeration factors in regions (I) and (111).

-E
5
d

3
kn
n

-
a
L

c
Y
Working life
a
4

B
a
--
3
a Figure 2-9.
LL
The “bathtub” curve depicting the
REGION I REGION I1 REGION 111 failure rate of semiconductor
TIME devices as a function of time.

2.4.3 Screening [33, 391

The following steps are taken as the initial stress and burn-in procedure to detect defective
transducers before subjecting them to more extensive accelerated aging tests: (1) high-
temperature burn-in, (2) high-temperature storage bake, (3) overstressed electrical test, (4)
temperature cycling and thermal shock tests, and (5) mechanical shock tests. It should be em-
phasized that specific temperatures, durations, and procedures given throughout this section
are for devices and transducers fabricated using standard silicon integrated circuit
technologies. However, they may also serve as examples and guides for other technologies.
1. High-tempemture burn-in consists in subjecting the transducers to a temperature of T T
for t h, where T and t are different for different transducers. For transducers that are
fabricated using standard silicon technologies, usually one chooses T = 125 “C and t = 48 h.
Burn-in tends to eliminate the defective devices, significantly improving the failure rate in the
infant mortality region.
2. A high-tempemture stomge bake for silicon integrated circuits consists in baking the
transducer at temperatures much higher than 125 “C, typically at 25OOC for several hours. The
purpose of this test is to activate instability mechdsms such as contamination, bulk defects,
and metallization problems.
3. The electrical overstress test is performed to detect oxide, interconnection and insulator
defects that are not very responsive to temperature stresses. Stressing may consist in applying
2.4 Reliability and Long-Term Stability Issues 37

progressively larger voltages, up to 50% in excess of the specification, for different periods
of time.
4. Thermal shock testing consists in subjecting the transducer to temperatures between - 65
and +125"C for about 10 s at each temperature to test the integrity of the packaging. In
temperature cycling, the time spent in each temperature level (- 65 and + 125 "C) is increased
to 10 min and the cycle is repeated ten times (for silicon-based integrated circuits and
transducers).
5. Another test of packaging and transducer substrate integrity is the mechanicalshock test
which typically consists in dropping the package from few .meters height or attaching the
package to a shaker table.

2.4.4 Accelerated Aging Test [35,361

The more reliable a part is, the more difficult it becomes to determine its reliability. Hence,
it is usually necessary to run accelerated tests. The task, then, is to observe the performance
of a transducer at a high stress level and predict how the component part would perform under
normal conditions. Clearly, despite the simplicity of the idea, the analysis of accelerated test
data is not straightforward. Care should be taken to accelerate the aging of the component
part so that the component becomes ,,older" in the same manner in which it becomes older
under the normal conditions without changing the relative importance of the aging
mechanisms. This is called the true acceleration. Also, the following questions should be
answered. From a set of high stress level measurements, how can one determine whether or
not the true acceleration has occurred? How can the observed performance at the higher
stresses be validly extrapolated to obtain the expected performance under normal conditions?
How does one determine the acceleration factor which signifies how many hours of normal
operation correspond to 1 h of accelerated testing?
Widely accepted models in the semiconductor industry are (i) the Arrhenius model (which
yields the exponential distribution) and (ii) the log-normal distribution model. A brief
description of these models follows.
1. Arrhenius model: Denoting a measure of quality, such as the inverse of leakage current
in a p-n junction, by Q, a function of Q is defined, g (Q),which yields an exponential degrada-
tion of quality over time [33, 351:

(2-23)

where g (Q) is some function of quality, A ( T ) does not depend on time and is a constant
degradation rate which depends on the temperature T (or any other relevant stress parameter),
and t is the time. The degradation rate can be redefined as the failure rate if some margin of
acceptability (error band) is defined and any component part with quality parameter Q out-
side this margin is considered to have failed. The Arrhenius equation is written as

A (7.) = eo-b'kgT (2-24)

where (I and b are empirical constants and k, is the Boltzmann constant (Figure 2-10). In
general, the constant b can be considered as an activation energy associated with the activation
of a particular failure mechanism (Figure 2-10).
38 2 Sensor Parameters

L I I

ACCEPTABLE ACTIVATED NOT ACCEPTABLE

QUALITY FACTOR

'E - ACTIVATION ENERGY

lOOOlT (K)
Figure 2-10. The Arrhenius model (a) and Arrhenius plot of a semiconductor device (transducer) (b).
(a) Potential diagram of aging process of silicon based transducer. According to this poten-
tial, the probability of transition from acceptable to not acceptable state is eAElksT(k, is
the Boltzmann constant and Tis the temperature in degrees Kelvin). (b) The Arrhenius plot.
I, is the projected failure rate at 30°C. There are two modes of failure and only one of
them is taken into account.

-'
The natural logarithm of A ( T ) as a function of (kBT ) can be plotted in an Arrhenius
plot to obtain a and b as the intercept and slope of In (A (r)). To calculate the relationship
between the accelerated time and normal time, let us assume that at some time t and under
normal operating temperatures T the quality factor Q is equal to Q' at some time t' under
high stress temperature condition T'.Hence one can write

(2-25)

However, from Equations (2-24) and (2-25) one obtains


(ea-b/kBT) t = (ea-b/kgT') tt (2-26)

which on solving for t and denoting the factor e(b(l'kBT-l'kBT')by a yields

t = at' (2-27)

Thus t' h of operation at elevated temperature is equivalent to t h of operation at normal


temperature with the above equation giving the relationship between t and t'. To analyze more
than one failure mechanism, see [43].
2.4 Reliability and Long-Term Stability Issues 39

2. Log-normal distribution model: According to the Arrhenius model, L (T) does not de-
pend on time and an acceleration is said to be a true acceleration if the Arrhenius plot is linear.
In practice, however, the degradation or failure rates are not constant and there is an aging
phenomenon that results in the time dependence of the failure rates. Therefore, usually the
acceleration effects are determined by examining the degree to which the life distribution, ie,
the distribution of time-to-failure, moves with time under the influence of a variation in stress.
This implies that the failure distribution must be measured at sufficiently high stress levels.
Also, the longer the time one can test the distribution of time-to-failure, the more confidently
the distribution parameters can be estimated. The distribution of the time to failure is a log-
normal distribution in the case of semiconductor component parts. In its simplest form, the
log-normal distribution may be defined as the distribution of a random variable whose
logarithm obeys the Gaussian distribution [38, 391. The probalitity density function of a log-
normal distribution is given by 1381

(2-28)

where cGand ,uG are the standard deviation and the median of the corresponding [39] nor-
mal distribution. It can be shown that the expectation value of the log-normal distribution
is given by

and its variance is given by

It has been shown that if the time-to-failure of silicon-based devices, which has a log-nor-
ma1 distribution, is plotted as a function of cumulative percentage failures in a log-log plot,
a straight line is obtained [36] (Figure 2-11). In some cases, however, it has been observed that
these plots have two linear regions (bimodal distributions), with the first region having a

1 o6

10

1
Figure 2 4 .
Log-normal plot of a silicon n-p-n transistor. CUMULATIVE X FAILURES
40 2 Sensor Parameters

steeper slope 133, 341. This first region is usually identified with the freak and infant mortality
failures, representing the products that have relatively shorter lifetimes. To analyze the freak
data, the Weibull probalitity distribution is usually used. For a detailed discussion of handling
the data, including the freaks, see [33] and [44].The median life and the standard deviation
are the distribution parameters of the log-normal distribution. The median life is estimated
from the log-normal graphs and it is the time corresponding to the 50% cumulative failure.
The standard deviation is given by

17 = In (time to 50% failurehime to 16% failure) . (2-31)

To predict the time-to-failure of component parts under normal stress conditions, the standard
deviation and median life are determined for log-normal distributions under different stress
levels.
Using log-log graph paper, one chooses the ordinate as the relevant stress variable, such as
T I , and the abscissa as life in hours. Median life plotted in this coordinate system usually
appears as a linear line [32]. By extrapolation, one therefore determines the median life of the
component parts at normal stress levels. To calculate the standard deviation and failure rates
at normal stress levels, however, one needs to construct a graph of instantaneous failure rate
versus median life using the log-normal distribution [36].

2.4.5 Failure Mechanisms

Among various mechanisms that might be responsible for the degradation of transducers,
the electrical and mechanical failure mechanisms are the most important.

2.4.5.1 Electrical Failure Mechanisms [18, 24, 33-36]

Electrical failure mechanisms in silicon-based transducers are of the same type that exist
for microelectronics. It is arbitrary to divide the failure mechanisms into electrical and
mechanical. All the mechanisms that result in mechanical damage to the transducers package
or its substrate are considered as mechanical mechanisms. In semiconductor devices, the elec-
trical failure mechanisms are as follows. (i) Insulator breakdown, which is explained in Section
2.3.1. (ii) Ionic contamination, usually caused by Na , C1 -, and K + ions. Na + , owing to its
+

small radius, is the most mobile ion. These contaminants are introduced through the environ-
ment, human contact, materials used in processing and fabrication, and packaging materials.
(iii) Surface charge spreading is observed between biased metal conductors at the surface of
insulators, reducing inter-electrode resistances and leading to electrode meltdown. Moisture
and ionic contamination have been found to be responsible for surface charge spreading.
(iv) Charge effects lead to leaky gate oxide, explained in Section 2.3.1. (v) Hot electron effects
lead to charge injection from the channel of a metal-oxide-semiconductor field-effect tran-
sistor into the gate oxide. This also leads to the breakdown of oxides. (vi) Piping is an electrical
shunt path observed along crystal defects that are perhaps generated during the wafer process-
ing stage. (vii) Dislocations caused by diffusion of impurities within the silicon.
2.4 Reliability and Long-l2rm Stability Isnres 41

In addition to the above mechanisms, the following are responsible for the failure of metal
interconnects. (i) Corrosion, which is an electrochemicalmechanism, is observed in microelec-
tronics in DC biased metal interconnection in the presence of ionic contaminants and
moisture. This mechanism is very important in chemical sensors. (ii) Electromigration is the
migration of metal in metallic interconnections caused by large current densities. The con-
tinuous impact of electrons on the aluminium atoms causes them to move in the direction of
electron flow, piling up in one end of the track and creating a void at the other end. This is
a regenerative process since smaller cross-sectional areas lead to larger current densities and
larger current densities result in enhanced electromigrationand, hence, in even smaller cross-
sectional areas. (iii) Contact migration is observed in metal (aluminium)-silicon junctions,
which is caused by migration of aluminium into the silicon or vice versa. Contact migration
leads to the formation of aluminium spikes inside silicon, electrically shorting narrow p-n
junctions. (iv) Particulate contaminants, which are usually particles of silicon from scribing,
gold flakes, and ends of bond wires, all of which lead to electrical shorts. (v) Radiation results
in electron-hole generation. Depending on the energy of these charge carriers, they result in
either permanent damage, such as in oxides and the generation of microvoids in semiconduc-
tor materials, or they result in a momentary increase in leakage currents and noise levels.

2.4.5.2 Mechanical Faiiun? Mechanisms [24, 33, 34, 411

In microtransducers fabricated using micromachining technology, there is not yet sufficient


reliability data to identify the main failure mechanisms. In these transducers, however, it is ex-
pected that the failure mechanisms are basically of the type that exist for the microelectronics.
Among the failure mechanisms that have been studied so far and reported in microsensors are
the following. (i) Interconnect failures, which are caused by thermal cycling which leads to ex-
cessive mechanical stresses on the interconnects. (ii) Cracked transducer/package caused by
differences in the thermal expansion coefficients of the microtransducer and the packaging
material resulting in low cycle fatigue and cracks. (iii) Delamination between the transducer
and the packaging material, which is usually caused by the presence of moisture at the relevant
surfaces before the lamination and temperature cycles. (iv) Generation and propagation of
microcracks has also been observed in transducers with silicon diaphragms and cantilever
beams. (v) In these transducers, the packaging sometimes results in deformation of thin silicon
diaphragms. (vi) Mechanical stress relaxation is another form of metal migration; atoms in
the areas of high stress migrate to equalize the stress, which results in the deformation of
metals.
Some of the above mechanisms are also responsible for the failure of the device while in
storage. It appears that the only available techniques to test the storage reliability are either
temperature or humidity cycling. Such accelerated tests provide an estimate of device
reliability. The most difficult task in determining the storage reliability is in understanding the
exact cause of the failure.
42 2 Sensor Parameters

2.5 References

[l] Norton, H. N., Sensor and Analyzer Handbook; Englewood Cliffs, NJ: Prentice Hall, 1982.
[2] Electrical Tmnsducer Nomenclature and Terminology, ANSI MC6.1-1975; Research Triangle Park,
NC: Instrument Society of America, 1975.
(31 Terms and Definitions in Industrial-Process Measurement and Control, IEC 65-Secretariat-84 draft,
International Electrotechnic Committee, 1982; and VDI/VDE 2600, Metrologie, Dusseldorf, 1973.
[4] KO, W. H., “Biomedical Transducers” in: Handbook of Biomedical Engineering, New York:
Academic Press, 1988.
[5] Doebelin, E. O., Measurement Systems: Application and Design, 3rd ed.; New York: McGraw-Hill,
1983.
[6] Gray, P. R., Meyer, R. G., Analysisand Design of Analog Integrated Circuits; New York: Wiley, 1977.
[7] van der Ziel, A., Noise in Solid State Devices and Circuits; New York: Wiley, 1986.
[8] Chau, H. L., Wise, K. D., “Noise Due to Brownian Motion in Ultrarensitive Solid-state Pressure
Sensors”, IEEE Trans. Electron Devices ED-34 (1987) 859.
[9] Spencer, R. R.. Fleischer, B. M., Barth, P. W., Angell, J. B., “A Theoretical Study of Transducer
Noise in Piezoresistive and Capacitive Silicon Pressure Sensors”, IEEE Trans. Electron Devices
ED-35(1988) 1289.
[lo] Lathi, B. P., Signals, Systems and Communication; New York: Wiley, 1961.
[I11 Tasker, R. E., “What Transducer Performance Specs Really Mean”, Sensors, November (1988)
22-27.
[12] Morrison, R., Grounding and Shielding Techniques in Instrumentation, 2nd ed.; New York: Wiley,
1977.
[13] Industrial Platinum Resistance Thermometers, IEC publication 751; International Electrotechnical
Commission, 1983.
[I41 Danzer, K., Than, E., Molch, D., Kuchler, L., Analytik; Stuttgart: Wissenschaftliche Verlagsgesell-
schaft, 1987.
[15] Frederick, D. K., Carlson, A. B., Linear Systemsin Communicationand Controt New York: Wiley,
1971.
[I61 Zeitverhalten von Beriihrungsthermometern; VDI/VDE Standard 3522, VDI, Dusseldorf, FRG.
117) Keithly, J. F., Low Level Measurements; Cleveland, OH: Keithly Instruments, 1984.
1181 Wolters, D., Verwey, J. F., “Breakdown and Wear-Out Phenomena in SiO, Films” in: Instabilities
in Silicon Devices, Barbottin, G., Vapaille, A. (eds.); New York: North-Holland, 1986.
[I91 Nicolian, E. H., Brews, J. R., MOS Physics and Technology; New York: Wiley, 1982.
[20] Ngyen, T. N., Olivo, P., Ricco, B., “A New Failure Mode of Very Thin (< 50 A) Thermal SiO,
Film” in: Proceedings of IEEE 1987 International Reliability Physics Conference, Publication No.
87CH2388-7, 1987, p. 66.
1211 Wolters, D. R., “Breakdown and Wearout Phenomena in SiO,” in: Insulating Films on Semicon-
ductors, Schulz, M., Pensl, G. (eds.); New York: Springer, 1981, p. 180.
[22] Ghandi, S. K., The Theory and Practice of Microelectronics; New York: Wiley, 1968.
[23] Sze, S. M., Physics of Semiconductor Devices; New York: Wiley, 1981.
[24] Modern Microelectronic, IC Application, Fabrication Technology Vol. 1; New York: Research and
Education Association (REA), 1981.
[25] Modern Microelectronic, IC Application, Fabrication Technology Vol. 2; New York: Research and
Education Association (REA), 1981.
[26] Wolf, S., Guide to EIectronicMeasurementsand Laboratory Practice; Enlewood Cliffs, NJ: Prentice-
Hall, 1983.
1271 Zernike, F., “Fabrication and Measurement of Passive Components” in: Integrated Optics, Tamir,
T. (ed.); New York: Springer, 1975.
[28] Goell, J. E., Standley, R. D., “Integrated Optical Circuits”, Proc. IEEE 58 (1970) 1504.
[29] Silver, M. D., Chow, E. K., “Thickness Measurement of Thin Permalloy Films: Comparison of X-
Ray Emission Spectroscopy, Interferometry, and Stylus Method”, J. Vac. Sci. Techno12 (1965) 203.
[30] Abeles, F. A., “Optics of Thin Films” in: Advanced Optical Techniques, van Heel, A.C.S. (ed.); New
York: Wiley, 1967, p. 143.
2.5 References 43

[31] A u a m , R. M. A., Bashara, N. M., Ellipsometry and Polarized Light; Amsterdam: North-Holland
Physics, 1987.
(321 Bevington, P. R., Data Reduction and Error Analysis; New York: Wiley, 1987.
(331 Amerasekera, E. A., Campbell, D. S . , Failure Mechanisms in Semiconductor Devices; New York:
Wiley, 1987.
(341 Sinha, S. K., Reliability and Life Testing; New York: Wiley, 1985.
(351 Thomas, R. E., Gorton, H. C., “Research Toward a Physics of Aging of Electronic Component
Parts” in: Proceedings of Symposiumon the Physics of Failure in Electronics, Chicago, Goldberg,
M. F., Vaccaro, J. (eds.); 1963, Vol. 2, p. 25.
(361 Peck, D. S., “The analysis of Data From Accelerated Stress Tests” in: Proceedings of ZEEE 1971 Zn-
ternational Reliability Physics Conference; Publication No. 71-C-9-PHY, 1971, p. 69.
(371 Platteter, D., “Basic Integrated Circuit Failure Analysis Techniques”, Proceedings of ZEEE 1976 Zn-
ternational Reliability Physics Conference; Publication No. 76CH1072-8, 1976, p. 248.
(381 Larson, H. J., Shubert, B. O., Probabilktic Models in Engineering Sciences; New York: Wiley, 1979.
(391 Aitchison, J., Brown, J. A. C., TheLognormal Distribution; New York: Cambridge University Press,
1969.
[40] O’Coonor, P. D. T., Practical Reliability Engineering, 2nd ed.; Chichester: Wiley, 1985.
(411 Uebbing, J., “Mechanisms of Temperature Cycle Failure in Encapsulated Optoelectronic Devices”
in: Proceedings of ZEEE 1981 International Reliability Physics Conference; Publication No.
CH1619-6, 1981, p. 149.
(421 Boresi, A. P., Sidebottom, 0. M., AdvancedMechanicsof Materials, 4th ed.; New York: Wiley, 1984.
(431 Kaplan, E. L., Meier, P., “Non-parametric Estimation from Incomplete Observations”, JASA 53
(1958) 457.
(441 Parker, C. D., Integrated Silicon Device Technology-VolumeX K Reliability; Springfield, VA: Clear-
inghouse, 1967.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

3 Sensor Modeling
HENRY
BALTES.Swiss Federal Institute of Technology (ETH). Zurich.
Switzerland.
AROKJA University of Waterloo. Waterloo. Ontario. Canada
NATHAN.

Contents
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.1.1 Motivation for Microsensor Modeling . . . . . . . . . . . . . . . 46
3.1.2 Overview of Numerical Modeling Work . . . . . . . . . . . . . . . . 47
3.2 Model Equations for Electronic Phenomena . . . . . . . . . . . . . . 49
3.2.1 Temperature Effects . . . . . . . . . . . . . . . . . . . . . . . . 50
3.2.2 Optical Radiation . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2.3 Magnetic Field Effects . . . . . . . . . . . . . . . . . . . . . . . 52
3.2.4 Mechanical Effects . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2.4.1 Modulation of Minority Carrier Injection . . . . . . . . . . . . . . 53
3.2.4.2 Piezoresistive Effect . . . . . . . . . . . . ............ 54
3.2.4.3 Capacitive Effects . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3 Model Equations for Mechanical Structures . . . . . . . . . . . . . . 56
3.4 Numerical Procedures . . . . . . . . . . . . . . . . . . . . . . . 57
3.4.1 Mesh Generation . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.4.2 Discretization Procedures . . . . . . . . . . . . . . . . . . . . . . 60
3.4.3 Solution Procedures . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5.1 Photodiode . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5.2 Magnetic Sensors . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5.3 Mechanical Sensors . . . . . . . . . . . . . . . . . . . . . . . . 67
3.6 Numerical Modeling as a Microsensor CAD Tool . . . . . . . . . . . . 74
3.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
46 3 Sensor Modeling

3.1 Introduction

In contrast to the mature field of semiconductor IC device modeling, the art of sensor
modeling is in its infancy. With the growing demand for sensors, adequate modeling tools
facilitating sensor design and analysis become more and more desirable. This is particularly
true for microsensors, i. e. sensors designed and manufactured using methods and processing
steps of semiconductor IC technology.
The’modeling of microsensors is the object of this chapter, where we review a variety of
physical approximations, model equations, numerical techniques, and typical results. We
focus on physicuf microsensors, i. e. thermal, optical, magnetic, and mechanical measurands.
Although the models described here are pertinent only to semiconductor sensors (and to
silicon sensors in particular), the concepts related to numerical methods and thermal stress
analysis can be readily applied to a wider class of sensors. The role of modeling in microsensor
development is demonstrated by selected examples.

3.1.1 Motivation for Microsensor Modeling

The development of new semiconductor microsensors usually involves several design and
fabrication cycles until the specifications are satisfied. Numerical modeling can reduce the
number of costly trial-and-error steps in this development and provide a unique insight into
the functioning of microsensors by means of the distributions of various physical quantities
in the interior of the device not readily accessible to experiments.
Simple analytical models of microsensor operation are usually the first choice and they in-
deed serve as valuable heuristic tools for the design of trial devices. Unfortunately, they may
turn out to be correct only under very special geometric and operating conditions. In more
general situations, and for an in-depth investigation of the physical principles and transduc-
tion mechanism(s) involved, it is highly desirable to solve the equations numerically (under
pertinent boundary conditions) that adequately describe the relevant physical effects in the
sensor. Closed-form solutions to the equations may be made possible by simplifying assump-
tions. These assumptions, however, may not do justice to the physical nature of practical
devices. For example, the assumption of homogeneous material is inadequate for most
micromechanical structures since they inherently contain material inhomogeneities such as
multiple layers. In addition, there can be variations in the thickness of the structure. Moreover,
analytical solutions are valid only for the specific boundary conditions and geometry for
which they have been derived.
The goal of numerical modeling is to understand the sensor’s operating principles and, in
particular, how the design, fabrication, and operating parameters determine, enhance, or limit
its sensitivity with respect to the measurand under consideration. By design, the presence of
the physical or chemical input signal is meant to “upset”, in a controlled manner, an “or-
dinary” IC device as much as possible, making the task of modeling the device even more
essential. The various signals (such as force, magnetic field, temperature, and chemical con-
centration) interact with the inherent electrical transport in a complicated and involved
fashion. For example, magnetic induction disturbs the carrier transport by the b r e n t z force;
incident radiation alters the generation-recombination balance in photodetectors ; mechanical
3.1 Introduction 47

stress modulates the electric conductivity, the capacitance of a microcavity, or some


mechanical resonant frequency. The measurand often appears in the form of an external field
that reduces the symmetry of the device operation, hence making the choice of appropriate
model equations, physical and material parameters, and boundary and interface conditions
crucial. Modeling results based on the pertinent material and process parameters can help to
assess the portability of sensor designs from one chip fabrication process to another.
Sensitivity estimates and related optimization of pertinent sensor structures are obvious
goals of sensor modeling [l]. One example is to find the optimal shape of an integrated Hall
plate with minimum offset and to predict the impact of fabrication geometry tolerances on
the performance of the device. Another example is a diaphragm where simulation of the stress
distribution aids in limiting the nonlinearity by optimizing the shape and size, in addition to
resistor position for a given pressure range [2]. This approach avoids having to provide com-
pensation by resistor trimming or by electronic circuitry.
In addition, valid sensitivity models may be obtained which are required for the evaluation
of material parameters from sensor performance measurements [3]. Also, the performance
sensitivity to process parameters and their variations, and the effects of cross sensitivity
associated with microsensors, can be adequately predicted. Through appropriate
mathematical reformulation, modeling results can finally be incorporated into existing CAD
(computer-aided design) and simulation tools [4].

3.1.2 Overview of Numerical Modeling Work

Solar cells and photodiodes are probably the best understood semiconductor sensors and
the latter can often be treated adequately by one-dimensional numerical modeling. A variety
of one-dimensional (PClD [5]) and two-dimensional [6] numerical codes have been developed
that solve the system of nonlinear partial differential equations (Poisson’s equation and the
time-dependent carrier continuity equations) in the presence of optical radiation. PClD is a
finite element code tailored to the personal computer environment (it runs on IBM XT com-
patible computers). It is menu driven and provides the user with a variety of sophisticated
physical models for both silicon (including a-Si :H) and GaAs photodiodes. It also allows
user-input tables of data. The models for the various physical processes take into account
temperature dependence. However, the package does not include a physical model for electron-
hole pair generation through impact ionization and hence avalanche effects in photodiodes
cannot be modeled. The effects of high doping are accounted for through band-gap narrowing
and carrier-mobility reduction, and band-energy variations and Fermi-Dirac statistics. Besides
the usual boundary conditions at contact regions, default and user-supplied surface recom-
bination velocities can be entered for the photodiode front surface.
Magnetic sensors: require at least two-dimensional modeling in view of the vector character
of magnetic induction [l, 71. In certain limiting cases, the system of partial differential equa-
tions governing galvanomagnetic carrier transport in semiconductors can be simplified under
various assumptions (such as negligible concentration gradients, negligible minority carrier ef-
fects, negligible generation, recombination, and space-charge effects). Moreover, in the
presence of homogeneous magnetic fields, the problem reduces to Laplace’s equation for the
electric potential, viz., the classical Hall plate equation. Both finite difference [8- 141 and
finite element techniques [15, 161 have been employed to obtain the desired potential and cur-
48 3 Sensor Modeling

rent-density distributions in the Hall plate. For magnetic-field-sensitive split-drain MOSFETs


(dual- and triple-drain structures) operating in the linear region, two-dimensional distribu-
tions of electric potential, current density, and surface charge in the inversion layer have been
presented [13, 141 for magnetic induction perpendicular to the chip surface. In the presence
of inhomogeneous induction perpendicular to the device surface (such as in magnetic domain
detectors), the modeling problem is more involved [17, 181. Using a finite difference scheme,
the Hall voltage in a Hall cross geometry was computed for the case when the magnetic field
distribution is inhomogeneous in one direction and constant in the other [19]. For magnetic-
field-sensitive bipolar devices (magnetotransistors), the analysis has to take into account the
complete system of partial differential equations with judiciously chosen boundary condi-
tions. Solutions based on finite difference (20-231 or finite box discretization procedures
[24-261 have been presented. A code developed for the simulation of semiconductor magnetic
field sensors is ALBERTINA, which provides numerical solutions to the system of equations
governing carrier transport in Hall-type and bipolar magnetic field sensors in the presence of
both homogeneous and inhomogeneous induction [l].
Mechanical sensors: there have not been many reports on a full numerical analysis of the
interactions of mechanical, thermal, and electric effects in pressure transducers and related
devices. We are only aware of a PhD Dissertation [27] and a limited number of publications
[28, 291 from which the package SENSIM evolved. The package, based on a two dimensional
finite difference code, solves the system of partial differential equations that govern the stress
resultants and deflection in a diaphragm. It yields the output response of piezoresistive or
capacitive diaphragms to pressure and temperature variations. The equations employed are
based on the thermoelastic thin-plate theory [30]. They are valid as long as the mechanical
device satisfies the quasi-static, plane-stress, isothermal, and small-deflection conditions. The
available user-specified outputs include diaphragm deflection, stress distributions over the
diaphragm, average stress over the piezoresistors, output voltage from piezoresistive network,
stress induced resistance changes, capacitance changes with pressure, and thermally induced
stresses or bending moments. A variety of two- and three-dimensional finite element packages
have been employed to simulate mechanical behavior, eg, stress distribution in diaphragms
[31-331, stresses due to packaging of diaphragm [2], thermal stresses in the diaphragm [34],
and stress and modal analysis of cantilever accelerometers [35], and to extract (and verify)
values of pertinent mechanical parameters such as Young’s modulus [36]. In [35], a general-
purpose finite element package ANSYS [37], from Swanson Associates, was employed. There
are analytical solutions based on Fourier expansions that describe the static and dynamic
behavior under the various boundary conditions for the different geometries [30, 381.
However, obtaining such solutions requires a good understanding of the physical nature of the
problem and can turn out to be very involved (if possible at all), particularly if there are struc-
tural inhomogeneities and thermal effects.
Analytical models describing the interactions of electrical and mechanical phenomena in
piezoelectric media, are not adequate for the design of piezoelectric transducers with realistic
geometries and material parameters (see (391 and references therein). For a rigorous analysis
of the underlying device physics and accurate prediction of the electrical and mechanical
device characteristics, it is mandatory to seek numerical solutions in two or three dimensions,
to the system of fundamental equations that couple the electric and mechanical phenomena
in the piezoelectric element. In particular, a finite element scheme is desirable because of its
inherent flexibility in handling arbitrary device geometries as well as anisotropies in the
3.2 Model Equations for Electronic Phenomena 49

piezoelectric material. To account for the interactions of the transducer with the ambient
acoustic media (as in the case of transducers emitting or sensing sound fields), solutions to
the wave equation governing the propagation of acoustic waves in the ambient fluid are re-
quired. A finite element scheme providing solutions in two as well as three dimensions to such
a system of coupled equations has been successfully employed in the development of
telephone and ultrasonic transducers, mechanical delay lines, and surface acoustic wave
devices [39, 401. The simulations yield the pertinent mechanical and electrical distributions
in the piezoelectric as well as the ambient acoustic media. Simulations and measurements of
electrical impedance, mechanical displacement, as well as pressure distribution in the ambient
acoustic fluid appear to be in good agreement.

3.2 Model Equations for Electronic Phenomena

The fundamental system of partial differential equations that describe the physical pro-
cesses occuring in semiconductor devices in general, are usually given in the following
form [41]:

div(egradty) = -q@ - n + ND - NA)


div J, - q an/at = q ( R - G ) (3-1)
divJp + q a p / a t = -q(R - G),

with J,, Jp given by the transport relations

In Equations (3-1) and (3-2), E denotes the material permittivity, q the elementary charge,
ty the electrostatic potential, NDand N A denote the fully ionized impurity distribution, D,,
Dp denote the diffusion constants, and R and G denote the recombination and generation
rates, respectively. The carrier concentrations n and p are assumed to follow Maxwell-Boltz-
mann statistics:

where Vt denotes the thermal voltage (kBT/q),nie the effective intrinsic concentration (taking
into account heavy doping effects and gap energy narrowing), and p, pp denote the respec-
tive Fermi potentials. Although Equations (3-1) and (3-2) are a result of many approximations
[41], they are nevertheless justified in view of the physical dimensions and operating condi-
tions normally encountered in practical integrated sensing structures. In semiconductor sen-
sors, depending on the nature of the measurand, the various quantities in Equations (3-1) and
50 3 Sensor Modeling

(3-2) have to be suitably modified to account for the variety of physical effects (and their in-
teractions) taking place in the device.
Equations (3-1) and (3-2) are generally solved subject to a mixture of Dirichlet and homoge-
neous Neumann boundary conditions. The electric potential and carrier densities at ohmic
contacts (assumed ideal) are prescribed by the usual Dirichlet conditions [42]:

w = V, + ( k B T / q )sinh-‘ (N/2nie)
n = (N2/4 + r ~ k ) ”+~ N / 2 (3-4)
p = (N2/4 + rife)"' - N / 2 ,
where V, is the applied voltage and N (= ND - NA) denotes the net ionized impurity con-
centration. For contacts with an external resistor tied to the voltage source or for contacts with
a non-negligible contact resistance, the electrostatic potential in Equations (3-4) have to be
duly modified. In the case of contacts with a Schottky-barrier junction, a different set of rela-
tions based on thermionic-diffusion theory has to be employed [42].
At the interface between two different media, the net electric displacement normal to the
interface is assumed to be equal to the interface charge density. The current densities at in-
sulating boundaries are determined by the recombination at the interface, for nonideal inter-
faces. At the outer edges of the device’s simulation domain, homogeneous Neumann boun-
dary conditions are applied for the electrostatic potential, viz, grad v/ . n = 0, with n denoting
the outward normal vector. If this outer edge happens to be part of the active device region,
zero normal electric current densities are assumed for ideal interfaces. An elaborate account
of physical models for the various boundary conditions can be found in [42].

3.2.1 Temperature Effects

Semiconductor devices are sensitive to variations in temperature, whether applied externally


or generated within the device. A variety of methods are possible for utilizing devices and cir-
cuits in standard IC technology, for thermal sensing applications. For measurement of en-
vironmental temperature, the highly predictable and stable temperature dependence of the
base emitter voltage, VBEof a bipolar junction transistor lends itself well to temperature sens-
ing. Together with pertinent circuitry (for amplification, biasing, and signal correction) in-
tegrated with the transducer, an output voltage or current proportional to absolute tempera-
ture can be achieved (see also Chapter 10). On the other hand, the thermoelectric effect, eg,
integrated thermopiles can be exploited for the transduction of a nonthermal signal into an
on-chip temperature difference. A review of principles and applications of thermal sensors can
be found in [44, 451.
To allow for the effects of temperature, the transport Equations (3-2) are modified and for
not too large temperature gradients they read
3.2 Model Equations for Electronic Phenomena 51

These equations include a component of electric current density with the temperature gra-
dient as a driving force, and this current component is crucial in the analysis of thermopiles.
Here, q,,, aspdenote Seebeck coefficients for the respective n- and p-type material and are
dependent on both temperature and material composition. In general, for nondegenerate
silicon the coefficients can be represented in the form [44]

where s,,,sp(the exponents describing the relaxation time and carrier energy) are of the order
of -1 to 2, N,.,Nv denote the density of states in the respective conduction and valence
bands, and @ is the phonon-drag contribution ranging from 0 (highly doped silicon) to ap-
proximately 5 (low doped silicon) at room temperature At low temperatures, the correspond-
ing range of @ can be from 0 to 100. In Equation (3-9, the temperature-dependent concentra-
tion diffusion constants, On, Dp,and the drift mobilities, p,,, pp, are assumed to be related
by Einstein's relations, viz.,

Electrical and thermal interactions in the device can be accounted for by an additional heat
flow equation, viz.,

where x ( 7')denotes the thermal conductivity, p the mass density and c the specific heat; H ac-
counts for the various heat sources and sinks in the system. An elaborate account of the
temperature dependence of the various terms in the transport relations can be found in [41,
421.
As for the boundary conditions, system (3-4) is employed for the electrostatic potential and
carrier densities, but with nie suitably modified to account for its temperature dependence
Depending on the particular problem, boundary conditions for Equation (3-8) may consist
of Dirichlet conditions, adiabatic conditions, or mixed conditions [45, 461. Mixed boundary
conditions are employed when simulation of thermal behavior over a large domain can lead
to high computational requirements [&I. Equations (3-1), (3-9, and (3-8), including the
dependence of the various terms on temperature, together with the pertinent boundary condi-
tions adequately describe the interactions of electrical and thermal phenomena in integrated
thermal sensors.

3.2.2 Optical Radiation

Physical processes in optoelectronic devices such as photodiodes and solar cells can be
described by Equations (3-1) and (3-2), but optical generation of carriers has to be taken into
52 3 Sensor Modeling

account. For monochromatic radiation with zero reflectance at the diode back surface, the op-
tical generation rate G at depth x into the device can be generally expressed as [5, 6, 471

G(x) = 5 (1 - T)pae-axdA, (3-9)


0

where (p is the incident photon flux per unit area normal to the device, r is the front surface
reflection coefficient, and a is the absorption coefficient, all being functions of the wave-
length, A. For multiple reflections between front and back surfaces, the integrand in Equa-
tion (3-9) becomes more involved [47, 481. In avalanche photodiodes, the high reverse bias
voltage across the depletion zone leads to electron-hole pair generation from high-energy elec-
trons and holes in the depletion region, thus requiring the inclusion of impact ionization in
the continuity equations. High doping effects in solar cells result in the recombination term,
R, accounting for both Auger and Shockley-Read-Hall processes. The dependence of the in-
trinsic carrier concentration and carrier lifetimes, T,, T ~ on
, doping concentration are modi-
fied accordingly [41, 421. Also, the electric potential in the current density Equations (3-2)
have to be replaced with effective potentials accounting for variations in band structure and
the Fermi-Dirac statistics [49].
The boundary conditions discussed in Section 3.2 also hold for the simulation of such op-
toelectronic devices.

3.2.3 Magnetic Field Effects

In the presence of a magnetic field, moving charge carriers in a semiconductor are subject
to the Lorentz force. Several phenomena can result as a consequence of the Lorentz force, viz.,
Hall effect, carrier deflection, magnetoconcentration, and magnetoresistance. Integrated
magnetic sensors (such as Hall plates, magnetotransistors, and magnetodiodes [7]) exploit one
or the other of these phenomena. The Lorentz force on moving carriers manifests itself in the
transport equations. Under various assumptions [50], the magnetic-field-dependent electric
current densities can be expressed in the classical drift-diffusion formulation as

(3-10)
J p ,-
~ P p * J p .X~ B = - 4 0 , [gradp + p g r a d ( q y / k , T ) ] .

These transport relations are a good approximation only in the weak field limit, ( p t B ) ’ ,
(p,*B)’ Q 1, with a relative error of the order of (p,*B)*,( P , * B ) ~ . equations take into
The
account the indirect effects of temperature through the T dependence of the various coeffi-
cients, but they do not include thermoelectric and thermomagnetic effects. The Hall mobility,
,ut is assumed to be proportional to the drift mobility, p,, with the constant of propor-
tionality being the Hall scattering coefficient. The experimental value of this coefficient has
been found to be significantly different from theory, particularly in the limit of ionized im-
purity scattering [51]. In general, this coefficient depends on the nature of the scattering
mechanism, the band structure, the degree of degeneracy, and the statistics characterizing the
velocity distribution of carriers (521.
3.2 Model Equations for Electronic Phenomena 53

The boundary conditions at contact regions in integrated devices take the usual form (see
Section 3.2). At insulating boundaries, particularly in bipolar devices, application of the stan-
.
dard boundary conditions on the electric field (E n = 0, with n denoting the outward nor-
mal vector) may yield solutions that are not physically plausible. The presence of a magnetic
field could result in a significant Hall field at the boundaries. Therefore, a procedure of the
kind introduced in [24] has to be adopted. The actual boundary condition on w at these boun-
daries is treated as unknown and to deal with the problem an “artificial” oxide region is in-
troduced, which completely encloses the device domain. In this way, one avoids imposing ar-
tificial boundary conditions at the device/oxide interface which could a priori affect the
results. The discontinuities in the normal component of electric field at the interface are
handled in the so-called weak formulation of the equations (see [53] for a description of the
equations in weak form). Poisson’s equation is solved over the entire domain (consisting of
the device plus oxide), with the normal component of electric field taken to be zero at the
oxide’s outer edges. The nature of this condition could, in principle, affect the solution in the
device’s active region. These effects, however, can be made minimal by an appropriate choice
of the oxide thickness. The solutions of the continuity equations are restricted to the device
domain. At the semiconductor/oxide interface, the zero normal current condition Jn . n =
Jp . n = 0 is imposed for ideal interfaces.

3.2.4 Mechanical Effects

Mechanical stress, induces changes in the energy band structure and a repopulation of
mobile carriers in silicon [54], the effects of which are reversible within the elastic limit. These
effects can be exploited in p-n junction-based devices where the variations in gap energy with
stress gives rise to a modulation of minority carrier injection or in nondegenerate extrinsic
samples where stress modulates the resistivity, which is the so-called piezoresistive effect.
These effects are described in what follows.

3.2.4.1 Modulation of Minority Carrier Injection

The variation of the energy gap, Eg with stress, (T affects the injected minority carrier con-
centration and hence the output characteristics of p-n junction diodes and transistors [55-591.
The advantage in using such effects in pressure sensing is the reduced power consumption,
making such devices potentially useful in biomedical applications. With a suitable bias con-
figuration, the linear variation of base emitter voltage in a bipolar transistor with applied
pressure has been employed for high resolution, temperature stable accelerometers for air-
borne and robotic applications [O].
Stress-induced changes in band structure have to be accounted for in the transport Equa-
tions (3-2), by suitably replacing the electrostatic potential with effective potentials that
describe the variations in band energy. The stress dependent minority carrier concentrations,
np,o,P , , , are
~ given by

(3-11)
54 3 Sensor Modeling

where n,, and p,,, are the minority carrier concentrations in p- and n-type material under
zero stress [57]. The change in gap energy, AEg can be assumed to be proportional to the
pressure. The constant of proportionality has been obtained theoretically for various stress
conditions and for uniaxial compression stress in the (100) and (111) directions; the value
is approximately eV/bar (571. Assuming that the majority carrier concentrations, n, p
remain independent of stress (ie, that the stress does not affect the ionization energy of im-
purities), the intrinsic concentration under stress, ni,u can be approximated as

In addition to these exponential stress-induced variations in the intrinsic concentration,


other parameters such as carrier mobility and lifetimes (and consequently the diffusion con-
stants and diffusion lengths) also vary with pressure, but in a linear fashion. The changes in
the minority carrier mobilities are expected to be of the same order of magnitude as those en-
countered in the majority carriers with the piezoresistive effect. For silicon, the relative change
in resistance is of the order of (10 - 4 bar - l ) 6,as a consequence of a corresponding relative
change in mobility [57]

-Ap/p - bar - I ) CJ . (3-13)

This indicates that the mobility may vary by an order of magnitude for stress levels of
lo5 bar. The minority carrier lifetimes, T,, T ~ change
, owing to stress-induced generation-
recombination centers. In the case of germanium, the density of dislocations induced is pro-
portional to the stress [57].

3.2.4.2 Piezoresistive Effect

In nondegenerate n-type Si, according to the many-valley model, stress induces a repopula-
tion of electrons in the valleys of the conduction band. This gives rise to variations in the
isotropic electron mobility as a result of the different electron effective masses. In addition,
the intervalley scattering (which determines the average momentum relaxation time) changes
with mechanical stress. In p-type Si, the dependence of resistivity on stress is believed to be
owing to a repopulation between heavy and light holes in the valence band. The energy sur-
faces at the valence band are intricate and calculation of the corresponding piezoresistive coef-
ficients is not straightforward.
The piezoresistive effect in diffused or ion-implanted layers in silicon micromechanical
structures, is currently being exploited in a variety of applications. With knowledge of the
stress components, the fractional change of resistivity in a piezoresistive layer can be
evaluated. Considering first order terms only, the relative change in resistivity can be expressed
as 121,

(AP/p)ij = 1
kl
Zijkl Okl (3-14)

where 7c is the tensor of piezoresistive coefficients which are functions of doping level and
temperature. Exploiting the symmetry conditions and the principal coordinate system of cubic
3.2 Model Equations for Electronic Phenomena 55

silicon, there are only three linearly independent components, and the relative change of
resistance can be expressed as

Here, o1and u, denote the average normal stresses longitudinal (parallel) and transverse
(perpendicular) to the current flow, respectively, and n1 and n, denote the corresponding
longitudinal and transverse piezoresistive coefficients. These coefficients are functions of the
fundamental coefficients n l l , nIz, and n-, and can be evaluated by coordinate transforma-
tion [27]. Equation (3-15) does not account for the effects associated with shear stresses and
described by cross terms. Depending on the resistor geometry and orientation, these effects
can be neglected.
Equation (3-15) is based on a linear expansion and hence does not include nonlinearities
arising from any higher order piezoresistive coefficients and stress terms. For a structure with
piezoresistors arranged parallel to the (110) axis, the relative resistance change under various
assumptions and in terms of modified coefficients [31, 611 can be approximated as

(3-16)

where eventual fourth order terms (i = 4) do not significantly contribute to A R .


Besides detecting induced stresses by the piezeorestistive effect, capacitive effects associated
with the deflecting plane and a suitable reference plane, can also be used.

3.2.4.3 Capacitive Effects

With knowledge of the diaphragm deflection, the total effective capacitance between the
diaphragm and the reference plate, for capacitive based structures is given by [27]

c = EO/S jj dx dy + cp,
1
(3-17)
w(x9 Y)
[I- s

where Cpis some parasitic capacitance, s is the zero pressure separation between diaphragm
and reference plate, E, is the dielectric constant in the cavity, and w (x, y) denotes diaphragm
deflection in the z-direction. The relative change of capacitance due to deflection
(C - Co)/C can then be evaluated with the zero pressure capacitance C, being E, L 2 / s and
t the diaphragm length for a square diaphragm.
The calculation of the various electrical parameters (piezoresistivity and capacitance)
described above, would require knowledge of the pertinent mechanical variables (stress and
deflection) in the micromechanical structure' under consideration. The procedures employed
in determining these mechanical variables are outlined below in Section 3.3.
56 3 Sensor Modeling

3.3 Model Equations for Mechanical Structures

The basic effects exploited by integrated silicon micromechanical structures are related to
the modulation of piezoresistivity or capacitance, or modulation of the structure’s mechanical
resonance frequency. The piezoresistive (AR) or capacitive (AC) effects are most commonly
utilized, although resonant micromechanical structures [62] are becoming increasingly attrac-
tive because of the high sensitivity of resonant frequency to some physical or chemical input
signals.
By virtue of the structural geometry of some micromechanical devices such as thin
diaphragms (thicknesses much smaller than other physical dimensions), analysis can be
greatly simplified by reducing the problem to two dimensions using the plane stress approx-
imation. Based on the assumptions employed in the classical thermoelastic thin-plate theory
(27, 301, the equations governing stress and deflection for a multilayered diaphragm in the x-y
plane under static conditions are

(3-18)

where F is the stress function and w denotes the deflection.


In Equations (3-18) and (3-19), E‘ and G denote the effective Young’s and shear moduli
(averaged over the thickness), respectively, NT and MT the respective thermal load and bend-
ing moment, q is the effective loading (which includes the applied pressure, the packaging and
thermal loads), v denotes Poisson’s ratio, and D,, D,,and D,are constants [27].The equa-
tions account for thermal stresses caused by the mismatch of thermal expansion coefficients
of the various material layers (provided that the thickness of the silicon layer is much larger
than the other layers) and the supporting substrate. Thermal stresses at the supporting plate
are treated as equivalent uniform surface traction forces along the edges of the diaphragm and
the forces are assumed to be proportional to the difference in expansion coefficients of the
silicon layer and substrate. Bending and stretching effects arising in the diaphragm have been
decoupled by assuming a constant Poisson’s ratio, v, over the thickness, and that the shear
modulus is proportional to Young’s modulus.
The resulting plane stress components ,,a ayy,and ao, are [27, 291:
3.4 Numerical Procedures 57

where Esi and Gsi are the Young's and shear moduli of silicon, respectively. The thermal
stress, a,, can be cast in the form

(3-21)

where a,, is the coefficient of expansion of silicon and T and To denote the processing and
ambient temperatures, respectively.
The boundary conditions for Equation (3-18) are as follows:

a 2 F / k 2= a2F/ay2 = N, and a2F/&ay = o (3-22)

where N, is the equivalent uniform surface traction force due to thermal stresses at the sup-
porting plate. The boundary conditions for Equation (3-19) are

w = o and g D , a 2 W / a x 2 + awl& =gD, a2w/ax2 + a w l & = -g [ M ~ / -v)]


(I
(3-23)

where g is the edge factor, which represents in effect the degree of elastic deformation of the
supporting rim and is believed to be a function of diaphragm thickness [27] and also depends
on the geometrical parameters of the supporting rim. It is zero for rotation free edges (rigidly
clamped) and infinity for simply supported edges allowing full rotation. Equations (3-18) and
(3-19) solved with boundary conditions (3-22) and (3-23) yield the stress components, ,a
oyy,and ax,,, in the diaphragm.
We recall that the equations outlined above are only valid if the geometrical parameters of
the structure considered satisfy certain well prescribed conditions. The more general case can
be adequately treated by three-dimensional finite element schemes, examples of which are
discussed in Section 3.5.3.

3.4 Numerical Procedures

The numerical solution to the systems of equations described in Sections 3.2 and 3.3 can
be obtained using a variety of discretization schemes. Qpical schemes employed for the
simulation of both electronic and mechanical transducer phenomena have been based on
finite-difference, finite-element, and finite-box procedures. An important and resource-con-
suming aspect of obtaining accurate numerical solutions lies in efficiently discretizing the
spatial simulation domain. The mesh generation procedure is described in the following
section.

3.4.1 Mesh Generation

The generated mesh should adequately resolve the variation of physical and material
parameters and any nonlinear behavior of the variables involved to avoid convergence pro-
blems. At the same time, it is necessary to minimize discretization error [63] by making all
58 3 Sensor Modeling

angles in an element as large as possible. Also, in view of speed and memory requirements,
the generation of an excessive number of nodes should be avoided, particularly considering
the relatively large physical dimensions that microsensors generally have.
To meet the above criteria, it is desirable to employ a hybrid mesh-generation procedure
which is based on a combination of the well-known regular grid triangulation and Delaunay
triangulation schemes. A detailed description of the Delaunay and hybrid schemes can be
found in [64, 651 and references cited therein. The above hybrid scheme employed for the
spatial discretization of a bipolar magnetotransistor is illustrated in Figure 3-1. Owing to the

Figure 3-1. Mesh generated for emitter base


junction of a magnetotransistor,
using a hybrid scheme [65].
3.4 Numerical Procedures 59

awkward structural nature of magnetotransistors (large junction aspect ratios), neither of the
above procedures on their own is efficient in optimizing the total node count. Regular grid
triangulation is economical on total node count but fails to efficiently resolve curved regions
of p-n junctions. Delaunay triangulation, on the other hand, provides a good resolution of
nonrectangular regions, but has the drawback of generating an excessive number of nodes
because of its inherent property to generate equilateral triangles. In view of these drawbacks,
a combination of both schemes is employed, where the latter is used in the triangulation of
curved regions (such as junction sidewalls) and the former everywhere else (see Figure 3-1).
An outline of the hybrid mesh generation procedure is as follows. As a first step, the general
simulation domain undergoes regional decomposition, leading to smaller regions of various
shapes and areas depending on the physical nature of material parameters and solution
variables. The next step is to triangulate each region according to the type of triangulation
scheme and the required number of elements. The regular grid triangulation scheme sub-
divides a rectangular decomposed region into right-angled triangles. In Delaunay triangula-
tion, the decomposed regions are subdivided into triangular elements that are as far as possi-
ble close to equilateral triangles (see Figure 3-2).

Figuiw 3-2. Delaunay triangulation for a micromechanical


bridge with apertures [65].
60 3 Sensor Modeling

For a reliable description of the underlying physical effects and in accurately predicting the
output characteristics, most microsensors, and in particular, micromechanical structures de-
mand a three-dimensional simulation. However, the development of versatile and efficient
grid generators in three dimensions is highly resource consuming, and thus such a develop-
ment is only in its infancy.

3.4.2 Discretization Procedures

Using a finite element procedure, solutions have been obtained for Poisson’s equation and
the time-dependent carrier continuity equations in the presence of optical radiation for solar
cell and photodiode analysis [5]. In Hall plate analysis, five-point [8-11, 13, 14, 191 or nine-
point [12] finite differencing in addition to finite-element [15- 181 techniques have been
employed. In the analysis of bipolar magnetic sensors, Poisson’s equation and the steady-state
continuity equations have been solved using finite-difference [20-231 and finite-box pro-
cedures [24-261. In mechanical transducer analysis, the use of both finite-element and finite-
difference schemes has been reported. Using a twelve-point differencing scheme [27, 291,
numerical solutions to the system of fourth-order partial differential equations that govern
the stress and deflection in a diaphragm have been presented.
For the class of problems related to electronic effects, the finite-element or finite-box
discretization procedures are desirable because of their ability to handle nonrectangular
meshes. A nonrectangular mesh handles complicated geometric shapes with ease. Although
the superiority of these procedures over the finite-difference scheme cannot generally be
claimed, these methods have several distinct advantages, such as excellent current conserva-
tion, efficient node distribution to handle rapidly varying physical quantities, and the ease
with which higher order approximations can be constructed [66]. The most significant advan-
tage, however, lies in its ability to handle localized regions of high node density with ease. In
certain geometries, the Dirichlet boundary regions (electrodes) occupy only a small portion
of the entire simulation domain [15, 161 (eg, the vertical Hall device [67] employed in magnetic
field sensing). Without an adequate number of nodes at these regions, the resulting stiffness
matrix is almost singular. In such a case, direct methods can produce significant roundoff
error and iterative schemes converge very slowly or not at all. Hence, by allocating a
reasonable portion of the total node count on Dirichlet regions, the condition of the stiffness
matrix can be significantly improved (see [16] and references cited therein). Further, without
enough nodes at electrode vicinities, singularities occur in the solution where there are discon-
tinuities in the prescribed conditions (Dirichlet and Neumann conditions on the same line) [8,
121. This may lead to error propagation and possibly reduce the overall solution accuracy in
the bulk of the device [12, 661. In addition, the boundary conditions and any material in-
homogeneities in the device and inhomogeneities in the external input signal (eg, locally in-
verted and discontinuous magnetic induction - magnetic domains), can be handled naturally
in the weak formulation of the equations.
In the following, we illustrate one easy and reliable discretization scheme that can be
employed for the simulation of electronic phenomena. The finite-box discretization procedure
is also known as the control-region approximation [68] or more commonly the box-integration
method [69] when applied to the divergence Equations (3-1) transforms them by means of the
Gauss divergence theorem into a system of discrete algebraic equations expressed as a function
3.4 Numerical Procedures 61

of the respective unknowns. For a subdomain (see Figure 3-3), this yields a system of algebraic
equations found by solving for node i the equation:

(3-24)

where F denotes a vector field (eg, electric field, current densities, or temperature gradient)
and f denotes a scalar field (eg, space charge, generation, recombination, or heat), Qiis the
cell (or subdomain) formed by the union of perpendicular bisectors of element edges that con-
verge to node i (Figure 3-3), M idenotes the boundary of Qi, and n the outward normal vec-
tor to Mi. The choice of perpendicular bisectors of element edges keeps calculations of flux
emanating from node i relatively simple. The electric potential or temperature is assumed to
be linear in an element. Electric fields or temperature gradients are then piecewise constant
in the domain. By assuming a piecewise constant material permittivity E or thermal conduc-
tivity y the discretized form of the potential or heat flow equation can be easily evaluated [45,
701. The terms on the right-hand side can be assumed to be spatially constant in the sub-
domain, although higher order approximations can also be employed.

Figure 3-3. The cell or subdomain Ri (bounded by ani),


formed by the union of perpendicular bisectors of
elemental edges converging to node i 170).

The discretization of the carrier continuity equations is based on the two-dimensional ex-
tension of the Scharfetter-Gummel approach [71]. In magnetic field sensors, owing to the
asymmetries introduced by the field, variations in the standard procedure have to be employed
in determining the projected current density along the edge of an element under consideration
[70]. By integrating the right-hand side of Equations (3-10) over the pair of elemental sides
that converge to the node in question, Jn,B in the element can be constructed. By assuming
JnVB is constant in the element, and with an arithmetic average of the Hall mobility, pz, the
flux or the current emanating from node i is evaluated in the usual manner. The generation-
recombination terms on the right-hand side are treated as in usual device modeling. The
resulting stiffness matrix for the continuity equations in system (3-1) is positive definite non-
*
symmetric, and this is due to the magnetic field (pu,*B 0, ,u:B * 0) which explicitly shows
up in the current density relations.
In the analysis of the mechanical behavior of microstructures, finite-element procedures are
suited for the discretization of the fourth-order partial differential equations that govern stress
and deflection in the structure. These procedures can be conveniently applied to nonrec-
tangular meshes and are therefore potentially useful for the simulation of realistic
micromechanical structures. This is particularly true for design geometries that have apertures
(holes) distributed on the mechanical structure, a feature that is desired for a variety of reasons
62 3 Sensor Modeling

[62]. Further, the boundary conditions and material inhomogeneities (typically encountered
in multilayered microstructures) can be handled naturally in the weak formulation of the
equations.

3.4.3 Solution Procedures

The discretized equations can be solved using either successive [24-261 or simultaneous [5,
20-23, 27-29] procedures. The successive (or decoupled) scheme is relatively easy to imple-
ment. The main disadvantage lies in the slow overall convergence of the variables involved
when the mutual coupling between them is strong. Such convergence problems may be over-
come by employing a simultaneous (or coupled) scheme, although the price to pay is code
complexity and large storage requirements. We illustrate an example of the successive ap-
proach which has been extensively employed in bipolar magnetic field sensor simulations
[70, 721.
The procedure is initiated by solving the nonlinear Poisson’s equation for the electric poten-
tial, with the Fermi potentials kept constant. The speed of convergence can be improved by
employing acceleration algorithms [73]. When the nonlinear loop has been satisfactorily ter-
minated, the concentrations n and p are updated with the new value of the potential Y,I which
in turn is kept constant during the subsequent solution of the continuity equations. The pro-
cess is repeated until the variables have achieved self-consistent values. To obtain the solutions
of the variables for a desired device operating point, the applied bias is incremented in steps
starting from the zero bias (or equilibrium) solution.
The resulting matrix equations can be solved by a family of direct methods (see [27, 701)
or by relaxation methods (eg, Successive Line Over Relaxation [13,23], Successive Over Relax-
ation [8-11, 16, 181). Direct methods have been found to give very good results but the storage
requirements encountered in the simulation of sensors requiring large node density (par-
ticularly in 3-D simulations) indicate that an iterative scheme such as the generalized con-
jugate gradient method would be preferable. The convergence criteria may be chosen depend-
ing on the problem at hand [72].
Before embarking on a simulation with either a newly acquired package or newly developed
code, it is essential to validate the software. The best test vehicles that can be simulated for
program verification are simple structures whose electrical or mechanical response is analyti-
cally well known. The ultimate validity of the simulations lies in how well the numerical and
experimental results corroborate [18, 26, 271.

3.5 Examples

3.5.1 Photodiode

Selected numerical results for photodiodes using the one-dimensional PC-ID simulator are
illustrated in Figure 3-4 [47]. The electric field and carrier concentrations at equilibrium are
illustrated for the Hamamatsu 1337 photodiode using the default parameters in PC-1D [S].
3.5 Examples 63

10’0 0 I I I I

-
1015 -

-
i
w
0
z -
0
0
-
4 I I I
-14
’- 0 1 2 3 4 5
DEPTH ( p m )
Figure 3-4. The electric field and carrier concentrations at equilibrium as a function of depth. The
nominal front region parameters used are characteristic of the Hamamatsu 1337 photodiode
(see [47]).

The results were obtained using a convergence criterion of 10 - 7 and a total node count of 78,
of which 11 are placed within 0.1 pm of the diode’s front surface where nonlinearity in the
charge density is the highest. The limitations of PC-lD, particularly in simulating the short-
circuit photocurrent in high-accuracy photodiodes (such as those used for self-calibration), are
discussed in [47]. Current efforts in silicon photodiode modeling aim at high-accuracy predic-
tion of the spectral quantum efficiency for the purpose of absolute calibration in radiometry.

3.5.2 Magnetic Sensors

Numerical results of magnetic field sensor simulations using ALBERTINA, are illustrated for
both unipolar and bipolar devices in Figures 3-5 to 3-13. Distributions of carrier transport
were computed for a variety of Hall devices under various configurations of locally inverted
and discontinuous magnetic induction (Figure 3-5). The configurations include the one-
dimensional locally inverted magnetic induction (referred to as longitudinal or transverse strip
domains depending on their orientation) and the two-dimensional locally inverted magnetic
induction of circular geometry (referred to as circular domains). In the case of the Hall cross
with a longitudinal strip domain, strips of Hall fields are observed with current flowing paral-
lel to the inversion boundaries, indicating that the distributions effectively resemble the “Hall
effect” analytical model. In the other limiting case of the transverse strip domain, there are
no Hall fields but the current lines are skewed by the local Hall angle. In this case, the distribu-
tions obey the “carrier deflection” intuitive model. In the general case where the magnetic
field is inhomogeneous in both directions, a mixture of both Hall effect and carrier deflection
are involved in a complex way on both sides of the inversion boundary. The analysis of such
configurations in terms of simple analytical models could become very complicated, if not im-
possible. The experimental and numerical output responses illustrated in Figure 3-6, for both
bubble and strip domain configurations, appear to be in good agreement.
Results of simulations of bipolar structures are illustrated in Figure 3-7 to 3-13. The simula-
tion domain is shown in Figure 3-7 and Figure 3-8 illustrates the actual device structure which
64 3 Sensor Modeling

1,mV 1( mV l1mV

Figure 3-5. Equipotential and current lines for


a Hall cross with longitudinal strip
domain, a split-electrode Hall
device with transverse strip do-
main, and a conventional Hall
device with circular domain [18].

was fabricated using bipolar IC technology. The oxide that surrounds the base region (Fig-
ure 3-7) is part of the model only and has been introduced to deal with the problem of boun-
dary conditions [24] discussed Section 3.2.3. To illustrate the effects of magnetic field in the
vicinity of the junction, equipotential lines for the region around the emitter of the device are
shown in Figure 3-9. The device is operated at V, = 0.85 V with a magnetic field of 2 tesla
parallel to the chip surface. The current through the device is 0.6 mA. The equipotential lines
appear to be symmetric around the emitter-base junction, indicating that there are no signifi-
cant Hall fields in that vicinity. The distribution of flow lines for minority carriers (electrons)
is illustrated in Figure 3-10. In contrast to the distribution of potential, the effect of the
magnetic field clearly manifests itself in the distribution of current density. The numerical
results discussed yield Hall voltages of the order of microvolts (see Figure 3-11) at the emitter-
base junction vicinity. This is too small in magnitude to substantiate the validity of the in-
tuitive emitter injection modulation analytical model [74]. These predictions gathered from
3.5 Examples

a
65

3.2mA.

MAGNETIC FIELD DISTRIBUTION

-200 -100 0 100


r (pn200
)
-200 -100 0 100 200
BUBBLE DISPLACEMENT (microns)

-150 -mo -50 o 50 100 150 -20 -10 o 10 20


RELATIVE STRIP DISPLACEMENT (microns)
r (pw)
Figure 3-6. Comparison of experimental and numerical results of output response for the locally inverted
circular and strip domain configurations. The inverted field distribution used in simulations
is denoted by curve a and the corresponding measured distribution is denoted by curve b [18].

c W'
I
Figure 3-7. Wo-dimensional simulation
geometry of the magneto-
transistor's base region sur-
rounded by an artificial 4
I ' I
oxide region [26]. The junc-
tion depth, xj, is 2.5 pm.

? ! V"
B C2

Figure 3-8. Cross-sectional view of


dual-collector magnetotran-
sistor fabricated in bipolar
technology [26] with
xi = 2.5 wm.
66 3 Sensor Modeling

I 1

Figure 3-9. Equipotential lines in the emitter vicinity for VBE = 0.85 V and B = 2T [26].

Figure 3-10. Minority carrier (electron) flow lines in the vicinity of the emitter for the same operating con-
ditions as in Figure 3-9 [26].

- 700-
5w 600i +
+
++++++
+
+
+
o V = 0.7

V = 0.75

5
0 500-
+ + v = 0.8
400' ++ +
+

9 300-
1 +
+ +
Figure 3 4 . The Hall voltage distribution as a
function of position, W: along
y = 0 for B = 500 mT and dif-
ferent injection levels [26]. V
denotes the base-emitter voltage
(in volts) and the junction is
0 30 60 90 120 150 located at the vicinity of
WPm) W" = 25 pm.

T
4Pm FIELD OXIDE
I

Figure 342. Simulation geometry (and top


view) of dual-collector NPN
magnetotransistor [24].
3.5 Examples 67

simulations are supported by experimental data obtained from in situ Hall probe measure-
ments (261. Simulation results for a CMOS NPN dual-collector magnetotransistor (Figure
3-12) are illustrated in Figure 3-13. The magnetic field perpendicular to the device surface
is 2 T and V,, = V,, = 0.9 V. There is again symmetry in the base-emitter voltage, but the
minority carrier trajectory is asymmetric, giving rise to collector current imbalance.
A systematic comparison of simulations performed for various device structures leads to the
conclusion that the dominant mechanism in magnetotransistors with linear response is carrier
deflection, while possible nonlinearities can be attributed to magnetoconcentration effects.

Figure 343. (a) Equipotential lines. Current


density flow lines of (b) holes
injected from base and (c) elec-
trons injected from emitter [24].

3.5.3 Mechanical Sensors

In special cases, the deflection for a rectangular diaphragm of uniform thickness can be
calculated analytically using superposition of Fourier series solutions [27, 301. The diaphragm
deflection and corresponding stress distribution for a single layered structure (100 pm x
100 pm x 10 pm) are shown in Figures 3-14 and 3-15, respectively, for an applied pressure
of 100 mmHg. The results shown are valid only when the diaphragm deflection is small. Also,
the calculations have been performed for an ideal built-in edge condition (fully clamped, edge
factor g = 0). The solutions depend strongly on the chosen value of g, which represents in
effect the degree of elastic deformation of the supporting rim. The superposition approxima-
tion employed can be extended to Si/SiO, structures by considering the thermal moment of
the additional thin layer [27]. When diaphragm deflections are large, there is nonlinearity in
the resulting stress distribution (due to the “ballon effect” [2]), contributing to nonlinearity
in the output response. In such cases, approximate solutions based on the Ritz strain-energy
method have been found [75). For a fully clamped square diaphragm, the deflection at the
diaphragm centre as a function of applied pressure, P, is shown in Figure 3-16. For
comparison, results based on small deflection theory (linear analysis) are also shown. Here,
a denotes the diaphragm half length (2 a = L ) and D is the bending rigidity, viz., Eh3/
[12(1 - v’)]. Using both small and large deflection models, the normalized bending and
membrane stress distributions induced at the middle of the diaphragm edge are illustrated in
Figure 3-17 for tension and compression conditions. The membrane stresses remain the same,
independent of tension or compression, but bending stresses are opposite in sign.
To accurately predict diaphragm deflection in practical structures fabricated using
anisotropic etching techniques, it is crucial to determine the edge factor. By fitting simulated
68 3 Sensor Modeling

results with measurement data for diaphragm deflection, the edge factor can be suitably ex-
tracted. For diaphragms formed on (100) N-type substrate with rim slopes oriented in the
(111) direction, g was evaluated numerically to be of the order of providing good
agreement with optical measurement data (Figure 3-18). The dashed curve in Figure 3-18 cor-
responds to the fully clamped condition (g = 0). Using an edge factor of simulation
and measurement data for deflection as a function of position correlate well for various ap-
plied pressures (Figure 3-19). Comparison of simulations and measurements of the output
response of a capacitive pressure sensor (2.1 x 2.1 mm2) are illustrated in Figure 3-20. The
diaphragm thickness is 27 pm and the corresponding edge factor is 3.6 x The device
has been electrostatically sealed with the reference cavity depth being 3.5 pm at 760 mmHg
absolute pressure. The simulations indicate that the dependence of the output response on
temperature appears to be dominated by the effects of trapped gas in the reference cavity.
Thermally induced stresses due to packaging were found to be less than 100 ppm/"C. The
results have been confirmed by experiments.
Simulations of temperature effects in piezoresistive pressure sensors are shown in Figure
3-21 for a 1 mm2 square diaphragm with a 1 pm oxide layer on 10 pm Si. The simulations
are performed for matched diffused piezoresistors in bridge configuration. The temperature

22
~ ( p =
) W' (-1L4 P x10-l' cm
h3

L = diaphragm l e n g t h ( c m )
h = diaphragm t h i c k n e s s ( c m )
P = a p p l i e d pressure (mmHg)

Figure 3-14. Deflection of a fully clamped diaphragm as a function of position. The result is based on
analytical computations using the small deflection theory [27].
3.5 Examples 69

- ox,= SI(,) L 2 P (dyne/crn')


--- r y y =S ' ( L 1 h ) 2 P
0
W L =diaphragm l e n g t h (cm)
h = d i a p h r a g m t h i c k n e s s (cm)
P = a p p l i e d p r e s s u r e (mmHg)

uyy ( Y I L r O . 5 )

'0.0 0.1 0.2 0.3 0.4 0.5


XIL

Figure 3-15. Normalized stress distribution for the structure in Figure 3-14 [27].

1.6

c 1.2
x

Figure 3-16.Normalized deflection at the center of a fully


clamped square diaphragm as a function of ap- ............ small dsf. th.
lorgo dot. th.
plied pressure, computed using small and large
deflection theories [75]. Young's modulus, 0 40 80 120 160 200
E = 1.7 x 10l2 dyn/cm2 and Poisson's ratio, Pa4/ D h
v = 0.066. u.0.066

sensitivity of the offset voltage at zero pressure, for a diaphragm with a convex thickness pro-
file turns out to be less than 0.1 mmHg/"C over the range of operating temperatures shown
(curve a). The temperature coefficient (TC) of the pressure sensitivity for the same structure
is 175 ppm/"C (curve b), while for the diaphragm with uniform thickness, the TC is 148
ppm/"C (curve c). When mounted on a package, the TC of pressure sensitivity of the structure
with uniform thickness is reduced to less than 14 ppm/"C (curve d) and with an increase in
I0 3 Sensor Modeling

15

10

" 5
D
\

--
Y O
I

\ -5
a
bE
-10
FFgure 3-17. Distribution of maximum stress induced in the
-15 diaphragm (same structure as in Figure 3-16)
0 40 00 120160200
Pa* / D h under the conditions of tension and compression
V.0.066 V51.

SAnPLE A 0 EXPERIIlENT
L -0.87 mm -SLOPED EDGE
h - 9.8pm ---- BUILT-IN EDGE

150.00 450.00 600.00 10.00


FIPPLIE?%ESSURE (mm Hg)
Figure 3-18. Numerically simulated center deflection as a function of applied pressure for a diaphragm
with fully clamped (g = 0) and rim supported (g - edge conditions [27]. The
diaphragm thickness is approximately 10 pm. The experimental results are based on optical
measurement data.

the pressure sensitivity. This is due to the compressive and tensile nature of thermal stresses
induced by the package and oxide, respectively, resulting in stress cancellation. The results
shown so far (curves a to d) are based on a zero TC of diffused resistors and piezoresistive
coefficients. Simulations performed for a TC of diffused resistors of 2000 ppm/"C, TC of
5 . ~fixampies

-
?
Q P-478.5 m m Hg
A P - 99.5 mmHg
8 - SrnULATION
0

a L -0.870 rnm
$ Wo(P-478.5mrnHg)= 4.27prn
==
-
$0

f
\
as
-%-
$0

Figure 3-19. Diaphragm


deflection (for 3
the structure 0

shown in 3
Figure 3-18) as
a function of
position for
different ap- ?
plied pressures 'b.00 0.10 0.20 0.30 0.40 I

~71. XIL

--
CRP. SENSOR
C(TOTRL)-13.6 PF

--
L Zlmm
h 27 pm
g 3.5 x 10-6
T 27'C

Q D(PERII"1
- SIMJJLFITION

Figure 3-20. Numerical


and ex-
perimental

d
results of cy
d
relative
capacitance
change for a
capacitive 3.00
? 100.00 200.00 300.00 400.00 I
pressure sen- 10.00
sor 1271. APPLIED PRESSURE (rnrnHg1
72 3 Sensor Modeling

piezoresistive coefficients of - 2500 ppm/ “C indicate that the major contribution to the TC
of pressure sensitivity is from the temperature dependent piezoresistive coefficients (see
Fig. 3-21, curve e). The effects of thermal stresses on the TC of pressure sensitivity were found
to be negligible for the structure considered.

0
0

b
d
C

i
a
dd
-30.00 am 30.00 60.00 90.00 .M
TEIIPERRTURE [ C 1
Figure 3-21. Temperature dependence of output response for piezoresistive pressure sensors [27].

Figure 3-22 illustrates the distribution of stress along the center line of the diaphragm as
a function of the distance from the diaphragm center. The distributions shown are based on
a three-dimensional finite-element scheme using “transient thick shell” elements which does
not employ an apriori assumed edge factor [31]. For the structure with the rim support, the
magnitudes of stress are small at distances larger than about 150 pm from the diaphragm
edge. Hence, signal processing circuitry can be placed at such distances to minimize perfor-
mance degradation.
In the design of a silicon cantilever accelerometer (singly fiied beam), the general-purpose
finite-element code ANSYS has been extensively used to predict stress patterns on the beam,
the sensitivity, the fundamental and higher order resonant modes, temperature coefficient,
and overrange characteristics [35]. Knowledge of the stress distribution on the beam aids
resistor placement. Modal analysis aids in determining a priori whether the higher order
modes would present serious problems for a given measurement bandwidth. Results of the
analysis are illustrated in Figure 3-23, where the fundamental bending mode is at a resonant
frequency of 839 Hz and a second (torsional) resonant mode at 68575 Hz.
3.5 Examples 73

X-DISTANCE [mm]

Figure 3-22. Simulated stress distribu-


tion along the axis of a -20 0.2 0.4 0.6 0.8 1.0
diaphragm for ideal and X-DISTANCE [mm]
supported rim edge con-
ditions. The abscissa -0- WITH BUILT-IN EDGES
denotes the distance ..... WITH RIM-SUPPORTED EDGES
from diaphragm center DIAPHRAGM 1mmX l m m X 3 0 v m
[311. RIM 2mmX2mmX350pm

Figure 3-23. The fundamental (ben-


ding) mode at 839 Hz and
the second (torsional)
mode at 68575 Hz for a
silicon cantilever
accelerometer [35].
74 3 Sensor Modeling

3.6 Numerical Modeling as a Microsensor CAD Tool

Numerical modeling serves as a useful design tool for optimising the microsensor’s perfor-
mance with respect to device structure and geometry for given fabrication and operating con-
ditions (see e.g. [76]), and hence reduces the cost involved in the trial and error steps of ex-
perimental device development.
Depending on the operating mode, the response of the sensor can be in the form of a dif-
ferential voltage (A V ) , current (AT), resistance ( A R ) , capacitance (AC), or frequency (Af). In
the case of magnetic sensors, the output response, AV or AI, is generally proportional to the
device current, the Hall mobility, the magnetic field, and some appropriate geometry factor.
For Hall plates under uniform field, the geometric factor has been widely investigated using
both numerical and analytical techniques and very good agreement with experimental data
has been achieved [7]. In the presence of inhomogeneous fields (such as in locally inverted
circular domains), the geometric factor depends on several additional parameters such as in-
version size and position and inverted field strength. Using numerical techniques, an expres-
sion for the geometric factor and hence the output response has been developed for such con-
figurations [77]. In magnetotransistors, for relatively simple structures, the geometric factor
has been extracted by fitting measured magnetic response with intuitive analytical models
[7, 781. For more complicated structures (eg, [79]) numerical modeling becomes essential. This
characterization approach expedites the CAD of microsensors integrated with signal-process-
ing circuitry.
Finally, for completeness, two examples are presented where standard circuit simulation
tools have been employed in predicting the output characteristics of a magnetic-field-sensitive
element [80]and a ring oscillator whbse output frequency is modulated by applied pressure
[81]. In the first example, a channel of a MOSFET Hall sensor is partitioned into cells, with
each cell containing the input magnetic signal represented in terms of variable current or
voltage sources configured in a suitable equivalent circuit form, and the boundary conditions
judiciously included in the circuit. The resulting network of circuits was then simulated using
SPICE. In the latter example, the dependence of the oscillator output frequency on pressure
was simulated via the stress dependent mobility of the MOS transistors in the ring. The
simulations were performed using SPICE.

Acknowledgements

It is a pleasure to thank Prof. Walter Allegretto, University of Alberta, Canada, for his ef-
forts and contributions to the magnetic sensor simulation project, and to express our ap-
preciation to Mr. Kris Chau, University of Alberta for assistance with the grid generation. We
thank Jon Geist, National Bureau of Standards, Washington DC, USA for providing
numerical results on photodiodes, Dr. K. W. Lee and Dr. K. Wise, University of Michigan,
An Arbor, USA for allowing us to reproduce Figures 3-14, 15, 18-21, Mr. K. Suzuki, Mr. T.
Ishihara, Mr. M. Hirata, and Mr. H. Tanigawa, NEC Corporation, Kanagawa, Japan for
Figures 3-16, 17, 22, and to Dr. F. Pourahmadi and Dr. P. Barth, NovaSensor, Fremont, CA,
USA for providing Figure 3-23.
3.7 References 75

3.7 References

[I] Baltes, H. P., Allegretto, W., Nathan, A., in: Simulation of Semiconductor Devices and Processes
Vol. 3, Baccarani, G., Rudan, M. (eds.); Bologna: Tecnoprint, 1988, pp. 563-577.
[2] Poppinger, M., in: Studies in Electrical and Electronic Engineering Vol. 23, Balk, P., Folberth, 0. G.
(eds.); Amsterdam: Elsevier, 1986, pp. 53-70.
13) Briglio, D. R:, Nathan, A., Baltes, H. P., Can. 1 Phys. 65 (1987) pp. 842-845.
[4] Briglio, D. R., Characterization of CMOS Magnetic Field Sensors, MSc Thesis, 1988, University of
Alberta, Edmonton, Canada.
[5] User’s Guide, PC-ID, 1985, Iowa State University Research Foundation (ISURF), Ames, IA.
(61 Gray, J. L., Two-DimensionalModeling of Silicon Solar Cells, PhD Dissertation, 1982, Purdue
University, West Lafayette, USA.
171 Baltes, H. P., Popovic, R. S., Proc. ZEEE 74 (1986) 1107-1132.
[8] Newsome, J. P., Proc. IEE lI0 (1963) 653-659.
[9] Mimizuka, T., Solid-State Electron. 14 (1971) 107- 110.
[lo] Mimizuka, T., Solid-state Electron. 21 (1978) 1195- 1197.
[11] Chwang, R., Smith, B. J., Crowell, C. R., Solid-state Electron. 17 (1974) 1217-1227.
[12] Grutzmann, S., Solid-State Electron. 9 (1966) 409-416.
(131 Nathan, A., Huiser, A. M. J., Baltes, H. P., ZEEE Trans. Electron Devices ED-32 (1985) 1212-1219.
1141 Nathan, A., Andor, L., Baltes, H. P., Schmidt-Weinmar, H. G., ZEEE 1 Solid-state Circuits SC-20
(1985) 819-821.
[IS] Allegretto, W., Mun, Y. S., Nathan, A., Bakes, H. P., Proc. NASECODEZVConJ (1985) 129-133.
[16] Mun, Y., Numerical Modeling of CMOS Magnetic Field Sensors by Finite Element Method, MSc
Thesis, 1986, University of Alberta, Edmonton, Canada.
1171 Nathan, A., Allegretto, W., Bakes, H. P., Sugiyama, Y., ZEEE Electron Device Lett. EDLS (1987)
1-3.
(181 Nathan, A., Allegretto, W., Baltes, H. P., Sugiyama, Y.,IEEE Trans. Electron Devices ED-34 (1987)
2077 -2085.
(191 Baird, A. W., ZEEE Trans. Magn. MAG-15 (1979) 1138-1141.
[20] Andor, L., Baltes, H. P., Nathan, A., Schmidt-Weinmar, H. G., ZEDM Tech. Dig. (1983) 635-638.
1211 Bakes, H. P., Andor, L., Nathan, A., Schmidt-Weinmar,H. G., ZEEE Trans. Electron Devices ED-31
(1984) 996-999.
[22] Schmidt-Weinmar,H. G., Andor, L., Baltes, H. P., Nathan, A., ZEEE Trans. Magn. MAG-20 (1984)
975-978.
(231 Andor, L., Bakes, H. P., Nathan, A., Schmidt-Weinmar,H. G., ZEEE Trans. Electron Devices ED-32
(1985) 1224-1230.
[24] Allegretto, W., Nathan, A., Baltes, H. P., Proc. NASECODE V Con$ (1987) 87-92.
1251 Nathan, A., Allegretto, W., Baltes, H. P., Sensors Muter. 1 (1988) 1-6.
1261 Nathan, A., Maenaka, K., Allegretto, W., Baltes, H. P., Nakamura, T., IEEE Trans. Electron
Devices ED-36 (1989) 108- 117.
[27] Lee, K. W., Modeling and Simulation of Solid State Pressure Sensors, PhD Dissertation, 1981,
University of Michigan, Ann Arbor, USA.
1281 Lee, K. W., Wise, K. D., IEDM Tech. Dig. (1981) 471-474.
1291 Lee, K. W., Wise, K. D., IEEE Trans. Electron Devices ED-29 (1982) 24-31.
1301 Timoshenko, S., Woinowsky-Kreiger, S., Theory of Plates and Shelk, New York: McGraw-Hill, 1959.
1311 Suzuki, K., Ishihara, T., Hirata, M., Tanigawa, H., ZEEE Trans. Electron Devices ED-34 (1987)
1360-1367.
1321 Suzuki, S., Yamada, K., Nishihara, M., Hachino, H., Minorikawa, S., in: Proceedings of the 1st Sen-
sor Symposium, IEE Japan, 1981, pp. 131-133.
76 3 Sensor Modeling

[33] Suzuki, S., Yagi, Y., in: Proceedings of the 2nd Sensor Symposium,IEE Japan, 1982, pp. 163-165.
[34] Bessho, M., Tsuru, Y., Horiike, H., Jinmon, M., Yamagami, K., Wataya, S., SAE Spec. Publ.,
NO. 536 (1983) 55-59.
[35] Barth, P. W., Pourahmadi, F., Mayer, R., Poydock, J., Peterson, K., in: Technical Digest, IEEE
Solid-state Sensor and Actuator Workshop, 1988, pp. 35-38.
[36] Maseeh, F., Schmidt, M. A., Allen, M. G., Senturia, S. D., in: Technical Digest, IEEE Solid-state
Sensor and Actuator Workshop, 1988, pp. 84-87.
[37] User’s Manual, ANSYSEngineering Analysis System, Houston: Swanson Analysis Systems Inc.
[38] Gorman, D. J., Free Vibration Analysis of Rectangular Plates, Amsterdam: Elsevier, 1982.
[39] Lerch, R., Proc. IEEE Ultrasonics Symp. (1988) 643-654.
[40]Lerch, R., Siemens Forsch.- u. Entwickl.-BeE 17 (1988) 283-290.
[41] Selberherr, S., Analysis and Simulation of Semiconductor Devices; Vienna: Springer, 1984.
[42] Baccarani, G., Rudan, M., Guerrieri, R., Ciampolini, P., in: Process and Device Modeling,
Engl, W. L. (ed.); Amsterdam: North-Holland, 1986, pp. 107-158.
[43] Meijer, G. C. M., Sens. Actuators 10 (1986) 103-125.
[44] Van Herwaarden, A. W., Sarro, P. M., Sens. Actuators 10 (1986) 321-346.
[45] Nathan, A., Allegretto, W., Chau, K., Baltes. H. P., in: Simulation of SemiconductorDevices and
Processes, Vol. 3, Baccarani G., Rudan, M. (eds.); Bologna: Tecnoprint, 1988, pp. 589-598.
[46] Ghione, G., Golzio, P., Naldi, C., Proc. NASECODE V Con$ (1987) 195-200.
[47] Geist, J., Baltes, H. P., “High accuracy modeling of photodiode quantum efficiency”, Appl. Opt.
28 (1989) in press.
[48] Heavens, 0. S., Optical Properties of Thin Solid Films, New York: Dover, 1965.
[49] Lundstrom, M. S., Schwartz, R. J., Gray, J. L., Solid-state Electron. 24 (1981) 195-202.
[50] Madelung, O., Introduction to Solid State Theory, Berlin: Springer, 1978.
(511 Zieren, V., Integrated Silicon Multicollector Magnetotransistors, PhD Dissertation, 1983, Delft
University of Technology, Delft, The Netherlands.
[52] Beer, A. C., Galvanomagnetic Effects in Semiconductors, New York: Academic Press, 1963.
[53] Zienkiewicz, 0. C., Taylor, R. L., The Finite Element Method, 4th ed., Vol. 1, London: MacGraw
Hill, 1987.
[54] Keyes, R. W., in: Solid State Physics, Vol. 11, Seitz, F., lbrnbull, D. (eds.); New York: Academic
Press, 1960, pp. 149-221.
[55] Ridner, W., Braun, I., J. Appl. Phys. 34 (1963) 1958-1970.
[56] Hall, H. H., Bardeen, J., Pearson, G. L., Phys. Rev. 84 (1951) 129-132.
[57] Wortman, J. J., Hauser, J. R., Burger, R. M.. J. Appl. Phys. 35 (1964) 2122-2131.
[58] Wortman, J. J., Monteith, L., IEEE l’jans. Electron Devices ED-16 (1969) 855-860.
[59] Sansen, W., Vandeloo, P., Puers, B., Sens. Actuators 3 (1982) 343-354.
[60]Puers, B., Reynaert, L., Snoeys, W., Sansen, W., IEEE l’jans. Electron Devices ED-35 (1988)
764-770.
[61] Yamada, K . , Nishihara, M., Shimada, S., Tanabe, M., Shimazoe, M., Matsuoka, Y., IEEE Trans.
Electron. Devices ED-29 (1982) 71-77.
[62] Howe, R. T., Digest of Technical Papers, 4th International Conference on Solid-state Sensors and
Actuators, 1987, pp. 843-848.
[63] Babuska, I., Aziz, A. K., SIAM (Soc. Ind. Appl. Math.) . Numerical Anal. 13 (1976) 214-216.
l
[64] Joe, B., SIAM J. Sci. Stat. Comput. 7 (1986) 514-539.
[65] Chau, K. Y. K., Finite Element Mesh Generation for Semiconductor Device Simulation , MSc
Thesis, 1988, University of Alberta, Edmonton, Canada.
[66] Barnes, J. J., Lomax, R. J., IEEE Trans. Electron Devices ED-24 (1977) 1082-1089.
[67] Popovic, R. S., IEEE Electron Device Lett. EDL-5 (1984) 357-358.
[68] McCartin, B. J., in : Problems and New Solutionsfor Device and Process Modelling, Miller, J. J. H.
(ed.); Dublin: Boole Press, 1985, pp. 72-82.
[69] Rudan, M., Guerrieri, R., Ciampolini, P., Baccarani G., in: Problems and New Solutionsfor Device
and Process Modelling, Miller, J. J. H. (ed.); Dublin: Boole Press, 1985, pp. 110-121.
[70] Allegretto, W., Nathan, A., Baltes, H., “Numerical analysis of magnetic-field-sensitive bipolar
devices”, IEEE Pans. Computer Aided Design of Integrated Circuits and Systems, submitted.
3.7 References 77

[71] Scharfetter, D. L., Gummel, H. K., IEEE nuns. Electron Devices ED-16 (1969) 64-77.
[72] Allegretto, W., Nathan, A., Baltes, H., in: Proceedings of 1st International Forum on ASIC and
Tmnsducer Technology, Honolulu, February 7-10, 1988, pp. 89-95.
[73] Bank, R. E., Rose, D. J., SIAM (SOC.Ind. Appl. Math.) .l Numerical Anal. 17 (1980) 806-822.
[74] Vinal, A. W., Masnari, N. A., IEEE %ns. Electron Devices ED-31 (1984) 1486-1494.
[75] Suzuki, K., Ishihara, T., 'hnigawa, H., Proceedings of the 5th Sensor Symposium(1985) 159-163.
[76] Puers, B., Peeters, E., Sansen, W., Sens. Actuators 17 (1989) 423-429.
[77] Nathan, A., Allegretto, W., Baltes, H. P., Sugiyama, Y., Tacano, M., lhnsducers '87, Digest of
Technical Papers, 1987, pp. 536-537.
[78] Zieren, V., Duyndam, B. P. M., IEEE Tmns. Electron Devices E B B (1982) 83-90.
[79] Ristic, L., Baltes, H. P., Smy, T., Filanovsky, I., IEEE Electron DeviceLett. EDL-8 (1987) 395-397.
[SO] Popovic, R. S., Solid-State Electron. 28 (1985) 711-716.
[81] Schorner, R., VDI Berichte Nr. 677 (1988) 93-97.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

4 Sensor Design and Packaging


ROSEMARYL . SMITH.University of California. Davis. CAYUSA;
SCOTTD. COLLINS.
Damien Associates Davis. CAYUSA

Contents
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.2 System Design . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.2.1 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.2.2 System Partitioning . . . . . . . . . . . . . . . . . . . . . . . . 82
4.3 Chip Layout and Package Design . . . . . . . . . . . . . . . . . . . 83
4.4 Technology Constraints . . . . . . . . . . . . . . . . . . . . . . 86
4.5 Packaging Constraints . . . . . . . . . . . . . . . . ...... 89
4.6 Sensing Elements . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.6.1 Transduction Mechanisms . . . . . . . . . . . . . . . . . . . . . 92
4.6.2 Scaling of Sensing Elements . . . . . . . . . . . . . . . . . . . . 94
4.6.3 Sensing Element Compatibility . . . . . . . . . . . . . . . . . . . 95
4.7 Selected Microsensor Examples . . . . . . . .
. . . . . . . . . . . 97
4.7.1 ChemFET with Micromachined Package . . . . . . . . . . . . . . . 98
4.7.2 In-vivo Pressure Sensor . . . . . . . . . . . .
. . . . . . . . . . 99
4.7.3 Floating Element Shear Force Microsensor . . .
. . . . . . . . . . . 101
4.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

4.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104


80 4 Sensor Design and Packaging

4.1 Introduction

Design and packaging are important aspects of sensor development. They are especially
critical for microsensors. Although the basic considerations discussed in this chapter may be
equally applicable to conventional sensors, the focus is on microsensors. In the following
discussion, a microsensor is a transducer which is fabricated using microfabrication
technologies, and microfabrication refers to the collection of techniques used by the elec-
tronics industry in the manufacture of integrated circuits (see Chapter 5, Sections 5.2-5.3).
Applying microfabrication to sensor design is attractive because it combines the economic ad-
vantages of batch fabrication with the dimensional precision of photolithography. In addition
to new manufacturing capabilities for existing types of sensors, microfabrication provides the
means for creating new types of sensors as a result of the properties of microelectronic devices,
and new fabrication techniques, such as micromachining. Most microsensors are either
custom electronic components, or use electronic devices for transduction and readout.
Therefore, this chapter will focus on the constraints which result from the merging of
microelectronics with sensing elements, and presents guidelines for successful microsensor
design.
The use of microfabrication enables the designer to place intelligence, i.e. circuits, at the
sensing site with the microsensor, improving sensor performance through amplification, off-
sets, compensation, etc. Although there can be advantages to merging circuits with microsen-
sors [l-31, integration also constrains the design and fabrication of these systems. Since
fabrication and encapsulation are complicated by the inclusion of active circuitry on chip, it
is critical that integration results in either improved performance or increased capability. The
amount of integration which achieves this is sensor, and application, specific. In the case of
biomedical sensors, increased information gathering capability per unit volume is.often the
driving factor for integration. However, the difficulty of protecting electronics in the
biomedical environment has limited the commercial success of biomedical microsensors. In
the most successful designs to date, integrating electronics has been limited to the amount
which is absolutely necessary to the transduction of the sensed parameter.
Every microsensor requires some degree of encapsulation and packaging in order for the
sensor to function in the desired test environment and to communicate with a remote operator
or actuator. Since microsensors are application specific, the package is also a custom design
feature, and must be designed along with the sensor from the very start, The package will
determine the placement of sensor, circuitry, and i/o pads with regard to function, isolation
and encapsulation. In turn, any packaging process performed on wafer will affect the design
and fabrication of any integrated microelectronic devices or circuits. It is difficult to isolate
the discussion of any aspect of microsensor design from any other, especially its package,
therefore repetition of issues has been unavoidable in this chapter.

4.2 System Design

The block diagram of a typical sensor is shown in Figure 4-1. If one, single type of in-
tegrated circuit technology can realize all the system components, integration of the entire
4.2 System Design 81

3 - POWWSOURCE DATA
INTERFACEClRCUlT H' ACQUlSmON
SImtwCcEmm
8 DlSPlAY
J

- I -
Figure 4-1. A schematic diagram of a sensor system, with its individual components. A microsensor
design may incorporate any amount of the system on the sensor chip, and in the sensing
environment, as indicated by the dashed lines. The system partitioning is determined by the
application and available technology.

system on a single chip is possible, for very low cost per die. This is the rule which drives the
integrated circuit industry to higher density chips. However, the microsensor designer should
not follow blindly down this path. Although it may be physically possible to incorporate the
entire system on a single chip, the resultant loss of modularity creates special difficulties for
microsensors. The design of microfabricated parts is done one layer at a time, with each layer's
layout dependent on the others. The microfabrication process is performed according to
previously developed, specified steps and the geometrical design is fixed by the mask set.
Therefore, every component of the sensor system which is incorporated on-chip must be
designed and fabricated at the same time. Any change in the design of either the sensor or
any circuit requires that an entirely new mask set be made and often, due to technology con-
straints, a new process flow must be developed. This is unlike conventional sensor systems,
where the transducer, its interface electronics, signal processing circuitry, and package can
each be designed and fabricated independently. These requirements make independent op-
timization of integrated sensor system components difficult, if not impossible, and often
results in compromises in performance. In addition, the microsensor environment is not the
well controlled interior of most electronic systems and integrated circuits are very susceptible
to moisture, heat, and reactive gases. The increased requirements for encapsulation, and hence
package design, usually negate the benefits one expects to derive from increased integration.

4.2.1 Calibration

Microelectronic components are themselves very sensitive to temperature, pressure,


humidity and chemicals, and usually non-selectively to all of these parameters. Response of
the transducer is not easily isolated from that of any integrated circuitry. The package may
also contribute to the response due to environmentally induced strain, e.g. the swelling of
epoxy encapsulants due to hydration or thermal expansion. Therefore the microsensor
presents a complicated calibration problem : the sum total of responses to the parameter of
82 4 Sensor Design and Packaging

interest of all the system components co-existing on chip, including packaging effects,
becomes calibrated, rather than only the component that was designed to sense the parameter
of interest, i. e. the sensing element. The resultant calibration is not likely to be the same as
the transducer alone, any individual component, or any subset of components of the system.
This poses several problems for the designer and eventual user. If any part of the integrated
sensor should fail or drift out of calibration, the entire device becomes useless. Even if the
failed component could be identified, it cannot be individually replaced (lack of modularity).
Since microsensors are often destined for hazardous or inaccessible locations, in situ calibra-
tion capability is an important design consideration. This is an especially difficult issue for
in vivo, biomedical sensing where calibration via parameter control is not possible. One ap-
proach to this problem is to design for maximum stability, in order to extend the calibrated
lifetime of the system.

4.2.2 System Partitioning

The designer can minimize those problems inherent to an integrated system, i. e. calibration
and packaging, by judiciously partitioning the principal parts of the system between on-chip
with the microsensor, and off chip. The latter can be designed so that the entire system is
located at the sensing site, in a hybrid package, or with all but the sensing element located
at a remote site. There are many issues to consider when making partitioning decisions. The
ones discussed here are function, performance, and the environment.
The desired function of the microsensor determines its design, including how the system will
be partitioned. For example, if a low cost, disposable sensor for a single parameter is desired,
it is best to keep the design simple - in function, fabrication and packaging. In this case, the
least amount of integration required to perform the given task is preferred. In fact, nothing
other than the sensor on chip may be the best choice. On the other hand, if a small, multi-
sensor is desired, adding a multiplexing circuit can reduce the package size by reducing the
number of i/o connections. In this case, the increase in design complexity is justified by an
otherwise difficult to achieve functionality per unit volume.
The system performance can be improved, unaffected or compromised by integration.
Sophisticated integrated circuits require very tightly specified fabrication sequencing in order
that they arrive at the test bench functioning as designed. When circuits are integrated with
sensing elements, the overall processing sequence is inevitably different from that for the cir-
cuit alone. Even minor sequence or process changes can result in significantly reduced perfor-
mance of an integrated circuit. Therefore, the most desired circuit, in terms of system perfor-
mance, often cannot be successfully integrated with a given sensing element due to fabrication
constraints. In this case, the on chip integration of circuit and sensing element requires the
employment of less sophisticated fabrication technologies and/or devices which may not pro-
vide the preconceived, improved performance. However placing the circuit off chip can also
hamper performance, regardless of its degree of sophistication. A premier example is the sens-
ing of very small capacitance or charge changes. These may occur in response to the physical
displacement of a mechanical structure [4] or a change in dielectric properties of a material
[ 5 ] . Maximizing sensitivity while converting capacitance (or charge) to voltage (or current) will
depend on the parasitic capacitances, and hence the length, of any charge transferring leads.
A comparison of off chip to on-chip lead lengths supports on chip placement of the conver-
4.3 Chip Layout and Package Design 83

sion devices, or circuit. Another example where placement of circuitry on chip is advantageous
is the use of buffer amplifiers for the reduction of the output impedance of the measurement
signal. Reduced output impedance improves the signal to noise ratio and makes the signal less
sensitive to changes in output lead and connection impedances, which can occur due to encap-
sulation failures.
It has already been mentioned that microsensor design is constrained by the intended en-
vironment. The degree of protection required, i. e. encapsulation and isolation, depends on the
type of environment and on the type and role of any on-chip electronics. The operation of
integrated circuits is limited in temperature range, primarily due to the exponentially increas-
ing behavior of the intrinsic carrier density and, consequently, reverse bias diode currents.
When structures are used which employ diodes to electrically isolate regions of a semiconduc-
tor, such as diffused resistors, leakage is a major concern. This limits the usefulness of most
integrated electronics to temperatures below 150°C [6].Therefore, if temperatures in the in-
tended environment exceed this, circuits should be partitioned off chip, if possible. Otherwise,
specially designed circuits [7] or thermally isolated structures [8] need to be implemented.
Corrosive environments pose severe constraints on the encapsulation integrity, and par-
ticularly on any current carrying leads. In these environments, the interconnections of on-chip
circuitry must be totally and effectively encapsulated. If the circuit need not be on chip, and
consequently in the corrosive environment, its lifetime, and hence the sensor’s, will be ex-
tended. The input/output (i/o) signal instability, due to chemical and/or mechanical instabil-
ities in connections and cable, may limit sensor lifetime and performance. The situation can
favor either more or less integration. For example, a sensor design which incorporates at-site,
signal modulation may be the best solution for eliminating signal sensitivity to changes in i/o
lead impedance [9]. Integration of radio telemetry circuits would eliminate i/o connections en-
tirely, and the entire chip could than be hermetically sealed in a metal or ceramic package [lo].
Sensor designs which incorporate circuitry for added performance and/or function are refer-
red to as “smart“ sensors. A more detailed discussion of smart sensors can be found in
Chapter 12.

4.3 Chip Layout and Package Design

The logistics of where one places the sensor, input/output connections and any integrated
circuitry on the silicon die are different for the integrated sensor designer and the IC designer.
For the latter, optimizing the performance of the circuit, perhaps by minimizing real estate,
resistance or parasitic capacitances, is the major concern. Bonding pad number and placement
are predetermined by the IC package manufacturers. Although it is true that the number of
i/o connections currently limits the minimum size of an IC chip, and that this is a major area
of current R & D, very rarely does the IC designer consider a custom package. Standardization
of packages enables automated assembly which keeps IC packaging and product costs down.
On the other hand, sensor packages are always custom items. The package is closely tied to
the environment in which it will go and the parameters that the system is designed to sense.
The sensor designer must design the package at the same time as the chip, and therefore is
concerned with how and where he will connect off chip, where a window to the sensed
parameter will be, and the size and shape of the chip. The placement and number of i/o con-
84 4 Sensor Design and Packaging

nections are determined by the desired ultimate size of the packaged part, the environment,
the lead attachment technology, and the packaging materials application technology.
The standard IC package is fast becoming obsolete for microsensors. They are often too
bulky, especially for biomedical applications, and have no window through which the desired
parameter can be sensed. In fact, they were designed to the contrary: to keep out the environ-
ment. Some parameters can penetrate through the ceramic or plastic packaging materials, such
as heat or magnetic fields, but they were not designed, and therefore not optimized, for that
purpose. Packages with windows to light are also available as a consequence of the Erasable
Programmable Read Only Memory (EPROM) technologies, where erasure is accomplished by
exposure to ultraviolet light. However, windows to gases and liquids and mechanical forces are
not available options with commercial packages.
Usually, the very same parameters which one would like to sense, or the environment where
the parameter is to be sensed, cause havoc with integrated circuits. Humidity is an excellent
example. In the presence of humidity, metal lines corrode, even in the absence of en-
vironmetally supplied ions, since there is usually a sufficient supply of residual, process related
chemicals on the surface of the chip. Therefore, the same package must provide both a window
to sensing elements on the chip, and encapsulation and protection to the circuitry and the i/o
connections. The major difficulty encountered in the design of such a package is the sealing
of the encapsulant to the surface of the chip. Standard IC packages totally encase the chip
in the encapsulating material, leaving no exposed interfaces. Insertion of a window implies
a local change in material and therefore interfaces. It is along these interfaces that encapsula-
tion failures most often occur. This is one of the many reasons why the wise sensor designer
places circuitry and i/o connections as far from the window as possible. A comparison of the
standard IC layout and a frequently employed microsensor layout scheme is shown in Figure
4-2.
Presently, there are but a few window technologies employed in microsensor packages. For
chemical sensing, the harsh environments require the use of the best known and available en-

I / O INTERCONNECT
Figure 4 2 . Examples of chip layouts for an integrated circuit and a microsensor. Integrated ciEuit
packages provide bonding connections along the entire perimeter of the chip to maximize the
available i/o. Microsensor encapsulation is usually facilitated by placement of sensor as far
from i/o as possible, allocating the space between any conversion devices, on chip
circuitry, and/ or test structures.
4.3 Chip Layout and Package Design 85

capsulating materials which can be, in some fashion, patterned to "open" a window. Encap-
sulation with epoxy resin [ll] is accomplished by painting, or flowing, the precursor over the
chip and wire bonds. With the bonding pads placed along one edge of the chip, as far as possi-
ble from the sensing area, the flow of epoxy can be restricted to outside the sensing window
area. Barriers can be formed, by building a wall around the window, with viscous epoxy, or
by means of a temporary window covering. Anodic bonding of glass to silicon is used to
package some microsensors, such as the piezoresistive pressure sensor [12]. The sensing win-
dow is a hole that has been drilled into the glass, and which aligns over the deformable silicon
membrane. The anodic bonding process hermetically seals the glass to the silicon substrate
wherever the two come into intimate contact. The hermeticity of the seal requires an extremely
planar surface (step heights < 1000 Angstroms), completely surrounding the sensing window.
It is evident that for either method of encapsulation, the chip layout must accommodate win-
dow definition.
Microsensor design often includes three dimensional, micromachined structures.
Micromachining refers to a sequence of deposition and etching processes which produce
microstructures in or on the surface of an otherwise planar substrate (see Chapter 5). Ex-
amples include cavities and wafer via holes, and moveable parts, such as diaphragms and can-
tilever beams. These designs are two-sided and multi-planar, resulting in layouts of additional
complexity. For example, the proof mass, which moves in response to acceleration in micro
accelerometers, is formed by anisotropic etching of the silicon substrate [13, 141. It is in-
herently important to the design of these structures that their dimensions, and hence their
mass, are controlled and reproducible. The shape of a microstructure which results after ex-
posure to an anisotopic etchant is determined by the geometry of the mask pattern, its align-
ment to the cystallographic planes, the time of etching, and the masking material (see
Chapter 5). Therefore, the designer must know the fabrication sequence, and be knowledge-
able of the particular micromachining technology to be implemented, in order to provide the
proper layout.
Integrated circuit packages (such as the plastic, dual in-line package (DIP) and the metal,
TO-5header) have external leads which were designed for printed circuit board, solder or wire-
wrap, connections. More recent innovations are the ceramic chip carriers with reflow solder
leads for hybrid circuit boards. In any case, the IC chip package enables it to be interconnected
with other chips and discrete components on a board, to and from which signals and power
travel in a cable. The goal is an ultra-high density of mix and match components to create
a system which is then placed in an environmentally controlled box. The microsensor, however,
is a custom designed system or component of a specific system, which is often intended
for a remote sensing application, in a hostile environment. Such applications require
custom cables to which the integrated sensor can be attached and which will survive the en-
vironment. Additional requirements may be reduced total size of the sensor plus cable, and
cable flexibility, especially for many biomedical sensors. Tape Automated Bonding (TAB) is
a commercial means to connect a single chip to a flexible cable [15]. However, it does not pro-
vide encapsulation of the i/o connections and the compression technique of bonding the cable
to chip limits its overall miniaturization.
In answer to these packaging challenges, microsensor designers have developed custom
cables and i-o connection schemes. For example, Barth et a1 [16] have integrated both cable
and sensor on wafer to create flexible arrays of thermal sensors. Silicon islands, each contain-
ing a p-n junction thermal sensor, are suspended from polyimide ribbon cables. This design
86 4 Sensor Design and Packaging

required a vastly different approach to layout than for a conventional IC. Here, a dispropor-
tionate amount of the silicon wafer surface is taken up by i/o structures. In fact, most of the
silicon substrate serves only as a temporary support during processing and is eventually etched
away. This is another example of where the layout is geometrically constrained by the package.

4.4 Technology Constraints

In many cases, the required sequence of process steps for electronic component and circuit
fabrication are not compatible with the sensing element’s fabrication. Integrating a sensing
structure is constrained by the limitations of planar processing and the very tight material pro-
cessing controls required for properly functioning electronics devices. The most sucessful way
to integrate sensing elements with circuitry is to minimally perturb the circuit process. This
is often referred to as IC compatible processing. In some cases, integration is readily
achievable by partitioning the fabrication steps into those which are accepted as standard in
the integrated circuit industry and can be performed during microelectronics fabrication with
no perturbation to that process, and those which are non-standard, and cannot be readily pro-
cess integrated. The latter are then performed either prior to, or after the IC compatible steps.
If the IC processing steps are performed by a commercial foundry, additional constraints will
be placed on the design and flow sequence by the foundry’s process engineers. Foundries are
reluctant to accept pre-processed wafers from an outside source into their fabrication lines due
to risk of contamination. Therefore, the use of a commercial fabrication line will undoubtedly
limit partitioning of non-standard process steps to after the devices have left the foundry.
The merits of technology partitioning become apparent when one attempts to integrate
micromachined parts with a circuit. If the micromachining produces structures which are
mechanically fragile or result in a non-planar wafer surface, it is best to design the process
such that these structures are created post-IC fabrication. In this way one can minimize the
number of times one must spin on photoresist onto a wafer with holes, or fragile structures.
Also, the accurate photolithographic transfer of patterns depends heavily on the planarity of
the substrate, for uniform photoresist thickness and optical image focusing. This is usually
most critical for microelectronic devices and circuits, since their dimensions are usually much
smaller than sensor structures and their dimensions are very tightly associated with function
and performance. On the other hand, if the integrated, micromachined structures do not
substantially effect planarity, such as silicon membranes, but do require the use of etchants
for which suitable low temperature depositable masks are not available, micromachining prior
to IC fabrication may be more suitable.
An example follows, where both pre- and post-IC processing of a mechanical structure are
performed in order that an integrated MOS device’s process is not perturbed. The microsensor
in Figure 4-3 is a pH ChemFET with an on-chip, miniature reference electrode [17]. The
mechanical structure is a porous silicon membrane, which will serve as the diffusion barrier
and liquid junction for the on-chip reference electrode. The major constraints placed on the
fabrication of this device are (1) the need for complete electrical isolation of the FET from
the test solution, (2) the process conditions for membrane formation, (3) porous silicon for-
mation and (4) the highly reactive nature of porous silicon. The first constraint is handled by
4.4 Technology Constraints 87

A.

B.

C.

PoRoUss l u m

Figure 4-3. Fabrication sequence for an integrated ChemFET and micro reference electrode. (A) mem-
brane is formed by anisotropically etching the substrate, (B) FET fabrication flows according
to standard CMOS process up to metallization,(C) membrane is made porous in HF solution.

the FET’s complementary structure and the use of silicon nitride as a surface layer encap-
sulant. These are both discussed in more detail in Section 4.7. The location, and size of the
silicon membrane on the chip is determined by the opening in the masking material on the
backside of the chip. The design requires rather thick (50 vm) membranes, which are not
fragile. Also, the top surface of the wafer remains planar, and does not interfere with standard
photolithographic processes. Therefore, the membranes were formed prior to FET fabrication
(Figure 4-3a.) The FET’s were fabricated according to a commercial, metal gate CMOS pro-
88 4 Sensor Design and Packaging

cess (Figure 4-3b.). Porous silicon formation requires the exposure of the silicon membrane
to concentrated hydrofluoric acid solutions [18]. Hydrofluric acid etches most IC materials
very rapidly. Although noble metals, e.g. platinum or gold, do not etch in HF, they do not
adhere well to silicon or its oxides, and the conventional adhesion layers are very susceptible
to HF. This makes masking of adjacent microelectronic structures very difficult for the length
of time required to make thick membranes completely porous. Therefore, it would be
preferable to form the porous silicon prior to FET fabrication. However, the highly reactive
nature of porous silicon precludes this, since it will oxidize during the oxidation process steps,
plugging up the pores [19], and etch during many of the requisite wet etching steps. Therefore,
the pore formation step was left to post FET processing, and the masking problem was solved
by the use of a mechanical jig to expose only the back side of the wafer to the HF solution.

A.

B.

C.

D.

Figure 4-4. Process sequence for the formation of a surface micromachined, free-standing microstruc-
ture. (A) spacer layer is deposited and patterned to form anchor, (B) the microstructurethin
film is deposited and patterned, (C) the spacer layer is selectively etched out from under the
microstructure, (D) the resultant free-standing structure, with dashed line indicating effects
of residual stress.
4.5 Packaging Constraints 89

Technology partitioning can also be applied to microsensor designs which employ the so-
called surface micromachining process [20, 21, 221 for micromechanical structure fabrication.
These structures are formed by the sequential deposition of sacrificial spacer layers and struc-
tural layers of thin film materials (Figure 4-4). Here, the process steps used for fabrication are
IC compatible and they do not result in a non-planar surface, or fragile structures, until their
release, which is the very last step performed. The major technological constraint encountered
in the integration of these structures with microelectronics is the selective removal of the
spacer layer. These layers may need to be microns thick, or to be etched out from under very
wide structures. In either case, the etch time is much longer than normally required for patter-
ning thin films. Therefore, etchant selectivity and the lifetime of the masking layer become
critical issues in the design of these devices.
Other, less readily solved, constraints result from high temperature steps in addition to those
required for circuit fabrication, the deposition of intermediary or additional thin film layers,
and the inclusion of new or usually considered contaminating materials. The interaction of
process steps and materials must be carefully considered in the process design. For example,
residual stress in thin films can result in structures which bend out of plane when released [23],
as shown by the dashed line in Figure 4-4 d. One technique to relieve residual stress is to anneal
the structures, before release, at high temperatures ( - 1000°C) [24]. At this temperature,
dopants will redistribute in the silicon substrate, and microelectronic device characteristics will
definitely be affected. Process modeling programs, such as SUPREM [25], are very helpful
in predicting such effects, but to date these programs are limited to conventional IC fabrica-
tion steps and materials. Hence, the design of microsensors is, at present, constrained by the
lack of design aids and material properties knowledge.

4.5 Packaging Constraints

The packaging and encapsulation of active components is critical to the usefulness of the
integrated sensor. It has been pointed out that the degree of integration (system partitioning:
Section 4.2.2) and the relative positioning of circuits, sensor, and i/o leads (chip layout : Sec-
tion 4.3) becomes intimately dependent on the packaging capabilities and the effects of the
environment on the functioning of the system. In addition, the process design is constrained
by the package design, technology and materials.
The trend in microsensor packaging is wafer level encapsulation and micromachined
packaging [26,27]. This is due to the need of a sensing window, the high quality encapsulation
afforded by thin film glasses and ceramics, and the cost advantages of batch assembly. The
wafer level process steps performed for the packaging of the device must also be integrated
in a compatible manner with the microsensor fabrication. Solid-state coatings with low water
and ion permeability, such as silicon nitride and aluminium oxide, are frequently chosen as
microsensor encapsulating layers. Chemical vapor deposition (CVD) techniques [28] have
been developed and perfected which produce pin-hole free, uniform and conformal thin films
of these materials. They can be patterned by standard photolithography and plasma etching.
Although films can be deposited by plasma or photo enhanced CVD at low temperatures, the
highest quality films, in terms of encapsulating capabilities, are deposited at temperatures
90 4 Sensor Design and Packaging

greater than 400°C. A typical deposition temperature for low pressure CVD silicon nitride is
750"C, and it takes around 30 minutes at this temperature to deposit lo00 Angstroms. For this
time and temperature, dopant redistribution in the underlying silicon is minimal. However,
temperatures greater than 400 "C preclude the coverage of any interconnect lines if aluminium
metallization is used, due to the high diffusivity of aluminium in silicon and oxides, and its
low melting point (660°C [29]). In order to use CVD silicon nitride as an encapsulant of in-
tegrated circuits, including their interconnect, a process must be employed which uses a com-
patible interconnect material, such as silicides [30].
The constraints of using solid state encapsulants are not limited to their high deposition
temperatures. In addition, it is known that the deposition of silicon nitride over MOS devices
can cause immediate and long term shifts in threshold voltages [31]. This is believed to be due
to hydrogen, which is a by-product of silicon nitride CVD deposition. Similarly, the CVD pro-
cesses of oxides, such as silicon dioxide, aluminium oxide or tantalum oxide, employ strong
oxidants, e. g. water, nitrous oxide and oxygen, which cause the oxidation of underlying silicon
structures. Finally, the strain induced in the substrate due to the high residual stress [32, 331
of these CVD films can affect semiconductor device behavior.
As mentioned earlier, batch assembly techniques are attractive from a packaging cost
perspective. An example of the application of wafer level processing to assembly is the provi-
sion of backside contacts to i/o ports on the front side of a wafer [34, 351. This technique
can greatly simplify the encapsulation of sensors, by placing i/o on the opposite side of the
chip from the sensing environment. However, the spatial and processing requirements of this
technique place constraints on the type of integratable circuitry and the number of i/o. One
manner for creating back side contacts is depicted in Figure 4-5. Via holes are etched into the
back side of the wafer, just opposite the front side contacting area. Alignment from front to
back requires special equipment, e. g. infrared imaging or a mechanical jig. The sides of the
via hole are either doped, or coated with a conducting layer, or both, to electrically short the
front to back contact. Isolation of the contact from the substrate can be provided by doping

A. pp
..............

........
........
.. .. .. .. .. ......
......... . . . ...........
.. .. .. . .
...........................
..........................

R I I

C. Figure 4-5.
P-tw I/O contact formed from back-to-front
diffusion of the microsensor chip by anisotropic
etching and selective doping, showing
(A) back, (B) front, and (C) cross-
Metal sectional view.
4.5 Packaging Constraints 91

the sides of the hole with the opposite type, thereby creating a diode. This diode has a large
area, with an associated large junction leakage current and capacitance, which must be con-
sidered in the electronics design. If a CMOS process is used to fabricate the circuits, this
method cannot be used to make contacts to diffused regions within isolating wells, since the
via crosses regions of both conductivity types. Also, this method consumes a great deal more
chip space per contact than the more conventional wire bonding method, and therefore
significantly reduces the number of contacts possible.
The use of spin-on coatings for encapsulation has the advantage of providing conformal
coatings of controllable thickness, at relatively low temperatures and low costs. Spin on
glasses, polyimides, epoxies and silicones have all been used as encapsulating or isolating
layers. Difficulties occur in patterning thick layers, spin coating over three dimensional struc-
tures, especially holes, and mismatch in thermal expansion coefficients to silicon, resulting in
high residual stress. The later can result in substrate warping (polyimide), or film cracking
(glasses).
The application of micromachining techniques to sensor packaging has resulted in new and
improved encapsulation methods and structures, which would be unattainable with conven-
tional machining and assembly techniques. One example is the use of a separate, micromach-
ined substrate as part of the package, which is bonded to the sensor containing substrate as
part of the assembly process [27, 361. The advantages of separate, micromachined packages
are: (1) they can be fabricated independently from the microsensor, allowing for great dimen-
sional and functional variety without the usual technological constraints, and (2) the micro-
machined substrate and the substrate can be bonded together at the wafer level, reducing
packaging costs by utilizing batch assembly. The constraints placed on microsensor design by
this approach to packaging result from the bonding and die separation processes. Die separa-
tion constrains the layout of the sensor chip with respect to the micromachined substrate. The
bonding process constrains the designer in less obvious ways, therefore a brief discussion of
bonding techniques follows.
Silicon and glasses of similar thermal expansion can be hermetically sealed to silicon
substrates by field assisted (anodic) bonding [37]. Pyrex 7740 has been used for many years
to hermetically seal the reference chambers of piezoresistive pressure sensors and as a ther-
mally matched mounting pedestal [12]. The process of anodic bonding employs high fields
(200- 1000 V) and temperature (45OoC-200"C) in order that sufficiently large anodizing
current will flow across the silicon substrate/glass (or silicon dioxide) interface to chemically
bond the glass to the substrate [38]. The use of field assisted bonding to attach the package
to the sensor wafer poses the following constraints: (1) the high fields and temperatures
required for bonding can be detrimental to electronic devices, particularly MOS devices,
(2) the bonding technique requires a conductive plane on the substrate surface which comes
in direct contact with the glass, (3) the bonding technique requires an extremely planar surface,
c 1000 gngstroms steps. These constraints can be met by appropriate design of the microsen-
sor wafer, e. g. providing shielding for MOSFET's, using doped polysilicons as the conductive
plane, and employing special planarization techniques.
Silicon can also be bonded to metal coated glass or ceramics, by eutectic sealing or by com-
pression sealing. The eutectic point corresponds to the composition, in a two component
phase diagram, which has the lowest melting temperature. Silicon and gold have a eutectic
point at 363 "C, with composition of 97.1% Au and 2.85% Si. Silicon can be bonded to silicon
or to glass by coating the substrate with gold and heating the two substrates while in intimate
92 4 Sensor Design and Packaging

contact, to the eutectic temperature [26]. Room temperature sealing can be achieved with com-
pression metallic bonding of indium or gold/aluminium coated substrates.
Another, newer bonding technique is silicon fusion bonding [39]. This process can be used
to chemically bond two silicon wafers together. The process requires highly planar structures,
temperatures greater than 700°C and an oxidizing ambient. Due to the extreme conditions of
the bonding process, this technique is most applicable to bonding two micromachined wafers,
devoid of electronic devices, to achieve additional three dimensional flexibility.

4.6 Sensing Elements

Sensing elements can be divided into two generic classes according to what the primary
sensing response is directed : a chemical species or a physical process. The physical processes
that have been successfully incorporated into integrated microsensors include heat, flow,
pressure, acceleration and electromagnetic radiation. These sensors have found a substantial
degree of success in commercial markets and have reached a degree of integration and
sophistication that greatly surpasses that of the chemical sensors. This results, at least in part,
from the compatibility of present packaging technologies to accomodate the sensing elements
of these physical processes with the simultaneous exclusion of the surrounding unwanted and
deleterious environment, i. e. chemical barriers. The common denominator among the various
physical sensors is that their packaging allows selective permeation of the physical environ-
ment through windows while at the same time excluding all chemical species. For the chemical
sensor this is not possible. At least one chemical species must be allowed to pass through the
window. This transfers the burden of selectivity from the sensing element to the packaging
level. It is exactly this obstacle that has impeded the development of successful chemical sen-
sors to date. Although most of the principles presented in this section apply equally to physical
or chemical sensors, the emphasis will be directed to chemical sensing technologies and their
relationship to microsensor design in terms of materials, processing, and IC compatibility.
The discussion which follows is instructive, not comprehensive, and is organized to provide
insight into design issues and principles.

4.6.1 Transduction Mechanism

A sensor is a transduction device which converts one or a combination of chemical and/or


physical signals into an easily quantifiable output, usually a voltage to control an electronic
display or chart recorder. The sensing element is that critical portion of the sensor where the
primary transduction occurs and, as such, is vitally important in the operation of the total
sensor. If the proper design and functioning is not accomplished at this level, then there is
little hope for the sensor no matter how sophisticated the signal processing or packaging may
be. Therefore, from the conception of the sensor, it is crucial to clearly define the optimal
transduction mechanism. When undefined or poorly understood transduction mechanisms
are employed in microsensor designs, development delays and fabrication problems are bound
to occur. For product development, it is best to consider only mature and/or well defined
4.6 Sensing Elements 93

transduction technologies, and delegate the more questionable sensing mechanisms to the
research bench.
The ChemFET (Chemically sensitive Field-Effect Transistor) is an excellent example of a
microsensor which has suffered several developmental setbacks due to the lack of understand-
ing of transduction mechanism, operation principles, packaging and long term drift behavior.
When it was discovered that solid state insulators utilized in MOSFET gate structures, i. e.
silicon dioxide and silicon nitride, exhibited an almost Nernstian sensitivity to H +, the
ChemFET issued forth as the first integrated chemical microsensor. In testing, it became ap-
parent that ChemFET pH sensitivity and ion selectivity depended strongly on the insulator
material. However, since the mechanism for the observed behavior was unknown, no explana-
tion or reliable method for controlling sensitivity was found. Only recently has the behavior
of the insulator/solution pH sensitivity been carefully studied and mechanisms have been pro-
posed [ a ] .
In the original ChemFET implementation by Bergveld [41], a reference electrode was not
included in the measurement system. This was later recognized by Janata et a1 [42] to be a
violation of first principles, and it has since been widely accepted that reliable ChemFET
operation requires a reference electrode. The most probable reason that the initial, no
reference electrode, ChemFET functioned at all was due to poor or faulty encapsulation of
the completed sensor package. Because of the high impedance of the FET gate insulator, any
resistive path between solution and ground served as a reference, including hydration paths
through the bulk epoxy used to encapsulate the FET and along interfaces between the FET
surface and epoxy. It is reasonable to believe that encapsulation failures also contributed to
the observed output instability of these devices. Encapsulation of solid state chemical
microsensors has dramatically improved in recent years, but the ChemFET response continues
to demonstrate a significant long term drift. The insulator/solution interface is believed to be
the source of this problem, however the mechanisms involved are still unknown. This par-
ticular incidence serves to emphasis two of the most important aspects of microsensor design
principles : to understand the particular transduction mechanism in question before using it,
and pay attention to the isolation and packaging of the sensor.
Although the above example occurred almost 20 years ago, during the infancy of microsen-
sor research, the application of poorly understood transduction mechanisms in microsensor
design is a recurring problem. The more recently proposed liquid-based acoustic oscillator for
immuno-sensing is another example of a chemical microsensor technology which is lacking
an acceptable theoretical basis of operation. Acoustic oscillators were initially used as sensors
in thickness monitors of evaporated films in vacuum deposition systems [43]. The technique
possesses ingstrom sensitivity and the transduction mechanism is beautifully simple and
easily understood. Evolution of the acoustic thickness monitors into chemical gas sensors oc-
curred with the realization that the oscillator resonant frequency was not only a function of
thickness, but also of the acoustic properties of the deposited film. Chemically selective gas
ad/absorbing films were deposited on the surface of the oscillator, and the coatings altered
the acoustic path length (resonant frequency) as a function of gaseous analyte [44].Many dif-
ferent types of acoustic oscillator chemical sensors, sensitive to a wide variety of gaseous
analytes [45-511 have been reported. The use of acoustic oscillators as chemical gas and vapor
sensors appears to be a working, commercially viable microsensor technology. However, the
extension of the sensing mechanism to liquid-based, acoustic chemical microsensors has not
met with an equivalent success. Bastiaans et a1 [52] were the first to demonstrate that an
94 4 Sensor Design and Packaging

acoustic device could be made to oscillate in aqueous environments, despite the resulting
dramatic attenuation in signal, and they promptly proceeded to produce antigen/antibody
sensors by immobilizing the corresponding antibody/antigen to the surface of the acoustic
device [53]. Results, so far, have been nonreproducible. A transduction mechanism similar to
that of the gas acoustic sensors was initially used to explain the experimental "mass loading"
response. However, with the exception of attenuation effects [54, 551, little serious considera-
tion has been given to the ramifications of intimate acoustic contact with a dense medium,
e. g. the effects of acoustic transmission and reflection back to the sensor, with resulting in-
terference. Recently, Martin et al. [56] have presented a liquid based acoustic sensor utilizing
a surface shear wave instead of the normal Rayleigh (Surface Acoustic) wave. This essentially
evanescent acoustic mode by-passes most of the "acoustic interferometry" problems en-
countered in the bulk of the liquid. Despite this improvement, Grabbe et al. [57] have recently
presented data which indicates that it may be fundamentally impossible to obtain reliable
qualification of surface ad/absorbed organic and protein analytes owing to variations in solva-
tion and viscoelastic effects at the sensor surface. In light of these arguments, it appears that
if liquid-based acoustic chemical sensors are feasible at all, their scope of application will be
severely restricted to specialty analytes under well defined test conditions. Again, the develop-
ment of these sensors could have proceeded more directly if early efforts had gone towards
understanding the transduction mechanism.

4.6.2 Scaling of Sensing Elements

There are practical limits to the extent of miniaturization that any sensing technology can
endure. The miniaturization of the piezoresistive pressure sensor to a microfabricated version,
results in a loss of modularity, as discussed in Section 4.2, so that the mechanical, thermal,
and electrical interactions between diaphragm, supporting chip and. the package, together
determine the sensitivity of the sensor. However, the actual transduction mechanisms have not
been changed by the reduction in size of the primary sensing element, which in this case is
the combined diaphragm and piezoresistors. Rather, other factors, which were negligible in
their effect on sensitivity for a macrosensor, have become significant. Another example of this
is the viscous damping of a micro-mechanical element due to the very small spaces through
which air must move when the element is displaced. However, there are cases where
miniaturization to the micron scale results in different governing chemical and physical pro-
cesses, and consequently they yield different responses. One example is the change in slope
of the variation of dielectric breakdown voltage of air with air gap distance, from positive to
negative, for distances less than about four microns (Paschen's Law). This phenomenon has
significant influence on the design of electrostatically driven, micromechanical elements.
Similarly, the scaling of macro sensing materials and technologies to the micro level in
chemical microsensors poses some unique design problems. In the development of the
ChemFET, this principle is best demonstrated by the application of Ion Selective Electrode
(ISE) membranes to the FET.
Macro ISE membrane systems are available for a host of different ions, and there are
numerous reviews of both theoretical and practical importance on the subject [58]. The
technology enjoys almost 100 years of maturity and has produced well behaved and well
understood ISE's for almost any ion imaginable [59]. Most of the macro ion selective mem-
4.6 Sensing Elements 95

branes are based on immiscible organic liquids. Either the liquid, itself, is electrochemically
active and capable of generating a selective potential, or the liquid is "doped" with an
ionophore or ion-exchange material. Usually, the liquid is incorporated into an inert polymer
matrix such as PVC [60] to produce a rigid or semi-rigid, mechanically robust "gel". Other
polymer matrices have also been used mainly to facilitate selective membrane deposition
and/or patterning, e; g. photoresist [61], silicone rubber [62], polyimides [63] and others [64,
651.
The expansion of the ChemFET sensing repertoire beyond H + by using ISE membrane
systems was a natural choice. In scaling the macro, thermodynamically defined ISE membrane
systems to the micro level, certain physical realities must be recognized. The small finite
volume of the micro membrane leads to pronounced dissolution and/or leaching of the active
components from the membrane. This results in significantly reduced sensor lifetimes and in-
creased long term sensor drifts. The finite reserve of the membrane ultimately results in the
catastrophic failure of the sensor. When designing microsensors with chemically active sensing
layers it is important to consider the effects of finite, non-zero solubility and chemical
degradation which may not be important in the large reserves of the macro world.
To minimize the effects of leaching and dissolution, membranes have been developed in
which the active components are covalently attached to the membrane [66]. This is an im-
provement, however systems utilizing covalent attachments show an inferior response and sen-
sitivity. An extension of covalent attachments are the solid state membrane systems, such as
silicon nitride, aluminium oxide, tantalum oxide [41, 671, or molecular organized systems like
Langmuir-Blodgett films [68, see also Chapter 51. These systems are on the order of 30 to
1000 Angstroms thick, and show considerable promise in the future development of the
ChemFET sensor. Although the Langmuir-Blodgett film technology is more speculative, the
solid state, insulating oxide systems have already shown a practical and realistic sensitivity to
H + . If ISE technology is to be continually scaled, covalent and solid state membrane systems
must be the answer. They are ultrathin, IC process compatible, and selectively patternable. Un-
fortunately, the range of analytes to which the present solid state membrane technology is
selectively sensitive is limited to H + . Work is presently underway to develop new solid state
membranes by the ion implantation of gate oxides [69]. However, material technology must
undergo considerable advances before a larger variety of solid state membranes are available
(including solid state reference electrodes).
The atomistic dimensions of these solid state films produce another problem associated
with scaling of chemically sensitive layers. In the macro world most chemical processes are
defined in terms of their bulk thermodynamic relationships. In scaling a membrane or
chemical system into the micro world one frequently passes out of the realm of ther-
modynamics, and statistical interpretations must be invoked. For example, it is not likely that
thermodynamic relationships are valid for a membrane thickness of roughly 100 atoms (100
to 300 Angstroms). However, dimensions on this order of magnitude are frequently employed
in microsensor structures and sensing elements. The familiar bulk thermodynamic relation-
ship can not be expected to automatically apply within the micro world.

4.6.3 Sensing Element Compatibility


From the moment a microsensor is conceived, it is crucial that fabrication flows from a col-
lection of basically compatible processes. It is important to consider all process flows, in-
96 4 Sensor Design and Packaging

cluding encapsulation and packaging. Thin film solid-state sensing layers, such as silicon
nitride, are deposited and patterned by microfabrication techniques, and are therefore
basically compatible with microsensor structures. On the other hand, the application of a li-
quid or semi-liquid organic layer to a solid state, planar microsensor structure represents a
basic incompatibility between technologies from the point of view of deposition, adhesion and
thermal stability. For example, the high temperatures normally used in the fabrication of solid
state circuits will destroy organic ISE membranes. Therefore, the deposition of organic mem-
branes must be delayed until all high temperature processing is completed. In addition, pro-
cessing steps involving organic solvents must be avoided after the deposition of an ISE mem-
brane to prevent its dissolution. This eliminates a host of general process steps including clean-
ing and standard photolithography procedures. In addition, the “liquid“ state of the standard
ISE membrane is not physically compatible with the planar realm of solid state fabrication
in regard to membrane deposition, adhesion and containment. The combination of ISE mem-
brane technology with ChemFET’s is a classic example of technology incompatibility. It is in-
structive to recount the development of ISE membranes and ChemFET integration, since
many ingenious designs and microstructures have evolved in the search for a solution to this
problem.
Originally, polymer based membranes were solution cast onto an epoxy resin encapsulated
ChemFET, leaving the membrane covering the gate region, as well as the rest of the chip, when
the solvent evaporated [41]. This particular deposition technique was quick and easily per-
formed, however the lack of selective placement of the membrane precluded a multisensor
device. In addition, attachment of the membrane to the sensor, particularly the active gate
region, relied solely upon physical adhesion which frequently failed upon immersion in
aqueous environments. Even minor solution shunts around/or through the membrane
diminishes sensor response and contributes to excessive long and short term potential drifts.
The physical adhesion was initially improved by surface silanization techniques (701, however,
it was soon realized that physical adhesion was insufficient to guarantee attachment of the
membrane, and researchers looked at the possibility of altering the sensor design and fabrica-
tion to accomodate the specific needs of the membrane technology. This was probably the first
time the entire sensor package was viewed as a comprehensive unit and not as a collection of
individual parts. It was here that the beginnings of a complete microsensor design concept
emerged.
The first attempts to enhance membrane deposition and adhesion through microsensor
design was to produce deep (approximately 100 microns) wells in an encapsulating Riston
polymer [71]. The wells were intended to hold and locally contain the membrane, and in addi-
tion, the geometry of the wells and polymer side walls were intended to enhance membrane
adhesion. The sensor worked reasonable well for short periods of time, but eventually, the
adhesion failed and the membrane lifted out of the well. Although the process by which the
Riston wells were fabricated was crude, it was accomplished at the wafer level. A more
sophisticated approach to membrane adhesion used a spun cast polyimide film in which a
suspended mesh was patterned over the active ChemFET gate area [72]. The mesh acted to
entwine the solvent cast membrane and anchor it to ChemFET gate. Shown in Figure 4-6 is
a diagrammatic sketch of the “suspended mesh” showing the positioning of the mesh above
the ChemFET gate region. The mesh seemed to solve the adhesion problem for polymer gel
membranes, but did not address the problem of containment of liquid membranes or selective
deposition. In addition, the small membrane volumes imparted an equally short lifetime to
4.7 Selected Microsensor Examples 97

Figure 4-6. Cross-sectional view of a ChemFET chip with polyimide mesh suspended over the gate
region, to improve the adhesion of a polymeric ion selective membrane.

the sensor. To overcome these problems, a three dimensional, silicon micromachined sensor
package has been developed [27]. This structure is described in detail in Section 4.7.1.
From the above discussion, it should be evident that there are very few known absolutes for
chemical microsensor design. Unlike the physical sensors, e. g. pressure, flow, temperature and
electromagnetic, the problems of chemical microsensing technologies are considerably more
profound, numerous and usually, are not inclined to a universal solution. But every sensor
design, whether physical or chemical, must be carefully analyzed in terms of the complete sen-
sor package and appropriate sensing mechanisms. To successfully implement a sensing ele-
ment in a microsensor design, one must :
(1) Understand the sensing transduction mechanism.
(2) Design the sensor as a whole unit, taking into account the interrelationships of the sens-
ing element(s) with the entire sensor package, including encapsulation.
(3) Be aware of problems that arise in scaling well known macrosensing technologies into
the micro world.
(4) Choose sensing technologies which maintain process compatibility when possible, or
isolate and compartmentalize when it is not possible.

4.7 Selected Microsensor Examples

It has been pointed out several times in this chapter that microsensors and their packages re-
quire custom designs. This is usually due to the constraints placed on the design by the in-
tended evironment, e. g. size and durability. The following examples were chosen because they
address each of the design issues addressed in this chapter, i. e. system partitioning, technology
constraints, sensing elements, and packaging. They are especially good examples of creative
approaches to microsensor packaging, using combinations of microfabrication, micromachin-
ing and thin film technologies. A chemical microsensor and two physical sensors are
98 4 Sensor Design and kckaging

presented. Their application determined design specifications are identified and the resultant
design and fabrication are discussed. References to sections of this chapter are given
where design and fabrication decisions were made according to system (Section 4.2), layout
(Section 4.3), technology (Section 4.4), packaging (Section 4.9, or sensing element (Section
4.6) considerations.

4.7.1 ChemFET with Micromachined Package

The ChemFET described here is the same, fundamentally, as in many previously reported
implementations [41, 67,70, 171. The differences lie in how the encapsulation and packaging
of this device is accomplished. This device [27] was designed for long lifetime, submerged in
a biological fluid environment, and for use with liquid ion exchange membranes. Therefore,
special attention was focused on encapsulation, and on containment of a small volume
(several wl) of liquid ion exchange membrane over, and in intimate contact with, the gate in-
sulator of the ChemFET. A cross-sectional diagram of the ChemFET with micromachined
package is shown in Figure 4-7.

n substrate

Fgrre 47. Cross-sectional view of a ChemFET substrate bonded to another silicon substrate, which
contains micromachined chambers for holding liquid ion exchange membranes. The FET is
formed in a p-well. which isolates the active devicc from the silicon substrate and the surroun-
ding solution.

In the Chemically sensitive PBT, the induced field is established by the series combination
of applied "gate" or solution potential and the chemically sensitive, solution/insulator inter-
face potential, with respect to the source or substrate. Electrical isolation of the ChemFET
from the surrounding conductive solution is critical to its proper functioning. In this design,
the electrically biased, or active, regions of the PBT are isolated from the substrate, and hence
the solution, by p-n junctions. The resultant semiconductor structure is called a well, and is
employed in Complementary MOS (CMOS)circuits. In addition, the gate insulator of the
4.7 Selected Microsensor Examples 99

device, and the entire top surface, is coated by silicon nitride. Silicon nitride is an excellent
barrier to ions and water, and acts as an insulating encapsulant. Both silicon nitride deposition
and well formation are standard IC processes, enabling a part of the encapsulation to be ac-
complished on wafer, by solid state materials and methods (Section 4.5).
Since the FET acts as an impedance transformation device, the ChemFET is a low output
impedance device, and therefore has a relatively high signal-to-noise ratio. This precludes the
need for additional on-chip signal amplification or conditioning circuits, despite long i/o leads
(Section 4.2). A MOSFET of similar geometry is included on each chip for process and
threshold monitoring. The chip layout places the ChemFET at the opposite end of the die
from the monitoring MOSFET and bonding pads to facilitate chamber (i. e. window) place-
ment and isolation, as well as encapsulation of lead wires (Section 4.3).
The liquid, ion selective membrane which is used in this design imparts specificity to the
chemical sensor. It is contained by chambers, micromachined in silicon. These chambers are
formed in a seperate silicon wafer, which is then bonded to the ChemFET containing wafer.
The use of a seperate silicon wafer imparts the necessary third dimension for liquid con-
tainment, without necessitating the integration of 3-D structures with the IC processing
(Section 4.4). The micromachining of the chambers employed the back-to-front alignment of
patterns, to realize a configuration where the chambers could have as large a compartment
as possible, 400 x 400 x 100 pm, with a small compartment opening, 30 x 30 pm. This
configuration was designed to minimize the leaching and dissolution of the membrane, and
thereby to increase functioning membrane lifetime (Section 4.6). The chamber containing
wafer layout complements that of the ChemFET containing wafer. A photo of the bonded
wafers is shown in Figure 4-8. Openings are provided over the bonding pad region, and in such
a way that after the separation, the bonding pads are freely exposed and wire bonding can
be performed with standard bonding tools. The bonding pads for the ChemFET and
MOSFET can be seen in the bottom photo of Figure 4-8. Bonding pads numbered 1 and 5
are the drain and source of the ChemFET, 3 and 4 are the source and gate of the MOSFET,
respectively. The FETs have a common drain, pad 5 , and the p-well contact is pad 2. The
micromachined chamber design provides a means for batch assembly and generic chemical
sensing structure, with selectivity imparted by the choice of chemically sensitive membrane.

4.7.2 In-vivo Pressure Sensor

A piezoresistive pressure transducer with on chip bridge circuit and hermetic packaging is
presented as the next example. This device by Burns et al. [73], was designed for long term,
in-vivo pressure monitoring. The design takes into consideration the requirements of small
size, corrosion resistance, ruggedness and long term stability. The pressure transducer is a thin,
deformable silicon diaphragm with piezoresistors positioned along the edges of the diaphragm
(Section 4.6). The piezoresistors are connected in a bridge configuration. So far, this describes
the generic piezoresistive pressure microsensor. This device’s unique features are its interface
circuit and packaging.
The bridge output voltage is converted to a current output and amplified, via an on-chip,
bipolar circuit. The current output signal is less sensitive to i/o lead impedance changes, which
occur from the permeation of the surrounding electrolyte into lead encapsulation. On-chip
voltage regulation is also provided, to improve circuit stability (Section 4.2). An anodically
100 4 Sensor Design and PIJckaging

-4-8.
Top view of a bonded Chem-FET and
chamber containing substrates. The lines
m the upper photo delineate die sepera-
tion lanes. The bottom photo shows the
exposed bonding pad area and the mem-
brane chamber opening over the gate
Photo reproduced from Smith and Col-
lins [27], with permission of the journal.

bonded, Pyrex glass covering is employed to protect the on chip electronics from the conduc-
tive and corrosive biomedical environment (Section 4.5). The use of anodic bonding requires
planar gladsilicon contact surfaces. To meet this constraint, the circuit is fabricated in an
anisotropically etched, 5 micron deep well, and i/o feedthroughs under the surrounding
glass/silicon seal are created by ion implanted lines (Section 4.4). A 5 micron deep, reference
pressure chamber is formed at the same time as the h i t well, and becomes sealed during
anodic bonding (Section 4.3).
A special lead attachment technique was implemented in this design, which is rugged, corro-
sion resistant, and compatible with the anodic bonding process. V-grooves were etched along
the edge of the chip, intersecting the ion-implanted feedthrough lines on one end and the die
separation streets along the other. The V-grooves are heavily diffused with dopant of the same
type as the feedthrough lines, making the two electrically c o ~ e c t e dwhile
, maintaining diode
isolation from the substrate. The V-grooves are coated with a tungsten barrier layer followed
by gold, to prevent gold/silicon interdiffusion during bonding, and to form a solder wettable
surface, respectively. After the glass is bonded to the silicon substrate, the silicon diaphragms
are formed by anisotropically etching. The glass protects the top, circuit containing surface
4.7 Selected Microsensor Examples 101

of the silicon wafer from being attacked in the etchant (Section 4.4). When the die are
separated, the i/o tunnels are exposed. Leads are inserted into the tunnels and are soldered
into place. The final, packaged part, shown in Figure 4-9, has a small cross-section and is
suitable for insertion into a catheter for in-vivo measurements.
SENSING DIAPHRAGM

V GROOMSFOR 1/0

Figure 4-9. Top and bottom views of an in-vivo pressure sensor with on chip signal conditioning, glass
encapsulation, and anisotropically etched V-groove i/o ports, after Burns et a1 [72].

4.7.3 Floating Element Shear Force Microsensor

This microsensor, by Schmidt et al. [74], was designed to detect the shear stress exerted at
the wall of a wind tunnel. This is accomplished by measuring the displacement of a
microfabricated floating element in response to the shear stress, shown in Figure 4-10.
A special consideration in the microsensor design was low profile, for minimal perturbation
of the flow. The same consideration was applied to the package, which was designed for flush
mounting into the wind tunnel wall.
The microfabricated floating element consists of a polyimide plate, with imbedded, metal
thin film, which is suspended 3 microns above a silicon substrate. It is attached to the substrate
at its four corners by polyimide tethers, which enable lateral plate displacement under shear
loading (Section 4.6). An AC drive voltage is capacitively coupled to two sense electrodes lying
under the polyimide plate. The amount of overlap of the plate and the sense electrodes deter-
mine the voltages across respective capacitive dividers which, in turn, modulate the current
flowing through two depletion mode MOSFETs (Figure 4-10 b). The current of each FET is
converted to voltage by means of an electrometer. In this design, the sense capacitances are
on the order of 1 picofarad. This capacitance is small compared to the parasitic capacitance
of i/o leads, necessitating the placement of the capacitive dividers and MOSFETs as close as
possible to the moving plate, i.e. on-chip. The relatively low output impedance of the
MOSFETs enables the placement of the electrometer circuit off chip (Section 4.2). This con-
102 4 Sensor Design and Packaging

SUSPENDED

POLYlMlDE
TEMERS

BURIED CONDUCTOR

h= Figure 4-10.
Schematic diagrams of the micro-
fabricated floating element, showing in
plan view (a), the displacement of the
suspended polyirnide plate under shear
stress and the cross-section (b), the
drive and sense capacitors formed by
the electrodes in the substrate and the
buried conductor in the floating ele-
ment (after M. A. Schmidt [74]).

siderably reduces the constraints placed on the microelectronic device fabrication, since
although each MOSFET pair should be matched in characteristics, their specifications are
otherwise quite flexible. Therefore, the process specifications are much less stringent than for
an electrometer circuit, and the likelihood of sucessful integration is higher (Section 4.4).
The floating element is fabricated after MOSFET fabrication is complete, using only low
temperature (i. e. <450°C) processing. The sacrificial spacer layer is deposited and patterned,
followed by spin coating of polyimide. A thin metal film is sealed between successively spun,
and partially cured layers of polyimide, to form the imbedded conductor. After patterning the
polyimide, the aluminium spacer is selectively etched out from under the floating element.
Knowledge of the residual stresses of these thin films and their process dependence is crucial
to the successful fabrication of a composite, released structure which remains suspended and
flat.
The insertion of the floating element microsensor into the wall of a wind tunnel, so as to
lie in its turbulent boundary layer, required a very planar package which could be flush
mounted. The sensor die itself was flush mounted into a mounting plate comprised of a
micromachined silicon wafer, of thickness equal to the die (Section 4.5). A polyimide flange
was created over the hole into which the sensor die is placed. The polyimide coating on the
sensor die has been patterned to match the flange such that the polyimide of the flange and
the polyimide on the chip meet edge to edge to form an equal height surface from sensor die
to silicon mounting plate (Section 4.3). Pressure taps were also etched into the mounting plate,
up stream from the sensor, for calibration purposes. The package fabrication flow is shown
in Figure 4-11.
*
4.8 Summary 103

SPIN-COAT POLYIMIDE

Polyimide

PLASMA-ETCH POLYIMIDE

4 PLASMA-ETCH SILICON

-Pressure Ta?
ASSEMBLE PACKAGE

Figure 4-ll. The process flow for fabrication of the micromachined sensor support (from Schmidt
et al. [73]).

4.8 Summary

Microfabrication technologies have been developed by the electronics industry, which is able
to support the economic burden of R & D by a large commercial market for general applica-
tion products. The sensor industry,’ which has a much smaller and more custom market, has
been able to glean from the electronics industry without having to provide high capital,
development costs for most of these process technologies. This has been crucial in the evolu-
tion of microsensors. As microsensor specific technologies become more highly developed,
and as markets appear and grow, circuit processing which is fully compatible with sensor
104 4 Sensor Design and Packaging

technologies, and more generic sensor packaging technologies will emerge. Towards this end,
considerable research and development is currently in progress to gather material properties
data and develop simulation programs to simplify microsensor design and fabrication.
It must be pointed out that the examples given above have been designed and fabricated in
academic, research laboratories and have not been commercially manufactured. However,
their unusual, yet successful, integrations of structures, technologies and materials stir the im-
agination. The use of organic thin films, micromachining and bonding technologies in addi-
tion to the more standard IC fabrication technologies provides the designer with an enormous
arsenal of materials and tools. But, the arsenal alone is not enough. In addition, successful
microsensor design requires knowledge of relevant material properties, understanding of the
transduction mechanism, evaluation of circuit requirements, attention to process com-
patibility and protection from the environment.

4.9 References

[l] Muller, R. S., Technical Digest of the ZEDM (1984) 206-209.


[2] Wise, K. D., 6th Sensor SymposiumZEE of Japan, June 1986, pp. 1-9.
[3] Middelhoek, S., Hoogerwerf, A. C., Sens. Actuators 8 (1985) 39-48.
[4] Smith, M. J. S., Bowman, L., Meindl, J. D., ZEEE Journal of Solid-state Circuits SC-21 (1986)
1045-1056.
[5] Sheppard N. F., Day, D. R., Lee, H. L., Senturia, S. D., Sens. Actuators 2 (1982) 263-274.
[6] Sugiyama, S., Takigawa, M., Igarashi, I., Proceedings of the 4th Sensor Symposium,ZEE of Japan;
Tokyo, 1984, pp. 231-236.
[7] Brown, R. B., Terry, F. L., Wu, K-C., ZEDM Technical Digest (1987) 214-277.
[S] Stemme, G. N., ZEEE nuns. Electron Devices ED-33 (1986) 1470- 1474.
191 Chau, H. L., Wise, K. D.. Technical Digest, 4th Znt. Conj on Solid-state Sensors and Actuators,
Tokyo, Japan, June 1987, pp. 344-347.
[lo] KO,W. H., Spear, T. M., Eng. Med. Bio. 2 (1985) 24-38.
[11] Owen, D. R., Zone, R. M., in: Micromachining and Micropackaging of Transducers, Fung, C. D.,
Cheung, P. W., KO, W. H., Fleming, D. G. (eds.); Elsevier, 1985, pp. 19-26.
[12] Tanigawa, H., Ishibara, T., Hirata, M., Suzuki, K., ZEEE nuns. Electron Devices ED32 (1985)
1191-1195.
[13] Wu, X-P., Wu, Q-H., KO, W. H., Transducers '85, International Corlference on Solid-state Sensors
and Actuators, Philadelphia, June 1985, pp. 291 -294.
[14] Chang S-C., Hicks D. B., Technical Digest, ZEEE Solid State Sensor Workshop,Hilton Head SC,
June 1988, pp. 102-103.
[15] Senturia, S. D., Day, D. R., Tmnsducers '85, Technical Digest, International Conference on Solid-
State Sensors and Actuators, Philadelphia. June 1985, p p. 198-201.
(161 Barth, P. W., Bernard, S. L., Angel], J. B., ZEEE Trans. Electron Devices ED-33 (1985) 1202-1205.
[17] Smith, R. L., Scott, D. C., ZEEE Trans. on BME BME33 (1986) 83-90.
1181 Watanabe, Y., Arita, Y.,Yokohama, T., Igarashi, Y., J. Electrochem.Soc. 122 (1975) 1351.
[19] Bomchil, G., Herino, R., Barla, K., in : Energy Beam-SolidInteractions and Transient Thermal Pro-
cesses, Nguyen, V. T., Cullis, A. (eds.); Les Ulis, France: Les Editions de Physiques, 1986, pp.
463-474.
[20] Petersen, K. E., ZBM Technical Disclosure Bulletin 21 (1979) 3768-3769.
4.9 Re$ewnces 105

[21]Horwe, R. T.,in :Micmmachining and Micmpackaging of l?unsducers, Fung. C. D., Cheung, P. W.,
Ko, W. H., Fleming, D. G. (eds.); Elsevier. 1985, pp. 169-187.
(221 Guckel. H.,Bums, D. W., I)ansducers 35, lkhnical Digest, International Codewnce on Solid-
State Sensors and Actuators, Philadelphia, June 1985, pp. 90-92.
[23]Jolly, R. D.,Muller, R. S., J. Electmchem. SOC. 127 (1980)2750.
[24]Howe, R. T.,Muller, R. S., Sens. Actuators 4 (1983)447-454.
[25]Ho, C. P.. Hansen, S. E., lkhnicalReport SEL 83-001, Stanford Electronics Laboratories, Stanford
University, July 1983.
[26]KO,W.H.,Suminto, J. T., Yeh, G. J., in: Micmmachining and Micmpackaging of nunsducers,
Fung, C. D.,Cheung, P. W., KO, W. H.,Fleming, D. G. (eds.); Elsevier. 1985,pp. 41-61.
[27]Smith, R. L., Collins, S. D., IEEE Zhzm. Electmn &vimED-35 (1987)787-792.
[28]Sze, S. M.,VLSI Technology, Wiley, 1981,Chapter 3.
[29]CRC Handbook of Chemistry and Physics, 51st edition; the Chemical Rubber Co., 1971.
[30]Bousse., L., Shott, J.. Meidl, I. D., IEEE Electmn Device Lett. EDL9 (1988)44-46.
[31]Nissan-Cohen, Y., Gorczyca, T.. IEEE Electron Device Lett., EDL9 (1988) 287.
[32]Pan. P.. Berry, W., J. Electmchem. SOC.I32 (1985)3001-3005.
[33]Hoffman, R. W.,Thin Solid Films 34 (1975)185-190.
[34]Van den Vlekkert, H.-H., Kloeck, B., Prongue, D., Berthoud. J.. Hu, B.. de Rooij. N. F., Gi,E.,
de Crousaz. Ph.. Sens. Actuators (1988)165-176.
[35]Huang, J. C-M.. Wise, K. D., IEDM Xkhnical Digest (1983)316-319.
[36]Petersen. K., Brown, J., Renken, W., lkhnical Digest, Proc. 3rd International Coderenre on Solid-
State Sensors and Actuators, Philadelphia, 1985, pp. 361-363.
[37]Wallis,G., Pomerantz, D. I., 1 Appl. Phys. 40 (1969)3946-3949.
[38]Albaugh, K. B., Cad% P. E., Rasmussen, D. H., IEEE Solid-state Sensor and Actuator Workshop,
Hilton Head, SC; June 1988, pp. 109-110.
[39]Petersen, K. E., Barth, P., Poydock, J., Brown, J., Mallon, J. Jr., Bryzek, J., IEEE Solid-state Sen-
sor and Actuator Workshop,Hilton Head SC, June 1988, pp. 144-147.
[40]Fung, C. D.,Cheung. P. W., and KO, W. H., IEEE Zhzns. Electron Devices, ED-33,(1986)8-18.
[41] Bergveld, P., IEEE i’?uns. Biomed. Eng. 17 (1970)70.
[42]Janata, J., Huber R. J., in : Zon SelectiveElectmdes in Analytical Chemistry,Freiser, H.(ed.) ; New
York: Plenum Press, 1980,Vol. 2, Chapter 3.
[43]Oberg. P., Ligensjo, J., Rev. Sci. InstK 30 (1959)1053.
[44]King, W.H. Jr., Anal. Chem 36 (1964) 1735.
[45]Gjessing, D.T.,Holm,C., Lanes, T.,Electma Lett. 3 (1967).
[46]Wohltjen. H., Jarvis, J. L., Snow, A., Giuliani. M.,Barger, W., PKJC.Int. Symp.Protection against
Chemical Wagaw Agents; Stockholm, 1985, p. 51.
[47]Tomita, Y., Guibault, G., Anal. Chem 52 (1980) 1484.
[48] Hlavay, J., Guibault, 0. G., Anal. Chem. 49 (1977)1890.
[49]Nieuwenhuizen, M. S., Barendsz, A. W., Sens. Actuators ll (1987)45.
[50]Chuang. C. T.,White, R. M.,Bernstein, J. J., IEEE Electron Device Lett. EDL3 (1982)145.
[51] Guibault, G. G., Jordan, J. M..CRC Crit. Rev. Anal. Chem. 19 (1988)1.
[52]Konash. P. L., Bastiaans, G. J., Anal. Chem. 52 (1980)1929.
[53]W e r e r , J. E.. Bastiaans G. J., AnaL Chem. 55 (1983)2333.
[54]Kanazawa, K. K.. Gordon. J. 0. II., Anal. Chem. 57 (1985)1771.
[55] Glassford, A. P. M.,J. fir. Sci. Technol. 15 (1978) 1836.
[56]Martin,S. J.. Ricco, A. J., Hughes, R. C., Xkhnical Digest, 4th Internat. Cod on Solid-StateSen-
sor and Actuators, Tokya Japan, June 1987, p. 478.
[57]Grabbe, E. S., Buck, R. P., Melroy, 0.R., J. Electroanal. Chem 223 (1987)67.
[58]Cammann, K., Working with Ion-Selective Electrodes; New York: Springer-Verlag. 1979.
[59]Buck, R. P., AMI. Chem. 46 (1974)28R.
106 4 Sensor Design and Packaging

[60] Moody, G. J., Thomas, J. D. R., in : Ion-Selective Electrode Methodology, Vol. 1, Covington, A.
K. (ed.); Boca Raton, FL: CRC, 1979, p. 111.
[all Wen, C. C., Lauks, I., Zemel, J. N., Thin Solid Films 70 (1980) 333.
[62] Pick, J., Toth, E., Pungor, M., Vasak and W. Simon, Anal. Chim. Acta 64 (1973) 477.
[63] Kawakami, S., Akiyarna, L., Ujihira, Y., Fresenius Z Anal. Chem. 318 (1984) 349.
1641 Mascini, M., Pallezzi, F., Anal. Chem. 73 (1974) 375.
(651 Oesch, U., Simon, W., Transducers '87, 4th International Conference on Solid-state Sensors and
Acutators; Tokyo, Japan, 1987, p. 755.
[66] Ebdon, L., Ellis, A. T., Corfield, G. C., Analyst 104 (1979) 730.
[67] Abe, H., Esashi. M., Matsuo, T., IEEE Trans. Electron Devices ED-26 (1979) 1939.
[68] Moriizumi, T., Thin Solid Films 160 (1988) 413.
[69] Pham, M. L., Hoffmann, W.,Sens. Actuators 5 (1984) 217.
[70] McBride, P. T., Janata, J., Comte, P. A., Moss, S. D., Johnson, C. C., Anal. Chim. Acta. 101 (1978)
239.
[71] Ho, N. J., Kratochvil, J., Blackburn, G. F., Janata, J., Sens. Actuators 4 (1983) 413.
[72] Blackburn, G. F., Janata, J., J. Electrochem. Soc. 129 (1982) 2580.
[73] Burns, B. E., Barth, P. W., Angell, J. B., Technical Digest of the IEDM (1984) 210-212.
[74] Schmidt, M. A., Haritonidas, J. H., Senturia, S. D., IEEE lhzns. Electron Devices ED-35 (1988) 755.
[75] Schmidt, M. A., PhD Dissertation, Massachusetts Institute of Technology, 1988.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

5 Semiconductor Integrated Circuit Technology


and Micromachining
WENH.KO. JAMES T.S U M m . Case Western Reserve University,
Cleveland. Ohio. USA

Contents
5.1 Introduction ........................... 109
5.2 I. C. Technology . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.2.1 Crystal Growth and Wafer Preparation . . . . . . . . . . . . . . . . 110
5.2.2 Oxidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5.2.3 Lithography . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.2.3.1 Mask Fabrication . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.2.3.2 Pattern Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.2.3.3 Resists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
5.2.4 Etching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.2.4.1 Wet Etching . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.2.4.2 Dry Etching . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.2.5 Diffusion and Ion Implantation . . . . . . . . . . . . . . . . . . . 125
5.2.5.1 Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.2.5.2. Diffusion Mask . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.2.5.3 Ion Implantation . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.2.6 Metalkation . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.2.6.1 Vacuum Evaporation . . . . . . . . . . . . . . . . . . . . . . . . 132
5.2.6.2 Sputtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
5.2.6.3 Chemical Vapor Deposition . . . . . . . . . . . . . . . . . . . . . 136
5.2.7 Assembly Techniques and Packaging . . . . . . . . . . . . . . . . . 136
5.2.7.1 Die Separation . . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.2.7.2 Die Attachment . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.2.7.3 Wire Bonding . . . . . . . . . . . . . . . . . . . . . . . . . . 137
5.2.7.4 Packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
5.3 Micromachining . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.3.1 Silicon Etching . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.3.1.1 Etching Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.3.1.2 Isotropic Etching . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.3.1.3 Anisotropic Etching . . . . . . . . . . . . . . . . . . . . . . . . 143
5.3.1.4 Etch Stop . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.3.2 Chemical Vapor Deposition . . . . . . . . . . . . . . . . . . . . . 151
5.3.2.1 Silicon Epitaxy . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.3.2.2 Dielectrics and Polysilicon . . . . . . . . . . . . . . . . . . . . . 152
108 5 Semiconductor Integrated Circuit Technology and Micromachining

5.3.3 Bonding Layers of Materials . . . . . . . . . . . . . . . . . . . . 155


5.3.3.1 Electrostatic Bonding . . . . . . . . . . . . . . . . . . . . . . . 155
5.3.3.2 Thermal Fusion Bonding . . . . . . . . . . . . . . . . . . . . . . 160
5.3.3.3. Low Temperature Glass Bonding . . . . . . . . . . . . . . . . . . 161
5.3.3.4 Metallic Bonding . . . . . . . . . . . . . . . . . . . . . . . . . 162
5.3.4 Connection between Layers . . . . . . . . . . . . . . . . . . . . . 162
5.3.4.1 Etched-through Hole . . . . . . . . . . . . . . . . . . . . . . . . 163
5.3.4.2 Laser Drilled Hole . . . . . . . . . . . . . . . . . . . . . . . . . 163
5.3.4.3 Thermomigration of Aluminum Column . . . . . . . . . . . . . . . 164
5.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.2 I.C. Technology 109

5.1 Introduction

The conventional sensor uses well-established machine shop and instrument manufacturing
technology. These will not be discussed in this book. Chapters 5 to 9 are aimed at surveying
and summarizing the new technology, material selection and fabrication methods of solid
state sensors developed in the last two decades. These include semiconductor microsensors,
as well as optical fiber, ceramic and chemical sensors using new materials or devices (such as
optical fibers). The standard integrated circuit (I.C.) technology for VLSI (very large scale in-
tegration) chips and some recently developed micromachining techniques are essential to the
fabrication of these new sensors. The reasons are:
(1) The photolithographic etching, deposition, metalization and assembly techniques for
I.C. production are not only essential for semiconductor microsensors but also the basis for
fabricating the structure of all other thin film, thick film, chemical and biological sensors.
(2) Many sensors, such as temperature, magnetic and flow sensors, use I.C. elements.
(3) The signal processing of sensor output will be electronic circuits. The trend is towards
integration of chips on the sensor, or packaging in a unit.
However, besides I.C. processes, the three-dimensional structure of most sensors will require
precise mircomachining techniques. These highly developed processes make possible new sen-
sors with: (a) better reliability, (b) mass production with high yield, thus reducing the cost,
and (c) better performance, including high sensitivity, uniformity, stability and ease of opera-
tion.
It is these advantages, as well as increased needs due to automation, that support new sensor
development and the growth of the sensor industry.
Chapter 5 summarizes integrated circuit (I.C.) technology and micromachining methods for
solid state sensor fabrication with selected references [l to 1001.
There are many volumes of information on VLSI technology published and easily available
throughout the world. Therefore, only a very condensed outline will be given and the reader
will be referred to specialized books at various levels for more detailed information [l to 161.
Micromachining techniques were originally developed for microsensors and are now also
used in VLSI manufacturing. These techniques have not been collected and published in
generally available book, except in some regional form (17) therefore more detailed outlines
are presented.

5.2 I.C. Technology

Figure 5-1 illustrates the major processing steps used in I.C. and semiconductor sensor
fabrication. Starting with polished semiconductor (Si or GaAs) wafers, the film formation
step includes: (1) epitaxial layer, (2) oxide film, (3) polysilicon and dielectric films, and (4)
metalization films. After film formation, the wafer often undergoes diffusion or ion implanta-
tion for impurity doping. Then a lithography process is used to transfer the pattern stored in
the masks to the film surface, which is then etched to remove unwanted film or substrate parts
to complete the pattern transfer process. This cycle is repeated many times, depending on how
many masks are in the set.
110 5 Semiconductor Integrated Circuit Technology and Micromachining

0 WAFER

MASK SET FORMATION

-- i ETCH1 NG

WAFER Figure 5-1.


OUT Major process steps in inta-
grated circuit fabrication. 111

An I.C. is made by sequentially transferring the patterns from masks, level by level, on the
semiconductor wafer and the deposited films. A finished wafer may contain hundreds or
thousands of identical I.C. chips (or dice) with varying shapes and sizes. These chips are
separated by a laser or diamond saw and then packaged into device units.

5.2.1 Crystal Growth and Wafer Preparation

The most common single crystal semiconductors used for I.C. and microsensors are silicon
(Si), germanium (Ge), and gallium arsenide (GaAs). For silicon, the starting material is SiO,
(sand or quartzite). It is reduced in a furnace with various forms of carbon to become
metallurgical-grade silicon. Then it is treated with HCl to form SiHCl, (trichlorosilane),
which is a liquid at room temperature, with a boiling temperature of 32°C. Purified by frac-
tional distillation, the SiHCl, is reduced by H, to form electronic grade silicon (EGS) in
polycrystalline form, with impurity in the range of lo-'. The polycrystalline EGS is then
grown into single crystal ingots of desired shape by either: (1) Czochralski technique or (2)
float zone process.
For GaAs, because Ga and As both have high vapor pressure at the melting temperature
of GaAs (1238"C), the material has to be grown in an evacuated sealed quartz tube system.
Because As has a higher vapor pressure than Ga, a two-temperature furnace is commonly used
to convert Ga and As into polycrystalline GaAs from a melt. The Bridgman technique is used
for growing single crystal GaAs. A seed crystal and a two-temperature furnace are needed in
this process.
In single crystal growing, the major precaution is to maintain high purity of the grown
ingot. Techniques, such as zone refining, were developed to refine the ingot to high purity.
5.2 LC. Ethnology 111

The ingot is ground to the desired diameter and with flat regions along the ingot to mark
the crystal orientation and conductivity type The ingot is sliced by a diamond saw into wafers.
After lapping and polishing to prepare the surfaces, the wafers become the starting material
of I.C.fabrication processes.
The identifying flats of a silicon wafer arc shown in Figure 5-2. The primary flat (larger
flat) is for the alignment of the wafer to the desired I.C.or device orientation, or to the scribe
and break direction. The secondary flat (smaller flat) is used to identify the type of silicon.

PriflUry
llar
f I 10) plane

flat
n-typeflll) ptypc ( I I I )

Primary
b t
( 1 10) plane

Flgure 5-2. Primary and secondary flat locations on silicon wafers.

The typical specifications for polished silicon wafers are given in Table 5-1. The material
properties and requirements for VLSI are given in 'Tbble 5-2. Other parameters may be of in-
terest, including:
- Mechanical. warp, edge contour, chips, indentations, pits, laser markings
- Surface: flatness, haze, stains, streaks, scratches, swirl.
For detailed description and testing methods see references [4] and [5].
Besides grown wafers Si and GaAs can be grown on various substrates by epitaxial pro-
cesses, including:
- vapor phase epitaxy,
- liquid phase epitaxy, and
- molecular beam epitaxy (MBE) and metalorganic chemical vapor deposition (MOCVD).
112 5 Semiconductor Integrated Circuit Technology and Micromachining

lhble 5-1. Specifications for Polished Single-Crystal Silicon Wafers [I]

Parameter 100 mm 125 mm 150 mm


Diameter (mm) 100 f 1 125 * 1 150 f 1
Thickness (mm) 0.5-0.55 0.6-0.65 0.65-0.7
Primary flat length (mm) 30-35 40-45 55-60
Secondary flat length (mm) 16-20 25-30 35-40
Bow (pm) 60 60 60
Total thickness variation (pm) 50 65 50
Surface orientation (100) f 1 " Same Same
(111) f l o Same Same

lhble 5-2. Material Properties and Requirements for VLSI of Silicon Wafers [I]
~~ ~

Property Czochralski Float zone Requirements for VLSI


Resistivity (phosphorus) n-type (a cm) 1-50 1-300 and up 5-50 and up
Resistivity (antimony) n-type (0 cm) 0.005-10 - 0.001-0.02
Resistivity (boron) p-type (0cm) 0.005-50 1-300 5-50 and up
Resistivity gradient (four-point probe) (Yo) 5-10 20 <I
Minority carrier lifetime (ps) 30-300 50-500 300- 1000
Oxygen (ppma) 5-25 Not detected Uniform and controlled
Carbon (ppma) 1-5 0.1-1 <0.1
Dislocation (before processing) (per cm2) Q 500 < 500 Q1
Diameter (mm) u p to 200 u p to 100 Up to 150
Slice bow (pm) Q 25 Q 25 <5
Slice taper (pm) Q 15 Q 15 <5
Surface flatness (pm) Q5 Q5 <I
Heavy-metal impurities (ppa) Ql Q 0.01 < 0.001

Single crystal and polycrystalline Si has been grown on the surface of insulators such as sap-
phire (silicon-on-sapphire, SOS), SO,, and other materials. GaAs also can be grown on the
top of other single crystal substrates and silicon by epitaxial processes.
The MBE/MOCVD is becoming an established technology to grow layers of single crystal
material on various substrates and to grow and support lattice crystals. These materials are
important for GaAs heterojunction high speed devices and optical sensors [13- 171.

5.2.2 Oxidation

Oxidation of Si wafers is generally the starting process of I.C. fabrication and it serves as:
- passivation of crystal surface,
- diffusion and ion implantation masks,
- dielectric for integrated MOS capacitors.
Silicon surfaces exposed to the air will grow an oxide between 20 to 50 A thick. Thicker
oxides can be thermally grown in dry oxygen or a water vapor environment.
5.2 I.C. Technology 113

Figure 5-3 plots the experimental data of SiO, thickness as a function of time and
temperature in (a) dry 0, and (b) steam vapor environment. If x is the thickness of oxide, t
is the oxidation time, and at t = 0 there is an oxide thickness, do, represented by T = f(d,-,).
Then,
x2 + A x = B ( t + T) . (5-1)

OXIDATION TIME (h)


(01

ul
w
2
Y
Y
I
I-

Figure 5-3.
SiO, thickness as a function of time
OXIDATION TIME (h)
and temperature in (a) dry 0, and (b)
steam vapor environments. (b)

In order to reduce processing temperature and oxidation time, high pressure oxidation (up
to 3 MPa) has been used.
During oxidation the dopant impurity near the silicon-oxide interface will be redistributed
because the segregation coefficient of impurity in Si and SO,, k, is not unity. For boron
doped Si the impurity in the silicon interface is depleted, while for phosphorous doped Si there
is a buildup at the interface.
The colors of various oxide thicknesses are given in Table 5-3. These are not accurate
measurements of thickness but are generally used as a quick check on oxide reproducibility.
114 5 Semiconductor Integrated Circuit Technology and Micromachining

Table 5-3. Observed Colors for Different Oxide Thicknesses for


Perpendicular Illumination with White Light [2]

Thickness (A)
Color 1 2 3 4
Grey 100
Tan 300
Brown 500
Blue 800
Violet lo00 2800 4600 6500
Blue 1500 3000 4900 6800
Green 1800 3300 5200 7200
Yellow 2100 3700 5600 7500
Orange 2200 4000 6000
Red 2500 4400 6200

5.2.3 Lithography

Lithography is the technique of transferring patterns of geometric shapes of a design to a


layer of radiation sensitive material (resist) which, in turn, transfers the patterns to the
underlying films or substrates through etching processes. Depending on the radiation used,
one has: photolithography, X-ray lithography, E-beam lithography, and ion beam lithography.
The major techniques used in lithography are: (1) fabrication of mask or pattern generation,
and (2) transfer of the pattern from the mask to the wafer. Figure 5-4 shows the steps or flow
diagram of’ photolithography commonly used by semiconductor device and microsensor
design research laboratories.
Recent VLSI and microsensor manufacturing uses computer aided design and pattern
generation, or E-beam direct writing to fabricate master masks, eliminating the hand drawing
or Rubylith cutting procedure.

5.2.3.1 Mask Fabrication

After a test design or computer simulation of a circuit or a sensor is completed, the com-
posite design of the layout [3, 101 is divided into mask levels, each corresponding to a subse-
quent processing step that requires a specific geometric pattern to be transferred to the wafer
or films on the wafer.
The specific geometrical layout of each mask level is converted into patterns on a mask by
hand or computer. Refer to Figure 5-4 for manual mask fabrication. An enlarged pattern (200
to 400 times the actual size) is drawn on a plastic laminate (Rubylith) consisting of a dimen-
sionally stable Mylar sheet bonded to a thin veneer of ruby color plastic. Cutting along the
outline of the pattern is then made manually or automatically by a computer-controlled
facility (coordinatograph). The ruby color film on the proper area is peeled off to form the
desired pattern of clear and red regions. This is the original Rubylith pattern, which is reduced
5.2 I.C. Technology 115

Pattern Drawing
x200 x400

Photoreduction Pattern Generator

1 Reticle Mask
x10

Step-and-Repeat

I Master Mask

I
XI

Working Mask

Proximity/
Contact Print Projection Printing

Figure 5-4.
Manual mask fabrication processes.
Wafer (11>
photographically (usually about 10 : 1 to 20 : 1) on a first reduction photographic facility to
form a glass reticle mask with high resolution, high contrast emulsion. For complex circuits
and sensors, the manual technique becomes impractical, and computer aided design and
automatic pattern generators which are commercially available are used to generate the reticle
mask or master mask directly.
The reticle mask is again reduced to its final size and duplicated on the master mask by a
step-and-repeat camera which is essentially an inverted microscope. Because each circuit is
generally 1 to 10 mmz in size, a wafer of 75 mm or 100 mm diameter can have many units
of the circuit made on it. Therefore, the mask may contain multiple images of the circuit. This,
then, is the master mask.
Contact prints of the master mask or submaster mask are used as working masks. The
working mask will be subject to wear and tear during lithographic processes. For contact
aligners, a working mask can only have a lifetime of 5 to 30 cycles. Besides photo-emulsion
116 5 Semiconductor Integrated Circuit Technology and Micromachining

on glass, other hard materials such as chrome, iron-oxide, etc. are etched to make masks. These
can last longer and have better performance, but are more expensive.

5.2.3.2 Pattern Transfer

There are two optical exposure methods: (a) shadow printing, and (b) projection printing.
For shadow printing, the mask and wafer may be either in direct contact or in close proximity.
In either case, the minimum line width that can be printed (transferred) is:

,z = (5-2)

where L is the wavelength of the light and g is the gap between the mask and the wafer surface.
For I = 0.4 pm, gap = 15 pm, minimum line width is 2.4 pm. For projection printing (usually
a ratio of 1 : 1 to 10 : l), the resolution is:

,
Z = A/NA; N A = n sin a (5-3)

where N A is the numerical aperture of the optic, n, the index of refraction in the medium (for
air n = l), and a is the half angle of the cone of light converging to a point image at the
wafer. The depth of focus is:

AZ = fn1/2(NA)’. (5-4)

In both cases, the minimum feature size decreases with wavelength, therefore for VLSI
ultraviolet, X-ray and E-beam are used to achieve smaller feature size.
For lithography other than optical, the readers may read references [l-91; references [l-31 are
at the undergraduate level and [5-91 at graduate student level.

5.2.3.3 Resists

Photoresist can be classified as positive and negative. For positive resist, the clear region
of a mask is exposed and becomes soluble and is removed during the development process.
Therefore, the pattern formed on the positive resist is the same as that on the mask. For
negative resist, the reverse is true. The pattern on the negative resist is the reverse of the mask
pattern. Figure 5-5 illustrates the steps of the transfer process from a mask to a wafer with
an insulating film on its surface.
A few commonly used resists are listed in Table 5-4 where y is the contrast ratio of the resist.
Larger y implies more rapid solubility of the resist with increased exposure, resulting in a
sharper image. The photoresist is not sensitive to lights with wavelength greater than 0.5 pm.
It is processed in a clean room with yellow light. Figure 5-6 shows the typical procedure of
pattern transfer. The wafer is placed on a spinner. The resist is applied to the center of the
wafer, then the wafer is accelerated to a constant rotational speed (lo3 - lo4 rpm), and
maintained for a time of 15-30 seconds to give a uniform resist film of the desired thickness.
After pre-baking (80 - lOOOC), the wafer is processed in an aligner or other lithographic
5.2 LC ikhnologv

POSITIVE RESIST NEGATIVE RESIST

1 I

FIgnre 5-5.
"hnsfer process from a mask to a
I
wafer with an insulating film on its
surfacc [l]

'Ibble 5-4. Negative and Positive Resists [l]

Lithography Name m Sensitivity Y


Optical Kodak 747 Negative 9 mJ/cm2 13
Az1350J Positive 90 mJ/cm2 1,4
PR 102 Positive 140 mJ/cm2 1.9
e-Beam COP Negative 0,3 @cm2 0,45
Gese Negative 80 pC/cm2 395
PBS Positive 1 pc/cm2 0,35
PMMA Positive 50 pC/cm2 1.0
x-Ray COP Negative 175 mJ/cm2 0,45
DCOPA Negative 10 mJ/cm2 0.65
PBS Positive 95 mJ/cm2 0.5
PMMA Positive lo00 mJ/cm2 1.0

system to be exposed to ultraviolet or other radiation. The exposed resist is developed and
post-baked (100 - 18OOC). The wafer is etched to transfer the pattern on the resist to the
underlying films (Si02, metal, etc) or substrates. After that, the resist is stripped. A
lithographic cycle is, then, completed.
118 5 Semiconductor Integrated Circuit Technology and Micromachining

WAFERS INSPECT
AND
T MEASURE

DEVELOP
RINSE AND DRY
STRIP
RESIST t
1 MASKi-1 PRINTER
ALIGN
EXPOSE
OR GROW

+ INCOMING^ I -
PRE BAKE
OVtN I
I
WAFERS APPLY
PHOTORESIST

L PHOTORESIST ROOM
Figure 5-6.
Typical procedure of pattern transfer. [7]

An alternative to etching for transfer of patterns from resist to films is the "lift-off" pro-
cess. The substrate is coated with positive resist and a pattern is defined. Then the metal (or
other film) is deposited, the photoresist is stripped, and the metal or film on top of the resist
is lifted off with the resist, leaving the metal (or films) on the desired area where there is no
positive resist.
In addition to chemical strip agents, oxygen plasma strippers are used extensively, typically
at 1.3 kPa for 30 minutes. The change of plasma color can be used to detect the end of the
stripping process. More about plasma etching is given in Sections 5.2.4 and 5.3.1.

5.2.4 Etching

Etching is used extensively for: (1) removing surface damage such as occurs in the polishing
process, (2) cleaning the surface to remove contamination prior to other processes, (3) delin-
eating patterns and opening windows in insulating materials, and (4) fabricating 3-dimen-
sional structures such as in micromachining, which will be discussed in Sections 5.3 to 5.3.3.
Both wet chemical etching and dry plasma etching are used.
The materials to be etched include: (1) silicon, GaAs semiconductors, (2) metals conductors,
and (3) insulators. Wet etching of silicon will be discussed in more detail in Section 5.3.1.

5.2.4.1 Wet Etching


The mechanism for wet etching involves three essential steps: (1) the reactants are transpor-
ted (e. g., by diffusion) to the reaction surface, (2) chemical reactions occur at the surface, and
5.2 I. C. Technology 119

(3) the products from the surface are transported away (e. g., by diffusion). Both agitation and
temperature of the etchant solution will influence the etch rate.
Most of the dielectrics and metals used in microcircuit fabrication are either vitreous, amor-
phous, or polycrystalline in nature. They lack long-range order, so that etching by wet
chemicals is usually isotropic in character, i. e., the etchant spreads out under the mask layer
by an amount roughly equal to the etched depth. In many instances, stresses at the resist-film
interface, combined with capillary action of the wet chemical, can cause excessive undercut-
ting at this point, and even lifting or tearing of the resist. Excessive undercutting can also
occur because of stress between the film and the semiconductor.
A wide variety of etching chemicals are available for each material used in I.C. fabrication.
Their characteristics will depend upon such film parameters as its microstructure, its porosity,
how it is formed, and the nature of previous processes to which it has been subjected. A brief
listing of useful etchants for dielectrics and metals is provided in Tables 5-5, 5-6,5-7and 5-9
[I,5 , 7, 9, 131.

Table 5-5. Oxide Properties and Etch Rates

Dielectric Strength Etch Rate in Buf-


Resistivity (a cm)
Density (g/cm3)
lo6 V/cm fered HF (Ah)*
Dry 0 2 2.24-2.27 3 x lOI5 - 2 x 10I6 2 6.8
Wet 0, 2.18-2.21 6.7
Steam (1 stm) 2.0-2.20 1015 - 1017 6.8-9 1.3
SiH,/CO, 2.1 -2.24 2.06-2.92**

* Buffered HF etch: 1 part 48% HF, 10 parts NH,F solution (1 lb NH,F/680 cc H,O).
** P-etch: 3 parts 48% HF, 2 parts 70% HNO,, 60 parts H,O.

Table 5-6. Etchants for Noncrystalline Films [6]

Material Ers.hant Remark


SiO, 28 ml HF BHF, 1000-2500 &min at 25°C
170 ml H,O
113 g NH,F
15 ml HF P-etch, 128 A/min at 25°C
10 ml HNO,
300 ml H,O
1 ml BHF 800 A/min
7 ml H,O
BSG 1 ml HF R-etch, 300 A/min for 9 mole Vo . B,O,, 50 A/min for SiO,
100 ml HNO
100 ml H,O
4,4 ml HF S-etch, 750 A/min for 9 mole % * B,O,, 135 A/min for SiO,
100 ml HNO,
100 ml H,O
120 5 Semiconductor Integrated Circuit Ethnology and Micromachining

lsble 5-6. Continuation

Material Etchant Remark

PSG 28 ml HF BHF, 5500 A/min for 8 mole Yo . P,O,


170 ml H,O
113 g NH,F
15 ml HF P-etch, 34000 A/min for 16 mole 9'0 P30,, 110 A/min for SiO,
10 ml HNO,
300 ml H,O
1 ml BHF 800 A/min
7 ml H,O
Si,N, HF 140 A/min, CVD at 1100"C, 750 A/min, CVD at 900°C
1000 A/min, CVD at 800°C
28 ml HF BHF, 5-10 A/min
170 ml H,O
113 g NH,F
H3PO4 100 A/min at 180°C
Polysilicon 6 ml HF 8000 A h i n , smooth edges
100 ml HNO,
40 ml H,O
1 ml HF 1500 A/min
26 ml HNO,
33 ml CH,COOH
SIPOS 1 ml HF 2000 A/min for 20% 0,film
6 ml H,O
10 ml NH,F (40%)
A1 1 ml Hcl SOT, fine line, can be used with gallium arsenide
2 ml H,O
4 ml H,PO, 350 A/min, fine line, will attack gallium arsenide
1 ml HNO,
4 ml CH,COOH
1 ml H,O
16-19 ml H,PO, 1500-2500 A/min, will attack gallium arsenide
1 ml HNO,
0-4 ml H,O
0.1 mol K2Br,0, 1 pm/min, pH 13.6, no gas evolved during etching
0.51 mol KOH
0.6 mol K,Fe(CN),
Au 3 ml HCI Aqua regia, 25-50 pm/min
1 ml HNO,
4 g KI 0.5-1 pm/min, can be used with resist
1 g 1,
40 ml H,O
5.2 LC. Technology 121

Table 5-6. Continuation

Material Etchant Remark

Ag 1 ml NH,OH 3600 A/min, can be used with resists, must be rinsed rapidly
1 mi H,O, after etching
4 ml CH,OH
Cr 1 ml HCI 800 A/min, needs depassivation
1 mi glycerine
1 ml HCI 800 A/min, needs depassivation
9 ml saturated
CeSO, solution
1 mi, 1 g NaOH in 250-1000 A/min, no depassivation, resist mask can be used
2 ml H,O
3 mi, 1 g K,Fe(CN),
in 3 mi H,O
Mo 5 mi H,PO, 0.5 pm/min, resist mask can be used
2 ml HNO,
4 ml CH,COOH
150 mi H,O
5 ml H,PO, Polishing etch
3 mi HNO,
2 ml H,O
11 g K,Fe(CN), 1 pm/min
10 g KOH
150 ml H,O

W 34 g KH,PO, 1600 A/min, high resolution, resist mask can be used


13,4 g KOH
33 g K,FE(CN),
H,O to make 1 liter
Pt 3 mi HCl Aqua regia, 20 pm/min, precede by a 30 s immersion in HF
1 mi HNO,
7 mi HCl 400-500 A/min, 85 "C
1 mi HNO,
8 mi H,O
Pd 1 ml HCI 1000 A/min
10 ml HNO,
10 mi CH,COOH
4 g K1 1 pm/min, opaque, must be used before visual inspection
1 g 1,
40 mi H,O
122 5 Semiconductor Integrated Circuit Technology and Micromachining

Table 5-7. Typical Etch Rates and Selectivity for Some Dry Etching Process for VLSI [7]

Selectivity
Etched Material (M) Gas Etch Rate (A/min)
Mhesist M/Si M/SiO,
Al, A M , Al-Cu BC1, + C1, 500 5-8 3-5 20-25
Polysilicon c
12 500-800 5 - 25-30
SiO, CF, + H, 500 5 20 -
PSG CF, + H, 800 8 32 -
GaAs CCI, + 0, 6000 - - -
Si SF, + C1, 1000 - 4500 5 - 80

5.2.4.2 Dry Etching

Dry etching can take on several forms:


(1) Physical etching:
- sputtering etching
- ion milling
(2) Physical and chemical etching:
- reactive plasma etching
- reactive ion etching
- reactive ion beam etching
A plasma is a fully or partially ionized gas composed of ions, electrons, and neutral atoms.
A plasma is produced when an electric field of sufficient magnitude is applied to a gas, caus-
ing the gas to break down and become ionized. The plasma is initiated by free electrons that
are released by some means such as field emission from a negatively biased electrode. When
the applied voltage is larger than the breakdown potential, a sustained discharge is formed
throughout the reaction chamber. The electron concentrations in the plasma for dry etching

_..-.-.-.-.-._.)
PLASMAS FLOWING GAS

f
( 1 ) GENERATION OF ETCHANT SPECIES

(2) DIFFUSION TO SURFACE

f
----- --------------- ---- I
STAGNANT GAS LAYER

i
(4) REACTION i

Figure 5-7.
Basic steps in a dry-etching
processing. [l]
5.2 I. C. Technology 123

are relatively low, typically on the order of lo9 to 10l2cm -3. At a pressure of 1 Torr, the con-
centrations of gas molecules are lo4 to lo7 times higher than the electron concentrations.
Typical electron temperatures of 10000°K are achieved, although the gas thermal temperatures
are only 50 to 100°C. A plasma can thus be considered as an ensemble of highly reactive par-
ticles in a relatively cool medium. The possibility of independently controlling particle
energies and gas temperature gives rise to more freedom in adjusting process parameters. It
is these reactive particles, and their characteristics, that can be used effectively in dry etching.
The reactive plasma etching process proceeds in five steps as illustrated in Figure 5-7.
(1) The process begins with the generation of the etchant species in the plasma. (2) The reac-
tant is then transported by diffusion through a stagnant gas layer to the surface. (3) The reac-
tant is adsorbed on the surface. (4) This is followed by chemical reaction (along with physical
effects such as ion bombardment) to form volatile compounds. ( 5 ) These compounds are
desorbed from the surface, diffused into the bulk gas, and pumped out by the vacuum system.
An example is shown in the following equations in which relatively inert CF, gas is
energized by electrons and dissociated into the active fluorine species that will dissolve away
Si,N, layers.
CF, + e- -+ CF;+ F + 2e- (5-5)
12 F + Si,N, + 3 SiF, + 2 N,. (5-6)

The fluorine free radical is the primary active species which etches the Si3N, by converting
it to N, and SiF,, which are both stable, volatile reaction products.
Dry etch systems vary widely according to the type of reaction used. Barrel etch configura-
tions (Figure 5-8a) stack wafers vertically and bleed gas through the tube-shaped chamber
with rf potential on top and bottom surfaces. Perforated sleeves are added to protect the
wafers from energetic ions and electrons and to equalize the activity of plasma reactions and
increase etch uniformity. The planar plasma reactor (Figure 5-8b) uses two parallel plates as
electrodes. Basic plasma etch configuration has the samples on the ground plate (Figure 5-8 b),
while reactive sputtering and reactive ion etch configurations (Figure 5 . 8 ~ )place the wafer on
the rf electrode. There are also etch configurations in which rf is applied to a gas mixture to
extract special species while wafers are positioned downstream (Figure 5-8 d). Wafers kept on
the grounded electrodes are subjected to minimal radiation damage, while placing them on the

RF ,-Gas inlet rReactive


.. I

r G a s inlet

Exhaust Exhaust
Barrel etcher
Figure 5-8. (a 1
124 5 Semiconductor Integrated Circuit Technology and Micromachining

[
Wafers on grounded
anode in plasma mode
ToRF
A f Insulator

Cooled Gas inlet


electrodes

Exhaust
Plasma etcher

Wafers on powered
cathode in R I E mode

Gas inlet Cooled


electrode

Exhaust
(C) Reactive ion etcher

Gas in

Exhaust
Ion milling design. Also used
(d)
for reactive ion beam etching.

Figure 5-8. Plasma etch reactor configurations. (a) barrel etcher (b) plasma etcher (c) reactive ion
etcher (d) ion milling machine
5.2 I. C. Technology 125

powered electrode involves much higher ion bombardment levels and an increasing potential
for radiation damage to the device. In all types of equipment approaches, the concept is
basically the same: subject the wafer to a field of chemically active species that will combine
with the film to be removed and volatize it away. Secondly, the role of the plasma in an etching
chamber is to create energetic electrons and ions that will also react physically with the wafer
surface. The energetic ions play the role as catalyst for more gas-surface chemical reaction.
In standard plasma chemical reactions, without the benefit of the physical action of highly
energized particles, the process is isotropic (or lacks directionality). Highly energized ions
create anisotropy (or directionality) and add new chemical reactions. The flux of energetic
particles will also accelerate the rate of a standard plasma chemical reaction, thereby increas-
ing the wafer throughput. Energetic ion enhanced plasma chemistry is technically what
separates plasma etching processes from reactive ion etching (RIE). Usually both modes are
available in commercial equipment.
Photoresist is often removed by placing the resist-covered wafers in an oxygen plasma. This
is often called plasma ashing. A barrel type plasma system (Figure 5-8a) is commonly used
for this purpose. A large number of energetic species are generated in this plasma. Of these
species, perhaps the most predominant are formed by

e + 0, 2 0 + e-
-+o+o-
-+

(5-7)

Temperatures as low as 40-50°C are sufficient to cause oxidation (or burning) of the resist
by these free radicals. The reaction products consist mostly of water, carbon monoxide, and
carbon dioxide.
One parameter of considerable importance in pattern delineation in I.C. technology is the
selectivity of an etching process. Selectivity is defined as the ratio of etch rates between the
material to be etched and the mask material.
In order to increase the selectivity of plasma etch of SiO, over Si, hydrogen gas is added
to the CF, plasma. It results in the subsidiary reaction
2F + H, -+ 2 HF, (5-8)
thus suppressing the fluorine concentration in the system, and hence the etch rate of silicon.
Using this approach, etch ratios for SiO, : Si of as large as 35 have been achieved. An addi-
tional advantage here is that this plasma has no oxygen so that photoresists are relatively unat-
tacked by it.
Table 5-7 shows typical etch rates and some selectivities for the dry-etching processes [ 5 , 6 ,
111. For aluminium etching, the native oxide (approximately 30 A) must be removed first by
sputtering or chemical reduction before dry etching can proceed.

5.2.5 Diffusion and Ion Implantation

An essential requirement of the planar I.C. process is the ability to introduce controlled
quantities of dopant impurity atoms into selected regions of the wafer. Diffusion and ion im-
plantation are the two key processes used to fulfill this function. They are used to dope selec-
tively the semiconductor substrate to produce either an n- or p-type region. Selectivity is pro-
vided by a mask on the top surface of the wafer through which the impurities are introduced.
126 5 Semiconductor Integrated Circuit Technology and Micromachining

5.2.5.1 Diffusion

Diffusion of impurities is typically done by placing semiconductor wafers in a high


temperature furnace and passing an inert carrier gas that contains the desired dopant through
it. The temperature usually ranges between 800 to 1200°C for silicon.
For diffusion in silicon, boron is the most popular p-type dopant, while arsenic and
phosphorous are n-type dopants. These three elements have solubilities above 5 x lozocm - 3
in silicon in the diffusion temperature range. These dopants can be introduced in several ways,
including solid sources (e.g., BN for boron, As,O, for arsenic, and P,O, for phosphorous),
liquid sources (BBr,, AsCl,, and POCl,), and gaseous sources (B,H,, ASH,, and PH,).
Usually, the source material is transported to the semiconductor surface by an inert carrier
gas (e.g., N,) and is then reduced at the surface. An example of the chemical reaction of a
solid source is

2 P20, + 5 Si -+ 4P + 5 SO,. (5-9)

In this reaction, an oxide layer is formed on the silicon surface.


There are two major steps in the diffusion, namely: deposition and drive-in. Deposition,
which is the first step, is also known by: pre-deposition, dep, or pre-dep. During the deposition
step, impurity atoms are transported from a vapor source onto the semiconductor surface and
diffused into a semiconductor wafer. The vapor source maintains a constant level of surface
concentration during the entire diffusion period. The amount of atoms that enter the water
surface is limited by the solid solubility of the dopant in the wafer. Solid solubility of common
dopant elements in silicon and at different temperature is shown in Figure 5-9.
The second step is the drive-in, diffusion, or reoxidation, where, after deposition, the wafer
is heated up in a diffusion furnace, usually at a higher temperature, with an oxidizing or inert
carrier gas to redistribute the dopant atoms in the wafer to reach a desired depth from the
surface.
There is usually a sub-step in between deposition and drive-in which is called deglaze. The
wafers coming out of deposition have a thin oxide on the silicon that is highly doped with
whatever dopant was used. This doped oxide, if left on, could serve as a diffusion source dur-
ing drive-in and change the intended concentration. The deglaze step consists of an HF dip
of the wafers to remove the unwanted oxide and dopant source.
The basis of diffusion theory is Fick’s Law. For the deposition step, the surface concentra-
tion, C,, ,is maintained constant, the initial condition is C(x,O) = 0, at t = 0; the boundary
conditions are: C (0, t) = C,, ; C (00, t) = 0. Then the distribution function or diffusion pro-
file is:

C ( x , t ) = C,, erfc [ x / 2 Dt] (5-10)

where the erfc is the complementary error function,

erfc y = 1 -- 2 e-X2 cix.


fi 0
5.2 I.C. Technology 127

I-

10"' I I I 1 I 1
800 900 lo00 1100 1200 1300 1400
r, t a n ~ m n urc)
~
Figure 5-9. Solid solubility of common dopant elements in silicon at different temperatures. [2]

For the drive-in step, the total dopant in the wafer, S, is constant and the diffusion profile
is then,
S x2
2
- (5-11)
c (x, t ) = -e - 4 0 1
~ = C, e- L t

where S = { C(x, t ) dw and C, = c ( 0 , t ) = S / f i Dt.


0
128 5 Semiconductor Integrated Circuit Technology and Micromachining

For two step diffusion, if the D, and t, are the diffusion constant and time for the deposition
and D2 and t, are those for the drive-in and D, t, 4 D2 t, then the final concentration profile
is [2]:
C(x,t , t,) = (2 CslIz)(Dl t1/D2t2)'12exp [-x2/4 D, t,] . (5-12)

In most practical cases, this simple theory does not describe the actual profile accurately.
In other words, D is not a constant, it depends on C (x, t), the presence of other dopants, and
the mechanism of dopant interactions with defects in the wafer. Figure 5-10a and Figure 5-lob
show the experimental profile of As, B and P in silicon.

5.2.5.2 Diffusion Mask


Selective diffusion requires a mask that is both impervious to the dopant and can withstand
the high temperature necessary for diffusion. A suitable mask for silicon is silicon dioxide
(SiO,) which is easily grown or deposited. In addition, it can withstand the high temperature,
is easily patterned using photolithographic techniques, and the diffusion coefficient for the
usual silicon dopant is several orders of magnitude lower in silicon dioxide than in silicon.
Oxide masking is important because it is the basis of present-day integrated-circuit technology.
If we etch windows in the oxide and use the remaining oxide as mask, we can incorporate
dopant impurities into a silicon substrate in selective areas to form p-n junction region.

Figure 5-10a.
Normalized diffusion profiles for
arsenic and boron in silicon. The erfc
distribution is shown for comparison.
PI
5.2 I. C. Technology 129

Figure 5-10 b. 4000%

I
Phosphorous diffusion profiles for 4hr
arious surface concentrations after 10'5
diffusion into silicon for 1 h at 0 i.0 1 D
1OOo"C. 111 DEPTH ( p m )

Figure 5-11 shows the minimum thickness (d) of dry oxygen-grown silicon dioxide required to
effectively mask against phosphorous and boron as a function of temperature and time.

5.2.5.3 Ion Implantation


Ion implantation is a technique for introducing a layer of impurities just below the surface
of the host material, in this case, silicon, by bombarding it with a beam of ions whose energy
is in the range of one to several hundred keV. Among the advantages of ion implantation are:
(1) it offers precise control over the number of impurities introduced into the silicon; (2) it is
a low temperature process; (3) impurity layers can be introduced completely below the surface;
(4)there is a wide choice of mask material, such as photoresists, oxides, nitrides, and any
material that can collide with the incoming ions; and ( 5 ) implanted junctions can be made self-
aligned to the mask edge. The disadvantages are that complex, expensive machinery is required,
the junctions are not automatically passivated, and damage of the crystal lattice occurs.
During the ion implantation, ions arrive at the wafer surface with a spread of momentums
and experience different collision histories; therefore, they come to rest in the wafer at varying
distances. A typical distribution of ions implanted in amorphous materials is illustrated in
Figure 5-12a and Figure 5-12b. The total number of ions and their distribution are approx-
imated by the Gaussian distribution:
n(x) = (S/2ARp) exp [-(x - R p ) 2 / 2 A R p 2 ] (5-13)
130 5 Semiconductor Integrated Circuit Technology and Micromachining

Figure 5 4 .
Minimum thickness ( d ) of dry oxygen-
grown silicon dioxide required to
effectively mask against phosphorous
03 and boron as a function of temperature
t (min) and time. [l]

HIGH -VELOCITY

MASK
(b)

Figure 5 - 0 .
I I Comparison of (a) diffusion and
X (b) ion implantation. [l]

where S is the ion dose per unit area; Rp is the projected range, that is, the distance of the
peak concentration below the surface; and ARP is the projected straggle, at (x - R,) =
& AR,,, n (x) is 60% of the peak value (at x = R,,). At the axis perpendicular to the direc-
5.2 I. C. Technology 131

tion of incidence, x, the distribution is also a Gaussian of the form exp ( - y 2 / 2 A R j ) , A Ry


is the lateral straggle. The dose, S (ions/cm2), is the ion implantation beam current density
(ions/cm2 sec) times the implant time (sec).
In single crystal material, a phenomenon called ionchanneling is observed when the incident
beam is aligned with a low index crystallographic direction with little opportunity for colli-
sion. The channeled ions can penetrate many times deeper than the equation indicates. To
avoid channeling, the substrate is usually misoriented by an angle of 7 to 10" so that the ion
beam is incident along the random direction in the crystal. Another way to avoid channeling
is to create an amorphous surface layer using inert ions (e.g., Ar ions) prior to the dopant im-
plantation. Even with the misalignment, an exponential tail in the concentration profile exists.
Figure 5-13a and Figure 5-13b plot the projected range, Rp, and straggles A R p and A R y as
functions of ion energy for common ions implanted in Si and GaAs.
Ion implantation is a mechanical process and there is crystal damage resulting from the im-
pact of the ions. The amount of damage is in proportion to the implant energy, ion weight,
and dose. Longer, deeper implants cause more damage, as do heavier dopant atoms. Apart
from substrate damage, implanted ions may not be on the proper substitutional lattice sites
and so may not all be ionized, electrically-active donors or acceptors. Some of the damage,
and a high proportion of the trapped carriers can be restored by a high temperature step called
annealing. Typical annealing processes take place at 1000°C in nitrogen or hydrogen at-
mosphere. Laser pulse annealing is another method used. Rapidly pulsing the wafer surface

Figure 5-Wa.
Projected range for B, P, and As in
Si and SiO, at various energies. The
results pertain to amorphous silicon
target and thermal SO,.[l]
132 5 Semiconductor Integrated Circuit Technology and Micromachining

Figure 5-I3 b.
Ion straggles R , (projected)
I0 100 and R (lateral) for As, P,
ENERGY ( kev) and B ions in silicon. [l]

to the required temperature restores the crystal structure without causing impurity redistribu-
tion.

5.2.6 Metalization

Metalization refers to the formation of metal films used for interconnection, ohmic con-
tacts, and rectifying metal-semiconductor contacts. Metal films can be formed by: vacuum
evaporation, sputtering, and chemical vapor deposition.

5.2.6.1 kcuum Evaporation

Vacuum evaporation is the most widely used deposition process. Both conducting metal and
resistive films can be deposited. The process is usually carried out in a high vacuum bell jar,
-
typically 1.3 . lo-* to 1.3 lo-’ Pa. A typical setup is shown in Figure 5-14.
The substrate (e.g., silicon) and source material (e.g., Al, Au, Cr, etc.) to be evaporated are
mounted in the bell jar. The system is pumped down to the appropriate pressure and the source
material is heated until it vaporizes. Frequently the substrate is also heated to improve film
adhesion.
When the source material’s vapor pressure exceeds that in the bell jar, the material vaporizes
rapidly. Under a high vacuum, the mean free path of the vaporized atoms or molecules is
_.-..&-.-.-

VACUUM
VESSEL
- 5.2 I. C. Technology

^.I----

AND HEATER
tart

CONDENSED DEPOSIT
MAY ACT AS A GETTEF
AND SORB U S E S
133

~ ~ ~ u s i ~ i HOLDER

Ei
THE EVAPORANT

rF-7 -\\ ! //I


GAS DESOR0ED FROI'

P
INTERNAL SURFACE
RATE INCREASES AS
THE SURFACES ARE
HEATED

I II "A
- POR SOURCE
BAFFLE

SOME GAS MAY BACKSTREAM


FROM THE FUMPING SYSTEM

TO PUMPING SYSTEM

Figure 5-14. Scheme of a vacuum evaporator.

greater than the distance between the source and the substrate. The vaporized atoms, radiated
in all directions, condense on all lower temperature surfaces with which they collide, including
the substrate and the bell jar, forming a uniform thin film.
The heating filament is generally made from a refractory metal having a high melting point
and low vapor pressure, for example, tungsten or molybdenum. R.F. heating of a crucible is
used extensively; when higher energies are required to vaporize the source material, electron
beam bombardment is used.
Vacuum evaporation can be used to deposit most single element conductors, resistors, and
dielectrics.. However, alloy depositions are difficult to control because each component has a
different evaporation rate at a given temperature.

5.2.6.2 Sputtering

Sputtering is the process of removing surface atoms or molecules from a solid cathode by
bombarding it with positive ions from a rare gas discharge. Some of the released atoms or
molecules are intercepted by and deposited on the nearby substrate to form a thin film layer.
The cathode may be made of metal or an insulator and, in contrast with thermal evaporation,
complex compounds such as Pyrex glass can be sputtered with a lesser degree of change in
chemical composition. This technique is used for deposition of compound materials such as
ceramics in microsensors. A more detailed outline is given because it is also an important
micromachining technique.
134 5 Semiconductor Integrated Circuit Technology and Micromachining

i. DC Sputtering
A typical cathode sputtering system is shown in Figure 5-15. The bell jar contains a low-
pressure inert gas, usually argon, at about 1.3 to 13.3 Pa. A glow discharge is formed by apply-
ing a dc potential of 2000 to 5000 volts across the anode and the heated cathode. The
positively charged argon atoms accelerate toward and collide with the cathode and cause
atoms or molecules of the cathode to “sputter.” The cathode atoms are attracted toward the
anode and deposit on the nearby substrate.
Owing to the relatively high gas pressure, the mean free path of the sputtered atoms is
shorter than in vacuum evaporation. This results in lower deposition rates using sputtering
processes.
The chemical composition of the deposited films can be modified by adding small amounts
of reactive gases such as oxygen, nitrogen, or hydrogen to the argon gas. For example, tan-
talum nitride is sputtered using a tantalum cathode in a nitrogedargon atmosphere. This pro-
cess is known as “reactive sputtering.”
Thin films of many refractive metals, such as tantalum and molybdenum, which are dif-
ficult to evaporate, can be obtained by the sputtering process. Also, sputtered films of alloys,
such as nickle-chromium, may retain the chemical composition of the original sputtering
target. Insulators cannot be deposited through dc sputtering techniques because the cathode
potential cannot be applied to the insulator’s surface.

ii. RF Sputtering
Insulator or conductor films can be deposited by rf sputtering. The geometry of the system
is essentially the same as that for a dc sputtering system. An rf field is applied between the
anode and the cathode. To confine the discharge to the target area, coils are placed around
the bell jar to create an axial magnetic field. If the target is an insulator, a net negative charge

BELL JAR

CATHODE
SHIELD

CATHODE
HIGH-VOLTAGE
ANODE
HIGH-VOLTAGE
HEATER

HIGH
VOLTAGE Figure 5-Ea.
TO Scheme of a diode sputtering
VACWM PUMP apparatus.
5.2 I. C. Technology 135

MATCHING NETWORK

SHIELD
VACUUM
CHAMBER
RF ELECTRODE WALL 13.56 MHz
WITH TARGET

HOLDER
(a1

OIELECTR IC TARGET
PLATE BONDED WITH
CONDUCTIVE COMPOUND
TO METAL BACKING PIATE

X L SUBSTRATE HOLOER

Figure 5-15b. Scheme of the rf sputtering apparatus.

builds up on the insulator, creating a negative bias between the target and the anode. Sputter-
ing occurs in a manner similar to the dc case. If a conductor is the target, it must be
capacitively coupled to the cathode so that a dc bias can be established. The overall efficiency
of an rf sputtering system for depositing a conductor is lower than that of a dc sputtering
system.

iii. Magnetron Sputtering


The deposition rate of a dc sputtering system can be significantly improved by using
magnetic fields to intensify the gas discharge. The magnetic fields are produced by permanent
magnets and are oriented such that they are approximately parallel to the exposed surface
of the target. Planar and conical targets are used in this type of system, and the targets
are water cooled. High deposition rates for aluminium, aluminium/t% silicon alloys, and
aluminium/4% copper/2% silicon alloys are possible with dc magnetron sputtering.
RF magnetron sputtering can be used for the deposition of SiO,. In conventional rf sput-
tering, electron and positive ion bombardment of the substrate causes degradation of the
deposited film and heating of the substrates. The addition of a strong magnetic field parallel
136 5 Semiconductor Integrated Circuit Technology and Micromachining

to the substrates protects them from charge particle bombardment, and permits a higher
deposition rate.

5.2.6.3 Chemical Vapor Deposition (CVD)

CVD metalization offers conformal coating with good step coverage and it can coat large
numbers of wafers at a time. The basic CVD setup is the same as that used for deposition
of dielectric and polysilicon discussed in Section 5.3.2. LPCVD is capable of producing con-
formal step coverage over a wide range of topographical profiles with lower electrical
resistivity than that from evaporation or sputtering.
One of the major new I.C. applications of CVD metal deposition has been in the area of
refractory-metal deposition. Many other metals such as tungsten (W), molybdenum (Mo),tan-
talum (Ta), and titanium (Ti) are of interest for I.C. applications. These metal films can be
deposited by hydrogen reduction in an LPCVD reactor. Aluminium can be deposited by using
organometallic compounds such as tri-isobutylaluminium.

5.2.7 Assembly Techniques and Packaging

The final stages in the fabrication of microelectronic circuits or microsensors are the series
of processes which transform the circuits from the wafer form to the rugged individual
package ready to withstand severe environments and provide for electrical connection to the
outside world. The assembly and packaging of LC!s and microsensors includes die separation,
die bonding, wire bonding, and package sealing.

5.2.7.1 Die Separation

The separation of an I.C. or microsensor into dice is accomplished by a scribe-and-break


technique. Scribing streets, devoid of oxide and metal, are provided on the wafer. These streets
are oriented along crystallographic cleavage planes to facilitate fracturing. The orientation is
accomplished by orienting the initial photolithographic pattern parallel to the primary orien-
tation flat on the wafer (see Figure 5-2). Scribing may be accomplished with a pulsed laser
beam, a diamond-tipped scribing tool, or a diamond-impregnated saw blade.
The breaking process after scribing consists of placing the wafer, scribed side down, on a
soft, flexible support, covering the wafer with a thin plastic sheet, and passing a roller over
the wafer under pressure. In some cases, an automatic wafer breaking machine is used.
In many cases, it is necessary to thin the wafers down to 200 pm or less thickness by back-
lapping before the scribing process.

5.2.7.2 Die Attachment

The attachment of device dice to packages provides thermal and electrical contacts between
the substrate and the package. If electrical contact is not required, glass frit or thermally con-
5.2 I.C. Technology 137

ducting epoxy can be used. If electrical contact is necessary, gold or silver filled epoxy or
metallic phase alloy is required.
Common die attachments use eutectic alloys such as gold-silicon, gold-tin, or aluminium-
germanium. The eutectic alloy can be plating on a metal package, thick film ink on a ceramic
package, plating on the backside of the die, or a thin alloy preform placed between the die
and the package. Commonly used preforms are listed in Table 5-8. The aluminium-germanium
eutectic, made at about 420”C, is used in an all-aluminium system that is not subject to the
formation of intermetallic compounds at high temperatures. It also provides the best radiation
hardness and good thermal impedance.

Table 5-8. Compositions and Melting Points for Die Attach Preforms

Temperature (“C) -
Composition Liquidus So1idus

80% Au 20% Sn 280 280


92,5% Pb. 2.5% Ag 5% In 300
97,5% Pb 1,5% Ag 1% Sn 309 309
95% Pb 5% Sn 314 310
88% Au 12% Ge 356 356
98% Au 2% Si 800 370
100% Au 1063 1063

Epoxy die bonding has become a popular die attachment technique, particularly for devices
which are not subjected to temperatures about 300°C. Epoxy compositions are available in
unfilled and filled forms. Thermally conducting, but electrically insulating, epoxies contain
aluminia or similar high-thermal conductivity power. Electrically conducting epoxies contain
either gold or silver. The epoxy is always dispensed by an automatic system. The disadvantage
of the epoxy bond is outgassing, usually while experiencing thermal stress, thus introducing
contaminants into the hermetic package with adverse effects on reliability and lifetime.

5.2.7.3 Wire Bonding

Wire bonding is an important interconnection step for I.C. or microsensors. It is performed


after die separation and die bonding. Wire bonding can be performed with gold wire (99.999%
pure gold and 12 to 30 vm in diameter) by thermocompression, ultrasonic or thermosonic
techniques, or with aluminium wire (99070AlA% Si) by the ultrasonic technique.
Thermocompression wire bonding makes use of a combination of temperature and pressure
to form welds between the wire and the bonding pad on the chip. Nail-held bonding and
wedge (or stitch) bonding are used.
The best results are obtained for gold-gold bonds; the gold-aluminium bonds are not as
reliable. In the presence of heat, gold and aluminium can react to form a number of in-
termetallic compounds. One of these compounds is purple in color and is often referred to
as the “purple plague.” The purple compound is a relatively good conductor, but its presence
is usually accompanied by a tan compound that is brittle and a poor conductor. Bonds that
138 5 Semiconductor Integrated Circuit Technology and Micromachining

include these compounds can fail after thermal cycling. The “plague” can be avoided by
limiting the temperature of the thermocompression bonding process.
Thermocompression pulse bonding is used when it is desirable to keep the bonding area at
a lower temperature. The substrate is maintained at 150°C and the bonding tip is pulsed to
450°C during the bonding process.
Ultrasonic bonding is a low temperature process. The heat required is generated by friction
between the wire and the bonding area. Aluminium wire with a small percentage of silicon
(typically 1%) is the most popular wire for this type of bonding, but some success is obtained
with gold wire. Ultrasonic bonds are reliable but not as strong as thermocompression bonds.
Thermosonic bonding is the combination of thermocompression and ultrasonic bondings.
The bonding area is maintained at 150°C and ultrasonic energy is added to form the bond.
The temperatures at which thermosconic bonding takes place is below the threshold for the
formation of Au-A1 intermetallics and does not disturb epoxy die bonds.

5.2.7.4 Packages

The selection of a package for microsensors or I.C. depends on the die size, number of bon-
ding pads, power dissipation, and anticipated environment. (Refer to Chapter 4 for a detailed
discussion on packaging.) In general, metal packages are superior for heat transfer, but are
more expensive than ceramic or plastic packages. For commercial applications, where a
hermetic seal is not required, injection molded plastic packages, usually in dual-in-line con-
figuration, are popular.
TO-series metal can packages are made of Kovar, an iron-nickel-cobalt alloy with a thermal
expansion coefficient closely matching the coefficient of sealing glasses. The gold-plated base
or header comes with gold-plated glass insulated leads in place. The I.C. die is eutectically at-
tached to the header and the connecting leads are wire bonded to the terminals. Following an
extensive cleaning procedure and bakeout to remove surface moisture, packages are sealed in

AV b R A M I C MOUNTING BASE

Figure 5-16. Exploded view of the 14-lead version of the flat package, showing the various com-
ponents as well as the completed flat package.
5.3 Micromachining 139

an inert atmopshere by welding a Kovar cap to the header. The welding process is rapid, and
results in a high yield hermetically sealed package.
Figure 5-16 is an exploded view of a typical ceramic flat package, used for hybrid I.C!s as
well as microsensors. In most larger flat packages, the package base is ceramic with lead
frames bonded in place using a high-temperature glass. The package seal is made eutectically
with a gold-germanium or gold-tin preform sandwiched between a gold-plated Kovar (or
ceramic) lid and a gold-plated ceramic window frame on the package base.
The dual-in-line package (DIP) was developed primarily for low cost and ease of equipment
assembly and is the most popular package. It is available in size ranging from 8 pins to more
than 40 pins. The leads are arranged in two rows on 2.5 mm centers and are sufficiently rugged
to withstand automatic insertion on printed circuit boards. A variety of ceramic and plastic
DIP packages are in existence.

5.3 Micromachining

Micromachining is the key technology for the rapid developments of solid-state sensors in
recent years. It has emerged as an important and expanding extension of integrated circuit
technology. Basically, micromachining is a combination of precise etching, insulator-to-silicon
or silicon-to-silicon bonding techniques, connection between layers, and standard integrated.
circuit methods to fabricate precise three-dimensional silicon-based microstructures of great
diversity, including thin diaphragms, microbridges, miniature cantilever beams, needles, gears,
springs, pits, grooves, orifices and pyramids [17, 191. These micromachined structures com-
bined with special purpose thin films and high performance read-out electronics, have been
successfully employed to realize a large variety of solid-state sensors for measuring pressure,
force, flow, magnetic field, temperature, pH, humidity, gas composition, and molecular con-
centration [20]. The following part will summarize the micromachining techniques to comple-
ment standard IC technology that has been outlined in Sections 5.2.1 to 5.2.7.

5.3.1 Silicon Etching

Wet etching of silicon is used for shaping and polishing, as well as for characterizing struc-
tural and compositional features.
The fundamental etch reactions are electrochemical in nature. They involve oxidation-reduc-
tion, followed by dissolution of the oxidation products, frequently by complexing. Both
anodic and cathodic microscopic sites exist at the semiconductor surface. Oxidation of
semiconductor atoms takes place at the anodic sites, while the oxidant is reduced at the
cathodic sites. The mechanism of silicon etching is generalized as follows:

1. Injection of holes into the semiconductor,

Si + 2 h + .+ Si2’ (5-14)
140 5 Semiconductor Integrated Circuit Technology and Micromachining

2. Attachment of hydroxyl group OH - to Si2 , e.g.,


+

Si2+ + 2 (OH-) + Si(OH), . (5-15)

The primary oxidizing species, O H -, is formed by the dissociating of water:

H2O c* OH- + H+ . (5-16)

3. Reaction of the hydrated Si with complexing agent (complexant) in the solution.


4. Dissolution of the reacted product into etchant solution.
The etching process is either reaction-rate limited or diffusion-limited. If the etching
depends on the chemical reaction rate, the process is called reaction-rate limited; if it depends
on the transport of etchant by diffusion to or from the surface through the liquid, it is called
diffusion-limited. Diffusion controlled processes have lower activation energies (of the order
of a few kJ/mol) than reaction-rate controlled processes, and are therefore relatively insen-
sitive to temperature variations. They are, however, affected by agitation to a greater extent.
Agitation increases the supply of reactant material to the semiconductor surface, thus increas-
ing the etch rate. Changes in etching conditions such as temperature, as well as relative propor-
tions of etchant components, can change the rate-limiting process. The supply of minority
carriers to the semiconductor surface can also limit the dissolution rate in etching reactions
that result in a depletion of electrons or holes. Creation of electron-hole pairs on the surface
by illumination or by application of electric currents or providing generation sites, can then
increase the rate of etching. Additional factors that determine the rate of etching of crystalline
semiconductor include crystal orientation, type and concentration of doping atoms, lattice
defects, and surface structure.
Dry etching is expected to be used extensively in micromachining in the near future. A sum-
mary has been given in Section 5.2.4.2.

5.3.1.1 Etching Profiles

One of the most important characteristics of the etching process is the directionality (or pro-
file) of the etching process. This characteristic is defined in Figure 5-17, where the lithographic
pattern is in the x-y plane and the z-direction is normal to this plane. If the etch rate in the
x- and y-directions is equal to that in the z-direction, the etch process is said to be isotropic
or nondirectional and the shape of the side wall of the etched feature is as shown in Figure
5-17a and b with weak and strong agitation levels, respectively. The etching of single crystal
silicon or polycrystalline and amorphous silicon in HNA etchant systems (HF, HNO,, and
acetic acid CH,COOH) will result in this profile. Etch processes which are anisotropic or
directional have etch rates in z-direction that are larger than the lateral (x- or y-direction) etch
rate (Figure 5-17c). An example of this etch profile is the etching of (100) single crystal silicon
in KOH/water or ethylenediamine/pyrocatechol/water (EDP) etchants. The extreme case of
directional etching in which the lateral etch rate is zero (to be referred to here as a vertical
profiire) is also shown in Figure 5-17d. This profile can be achieved by etching (110) single
crystal silicon with KOH/water or any silicon substrate by ion bombardment assisted plasma
etching (e.g., reactive ion etching or ion milling).
5.3 Micromachining 141

<lOO> Surface Orbtation

54.74.

(C)
<(lo> Surface Orientation

Figure 5-17. Summary of wet chemically etched hole geometries commonly used in micromechanical
devices; (a) isotropic etching with agitation, (b) isotropic etching without agitation,
(c) anisotropic etching on (100) surfaces, (d) anisotropic etching on (110) surfaces. [18]

5.3.1.2 Isotropic Etching

Isotropic etching of a semiconductor in liquid reagents is the most widely used process for:
(1) removal of work-damaged surfaces: (2) creating structures or planar surfaces in single-
crystal slices; and (3) patterning single-crystal or polycrystalline semiconductor films. For
isotropic etching of silicon, the most commonly used etchants are mixtures of nitric acid
(HNOJ and hydrofluoric acid (HF) in water or acetic acid (CH,COOH), usually called
HNA system.
In the HNA system of silicon etching, after the hole injection (Equation 5-14) and O H -
attachment (Equation 5-15) to the silicon to form Si(OH),, hydrogen is subsequently
liberated to form SiO,. Hydrofluoric acid (HF) is used to dissolve SiO, to form water soluble
H,SiF,.
142 5 Semiconductor Integrated Circuit Technology and Micromachining

The overall reaction is

Si + HNO, + 6 HF + H,SiF, + HNO, + H,O + H, . (5-17)

Water can be used as a diluent for this etchant. However, acetic acid (CH,COOH) is
preferred because it controls the dissociation of the nitric acid and preserves the oxidizing
power of HNO, for a wide range of dilution. Thus the oxidizing power of the etchant tends
to remain relatively constant during its operating life.
Extensive studies of the HF-HNO, system have been made [21]. Figure 5-18 shows the
results in the form of isoetch curves for various constituents by weight. It should be noted
here that normally available concentrated acids are 49.2 wt% HF and 69.5 wt% HNO,,
respectively. Either water (in dashline curves) or acetic acid (in solid-line curves) may be used
as the diluent in this system. From these curves the following characteristics are observed:
1. At high HF and low HNO, concentrations, the etch rate is controlled by HNO, concen-
tration. Etching tends to be difficult to initiate with uncertain induction period. In addition,
they result in relatively unstable silicon surfaces which proceed to slowly grow a layer of SiO,
over a period of time. Finally, the etch is limited by the rate of oxidation-reduction reaction,
so that it tends to be orientation dependent.

-
....---
CH3 COOH OILUENT
H,O OILUENT

Figure 5-18.
90 Bo 70 60 50 40 30 20 10 Isoetch curves for silicon in
c W E l G H T X. DILUENT HF : HNO, :diluent system. 1211

2. At low H F and high HNO, concentrations, the etch rate is controlled by the ability of
HF to remove the SiO, as it is formed. These etches are self-passivating in that the surface
is covered with a relatively thick layer of SiO, (30-50 A). The primary limit on the etch rate
is the rate of removal of the complexes by diffusion. Etches in this region are isotropic and
are true polishing etches.
3. Etch formulas in the HF : HNO, = 1 : 1 range are initially insensitive to the addition of
diluent. Eventually, they fall off very sharply in etch rate, until the system becomes critical
with respect to the diluent.
As seen from Figure 5-18, an almost infinite choice of compositions can be used for silicon
etching. A number of these are listed in Table 5-9a and Table 5-9b.
5.3 Micromachining 143

Silicon is also soluble to a small extent in HF solutions; for 48% HF, at 25°C a rate of
0.3 A/min was observed for n-type 2-ohm-cm (111)-Si. Diluted H F etches at a higher rate
because the reaction in aqueous solutions proceeds by oxidation of Si by OH - ions [22]. A
typical buffered HF solution (BHF) has been reported to etch at radiochemically measured
rates of 0.23 to 0.45 A/min, depending on doping type and dopant concentration [23].

5.3.1.3 Anisotropic Etching

Some etchants, such as EDP, KOH, and hydrazine are orientation dependent. This means
they etch the different crystal orientations with different etch rates. Etch rate ratios of about

'Fable 5-9a. Chemical Etchants for Silicon [18]

Anisotropic
Typical Masking Films
Etchant
Compo- Temp Etch Rate (100)/(111) Dopant Dependence (etch rate of
(Diluent) "C (pm/min) Etch Rate
sitions mask)
Ratio

HF 10 ml =1017 cm-, n or p
HNO, 30 ml 22 0.7-3.0 1 :1 reduces etch rate by SiO, (300 A/min)
(water, 80 ml about 150
CH,COOH )
25 ml
50 ml 22 40 1 :1 no dependence Si,N,
25 ml

9 ml
75 ml 22 7.0 1:l
30 ml
Ethylene 750 ml
diamine 120g 115 0.75 35 : 1
Pyrocatechol 100 ml 2 7 x 1019 cm-3 SiO, (2 A/min)
(water) boron reduces etch Si,N, ( A/min)
750 ml rate by about 50 Au, Cr, Ag, Cu, Ta
120g 115 1.25 35: 1
240 ml

KOH 44g
(water, 100ml 85 1,4 400 : 1 2 10,' cm-, boron Si,N,
isopropyl) reduces etch rate by SiO, (14 A/min)
50 g about 20
100ml 50 1.o 400: 1
H2N4 100 ml SiO,
(water,
isopropyl) 100 ml 100 2.0 - no dependence Al
NaOH 10 g 2 3 x 1020cm-3 Si,N,
(water) 100 ml 65 0.25-1.0 - boron reduces etch SiO, (7 A/min)
rate by about 10
144 5 Semiconductor Integrated Circuit Echnology and Micromachining

'hble 5-9b: Oxide Properties and Etch Rates

Semi- Etch Rate


Etchant Purpose Composition
Conductor (pm/min)
Si CP-4A Polishing or 3 ml HF 34,8
lapping 5 ml HNO,
3 ml CH,COOH
CP-8 Polishing 1 ml HF 7.4
5 ml HNO,
2 ml CH,COOH
0,3 g I,/250 ml
solution
Junction-staining Measurement of HF + 0.1% HNO,
etch Junction depth
Orientation- Groove 23.4 wt% KOH 0.6 for (100)
dependent etching 13.3 wt% 6 x for (111)
etch Propyl alcohol
63.3 wt% H,O
GaAs H,SO,-H,O,-H,O Polishing 8 ml H,YO, 0.8 for (111)-Ga
System 1 ml H,O, 1.5 for all other
1 ml H,O
H,PO,-H,O,-H,O Polishing 3 ml H,PO, 0.4 for (111)-Cia
System 1 ml H,O, 0.8 for all other
50 ml H,O

400 : 1 for (100) to (111) orientations have been published [24]. The etch rate for (110) surfaces
lies between those for (100) and (111) surfaces.
The mechanism behind the anisotropic etching is not fully understood. In a diamond lattice
of silicon, the (111) plane is more closely packed than the (100) plane; this density variation
may be responsible for the anisotropic etching behavior [MI. Thus the etch rate is expected
to be slower for the (111) plane. Another possible factor is the energy needed to oxidize an
atom on the surface. The differences in termination of the lattices for the (100) and (111) faces
are illustrated in Figure 5-19 [48]. The number of dangling bonds per Si atom on the surfaces
are one for (111) and two for (100) faces. The corresponding numbers of back bonds which
need to be broken to oxidize a silicon atom are three for (111) and two for (100) faces.
Therefore, the (111) surface is expected to etch more slowly. However, these differences in atom
and bond densities cannot be totally responsible for the high etch ratio as 400 : 1. Screening
or passivating effects also may be of importance.
The possible profiles of anisotropic (orientation-dependent) etching are shown in Figure
5-17c and Figure 5-17d. Anisotropic etching of (100)-oriented silicon through a patterned
silicon dioxide mask creates precise V-shape grooves. If the window in the mask is sufficiently
large or if the etching time is short, a U-shaped groove will be formed, as shown in Figure
5-17c. The width of the bottom surface is given by

w, = w, - 21 cot 54.74" = w, -m (5-18)


5.3 Micromachining 145

< 100 > surface

H H H H H H H H
0 0 0 0 0 0 0
S
,i \ / ,Si, \ / ,Si, \/ ,Si, \/"

< 1 1 1 > surface

H H H H
0 0 0 0
Figure 549. I I I I
The different terminations for the (100) and (Ill) silicon
surfaces. 1481

where W, is the width of the window on the wafer surface and I is the etched depth. If (110)-
oriented silicon is etched in KOH-water etchant, essentially straight-walled grooves with sides
of (111) planes can be formed.
Anisotropic etchants for silicon are usually alkaline solutions used at elevated temperature.
As in the case of isotropic etchants, the two principal reactions are oxidation of the silicon,
followed by dissolution of the hydrated silica. The oxidant can be H,O in aqueous alkaline
systems such as NaOH [25, 261 or preferably, KOH [27-351, CsOH [36], hydrazine [35, 371,
ethylenediamine [35,38], quaternary ammonium hydroxides [39], or sodium silicates 1261. The
complexing or chelating agent to effect dissolution usually consists of an alcohol such as
isopropanol [31, 371, n-propanol [30, 341, sec-butanol [30], or pyrocatechol [38]. Water is im-

7
EDP
6

Figure 5-20. I
Lateral etch rate versus 1
1
146 5 Semiconductor Integrated Circuit Technology and Micromachining

portant in these dissolution reactions, which form hydrogen as the gaseous byproduct [27].
Alkalis of high concentration, hydrazine, and quaternary ammonium hydroxide can act in
water solutions as both oxidant and complexant so that no additional dissolving agents may
be needed. Table 5-9 lists the most commonly used anisotropic and some common isotropic
etchants for silicon.
It is shown in the literature that KOH and water system exhibit much higher (110) to (111)
etch ratios than the EDP system (the etch rate ratio of the [loo), [IlO],and [Ill) planes in EDP
are approximately 50, 30, and 1 [40]; while in KOH, they are approximately 100, 600, and 1
[41]). For this reason, KOH is especially useful for groove etching on (110) wafers. The large
differential etch ratio permits deep, high aspect ratio grooves with minimal undercutting of
the mask. A disadvantage of KOH is that SiO, is etched at a rate which precludes its use as
a mask in many applications. In structures requiring long etching times, Si,N, is the prefer-
red masking material for KOH, while EDP can be masked by a wide variety of materials (eg,
SO,, Si3N4,Cr, and Au). Hydrazine is attractive because it does not attack both aluminium
and SO,. However, EDP is expensive and hydrazine is very toxic. In the anisotropic etch the
alignment of mask to crystal orientation is important. Misalignment will change the etch rate
greatly, as shown in Figure 5-20 [43].

5.3.1.4 Etch Stop

i. Dopant Dependent Etch Stop


The etching process is fundamentally a charge-transfer mechanism and etch rates depend
on dopant type and concentration. Highly doped material might be expected to exhibit higher
etch rates because of the greater availability of mobile carriers. This has been shown to occur
in the HNA system (HF : HNO, : CH,COOH or H,O = 1 : 3 : 8) [45], where typical etch
rates are 1-3 pm/min at p or n concentrations > 10" cm - 3 and essentially zero at concen-
trations < 10" cm b 3 .

On the other hand, anisotropic etchants such as EDP [46, 471 and KOH [31] exhibit a dif-
ferent preferential etching behaviour which has not yet been adequately explained. Silicon
heavily doped with boron (1lozocm -3) will reduce the etch rate by about 5 to 100 when et-
ching with KOH; when etching in EDP, the factor is about 250. Figure 5-21 a and Figure 5-21 b
[44] show the relative silicon etch rate as a function of boron concentration for KOH and EDP
etchants, respectively. Several proposals to explain the boron etch stop mechanism have been
published but not verified [18, 48, 491.

ii. Electrochemical Etch Stop


When etching electrochemically, a voltage is applied across the silicon wafer and an elec-
trode (usually platinum) in the etching solution. The fundamental steps of the etching
mechanism are the same as in chemical etching. Oxidation is promoted by a positive voltage
applied to the silicon wafer, which causes an accumulation of holes in the silicon at the
silicon/solution interface. Under this condition, oxidation at the surface proceeds rapidly
while the oxide is readily dissolved by the solution. Holes transported to the negative platinum
electrode as H ions are released there as hydrogen gas bubbles. Excess hole-electron pairs
+

can also be created at the silicon surface, eg., by optical excitation, thereby increasing the etch
rate.
5.3 Micromachining 147

Figure 5-21 a.
Relative (100) silicon etch rate as a function of the
boron dopant concentration for Ethylenediamine-
Pyrocatechol-Water etchant. [44]

Figure 5-21 b.
Relative (100) silicon etch rate as a function of the
boron dopant concentration for various KOH
solutions. 1441 BORON CONCENTRATION

For an isotropic etchant, such as HF, heavily doped silicon substrates with high conductivity
can be etched more quickly than lightly doped material, which has lower conductivity. This
technique has been used successfully to remove heavily doped layers leaving more lightly
doped membranes in all possible dopant configurations: p on p +,p on n , n on p + , and
+

n on n + [SO].
A process to form an n-type silicon diaphragm with a thick n+-silicon rim using elec-
trochemical etch-stop method is depicted in Figure 5-22a. The typical etching condition is as
follows: electrolyte: 5% aqueous HF; bath temperature: room temperature; complete
148 5 Semiconductor Integrated Circuit Technology and Micromachining

darkness; anode: silicon wafer; cathode: platinum gauze; distance between anode and
cathode: 5 cm; and voltage across anode and cathode: 10 V.
Figure 5-22 b schematically shows the current density versus applied voltage across
anode and cathode during the electrochemical dissolution of silicon in 5% aqueous HF [51].
Since the current density is related to the dissolution rate of silicon, it can be seen that
p-type and heavily doped n-type materials can be dissolved, whereas n-type silicon does
not dissolve at sufficiently low voltages (for example, 10 V). It should therefore be possible
to separate n - or p-type regions from n-type region of silicon by electrochemical
+

dissolution. An experiment [50] on homogeneously doped silicon wafers shows that


n-type silicon of about 3 x 10l8 cm -3 (< 0.01 a cm) is completely dissolved in the above
mentioned etching condition, whereas n-type silicon of donor concentration lower than
2 x 10l6cm -3 (> 0.3 cm) is not dissolved; n-type silicon is dissolved when the acceptor
concentration is higher than 5 x 10” (< 3 a cm) and the dissolution rate increases

FROM BY TO

I
N,
.
N-epitaxial Si layer Remaining thin
On N-S1
N+-Si aubatrate

Anodic dissolution of N+-Si

Figure 5-22a. Forming an n-type silicon membrane using electrochemical etching in 5% H F electrolyte.
~511

5 lo b Yo -8

Figure 5-22b. Current-voltage characteristic of the electrochemical etching of silicon in 5% H F elec-


trolyte. [51]
5.3 Micromachining 149

with the acceptor concentration, and at acceptor concentration below 5 x 1015cm-3 a layer
of brownish porous silicon is formed on the surface of the silicon substrate.
Figure 5-23 schematically shows the profiles of the net impurity concentration as they
will appear in n n, p +n, n +p, and p + p epitaxial silicon substrate [50]. The picture indicates
+

the regions of etching, non-etching, and formation of a brownish porous silicon layer. The
brownish porous silicon that remains after the electrochemical etching is usually removed
in a freshly prepared etchant consisting of 50 ml HF (48070), 50 ml acetic acid (990700)and
200 mg KMnO, (99070)and the etch rate is approximately 0.2 Fm/min.

Figure 5-23. Electrochemical etching of epitaxially grown n +n, n +p, p +n, and p + p structures under
standard etching conditions. [50]

A more useful electrochemical etch-stop method employs anisotropic etchants such as KOH
or EDP. It is also called p-n junction etch-stop [52, 531. This has the advantage of retaining
all the anisotropic etching characteristics described in earlier sections. The current-voltage
characteristics of both p- and n-type silicon samples of < 100> orientations in EDP etchant
is shown in Figure 5-24. There are two voltages of special importance: the open circuit poten-
tial (OCP), which is the potential at which current is zero, and the passivating potential (PP),
where current suddenly drops (or the surface is passivated from further etching). At potentials
(or applied voltages) lower than the PP, the sample is etched, while at potentials higher than
the PP, an oxide grows and the surface is passivated (because anisotropic etchant, such as
EDP, usually etches silicon dioxide extremely slowly). The growth of an oxide results from
the progressing race between oxidation of the silicon and dissolution of the oxide at the
silicon/electrolyte interface. The OCP and the PP are dopant type-dependent. The PP for p-
type silicon is more anodic than that for n-type silicon. This difference suggests a selective
etch technique used to etch only p-type silicon and leave n-type silicon unetched. When apply-
ing a voltage between the PP’s of n- and p-type materials, one expects that only the p-type
150 5 Semiconductor Integrated Circuit Technology and Micromachining

10
L' ' ' ' I ' I ' ' ' I " ' ' " ' ' ' I ' ' ' I '-I
PP I
f I
A I
p: 0.01 ohm-cm
5
n: 0.01 ohm-cm,p

-5

-10
-2.0 -1.5 -1 .o -0.5 0 0.5

Applied Voltage VS SCE (V1


Figure 5-24. Current-voltage characteristics of both n- and p-type silicon with various concentrations of
Ethylenediamine-Pyrocatechol-Waterelectrolyte.

silicon would etch while the n-type silicon would not. An n-type silicon diaphragm can be
formed as depicted schematically in Figure 5-25a. An n-epi on p-substrate silicon wafer (or
n-diffused in p-substrate) is used for this purpose. The anode is applied to the n-epi (or n-
diffused) side of the wafer. Figure 5-25 b shows the typical record of the anode current during
the etching process. The etching is stopped (or the wafer is passivated) when the current is
FROM BY TO
Anode SCE Cathode

N-epilaxial Si layer Remaining thin


on N-SI
P-Si subalrale

KOH or EDP

DIasolution of P-Si
Figure 5-25a. Forming n-type silicon membrane using p-n junction etch-stop method.
5.3 Micromachining 151

7
A
m-

<
55
6

I
p-n junction etch-stop
8nodc potential -0.6
conplclcly d8rk
v I\
/ I

i
w 3
K

ELAPSED TIME Inin)

Figure 5-25b. A typical anode-current record in the fabrication of silicon membrane using p-n junction
etch-stop method.

drastically increased and then dropped down from a very high to a low value when the oxide
is formed on the surface [46, 51, 531.

5.3.2 Chemical Vapor Deposition

Chemical vapor deposition (CVD) is the deposition of a solid material onto a heated
substrate via decomposition or chemical reaction of compounds contained in the gas passing
over the substrate. Many materials can be deposited via CVD, but those of most interest to
the sensor and semiconductor industries are single crystal silicon (epitaxy), silicon nitride,
silicon dioxide, and polycrystalline silicon (usually called polysilicon).

5.3.2.1 Silicon Epitaxy

The word epitaxy is derived from the Greek words, epi, meaning “upon,” and taxis, mean-
ing “ordered.” It is a term applied to processes used to grow a thin crystalline layer on a
crystalline substrate. The doping of the epitaxy layer is relatively independent of the substrate
doping, thus permitting high-quality, lightly-doped layers to be grown on heavily doped
substrates.
There is various epitaxial equipment available commercially using horizontal, pancake, and
barrel susceptors that support the wafers. The susceptor is usually radio frequency (rf) induc-
tively heated. Radiant heated systems are also in use.
Four silicon sources have been used. They are silicon tetrachloride (SiCIJ, dichlorosilane
(SiH,Cl,), tri-chlorosilane (SiHCl,), and silane (SiH,). Silicon tetrachloride has been the
152 5 Semiconductor Integrated Circuit Technology and Micromachining

most studied and has the widest industrial use. The typical reaction temperature is 1200°C.
Other silicon sources are used because of lower reaction temperatures.

5.3.2.2 Dielectrics and Polysilicon

Deposited dielectric films (e.g., silicon nitride and silicon dioxide) are used for insulation
and passivation of devices and I.Cls and are used extensively in physical and chemical
microsensors. There are three commonly used deposition methods: atmospheric-pressure
chemical vapor deposition (CVD), low-pressure chemical vapor deposition (LPCVD), and
plasma-assisted chemical vapor deposition (PECVD). PECVD is an energy-enhanced CVD
method because plasma energy is added to the thermal energy of a conventional CVD system.
Considerations in selecting a deposition process are the substrate temperature, the deposition
rate and film uniformity, the morphology, the electrical and mechanical properties, and the
chemical composition of the dielectric films.

i. Reactors
In a hot-wall, low-pressure reactor as shown in Figure 5-26a, the quartz tube is heated by
a three-zone furnace, and gas is introduced at one end and pumped out at the opposite end.
The semiconductor wafers are held vertically in a slotted quartz boat. Typical process
parameters are: pressure from 30 to 250 Pa, gas flow rates of 1 to 10 cm3/s, and temperatures
from 300 to 900°C. It deposits films with excellent uniformity and its large batch size allows
processing of several hundred wafers each run. However, the deposition process is slow and
the gases used may be toxic, corrosive, or flammable and may need special treatment before
being discharged into the environment.
The parallel-plate, radial-flow, plasma assisted CVD reactor shown in Figure 5-26 b consists
of a cylindrical glass or aluminium chamber sealed with aluminium endplates. Inside are two
parallel electrodes. An rf voltage is applied to the upper electrode, while the lower electrode

Vibration
absorbing

Figure 5-26a. A hot-wall, low-pressure reactor (LPCVD System).


5.3 Micromachining 153

INSULATED RF INPUT

GLASS
CYLINDER
(b)
ALUMINUM
ELECTRODES

HEATED
SAMPLE
4
.. .-
Figure 5-26 b. HOLDER
A parallel-plate, radial flow, GAS GAS
plasma-assisted CVD reactor. INLET INLET

is grounded. The rf voltage causes a plasma discharge between the electrodes. Wafers are
placed on the lower electrode, which is heated between 100 and 400°C by resistance heaters.
Process gas flows through the discharge from outlets located along the circumference of the
lower electrode. The main advantage of this reactor is its low deposition temperature.
However, its capacity is limited, and the wafers may become contaminated if loosely adhering
deposits fall on them.
ii. Silicon Dioxide
CVD silicon dioxide does not have the electrical properties of thermally grown oxides. It
is used to complement the thermal oxides. A layer of undoped silicon dioxide is used to in-
sulate multilevel metalization, to mask ion implantation and diffusion, and to increase the
thickness of thermally grown field oxides. Phosphorous-doped silicon dioxide is used both as
an insulator between metal layers and as a final passivation layer over devices. Oxides doped
with phosphorous, arsenic or boron are used occasionally as diffusion sources.
Silicon dioxide films can be deposited by several methods. For low-temperature deposition
(300 to 50O0C), the films are formed by reacting silane, dopant, and oxygen. The deposition
process can be performed either at atmospheric pressure (CVD) or at reduced pressure
(LPCVD).
For intermediate-temperature deposition (500 to 800 "C), silicon dioxide can be formed by
decomposing tetraethylorthosilicate, Si(OC,H,),, in an LPCVD reactor. This is suitable for
polysilicon gates requiring a uniform insulating layer with good step coverage. The oxides can
be doped by adding small amounts of the dopant hydrides (phosphines, arsine, or diborane)
similar to the process in epitaxial growth.
For high temperature deposition (9OO"C), silicon dioxide is formed by reacting
dichlorosilane, SiCI,H,, with nitrous oxide at reduced pressure:

SiCI,H, + 2 N,O 9yc


SiO, + 2 N, + 2 HC1. (5-19)

This deposition gives excellent film uniformity and is sometimes used to deposit insulating
layers over polysilicon.
iii. Silicon Nitride
Silicon nitride films can be deposited by an intermediate-temperature (750 "C) LPCVD pro-
cess or a low-temperature (300°C) PECVD process. The LPCVD films are of stoichiometric
154 5 Semiconductor Integrated Circuit Technology and Micromachining

composition (Si,N,) with high density (2.9 to 3.1 g/cm3). These films serve as good barriers
to the diffusion of water and sodium, and as masks for the selective oxidation of silicon. The
films deposited by PECVD are not stoichiometric and have lower density (2.4 to 2.8 g/cm3).
These films can be deposited over the completely fabricated devices and serve as their final
passivation. The PECVD nitride provides excellent scratch protection, serves as a moisture
barrier, and prevents sodium diffusion.
In the LPCVD process, dichlorosilane and ammonia react at reduced pressure to deposit
silicon nitride at temperatures between 700 and 800°C. The reaction is:

3 SiCl,H, + 4 NH, + Si,N, + 6 HCl + 6 H, . (5-20)

In the PECVD process, silicon nitride is formed either by reacting silane and ammonia in
an argon plasma or by reacting silane in a nitrogen discharge. The reactions are:

SiH, + NH, + SiNH + 3 H, (5-21 a)


2 SiH, + N, + 2 SiNH + 3 H, . (5-21b)

The product properties depend strongly on deposition conditions. The radial-flow parallel
plate reactor (Figure 5-17) is used to deposit the films. The deposition rate generally increases
with increasing temperature, power input, and reactant gas pressure. The plasma nitride used
in semiconductor processing generally contains 20 to 25% atomic hydrogen. Films with low
tensile stress (approximately 2 x 10' N/mm2) can be prepared by plasma deposition.
Film resistivities range from 10 to 10,' ohm-cm depending on silicon-to-nitrogen ratio,
while dielectric strengths are between 1 x lo6 and 6 x lo6 V/cm.

iv. Aluminium Oxide


CVD of aluminium oxide (Al,O,) films can be based on pyrolysis or oxidation of
organometallic reactants, or on hydrolysis of aluminium halides at higher temperature [6, 7,
111.
Pyrolytic CVD of amorphous Al,O, films is particularly suitable with aluminium
isopropoxide as the starting material at a reaction temperature of 420°C in N, or 0, [5, 6,
7, 121. Pyrolysis of aluminium triethoxide at 300-750°C and aluminium acetylacetonate at
about 480°C have also been used.
Oxidative CVD reaction based on tri-isobutylaluminium and 0, at 25O-50O0C, tri-
methylaluminium and N,O at 650"C, and tri-methylaluminium with 0, at 275-475°C all
yield A1,0,. The films derived from tri-methylaluminium have much better quality than
those from the tri-isobutyl derivative. In general, Al,03 films deposited at low temperature
are amorphous but convert to crystalline when heat treated at or above 800"C, with substan-
tial changes of their properties.
Deposition of Al,O, films by the AlC1,-C0,-H, reaction at 850 to 1200°C has been used
extensively for preparing dense, polycrystalline deposits of high quality.

2 AlCl, + 3 H, + 3 CO, + A1,0, + 6 HCl + 3 CO . (5-22)


5.3 Micromachining 155

Amorphous A1,0, films have been formed in PECVD systems by vaporizing AlCI, into an
oxygen plasma. The pressure of the reaction chamber and the AlCI, vaporization rate greatly
affected the rate of film formation, which varied from 70 to 500 A/min, and was linearly
dependent on rf power. Adherent films up to thicknesses of several microns were prepared at
the optimum substrate temperature of 480°C.

v. Polysilicon
In MOS circuit technology, polysilicon is used as the gate electrode of MOS devices. It sur-
passes aluminium for electrode reliability. It is also used as a diffusion source to create shallow
junctions and to ensure ohmic contact to crystalline silicon. Additional uses include the
manufacture of conductors and high-value resistors. It is used extensively for microsensors.
An LPCVD system (Figure 5-26a) operated between 600 and 650°C is used to deposit
polysilicon by pyrolyzing silane:

SiH, -+ Si + 2 H, . (5-23)

Of the two most common low-pressure processes, one operates at a pressure of 26 to 13 Pa


using 100% silane, while the other process involves a diluted mixture of 20 to 30% silane in
nitrogen at the same total pressure.
Polysilicon can be doped by diffusion, ion implantation, or the addition of dopant gases
during deposition, which is referred to as in-situ doping. The implantation method is the most
commonly used because of its lower processing temperatures.

5.3.3 Bonding Layers of Materials

In integrated circuit (I.C.) packaging, the bonding technique is used to attach the I.C. die
to the leadframe and for the final seal on hermetic packages (Section 5.2.7). In silicon-on-in-
sulator (SOI) I.C. fabrication, the silicon-to-silicon bonding method and the preferential et-
ching method are used to produce a high quality thin silicon layer on insulator. A silicon wafer
is bonded to another silicon wafer via a layer of glass or SiO, film. One of the silicon wafers
is then thinned down by etch-back method to form a thin layer of single crystal silicon on an
oxide layer of another silicon substrate [54]. In the fabrication of a silicon sensor, e.g., solar
cell, it will often be necessary to encapsulate the sensor chip by glass-to-sensor chip bonding
method since the sensor chip will be used in exposed, hostile, and potentially abrasive en-
vironments [MI.Sometimes the substrate is bonded to the silicon substrate to form the active
part or supporting part (to relieve stress from the package) of the sensor itself, e.g., a silicon
capacitive pressure sensor and a piezoresistive pressure sensor [56, 571. In this section, bonding
using field-assisted method, low-temperature glass, metal film or preform, and thermal fusion
method will be summarized.

5.3.3.1 Electrostatic Bonding

Field-assisted thermal bonding [58,59], also called anodic bonding or electrostatic bonding,
is commonly used for joining glass to silicon for micromechanical application [HI. The utility
156 5 Semiconductor Integrated Circuit Technology and Micromachining

of field-assisted bonding arises from the low process temperature. Since the glass and silicon
remain rigid during the process, it is possible to bond glass to a silicon surface preserving
grooves and cavities in either the glass or silicon, which allows formation of devices such as
pressure transducers.

i. Glass-to-Silicon Bonding

The field-assisted thermal bonding process of glass to silicon can be accomplished in at-
mosphere or vacuum at temperatures between 180 and 500°C, i. e., near the annealing point
but well below melting point of the glass. The bonding set-up is shown schematically in Figure
5-27.The polished glass substrate is placed against the polished surface of the silicon. A
cathode electrode is held against the outer surface of the glass wafer and the whole assembly
is heated on a temperature controlled hot plate, which also serves as an anode. A 200 to 1OOOV
potential (depending on the thickness of the glass) is then applied between the electrodes. The
electrostatic force created by the electric potential pulls the two substrates into intimate con-
tact. At elevated temperature, the two wafers are bonded almost instantly. Looking through
the glass, the bonded region will become a dark grey color; when this region expands
throughout the whole wafer, the bonding is completed. A constant current, instead of constant
voltage, could also be used, but is avoided since dielectric breakdown may occur after the bon-
ding is completed, and the interface region becomes an insulator.

e (Beee e (B+
Figum 5-27.
Schematic description of the electrostatic
bonding of silicon-to-glass.

At elevated temperatures, near the annealing point the glass behaves like a solid electrolyte
and is moderately conductive. The positive sodium ions (Na +) in the glass become mobile
and are attracted to the negative electrode on the glass surface where they are neutralized by
5.3 Micromachining 157

the current carrying electrons. The more permanently bonded negative ions in the glass are
left behind, forming a sodium depletion layer in the glass adjacent to the silicon surface.
Capacitance [60] and microprobe [61] studies show that this depletion layer extends several
micrometers into the glass. The potential drop across this high resistance depletion region
creates a large electric field which pulls the glass and metal together.
The extremely high field developed in the depletion region transports oxygen out of the glass
to bond with the silicon surface. The seal appears to be chemical in nature, possibly Si-0
bonds. When a point cathode is used near the center of the glass plate, the bonding starts at
the initial point of contact (center) and spreads outward. This has the advantage of preventing
trapped air pockets.
During the bonding process the temperature and potential are kept constant. The typical
current versus time characteristic is shown in Figure 5-28 [63]. The current decays rapidly at
the start of the process due to the initial charging of the depletion layer. The exponential tail
shown in the current curve is believed to be due to the leakage within the depletion layer. The
area under the initial charging peak corresponds to the charge removed from the depletion
layer. The shape of the curve is determined by the series resistance (of the bulk glass) and the
charge concentration in the depletion layer. Since, effectively, no current is observed when the
applied voltage is reduced to zero, the depletion layer is charge neutral, which indicates that
any oxygen counter-ions present in the depletion layer are delivered to the anode (the silicon).
The seal is formed when the driven-out oxygens oxydize the silicon surface to glass network.
As can be deduced from Figure 5-28, bonding occurs after the delivery of 2 mC/cm2 to the
anode, which suggests that about 20 nm of oxide needs to be grown on the silicon to form
a good bond to glass.

P m s s Condltlons
m *C
1 kV
Sillcon Anode
Pymx G I s u

Figure 5-28.
Current response during the electrostatic
bonding of silicon to glass. [63] TI- Is1

The two conditions for bonding are: (1) the surfaces to be bonded must be smooth in
microns, and (2) sufficient oxidation must occur at the anode to provide permanent bridging
bonds. External pressure may also be used to secure this initial contact.
Bonding process parameters vary widely with applications and materials. Temperature,
voltage, current density, time, and atmosphere are important. Bonding temperature as low as
200°C has been used successfully in bonding Pyrex # 7740 to silicon. Values of 500 to 600°C
158 5 Semiconductor Integrated Circuit Technology and Micromachining

are used for irregularly surfaced devices and rough glasses. Higher temperatures can be used,
but process advantages are reduced and operational problems such as handling soft glass are
introduced. For silicon devices with aluminium (or gold) metalization, temperature should not
exceed the eutectic point of aluminium (or gold) and silicon, i.e. 577°C (or 370°C).
The voltage requirement depends upon temperature and glass type and thickness. Qpically
500 V to 1 kV is used for 7 to 60 mm thick glasses at moderate temperature bonding. Current
density varies in the same manner as voltage, typically being 0.1 to 1 mA/cm2. Process time
is dependent on temperature, voltage, and bonded area. Qpically 5 to 10 minutes is enough.
A longer time is required when using lower temperatures.
An air atmosphere is adequate for bonding and may be preferable due to the presence of
free oxygen. However, special applications may require a controlled atmosphere; for example,
bonding a semiconductor devise with corrosion sensitive metallization and bonding where
vacuum sealing is required. Ambients that have been used successfully in silicon-to-glass bon-
ding include air, nitrogen, forming gas, argon, and helium [64].
Laser heating techniques have also been used in field-assisted thermal bonding [55, 651.
This laser-assisted anodic bonding process combines the application of an electrical field with
heating of Pyrex glass via a CW CO, laser exposure to produce bonds similar to those
created by the standard field-assisted bonding process.
The main requirements for the materials to be bonded are:
1. The glass must be slightly conductive at the process temperature.
2. The surface roughness of both the glass and silicon should be less than 1 pm rms and
free from dust, organic residues, and other contamination.
3. The thermal expansion coefficients of two materials should be closely matched. A major
mismatch of expansion coefficients will build up stress when cooled down from the bonding
temperature and cause one or both materials to crack. For this reason, Pyrex # 7740 (anneal-
ing point 565OC) and # 1729 (annealing point 853OC) are well suited for bonding to silicon.
Figure 5-29 shows the linear expansion coefficient as a function of temperature for several
commercial glass substrates identified by Corning Glass Works code numbers and silicon [66].
As can be seen, the thermal expansion of # 1729 glass has the best match with silicon.
Field-assisted bonding of semiconducting GaAs to glass needs a special treatment prior to
the process [67]. The glass used for bonding is Corning # 0211. Even in a reducing atmosphere
of H, and N,, the surface of GaAs will form a non-adherent oxide layer which prohibits the
bonding. A solution to the problem is to prebake the glass prior to the bonding process in
a reducing atmosphere (H, or N,) at 400°C for 15 hours. This procedure is believed to
remove the complexes contained in the glass which are amenable to chemical reaction. The
bonding temperature is 360°C and the applied potential is 800 volts. The charge build-up takes
5 to 10 minutes and the entire operation is completed within 30 minutes.

ii. Silicon-to-Silicon Bonding

Silicon and silicon can also be anodic-bonded together via a thin layer of borosilicate glass
[68, 711 and SiO, [62, 69, 701.
The surfaces of the silicon to be bonded should be polished and one of the silicon surfaces
coated with a thin layer of borosilicate film [67]. The film is deposited by rf sputtering method
(Section 5.2.6.2) using a Pyrex 7740 target. The sputtering is carried out in a 1% oxygen in
argon atmosphere. For a satisfactory sealing, a film of 2 to 4 pm is needed. As-sput-
5.3 Micromachining 159

Figure 5-29.
Linear expansion coefficient as a function of tem-
perature for several commercial glass substrates and
silicon. [66]

ter-deposited borosilicate glass films are usually silicon rich and difficult to bond. The Pyrex-
coated silicon substrate is annealed at the annealing point (565°C) in a wet oxygen at-
mosphere. One hour annealing is usually adequate for good bonding.
To carry out the bonding, the two members are aligned in the desired orientation and held
in position by a point contact cathode. The cathode is applied to the film-coated silicon
member. After the sandwich is stabilized at its bonding temperature, i.e., 450 to 550"C, a
slowly increasing dc voltage is applied across the silicon-Pyrex-siliconsandwich. The primary
control during the bonding process is keeping a constant current density of 1 mA/cm2. The
voltage is advanced in steps as the current decreases with time. A maximum of 50 volts is ade-
quate for a good sealing. After it reaches the maximum voltage, the assembly is held at this
condition for 5 minutes. The substrate heater is then shut off so that the temperature can
decrease to near room temperature before the voltage is turned off. This sealing is proved to
be hermetic [67].
A sputter-deposited film of 2 to 4 pm thickness is neccessary to form a successful bonding
[71]. To make seals with this film thickness, a thermally grown SiO, (lo00 to 3000 A) may be
included in between the glass and the silicon to prevent electrical breakdown through the glass
layer. In this case the bonding is performed at 400°C with 200 volt dc supply for 5 minutes.
The yield of the bonding is approximately 90%.
Field-assisted bonding of two silicon wafers with thermally grown oxide is also reported to
be successful [62, 69, 701. Both silicon members are covered with 1 pm-thick thermally grown
SO, films (bonding of a bare silicon wafer to the second silicon wafer with thermally grown
oxide failed because of oxide breakdown under very small applied voltage). Temperatures in
the range of 850 to 950°C are selected as the best anodic bonding temperature. An anodic
voltage of 30 volt is applied for 45 minutes at the chosen bonding temperature.
160 5 Semiconductor Integrated Circuit Technology and Micromachining

5.3.3.2 Thermal Fusion Bonding

A silicon-to-silicon bonding by fusion in a high temperature furnace has been reported [54,
72-77]. Wafer bonding, as shown in Figure 5-30, is achieved simply by placing the oxidized
surfaces of two wafers in close contact and inserting them into an oxidizing ambient at
temperatures greater than 700°C. Wafers processed in a clean environment can realize voidless
bonding with high yields [75]. This method has given a new fabrication method for the silicon-
on-insulator devices [54, 72-76] and micromechanical sensors [77, 781. This bonding method
is simple and has many advantages, such as (1) the bonding temperature is below that which
forms defects, (2) voidless bonding is formed with high strength, and (3) bonded wafers are
stable in thermal and chemical treatments.

Si-Si02 Si02-Si Bonding

si I icon > 700 OC

Figure 5-30. Schematic illustration of the silicon-to-silicon bonding by fusion.

Silicon-to-silicon fusion bonding is performed as follows: A wafer is first chemically


polished and thermally oxidized in wet oxygen. The oxidized wafer and another chemically
polished bare silicon then undergo a hydration process by soaking them in a HzOz-HzS04
mixture [75], diluted HzS04 [74], or boiling nitric acid [77]. Then the wafers are rinsed in
deionized water and dried. Contacting the mirrored-surfaces at room temperature in clean
air forms self-bonding throughout the wafer surface with considerable bonding forces. The
self-bonded wafer pairs are heat treated in a nitrogen atmosphere [74-761 or oxidizing
atmosphere [54, 731 to form a strong bonding. The heat treatment temperatures are in the
range of 700 to 1100°C. Higher temperature (above 1000°C) is usually required to get voidless
and high strength bonding (reaching the fracture strength of single crystalline for a silicon:
100-200 kg/cmz) [75]. The bonding can be done successfully with one oxidized silicon wafer
to another bare silicon wafer [74-76, 781, two oxidized wafers [54, 73, 771, or two bare silicon
wafers [72, 751.
A method for bonding silicon by using high pressure and high temperature has been
reported [go, 811. A high pressure of 7 . 105-1.4 . lo7 Pa was applied at about 1200°C for
one minute. However, silicon pressure bonding at high temperatures has fundamental pro-
blems in regard to the increase of defects and the distortion of impurity profiles. It is found
5.3 Micromachining 161

experimentally that the required fusion temperature in oxidation atmosphere is approximately


proportional to R 6 and h - 3 , where R is the wafer radius and h the wafer thickness [82].

5.3.3.3 Low Temperature Glass Bonding

i. Phosphosilicate Glass
Phosphosilicate glass (PSG) films with low melting point have been used for silicon-
to-silicon wafer bonding [83]. A SiO, film of 1-2 pm thick was thermally grown on the
wafers. A PSG layer containing approximately 10% P,O, was formed on the SiO, surfaces by
exposing the wafers to a mixture of phosphorus exychloride (POCl,) vapor and 0, gas at
900°C. The wafers were then aligned and clamped together inside a quartz vacuum-chuck
assembly which forced the wafer surfaces into intimate contact by connecting them to a
vacuum pump throughout the bonding process. The bonding assembly was pushed into a fur-
nace at 1100°C and heated for 30 minutes, then slowly cooled to room temperature. Fusion
of the PSG layers took place readily and the resulting bond strength was excellent provided
the wafer surfaces were clean and reasonably flat.

ii. Glass Frit


Corning Glass Works has introduced a series of glass frit ( # 75xx) suitable for hermetic
low temperature sealing in a variety of applications [84]. The sealing temperature of these
glass frits ranges from 415-65OoC, and their thermal expansion coefficient ranges from 2 to
5 times that of the thermal expansion coefficient of silicon. Technology Glass Corporation
has also introduced a suspension of high purity, low sodium, ultrafine glass powder in
isopropyl alcohol [MI. It is designed for silicon-to-silicon wafer bonding. The thermal expan-
sion coefficient is about 3 times that of silicon and the sealing temperature is 510-540°C.
The glass frits for hermetic low temperature sealing can be divided into two types: vitreous
and devitrifying. Vitreous glasses are thermoplastic materials which melt and flow at sealing
temperature each time they are thermally processed. Devitrifying glasses are thermosetting
materials which crystallize by surface nucleation in a time-temperature relationship to produce
their specified properties. Once the sealing glass crystallizes, its thermal stability is improved
because its softening point is not that of the original glass frit, but is that of the crystalline
material which exhibits increased chemical durability compared to the original glass frit.
These glasses can be applied in four different ways: spraying, screen printing, extrusion, and
sedimentation. After the glass is applied, it has to be preglazed to remove the organic residues
produced by vehicle and binder decomposition. The sandwich of substrate-glass-substrate is
then heated to the sealing temperature and a pressure of at least 7 . lo3 Pa is applied to the
sandwich. Sealing cycles depend on the geometry of the seal and the composition of the solder
glass used.

iii. Spin On Glass


Wafer bonding with spin-on-glass (SOG) was reported by Yamada et al. [86]. SOG, which
consists of Si(OH), (2 < x < 4), was coated on the wafer surface to be bonded. After baking
the SOG film at a temperature of 250°C for 10 minutes with a thickness about 50 nm, the
162 5 Semiconductor Integrated Circuit Technology and Micromachining

wafers were washed and contacted in vacuum and then pressurized to 1 MPa at 250°C for
1 hour. The bonding became strong enough after this step; however, wafers were annealed for
1 h at 1150°C in air to sinter the SOG layer and to improve the dielectric susceptibility and
the breakdown voltage of the layer.
One of the advantages of SOG bonding technique is that there is no limitation of the sur-
face material. A silicon nitride film, which cannot be bonded by the thermal fusion bonding
method (Section 5.3.3.2), can be bonded without any difficulty.

5.3.3.4 Metallic Bonding

Another general class of bonding for micromachined parts is metal sandwiched between
layers of silicon. Solder and gold eutectic bonding are examples of metallic bonding.
Eutectic bonding using Au/Sn preform has been used in the fabrication of miniature
pressure sensors [87]. A silicon diaphragm with a diffused piezoresistor bridge is bonded to
a silicon substrate etched to have a cavity, forming a reference chamber. The bonding is per-
formed by sandwiching a 80% Au/20% Sn preform about 25 pm thick between the two
substrates. A pressure of about 1 MPa is applied. This assembly is put into a vacuum chamber
and gradually heated to 280°C and then cooled down to room temperature in four hours: one
hour to raise the temperature and three hours to decrease it. Evaporated aluminium film
(5 pm) can be used to bond silicon wafers by heating the assembly in vacuum to about the
eutectic point (T > 580°C) to form Si-A1 alloy layers [88].
Bhagat et al. reported silicon-to-silicon bonding by solid phase epitaxy [89].Aluminium,
germanium, or platinum was sputtered on silicon wafers with thickness of 500 nm. Wafers
were contacted and were pressurized to 0.3 MPa at 900°C for 30 minutes in vacuum. A
hermetic seal was obtained.
It has been found that oxidizing ambient atmospheres will produce good bonding of one
titanium surface to another titanium surface, even at temperatures as low as 700°C. The con-
cept of metal-to-metal bonding can be employed to bond silicon to silicon as well as to various
metallic surfaces [90].The bonding was performed by first depositing a 500 nm thick titanium
film on oxidized silicon. The two surfaces to be bonded were placed in physical contact and
were pressed together manually to remove air in between them (400 .. . 500 Pa). The sand-
wiches (Si-SiO,-Ti-Ti-SiO,-Si) were heated by placing them in an oxygen or steam atmosphere
in a rapid thermal annealing chamber for 20 minutes. Ti-Ti bonded very well in oxidizing at-
mosphere at temperatures ? 700°C. At a higher temperature (1150°C), Ti-Ti bond formation
was also successful in N, atmosphere [90].Cold welding of silicon wafers coated with Ti-Au
film was accomplished by compressing the wafers together in vacuum at room temperature
[881.

5.3.4 Connection between Layers

The sensor will sometimes necessarily be exposed or be in contact to the measurand which
may be hostile or abrasive. It will often be necessary to avoid the placement of the output leads
on the same side as the active sensor part so that they do not corrode or stand in the way bet-
ween sensor and measurand. Some sensors have their active regions on one side of the wafer
5.3 Micromachining 163

and the circuitries on the other side. Connection is thus needed to join the active region to
the circuit. A through-wafer via connector is the solution. Over the years, several techniques
have been developed to create via connectors in a silicon wafer, eg., (1) chemically etched-
through holes, (2) laser drilled holes, and (3) diffused aluminium columns with thermal gra-
dients to form conductive parts.

5.3.4.1 Etched-through Hole

A back-contact Ion-Sensitive Field Effect lfansistor (ISFET) using the etch-through hole
is shown in Figure 5-31 [91]. After the ISFET is formed on one side of the wafer, two holes
are etched from the backside of the wafer using anisotropic etchant (eg., EDP). The holes
are aligned with the source and drain contact regions of the ISFET and the etching will
automatically stop when the etchant reaches the heavily boron doped source and drain layers
(see Section 5.4 for dopant dependent etch-stop). A layer of metal film is then deposited on
the backside of the wafer and patterned to form the contact to the source and the drain of
the ISFET. In this case, only pchannel ISFET can be fabricated because heavily doped boron
(p ++)diffusion is needed for the source and the drain.

I t e Sensing Area

CVO Layer
Encapsulant

511 icon
Dloxlde

Figure 5-31.
A backside contact p-channel ion-
sensitive field-effect transistor which Drain Substrate So&
uses etched-through-hole via. [9] Contact Contact Contact

5.3.4.2 Laser Drilled Hole

The through-wafer via connector fabricated by the anisotropic etching has a large opening
angle (54.7") associated with the slow-etching crystallographic (111) planes of the substrate.
Qpically, the closest spacing for vias wet etched in a 300 Vm-thick wafer is approximately
425 pm, thus limiting those methods to low effective aspect ratio via holes. A laser drilled hole
can produce high aspect ratio through-wafer via conductors (93-951. The electrical contact
topology for a solid state capacitive pressure sensor using laser drilled vias is shown in
Figure 5-32 [94].
When holes are drilled in a silicon substrate by a laser beam in air or in an inert atmosphere,
solid debris collects around the holes. But if the procedure is done in chlorine gas, the silicon
vapor and droplets leaving the hole combine with the chlorine to form SiCl, gas and flow
164 5 Semiconductor Integrated Circuit Technology and Micromachining

1-Yide Anodic Bond -1

Figure 5-32.
Electrical contact in a silicon
capacitive pressure sensor using
laser-drilled vias. [94]

away from the work area. The hole is cleaner and more sharply defined, and debris does not
spatter on the surface of the substrate [95].
Vias can also be formed by a laser photochemical etching method [96, 971. The silicon
substrate is enclosed in a C1, atmosphere and irradiated with an argon-ion laser. The laser
photolyzes the C1, into reactive C1 atoms and at the same time heats a several-micrometer
spot on the silicon wafer to near the melting point. A highly localized reaction results in rapid
etching accompanied by the evolution of volatile SiCl,.
Br,, HCl, XeF, and KOH can be used as the etchant in the laser photochemical etching of
silicon in addition to Cl,. Table 5-10 summarizes several kinds of substrate material etched by
laser process with different kinds of etchant and laser wavelengths [98].

'Lgble 5-10. Summary of Etchants and Several Substrates Etched by Laser Photochemical Process I981

Substrate Etchant(s) Laser Wavelength (nm)


Si Br,, CI,, HCI, XeF,, KOH 488
Si c12 308, 331
Si COF, 193
Ge Br, 488, 514
GaAS Cl,, CCl,, H,SO, : H,O, (Aqueous) KOH, CH,Br, CH,CI, HNO, 488
GaAS CH,Br, CH,CI 193
GaP KOH 488
InP CF,I, H,PO,, HCI : HNO, 488
CdS H,SO, : H,02, HNO, 488
SiO, (32 488
SiO, CCl,F, 248
W COF, 193
A1 H,PO, :HNO, : K,Cr,O, 488
Ag c12 331, 355

5.3.4.3 Thermomigration of Aluminium Column

Anthony and Cline have published a series of reports on the migration of liquid eutectic
AVSi alloy droplets through single crystal silicon [93, 99, 1001. At sufficiently high
temperatures, aluminium will form a molten alloy with the silicon. If the silicon slice is sub-
5.5 References 165

jected to a temperature gradient (approximately 5O"C/cm, or 2.0"C across a typical 400 pm


thick wafer), the molten alloy zone will migrate to the hotter side of the wafer. As the AI/Si
liquid region transverses the bulk, aluminium will solidify with the silicon at the colder inter-
face. Thermomigration results in a p-doped trail extending through, for example, an n-type
wafer. The thermomigration rate is typically 3 pm/min at 1100°C. At that temperature, the
normal diffusion rate of A1 in silicon will cause a lateral spread of the p-doped region of only
3-5 pm or a migration of 400 pm (the full thickness of standard silicon wafers).

5.4 Conclusion

This chapter summarizes the major I.C. technology and micromachining that are com-
monly used for sensors and microsensors, with selected tables and figures for quick reference,
and selected reference papers on micromachining. For those who want to obtain a more
thorough understanding as well as more detailed information, see reference papers [18 to 1001.
The technology is evolving and new techniques are being developed; for example,
MBE/MOCVD, LPCVD, plasma deposition, dry etch with sacrificial layers, low temperature
bonding, silicon-silicon fusion, as well as new packaging techniques, all of which will be
developed in the near future.
On the other hand, X-ray lithography, ultrafine particle metallurgy and Langmuir-Blodgett
film (see Chapter 7), as well as other polymeric materials and techniques to form microstruc-
tures will be important areas for future research.
Technology is the tool; it needs to be combined with new materials and new principles,
either physical or mechanical, in order to develop new sensors and microsensors with better
performance to meet the needs of the future.

5.5 References

[I] Sze, S. M., Semiconductor Devices - Physics and Technology, New York: John Wiley, 1985.
[2] Colclaser, R., Microelectronics - Processing and Device Design, New York: John Wiley, 1980.
[3] Reinhard, D., Introduction to Integrated Circuit Engineering, Boston: Houghton Mifflin, 1987.
[4] Quick Reference Manual for Silicon Z.C. Technology, Beadle, W., Tsai, J., Plummer, R. (eds.); New
York: John Wiley, 1985.
[5] Wolfe, S., Tauber, R., Silicon Processing for the VLSZ Era, Sunset Beach, California: Lattice Press,
1986.
161 Ghandi, S., VLSZ Fabrication Principles, New York: John Wiley, 1983.
[7] Sze, S. M., VLSZ Technology, New York: McGraw-Hill. 1983.
[8] VLSZ Electronics: Einspruch, N. (ed.); New York: Academic Press, 1981 to 1988.
191 Mead, C., Conway, L., Introduction to VLSZ Systems, Reading, Massachusetts: Addison-Wesley,
1980.
[lo] Microcircuit Processing: Lithography and Dry Etching, Thompson, L. E, Wilson, C. G. (eds.);
American Chemical Society, 1984.
[Ill Chemical Vapor Deposition 1984: Robinson et al. (eds.); Pennington: Electron Chemical Society,
1984.
[12] Technologyand Physics of MBE: Parker, E. H. C., Dowsett, M. G. (eds.); New York: Plenum Press,
1986.
166 5 Semiconductor Integrated Circuit Technology and Micromachining

1131 Pamplin, B. R., Molecular Beam Epitaxy, Oxford, England: Pergamon Press, 1980.
[14] Manaserit, H. M., J. Crystal Growth, 55, No. 1 (1981) 1-9.
[I51 Takahasi, K., Dig. Tech. Papers, Transducers '85; Philadelphia, Pennsylvania, 1985, pp. 274-277.
[16] Capasso, F., Tsang, W. T., IEDM Tech. Dig., 1982, pp. 334-337.
1171 Mikromechanik: Heuberger, (ed.). , Springer, 1989.
[18] Peterson, K. E., ZEEE Proceedings No. 70, (1982) 420-457.
[I91 Fan, L. S., Tai, Y. C., and Muller, R. S., Dig. Tech. Papers, Tmnsducers '87, Tokyo, Japan, 1987,
pp. 849-852.
(201 (1) Dig. Tech. Papers, Transducers '85, Philadelphia, Pennsylvania, 1985.
(2) Dig. Tech. Papers, Transducers '87, Tokyo, Japan, 1987.
[21] Robins, H. and Schwartz, B., Journal of the Electrochemical Society (1960) 107-108.
1221 Hu, S. M. and Kerr, D. R., Journal of the Electrochemical Society 114, (1967) 414.
[23] Hoffmeister, W., International Journal of Applied Radiation and Isotopes 2 (1969) 139.
[24] Kendall, D. L., Applied Physics Letters 26 (1975) 195.
1251 Runyan, W. R., SemiconductorMeasurements and Instrumentation, New York: McGraw-Hill, 1975,
Chapters 1, 2, 7 & 9.
[26] Pugacz-Muraszkiewicz, I. J. and Hammond, B. R., Journal of Vacuum Science Technology 14
(1977) 49.
[27] Zwicker, W. K. and Kurtz, S. K. in: Semiconductor Silicon, 1973, Huff, H. R. and Burgess, R. R.
(eds.); Princeton: Electrochemical Society, 1973; pp. 315-326.
[28] Waggener, H. A., Kragness, R. C. and bler, R. C., International Electron Device Meeting Abstracts
11, No. 1, (1967) p. 68.
[29] Stoller, A. I., RCA Review, 31, (1970) 271.
[30] Waggener, H. A. and Dalton, J. V., Electrochemical Society Ext. Abstracts, No. 72-2, (1972) 5817.
[31] Price, J. B. in: Semiconductor Silicon, 1973, Huff, H . R. and Burgess, R. R. (eds.); Princeton: Elec-
trochemical Society, 1973, pp. 339-353.
[32] Clemens, D. P., Electrochemical Society Ext. Abstracts, No. 73-2, (1973) 407.
[33] Weirrauch, D. F., Journal of Applied Physics. 46 (1975) 1478.
(341 Bean, K. E., Yeakley, R. L. and Powell, T. K., Electrochemical SocietyExt. Abstracts, No. 74-1 (1974)
68.
[35] Declerq, M. J., De.Moor, H. P. and Lambert, J. P., EIectrochemicalSocietyExt. Abstracts, No. 75-2
(1975) 446.
[36] Clark, L. D., Lund, J. L. and Edell, D. J., Tech. Dig. IEEE Solid-state Sensor and Actuator
Workshop, 1988 pp. 5-8.
[37] Declercq, M. J., Gerzbery, L. and Meindl, J. D., Journal of the Electrochemical Society, I22 (1975)
545.
[38] Finne, R. M. and Klein, D. L., Journal of the Electrochemical Society, 114 (1967) 965.
[39] Asano, M., Dho, T. and Muraoka, H., Electrochemical SocietyExt. Abstracts, 76, No. 2 (1976) 911.
[40] Bassous, E., IEEE Trans. Electron Devices, ED-25 (1978) 1178-1885.
[41] Bean, K. E., IEEE Trans. Electron Devices, ED-25 (1978) 1185-1193.
[42] Wu, X., Wu, Q. and KO, W. H., Dig. of Tech. Papers, Transducers '85 ,Philadelphia, Pennsylvania,
1985. pp. 291-293.
[43] Seidel, H., Csepregi, L., Sens. Actuators, 4 (1983) 455-463.
[44] Seidel, H., Dig. of Tech. Papers, Transducers '87, Tokyo, Japan 1987 pp. 120-125.
1451 Muraoka, H., Ohhashi, T. and Sumitomo, Y.in: SemiconductorSilicon, Huff, H . R., Burgess, R. R.
(eds.); Princeton: Electrochemical Society, 1973 327.
I461 Greenwood, J. C., Journal of the Electrochemical Society, 116 (1969) 1325.
[47] Bohg, A., Journal of the Electrochemical Society, 118 (1971) 401.
[48] Palik, E. D., Bermudez, U. M. and Glembocki, 0. J., Journal of Electrochemical Society,132,No. 1,
(1985) 135-141.
[49] Palik, E. D., Faust, J. W., Gray, H. F. and Green, R. F., Journal of the Electrochemical Society, I29
(1982) 2051-2059.
[SO] Theunissen, M. J. J., Appels, J. A. and Verkuylen, W. H. C. G., Journal of the Electrochemical
Society, 117 No. 7, (1970) 959-965.
[51] van Dijk, H. J. and deJonge, J., Journal Electrochemical Society, 117, No. 4 (1970) 553-554.
1521 Waggener, H. A., Bell System Tech. Journal, No. 50 (1970) 473-475.
5.5 References 167

153) Jackson, T. N., Tischler, M. A. and Wise, K. D., ZEEE Electron Device Letter, DEGZ, No. 2, (1981)
44-45.
1541 Lasky, J. B., Stiffler, S. R., White, F. R. and Abernathy, J. R., International Electron Devices
Meeting, 1985 684-687.
1551 Engelkrout, D. W., Day, A. C. and Horne, W, E., 16th ZEEE Photovoltaic Specialist Conference,
1982 pp. 108-114.
[56] Sander, G. S., Knutti, J. W. and Meindl, J. D., ZEEE Trans. Electron Devices, ED-27 (1980) 927-930.
1571 Yamada, K., Nishihara, M., Kanzawa, R. and Kobayashi, R., Sens. Actuators, 4 (1983) 63-69.
[58] Wallis, J. and Pomerantz, D. I., Journal of Applied Physics, 40 (1969) 3946-3949.
1591 DeNee, P. B., Journal of Applied Physics, 40 (1969) 5396-5397.
[60] Wallis, J., Journal of American Ceramic Society, No. 153 (1970) 563-567.
1611 Gossink, R. G., Journal of American Ceramic Society, No. 61 (1978) 539-540.
1621 Frye, R. C., Griffith, J. E. and Wong, Y. H., Journal of the Electrochemical Society, 133 (1968)
1673-1677.
1631 Albangh, K. B. and Rasmussen, D. H., Tech. Dig. ZEEE Solid-state Sensor and Actuator Workshop
(1988) 109-110.
1641 Minucci, J. A., Kirkpatrick, A. R. and Kreisman, W. S., 12th ZEEE Photovoltaic Specialist Con-
ference, (1976) p. 309.
1651 Horne, W. E., US. Patent No. 4294602, 1981.
1661 Troxell, J. R., Harrington, M. I. and Miller, R. A., ZEEE Electron Device Letters, EDGS (1987)
576-578.
1671 Hok, B., Dubon, C. and Ovren, C., Applied Physics Letters, No. 43 (1983) 267-269.
[68] Brooks, A. D. and Donovan, R. P., Journal of the Electrochemical Society, ll9 (1972) 545-546.
1691 Anthony, T. R., Journal of Applied Physics, 58 (1985) 1240-1247.
1701 Ozawa, K., Ito, T. and Ishikawa, H., Japanese Journal of Applied Physics, 26 (1987) 1309-1312.
1711 Hanneborg, A. and Ohlckers, P. A., “Anodic Bonding of Silicon Chips Using Sputter-Deposited
Pyrex 7740 Thin Films” (to be published).
1721 Cohen, C. L., Electronics, December 23 (1985) 20-21.
1731 Lasky, J. B., Applied Physics Letters, No. 48 (1986) 78-80.
1741 Shimbo, M., Furukawa, K., Fukuda, K. and Tanzawa, K., Journal of Applied Physics, 60 (1986)
2987-2989.
1751 Ohashi, H., Ohura, J., Tsukakoshi, T. and Shimbo, M., Znternational Electron Devices Meeting,
1986 pp. 211-213.
1761 Ohura, J., Tsukakoski, T., Fukuda, K., Shimbo, M. and Ohashi, H., ZEEE Electron Device Letters,
EDGS (1987) 454-456.
1771 Tenerz, L. and Hok, B., Electronic Letters, No. 21 (1986) 615-616.
1781 Shoji, S., Nisase, T., Esashi, M. and Matsuo, T., Dig. of Tech. Papers, Transducers ’87, Tokyo,
Japan, 1987 pp 305-308.
1791 Iler, R. K., The Chemistry of Silica, New York: Wiley, 1979, Chapter 9.
1801 Kenney, D. M., US. Patent No. 3332137, 1967.
[Sl] Becke, G., Cave, E. F. and Stolnitz, D., US. Patent No. 3 355 636, 1967.
1821 Kido, T., MSc Thesis, Case Western Reserve University, Cleveland, Ohio, 1988.
183) Anacker, W., Bassous, E., Fang, F. F., Mindie, R. E. and Yu, H. N., ZBM Tech. Disc. Bull., No. 19
(1976) 372-374.
1841 Sealing Glass, Corning Technical Publication, 1981.
1851 7G-120 Si-to-Si Sealing Glass, Technology Glass Corporation Technical Publication.
1861 Yamada, A., Kawasaki, T. and Kawashima, M., ElectronicLetters, No. 23 (1987) 39-40 and 314-315.
1871 KO, W. H., Hynecek, J. and Boettcher, S. F., ZEEE Trans. Electron Devices, ED-26 (1979) 1896-1905.
1881 Binder, J., Ehrler, G., Heimer, B., Krisch, B. and Poppinger, M., SiemensForsch- u. EntwickL-Ber,
6, NO. 13 (1984) 294-302.
1891 Bhagat, J. K. and Hicks, D. B., Journal Applied Physics 61 (1987) 3118-3120.
1901 Nayak, D., Reisman, A. and lbrlik, I., Journal of the Electrochemical Society, No. 135 (1985)
1023-1025.
1911 Wong, A., PhD Dissertation, Case Western Reserve University, Cleveland, Ohio, 1985.
168 5 Semiconductor Integrated Circuit Technology and Micromachining

1921 van Den Vlekkert, H. H., Kloeck, B., Prongue, D., Berthoud, J., Hu, B. and DeRooij, N. F., Sens.
Actuators, No. 14 (1988) 165-176.
[93] Anthony, T. R. and Cline, H. R., Journal of Applied Physics, (a) 47 (1976) 2556; (b) 47 (1976)
2316-2324; (c) 48 (1977) 3943-3949; (d) 49 (1978) 5774-5782.
[94] Bowman, L., Schmitt, J. M. and Meindl, J. D. in: Micromachining and Micropackaging of Trans-
ducers, Fung, C. D., Cheung, P. W., KO,W. H. and Fleming, D. G. (eds.); 1987 pp. 79-84.
[95] Shlichta, P. G., NASA Tech. Briefs, No. 12 (1988) 84.
[96] Ehrlich, D. J., Silversmith, D. J., Mountain, R. W. and Tsao, J., ZEEE Tmns. Components,
Hybrids, and Manufacturing Technology, CHMT-5(1982) 520-521.
[97] von Gutfeld, R. J. and Hodgson, R. T., Applied Physics Letters, No. 40 (1982) 352-354.
[98] Eden, J. G., IEEE Circuits and Devices Magazine, No. 2, (1986) 18-24.
[99] Cline, H. E. and Anthony, T. R., Journal of Applied Physics, 49 (1978) 2412-2419 and 2777-2786.
[lo01 Mizrah, T., Journal of Applied Physics, 51 (1980) 1207-1210.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

6 Thin and Thick Films


SHIH-CHIA
CHANG.General Motors Research Laboratories. Warren. MI.
USA
WEN H . KO. Case Western Reserve University. Cleveland. OH. USA

Contents
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

6.2 Thick Film Processes . . . . . . . . . . . . . . . . . . . . . . . 171

6.3 Thin Film Processes . . . . . . . . . . . . . . . . . . . . . . . . 176


6.3.1 Thin Film Deposition Methods . . . . . . . . . . . . . . . . . . . 176
6.3.2 Thin Film Characterization Methods . . . . . . . . . . . . . . . . 178
6.3.3 Thin Film Delineation Techniques . . . . . . . . . . . . . . . . . . 179

6.4 Compatibility Issues . . . . . . . . . . . . . . . . . . . . . . . . 182

6.5 Langmuir-Blodgett Films for Sensors . . . . . . . . . . . . . . . . 183


6.5.1 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
6.5.2 Film Forming Apparatus (Film Balances) . . . . . . . . . . . . . . 187
6.5.3 Dipping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
6.5.4 Ion Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
6.5.5 Gas Sensors . . . . . . . . . . . . . .............. 190

6.6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . 191

6.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191


170 6 Thin and Thick Films

6.1 Introduction

A sensing system generally consists of three different functional groups : active transduction
element, passive supporting element and control/logic electronic circuit (either on-chip or off-
chip). The formation of these three components may require thin film and/or thick film
technologies [l-61. For instance, a semiconductor gas sensing system may have a thick film or
thin film semiconductor (eg, tin oxide or titania) as the transducing element, a thin film or
thick film resistor (eg, ruthenium oxide or platinum) as the heating element, and an insulating
ceramic plate or thin membrane as the substrate [7, 8, 9, 101. A silicon-based pressure
transducer may have a boron doped silicon thin film (piezoresistive) or zinc oxide thin film
(piezoelectric) as the transducing element on top of a micromachined single crystal silicon
membrane [ll, 12, 131. The membrane transforms mechanical force (pressure) into structural
deformation, thereby changing the electrical resistance (piezoresistive) or charged-state
(piezoelectric) of the transducing element. At present, the majority of the sensing systems rely
on off-chip electronic circuits for controVlogic function, although a certain level of on-chip
circuitry is preferred (and in some cases required) for the improvement of the reliability and
lowering the system’s cost [14, 15, 161.
The difference between thick film and thin film is not so much in the difference of the film
thickness per se, it is more in the difference of the deposition methods and the subsequent
device fabrication techniques. The electrical as well as mechanical properties of thick films
are generally less sensitive to the substrate as compared to those of the vapor deposited thin
films. This is mainly attributed to the printing and firing processes used to produce thick
films. Since thick film technology is a mature technology, and for the past ten years there has
been no fundamental change in the process, only a brief review on this subject will be
presented here. In contrast, thin film technology, mainly driven by the continuous advance-
ment of microelectronics, has been expanding and improving. In the sensor fabrication area,
the emerging technology for the past few years has been silicon micromechanics and on-chip
integration, of which the compatible technology for film preparation is thin film technology.
Consequently, thin film technology will be reviewed more extensively in this chapter. The
deposition techniques, such as CVD, PECVD, LPCVD, sputtering, and others, have been
discussed in Chapter 5 .
In the thick and thin film areas, progress in technology has outpaced science. The situation
is particularly serious in the sensor fields. This is the primary factor preventing a large variety
of sensors to reach a high quality performance. Examples will be given throughout the text
to illustrate this point.
This review confines the discussion of the thick film and thin film technologies to the single
device level. The fabrication processes for thick film and thin film circuits, which involve a
variety of other engineering issues have been extensively reviewed [ll, 121, and will not be ad-
dressed here. In this chapter, thick film technology is reviewed first through the description of
tin oxide gas sensors and cermet resistors, followed by the discussion of thin film technology
covering film deposition, characterization and delineation techniques. In Section 6.3 the
fabrication of tin oxide microsensors and polycrystalline silicon micro-structures are described
to illustrate the compatibility issues. A technology for organic films, Langmuir-Blodgett film
deposition, is reviewed as an example of developing new techniques potentially useful for
chemical and biological sensors.
6.2 Thick Film Processes 171

6.2 Thick Film Processes

Thick film technology has been used for the fabrication of metal interconnects on printed
wiring boards and passive components (eg, conductor, capacitor, and resistor) in hybrid cir-
cuits [l, 2, 3, 4, 18, 19, 201. It has also been used to produce various physical and chemical
sensors [21-251. The most commonly used thick film process consists of

- selection and preparation of a substrate,


- preparation of starting material in ink or paste form,
- screen printing or painting the ink or paste on the substrate, and
- firing the printed sample at a preset temperature-time profile, usually in an oxidizing am-
bient atmosphere.

The common substrates used for thick and thin films and their key properties are shown
in Tables 6-1 a and 6-1 b. Porcelain steel (enameled steel) is becoming popular as a thick film
substrate. It is a low carbon steel coated with low alkali content glass frit and fired at about
850°C to form a smooth glazed surface. It can be fired up to 625°C. The advantages are:

- high mechanical strength,


- easy machinability,
- good thermal conductance, and
- low cost.

The main disadvantage is the capacitive coupling of components to the steel substrate through
the thin (approximately 125 pm) coating.

Table &la. Properties of thick film substrates

96% 99.5% 99.5%


Characteristic Unit Conditions
Alumina Alumina Beryllia
Dielectric constant 1 MHz 9.3 9.9 6.9
1 GHz 9.2 9.8 6.8
Dielectric strength V/pm 5300 5600 5800
Dissipation factor 1 MHz 0.0003 0.0001 0.0002
1 GHz 0.0009 0.0004 0.0003
Thermal conductivity W/(cm K) 25°C 0.351 0.367 2.5
300°C 0.171 0.187 1.21
Thermal coefficient "C 25-300°C 6.4 x 6.6 x 7.5 x
of expansion
Bulk resistivity Wcm 25 "C 1014 1014 10 I4
100°C 2 x 1013 7.3 x 1013 loi4
Tensile strength MPa 170 193 I60
Surface finish nm 650-1000 250 500
Camber lrn/mm 4 4 3
c.
'hble 6-1b. Properties of thin film substrates 4
h,

Tape cast Corning 7059 Fused silica


Characteristic Unit Conditions Alumina Sapphire
beryllia glass (quartz)
~~

Dielectric 10.1 at 1 GHz 6.9 at 1 MHz 5.84 at 1 GHz 3.826 at 1 MHz 9.39 at 1 GHz
constant 10.7 at 9.9 GHz 6.80 at 1 GHz 5.74 at 8.6 GHz 3.824 at 6 GHz 9.39 at 10 GHz 9
2
Dielectric strength V/pm 19500 5800 10500 4800 ZI
Dissipation 1 MHz 0.0002 0.0001 o.oooo15 0.0001 &
factor 8.6 GHz 0.0002 0.0003 0.0036 0.00012 o.oooo5
Thermal
2
W/(cm K) 25 "C 0.367 2.5 0.017 0.014 0.417 R
conductivity 300 "C 0.187 1.21 0.008 0.008 2
Thermal coefficient 25 "C 6.7 x 7.5 x 4.6 x 0.49 x
of expansion
B
Bulk resistivity Wcm 25 "C 3.16 x 10" 1014 1014 3.16 x 10" 10l4
Thickness available w 250-1000 625 825 & 1650 Specify Specify
Tensile strength MPa 160 48 400
Surface finish nrn 25 380-500 25 25 25
Camber pm/mm 3 3 1 1 1
6.2 Thick Film Processes 173

Various pastes and inks for conductors, resistors, insulators (for capacitance) are commer-
cially available. Special purpose inks for sensor elements can be formulated with the proper
binder (low melting point glass) vehicle (define the printing characteristics of the material) and
special function materials. Screen printing usually uses stainless steel mesh with line widths
from 50 km to 200 pm.
The firing profile usually includes a pre-heat period when the temperature is ramped up
from room temperature to 250-350°C to drive off organic constituents, followed by a hot fir-
ing period with temperatures from 500°C to 100O"C, depending on the materials used. The
end period is for properly cooling the substrate to room temperature to minimize the thermal
stress of the films and the substrate.
Specific design rules and detailed selection of various materials and processes, including
trimming, are given in books and references [l-41.
The major advantages of the thick film process are the versatility in materials preparation
and the simplicity in film formation, which generally result in lowering the manufacturing
cost. The versatility in materials preparation is especially attractive for the fabrication of thick
film resistors as well as gas sensors. For thick film resistors, a wide range of resistivity values
are needed for different hybrid circuits, and for gas sensors, high differential sensitivity to
various gaseous species are desired. Both of these requirements can be achieved to some extent
through material modification, either by doping (for semiconductor gas sensors) or multi-
component mixing (for gas sensors or cermet resistors) [25,26]. These processes can be readily
and cost-effectively implemented by thick film technology at the ink or paste preparation step.
The fabrication of thick film tin oxide-based gas sensors and cermet type resistors will be
discussed as examples to further illustrate this point.
A. Figure 6-1 shows the processing flow chart for the fabrication of thick film tin oxide-
based gas sensors intended for the detection of hydrogen (H,), ammonia (NH,) and carbon
monoxide (CO) in the air [25]. The common base material is tin oxide mixed with palladium
(from palladium chloride, PdCI,) and magnesium oxide (from magnesium nitrate, MgNO,).
The purpose of incorporating Pd, a strong catalyst, is to promote the overall gas sensing ac-
tivities of the tin oxide. The function of MgO is not clear. It may be for the adjustment of
the sensor resistivity range (Mg with valence of 2 is an acceptor type dopant in the tin oxide
matrix). The selective sensitivities to H,, NH, and CO are enhanced by adding a few percent
of rhodium (Rh), zirconia (ZrO,) and thorium oxide (Tho,), respectively. These processes
are conveniently carried out in the paste preparation level. The porosity of the films, which
affects the overall sensor sensitivity, is mainly controlled by the binder materials (eg, silica or
organic vehicle) and the heat treatment procedure (temperature ramping rate, firing
temperature and time). The major problem confronting current semiconductor gas sensors is
that this selectivity level cannot be accurately controlled, and they are still unacceptable for
most practical applications. Since a detailed understanding of gas sensing mechanism@) and
the function of additives is lacking, a truly scientific approach to solve this problem has yet
to be developed.
B. Thick film cermet type resistors consist of a conductive component (gold (Au), silver
(Ag), palladium (Pd), or ruthenium (Ru)-based complex oxides) embedded in an insulating
medium (glass) [4]. Compared with thick film conductors or dielectrics, thick film resistors
are more sensitive to firing conditions and thus require more precise control of heat treatment
processes (temperature ramping rate, maximum temperature and firing time). The resistivity
of a cermet material depends on the concentration, size and distribution of the conductive
174 6 Thin and Thick Films

-
fire 600 "C. I h

fire800"C. I h

material

=-i
zro*: 5 wt%

fire 800 "C. 1 h


.I fire 800°C. I h

1 I I
I

-
leveling

-I calcination

600°C. I h -I calcination
5WoC.2h

calcination
400°C l h

mI
mtection
Figure 6-1. Flow chart for the preparation of thick film tin oxide-based gas sensors [MI.

component, as well as the resistivities of the conductive element and the insulating medium.
Figure 6-2 shows the resistivity versus the concentration of the conductive component for
several different thick film cermets. A wide resistivity range (more than 5 orders of magnitude)
is obtained by varying the concentration of the conductive element. The insulating medium
used in these films is a lead boro-silicate glass. For metal/insulator cermets (Au, Ag,
Pd + Ag/glass), a sharp increase in resistivity occurs when the concentration of the respective
conductive component is below a certain critical value (generally referred to as percolation
threshold) [27]. This is caused by the breaking up of the metal continuum, the current con-
ducting channel, into isolated islands. In the cases of Pd or Ru-based oxide/insulator com-
posites, there is no such sharp change in the resistivity. Evidently, a truly metallic conducting
continuum is not formed in these cases. Instead, it is suggested that thin insulating barriers
are formed between the conductive particles. The insulating barriers may be glass or PdO.
6.2 Thick Film Processes 175

lOOk
Au

10k \
lk

100

10

1
20 30 40 50 60 70 80
wt% Metal (Pigment)

Figure 6-2. Resistivity versus concentration of metal powders and resistor pigments in
PbO.B,O,.SiO, frit [ll].

When the insulating barrier is thin enough (< 50 A), the conduction is controlled by tunneling
processes, either through direct particle-to-particle tunneling or tunneling through in-
termediate states in the insulating barrier [4, 181.
The substrate effect on the electrical properties of the screen printed thick films is not yet
completely understood. It seems to be unimportant, partly due to the fact that substrates used
in thick films are inert insulating ceramic plates (eg, alumina). The possibility of the impurity
diffusion between the substrate and thick film is low. In addition, the crystal structure of the
screen printed thick film is not controlled by the nucleation processes initiated at the
filmhbstrate interface as is the case for the deposited thin film; rather, it is mainly controlled
by the sintering processes among the film constituents. However, for certain physical sensors,
such as thick film pressure transducers, compatibility of the material properties between the
substrate (eg, alumina) and the transducing element (thick film piezoresistive material, eg,
Ru0,-ruthenate) is one of the key requirements. For example, difference in thermal expan-
sion coefficients induces built-in stress in the transducing element which, in turn, introduces
zero point offset, long term drift and instability of the transducer [22, 28, 291.
The principal shortcomings of thick film technology are the limitation on the achievable
miniaturization of the device feature size and the poor dimensional control (in the silicon
microfabrication sense). These are mainly due to the intrinsic resolution limitation of the
screen printing technique (- 20 pm), large film thickness, and severe and hard to control film
176 6 Thin and Thick Films

thickness shrinkage during the firing process. Because of these, thick film processes are not
compatible with conventional silicon microfabrication, as active component, but are used ex-
tensively as a major part of hybrid assembly. On the other hand, the possibility of low cost
mass production of sensors by thick film technology has rendered a large number of
developments, some of which are compiled in Table 6-2.

Table 6-2. Examples for thick film sensors

Sensed quantity Principle Important Material(s) Ref.

Temperature Thermopile Au + PtAu alloy ~301


Thermistor Mn, Ru, Co oxides (spinel) [21, 31)
Temperature dependent
resistance Au, Pt, Ni
Pressure Diaphragm/capacitive
Diaphragm/piezoresistive Bi,Ru,O,
Concentration of Stoichiometry/electrical
conductivity depend on
CH,, CO, C2H,0H concentration SnO, + Pd WI
co SnO,/ThO, + hydrophobic [36]
silica
H,, CO, C,H,OH, Isobutane SnO, + Pd, Pt [371
0 2 Ba-, Sr-, CaTiO, 1381
H2 Nasicon [391
Humidity Resistive RuO, (spinel type)/glass [31]
Capacitive glass ceramic/Al,O, [401
Dew point BaTiO, /RuO,-glass [351

6.3 Thin Film Processes

Thin film processes, including deposition, characterization and device feature delineation,
have been extensively reviewed by many authors [5, 61. In this section, a brief general discus-
sion on these subjects is given. The emphasis is on the important aspects in thin film processes
which are relevant to sensor development.

6.3.1 Thin Film Deposition Methods

The most commonly used thin film deposition techniques can be grouped into four prin-
cipal categories :
1. Thermal evaporation : resistive heating
e-beam
2. Sputter-deposition : DC (plus magnetron)
RF (plus magnetron)
6.3 Thin Film Processes 177

3. Chemical methods : chemical vapor deposition (CVD)


plasma enhanced CVD
metallo-organic deposition (MOD)
4. Langmuir-Blodgett technique
The processes and facilities of 1, 2, and 3 have been discussed in Chapter 5 , Sections 5.2.6
and 5.3.2. Item 4 is summarized in Section 6.5 of this chapter.
Thermal evaporation and conventional sputter-deposition are well established techniques
(see Section 5.2.6). The film formation is through nucleation and growth. Atoms from the
vapor phase of a material condense on the substrate. In the initial stage (nucleation), the con-
densates form nuclei in a configuration with the lowest free energy, determined by the proper-
ties of the condensate and substrate, evaporation rate, and substrate temperature. The subse-
quent film growth is through diffusion-controlled processes. Magnetron sputtering is a
relatively new technique in which a magnetic field parallel to the cathode (and hence perpen-
dicular to the applied electric field) is added to the conventional sputtering system. The
Lorentz force, E x B, restrains the primary electrons near to the vicinity of the cathode,
thereby increasing the ionization efficiency of the electrons and thus increasing the sputtering
yield.
In chemical vapor deposition (CVD) (see Section5.3.2), a volatile compound of the material
to be deposited is vaporized, and the vapor is thermally (or assisted by plasma) decomposed
and then reacts with other gaseous species at the substrate to yield a non-volatile reaction pro-
duct on the substrate surface. Plasma enhanced chemical vapor deposition (PECVD) is a low
temperature process in which plasma is introduced mainly to enhance the pyrolytic process.
The deposited films are usually amorphous. The film stoichiometry can be controlled by vary-
ing the deposition parameters (eg, flow rates of the constituent vapors). Since the electrical,
mechanical and chemical properties of thin films are closely related to the film stoichiometry,
it is suggested that PECVD can be used to prepare thin films with properties tailored for
specific devices. This aspect makes PECVD particularly attractive for sensor fabrication.
Metallo-organic deposition (MOD) is a very powerful thin film deposition technique which
offers a means for depositing high quality thin films without the use of vacuum evaporation
or sputter equipment [41, 421. This deposition method has the versatility and the convenience
of the thick film technique discussed earlier. In addition, it is compatible with micro-fabrica-
tion processes. Therefore, MOD is a very attractive technique for the preparation of oxide thin
films intended for gas sensing, either for the traditional single sensor or for the emerging in-
tegrated sensors. A description of the MOD processes is presented below.
Metallo-organics are compounds consisting of two parts: (1) a central metal ion, and (2)
a ligand bonded to the central metal ion through a heterobridge such as oxygen, sulfur,
nitrogen, phosphorous or arsenic. The bond between the ligand and the metal ion is formed
by a pair of electrons donated by the ligand. The bond is a coordinate covalent bond in nature.
The metallo-organic inks are prepared by dissolving metallo-organic compounds in organic
solvents. The solvation results mainly from the intermolecular forces (dipole-dipole interac-
tions) between the solute and the solvent. Therefore, solute-solvent pairs having a high
polarizability are readily miscible. In general, spherically symmetrical molecules possessing
high molecular weight exhibit high polarizability. Consequently, tertiary carboxylates such as
neodecanoic acid and secondary carboxylates such as 2-ethylhexanoic acid are more
polarizable and are chosen as the solutes. Solvents such as xylene or pyridine which possess
high polarizability are chosen as the solvents.
178 6 Thin and Thick FiIrns

The MOD process basically consists of (1) application of ink to the substrate (eg, silicon
wafer coated with silicon dioxide), (2) spinning (eg, at 3000 rpm), (3) heat treatment of
the deposit. The MOD method has been used to prepare various different thin films such as
yttrium stabilized zirconia (ZrO,) and titania (TiO,) for oxygen sensors, barium titanates
(BaTiO,) and its derivatives for capacitor or pyroelectric sensor applications, tin oxide
(SnO,.) for NO, or reducing gas sensors, and high T, superconducting oxides such as yttrium
barium copper oxide (YBa,Cu,O,).
Some examples of sensors fabricated by thin film technology are given in Table 6-3.

'lhble 6-3. Examples for thin film sensors

Sensed quantity Principle Important Material(s) Ref.


Temperature Temperature dependent
resistance Pt I431
Radiation Bolometer Au [441
Flow Thermal anemometer Au 1441
Strain Piezoresistive effect CrNi 1451
Pressure Diaphragm/piezoresistive Poly-Si 1461
TiON 1471
Magnetic field Magnetoresistive effect Ni,,Fe,,, NiCo, 1481
Co72Fe8B20
Concentration of 0, Stoichiometry/electrical ZnO
conductivity depend on
concentration
Humidity Capacitive Ta205 1501

6.3.2 Thin Film Characterization Methods

Thin film characterization includes [5, 61 :


1. Microstructure determination : X-ray diffraction, electron diffraction (LEED, TED and
RED), electron microscopy (TEM and SEM).
2. Composition analyses : Auger electron spectroscopy (AES), X-ray photoelectron spec-
troscopy (XPS), electron probe microanalysis (EPM), secondary ion mass spectroscopy
(SIMS), Rutherford backscattering spectroscopy (RBS).
3. Mechanical property analyses : stress, adhesion and hardness.
All of these characteristics may have strong implications on the performance of thin film
devices. For instance, the sensing capability of the thin film tin oxide, and the transition
temperature and the current carrying capacity of high T, thin film superconductors, are very
sensitive to film composition. The intrinsic stress of polycrystalline silicon is strongly in-
fluenced by the size and orientation of the silicon grains [51, 52, 531.
One has to be very cautious when using high vacuum analytical instruments to determine
the film composition and to correlate the film composition and device performance. This is
because film composition may change when the ambient condition changes (eg, the
stoichiometry of oxide film changes when the ambient oxygen concentration changes) [9, 54,
551. The extent and rate of the change increases as the sample temperature increases. This
6.3 Thin Film Processes 179

poses a serious difficulty when high vacuum systems are used to study the sensing mechanisms
of the oxide semiconductor gas sensors. Gas sensors usually operate at elevated temperatures
(150-600°C) in high oxygen-containing environment (eg, air). At such high temperatures,
oxide films may have been greatly reduced under high vacuum in a relatively short period of
time. Consequently, the conclusion derived from vacuum experiments may not be applicable
to the actual sensing situation. This is one of the major conditions which hinders the establish-
ment of a comprehensive gas sensing model.

6.3.3 Thin Film Delineation Techniques

Thin film delineation techniques can be broadly categorized into two groups : wet etch and
dry etch (see Chapter 5, Sections 5.2.4. and 5.3.1). The wet chemical etch can achieve high dif-
ferential etch rates among different materials. For devices composed of many layers of dif-
ferent materials (eg, integrated circuits), device feature delineation is conveniently done by
using wet chemical etch. For instance, nitric acid (HNO,) with a few percent of ammonium
fluoride (NH,F) has a very high etch rate ratio between silicon and silicon dioxide. When
such a solution is used to delineate a silicon layer a few microns thick, a thin layer of SiO,
with a thickness less than 1000 A is used as the etch mask. The conventional wet etch is highly
isotropic (as opposed to the so-called orientation-dependent wet etch, discussed in Section 5.2.4).
Consequently, the smallest feature size generated by wet etching is limited by lateral etching
and undercutting. As the requirement for the device feature size reduction continues due to
the unabated advancement of VLSI, wet chemical etch becomes inadequate for the delinea-
tion of certain device components (eg, gate feature in a MOSFET). The development of dry
etching techniques is mainly to remedy this shortcoming of the wet chemical etch. Figure 6-3
shows a wet chemically etched phosphosilicate glass (PSG)/silicate glass (SG) double layer
pedestal with an overlay of polycrystalline silicon (polysi). The thicknesses of PSG, SG and
polySi are -0.4 pm, 1.6 pm and 1.6 pm, respectively. The phosphorous concentration is
-7%. In this particular case, a gradually sloped step of the etched pedestal is intended for
better step coverage of the polySi overlay. It is obtained by using a two-layer approach in
which a thin, faster etching layer (PSG) is deposited on top of a slow etching layer (SG) [5].

Figure 6-3.
A chemically etched microstructure.
The pedestal, composed of PSG
(0.4 pm)/SG(1.6 wm) was etched by BHF.
The gradually sloped step (desired for
better coverage of the polySi overlay)
results from the double-layer scheme.
180 6 Thin and Thick Films

The fast etching PSG layer facilitates the undercutting process, resulting in a gradual etched
step. The slope can be controlled to some extent by adjusting the thickness and the etch rate
ratios between the two layers.
The most commonly used dry etching techniquesare plasma etching and reactive ion etch-
ing (see Section 5.2.4). Plasma etching processes in which active species are generated from
certain inert molecules by glow discharge (plasma) have been relatively well understood. The
active species then react with the sample material producing volatile products which are subse-
quently pumped away. For example, in plasma etching of polysi, the inert gas (eg, CF,) is
dissociatively ionized,
CF, CF,+ + F + 2e.
+ e -----+
The radical species, fluorine 0 then reacts with silicon,

Si + 4F ----- + SiF4 (volatile).


Plasma etching is predominantly chemical in n a t w and, hence, is highly isotropic.
However, for planar plasma etching systems, in which samples to be etched are placed on the

Figure 6-4.Dry etch of polymer layer (HPR 204 photoresist) using oxygen plasma: (a) planar plasma
etch, with masking layer of silicon still on (note the highly isotropic etching nature) ;(b) RIE,
with masking layer of Si02 (note the highly anisotropic etching nature).
6.3 Thin Film Processes 181

unpowered electrode (anode), a certain degree of anisotropic etching has been observed in-
dicating that a minor ion bombardment-enhanced chemical etching has occurred 1371.
Reactive ion etching (RIE)is generally highly anisotropic. In RIE,samples to be etched are
placed on the powered electrode (cathode). Hence, ions generated by the glow discharge are
accelerated toward the sample surface by a high electric field (usually a few hundred volt/cm).
The ion bombardment facilitates the chemical reactions between the radicals and the sample
material and thus greatly enhances the etch rate in the direction of the electric field. This is
the most commonly accepted explanation of the cause of the anisotropy. Figure 6-4shows the
etched features of HPR photoresist layers by using (a) plasma etching, and (b) reactive ion
etching, using oxygen as the etch gas. The isotropic and anisotropic etching characteristics of
plasma etching and reactive ion etching, respectively, are clearly exhibited 156, 57, 581.
More often than not, both wet chemical etching and dry etching are required in the fabrica-
tion of a specific electronic device composed of several layers of different materials. Figure 6-5

Figure 65. SEM pictures of a tin oxide microsensor on a thin silicon membrane : (a) top view ;(b) cross-
sectional view.
182 6 Thin and Thick Films

shows a tin oxide based microsensor on a thin silicon membrane (a), top view; (b), cross sec-
tional view). The sensing element, a tin oxide thin film, is deposited by MOD method and
delineated by RIE technique. The built-in heater (a polySi layer) and the metal interconnect
(an aluminium/chrome double layer) are delineated by wet chemical etch. The silicon mem-
brane is generated by the orientation-dependent wet chemical etch using ethylene diamine
pyrocatechol (EDP) as the etchant (see Section 5.3.1).

6.4 Compatibility Issues

The fabrication of electronic devices usually requires tens or even hundreds of sequential
processing steps, and involves a variety of different materials. Hence, compatibility among the
fabrication processes and the different materials are the key factors determining the device per-
formance. In this section, the fabrication of thin film tin oxide microsensors (Figure 6-5) will
be used to illustrate the process compatibility issue and the fabrication of polySi microbridges
will be used to illustrate the material compatibility issue. The compatibility consideration is
also needed for thick film devices. The following discussion, therefore, is also applicable to
thick film device design.
Wet chemical etch of thin film tin oxide is done by placing zinc (Zn) powder on top of the
film and dipping the sample in hydrochloric acid (HCI) solution. The atomic hydrogen,
generated by the chemical reaction between Zn and HCl, reduces the tin oxide to tin which
is dissolved by HCl. This technique is tedious, time consuming and inaccurate. Consequently,
it is not suitable for batch fabrication (especially microfabrication). In the fabrication of tin
oxide microsensors, a RIE technique is applied to delineate the tin oxide thin film using SiCI,
as the etch gas, and a high temperature treated photoresist as the etch mask (a regularly treated
photoresist, once exposed to SiCI, plasma, becomes chemically very inert and cannot be
cleaned off chemically or by plasma) [59]. Oxygen plasma which is routinely used to etch off
photoresist has a significant effect on the resistivity of thin film tin oxide. Hence, after the
thin film tin oxide is deposited, oxygen plasma processing is avoided (or tightly controlled).
Thermal expansion coefficient, impurity diffusion and adhesion between the various layers
in a device are the most important aspects pertaining to materials compatibility. They have
profound effects on the mechanical and electrical properties of the individual layers. In
fabricating polySi micromechanical structures, such as bridges, cantilevers or membranes, the
so-called surface micromachining technique is used [34]. It consists of depositing a polySi
layer on top of a sacrificial layer (eg, PSG), delineating the polySi feature, annealing the sam-
ple at high temperature (llOO°C in dry nitrogen for 30 min.) to minimize the built-in stress
in polySi layer, and etching off the sacrificial layer using a highly selective wet chemical etch
(eg, hydrofluoric acid). Figures 6-6 b and c are the SEM photographs of three polySi bridges
with different underlying materials. The bridges have gone through the high temperature an-
nealing process. For polySi/PSG/SiO, and polySi/PSG/SG/Si,N, combinations (Figures
6-6a and b), the polySi layers are flat. The bridges remain flat after the underlying sacrificial
layer is etched off as shown in Figure 6-6(d). This implies that the built-in stress in the polySi
layer has been effectively reduced by the annealing process. However, for polySi/PSG/Si,N,
combination (Figure 6-6 c), the “stress annealing” process has resulted in a severely buckled
6.5 Lungmuir-Blodgett Films for Sensors 183

C d
ngUre 6 4 SEM photographs of polySi bridges with different underlayer combinations: (a)
polySi/PSG/SiO, ; @) polySi/PSG/SG/SiO,; (c) polySi/PSG/Si,N,; (d) a free standing
bridge.obtained from (a) or (b) after etching of the underlying sacrifical layer.

- -
polySi layer (the highest point of the buckled bridge is 18 pm, for a bridge 300 pm long
and -1.5 pm thick). The exact cause@)of the buckling is not known. It is speculated that
at 1100°C (annealing temperature), PSG flows freely due to the weak adhesive force between
the PSG and the Si,N, layers. The relaxation of the compressive stress in the polySi layer
causes the structure to buckle

6.5 Langmuir-Blodgett Films for Sensors

It has been known for centuries that oil films will spread on water. The first scientific paper
on this subject was presented to the Royal Society by Benjamin Franklin in ,1774 [a]. It was
184 6 Thin and Thick Films

not until the 1890’s that the methods for the control and manipulation of these monolayer
films were proposed. Anna Pockels was the first to use a form of the Langmuir trough [61].
Lord Rayleigh was the first to propose that these films were only one monolayer thick [62, 631.
Irving Langmuir is considered the major pioneer of modern monolayer research. He
demonstrated that the molecules in a compressed fatty acid monolayer were oriented with
their polar groups in the water and the hydrocarbon tails sticking up vertically out of the water
[64,651. Langmuir and Katherine Blodgett began the study of deposited monolayers in the
1930’s [66, 671. The paper by Blodgett is the classic description of the fundamentals of
monolayer deposition [67].
The standard reference text on monolayers is the book by Dr. G. L. Gaines, Jr. [68]. In 1985
the American Chemical Society began publishing a journal dedicated to surfaces and colloidal
films called “Langmuir”.
The Langmuir-Blodgett (LB) technique met the microelectronic age in 1972 when the first
metal-insulator-semiconductor capacitors were studied [69]. Starting in the late 1970’s, LB
films have been studied quite extensively as insulating materials in metal-insulator-semicon-
ductor capacitors and field effect transistors [70-781. A list of semiconductor substrates and
insulating LB films is given in Table 6-4. About this time, research in semiconducting
LB materials as active materials in electronic devices and solid state sensors began. There have
been many reviews published on the electronic and optical properties and sensor application
of LB films [79-831.

Table 6-4. LB-metal-insulator structure: substrates and insulator films

Substrate Reference Insulator Reference


~

Si 175, 761 Cadmium Stearate [70, 73, 741


amorph. Si U31, 74 Cadmium Arachidate [71, 73, 741
InP 171, 721 Calcium Behenate ~ 9 1
GaP [721 22-tricosanoic acid [77, 781
GaAs [771 Orthophenanthroline ~ 9 1
Ge [781 12.8-diacetylene [721
16.8-diacetylene [75, 76, 781

6.5.1 Materials

The materials which form LB films are amphiphilic; that is, they have a polar, hydrophilic,
head group and a hydrophobic tail, schematically shown in Figure 6-7a. The classical example
of these types of materials are the fatty acids. A list of these is given in Table 6-5. The polar
head group is the carboxylic acid and the straight chain hydrocarbon is the tail. When the
materials are spread onto the surface of a trough of water, the molecules orient themselves
with the head group into the water and the tail group out of the water (Figure 6-7 b). It is the
balance of the hydrophilic forces of the head group and the hydrophobic forces of the tail that
allows the material to form an oriented monolayer. If the head group is too strongly attracted
to the water, the material will dissolve, and if the tail is too hydrophobic, it will not form a
monolayer. A list of head group strength versus a sixteen chain hydrocarbon tail is shown in
Table 6-6.
6.5 Langmuir-Blodgett Films for Sensors 185

a- HYDROPHOBIC TAIL

y HYDROPHILIC HEAD

Figure 6-7.
Langmuir-Blodgett film: (a) schematic representation;
(b) the molecules orient themselves such that the head
group is on the water’s surface and the tail is sticking
upward. WATER SUBPHASE
b

Table 6-5. Saturated fatty acids

H3C - (CH2)n-z - COOH


n = 16-20
n Common Name (Naturally Occuring n =even)
16 Palmitic acid
17 Magaric acid
18 Stearic acid
19
20 Arachidic acid
21
22 Behenic acid

Table 6-6. Head group strength versus a C,, Hydrocarbon Chain (after Gaines [68])

Very Weak Weak Strong Very Strong


(no film) (unstable) (stable films) (dissolves)
hydrocarbon - CHZOCH, - CH,OH - so, -
- CHJ - C,H,OCH, - COOH - oso,-
-CH,Br - COOCH, - CN -C6H4S04-
- CHZC1 - CONH, - NR, +

-NO3 -CH=NOH
- C6HdOH
- CH2COCH3
- NHCONH,
- NHCOCH,

Because the film is constrained to move on the surface of the water, it can be thought of
as a two dimensional state of matter. The film can have the 2-D analogues of the three states
of matter - gas, liquid and solid - which are gaseous, expanded and condensed monolayers.
The state of the film is measured by surface pressure versus molecular area isotherm, as seen
in Figure 6-8. In the gaseous phase the molecules have a large area per molecule. Since the
186 6 Thin and Thick Films

4- COIIDEEISED

Figure 6-8.
A hypothetical surface pressure vs. molecular area
isotherm. The three phases of a monolayer are
shown: condensed, expanded and gaseous. The
CO-AREA MOLECULAR AKA co-area of the condensed film is also marked.

molecules are spread out far enough apart that they do not interact, the film can be com-
pressed with little resistance and the isotherm is flat. In the expanded phase, the film behaves
as a 2-D liquid, thus it has some viscosity and the isotherm shows pressure increases non-
linearly as the molecular area is decreased. The condensed film behaves as a solid, thus the
isotherm is nearly vertical.
At some point, as the surface pressure rises, enough force is applied and the film buckles.
This point is called the collapse pressure. It sets the upper limit of surface pressures at which
the film can be deposited.
There is also another way the film can collapse. When the film is held above its equilibrium
spreading pressure, it is no longer in thermodynamic equilibrium, and will tend to go into its
three dimensional bulk phase. The equilibrium spreading pressure (ESP) is the pressure at
which the LB film phase is in equilibrium with its three dimensional solid phase. It is
measured by putting a few crystals of the solid on a clean water surface and watching the
pressure rise, which can take many hours. This type of collapse usually starts at areas of high
stress, such as the corners of the moving barrier in a Langmuir trough.
This slow collapse can be noticeable in many systems, since most materials either have a
very low or almost zero ESP (ie, they prefer to remain as bulk crystals) or the ESP is in the
expanded phase, not in the condensed phase. Because of this, any impetus for the film to go
into its three dimensional phase must be eliminated: (1) the film balances must be carefully
designed to minimize the high stress areas in the film due to the mechanical design of the
trough; (2) the water for the subphase must be very pure (18 megaohm-cm and without any
surfactant impurities); and (3) the spreading solvents and any added impurity ions must be
very pure.
The low ESP of most materials makes it necessary to spread the material using a solvent.
The requirements for the spreading solvent are: (1) that it will disolve the material, which
seems obvious, but many of the large organic molecules are insoluble or very slightly soluble
in the common solvents; (2) that it be inert with respect to the material and the water sub-
phase; and (3) the solvent must be volatile - all of the solvent must leave the surface before
the film can be compressed or the properties of the film will be altered.
The molecular area which can be extrapolated from the vertical portion of the isotherm
back to zero pressure is called the co-area. The co-area is assumed to be the molecular area
per molecule in the close packed film. For a saturated fatty acid the co-area is about 0.2 nm2
WI.
6.5 Langmuir-Blodgett Films f o r Sensors 181

The pH and ionic content of the water subphase can greatly affect the properties of the
monolayer. Some head groups can be ionized depending on the pH of the -water subphase,
for example, an acid head group can be ionized by a basic subphase and vice versa [68]. This
causes the molecules to be more strongly attracted to the water surface. Some ionic impurities
can strengthen the film. It is well known that divalent cations will bind pairs of fatty acid
molecules together at the surface, for example, cadmium stearate [84].
The deposited fatty acid films are soft and have low melting points, around 70°C [49]. Thus,
more robust films are needed for electronic devices and sensors. Polymerizable derivatives of
the traditional fatty acids have been used, such as vinyl stearate and the diacetylenic acids
(H-(CH,),-C=C-C=C-(CH,),-COOH, n = 0-16 and m = 0-8). Both of these
materials can be polymerized by radiation (gamma for vinyl stearate and UV for the diacety-
lenes) before or after deposition.
Another method to increase the temperature stability and robustness of the LB films is to
use large aromatic compounds. One example of these molecules are the phthalocyanines
(Figure 6-9). The first work involved adding pendant groups to the ring to make a film forming
derivative. These phthalocyanine derivatives did not sit flat on the water surface as determined
by their isotherms [HI. Thus, the quality of the deposited film was in question. Other
“molecular engineering” approaches followed. One of the most successful ways was to stack
more than one phthalocyanine ring on top on one another using a (-SiO-) backbone for the
molecule [86].

Figure 6-9. The phthalocyanine ring.

The major reason for interest in the phthalocyanine molecule for sensors is that a p-type
semiconductor and its conductivity is altered when exposed to oxidizing gases. Another ad-
vantage is that a metal ion can be placed in the center of the ring, which will alter the intrinsic
conductivity of the molecule, giving a wide range of different possible conductivities.

6.5.2 Film Forming Apparatus (Film Balances)

The typical film balance consists of a trough, a surface pressure transducer, a movable bar-
rier, for compressing the film, and a dipping apparatus.
There are two types of pressure transducers commonly used: the Wilhelmy plate and the
Langmuir style float connected to torsion wire or a spring, as shown in Figure 6-10. In the
Wilhelmy plate method, the film pulls down on a piece of filter paper which is suspended from
an electro-balance through the film. The downward force is a direct measure of the surface
pressure. In the Langmuir method, a float separates the film from the clean surface of the
water. The float is connected to a torsion wire or a leaf spring so the deflection of the float
by the film is a measure of the surface pressure. The deflection of the float is usually measured
by a LVDT (linear variable differential transformer).
188 6 Thin and Thick Films

ELECTROBALANCE

~ be LEAF SPRING

Figure 6-10.
The Wilhelmy plate (upper) and the
Langmuir float style (lower) pressure
I TROUGH transducers.

There are two basic types of troughs, the Langmuir trough [49], and the constant perimeter
trough [87]. The Langmuir trough is usually machined out of metal about a centimeter deep
and coated with a hydrophobic material, such as teflon. The float is placed about 10 or 15 cm
from one end and the dipping well (for the substrate to enter when it goes through the film)
is usally placed about 10 cm from the float. This distance is necessary to allow the film to
flow behind the substrate and deposit on the back of the substrate. The moving barrier slides
along the edges of the trough, therefore square edges and a flat barrier are imperative.
The constant perimeter trough uses a set of moving rollers and a continuous tape to com-
press the film. The water level is such that about half the tape is in the water. By moving the
rollers, the area of the trough is decreased without changing the perimeter. The major reason
for this design was to eliminate the possibility of leakage which exists in a poorly constructed
or poorly maintained Langmuir trough. This trough must use the Wilhelmy plate as its
pressure transducer.
The dipping apparatus usually consists of an electric motor or hydraulic piston elevating
a shaft or wire on which the substrate is attached. The apparatus is usually computer con-
trolled with adjustable dipping rates (typically, from one millimeter to a few centimeters per
minute) for both up and down (these times may be different to allow for water drainage on
the up trip) and adjustable drying times between round trips.
The film balance must be isolated from the environment to produce high quality repeatable
LB films. The major disturbances to a LB film are vibration, temperature variations and air-
borne contaminants, which can be eliminated just by using a cover, but the person working
on the trough is a major source of particles, so keeping the covered trough in a clean room
is advisable. Temperature variations can be eliminated by either keeping the trough in a con-
stant temperature enclosure or by circulating a constant temperature fluid under the bottom
of the trough. Vibrations have two sources, the room and the dipper. The room vibrations can
6.5 Langmuir-Blodgett Films for Sensors 189

be reduced by vibration isolation tables. Passive isolation tables are usually sufficient, and
much more inexpensive than active tables. The dipping apparatus induced vibration can be
reduced by careful mechanical design.

6.5.3 Dipping

Since the goal of the LB technique is to deposit solid films on a substrate, the films are
deposited from the close packed phase, the condensed monolayer. There are two methods of
depositing the monolayers, vertical and horizontal dipping. In vertical dipping the film is com-
pressed and the substrate is dipped through the film perpendicular to the film (Figure 6-lla).
The substrate can either start out of the water or in the water. This depends on the type of
substrate. The hydrocarbon tails will not stick to a hydrophilic substrate, so starting out in the
water will save the time of one down trip. In this manner, many layers can easily be added
from one film.
Vertical dipping is the usual manner of deposition, but when either a particular order is
desired in the film (such as alternating layers of two different materials) or the film will not
deposit well in vertical dipping, horizontal dipping becomes necessary. The film is compressed
and the substrate is plunged through the film while it is parallel to the film (Figure 6-llb).
Again, it can either start out in the water or out of the water. The major disadvantage is that
since the substrate punches a large hole in the film, it usually collapses, and each film will
yield only one deposited monolayer. These depositions require much time and material.

Figure 6-11.
Deposition of Monolayers: (a) vertical
deposition showing both the down trip and
the up trip; (b) horizontal deposition.
190 6 Thin and Thick Films

There are three types of LB films - X, Y, and Z. These are characterized by the way in
which the film is oriented on the substrate after deposition, as seen in Figure 6-12. Normal
vertical deposition results in Y deposition: on the down trip the film deposits with the tail
group sticking to the substrate and on the up trip the head groups are touching (a head to
head and tail to tail arrangement). The other two types of deposition involve one way deposi-
tion. X deposition is the more common one - the film only deposits in the down trip. The
Z type is just the opposite, though it is very rare.

Figure 6-12. The three types of Langmuir-Blodgett films: X-type, Y-type and Z-type.

The deposition ratio is the ratio of the area on which the film is deposited to the area of
material removed from the trough. Ideally, the ratio should be one for deposition and zero
for no deposition.

6.5.4 I o n Sensors

Since a LB bilayer is analogous to a cell wall, biosensors are a natural application for
LB films. One obvious application is for an ion selective membrane on the insulator of an ion
sensitive field effect transistor (ISFET). Since the standard ionophores will not form LB films,
one possible method is to form a mixed monolayer of ionophore and LB material, such as
valinomycin and stearic acid [88]. This allows the controlled deposition of thin layers of a film
containing ionophores which can selectively transport ions from the solution to the gate in-
sulator, decreasing the interference effects of other ions in solution.
Another use for LB films is to support immobilizing enzymes on an ISFET gate. In par-
ticular, barium stearate multilayers have been used on the Si,N, gate of an ISFET [89]. The
LB film reduced the pH sensitivity of the ISFET, but showed stable, long term immobilization
of penicillinase without the inactivation that is usually seen when proteins and peptides ad-
sorb on a surface and unfold.
Stearic acid layers have also been used to protect the quartz substrate and aluminium elec-
trodes of a surface acoustic wave (SAW) immunosensor [90]. The antigen and antibodies were
immobilized in a polyacrylamide gel on top of the LB layers. The stearic acid layers also im-
proved the adhesion of the polymer gel on the hydrophilic quartz.

6.5.5 Gas Sensors

LB based gas sensors reported to date can be divided into two groups, chemiresistors and
surface acoustic wave (SAW) devices. The chemiresistors use semiconducting LB materials
6.7 References 191

which are oxidized (doped) by the gas to be sensed. The change in conductivity is directly
related to the concentration of the test gas. Examples of these materials are the phthalo-
cyanines and the porphyrins [91-961, which are sensitive to oxidizing gases.
SAW gas sensors depend on the change in vibrational frequency of the piezoelectric delay
line due to the adsorption of gas by the LB film deposited on the piezoelectric. Again,
phthalocyanines have been used as the adsorbing layer since they are very reactive to oxidizing
gases [97, 981.

6.6 Concluding Remarks

In the thick film and thin film areas, progress in technology has outpaced science. This
situation is particularly serious in the sensor areas. Semiconductor gas sensors using
thick/thin film material and processes were invented more than a quarter of a century ago,
and since then a wide variety of such sensors have been produced and used in many different
places. However, comprehensive sensing models to understand materials and technology have
yet to be developed. Crucial issues, such as selectivity, stability and doping effects of material
composition, deposition processes (rates, ambience, temperature, and methods), and the ensu-
ing thermal treatment remain partially solved. Consequently, applications of semiconductor
gas sensors (including chemFETS) have been quite limited. The recent push of using silicon
micromechanical structures for the active elements of sensors and actuators seems to be
heavily biased toward technology development. To ensure success in this new sensor field,
serious and systematic programs on the micromechanical sciences related to technologies (in-
cluding thick and thin film technologies) must be established and vigorously implemented.

6.7 References

[l] Jones, R. D., Hybrid Circuit Design and Manufacture; New York: Marcel Dekker, Inc., 1982.
[2] Topfer, M. L., Thick-filmMicroelectronics. Fabrication, Design and Applications; New York: Van
Nostrand Reinhold, 1971.
[3] Rikoski, R. A., Hybrid Microelectronic Circuits: The Thick-film; New York: John Wiley, 1973.
141 Holmers, P. J., Loasby, R. G., Handbook of Thick Film Technology;Electrochemical Publications
Ltd., 1976, Chapters 5, 6 and 7.
[5] Vossen, J. L., Kern, W. K., Thin Film Processes; New York: Academic Press, 1978.
[6] Chopra, K. L., Thin Film Phenomena, Melbourne: Krieger, 1979.
17) Chang, S. C., Hicks, D. B., Proc. IEEE Solid State Sensors Workshop, Hilton Head, SC, 1986.
[8] Oyabu, T., Kurobe, T., Hidai, T., Proc. Intl. Meeting on Chem. Sensors, Fukuoka, Japan, 1983,
p. 13.
[9] Morrison, S. R., Sens. Actuators 2 (1982) 329.
[lo] Tsuchitani, S., Sugawara, T., Kinjo, N., O’Hara, S., Tech. Digest, Intl. ConJ on Solid State Sensors
and Actuators, 1985, p. 210.
1111 Jaffe, J. M., Electronic Letters 10 (1974) 420.
192 6 Thin and Thick Films

1121 Muller, R., Sens. Actuators 4 (1983) 173.


113) Shimazoe, M., Matsuoka, Y., Yasukawa, A., Tanabe, M., Proc. 1st Sensor Symposium(1981) 135.
1141 Aske, V. H., Scientvic Honeyweller (1987) 53.
1151 Polla, D. L., Muller, R. S., White, R. M., IEDM Technical Digest (1984) 382.
1161 Petersen, K. E., Shartel, A., Raley, N. F., IEEE Trans. on Electron Devices 29 (1982) 23.
1171 Heidler, G. R., “Depositions for Microelectronics” in Handbook of Electronic Packaging, Harper,
C. A. (ed.); McGraw Hill, 1969.
1181 Pike, G. E., Seager, C. H. Journal of Applied Physics 48 (1977) 5151.
1191 Tsunaga, M., Kato, S., Hybrid Circuits 16 (1988) 32.
1201 Vest, R. W., Chitale, S. M., Kollipara, A. K., Proc. 5th European Hybrid Microelectronics ConJ,
Stresa, Italy (1985).
1211 Morten, B., Prudenziati, M., Tarani, A., Sens. Actuators, 4 (1983) 237.
1221 Dell’Acqua, R., Hybrid Circuits, 12 (1987) 11.
1231 Belford, R. E., Kelly, R. G., Owen, A. E., “Thick Film Devices” in: Chemical Sensors, T. E. Ed-
wards (ed.); New York: Chapman and Hall, 1988.
[24] Kanefusa, S., Nitta, M., Haradime, M., Proc. Intl. Meeting on Cbem. Sensors, Fukuoka, Japan,
1983, p. 84.
1251 Brignell, J . E., White, N. M., Cranny, A. W. J., Proc. IEE W5, No. 4 (1988) 77-84.
1261 Prudenziati, M., Tech Digest, 4th Intl. ConJ on Solid State Sensors and Actuators (Transducers
’87), Tokyo, Japan, 1987, p. 85.
[27] Kirkpatrick, S., Reviews ofModern Physics 45 (1973) 574.
[28] Dell’Acqua, R., Dell’orto, G., Forani, F., Puzone, A., Tech. Paper 920319, 1978, Society of
Automotive Engineers.
1291 Prudenziati, M., Morten, B., Taroni, A., Sens. Actuators 2 (1981/82) 17.
1301 Van Dorth, A. C., van der Graaf, F., Steenvorden, G. K., Sens. Actuators 4 (1983) 323-331.
131) Arima, H., Ikegami, A., Abe, K., Iwanaga, S., Isogai, T., Proc. ISHMSymposium.New York, 1980,
272-278.
[32] Gondek, J. J., Wojcicki,M. A., Electrocomp. Sci. Technol. 10 (1983) 95-102.
[33] Marx, G. M., Bell, R. L., SAE Paper 780214.
1341 Dell’Acqua, R., Dell’Orto, G., Simonetta, A., Canali, C., Int. J. Hybrid Microel. 58 (1982) 82-84.
(351 Ikegami, A., Arima, H., Iwanaga, S., Kaneyasu, M., 4th European Hybrid Microel. Con$,
Copenhagen, 1983, 211-218.
1361 Nitta, M., Haradome, M., IEEE Trans. Electron. Dev. ED-26(1979) 247-249.
I371 Oyabu, T., Kurobe, T., J. Appl. Pbys. 53 (1982) 7125-7130.
1381 Schonauer, U., Techniscbes Messen 56 (1989) 260-263.
1391 Chu, W. F., Technisches Messen 56 (1989) 255-259.
1401 Channon, N. D., I Soc. Environ. Eng., (Sept. 1979) 23-25.
[41] Micheli, A. L., Chang, S. C., Hicks, D. B., Ceramic Engineering and Science Proc. 8 (1987) 1095.
[42] Hamdi, A. H., et al., Applied Physics Letters 51 (1987) 2152.
1431 Hoffmann, D., Dynamische Ternperaturmessung, Berlin: VEB-Verlag Technik, 1976.
1441 Tschulena, G., Selders, M., Technisches Messen 50 (1983) 127.
1451 Dossel, O., NTGFachberichte 93, Berlin: VDE-Verlag, 1986, pp. 143- 147.
1461 Graeger, B., Schafer, H., Kobs, R., VDI Bericbte 677, Dusseldorf: VDI, 1988, pp. 99-104.
1471 Fischer, H., Miiller, J., WeiRenrieder, S., Kettner, T., VDI Berichte 677, Dusseldorf: VDI, 1988,
pp. 105-113.
(481 Dibbern, U.,Sens. Actuators 10 (1986) 127-140.
(491 Lampe, U., Miiller, J., NTGFachberichte 93, Berlin: VDE-Verlag, 1986, pp. 87-92.
[SO] Luder, E., KallfaD, T., Borgwardt, C., N7G-Fuchberichte79, Berlin: VDE-Verlag, 1982, pp. 299-303.
1511 Fan, L. S., Muller, R. S., IEEE Solid State Sensor and Actuator Workshop, Hilton Head, SC, 1988.
1521 Guckel, H., Burns, D. W., Tilmans, H. A. C., Deroo, D. W., Rutigliano, C. R., IEEE Solid State
Sensor and Actuator Workshop,Hilton Head, SC (1988).
6.7 References 193

(531 Howe, R. T., Muller, R. S., Journal of the Electrochem. Society I30 (1983) 1421.
[54] Chang, S. C., Proc. Int. Meeting on Chemical Sensors, Fukuoka, Japan, 1983, p. 78.
[55] Capehart, T. W., Chang, S. C., Journal of Vakuum Science Technology 18 (1981) 393.
(561 Bruce, R. H., Reinberg, A. R., Journal of the Electrochem. Society 129 (1982) 393.
[57] Burton, R. H., Smolinsky, G., Journal of the Electrochem. Society 129 (1982) 1599.
[58] Curran, J. E., Thin Solid Films 86 (1981) 101.
[59] Chang, S. C., LI S. Patent 4544444, 1985.
[60] Franklin, B., Philosophical Transactions of the Royal Society 64 (1774) 445.
[61] Pockels, A., Nature 43 (1891) 437.
[62] Lord Rayliegh, Proceedings of the Royal Society, 47 (1890) 364.
(631 Lord Rayliegh, Philosopy Magazine, 48 (1899) 337.
[64] Langmuir, I., Journal of the American Chemical Society 38 (1916) 221.
[65] Langmuir, I., Journal of the American Chemical Society 38 (1917) 1848.
[66] Blodgett, K., Langmuir, I., Physics Review 51 (1937) 964.
[67] Blodgett, K., Journal of the American Chemical Society 57 (1935) 1007.
[68] Gaines, Jr., G. L., Insoluble Monolayers at Liquid-Gas Interfaces; New York: John Wiley 1966.
[69] Tanguy, J., Thin Solid Films I3 (1972) 33.
[70] Lundstrom, I., McQueen, D., Chemistry and Physics of Lipids 10 (1973) 181.
[71] Roberts, G. G., Pande, K. P., Barlow, W. A., Solid State and Electronic Devices 2 (1978) 169.
[72] Kan, K. K., Petty, M. C., Roberts, G. G., “Polymerized Langmuir Film MIS Structures” in: The
Physics of the MOS Insulators, Proc. Intl. Topical ConJ, NC, June 18-20, 1980, Lukovsky, G.,
Pantelides, S. T., Galeener, F. L. (eds.); New York: Pergamon Press, 1980, p. 344.
(731 Lloyd, J. P., Petty, M. C., Roberts, G. G., Lecomber, P. G., Spear, W. E., Thin Solid Films 89 (1982)
395.
[74] Lloyd, J. P., Petty, M. C., Roberts, G. G., Lecomber, P. G., Spear, W. E., Thin Solid Films 99 (1983)
297.
[75] Dewa, A. S., Fung, C. D., DiPoto, E. P., Rickert, S. E., Thin Solid Films I32 (1985) 27.
[76] Fung, C. D., Larkins, G. L., Thin Solid Films I32 (1985) 33.
[77] Tabib-Azar, M., Dewa, A. S., KO,W. H., Applied Physics Letters 52 (1988) 206.
[78] Shutt, J. D., Rickert, S. E., Journal of Molecular Electronics 4 (1988) 201.
[79] Vincent, P. S., Roberts, 0. G., Thin Solid Films 68 (1980) 135.
[80] Roberts, G. G., “Langmuir-Blodgett Films on Semiconductors” in : Insulating Films on Semicon-
ductors, Proc. 2nd Intl. Conj, INFOS 81, Erlangen, Fed. Rep. of Germany, Shulz, M., Pensl, G.
(eds.); New York: Springer-Verlag, 1981, p. 56.
[81] Roberts, G. G., Sens. Actuators 4 (1983) 131.
[82] Sugi, M., Journal of Molecular Electronics 1 (1985) 3.
[83] Roberts, G. G., Advances in Physics 34 (1985) 475.
[84] Langmuir, 1. Schaefer, V. J., Journal of the American Chemical Society 59 (1937) 2400.
[85] Roberts, G. G., Petty, M. C., Baker, S., Fowler, M. T., Thomas, N. J., Thin Solid Films I32 (1985)
113.
(861 Shutt, J. D., Batzel, D. A., Sudiwala, R. V., Rickert, S. E., Kenney, M. E., Langmuir 4 (1988) 1240.
[87] Roberts, G. G., Vincent, P. S., Barlow, W. A., Phys. Technol. 12 (1981) 69.
[88] Brown, A. D., Sens. Actuators 6 (1984) 151.
[89] Anzai, J., Hashimoto, J., Osa, T., Matsuo, T., Analytical Sciences 4 (1988) 247.
[90] Thompson, M., Dhaliwal, G. K., Arthur, C. L., Calabrese, 0. S., IEEE Trans. on Ultrasonics, Fer-
roelectrics and Frequency Control UFFC-34 (1987) 127.
[91] Baker, S., Roberts, G. G., Petty, M. C., IEE Proceedings I30 (1983) 260.
[92] Wohltjen, H., Barger, W. R., Snow, A. W., Jarvis, N. L., IEEE Trans. on Electron Devices 32 (1985)
1170.
[93] Tredgold, R. H., Young, M. C. J., Hodge, P., Hoorfar, A., ZEEE Proceedings I32 (1985) 151.
194 6 Thin and Thick Films

[94] Grate, J. W., Rose-Pehrsson, S., Barger, W. R., Lmgmuir 4 (1988) 1293.
[95] Fu, C. W., Batzel, D. A., Rickert, S. E., KO, W. H., Fung, C. D., Kenney, M. E., Tech. Digest, 4th
Int. ConJ on Solid State Sensors and Actuators (Transducers '87), Tokyo, Japan, June 1987, 1987,
p. 583.
[96] Wang, H. Y., Batzel, D. A., KO, W. H., Lando, J. B., Kenney, M. E., Abstracts from the 5th Int.
Con$ on Solid State Sensors and Actuators (Tkansducers '89), Montreux, Switzerland, June 25-30,
1989, in press.
[97] Snow, A. W., Barger, W. R., Klusky, M., Wohltjen, H., Jarvis, N. L., Lungmuir 2 (1986) 513.
[98] Roberts, G. G., Holcroft, B., Ross, J., Barraud, A., Richard, J., British Polymer Journal 19 (1987)
401.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

7 Integrated Magnetic Sensors


HENRYBALTES.Swiss Federal Institute of Technology (ETH). Zurich.
Switzerland.
AROKIA University of Waterloo. Waterloo. Ontario. Canada
NATHAN.

Contents
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
7.1.1 Overview of Magnetic Field Sensor (MFS) Technologies . . . . . . . . 196
7.1.2 Summary of Pertinent Semiconductor Magnetic Effects . . . . . . . . 197
7.2 Semiconductor MFS Materials and Technologies . . . . . . . . . . . 202
7.2.1 MFS Design Selection Criteria . . . . . . . . . . . . . . . . . . . 202
7.2.2 Materials and Figures of Merit . . . . . . . . . . . . . . . . . . . 202
7.2.3 New Technologies and Examples . . . . . . . . . . . . . . . . . . 204
7.3 Standard Silicon MFS Technologies . . . . . . . . . . . . . . . . . 207
7.3.1 MFS Design by Standard IC Technologies . . . . . . . . . . . . . . 207
7.3.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
7.3.3 Limitations Imposed by Standard Technologies . . . . . . . . . . . . 210
7.3.4 Noise in Magnetotransistors . . . . . . . . . . . . . . . . . . . . 211
7.3.5 3-D Vector Sensors . . . . . . . . . . . . . . . . . . . . . . . . 212
7.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 13
196 7 Integrated Magnetic Sensors

7.1 Introduction

Magnetic field sensors can be realized using a wide variety of different technologies. The
various kinds of magnetic sensors and their pertinent technologies are discussed in Volume 5
(devoted entirely to magnetic sensors) of this series. In contrast, this chapter reviews the
materials, technologies, and the potential of integrated magnetic field sensors or semiconduc-
tor magnetic microsensors. Although this is just one of many possible choices for magnetic
sensors, we felt that in conjunction with the preceding discussion of general semiconductor
technologies in Chapter 5 , this choice fits in well with the present volume. Figures of merit
and the advantages and limitations of the different semiconductor materials are discussed,
together with selected examples. Special emphasis is given to magnetic sensor design and
fabrication using standard IC technologies.

7.1.1 Overview of Magnetic Field Sensor (MFS) Technologies

Most magnetic field sensors (MFS) exploit the Lorentz force F = q u x B on moving charge
carriers in a metal, semiconductor, or an insulator in one way or another. Here, q denotes the
electron charge, u the electron velocity, and B the magnetic induction vector. In view of the
relation B = bpo H where ,upo denotes the magnetic permeability of the sensor material and
H the magnetic field vector (the measurand), we can readily distinguish two major classes of
MFS [l]:
(i) Magnetic field sensing using high-permeability (ferro- or ferrimagnetic) materials, where
p 1 brings about a corresponding enhancement of sensitivity. Examples are MFS based on
the magnetoresistance of NiFe thin films, the magnetostriction of the nickel cladding of an
optical fiber, or the magnetooptic effects in garnets, and any MFS combined with a flux-con-
centrating device.
(ii) Magnetic field sensing using low-permeability (dia- or paramagnetic) materials, where
p = 1 does not provide any appreciable “leverage”. For example, all MFS based on
galvanomagnetic effects in semiconductors belong to this class.
Major types of MFS are summarized as follows:
Thin-metalfilm or wire MFS are based on ferromagnetic materials. The low-magnetostric-
tion alloy Ni,,Fe,g is preferably used for thin-film MFS. The most successful sensor effect is
the magnetoresistive (MR) switching of anisotropic NiFe or NiCo films [2-61. High spatial
resolution applications require the reduction of the MR element to micrometer dimensions,
where Barkhausen noise may degrade the MR switching characteristics. This problem can be
overcome by using sandwich structures of two NiFe layers with proper orientation of easy axis
separated by a silicon monoxide or conductive layer [3, 71. This type of structure reduces
demagnetizing fields and Barkhausen noise.
Optoelectronic MFS use light as an intermediate signal carrier. Magnetooptic MFS are
based on the Faraday rotation of the polarization plane of linearly polarized light due to
Lorentz force on bound electrons. Useful MFS can be realized by using optical-fiber coils pro-
viding a long light path and an accordingly large rotation per unit magnetic field. Magnetoop-
tic sensors for the current in high voltage transmission lines have‘been realized in this way [S,
91. A much larger Faraday rotation angle per unit path length is obtained in optically transpa-
7.1 Introduction 197

rent ferrimagnetic garnet materials [lo]. The most sensitive optoelectronic MFS are those
using optical fibers with magnetostrictive jacketing material such as nickel or an optical fiber
wound under tension on to a cylinder of magnetostrictive material. The strain transferred to
the fiber from the magnetostrictive material results in a change of optical path length, leading
to a phase shift detected with an optical-fiber interferometer [ll]. Minimum detectable fields
of less than 1 nT have been reported [12].
Semiconductor MFS exploit the galvanomagnetic effects such as Hall voltage, carrier
deflection, magnetoresistance, and magnetoconcentration, all of which are due to the action
of the Lorentz force on moving charge carriers (electrons and holes). Semiconductor MFS
have several advantages: they are flexible in design and application, small in size, rugged, and
they provide an electronic signal output. Semiconductor MFS can be fabricated using silicon,
germanium, and 111-V materials. Silicon MFS are by far the least expensive because of the
highly reliable and well-established standard IC technology. Fabricated devices include bulk
and inversion layer Hall elements, magnetotransistors, magnetodiodes, and carrier domain
magnetometers. Elaborate reviews of integrated silicon MFS are available [l, 13, 141. Certain
magnetic sensors fabricated using 111-V compounds (such as GaAs, InAs, and InSb) are more
sensitive to the magnetic field because of the relatively higher carrier mobility in these
materials. 111-V materials are used in the form of Hall and magnetoresistive devices in a variety
of applications [15].
Important high-resolution MFS worth mentioning, but not fitting in the above groups, are
SQUID magnetometers, the most sensitive of all MFS (see [16] or Chapter 10 of Volume 5
of this series). SQUIDS are sensitive to magnetic fields in the picotesla range and hence par-
ticularly useful in biomedical applications [17]. They exploit the quantum-mechanical galva-
nomagnetic effects observed in certain metals that are cooled to temperatures close to absolute
zero, where they become superconductive. The periodic variations in the induced current flow-
ing through a superconducting ring (lead or niobium), interrupted by a weak link (tunnel junc-
tion), is a function of the magnetic flux density threading the ring [HI. In an rf SQUID, the
ring is inductively coupled to an rf circuit and the induced current modulates the resonant fre-
quency of the circuit. By employing appropriate feedback circuits, the feedback current can
be monitored as a measure of the flux density. The superconducting ring can be coupled to
a superconducting search coil to achieve improvement in sensitivity. There have been various
successful attempts to integrate the SQUID and the superconducting coil together with the
feedback and signal conversion circuitry on a single chip ([19] and references cited therein).

7.1.2 Summary of Pertinent Semiconductor Magnetic Effects

The action of the Lorentz force manifests itself in the carrier transport relations ([l] and
references cited therein). We assume an isotropic n-type material with zero temperature
gradient. Let us denote the electron current density for zero magnetic induction ( B = 0) by
Jnp. The diffusion approximation of the Boltzmann transport equation leads to [20, 211

where on = qpnn denotes the electronic conductivity for B = 0, E denotes the electrical
field, Dn = p n k, T/q the diffusion coefficient, n the electron density, and pn the electron
198 7 Integrated Magnetic Sensors

drift mobility. For non-zero magnetic induction, the electron current density J,, obeys the
equation

Jn, B = Jn.0 - PU,* [ J n , B X BI, (7-2)

where p: is the Hall mobility for electrons. The Hall mobility is proportional to the drift
mobility, viz, p: = rnpnwith r, denoting the Hall scattering coefficient [l, 141. Equation
(7-2) can be solved with respect to J,, B , viz,

This equation expresses the isothermal galvanomagnetic effects for electrons. It accounts
for the direct effects of temperature on carrier concentration, diffusion coefficient, and
mobility, but does not include thermomagnetic or thermoelectric effects. An analogous equa-
tion holds for the hole current density. If carrier concentration gradients can be neglected, as,
eg, in n-type slabs with ohmic contacts, Equation (7-3) becomes

where on,B= on [l + (p: B ) 2 ]-'. If B is parallel to E, B x E = 0 leads to J n , B = o , E =


J,, : no longitudinal galvanomagnetic effect is observed in isotropic semiconductors. For B
-
perpendicular to E, B E = 0, and we obtain

This equation describes the transverse galvanomagnetic effects in the case of negligible dif-
fusion. In terms of B = (0, 0, B), E = (Ex, Ey, 0), and J,, = (Jn,x,J n , y , O), Equation (7-5)
reads

The following two limiting cases are usually distinguished:


(i)HaNfield. It is assumed that the current density has only an x-component, i. e., J,, =
0. This condition can be approximated by a sample that is long and whose cross section is
narrow, with current electrodes at the small faces (see Figure 7-1). The resulting Hall field is

where

R, = -p;/o, = -rn/qn (7-8)

denotes the Hall coefficient. For mixed n- and p-type conduction, the Hall coefficient is
suitably modified [ l , 141. The presence of the Hall field results in a rotation of the equipoten-
tial lines by the Hall angle 0, with
7.1 Introduction 199

tan 8, = E y / E x = -& B = 6,RH B. (7-9)

From Equations (7-7) and (7-8), it can be seen that a large Hall mobility and small carrier
concentration produce large Hall fields Ey . This explains why semiconductors are more
useful here than metals. A detailed timely review of the Hall effect and related sensors can
be found in [14].
(ii) Carrier deflection and magnetoresistance. Here a zero Hall field, Ey = 0, is assumed.
This condition can be realized approximately by a sample that is short and whose cross section
is wide, with current electrodes at the large faces (see Figure 7-2). Under this condition, the
Equations (7-6) lead to a rotation of the current lines (the so-called carrier deflection or
Lorentz deflection), described by the ratio

-J,,y/J,,x = pu,*B = tan 8,. (7-10)

Carrier deflection is commonly exploited in a wide variety of dual-collector and dual-drain


magnetic field sensitive transistors [l]. In addition, the carrier deflection due to the Lorentz
force enhances the total length of the carrier drift path. This leads to the geometrical
magnetoresistance effect:
(pn,B - pn)/pn = ( ~ u , * BIZ 9 (7-11)

where pn = I/a, is the resistivity for B = 0 and p,, = E x / J,, = Wan, the resistivity
enhanced by the magnetic induction. This effect is very small for silicon (eg, pn, - 1.02 pn
for B as large as 1 T), but large in some 111-V compounds, notably InSb and InAs.
It should be noted that the above relations are valid only in very special configurations
characterized by certain device geometry, doping, magnetic field distribution, boundary, and
operating conditions, where the one or the other galvanomagnetic effect may prevail [I, 14,
151. In order to clarify the role of the pertinent galvanomagnetic effects (and their possible
interactions) for more general device geometries, magnetic field distributions, and operating
conditions, the electron and hole continuity equations together with Poisson’s equation have
to be numerically solved subject to realistic material parameters and boundary conditions. The
corresponding microsensor modeling problems are further discussed in Chapter 3 of this
volume.
Strictly speaking, the relations derived for the Hall field [Equations (7-7) to (7-9)] and car-
rier deflection [Equation (7-lo)] for semiconductor plates are valid only in the limit of infinite
and zero length-to-width ratio L / W , respectively. These relations, however, are reasonable ap-
proximations for L / W L 4 and L / W 5 1/4, respectively, as is demonstrated by the numerical
modeling results [22] shown in Figures 7-1 and 7-2. The former illustrates the Hall effect in
a semiconductor slab of length L = 4 W with metal electrodes at the narrow top and bottom
edges. It shows the resulting current line and equipotential line distributions in two identical
n-type silicon plates when subject to perpendicular magnetic inductions, flu,*B = 0.21 and
0.42. Apart from boundary effects close to the electrodes, the vertical current lines (which start
and end at the electrodes) are the same as for zero induction, in support of the previous
assumption, Jn,u = 0, whereas the equipotential lines are rotated by the Hall angle, in agree-
ment with the analytical result (7-9). Figure 7-2 illustrates the (opposite) carrier deflection ef-
fect for two different short (L = W/4) plates, one of n-type material with pu,*B = 0.21 and
one of p-type material with pu,*B = 0.15, with contacts at the wide top and bottom edges.
200 7 Integrated Magnetic Sensors

L L

Figure 7-1.
Modeling results of carrier transport under magnetic field [22]
demonstrating the Hail effect in bulk silicon plate (n-type,
doping level 10l6cm--’) with long geometry (L = 4 W) and
ohmic contacts at top and bottom (0.1 V supply voltage).
Current lines (connecting the contacts) and equipotential lines
(running from left to right) are shown for two different values of
the product of Hall-mobility and magnetic induction (induction
vector perpendicular to the drawing plane): (a) fit B = 0.21, and
(b) ,ut B = 0.42. Rotation of the equipotential lines from the
B = 0 horizontal position (Hall effect) is the prevailing effect.
0 w 0 W The vertical current lines are hardly affected by the magnetic
a b field.

a Figure 7-2.
(a) Same as Figure 7-1, but demonstrating the
carrier deflection effect for a wide geometry
0
w (W = 4 L) and B = 0.21. (b) Same wide
geometry, but p-type doping of 10l6cm --’ and
vx B = 0.15. Rotation of current lines from the
B = 0 vertical position (carrier deflection) prevails.
The different directions of deflection of current
lines in (a) and (b) are due to the opposite charge
of the carriers. The vertical equipotential lines are
U
w hardly affected by the magnetic field.

In either case, the equipotential lines are almost horizontal, in support of the earlier assump-
tion Ey = 0, but now the current lines are rotated in agreement with Equation (7-10).
Similarly, Hall voltage and carrier deflection are dominant galvanomagnetic effects in very
long and very short inversion layers [23], respectively. Under certain specific configurations
of nonuniform magnetic fields such as longitudinal and transverse strip domains, Hall effect
and carrier deflection, respectively, dominate the transport pattern in the device [24]. With
more general device geometries and field configurations, one cannot expect either the Hall
rotation or carrier deflection models to describe adequately the current and potential distribu-
tions. For such geometries, numerical modeling becomes indispensable in predicting the trans-
port pattern. Indeed, carrier transport modeling results for a square (L = W)device geometry
under uniform fields [22, 231, given in Figure 7-3, show features reminiscent of both Hall ef-
fect and carrier deflection that blend in a unique way.
Diffusion phenomena (terms containing Vn) become important in the case of
rnagnetoconcentmtion or space-charge effects. These effects occur in magnetodiodes and in
bipolar magnetotransistors under the condition of high double injection of both electrons and
holes. An analytical description of magnetoconcentration and related effects can be involved,
even with the use of simplifying assumptions. Figure 7-4 demonstrates the magnetoconcentra-
7.1 Introduction 201

tion effect for a slab of nearly intrinsic material, where this effect dominates [25]. In contrast
to the extrinsic material (Figures 7-1 to 7-3), there is a build-up of electron and hole concentra-
L

Figure 7-3.
Same as Figure 7-1, but with square geometry ( W = L, p i B = 0.21).
Both current lines and equipotential lines are likewise affected by the
action of the magnetic field. The resulting pattern shows features
remir;iiscent of both Hall effect and carrier deflection. 0
W

L
P"
t
Y

r\

" i __
- x - w

1014
27

0
Figure 74. Modeling results of carrier transport under magnetic field demonstrating the magneto-
concentration effect [25] in a nearly intrinsic (T = 500 K) bulk silicon plate with square
geometry, p: B = 0.21, p; B = 0.07, and 0.1 V applied voltage. (a) Current lines (connec-
ting top and bottom ohmic contacts) and equipotential lines (approximately parallel to the
contacts). Current lines crowd to the right side of the plate (magnetoconcentration effect). (b)
Hole concentration. (c) Space-charge distribution. (d) Electron concentration.
202 7 Integrated Magnetic Sensors

tions to one side, leading to localized regions of high conductivity and consequently to current
crowding. The reverse happens on the other side, where there is a “depletion” of mobile
charge concentrations, leading to a localized increase in resistance.
Finally, we recall that Equations (7-2) to (7-11) correspond to the leading terms of a weak
field expansion and involve relative errors of (,ux B)’. They provide a good approximation
for (px B)’ < 0.1 which holds for n-type Si below B = 2 T, n-type GaAs below 600 mT, and
n-type InSb below 40 mT.

7.2 Semiconductor MFS Materials and Technologies

Each MFS application has particular sensitivity and resolution requirements. Switching and
displacement detection applications involve permanent magnets with fields of about 5 - 100 mT,
whereas 10 pT to 10 mT is the range of stray fields of magnetic domains in recording media.
A conductor carrying’ a current of 1 A produces a magnetic field of the order of 1 mT at the
surface of the conductor. The selection of the appropriate MFS technology depends on a
number of specifications that may vary widely from one application to the next.

7.2.1 MFS Design Selection Criteria

MFS share the general design selection criteria (environment, sensitivity, resolution,
linearity, etc., as outlined in Chapter 2 of this volume) pertinent to the majority of all sensors.
In addition, the design geometry of semiconductor MFS is often required to provide for sens-
ing desired component(s) of the magnetic field vector H. Once a specific MFS has been
designed and manufactured, its performance with respect to the crucial specifications must
be checked by appropriate measurements.
As an outstanding example, we mention a GaAs Hall device [26] fabricated by ion implanta-
tion and intended for magnetic flux meters operating between room temperature and 4 K. For
this purpose, high linearity, low temperature coefficient, and small, temperature-independent
offset are crucial requirements. Linearity and offset were checked by comparison with a Hall
device calibrated by NMR magnetometry. The specifications were achieved by tight control
of the device manufacturing technology and by exploiting the temperature stability of GaAs.
For other applications, a less costly strategy may be adopted. Using silicon IC technology,
in situ compensation of nonlinearity, offset, and temperature dependence can be attempted
by appropriate circuitry. Examples are the compensation of the non-linearity in integrated
Hall sensors caused by the junction field effect (see [14] and references cited therein) and the
offset reducing circuitry reported in [42].

7.2.2 Materials a n d Figures of Merit

The choice of semiconductor magnetic sensor material has to start from key figures of merit
such as output sensitivity and signal-to-noise ratio, temperature range and coefficient, and
inputloutput impedance for matching to circuitry components.
7.2 Semiconductor MFS Materials and Technologies 203

The magnitude of the underlying sensor effect (whether Hall effect, carrier deflection, or
magnetoresistance) is governed by the product of mobility and magnetic induction, p: B.
This favors materials with high mobility (given below in m2/Vs at room temperature). Thus
InSb (8 in bulk, 6.5 in thin film) and InAs (3) seem to be superior to GaAs (0.5 bulk, 0.7 in
the two-dimensional electron gas at the AlGaAs/GaAs hetero-interface), let alone Si (up to
0.14 in bulk and 0.07 in n-channel inversion layers). Generally, n-type semiconductor material
is superior to p-type material because of the much lower hole mobility (0.05 for bulk Si, 0.04
for GaAs, 0.14 for InSb) [2].
Apart from Hull mobility p:, three further figures of merit defined for Hall devices are
supply-current-reluted sensitivity VH/ZBwith Hall voltage VH and operating current I,
mktunce p/t, relevant for impedance matching, with resistivity p and effective plate thick-
ness t, and power-related sensitivity VH/(ZV) lIZBwith supply voltage K The latter figure is
proportional to M = (p: V, /ZB)‘I2 [27].
Figure 7-5 (adapted from [27]) offers a systematiccomparison of bulk silicon with different
111-V technologies in terms of these four parameters. The figure locates the materials Si, GaAs
(epi and implanted), InSb (bulk and film), and the two-dimensional electron gas (2DEG) at
the AlGaAsIGaAs hetero-interface, at room temperature and at 77 K, in a logarithmic net-
work of coordinates formed by the above four figures of merit. For example, Figure 7-5 in-
dicates that the 2DEG at 77 K shows a Hall mobility of about 15 mZ/Vs, a sensitivity of
about lo00 V/AT, and a resistance of 100 n, and that M is about 200 Q’”/T.
In addition to mobility, the bandgap is another important material figure of merit.
Although InSb and InAs show a large electron mobility, the drawback with these materials
is the small bandgap (0.2 eV for InSb and 0.4 eV for InAs). With room temperature operation,
intrinsic behavior prevails, which excludes the use of such materials other than for
magnetoresistive applications. In this respect, Si (1.12 eV) and GaAs (1.42 eV) are outstanding
materials with temperature coefficients that are far superior to InSb or InAs bulk and thin
film devices. The larger bandgap for GaAs permits device operation up to 250°C whereas

Flglue 7-5.
Comparison of Hall sensors fabricated with
different materials and technologies [27]. Current-
related sensitivities V,/ZB (Hallvoltage V,,
applied current I, magnetic induction B) are plot-
ted over .the Hall mobility pt. Also shown are the
plate mistance p / t (resistivity A effective plate
thickness t ) relevant for impedance matching and
the figure M = 01’ V,/IB)l’zcharacterizing the 16’ 1 10’ lo2
power-related sensitivity. MOBILITY (m?’V.s)
204 7 Integrated Magnetic Sensors

counterpart devices in Si stop working at about 150°C. Also, the temperature coefficient of
GaAs (bulk or implanted) Hall plates seems to be slightly lower than that of comparable
silicon devices, although this point may need further clarification. For further discussion of
the temperature coefficient of output sensitivity see [14].
Signal-to-noise ratio (SNR) controls the detection limit of MFS. For most applications the
SNR (related to the minimum detectable magnetic induction) rather than the sensitivity is the
leading figure of merit. Flicker (l/f) noise is dominant at low frequency operation. The cor-
responding SNR for the Hall voltage is proportional to the mobility for bulk-generated l/f
noise, hence favoring high-mobility material. More important, however, is the control of
flicker noise by the Hooge parameter a. This parameter can generally be defined in terms of
the square of the relative fluctuation of a two-pole of resistance R, viz,

(7-12)

where n denotes the total number of free electrons (assuming n-type material) in the resistance
block [48]. Values of a between and as low as have been reported depending on
material, technology, and geometry ([l, 141 and references cited therein). For Hall elements,
an appropriate choice of material, processing technology, and design geometry (eg, a buried
active layer reminiscent of the junction field effect transistor) which reduces the Hooge
parameter, could provide a much stronger improvement in SNR than that achievable by a
material with higher mobility [l, 141. At higher frequencies, thermal noise dominates and the
corresponding SNR is proportional to the mobility, thus again favoring high-mobility
materials.

7.2.3 New Technologies and Examples

Molecular beam epitaxy (MBE) is beginning to make an impact on semiconductor magnetic


microsensors. Highly sensitive Hall elements have been fabricated [27] which use the very thin
(10 nm) layer close to the AlGaAs/GaAs heterojunction and the related two-dimensional elec-
tron gas (2DEG) as the active sensor region. The device fabrication details are described in
[27]. The electron mobility in the active region at room temperature is 0.72 m2/Vs and the
device reaches a supply-current-related relative sensitivity of 100 V/AT. The temperature
dependence of the resistance p / t , Hall mobility p;, and sheet carrier concentration is il-
lustrated in Figure 7-6 for a lo00 nm thick AlGaAdGaAs layer with an impurity concentra-
tion of lo1*ern-,. The mobility exhibits a T-2.45 temperature dependence and the resulting
device temperature coefficient is around -0.7 %/K. From noise measurements for the
Al,, Ga,, As/GaAs heterojunction Hall element [28], a minimum detectable magnetic field
of about 2 nT at 1 kHz has been claimed.
Recently, 1200 V/AT sensitivity with a -O.l%/K temperature coefficient has been achieved
[29, 301 using the AlAs/GaAs heterojunction in the modulation-doped AIAs/GaAs superlat-
tire structure shown in Figure 7-7 as the Hall element’s active layer. Moreover, split-contact
magnetic sensors have been realized [31] using the same technology, with a design geometry
reminiscent of the dual-drain MOSFET discussed in Section 7.3.2. As one would expect from
the high electron Hall mobility of this structure (0.64 m*/Vs at room temperature), the
observed relative current imbalance per tesla (48%/T) is an order of magnitude larger than
7.2 Semiconductor MFS Materials and Technologies 205

I1
PDEG
t

10: -

P't

Figure 7-6.
(n)
Temperature dependence of parameters - 10"
characterizing the active region of 10'
AIGaAs/GaAs heterojunction Hall sen- . "r
sors [27]. The resistance p/t (resistivity p, (crn-2)
effective layer thickness 1, outer left hand
scale), the electron Hall mobility flux (inner
- t
left hand scale), and the sheet carrier I , I , ' 110"
concentration ns (right hand scale) are 1 10 100 1000
plotted as function of temperature. T ( K )

Ohmic contact

Quantum-Well

1
GaAs Buffer (X)Onm)T

1 S.1.-GaAs I
Figure 7-7. Cross-section of superlattice structure [29, 301;the two-dimensional electron gas (2 DEG) is
used as the active layer for the Hall element.

that of comparable split-drain MOSFET sensors with only about 0.07 m2/Vs n-channel Hall
mobility.
Maskless implantation of Si2+ ions into semi-insulating GaAs using focused ion beam
technology has been successfully employed in the fabrication of miniature Hall elements with
active regions of submicron dimensions [32]. Apart from its application potential in the
semiconductor IC industry, this is a promising microsensor (actually, "submicron" sensor)
technology, particularly for the design of magnetic microsensors with high spatial resolution.
206 7 Integrated Magnetic Sensors

Moreover, it requires a minimum number of process steps, provides accurate pattern definition
(crucial for minimizing offset), and allows variation in the implanted profile to optimize the
magnetic sensitivity and device impedance for integration with circuit components. The
fabricated device is illustrated in Figure 7-8. The shaded region of the cross-shape denotes the
active sensor region which is approximately 0.3 pm in diameter with a sheet carrier concentra-
-’
tion of the order of lOI3 cm [32].The relative sensitivity of the device is below 100 WAT.
The minimum width of the implanted region is limited by the diameter of the focused beam.
The discovery of high-T, superconductors may bring about a further technology for
magnetic microsensors which operate at the temperature of liquid nitrogen. The first super-
magneto-mistor highly sensitive to small fields (below 5 mT) has been reported recently [33].

Flgure 7-8.
Structure of Hall element fabricated using
maskless implantation with focused ion beams
[32].Inset shows calculated equiconcentration
contours within 0.5 pm from center of the
device

Fngure 7-9.
Resistance as function of magnetic induction and
operating current of the super-magneto-mistor [33]
fabricated using the high-T, superconductor Y-Ba-Cu-0.
7.3 Standad Silicon M .Ethnologies 201

The device exploits the property that a weak magnetic field (typically 100 pT) disrupts the
superconductivity of a granular Y-Ba-Cu-0 ceramic sample by increasing the resistance bet-
ween the (superconducting)grains. This leads to an abrupt change in resistance of the sample
with magnetic field. The resistance of the device as a function of magnetic field and operating
current is illustrated in Figure 7-9. The reported range of magnetic fields that can be sensed
is down to 10 nT, and further improvement is expected.

7.3 Standard Silicon MFS Technologies

At present, silicon offers the unique advantage of inexpensive batch fabrication by allowing
the integration of one or several basic sensor elements together with appropriate support and
signal processing circuitry in standard IC technologies of established reliability, such as
bipolar or CMOS technologies. Indeed, a large number of integrated MFS have been realized,
following the design rules of standard chip manufacturing processes offered by custom chip
manufacturers and university laboratories. Integrated silicon Hall devices are currently sold
in large quantities by a number of component manufacturers. The ever advancing silicon IC
technologies, and in particular the proliferation of the BICMOS process, will continue to offer
further sensor design opportunities.
Magnetic field sensor development outside established mainstream IC technologies has to
face the extra cost of developing specific manufacturing technologies and tools for mass pro-
duction in addition to appropriate test and reliability procedures [l]. This development cost
is usually beyond the financial possibilities of small- and medium-sized companies. This in-
vestment may become justifiable when the applicability of silicon MFS can be clearly ruled
out as, eg, in the case of high operating temperature or very high resolution requirements. In-
tegrated GaAs MFS seem to be the next proper choice if operating temperatures above 150K
are required. An example is the GaAs Hall IC chip reported in [34,35]. A non-standard chip
combining silicon IC and magnetoresistive thin-film technology (Ni,,Co,) has been realized
[36] for digital applications.

7.3.1 MFS Design by Standard IC Technologies

The constituent parts of the well-establishedbipolar and CMOS technologies include com-
puter-aided design (CAD) with pertinent CAD tools, device models for the simulation of
device and circuitry functions, mask and chip fabrication, assembly (including packaging),
and test and quality procedures. By definition, the integration of semiconductor magnetic sen-
sors means designing such devices for a given IC chip fabrication process as offered by an IC
chip manufacturer. Postprocessing of semiconductor magnetic sensor chips by, eg, applying
ferromagnetic thin films to the f i s h e d chip is beyond the scope of this chapter. However,
nonstandard procedures required by magnetic sensor integration include the design of novel,
unconventional device structures compatible with the given fabrication process, specific CAD
tools with models allowing the simulation of such devices, special packaging minimizing
mechanical stress and avoiding ferromagnetic materials, and test of the specific sensor func-
208 7 Integrated Magnetic sensors

tions, eg, by comparison with a calibrated magnetic probe. For component reliability, the
mature quality of the standard manufacturing process can be taken for granted, which is
another important advantage offered by integrated semiconductor magnetic sensors.

7.3.2 Examples

The inversion layer or channel provided by metal oxide semiconductor (MOS) technology
can be readily used as an extremly thin Hall plate in the form of a MOS field effect transistor
(MOSFET) with additional Hall contacts. However, the supply-voltage-related relative sen-
sitivity V,/VB is proportional to the channel Hall mobility P:,~, and is hence lower than
that of bulk Hall plates [l]. Split-drain or dual-drain MOSFET structures of the kind shown
in Figure 7-10 have been studied extensively [37, 381 in order to assess the noiseproperties of

Subsiraic Sourcc A

Figure 7-10.
Split-drain n-channel MOSFET;
magnetic induction vector perpen-
. - .
dicular to chip surface produces
current imbalance between drains
p wcll
1371.

such magnetic sensors and determine the n-channel Hall mobility &. In these devices, the
magnetic induction vector perpendicular to the inversion layer, by virtue of the Lorentz force,
produces a current imbalance AID between the two drains, viz,
AID = KpuXCBID (7-13)
where IDdenotes the total drain current and K is a dimensionless geometric factor depending
on the channel width and length and the separation of the drain regions. The magnetic sen-
sitivity can be assessed in terms of the relative drain current difference per magnetic induction,
I AI/ID B I, and is typically about 5%/T. From sensitivity measurements [38] and modeling
results for the factor K , the channel Hall mobility and consequently the Hall scattering coeffi-
cient (rn = pt,c/pn,c)can be determined (see Figure 7-11). In silicon inversion layers, scatter-
ing at the Si-SiO, interface contributes to r,, in addition to phonon and impurity scattering.
Thus the value of r, may be an indicator of the quality of the interface and hence the
underlying fabrication process. For a variety of device geometries and operating conditions
ensuring strong inversion in the linear regime, p,*,,,has been found to range between 600 and
850 cm2/Vs and the corresponding r,, between 1.1 and 1.4 [38]. The dual-drain MOSFET has
been integrated with a current-controlled oscillator on a single chip [39] fabricated with 5 pm
CMOS technology. This “magnetically controlled oscillator” (MCO) is sensitive to magnetic
fields perpendicular to the chip surface, and its output is a square wave whose frequency varies
linearly with, and is modulated by the input magnetic signal. The drawback faced with
MOSFET-based MFS is the high l/f noise in the inversion layer.
Homogeneous bulk silicon as the active sensor region is superior to the MOS inversion layer
in view of the lower output sensitivity and high noise level of the latter. On the other hand,
7.3 Standard Silicon MFS Technologies 209

Figure 74.
Electron channel mobilities of dual-drain
MOSFET (L = W = 100 pm) as functions of
gate voltage V, with drain voltage V , between
2 and 5 V [38]. Upper curve: channel Hall
mobility ,ut, c , lower curve: channel drift
mobility ,un,c . VG (VOLTS)

CMOS has become the leading digital IC technology and magnetic sensors compatible with
a standard CMOS manufacturing process are highly desirable. This has triggered the develop-
ment of CMOS-compatible bulk Hall-type devices (eg, the vertical Hall device [40])and
lateral magnetotransistors like the SSIMT [41] shown in Figure 7-12.
The suppressed-sidewall injection magnetotransistor (SSIMT) shown in Figure 7-12 il-
lustrates the sensor design possibilities offered by a commercial standard CMOS process. The

t'

N-SUBSTRATE

Figure 742. Suppressed sidewall injection magnetotransistor(SSIMT) with emitter contact E, base con-
tacts B 1, dual collector contacts C 1 and C2, and suppression stripe contacts B 2 (411.
210 7 Integrated Magnetic Sensors

SSIMT is a lateral transistor with dual collector and base contacts. In the presence of a
magnetic field, by virtue of the Lmentz force, carrier transport is enhanced for one collector
at the expense of the other; the resulting current difference A Z between the two collectors is
proportional to the magnetic induction B. The magnetic sensitivity of the device can be
described in terms of the relative collector current imbalance per unit magnetic induction,
1 AZ/Zc B I where Zc denotes the total collector current. The two P + stripes placed between
the emitter and each of the two collectors suppress the injection of carriers through the
sidewall of the emitter; these “parasitic” carriers would reach the collector independent of the
magnetic field and therefore not contribute to the sensor signal. A reverse potential applied
to these P + stripes confines injection to the centre of the bottom of the emitter base junction
and produces a partly focused carrier beam. Moreover, a lateral accelerating field in the
neutral base region is created in this way. All these effects combine to boost the sensitivity of
the device to over 1000%/T or l%/mT. By applying two different potentials to the two
P + stripes, one can shift the position of the emitted beam and thus correct for offset [49, 501.

7.3.3 Limitations Imposed by Standard Technologies

Although standard silicon IC technologies offer impressive advantages for integrated


magnetic sensor manufacturing, they also impose their limitations on the resulting microsen-
sor performance. Apart from the limited sensitivity and resolution mentioned earlier, there are
the undesirable phenomena of offset, non-linearity, temperature drift, and limited frequency
response [l, 141.
Offsetin integrated Hall elements, for example, is mainly caused by imperfections in the
fabrication process (geometry and material uniformity) and piezoresistive effects (usually
enhanced by the mechanical stress resulting from packaging). Errors in geometry can be kept
under control by designing fairly large devices and using the best lithography and etching pro-
cess available. The piezoresistive effect can be minimized by judicious choice of the
crystallographic orientation of the integrated Hall plate on the silicon wafer, viz. (110) crystal
plane, (110) current direction. Of course, integrated sensors will automatically benefit from
any further progress in lithography tolerances and semiconductor material quality. With pre-
sent technology, the typical offset is several mT, in practice (costly) laser trimming is still used
for further reduction of offset in commercial integrated Hall sensors. Offset in MT devices
is caused by the same basic effects as offset in Hall plates. Offset reducing integrated circuitry
on the sensor chip [42] is another possibility.
Linearity errors in Hall elements with a constant supply current are mostly due to short-
circuit effects and can be minimized by loading the sensor output with a properly valued
resistor. Dual-drain or dual-collector devices based on the carrier deflection effect show good
linearity as long as the current imbalance is only a few percent; the linearity error can be as
small as 5 x for fields up to 1 T. These devices, however, can be driven into saturation
by high magnetic induction such that the one drain or collector takes all the current and the
other is left with virtually zero current. This effect has been observed for the highly sensitive
SSIMT [49, 501.
Temperaturedriftin semiconductor magnetic sensors is due to the temperature dependence
of the material parameters (drift mobility, Hall mobility, and scattering factor); in the case
7.3 Standard Silicon MFS Technologies 211

of magnetotransistors there is, in addition, the exponential temperature dependence of the


junction diode characteristics.

7.3.4 Noise in Magnetotransistors

Despite its high output sensitivity, the intrinsic noise (l/for white) in the magnetotransistor
ultimately limits the magnetic field resolution and consequently imposes restrictions on its ap-
plication range, particularly at low frequencies, where l/f noise predominates. The only re-
maining option for high field resolution using silicon is to seek ways to reduce the intrinsic
noise. It has been shown [43, 45, 511 that a careful choice of device geometry and operating
configuration can reduce the intrinsic noise power in the MT's output by as much as four
orders of magnitude. This indicates that the resolution of the lateral MT may be pushed into
the nT range [52]. This has been achieved by exploiting a unique feature, namely the positive
correlation between the output noise voltages of the individual collectors observed in dual-
collector MTs whose geometry is of the kind shown in Figure 7-13.

c1 B E B CZ
R R

P
2@
10
Figure 743. P-WELL
Cross-sectional view of dual-col-
lector CMOS MT used in noise
- \ 1 -
correlation measurements. N -SUBSTRATE

Noise spectra for an MT fabricated with 3 wm CMOS technology are shown in Figure 7-14
for the device operated in single-ended and differential modes in a common emitter configura-
tion. The individual collectors have identical noise voltage power spectral densities (PSD) of
S,, ( w ) = S22(w),which show an approximately 1/f behavior. The corresponding differential
power spectral density S , ( w ) was found to be much smaller than the single-ended PSD, viz,

--
S, ( w ) 5 10 - 4 S,, (a). From this experimental result, a strong positive correlation,
Re &, (w)]/ S,,( w ) - 1, between the collector noise voltages has been derived [43].
Likewise, the correlation coefficient estimated from time domain measurements turns out to
be close to unity. This unique noise behavior has been reproduced for MTs of similar geometry
fabricated in other IC processes [43].

Single ended
\
Q
-90
Differential

-110 - S-m
P
D 0
-130qr I

2 .*-*
Figure 7-14.
-150 ..**c**-
Power spectral density (PSD) in single-ended -170 ' '''.*' ' '.''*' '
**"s**-
' '
212 7 Integrated Magnetic Sensors

These results are in striking contrast to the case of the dual-drain MOSFET [44,451, where
the differential spectral densities are larger than the single ended ones [S, (w) -
4 S,, (w)]
and a correspondingly negative correlation of Re {S12 (w)]/Sl, (w) -
-0.9 is found (see
Figure 7-15). A tentative interpretation of the strong correlation between the fluctuations in
output currents may be as follows [43]. In the case of magnetotransistors, the device noise cur-
rent is governed by events localized near the emitter base junction. These manifest themselves
as fluctuations in the emitter injected current which are felt simultaneously by both collectors.
For the dual-drain MOSFET, fluctuations in the drain current are due to noise sources
distributed throughout the inversion layer. The carriers in the vicinity of the channel pinch-off
undergo a partition effect, ie, they land on one drain or the other. Consequently, the fluctua-
tion at one drain is complemented by a fluctuation in the opposite sense at the other drain,
and the difference signal shows an even larger fluctuation.

-1301 Figure 7-15.


Power spectral density (PSD) in single-
-150
10” loo 10’

l o 2 103
““-I . vvvl
104
’’ ’
lo5
ended and differential operation for a
CMOS dual-drain MOSFET (geometry as
Frequency (Hz) in Figure 7-10).

7.3.5 3-D Vector Sensors

By employing combinations and arrays of magnetic sensor elements, 3-D magnetic field
detection can be made possible. Magnetic vector sensors have been realized using a multiple-
collector MT geometry [46], and also by a combination of lateral and vertical Hall-type sen-
sors with pertinent circuitry [47].
For the vector detection of magnetic fields with strong spatial variation, the active device
region should be minimal in area. For such applications, the magnetotransistor in bipolar
technology is a desirable choice because both vertical and lateral transistor action can be ex-
ploited to sense the three components of the magnetic field simultaneously. Laterally injected
currents of a multiple collector MT can be utilized to detect the magnetic field component
perpendicular to the chip surface. The inplane magnetic field components (parallel to the chip
surface) influence the vertical component of the injected current. Cross-sensitivity effects due
to the various field components do not appear to be a serious concern [46]. The spatial resolu-
tion of the fabricated 3-D detector is 8 x 10 x 20 vm, which is much higher than the resolu-
tion achieved with an array of three 1-D detectors (Hall elements) or with a combination of
2-D detectors (eg, the vertical MT) and a Hall element alongside each other.
For applications where only the absolute value of the vector magnetic field is to be detected
(such as in omnidirectional magnetometry and contactless switching), the size of the device
active region is not a serious constraint. For such cases, a combination of three Hall elements
7.4 References 213

(two lateral and a vertical) integrated with signal processing circuitry has been realized in
bipolar technology [47]. The lateral Hall element, which relies on the lateral current flow in
the epitaxial region, is sensitive to the component of field perpendicular to the chip surface.
For detection of field components parallel to the chip surface, vertical Hall cells are employed
where the direction of current path in the epitaxial region is vertical due to the presence of
a buried layer. The Hall elements are arranged as shown in Figure 7-16, to minimize the sen-
sitive area (ca. 200 vm in diameter) of the integrated detector. Cascaded translinear circuits,
integrated with the Hall elements, provide the necessary analog operations to retrieve the
magnitude of the magnetic field vector.

Figure 7-16.
Top view of three-dimensional
magnetic detector design 1471.

7.4 References

[I] Bakes, H. P., Popovic, R. S., Proc. IEEE 74 (1986) 1107-1132.


[2] Fluitman, J. H., in: Solid-State Sensors and Transducers Vol. 11, Sansen, W., Van der Spiegel, J.,
(eds.); Leuven, Belgium: Katholieke Universiteit, 1982, pp. IX-1-23.
[3] Vinal, A. W., IEEE Trans. Mugn. MAG-20 (1984) 681-686.
[4] McGuire, T. R., Potter, R. I., IEEE Trans. Mugn. MAG-11 (1975) 1018-1038.
[5] Thompson, D. A., Romankiw, L. T., Maydas, A. F., IEEE Trans. Mugn. MAG-11 (1975) 1039-1050.
[6] “Session BE-Magnetic Sensors”, in: International Magnetics (INTERMAG) Conf: Proc., MacNeal,
B. E., Fontana, R. E., Smits, Jr., and J. C., (eds.); IEEE Trans. Magn., MAG-20 (1984) 954-974.
214 7 Integrated Magnetic Sensors

[7] Berchier, J. L., Solt, K., Zajc, T., J. Appl. Phys. 55 (1984) 487-492.
[8] Rashleigh, S. C., Ulrich, R., Appl. Phys. Lett. 34 (1979) 768-770.
[9] Papp, A., Harms, H., Appl. Opt. 19 (1980) 3729-3834.
[lo] Castera, J. P., Hepner, G., “Device for modulating optical radiation by a variable magneticfield”
US. Patent 4236782, 1980.
[Ill Jones, R. E., Willson, J. P., Pitt, G. D., Pratt, R. H., Foulds, K. W. H., Batchelder, D. N., in: Optical
Fibre Sensors (IEE Conf. Publ. No. 221), D. E. N. Davies (ed.); London: Inst. Elec. Eng., 1983, pp.
33-37.
[I21 Dandridge, A., Weten, A. B., Sigel Jr., G. H., West, E. G., Giallorenzi, T. G., Electron Lett. 11(1980)
408-409.
[I31 Kordic, S., Sens. Acfuators 10 (1986) 347-378.
1141 Popovic, R. S., Sens. Actuators 17 (1989) 39-53; see also: Popovic, R. S., Heidenreich, W., Chapter
3, Volume 5, this series.
[I51 Sugiyama, Y., “Fundamental Research on Hall Effects in Inhomogeneous Magnetic Fields” in: Res.
Electrotech. Lab., No. 838, Tokyo: Electrotech. Lab., 1983.
[la] Clarke, J., IEEE Trans. Electron Devices ED-27 (1980) 1896-1908.
[I71 Romani, G. L., Williamson, S. J., Kaufman, L., Rev. Sci. Znstrum. 53 (1982) 1815-1850.
[I81 Lenz, J. E., ScientificHoneyweller 6 (1985) 16-25.
[I91 Fujimaki, N., Tamura, H., Imamura, T., Hasuo, S., ZEEE Trans Electron Devices ED-35 (1988)
2412-2418.
[20] Madelung, O., Introduction to Solid State Theory, Berlin: Springer Verlag, 1978.
[21] Selberherr, S., Analysis and Simulation of Semiconductor Devices, Vienna: Springer Verlag, 1984.
[22] Baltes, H. P., Andor, L., Nathan, A., Schmidt-Weinmar, H. G., ZEEE Trans. Electron Devices ED-31
(1984) 996-999.
1231 Nathan, A., Huiser, A. M. J., Baltes, H. P., ZEEE Trans. Electron Devices ED-32 (1985) 1212-1219.
[24] Nathan, A., Allegretto, W., Baltes, H. P., Sugiyama, Y., IEEE Trans. Electron DevicesED-34 (1987)
2077-2085.
(25) Andor, L., Baltes, H. P., Nathan, A., Schmidt-Weinmar, H. G., ZEEE Trans. Electron Devices ED-32
(1985) 1224-1230.
[26] Hara, T., Mihara, M., Toyoda, N., Zama, M., ZEEE Trans. Electron Devices ED-29 (1982) 78-82.
(271 Sugiyama, Y., Taguchi, T., Tacano, M., Proc. of the 6th Sensor Symposium (1986) 55-60.
[28] Tacano, M., Sugiyama, Y., Taguchi, T., ZEEE Electron Device Letts. EDL-8 (1987) 22-23.
1291 Sugiyama, Y., Soga, H., Tacano, M., Workbookof the Fifth Znt. ConJ on Molecular Beam Epitaxy
(1988) 522-555.
[30] Sugiyama, Y., Taguchi, T., Tacano, M., Transducers ’87Digest of Technical Papers (1987) 547-550.
[31] Sugiyama, Y., Soga, H., Tacano, M., Baltes, H. P., Technical Digest of the 7th Sensor Symposium
(1988) 47-50.
(321 Kanayama, T., Hiroshima, H., Komura, M., J. VacuumScience & Technology I36 (1988) 1010-1013.
[33] Kataoka, S., Tsuchimoto, S., Nojima, H., Kita, R., Nagata, M., Shintaku, H., Sensors Muter. 1
(1987) 7-12.
1341 Heywang, W., Sensorik, Halbleiter-Elektronik Vol. 17, Berlin: Springer Verlag, 1984.
[35] Pettenpaul, E., Huber, J., Weidlich, H., Flossmann, W., v. Borcke, U., Solid-state Electron 24 (1981)
781 -786.
[36] Usuki, T., Sugiyama, S., Takeuchi, M., Takeuchi, T., Igarashi, I., Proc. of the 2nd Sensor Sym-
posium (1982) 215-217.
[37] Briglio, D. R., Characterisation of CMOS Magnetic Field Sensors, M. Sc. Thesis, University of
Alberta, Edmonton, Canada 1988.
4381 Briglio, D. R., Nathan, A., Baltes, H. P., Can. J. Phys., 65 (1987) 842-845.
[39] Nathan, A., McKay, I. A., Filanovsky, I., Baltes, H. P., IEEE 1 Solid-State Circuits SC-22 (1987)
230-232.
1401 Popovic, R. S., IEEE Electron Device Letts. EDL-5 (1984) 357-358.
[41] Ristic, Lj., Baltes, H. P., Smy, T., Filanovsky, I., ZEEEEIectron DeviceLetts. EDG8 (1987) 395-397.
[42] Kordic, S., and van der Jagt, P. C. M.,Sens. Actuators 8 (1985) 197-217.
[43] Nathan, A., Baltes, H., Briglio, D. R., Doan, M., ZEEE Trans. Electron Devices ED-36 (1989)
1073-1075.
7.4 References 215

I441 Briglio, D. R., Nathan, A., Baltes, H. P.,in: Noise in Physical Systems, Van Vliet, C . M. (ed.);
Singapore: World Scientific, 1987, pp. 453-456.
[45] Baltes, H. P., Nathan, A., Briglio, D. R., IEEE Solid State Sensors and Actuators Workshop
Technical Digest (1988) 104- 105.
[46] Kordic, S . , ZEEE Electron Device Letts. EDG7 (1986) 196-198.
[47] Maenaka, K., Ishida, M., Nakamura, T., nansducers ’87Digest of Technical Papers (1987) 523-526.
[48] Ambrozy, A., Electronic Noise, New York: McGraw-Hill 1982, p. 114.
(491 Ristic, Lj., Smy, T., Baltes, H. P., Sensors Muter. 1 (1988) 83-92.
[50] Ristic, Lj., Smy, T., Baltes, H. P., ZEEE Trans. Electron Devices ED-36 (1989) 1076-1086.
[51] Nathan, A,, Baltes, H., “Integrated silicon magnetotransistors: High sensitivity or high resolu-
tion?’’, Sens. Actuators 21/22 (1989/90) in press.
[52] Nathan, A., Baltes, H., “How to achieve nanotesla resolution with integrated silicon magnetotran-
sistors”, IEEE International Electron Devices Meeting (IEDM) 1989, Technical Digest, in press.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

8 Optical Fibers and Integrated Optics


EMERYLIGHTNERMOORE.
Thousand Oaks. CA. USA
RAMONPEREZDEPAULA.
Bethesda. MD. USA

Contents
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
8.2 . . . . . . . . . . . . . . . . . . .
Definition and Classification 218
8.3 Optical Fiber Waveguides . . . . . . . . . . . . . . . . . . . . . 219
8.4 An Assortment of Sensors . . . . . . . . . . . . . . . . . . . . . 220
8.4.1 Transduction Coatings for Optical Fibers . . . . . . . . . . . . . . 220
8.4.2 Intrinsic Sensors . . . . . . . . . . . . . . . . . . . . . . . . . 221
8.4.2.1 Non-Interferometric Sensors . . . . . . . . . . . . . . . . . . . . 221
8.4.2.2 Interferometric Sensors . . . . . . . . . . . . . . . . . . . . . . 223
8.4.3 Extrinsic Sensors . . . . . . . . . . . . . . . . . . . . . . . . . 227
8.4.3.1 Image Translating Fiber Bundles . . . . . . . . . . . . . . . . . . 221
8.4.3.2 Translational Motion Sensors . . . . . . . . . . . . . . . . . . . . 221
8.4.3.3 Liquid Level Detectors . . . . . . . . . . . . . . . . . . . . . . . 229
8.4.3.4 ChemicalSensors . . . . . . . . . . . . . . . . . . . . . . . . . 230
8.5 Accuracy Limitations . . . . . . . . . . . . . . . . . . . . . . . 232
8.6 Optical Fiber Sensor Sources . . . . . . . . . . . . . . . . . . . 232
8.7 Optical Fiber Detectors . . . . . . . . . . . . . . . . . . . . . . 234
8.8 Sensor Components . . . . . . . . . . . . . . . . . . . . . . . . 236
8.8.1 Graded Index (GRIN) Lens . . . . . . . . . . . . . . . . . . . . 236
8.8.2 Optical Fiber Couplers . . . . . . . . . . . . . . . . . . . . . . 238
8.9 Integrated Optics (10) Circuits for Sensors . . . . . . . . . . . . . . 239
8.9.1 IOC Materials . . . . . . . . . . . . . . . . . . . . . . . . . . 240
8.9.2 LiNbO, Fabrication Process . . . . . . . . . . . . . . . . . . . . 241
8.9.3 LiNbO, Components . . . . . . . . . . . . . . . . . . . . . . . 242
8.9.4 An IOC Application . . . . . . . . . . . . . . . . . . . . . . . 243
8.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
8.11 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
218 8 Optical Fibers and Integrated Optics

8.1 Introduction

The development of low-loss optical fibers for communication purposes has spawned a
broadbased optical fiber sensor spinoff technology which happily has been able to build on
the communication fiber R & D base. Without this synergism fiber sensors might never have
garnered the necessary financial assets to initiate their own R & D incubation period.
It is appropriate in introducing the subject of fiber sensors to ponder: “When did fiber
optic sensors first begin development?” Such a question immediately leads to another ques-
tion. “What are optical fiber sensors?” If we can answer the question, “What are they?”, ie,
define optical fiber sensors, then perhaps we can decide when their development first began.

8.2 Definition and Classification

The broadest definition of optical fiber sensors includes “all sensory devices in which op-
tical fibers are a component”. For example, some authors divide optical fiber sensors into two
all encompassing categories, intrinsic and extrinsic. In this breakdown, intrinsic sensors are
‘distinguished by a measurand (the parameter being measured) which directly interacts with the
light in the fiber, generally by locally changing the fiber index through pressure, changing the
length of the fiber by pressure, force or dynamic motion, or by changing the intensity of light
in the core by way of a sharp bend in the fiber (microbend sensor). Extrinsic sensors are
distinguished by a measurand which affects the light properties through a medium external
to the fiber; ie, the fiber acts mainly as a delivery system to and from a device which in some
way alters the properties of the transmitted or reflected light. It is usually clear upon first ex-
amination of a sensor containing optical fibers which of the two categories, intrinsic or extrin-
sic, is appropriate. When it is not clear, the choice must be left to the observer. There are other
ways to categorize fiber sensors so as to encompass a complete set [I]. Vpically two other
popular categorizations are used. One categorization separates sensors into intensity
modulated and phase modulated sensors depending on whether the light intensity or its phase
is chinged by the measurand. Examples of each kind are:
Intensity Modulated Phase Modulated
- Attenuation/Loss - Interferometric
- Scattering - Resonant
- Grating Plates - Polarimetric
The other popular categorization separates sensors into interferometric and non-in-
terferometric depending on whether interferometric principles are employed or not. Some sen-
sors fitting this categorization are:
Interferometric
- Mach-Zehnder - Ring Resonator
- Michelson - Fabry Perot
- Sagnac - Multimode
- Polarimeter
8.3 Optical Fiber Waveguides 219

Non-Interfeometric
- Attenuation - Impurity Diffusion
- Scattering - Cladding Modification
- Grating Plates - Fluorescence
- Frustrated Internal - Radiation Loss
Reflection
For this discussion we will choose the broadest definition of fiber sensors, ie, all sensory
devices in which optical fibers are a component, and relate further discussion of optical fiber
sensors to the categorization of intrinsic and extrinsic sensors. Accepting the broad definition
of fiber sensors the earlier question of when fiber sensor development got underway is easily
addressed. In the early 1960s when fiber losses were still orders of magnitude higher than
today’s losses of 1 dB/km or less, the technology of transmitting images through fiber bundles
was being developed for many applications. For example, in the medical field various primitive
endoscope instruments for invasively viewing hard to reach locations in the human body were
developed. In addition military and civilian applications were pursued where images had to
be transported over short distances and bent around corners where flexibility (eg, in the
presence of vibration) of the light path was mandatory. These early instruments can be in-
cluded as part of the set of optical fiber sensors which today are considered light-delivery in-
struments. So it is rational to say that development of optical fiber sensors began in the early
1960s.

8.3 Optical Fiber Waveguides

Optical fibers today come in many forms - single mode, multimode, polarization preserv-
ing, polarizing and multicore silica cover the bulk of fiber types. However others exist for very
special purposes; eg, fluorescing for sensing applications, specifically doped for lasing, plastic
(or polymers) for low cost, halide glasses for low loss, rubber for extreme bendability etc. A
basic multimode and single-mode fiber is shown in Figure 8-1.
In order to act as a waveguide for light the fiber has a core with a slightly higher index of
refraction than the cladding. Qpical core and cladding indices in silica fibers might be 1.40
and 1.46, respectively. This index difference is sufficient to represent a dielectric boundary.
The core then becomes a waveguide wherein the light (electromagnetic energy) interacts with
the core-cladding interface according to the rules for dielectric boundary conditions. In
multimode fibers the core diameter is large (50-200 microns) and several modes (ie, traveling
electromagnetic wave patterns) can be sustained. In single-mode fibers the core diameter is
small (5-10 microns) so that only one traveling wave pattern may be sustained, albeit with
possibly two orthogonal polarizations.
Silica multimode and single-mode fibers are the standard fiber types in optical fiber sensors
so that we will limit our discussion to sensors utilizing such fibers. Details of fibers other than
standard silica multimode and single-mode can be found extensively in the literature.
B. Culshaw [2] gives a good treatment of the basics of optical fibers while Wysocki [3] presents
a treatise on optimization of fiber parameters. Ultralow-loss fibers are treated by Miller [4],
polarization preserving fiber is addressed by Tajima and Sasaki [S], and Aoyagi [6] discusses
plastic optical fibers. References to further related literature are given by each of these authors.
220 8 Optical Fibers and Integrated Optics

\ JACKET

,/ b. SISZkEMODE
Figure 8-1. ' h o basic fiber types: (a) multimode fiber; (b) single-mode fiber.

8.4 An Assortment of Sensors

We will attempt to provide an overview of optical fiber sensors by presenting an assortment


where most of the principal types are represented. A number of earlier overviews [l, 71 of op-
tical fiber sensors can be consulted as a supplement to the material presented here.

8.4.1 Transduction Coatings for Optical Fibers

Before we begin our discussion of optical fiber sensors, it may be helpful to mention the
concept of optical fiber coatings which are sometimes utilized to enhance the measurand sen-
sitivity of intrinsic sensors, particularly Mach-Zehnder and Michelson interferometers which
will be discussed later. The transduction portion of the optical fiber is usually sensitized in
some fashion to the measurand. There exist numerous ways to accomplish this transduction,
from measurand to fiber strain, however, two schemes are particularly prevalent. One is to
wrap the fiber around a mandrel which is sensitized to the measurand, while a second method
inverts the process and wraps a sensitized coating around the fiber. Of course it is not
unreasonable to use both methods and wrap a coated fiber around a mandrel thereby increas-
ing the effect. Figure 8-2, after J. A. Bucuro, presents a brief summary of coatings utilized
to achieve sensitivity to four physical measurands or fields - acoustic, magnetic, thermal, and
electric.
8.4 An Assortment of Sensors 221

SENSOR ACOUSTIC MAGNETIC THERMAL ELECTRIC

I COATING ELASTOMERS MAGNETOSTRICTIVE THERMALLY PIEZOELECTRIC

I
EXPANSIVE,
CONDUCTIVE

1 I 1 1 EEzz!$NG
0 POLYSTYRENE 0 NICKEL ALLOYS NICKEL 0 PVFz
0 NYLON 0 MET GLASS ALUMINUM 0 COPOLYMERS

ELECTROPLATING
~ , !
EXTRUS1oN ~ ~
EVAPORATION ~ ~ ~
CHEM. DEPOSITION
~
ION SPUTTERING
DIP COATING

I SPECIAL
CONSIDERATIONS II I ANNEALING
BIASING I POLING
MOLECULAR.
ORIENTATION
I I I
AFTER J. A. BUCARO. 1983

Figure 8-2. Coatings for fiber optic transduction.

8.4.2 Intrinsic Sensors

8.4.2.1 Non-Interferometric Sensors

As previously stated, with intrinsic sensors the measurand acts directly on the fiber; both
interferometric and non-interferometric sensors are included in this category. Probably the
simplest type of fiber optic sensor is a microbend sensor.

Microbend Sensors
Microbend sensors can be classified as belonging to the subset intrinsic, non-in-
terferometric, intensity modulated sensors. The basic operation is derived from the principle
of total internal reflection. This principle simply states that light traveling in the core of the
fiber will remain traveling in the core as long as the angle-of-incidence (O), Figure 8-3a, with
the normal to the core cladding interface exceeds a critical angle (Oc). When this angle of in-
cidence is less than 0, at least part of the light leaks out and is lost from the core as in Figure
8-3b. The microbend sensor uses this principle to advantage by detecting the light intensity
change when microbends are introduced into the fiber. A simple example of one type of micro-
bend sensor is shown in Figure 8-4. Such sensors have many uses, some of which are robotic
tactile sensors, vibration monitoring and strain sensing.

Distributed Temperature Sensor


Another type of intrinsic, non-interferometric sensor utilizes a standard, multimode, doped
silica-core fiber where the dopant (typically germania) contributes to a backscattered Raman
spectrum.
222 8 Optical Fibers and Integrated Optics

C-)
CLADDING
CORE /

e>ec
TOTALREFLECTANCE
a.

Figure 8-3.
Microbend light loss:
(a) straight fiber has no bend loss;
PARTIAL LOSS (b) small bend creates transmission
b. loss.

APPLIED FORCE
TRANSDUCER
ESTORING SPRING

Figure 8-4.
Microbend sensor.

Raman scattering is classified as Stokes or anti-Stokes dependent on whether the incident


photon loses energy (shifts to longer wavelength) or gains energy (shifts to shorter
wavelength), respectively. Although the Raman scattered light is several orders of magnitude
less than the Rayleigh scattered light, the ratio of the Stokes and anti-Stokes components con-
tains reliable temperature information which is independent of laser power, geometry of the
fiber or scattering and backscatter capture cross-sections [8]. The rationalization process nor-
malizes most significant variables leaving the ratio as a temperature determinant.
Figure 8-5 shows a general concept of an optical fiber distributed temperature sensor. A
pulsed laser source sends an optical pulse along an optical fiber over distances up to several
kilometers. If a localized temperature differential A T exists somewhere along the fiber, the
Stokedanti-Stokes Raman backscattered intensity ratio will vary. By filtering the backscat-
tered light, the Raman components will be detected by the photodetector and the absolute
temperature and location of the temperature differential (based on pulse delay time) can be
8.4 An Assortment of Sensors 223

FIBER
COUPLER
LASER
SOURCE

SENSOR FIBER

INTERFERENCE PHOTO
FILTER DETECTOR

PULSE AMPLIFIER
GENERATOR DEMODULATOR
I
Figure 8-5.
Optical fiber distributed
temperature sensor.

determined. Temperature resolution of 1 K and spatial resolution of several meters have been
reported. Applications of this technology to ships and large buildings have been successfully
tested. The low cost of optical fiber makes this distributed temperature sensor appear attrac-
tive, especially if it can be adapted for use with off-the-shelf optical time domain reflec-
tometers (OTDRs) rather than expensive custom optoelectronics.

8.4.2.2 Interferometric Sensors

There are numerous types of fiber optic interferometers being developed as sensors. The
four most popular types will be introduced here.

Mach-Zehnder Interferometer
Shown in Figure 8-6 is the Mach-Zehnder fiber optic interferometer. This interferometer is
rather simple to construct. It consists of a reference arm and a sensor arm between two 3 dB
couplers. The reference arm is generally isolated from the measurand while the sensor arm is
exposed to such effect, often with the addition of a transduction device which is especially
sensitive to the effect. Light from the source, usually a narrow line laser diode, is split by the
first coupler and travels through both arms and is recombined by the second coupler. One or

3 dB COUPLER

Figure 8-6.
Mach-Zehnder optical fiber
interferometer. 3dB COUPLER
224 8 Optical Fibers and Integrated Optics

more photodetectors convert the light signal to an electronic signal for further processing.
Phase shifts as small as 1 microradian of the light source wavelength can be detected with the
interferometer. The sensor arm need only experience a very small effect for the effect to be
detectable. Any effect (pressure, electric or magnetic field, temperature .. .) which can cause
the sensor arm to change length (or equivalently refractive index) by as little as 0.001
Angstroms is a candidate for measurement with a Mach-Zender interferometer.

Michelson Interferometer
The Michelson fiber optic interferometer shown in Figure 8-7 is similar in operation to the
Mach-Zehnder. It has two arms as in the Mach-Zehnder. However, only half as much fiber
is needed in each arm for equivalent sensitivity since the mirrored ends reflect the light back
through the arms. In addition, only one coupler is required. The penalty, if any, is in the need
to mirror the ends of each fiber arm. Mirroring of the arms is quite simple so that this penalty
is not severe.

MIRRORED
3 dB COUPLER

Figure 8-7.
Michelson optical fiber
TRANSDUCER interferometer.

Sagnac Interferometer
The Sagnac fiber optic interferometer shown in Figure 8-8 is a rotation sensor (or
gyroscope) first investigated by G. Sagnac [9] using bulk optic devices. In modern times the
practicality of the Sagnac interferometer was demonstrated using a fiber optic coil to for-
mulate the counterpropagating light paths by Vali and Shorthill (lo] working at the University
of Utah. This Sagnac interferometer is a phase sensitive interferometer like the Mach-Zehnder
and Michelson. The basic sensor is no more than a coil of fiber (or equivalently, a multi-turn
light path). Light from a single source is injected into both ends of the fiber coil through a
coupler (beam splitter). After transiting the fiber coil, light from both ends of the coil is
passed back through the coupler where both return beams interfere. The interference signal,
now in both of the two return fiber paths out of the coupler, is transmitted to a photodetector
through just one of the return paths.

POLARIZER

Figure 8-8.
I \ ‘FIBER
3 dB END 3dB Sagnac optical fiber rota-
COIL
COUPLER COUPLER tional phase interferometer.
8.4 An Assortment of Sensors 225

If the coil should happen to rotate at rate 52 (degrees per second) while the counter-
propagating light beams are in transit through the coil, the pathlength for one beam becomes
elongated while the opposing path becomes shortened. When the counterpropagating beams
reach the coupler after transiting the rotating coil they are out of phase with each other by
an amount A@ where

and where L is the length of the fiber coil, R is the coil radius, I is the source wavelength and
c is the speed of light in the fiber. The amplitude of the combined return beams out of the
coupler, varies as the phase difference (A@) varies, thus providing rotation rate information
to the detector. A polarizer is included to remove one of the two polarization modes, thus
eliminating crosstalk noise, in the single-mode fiber device.

Resonant Ring Interferometer


Shown in Figure 8-9 is a fiber optic ring resonator. Here light from a highly coherent source
is transmitted along the fiber to a coupler. At the coupler, part of the light continues to a fiber
coil and part couples to the detector end of the fiber. The light going through the coil even-
tually returns to the coupler again. At the coupler the light which transited the coil is partially
coupled back into the coil and partially transmitted on to the detector. Hence at the coupler
both the light from the source and the light from the coil interfere. The interference may be
constructive, destructive or any combination thereof dependent on the optical length of the
fiber coil. To make a sensor out of the device one needs to expose the fiber coil or a portion
thereof to a measurand which acts directly on the fiber or through an enhancement
transducer. The measurand must act on the coil to effectively change the optical length of the
fiber and thus change the state of the interference of the two light beams entering the coupler.
The detector will then sense a change in the amplitude of light reaching it. The fiber optic
resonator was first reported on by Stokes et al. Ill] who noted the need for a very low-loss
coupler and a very coherent source to even approach the shot noise limit. Not yet empirically
analyzed, is the implementation of the resonant fiber optic ring as the sensor arm of a Mach-
Zehnder or Michelson interferometer. Certainly some special sensitivity is possible with such
a configuration.
FIBER

COUPLER

Figure 8-9. TRANSDUCER


Resonant ring optical DETECTOR
fiber interferometer.

Resonant Fiber Optic Gyro, RFOG


The resonant fiber optic sensor can also be implemented as a rotation sensor, which is
known as the Resonant Fiber Optic Gyro or RFOG, Figure 8-10 [12]. In the RFOG a
resonating gyro cavity is formed by making a small optical fiber loop. Light is coupled into
226 8 Optical Fibers and Integrated Optics

COUPLER
SIGNAL
PROCESSING

Figure 8-10. Resonant fiber optic gyro, RFOG; a complex implementation of the resonant ring inter-
ferometer.

the loop in two directions after a modulation signal is first imposed. At a second coupler light
in both directions is partially coupled to a detector and partially coupled back into the loop.
The coupling ratio of the couplers can be on the order of 90 to 99 percent so that most of
the light stays in the loop.
When the wavelength of the light is properly matched to establish standing optical waves
over the cavity (loop) length, one achieves construcfive interference or resonance at that
specific wavelength (or frequency). This resonance can be achieved for light propagating
around the loop in either direction (say clockwise and counterclockwise). This bidirectional
light propagation distinguishes the simple ring resonator from the RFOG. If the loop is caused
to rotate about an axis perpendicular to the plane of the loop, the pathlength in one direction
of rotation gets longer while the pathlength in the opposite direction gets shorter.
Hence, the loop or resonant fiber cavity will resonate at two different frequencies, one for
the longer wavelength (lower frequency) and one for the shorter wavelength (higher fre-
quency). Detection of the two frequencies by any of several techniques, provides a measure
of the rotation rate.

Sensitivity Enhancement
For the phase interferometric sensors of Figures 8-6 and 8-7, ie, the Mach-Zehnder and
Michelson, and the resonant ring interferometer of Figure 8-9 it is typical to enhance the sen-
sitivity by devising some means which increases the sensitivity of the bare fiber. A simple
means of enhancing the sensitivity is to coat the fiber with a jacket which is particularly sen-
sitive to the measurand as discussed in Section 8.4.1. For example, the fiber may be coated
with elastomeric, magnetostrictive or piezoelectric jackets to enhance the sensitivity to
pressure, magnetic or electric fields, respectively. In addition the fiber may be wrapped on a
mandrel having elastometric, magnetostrictive or piezoelectric characteristics for further sen-
sitivity enhancement. There are many other means of achieving increased sensitivity beyond
that of the bare fiber. The main goal however is to amplify or enhance the phase response of
the fiber to the measurand. Anything that accomplishes this and can be made to satisfy the
specific application conditions, eg, cost, weight, volume, environment, dynamic range, will

i i g accompiisnes rnis ana can be maae to satisry the


LUG IIUCI LU LIIC I I I G ~ ~ S U M I I U . ~ ~ ~ y r i i iinat

specific application conditions, eg, cost, weight, volume, environment, dynamic range, will
8.4 An Assortment of Sensors 227

suffice. Several forms of optical fiber interferometers are now commercially available and
numerous others are in various stages of research and development.

8.4.3 Extrinsic Sensors

Recall from earlier discussion that extrinsic optical fiber sensors are those where the
measurand alters the light properties by way of a medium external to the fiber. Extrinsic sen-
sors come in many more formulations than the rather limited number of generic intrinsic sen-
sors. Again we will attempt to give an overview so as to introduce the subject rather than to
present an exhaustive treatise attempting to illustrate all known varieties.

8.4.3.1 Image Translating Fiber Bundles

The concept of translating an optical image from one spatial point to another over a very
flexible, dynamic path is greatly facilitated by optical fibers. A need exists for this type of op-
tical image transport in medical diagnosis and treatment, in vibrating or rotating instruments
and in industrial inspection of hard to see components. Medical technology has developed this
type of optical fiber sensor to the point where off-the-shelf instruments of high quality are
readily available. Optical fibers facilitate this technology because they are very thin (core
diameters of several microns) and flexible. Image translation is performed by simply sensing
an image as a matrix of points at one end of a fiber bundle, translating each point image in
a single fiber core over the desired pathlength and then reassembling the image at the terminal
end of the transport path point-by-point (fiber-by-fiber) in a matrix pattern which is con-
gruent with the starting matrix. The matrix points can be thought of as picture cells (pixels)
in an image plane with an individual fiber translating the optical information of each separate
pixel over the desired transport path.
Figure 8-11 shows a fiber image translator with an endface blowup. Such fiber bundles for
image translation are not typical of communication fibers in composition. For example, in
order to achieve good light-collection quality and to make the pixel core diameter as small as
possible (for higher resolution in smaller diameter bundles) the core to cladding refractive
index ratio may be approximately 1.5. The same ratio for silica-communication fiber is closer
to 1.01. In some newer image bundles for medical endoscopes, 2,000 pixel cores are contained
in a finished translator bundle of 0.27 millimeter diameter. These devices along with the latest
in tomography technology are a great asset for in vivo diagnostics.

8.4.3.2 Translational Motion Sensors

A wide variety of extrinsic motion sensors can be created which basically detect changes
in the amplitude of light due to translational motion affecting the light transmittance between
light source and photodetector with optical fibers acting as the principal light carrying chan-
nels.
In Figure 8-12a is shown a reflecting surface at a distance d from the end of an optical fiber.
Light is transmitted to this surface through an optical fiber, then reflected from the surface
228 8 Optical Fibers and Integrated Optics

SILICA JACKET
/

Figure 8-ll. Flexible fiber bundle image translator.

so that some of the light is recaptured by the fiber. The return light is split by a fiber optic
coupler (beam splitter) to provide a portion of the light for sensing by a photodetector. If the
distance d increases, the amount of light returned to the photodetector will decrease. The light
to the photodetector increases when the distance d decreases, hence the detector senses the
change in proximity or motion of the reflecting surface. Depending on the use of focusing
devices and reflectivity of the reflecting surface the sensitivity can be changed. Figure 8-12b
shows another variation of a motion sensors where the source fiber and the detection fiber
are referenced to surfaces which may move or translate vertically or longitudinally relative to
one another. In either case of translation, the light collected by the detector fiber will vary.
Another variation on the motion sensing theme is presented in Figure 8-12c. Here a mask is
allowed to move in the transmittance path between a source and detector. Optical fibers are
used as the channels to transport light to and from the vicinity of the mask. As the mask
moves, the degree of light obscuration between the source and detector changes thus providing
a measure of the mask motion at the detector. One can increase the sensitivity of the mask
(or obscurance sensor) by clever design of the mask. A rather simple increase in sensitivity
is obtained by using both a fixed and moveable mask having a bar like grid pattern on each
mask. The periodic transmittance pattern can provide a rather high-resolution linear position
sensing capability.
Depending on the details of implementation, the three sensors of Figure 8-12 can be used
to detect a number of measurands including: vibration, linear position, temperature, pressure
8.4 An Assortment of Sensors 229

REFLECTING
SURFACE,
LIGHT
COUPLER
SOURCE
/-
PHOTO
1 <
-/ Ill
DETECTOR

LIGHT
SOURCE
+
VERTICAL

LONGITUDINAL

DETECTOR

b.

VERTICALLY

Figure 8-u.
Translational motion sensors:
(a) reflecting motion sensor;
(b) transmitting motion sensor;
(c) obscurance motion sensor. C.

and stress or strain. Although similar sensors can be and have been constructed without using
optical fibers, use of optical fibers generally provides the more cost-effective light channel.

8.4.3.3 Liquid Level Detectors

It is known that in step index, multimode fibers both core and cladding propagation modes
exist. The light propagating in the cladding modes continues to propagate in the cladding so
long as the cladding is surrounded or jacketed with a material having a lower refractive index
than the cladding. However if the cladding is surrounded with a material of higher index than
the cladding, the cladding modes will leak out or be stripped off. This phenomena provides
a means for liquid-level detection which is particularly useful for fluids which may be
dangerous in the presence of electrical signals or may be corrosive to non-silica based
materials.
In Figure 8-13a is shown a simple liquid-level detector which senses when the level of the
fluid in the beaker (any container) is above or below the level A. Over a short distance near
level A the fiber is stripped of its jacket so that the cladding is exposed either to air or liquid
in the beaker. When the liquid is below level A the cladding is surrounded by either air or
jacket (indices less than the cladding) and all the modes, both core and cladding, propagate
230 8 Optical Fibers and Integrated Optics

EXPOSED

-
FIBER
CLADDING

-/ 1
1-JL DETECTOR

- 7 -L A
BEAKER BEAKER FILL HEIGHT

a. b.
Figure 843. Optical fiber discrete level detector (a) and its intensity versus fill height response (b).

on to the detector giving maximum signal as shown in Figure 8-13b. As soon as a liquid with
a refractive index greater than the cladding goes above level A the cladding modes are stripped
off into the liquid and the power at the photodetector drops. When the liquid drops below
level A the detector signal will again increase. This liquid level detector is thus a discrete level
detector which tells only that the liquid is above or below level A. With a little ingenuity one
can design a scheme based on this same concept which can sense numerous discrete steps or
continuous level changes along a length of fiber.
A liquid level sensor capable of sensing multi-discrete liquid levels is shown in Figure 8-14.
This design is based on similar principles as the design tested by Belkerdid et al. [13]. Here
the cladding, as shown in Figure 8-14a, is exposed on the return path of the step index,
multimode fiber. The return path fiber is made to conform to a zigzag contour so that it
takes on a small bend radius at regular intervals. As soon as the exposed cladding is sur-
rounded by a fluid of higher index, the cladding modes are stripped away lowering the detected
power level. However, each time the fiber curves around a small bend radius (as at levels,
A, B, . .. L) new cladding modes are generated. If no liquid exists above a given level the clad-
ding modes continue to propagate to the photodetector. Each time the liquid goes above a
bend level the cladding modes generated at that bend are lost to the liquid. As the liquid level
goes up above each bend the detector power decreases, Figure 8-14b, having lost the cladding
mode power generated at that bend level. With a little further thought many other configura-
tions of liquid level sensors, based on the principles exhibited here and especially suited to
adverse environments, can be conceived.

8.4.3.4 Chemical Sensors

Another type of extrinsic optical fiber sensor, of which there have been concocted numerous
varieties, is the chemical sensor. These sensors are finding application in industrial process
c
8.4 An Assortment of Sensors 231

EXPOSED CLADDING

/ JACKETED

/FIBER

C
B
A I I I I I I I I 1 I l l
A B C K L
DEAKER BEAKER FILL HEIGHT
a. b.

Figure 8-14. Multi-Discrete liquid level detector (a) and its intensity versus fill height response (b).

monitoring and the biomedicaVbiochemica1 field. Of the many varieties, those based on
fluorescence are quite prevalent. The basic concept of an optical fiber fluorescent chemical
sensor is illustrated in Figure 8-15. The sensing mechanism is in the form of a chemical reagent
which is somehow bound to the distal tip of the fiber. The binding of the reagent can be via
a covalent bond to the fiber surface or to an optically transparent substrate which is
mechanically fastened to the fiber. Techniques also exist for capturing reagent on the fiber tip
behind a membrane permeable by the analyte. The reagent is chosen for its specificity in reac-
tion with the analyte. The sensor detects the presence of analyte by observation of induced
changes in the optical properties of the reagent. In particular, following exposure of the
reagent to the analyte, the reagent is illuminated by the light source, usually a laser, and the
fluorescence spectrum is analyzed. The fluorescence spectrum will be altered by reagent ex-
posure to the analyte indicating the presence of the analyte. In some experiments im-
munochemical reagents have been in antibody/antigen and antibodylhapten interactions to
obtain highly sensitive and selective detection of trace amounts of large molecules [14].
Optical fiber chemical sensors exhibit an advantage over other forms of fluorescent sensors

COUPLER

Figure 8-15. REAGENT COATED


Optical fiber fluorescent DISTAL TIP
chemical sensor.
232 8 Optical Fibers and Integrated Optics

because of the large optical excitation and light collection efficiencies of the experimental
sample. Keen interest continues to build in optical fiber chemical sensors both for in vitro and
in vivo sensing.

8.5 Accuracy Limitations

Non-interferometric, intrinsic sensors as well as extrinsic optical fiber sensors are generally
intensity type sensors [15, 161; ie, they depend on the light intensity reaching the detector
rather than a phase interference signal. A number of intensity type sensors are low-accuracy
devices when compared to interferometric devices. Accuracies of 0.1 percent and dynamic
ranges exceeding 60 dB push the limit of most intensity sensors. On the other hand, in-
terferometric sensors can achieve accuracies of 0.001 percent and dynamic ranges exceeding
100 dB.
Particularly exemplary of low-accuracy intensity sensors are translational motion sensors as
discussed in Section 8.4.3.2 (Figure 8-12). Should dust or dirt particles get into the transmissive
or reflective light paths of such sensors, the response at the detector will be altered as com-
pared with previous measurements. In essence the response of the sensor is not highly
repeatable over time. If accuracy is not paramount this situation can be tolerated. Otherwise
some method of isolation of the sensor light path must be achieved. Hermetic sealing may
sometimes be feasible or a choice of an intrinsic sensor (eg, microbend sensor) to inhibit ex-
posure of the light path to unwanted variables may be the solution. If no method can be found
for achieving the desired accuracy (and long-term repeatability) with an intensity sensor, the
choice of an interferometric sensor may be appropriate. Unfortunately this tradeoff normally
imposes a higher cost.

8.6 Optical Fiber Sensor Sources

Generally there are two kinds of optical fiber sensor sources or light emitters. These are the
light emitting diode (LED) and the laser diode both of which are semiconductor devices. The
LED emits incoherent light and the laser diode emits coherent light. True coherent light has
the characteristic of constant light phase across the beam cross section. The constant cross-
sectional phase must exist at any point at a given time, or at any time at a given point, along
the beam. Laser diodes (and lasers in general) approximate this condition of coherence, some
more than others. LED’S do not emit light with a constant phase front and thus emit in-
coherent light.
Shown in Figure 8-16, the LED and the laser diode both are formed by bringing together
a p-type semiconductor and an n-type semiconductor to form a p-n junction (stripe type is
illustrated) or active region of the source. The light emitting region is roughly defined by the
intersection of the “stripe’s” vertical projection (dashed lines) through the active region or
junction. Light can be emitted in any direction in an LED, most ofted:f#om the edge (cross
- 233
8.6 Optical Fiber Sensor Sources

CONTACT (+I

TYPICAL LIGHT EMISSION REGION


MATERIALS:

ACTIVE REGION OR JUNCTION

Figure 8-16.
Stripe p-n junction.
' STRIPE CONTACT (-1

hatched) or sometimes the contact surface when the surface is properly prepared. To achieve
lasing from a device like that shown in Figure 8-16 the light emitting region must form a reso-
nant cavity. This is done by cleaving or dielectric coating the end surfaces of the semiconduc-
tor chip to obtain partial reflection (typically 35 percent) at the ends of the light emitting
region. The laser diode is thus limited to edge emission, emitting through the partially reflec-
ting resonant cavity ends.
In Figure 8-17, a homostructure (common semiconductor material for both p and n sec-
tions) p-n junction is represented to illustrate the means of photon emission from such a junc-
tion. The junction is shown as simply a side-by-side mating of p and n type:semiconductors

JUNCTION REGION

UNBIASED
JUNCTION
BARRl ER
POTENTIAL

7 ' 1 1 VALENCE

(c) FORWARD BIASED


FREEHOLES I I
.f BAND

Figure 8-17.
JUNCTION
p-n junction semicon-
- + + + + + I-,
ductor optical source: BARRIER {- +++++
(a) homostructure p-n POTENTIAL
junction;
(b) unbiased junction;
(c) forward biased junc-
tion.
-
I I
JUNCTION REGION
234 8 Optical Fibers and Integrated Optics

in the top (a) part of the figure. In the (b) or middle part of Figure 8-17, the important energy
bands of the p-n junction are illustrated for an unbiased (no voltage applied) junction. Free
electrons are shown in the conduction band and free holes (lack of electrons) are shown in
the valence band. The forbidden band, where no quantum theoretically allowed carrier (elec-
trons or holes) energy states exist, is characterized by an energy gap, Eg.Eg has values of ap-
proximately 1.5 to 0.8 electron volts for junctions emitting at respective wavelengths of 0.8 to
1.5 microns.
In the unbiased condition, no holes or electrons will conduct or flow across the barrier since
they have insufficient energy (as illustrated by the vertical axis) to surmount the barrier poten-
tial. Holes have increasing energy toward the bottom of the figure and electron energy in-
creases toward the top. When the junction is forward biased (ie, a voltage is applied + to -
from the p to n side) the barrier potential is lowered as shown in part (c) of Figure 8-17. Some
electrons and holes can now move into the junction region. When electrons and holes are in
close enough proximity, they may recombine releasing energy equivalent to the gap energy
E g . This energy release takes the form of a photon of light with energy E g . The junction
when forward biased thus becomes a light emitter such as an LED or laser diode.
For emission in the 0.8 micron spectral region, AlGaAs is the standard compound whereas
in the 1.3 to 1.5 micron spectral region InGaAsP is the typical compound. AlGaAs and
InGaAsP are 111-Vcompounds (compounds composed of elements from groups 111 and V of
the periodic table). Emission at wavelengths from 5 to 12 microns is possible with IV-VI com-
pounds (eg, PbSnSae). LED’s are the more robust and longer lived emitters. They emit a
multimode structure with spectral widths of 50 to 100 nm in the 1.3 micron region and propor-
tionally smaller spectral widths at shorter emission wavelength. LED’s couple better to
multimode fibers because the larger core diameters of the multimode fibers are more receptive
to the larger beam spread of LED’s. Full beam at half power level cone angles for LED’s are
typically 120 degrees. Laser diodes, on the other hand, couple better to single mode fibers.
Laser diodes have much smaller spectral widths, from approximately 5 nm down to less than
0.1 nm, and can emit in a single mode structure. Full beam cone angles from 10 degrees to
35 degrees are common. A good description of the materials and structures of semiconductor
light sources has been given by Suematsu [17].

8.7 Optical Fiber Detectors

Optical detectors for detecting signals in optical fibers are generally limited to semiconduc-
tor devices know%as PIN or avalanche photodiodes, both of which are variants of the simple
p-n junction photodiode. These should be distinguished from earlier vacuum tube photodetec-
tors like the vacuum tube photodiode and photomultiplier. Materials in use for semiconductor
photodetectors are chosen for their responsivity to the photon wavelength of interest. Silicon
will efficiently absorb photon energy in the wavelength region from 0.3 to 1.1 microns.
However, it is transparent beyond 1.3 microns so that other materials are required for the 1.3
and 1.55 micron fiber windows. Both germanium and the InGaAsP system are photon sen-
sitive for the 1.3 and 1.55 micron windows. The InGaAsP system most often is chosen because
of its lower dark current (ie, thermally generated current in the absence of light excitation).
8.7 Optical Fiber Detectors 235

The basic concept of semiconductor photodetection is illustrated in Figure 8-18a with the
p-n junction. The process is simply the reverse of p-n junction, semiconductor optical sources.
A photon incident on the junction with energy equal to or greater than the forbidden gap
energy, E g , may excite an electron from the valence band into the conduction band leaving
a hole behind in the valence band, Figure 8-18 (b). The hole-electron pair created by the
photon absorption is now free to flow in response to the external reverse bias. The resulting

-
current amplitude is a direct indication of the photon flux density striking the photodetector
surface.

JUNCTION
REGION

I I
(a) p-n JUNCTION

REVERSE BIAS VOLTAGE

CONDUCTION
(b) REVERSE
REVERSE BIASED
JUNCTION

FORBIDDEN
BAND (Eg)

HOLE- ELECTRON
HOLE-ELEC:TRON
PAIR CREATION
CREATlON VALENCE
BAND
7
Figure 8 - 1 The p-n junction photodetector.

The simple p-n junction has several pitfalls best described by being slow (ie, a long rise time)
and inefficient. There is nothing to stop the photons from being absorbed in the p or n regions
as well as the junction region. Of course the applied voltage is felt mainly across the junction
region. Electron-hole pairs created in the p and n regions do not move rapidly because there
is relatively low voltage present in these regions, hence most of them recombine before being
able to contribute any current flow. Electron-hole pairs created in the junction region, where
a larger electric field exists, move rapidly through the junction before any significant recom-
bination can take place thus contributing to current flow. The lack of current contribution
from the photons absorbed in the p and n regions makes the simple p-n junction an inefficient
and slow response photodetector.
An improvement on the p-n junction photodetector is the PIN photodiode. This is again
a p-n junction with a rather thick intrinsic semiconductor (eg, Si or Ge) layer separating the
p and n regions, hence the designation PIN. An intrinsic semiconductor is relatively a non-
conductor at room temperature compared to the impurity semiconductors (p and n) which
have excess free charge carriers (holes or electrons) at room temperature. Therefore, the ap-
236 8 Optical Fibers and Integrated Optics

plied voltage across a PIN photodiode appears mainly across the intrinsic layer. Most of the
electron-hole pairs created by photon absorption are generated in the large intrinsic layer
where they spend insufficient time to recombine due to the presence of the applied electric
field. The large proportion of electron-hole pair creations in the intrinsic region compared to
the much smaller p and n regions makes the PIN photodiode a much faster and more efficient
photodetector than the p-n junction.
A variation of the PIN photodiode is the avalanche photodiode (APD).This device is made
to work much like a photomultiplier tube in that it has internal gain. A large, nearly intrinsic
layer, like that in a PIN photodiode, is used to localize electron-hole pair generation. In one
type of APD the free electrons move to a junction region where a high-voltage drop exists.
The voltage drop across the junction is controlled by the density of impurity dopants in the
materials on either side of the junction. In the junction region, electrons are highly accelerated
and many collide with neutral atoms to form electron-hole pairs. More than one electron-hole
pair can be created per accelerated electron. These newly created charge carriers can collide
with neutral atoms to create additional electron-hole pairs, thus an avalanche effect occurs
resulting in internal gain. The reverse bias voltage required across an APD is necessarily larger
than that across a PIN photodiode and is directly associated with the avalanche gain. The bias
voltage, however, should not exceed the breakdown voltage which may vary from 10’s to 100’s
of volts.
The APD has a temperature-dependent mean free path between electron-hole pair
generating collisions; thus the APD may require temperature stabilization in some applica-
tions. The design engineer must trade this phenomenon against the lower noise generation
available from an APD with its internal gain mechanism versus a PIN photodiode used in con-
junction with an external preamplifier.

8.8 Sensor Components

There are many components which have been conceived for optical fiber sensors. Indeed,
at least one prominent industrial concern has applied for more than 150 United States patents.
There are many less components generally seen in practical use. Here we will mention two of
the most common.

8.8.1 Graded Index (GRIN) Lens

To make a lens, it is possible to start with a cylindrical glass rod and cut it into disks of
various thickness. Then the glass disks can be ground and polished so that from the center
of the disk to the edge, the glass tapers (convex lens) or becomes thicker (concave lens). The
convex lens is known as a converging lens, and the concave lens as a diverging lens since they
will cause parallel rays of light to converge or diverge, respectively, upon passing through the
lens.
Another method to accomplish the same result is to cause the index-of-refraction (n)of the
glass to vary radially from the center to the edge of the disk. The disk can remain a right cir-
8.8 Sensor Components 237

cular cylinder and still have focusing power. In Figure 8-19a a right circular-cylinder glass rod
is shown with its refractive index profile. The index takes a step function increase at the air
glass boundary and then is parabolic from the edge to the center of the cylinder. The index
is made to vary by doping the glass rod, usually at elevated temperatures, with any of several
impurity compounds. The variable index is known as a graded index, hence components like
that shown in Figure 8-19a are referred to as GRIN rod lenses.
In Figure 8-19b several rays are shown traversing a GRIN lens. Typically the rays on the
optic axis (OA) will not be curved. Those not parallel to the optic axis will undergo greater
curvature (refraction), the greater the angle they make with the optic axis due to experiencing
a greater change in refractive index. GRIN rods are usually doped to obtain a parabolic index
profile. Such a profile will cause all rays diverging from a point on the optic axis to transit
a complete curvature cycle in a distance defined as the pitch of the lens. The lens pitch is il-
lustrated in Figure 8-19b.
In Figure 8-19c a cross section of a quarter pitch GRIN lens is shown. The qbarter pitch
lens will collimate the light from a point source, or focus a collimated beam to a pojnt, within
its 1ength.it.k rather common to use GRIN lenses which are slightly longer than quarter pitch

Figure 8-19.
Graded index (GRIN) lens:
(a) index profile;
(b) GRIN lens pitch;
(c) quarter pitch GRIN
lens.
238 8 Optical Fibers and Integrated Optics

to focus a light source (eg, a semiconductor laser or LED) onto a fiber core or the output of
a fiber onto a semiconductor (eg, PIN or avalanche photodiode). The GRIN lens geometry
is naturally suited to optical fibers and has become a standard feature of optical fiber sensor
systems.

8.8.2 Optical Fiber Couplers

The fiber optic coupler performs the same function for light traveling in optical fibers as
the classical beamsplitter performs in traditional optics. It splits a beam traveling in one fiber
so that it branches into two fibers. Any splitting ratio is possible; 50-50, 40-60, 30-70, ----.
In addition, many splitting structures are possible, four port (2 x 2) and star couples (1 X n)
are common. The fourt port is the coupler most often seen in fiber sensors. Here two types
of four port couplers will be briefly described.
Common in laboratory and experimental use is the polished silica block (PSB) coupler,
Figure 8-20. The principle of the PSB coupler is illustrated in Figure 8-20a. Here two fiber
cores are shown in close proximity to each other. Light traveling in the top fiber is shown to
have an energy field with exponentially decaying tails beyond the core and into the cladding.
This is the evanescent field associated with the phenomenon of total internal reflection. The
evanescent field Poynting vector is parallel to the corekladding interface so that no energy is
radiated out of the core. The evanescent field simply propagates in the cladding parallel to
the corekladding interface. However, if a second (bottom) fiber core (ie, waveguide) is brought

->*q
close to (within approximately one micron or less) the energy propagating core (top), so that
the second core intersects the evanescent field of the first core, some light energy will be
shunted into the second core. This shunted energy will continue to propagate in the second
core, and an effective coupling of energy will have taken place.

CLADDING

(b)

Figure 8-20.
Polished silica block (PSB)
coup1er :
(a) Fiber core and cladding with
energy profile;
(b) polished coupler half;
(c) adjustable coupler.
8.9 Integrated Optics (10) Circuitsfor Sensors 239

The PSB coupler accomplishes this coupling function as shown in Figures 8-20b and c. In
part (b) of the figure a fiber has been placed in a curved groove of a silica block. The groove
is typically filled with an epoxy compound and the block polished until the surface is within
somewhat less than a micron of exposing the fiber core. A second silica block is prepared in
the same manner. The two blocks are then placed polished surface to polished surface as in
part (c) of the figure. Sliding the polished surfaces over one another varies the distance bet-
ween cores and hence the coupling ratio. One thus has a variable ratio four port fiber coupler.
The silica blocks can be fused together if desired to maintain a fixed coupling ratio.
Another type of optical fiber coupler, which is based on evanescent field coupling, is the
fused biconical coupler illustrated in Figure 8-21. In the fused biconical coupler fabrication
process, two fibers are twisted about each other and heated until the silica is semi-molten.
Then the fibers are put under tension and drawn until they squeeze together sufficiently to
obtain the desired coupling ratio. During the drawing process the evanescent fields cross cou-
ple into the adjacent fibers due to the squeezing process. The fibers also taper somewhat to
a smaller diameter near the middle of the twist. Thus each fiber resembles two truncated and
elongated cones joined at their truncation points. The fibers fuse together once the molten
silica cools leaving a fused biconical coupler. These couplers are rugged, low loss, temperature
stable and simple to fabricate. The process works with single mode, multimode and polariza-
tion maintaining fibers. Such features make the fused biconical coupler the most common op-
tical fiber coupler in practical fiber optic systems and sensors.

Figure 8-21.
Fused biconical optical fiber coupler.

8.9 Integrated Optics (10) Circuits for Sensors

Research and development of Integrated Optics (10) technology has been ongoing for
20 years (since 1968). During this period, the technology has moved from the research
laboratory to a number of actual applications. A number of good general references exist on
integrated optics and Integrated Optic Circuits (IOC’s). For the reader interested in a more
detailed discussion, two references by Hutcheson [18, 191 may be helpful.
10 is the technology of integrating several optical devices with associated waveguides in a
single chip or substrate. Like integrated electronics, integrated optics promises the possibilities
of devices and systems that would otherwise be too cumbersome or costly to be made or used
in bulk optics form.
Integrated optics technology provides key advantages and unique capabilities for fiber-optic
sensor systems [20]. These advantages are large bandwidth, electromagnetic interference
(EMI) immunity, small size, ruggedness, consistent reliability compared to bulk optic systems,
and the potential for low-cost devices when mass produced. In addition, due to the small
geometry associated with the I 0 waveguides, electro-optic modulation can be achieved with
low-voltage and high-energy efficiency. I 0 has made some fiber-optic sensor systems practical.
240 8 Optical Fibers and Integrated Optics

Some of these systems, when built with bulk optics, are very cumbersome. Neither can some
be effectively implemented with “all-fiber” technology. In such cases I 0 technology is the best
answer. The most prominent of these systems today is the fiber optic gyroscope (FOG). A
Multifunction 10 Circuit (or MIOC) can incorporate all the functions of the FOG except the
source, detector and fiber coil as will be illustrated in Section 8.9.4.
However, I 0 technology has its own limitations, mainly in size due to the finite wavelength
dimensions of light, and the minimum interaction length required to produce useful effects
for components. These useful effects are: switchinghplitting, modulation, and polarization
filtering. Integrated optical waveguides and components can also be seriously limited by
waveguide scattering and losses.
The number of components that can presently be integrated in a substrate is severely
limited, first by losses due to bends, and second by the limited real estate available because
only small-size single-crystal substrates can be obtained. To obtain a large number of com-
ponents on a single substrate, waveguide bends must be used to guide light from one compo-
nent to another. Excess optical loss can occur due to discrete angle bends, curved waveguides,
and mode coupling.

8.9.1 IOC Materials

Optical waveguide structures and devices have been manufactured using various materials
and a variety of techniques. Figure 8-22 shows the cross section of three key channel waveguide
structures now being used to form integrated optic devices. The key substrate materials for
channel waveguide fabrication today, are LiNbO, (lithium niobate), 111-V compound
semiconductors (GaAs), Si (silicon) and glass. Each of these substrate materials for IOC
technology has its own strengths and weaknesses. Each will be briefly described in the follow-
ing paragraphs. A more extensive discussion on LiNbO, will then follow, and several com-
ponents will be shown.
LiNbO, is presently the leading IOC substrate material. Channel waveguides are defined in
LiNbO, substrates by photolithographic mask techniques. The optical channel waveguides
are fabricated by either titanium-in-diffusion or proton-ion exchange [21]. A titanium channel
waveguide is illustrated in Figure 8-22a.

OPTICAL BEAM
CONFINEMENT

UPPER I
CLADDING DOPED
WAVEGUIDE

LiNbOj LOWER Si
SUBSTRATE CLADDING SUBSTRATE SUBSTRATE
LAYER
(a) Ib) (C)

Figure 8-22. Cross sections of optical channel waveguides made on the following substrates:
(a) LiNbO,, (b) GaAs, (c) Si.
8.9 Integrated Optics (lo)Circuits for Sensors 241

LiNbO, devices are primarily used in fiber optic sensor and communication systems.
Devices that have been demonstrated, are: phase modulators, amplitude modulators, switches,
splitters, and polarizers. Presently, for all these devices, LiNbO, is the material of choice due
to its excellent waveguiding, piezoelectric, and electrooptical properties. It is the most desirable
material for light modulation in fiber systems because of its high electro-optic modulation
coefficient. The only devices that have not been demonstrated in LiNbO, are sources and
detectors.
In GaAs substrates the GaAlAs (III-V compound semiconductors) channel waveguide
structures are formed by either molecular beam epitaxy (MBE) or metal oxide chemical vapor
deposition (MOCVD) to grow multiple layers with each having a different fractional of
aluminium concentration. The refractive index of such a structure is determined by the
aluminium fractional concentration. After deposition, channel waveguides are formed by
etching to form a ridge, as shown in Figure 8-22b. The guided optical beam is confined to
the cross hatched area shown. Waveguide propagation losses for GaAs are being reported in
the range of 0.2 to 1 dB/cm. These very low values are comparable to what is achieved with
LiNbO, . III-V compound semiconductors are the ultimate substrate for integrated optical
circuits. These materials have the potential to provide sources, detectors, waveguide com-
ponents, such as splitters and modulators and high-speed electronic circuitry in the same
substrate. These circuits are the so-called “OEICs” (Opto-Electronic-Integrated-Circuits).
They are expected to revolutionize the optical world just as integrated circuits revolutionized
the electronics world.
Another substrate material being used for integrated optic channel waveguide structures
is silicon. The waveguiding material structure for Si substrate IOC’s is very similar to that
of an optical-fiber waveguide. A waveguiding core is formed with a doped silicon dioxide
(SiO,) region surrounded by a slightly lower refractive index undoped SiO, cladding as in
Figure 8-22c. For many passive applications, the use of silicon as a substrate for optical chan-
nel waveguides is an ideal approach. These applications include multiplexers and
demultiplexers for communication and sensor systems. In such applications silicon offers the
unique advantage of a very low loss and low-cost substrate as well as an established fabrication
technology.
The last type IOC substrate material is glass. Glass based I 0 devices have the potential to
be used for both single and multimode systems. Glass is a passive material and, therefore, the
technology is limited to very simple applications such as branching circuits. These branching
circuit devices are equivalent to fiber devices such as couplers and multimode stars.

8.9.2 LiNbO, Fabrication Process

LiNbO, IOC device fabrication begins with the design of the waveguide configuration in
the mask. This mask serves as the pattern for the photolithographic process. This process is
shown in Figure 8-23 for Titanium in-diffused waveguides. First the photoresist material is ex-
posed to UV light which passes through the mask pattern. In a subsequent development step
the exposed photoresist material is chemically dissolved away leaving a waveguide pattern in
the remaining photoresist. A metal coating is then laid on top followed by removal (lift off)
of the metal coated, UV exposed photoresist iayer. This leaves a titanium waveguide pattern
which is then diffused into the LiNbO, substrate by heat treatment. A similar mask and
242 8 Optical Fibers and Integrated Optics

MASK Li:T PHOTORESIST PHOTORESIST

/ ii
EXPOSURE DEVELOPMENT METALIZATION LIFTOFF DIFFUSION

Figure 8-23. Processing steps for titanium in diffusion optical waveguide fabrication using the lift-off
technique.

photoresist process is also used for the newer proton exchange waveguide fabrication process.
After the waveguide has been formed, the next step is the electrode fabrication. The steps
in the electrode patterning are similar to the waveguide patterning and are shown in Fig-
ure 8-24.
Two basic LiNbO, substrate orientations are used today: X-cut, Y-propagating, and Z-cut,
Y-propagating. The TE mode is typically the preferred propagating mode in Z-cut. Optical
modulation of the light beam propagating through the waveguide is accomplished in either
case by applying an electric field parallel to the direction of the optical field polarization.

uv
ILLUMINATION
PHOTORESIST

WAVEGUIDE WAVEGUIDE

EXPOSURE DEVELOPMENT METALIZATION LIFT OFF

Figure 8-24. Processing steps for electrode deposition.

8.9.3 LiNbO, Components

One type of passive component is a Y-junction. It serves the function of a power splitter
or combiner. The Y-junction consists of two adjacent waveguides that merge into one in a
shallow angle as shown in Figure 8-25. It is typically used to obtain an even split into the arms
of a Mach-Zehnder or a Sagnac optical-fiber interferometer. The symmetry of the Y-junction
determines the power splitting ratio.
Another device is the passive directional coupler as shown in Figure 8-26, used for switching
and power splittingkombining functions. The directional coupler is made of two adjacent
waveguides placed close enough to each other, such that there is an overlap in the evanescent
field of the two guides as with the optical fiber coupler of Section 8.8.2. In such a case there
is an exchange of optical energy between the two waveguides. The passive direction coupler
configuration can be turned into an active optical switch with the photolithographic applica-
tion of electrodes on the substrate surface adjacent to the indiffused waveguides. A voltage
applied to the electrodes creates an electric field which penetrates the waveguides and can alter
the splitting ratio so that an effective switching occurs.
8.9 Integrated Optics (IO) Circuits for Sensors 243

Figure 8-25. SUBSTRATE


Y-junction splitter. WAVEGUIDE

Figure 8-26.
Passive directional coupler. WAVEGUIDES SUBSTRATE

Figure 8-27.
Integrated optics phase SUBSTRATE
modulator.

A third, and probably most important device, is the I 0 phase modulator shown in Fig-
ure 8-27. It consists simply of a straight waveguide and a set of electrodes. A voltage applied
between the electrodes causes a change in the waveguide refractive index, creating a phase
change of the light propagating in the waveguide. This is the easiest and simplest active com-
ponent that can be fabricated.

8.9.4 An IOC Application

As discussed previously, integrated optics brings to fiber optic sensors many advantages.
Several fiber optic sensing systems incorporate these advantages. For example, sensor systems
244 8 Optical Fibers and Integrated Optics

which use IOCs are typically those for measuring velocity, displacement, position, electric
field and in particular rotation, ie, fiber optic gyroscopes. Some of the specific advantages in-
tegrated optics brings to fiber optic gyroscopes, are low drive voltage, larger bandwidth phase
modulation, low cost, and polarization preservation between components/devices in the same
substrate.
Multifunction IOC’s (MIOC’s) are incorporated into the developmental designs of fiber
optic gyroscopes by several manufacturers. Employment of specific IOC components varies
with the manufacturer. The Y-junction is the most popular IOC component because of its
simplicity. Among the component design variations is a 3 x 3 directional coupler functioning
as both the polarizer and a 3 dB (50: 50) splitter in an interferometric fiber optic gyroscope
as shown in Figure 8-28 [22]. The 3 x 3 directional coupler, with phase modulator electrodes
on either side of one of the output channels has been fabricated in Z-cut, Y-pro-
pagating lithium niobate using titanium in-diffused waveguides for single mode operation at
rl = 1300 nm.
Design of the 3 x 3 directional coupler maintains the TE polarization as the throughput
state (bar-state) and the TM polarization as the cross-state. The TM light entering the center
waveguide splits into the outer guides which are connected to the gyro coil. The TE light con-
tinues along the center waveguide and is dumped into an output fiber where it is lost. In order
to reduce Fresnel reflections at the fiber to lithium niobate substrate interfaces, angled edges
are cut on the substrate. This technique effectively decouples the reflections from the guided
wave structure.
The fiber optic gyroscope, with an MIOC integral to its design, has proceeded through
development to the point where it has been flight tested as part of an Inertial Measurement
Unit (IMU) by at least two manufacturers. It is clear that IOC’s will play a major role in many
optical fiber sensor systems of the future.

Figure 8-28. Phase interferometric fiber optic gyroscope.


8.11 References 245

8.10 Summary

We have attempted to present here only a representative sampling of intrinsic and extrinsic
optical fiber sensors, several commonly used components and a brief introduction to in-
tegrated optic (10) circuits. The intent is to introduce the reader to the subject so that a little
knowledge is imparted on what is available from optical fiber sensor and I 0 technology, not
to cover the technology in depth as might be done in a format of one or more volumes. Bear
in mind that most optical fiber sensor and 1 0 technology is still in the development stage.
Many proof-of-principle devices exist, lots of devices are in the prototype category and some
sensors are available off-the-shelf. The bulk of the technology has been represented herein,
albeit, in a somewhat cursory manner. It should be possible for the reader to now ascertain
if his need for, or interest in, sensors should further indulge the technology of optical fiber
sensors.

8.11 References

[l] DePaula, R. P., Moore, E. L., Proceedings SPIE 556 (1985) 2-15.
[2] Culshaw, B., Optical Fibre Sensing and Signal Processing, London: Peter Peregrinus Ltd., 1984.
131 Wysocki, J. A., Proceedings SPIE 988 (1988) 124-130.
141 Miller, S. A., Ultra Low-Loss Communications in the Mid-IR, Phontonics Spectra, (1986) 87-90.
[5] Tajima, K., Sasaki, Y., Lightwave Technology 7 (1989) 674-679.
16) Aoyagi, T., Proceedings SPZE 840 (1987) 10-18.
171 Giallorenzi, T. G., et al., IEEE Trans. on Microwave Theory and Techniques 30, (1982) 472-511.
181 Dakin, J. P., Pratt, D. J., Proceedings SPIE 556, (1985) 249-256.
191 Sagnac, G., C. R. Acad. Sci. 95, (1913) 708.
[lo] Vali, V., Shorthill, R. W., Appl. Opt. 15, (1976) 1099.
1111 Stokes, L. F., Chadorow, M., Shaw, H. J., J. Lightwave Technology Lt-1, (1983) 110.
1121 Moore, E. L., DePaula, R. P., Advanced Imaging 8, (1987) A48-A50.
1131 Belkerdid, M. A., Ghandeharioun, N., Proceedings SPIE 566, (1985) 153- 158.
1141 Tromberg, B. J., Sepaniak, M. J., Vo-Dinh, T, Proceedings SPIE 906, (1988) 30-38.
1151 Krohn, D. A., Proceedings SPIE 7l8, (1986) 2-11.
1161 Berthold, J. W., Proceedings SPIE 838, (1987) 2-8.
1171 Suematsu, Y., Advances in Semiconductor Lasers, Physics Today, May 1985, 32-39.
[18] Hutcheson, L. D., ed., Integrated Optical Circuits and Components:Design and Applications, New
York: Marcel Decker Inc., 1987.
1191 Hutcheson, L. D., Integrated Optics: Evolution and Prospects, Optics News, 14-2 (1988) 7-31.
1201 Sellers, G. J., Sriram, S., Integrated Optic Technology, Laser Focus, Sept. 1986, 74-82.
1211 Suchoski, P. G., Findakly, J. K., Leonberger, F. J., Proceedings SPIE 993, (1988) 240-243.
1221 Minford, W. J., DePaula, R. P., Bogert, G. A., Technical Digest, Optical Fiber Sensors (OFS’ 88),
FBB-2,(1988) 385-392.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

9 Ceramics and Oxides


JORG ARNDT.Universitat Tubingen. FRG

Contents
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
9.2 Electrical Conducting Materials . . . . . . . . . . . . . . . . . . 252
9.2.1 Ionic Conductors . . . . . . . . . . . . . . . . . . . . . . . . 252
9.2.1.1 Zirconia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
9.2.1.2 P-Alumina . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
9.2.1.3 NASICON . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
9.2.2 Semiconductors . . . . . . . . . . . . . . . . . . . . . . . . . 266
9.2.2.1 Titania . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
9.2.2.2 Tin Dioxide . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
9.2.2.3 Zinc Oxide . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
9.2.2.4 Perovskite-npe Oxides . . . . . . . . . . . . . . . . . . . . . . . 269
9.3 lnsulating Materials . . . . . . . . . . . . . . . . . . . . . . . . 271
9.3.1 Piezoelectric Materials . . . . . . . . . . . . . . . . . . . . . . . 271
9.3.1.1 Perovskite-npe Materials . . . . . . . . . . . . . . . . . . . . . 271
9.3.2 Pyroelectric Materials . . . . . . . . . . . . . . . . . . . . . . . 274
9.3.3 Ferroelectric Materials . . . . . . . . . . . . . . . . . . . . . . . 275

9.4 Negative Temperature Coefficient (NTC) Ceramic Thermistors . . . . . 275

9.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . 276

9.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277


248 9 Ceramics and Oxides

9.1 Introduction

Ceramic oxide materials play an important, steadily increasing role in almost all fields of
electronics [l]. Applications encompass ceramic parts which themselves do not have an elec-
tronic function, such as insulating substrates and integrated circuit packages, components
which are externally activated to perform desired functions such as piezoelectric and elec-
trooptic transducers, and active elements such as gas sensors and solid-state electrolytes.
Ceramic materials possess a number of unique advantages, especially the abundant
possibilities of optimizing performance and tailoring to specific demands by modification of
chemical composition and/or variation of microstructure by changing the parameters in the
fabrication processes. In order to meet the increasing requirements related to developments
in electronics, new non-conventional methods of preparation and processing of the starting
powders and sophisticated fabrication techniques are being developed. Also, it was recognized
that the standard requirements of the electronics industry, ie, high-purity raw materials, pro-
cessing and production in a clean-room environment, rigorous control of fabrication pro-
cesses, guarantee of reproducibility in the properties of components, and severe control of pro-
duct quality, must inevitably be adoped in the fabrication of ceramic products for electronic
applications.
A concern of basic importance is the scientific elucidation of the relationships between the
microstructural characteristics of the ceramics and their properties. A ceramic is a complicated
product consisting of more or less statistically oriented polycrystalline grains, intergranular
secondary amorphous or crystalline phases, chemical impurities, trapped gases, pores and
microstructural defects such as inclusions and cracks.
The property of interest may be predominantly controlled by phenomena taking place in
the volume, at grain boundaries, or at the surface. Generally, these various phenomena act
simultaneously with different relative importances. In order to gain an insight into these com-
plex microstructure-property interrelations and to understand the observed phenomena,
classical methods of investigation such as polarizing microscopy, scanning electron
microscopy (SEM), transmission electron microscopy (TEM) with energy-dispersive X-ray
analysis (EDX), X-ray diffraction (XRD), infrared (IR) and Raman spectroscopy, together
with more sophisticated methods of surface and depth profile analysis such as Auger electron
spectroscopy (AES), secondary-ion mass spectrometry (SIMS), X-ray photoelectron spec-
troscopy (XPS), and electron energy loss spectroscopy (EELS) must be used in conjunction
with highly reliable methods of property measurements.
The aim of this chapter is to give a survey of major basic developments in ceramic oxide
materials used for sensors. The most important of these materials, their functional properties,
and their major applications as sensor devices are listed in Table 9-1. Ceramics used for
substrate applications and for high-T, superconductors are not considered here. Also, a
general treatment of the various methods, mechanisms, and theories of sintering is not given,
as comprehensive representations of this complex may be found in the specialized literature
and text books (eg [2]).
9.1 Introduction 249

Explanations to the equations in the following Table 9-1:

Equation (9-1) EMF (mV): Electromotive force


R Gas constant
T (K) Absolute temperature
F Faraday constant
Equation (9-2) u Conductivity
0 0
Constant
E* Activation energy
kB Boltzmann constant
m Constant, varying from 4 to 6, dependent on the nature of the lat-
tice defects
Equations (9-3, 9-4) R Resistance at temperature T
Ro Resistance at standard temperature To
B EJkB
a Temperature coefficient of resistance
Equation (9-5) I Current density
F Applied electric field
a Characteristic varistor exponent
a = 1 ohmic varistor
a -, OD perfect varistor
25 Ia I 50 typical varistor value
Equation (9-6) Pi Components of polarization vector
Oj k Components of stress tensor
drjk Piezoelectric moduli
Equation (9-7) ejk Components of strain tensor
E, Components of electric field vector
Equation (9-8) AP, Change in the components of the polarization vector related with
temperature change A T
Pyroelectric coefficients
Table 9-1. Functional properties and general applications of important ceramic oxide sensor materials

Ceramic oxide Phenomenological representation of


Class of material Functional properties General application(s)
material functional property
_ _ ~ ~ ~ ~

Electrically conducting I. Volume properties


materials
Ionic conductivity ZrO, Solid electrolyte Nernst equation: (9-1)
Oxygen gas sensor (A-sensor)
Oxygen pump

p, v-alumina Solid electrolyte in sodium/


sulfur battery
NASICON Solid electrolyte in batteries
Gas sensor
Semiconductivity
TiO, Oxygen gas sensor Dependence of conductivity on
temperature and oxygen partial
pressure (9-2)
SnO,, ZnO Gas sensor for hydro-
carbon gases

Perovskite-type Oxygen gas sensors


oxides (eg, Sr TiO, ,
BaTiO, , SrSnO,)

Negative Temperature Ionic conducting Temperature sensors, Temperature dependence of ther-


Coefficient (eg, ZrO-Y,O,)anc temperature compensation mistor resistance: (9-3)
of Resistance (NTC) semiconducting R = R, exp [ B (1/T - l/To)]
(eg, NiO -TiO,)
(9-4)
thermistor materials
II. Grain boundary properties
Positive Temperature Doped BaTiO, Selfregulating heating
Coefficient of elements, temperature
Resistance (FTC) compensation
Varistor property ZnO Overvoltage protection, (9-5)
surge current absorber 1,/12 = ( F , / F J a
~

III. Surface effects


Varistor BaTiO, Electronic sensors

Surface ionic SO,, ZnCr,O, Humidity sensors


conductivity

Volume properties

Piezoelectricity perow kite-type Pressure sensors, capacitors, Direct piezoelectric effect: (9-6)
materials vibrators, oscillators, spark
(BaTiO, , PZT, generators
PLZT), Quartz Converse piezoelectric effect: (9-7)
ejk = dbk E,
Pyroelectricity PZT-like ferro- Infrared radiation detectors, Pyroelectric effect: (9-8)
electric materials temperature sensors
. . 1
PbTiO, , LiTaO, ,
SrNbO,
Ferroelectricity Solid solutions Capacitors, electrooptic
based on BaTiO, , applications (light memory
PZT solid solution elements, light modulation,
light shutters)
252 9 Cemmiu and Oxides

9.2 Electrical Conducting Materials

9.2.1 Ionic Conductors

9.2.1.1 Zirconia

According to the phase diagram shown in Figure 9-1, pure ZrO, appears in three
crystalline modifications at normal pressure. Between the melting point at 2680°C and
2372"C, ZrO, crystallizes in the face centered cubic fluorite structure. Each Zr4+ cation is

Tetragonal 800- \
\ rn=3O/kbar

Loo
- Mon.
I

0
0 20 40 60
Pressure (kbarsl

0 02-
ZrL+
MonoIinic
Figure 9-1. Phase diagram of ZrO, [3] and structures of the cubic, tetragonal and monoclinic ZrO,
phases.
9.2 Electrical Conducting Materials 253

surrounded by eight 0,-anions arranged in form of a cube and each 0,-is regularly
tetrahedrally coordinated to four Zr4+. Below 2372°C the structure is tetragonal, Zr4+ and
0 ions in distorted eight- and four-fold coordinations, respectively. This structure is stable
down to about 1200°C. Below this temperature, the symmetry of the structure is decreased
to monoclinic, with a seven-fold coordination of the Zr4+ cations. This monoclinic phase is
stable down to room temperature. At high pressure, an orthorhombic structure of ZrO,
becomes stable. However, there are still uncertainties in the basic phase relations in the ZrO,
system [3, 4, 51.

Partially and Fully Stabilized Zirconia

The martensitic phase transformation from the tetragonal to the monoclinic structure is ac-
companied by a considerable volume expansion of 3-5%, which invariably results in cracking
of the material even on a small scale. This volume instability prevents the production of large
pieces of pure zirconia.
By adding oxides such as Y,O,, CaO, or MgO, which form solid solutions with ZrO,, it is
possible to stabilize the cubic structure down to room temperature and thus to avoid the
tetragonal-monoclinic transformation. Figure 9-2 shows the low-yttria region of the phase

3000

-
-
Liquid IL)
f
I
I
L+F

-u
-
0

;
K
c

5
I
I

tetragonal IT') I
Figure 9-2.
Low-yttria part of the ZrO,-YzO, phase 0 5 10 15
diagram [ 5 ] . Mole % YO,,5
254 9 Ceramics and Oxides

diagram of the system Zr0,-Y203. In the compositional range from 0 to about 1.8 mol-Vo
Y,03, a monoclinic phase field exists above which a narrow field of monoclinic plus
tetragonal solid solutions appears with increasing temperature. This is followed by the stability
field of a tetragonal solid solution, which may transform on cooling to the monoclinic struc-
ture. For compositions containing more than about 2.5 mol-To Y203, a large field of mix-
tures of non-transformable tetragonal and cubic solid solutions exists. With further increase
in Y,O, concentration an extended stability field of cubic solid solutions is encountered
which is stable from room temperature up to the melting point. These phase equilibria in the
ZrO, - Y,03 system are fundamental for the phenomena of partial or complete stabilization.
For example, a partially stabilized zirconia (PSZ) ceramic consisting of an intimate mixture
of tetragonal, cubic, and/or monoclinic phase may be obtained either by introducing a Y203
concentration that is insufficient for complete stabilization of the cubic phase or by heat
treating a cubic phase to develop a two-phase ceramic.
In the system ZrO, - CaO and ZrO, - MgO phase equilibria involving monoclinic,
tetragonal, and cubic solid solutions also exist which provide possibilities of producing par-
tially or fully stabilized zirconia ceramics (eg, [4, 51). An understanding of the phase equilibria
in these binary systems and in systems of zirconia with other oxides is essential for the applica-
tion of zirconia both as a functional and structural ceramic material. The coexisting phases,
their amount, size, and distribution, the relationships of the phases and their transformation
kinetics and environment, and the kind of the cations added have a profound influence on
almost all properties of zirconia ceramics. However, as with the pure ZrO, system, phase
equilibria in both binary and ternary systems involving ZrO, are not well established.

Zirconia Ceramics as Oxygen Sensor Materials

The use of yttria-stabilized zirconia as a solid oxide electrolyte goes back to Nernst, who
in 1897 invented the “Nernst light”. However, only recent studies have provided a closer
understanding of the conduction phenomena in zirconia. The replacement of Zr 4 + cations
by cations of lower charge results in the formation of vacant sites in the anionic oxygen lattice.
Both the distribution of the stabilizing cations and the distribution of the resulting vacant sites
are statistical. These oxygen vacancies may move through the lattice, thus providing electric
conductivity by the transport of electric charge. In order that the energy required for the move-
ment of oxygen from a vacant site to the next through the lattice is minimal, it is necessary
that at least one of the neighboring anion sites is also a vacancy, a requirement that is best
provided by the statistical distribution of the vacant sites.
On the other hand, conductivity is affected by the ionic radius of the cation replacing zir-
conium, as the oxygen ion must move through a tetrahedral face formed by three cations in
order to reach the next vacant site. The energy required will therefore increase with increasing
radius of the cation. The radius of Zr4+ is 0.79 A, and those of Y3+ and of Ca2+ are 0.92
and 0.99 A, respectively. Therefore, the conductivity of zirconia stabilized by Y 3 + should be
higher than that of zirconia stabilized by Ca2+. This is actually observed, as shown in
Figure 9-3, which shows the ionic conductivity as a function of temperature of various
stabilized zirconia ceramics and of other ceramic oxides commonly used as ionic conductors.
The maximum in the ionic conductivity is observed at 8-9 mol-To Y,O, and 12- 13 mol -To
CaO, respectively.
9.2 Electrical Conducting Materials 255

Temperature
1200 800 600 500 400 30
I I I I I I I I I I

1c

1c
x
c
.-
u
4-

3
U

u
0

10

10’

Figure 9-3. Ionic conductivity as a function of temperature for various ceramic oxides [ 6 ] .
1. CeO, (0.80)Gd,O, (0.20),2. ZrO, (0.90)Sc,O, (0.10), 3. Bi,O, (0.75)WO, (0.25), 4. La,O,
(0.945)SrO (0.055), 5. ZrO, (0.90)Y,O, (0.10), 6. CeO, (0.99)Y,O, (0.01), 7. ZrO, (0.87)
CaO (0.13), 8. CaOTiO, (0.70)A1,0, (0.15), 9. ZrO, (0.79)Nd,O, (0.21), 10. La,O, (0.35)
CaO (0.30) Al,O, (OSO), 11. ZrO, (0.87)La,O, (0.13), 12. ZrO,, 13. ZrO, (0.75)Gd,O,
(0.25). 14. ZrO, SrO, 15. ZrO, CaO.

Zirconia-based oxygen sensors are widely used in combustion control, especially in


automobiles, atmosphere control in furnaces and as monitors of the oxygen concentration in
molten metals.

Growth of Cubic Zirconia Single Crystals

Growth of ZrO, single crystals from the melt is difficult, as temperatures up to 2750°C are
required. Methods employing plasma arc or solar heating have been used. Recently, the “skull
melting” process has become a commercially important method in which the mixture of
ZrO, and stabilizing material, eg, 87.5 mol-Yo ZrO, and 12.5 mol-Yo CaO, is heated and
melted directly by radiofrequency induction heating at 4 MHz and power levels as high as
256 9 Ceramics and Oxides

100 kW [7]. Crystal growth is achived by slow cooling of the melt batch. Single crystals of
cubic zirconia with diameters of 50 mm and lengths up to 100 mm are grown by this method.
Single crystals of cubic zirconia are used primarily as a diamond simulant in the gem in-
dustry owing to the high refractive index of 2.15-2.18, which is only slightly smaller than that
of diamond (2.41 -2.43). Important scientific applications of cubic zirconia crystals are as
substrates for thin-film superconducting oxides and as samples for research on the physical,
particularly the electrical, properties of cubic zirconia.

Fabrication of Zirconia Ceramics

Ideally, the stabilizing additives should be distributed homogeneously in the zirconia matrix
on an atomic scale. This is difficult to accomplish by blending milled powders of zirconia and
an additive containing the stabilizing cation and calcining the resulting mixture, although this
method is currently the most common. Better results are obtained by using non-conventional
methods of powder preparation [8]. Homogeneous distribution can be achieved by calcination
of zirconium-chloride and nitrate, controlled coprecipitation of hydroxides of the consti-
tuents, the sol-gel process, chemical vapor deposition (CVD), or hydrothermal oxidation.
These processes yield highly reactive powders with very small spherical grains (< 1 pm),
which may be sintered at much lower temperatures than conventional oxide powders.
Of the non-conventional powder preparation methods, the sol-gel process is very promising
for achieving highly sinter-reactive, monosized powders with spherical grains with sizes in the
submicron range. These powders yield ceramics of homogeneous packing density and texture
at much lower sintering temperatures than are required for conventional powders. With the
sol-gel method, a very homogeneous distribution of the various components on the atomic
scale is attained in polycomponent systems. This is an important prerequisite for the produc-
tion of partially or fully stabilized zirconia ceramics. For example, from powders prepared by
the sol-gel process, PSZ ceramics with more than 95% of the theoretical density and a grain
size of 0.3 pm could be sintered at 1180°C [9]. Conventionally prepared oxide powders of
grain size up to 8 bm demand firing temperatures of 1900°C or even higher to obtain com-
pletely nonporous, vacuum-tight ceramic bodies as required for zirconia electrolyte applica-
tions.
Fabrication of zirconia ceramics is performed by methods common to most oxide ceramics
(Figure 9-4). The powders prepared by conventional or non-conventional methods are pro-
cessed to meet the demands of the final product. The processing begins, for example, by the
addition of stabilizing and sintering aids to the pure oxide powder. As the next step, the
powders are generally mixed with organic materials in order to obtain a plastic mass that can
be formed into the green, ie, unfired body. This may be accomplished by various methods such
as axial or isostatic pressing in dies, extrusion, slip casting, or tape casting. To drive off the
organic materials, the green bodies formed are subjected to moderate temperatures and are
then sintered at high temperatures to give the final ceramic product.
Each of the various steps of fabrication is critical for the performance of the ceramic and
much effort has been devoted to improving the processing details. However, many of the pro-
cess parameters are based on empirical experience and often rely on results of trial and error
investigations. Also, the formulations of organic forming additives are in most cases pro-
prietary information and are not published.
9.2 Electrical Conducting Materials 257

Conventional Methods Unconventional Methods


(Calcining,milling, mixing) (Solutiontechniques (eg, sol-gel method),
vapor phase methods, salt decomposition)

\ /
I Powder Processing I
(Addition of sinter aids, dispersion components,
dopants, organic materials (binders, lubricants))

1
(Axial and isostatic pressing, slip casting,
tape casting, injection moulding)

I
I Sintering I
(Sintering in various gas atmospheres at low pressure or in
vacuum, hot pressing (HP), hot isostatic pressing (HIP))

Thermal Annealing and Aging


~ ~~

(Eliminationof stresses, stabilization of properties)

1
Application Specific Final
Operations
(Grinding,polishing, shaping to required tolerances,
metallizing, electroding, coating for protection)

Figure 9-4. Flowchart for standard processes in ceramics fabrication.

The ZrO, electrolyte device used in automotive engines for the control of the air-to-fuel
ratio commonly consists of a tube of yttria-stabilized zirconia which is operated in the engine
exhaust gas stream at temperatures of about 900°C. In this environment, the sensor is sub-
jected to hostile conditions, especially thermal shock, deposition of chemical combustion pro-
ducts, and poisoning of the Pt electrodes. These effects may lead to degradation of the sensor
ceramic, resulting in deterioration of the electrical sensor characteristics such as an increase
in response time and a decrease in the voltage output.
Degradation of sensor performance may also result from mechanical failure of the ceramic
caused by the formation and propagation of cracks induced by mechanical and thermal
258 9 Ceramics and Oxides

stresses. Failure is invariably catastrophic owing to the inherently brittle nature of ceramics.
Improvement of resistance to failure is a central issue in the development of mechanically
reliable ceramics, eg, [lo]. A common approach is to increase fracture toughness by transfor-
mation toughening [ll]. This can be accomplished by using the tetragonal-monoclinic phase
transformation in partially stabilized zirconia (PSZ) or in a zirconia ceramic consisting com-
pletely of tetragonal zirconia polycrystals (TZP). Transformation in adjacent tetragonal ZrO,
particles of a critical grain size is induced by the stress field at the tip of a propagating crack
that deadlocks in the matrix ahead since this is placed under compressive stress as a result of
the volume increase related to the transformation (stress-induced transformation toughening).
Toughening can also be achieved by transforming the tetragonal particles on cooling into the
monoclinic structure. As a result of the volume expansion, microcracks are formed around the
particles in the matrix. These microcracks can extend in the stress field of a propagating crack,
thus absorbing its energy or they can deflect the crack, thus dissipating its energy. As a result,
the toughness of the ceramic increases. Such toughening improves the reliability of ceramics
against failure, resulting not only from initial flaws induced by raw materials and processing
components but also from flaws resulting from installation or operation.
The most important sources of crack formation are particularly pores due to inadequate
sintering procedures and inclusions or precipitates of foreign phases resulting from impurities
inherent in the powder materials or introduced during powder processing. It is evident that
the latter kind of flaws should be avoided or at least be reduced by using high-purity powders
and improving the powder processing techniques.
Sintering at normal pressure yields pore-free, ie, completely dense, ceramics only if either
complex sintering methods are applied or sintering aids are added to effect liquid-phase sinter-
ing [12]. In most cases, however, sintering at high pressure is required. Hot pressing in a heated
piston-die assembly is most frequently used (eg, [13]). Recently new, more sophisticated
techniques employing hot isostatic pressing (HIP) in a compressed and heated gas medium
at pressures and temperatures up to 200 MPa and 2000°C, respectively, have been developed
(eg, [14]). The gas used is generally argon or nitrogen.
Several HIP techniques are in use. Their common basic purpose is the elimination of pores
but they also offer a variety of other important possibilities for the improvement of materials
such as healing of defects and, especially, control of grain size and grain boundary phases dur-
ing sintering by appropriate selection of the process parameters. One technique is the HIP
powder consolidation method in which powder encapsulated in a can consisting of metal or
glass is hot pressed and sintered simultaneously, achieving sintered parts of high density. In
the sinter plus HIP technique, powders compacted, for example, by cold isostatic pressing are
sintered at normal pressure until closed porosity is attained, corresponding to a density of
93-95'70 of the theoretical value. The ceramic is then isostatically hot pressed without encap-
sulation, thereby eliminating the residual porosity. Completely dense ceramics may be ob-
tained after 1 h of hot pressing at 20 MPa gas pressure at temperatures of 100°C below the
normal sintering temperature [15]. In the sinter-HIP technique, sintering and hot isostatic
pressing are performed in one processing step. The ceramic is sintered at high temperature but
low pressure to closed porosity and then the pressure is increased to eliminate the residual
porosity.
Comparison of the relative merits of the different HIP technologies with regard to transfor-
mation-toughened ZrO, ceramic materials such as PSZ and TZP revealed that generally HIP
methods must be applied in order to utilize fully the strength potential of these ceramics [14].
9.2 Electrical Conducting Materials 259

The sinter plus HIP method without encapsulation is most promising, but the usefulness of
the sinter-HIP technique has not yet been completely explored. However, development of this
technique in conjunction with oxygen-containing gases as pressure media seems very promis-
ing for successful application to zirconia ceramics.
One of the most challenging goals in ceramics development is the reliable control of grain
boundary phases, which essentially control all properties of both functional and structural
ceramics. Grain boundary phases may be crystalline or amorphous. Generally, they result
from a liquid phase formed during sintering either from added liquid-phase sintering aids or
from impurities. Zirconia ceramics may contain glassy grain boundaries resulting from im-
purities present in the starting powders of zirconia and the stabilizing additives or are in-
troduced by processing such as powder milling. During sintering, ions may segregate into the
liquid grain boundary and secondary crystalline phases may be formed on cooling. The
chemical and structural nature of grain boundaries are difficult to study owing to their often
extreme thinness down to a few nanometers.
Grain boundaries in most cases have deleterious effects on the electrical and mechanical
properties of ceramics. Especially at high temperatures, amorphous grain boundaries soften
and become liquid. As a result, the mechanical strength of the whole ceramic may decrease
drastically. In addition, amorphous grain boundaries represent preferred paths for diffusion
and segregation of ions. In some cases, however, they are used to advantage, such as in ZnO
varistors.
In order to be able to influence grain boundaries, for example by controlled crystallization
of amorphous phases, their chemical and structural nature must be known. For chemical
analysis, SIMS is a most important spectroscopic method owing to its high detection power.
At present, it is the only method that offers topochemical information at ultratrace levels and
is by far the most sensitive method for bulk analysis. For microstructural characterization,
high-resolution TEM in conjunction with EDX and, for example, EELS for quantitative
analysis of the oxygen content in grain boundaries and adjacent grains is indispensable.

9.2.1.2 j-Alumina

p-Alumina is a generic term currently used to refer to nonstoichiometric Na,O -A1,0,


compounds derived from the as yet unknown sodium aluminate, Na,O . 11 Al,O, or
NaAl,,O,,. All these compounds contain excess Na compared with this composition. They
have high ionic and low electronic conductivity, excellent chemical inertness and high
mechanical and thermal strength.
The p-aluminas are currently the most widely investigated family of solid electrolytes for
batteries, for purification of liquid sodium and for use in high temperature sodium heat
engines. They permit not only fast diffusion of Na ions but also of other alkali metal ions,
+

other divalent and trivalent cations, and protons.


P-Alumina has been discovered by Rankin and Merwin [16] and was originally thought to
be a polymorph of A1,0,. A compound structurally very similar to p-alumina was found
later [17]. Its composition is Na,O . 5 A1,0, or Na,Al,00,6 and it has been named
p"-alumina. In addition, two other p-aluminas have been reported to occur in the ternary
system Na,O -MgO -Al,O, [18].
260 9 Ceramics and Oxides

Structures of 8- and p-Aluminti

Stoichiometric p-alumina has a hexagonal layered structure [19]. The structure is built up
by two spinel blocks bridged by oxygen atoms at widely spaced intervals and related by a two-
fold screw axis (Figure 9-5). In the spinel blocks the oxygen ions form a cubic closest packing
in which the A13+ ions occupy octahedral and tetrahedral interstitial sites. The mobile and
thus conducting Na + ions occupy sites in the relatively unpopulated two-dimensional planes
between the spinel blocks which are termed the conducting planes.

0 AP+
(a)
Figure 9-5. (a) Structure of stoichiometric P-alumina. A, B and C indicate oxygen layers of the spinel
blocks; B' and C' show the conducting planes. Each conducting plane of the unit cell contains
one N a + ion; (b) structure of p"-alumina (idealized). A, B and C indicate oxygen layers of
the spinel blocks; A', B' and C' show the conducting planes. Each conducting plane of the
unit cell contains two Na' ions [20].
9.2 Electrical Conducting Materials 261

As can be seen from the phase diagram of the system Na,O-A1,0, (Figure 9-6), the S-
alumina solid solution exists over a narrow compositional range, the width of which is nearly
indepedent of temperature. The general composition of p-alumina is Nal+xA111017+x,2, with
x ranging from 0.1 to 0.3. The excess sodium content may thus vary between 10 and 30%. A
typical composition is Na,,2,A1,,0,,,,2. The excess Na + ions also populate the conducting
plane between the spinel blocks, but not all available sites are occupied. The excess positive
charge is compensated for by defects and by oxygen ions in the conducting plane. Ionic con-
ductivity due to the movement of the Na + ions can therefore only take place in the plane of
the bridging oxygen ions but not through the spinel blocks.

2100

1900

1700

1500 BNaAIO, + p
1410
-y 1300
-
E
L
3
c.

1100
0.

E
l- II
900 I1
II
II
700 II
II
II
II
500 II
I1
II
Figure 9-6. 30C I I I

Alumina-rich part of the 1 60 70 80 90 1 D


Na,O -A1,0, phase diagram [21]. Mole % AI,O,

S"-Alumina is a thermodynamically unstable compound in the binary system


Na20-A120, [21]. However, its structure can be stabilized by the addition of cations which
are capable of occupying either the octahedral or tetrahedral sites in the spinel blocks. Li +
and MgZ+ are the most effective of those cations.
The structure of Li-stabilized p"-alumina is rhombohedra1 [22]. The structure contains three
spinel blocks related by a three-fold screw axis parallel to the c-axis. The oxygen planes bet-
262 9 Ceramics and Oxides

ween the blocks are staggered and result in slightly larger spacings through which the sodium
ions can migrate (Figure 9-5). P”-Alumina thus has a higher ionic conductivity than p-alumina
(Figure 9-7).

Temperature (OCI
800 200 0 -100

Figure 9-7.
Ionic conductivity as a function
of temperature for various ionic
conductors [23].
1. RbAg,I,, 2. 0” alumina,
3. 0 alumina, 4. Na,Zr,PSi,O,,,
5. Li,N, 6. Li,,,Zr,,,Ta,,,P,O,,,
7. CaO-ZrO,, 8. KCl,
103/T IK-’) 9. LiAISiO,.

The general formula of p”-alumina is Nal+xMxAlll~xO17, in which M represents a


divalent cation such as Mg2+, Ni2+ or Zn2+. A typical composition of 0”-alumina is
Na,,7Mgo,6~Al,o,330,7. The excess charge of the Na ’ ions is compensated for by substituting
the divalent or monovalent cation into a spinel block site normally occupied by A13+.
In p- and p”-alumina the sodium ions may be replaced by a large number of monovalent
cations, including K + , Cs +,Rb ’, Ag , Cu ’, T1 ’, H ’, and NH, + . Sodium may also be
+

replaced by divalent cations in p-alumina but the resulting divalent p-aluminas are bad ionic
conductors. Divalent cations move, however, much faster in P”-alumina, which are thus much
better conductors for those cations. In p”-alumina sodium may even be replaced by trivalent
cations such as those of the lanthanides [23].

Fabrication of /3- and 8”-Alumina Single Crystals and Ceramics

Both compounds can be prepared as single crystals and as polycrystalline ceramic materials.
Polycrystalline material has a lower conductivity than single crystals. Crystals of p-alumina
may be grown with the Czochralski technique by pulling from the eutectic melt (65 mol-To
A1203)above the eutectic temperature of 1572°C [24].
9.2 Electrical Conducting Materials 263

Single crystals of p"-alumina have been grown at 1675"C from a melt of Na20, MgO, and
A120, [25]. Single p"-crystals have also been obtained from fluxes at 1500°C consisting of
mixtures of Na2C0,, NaA102, NaNO,, and LINO, [22].
Raw materials for p- and p"-alumina ceramics may be Na&O, and a-A1203powders which
are ball milled in the required proportions, calcined at 1250"C, pressed into pellets, and
sintered in air between 1550 and 1650°C [26]. Apparently, no j3"-phase is formed on sintering
at 1650°C. In order to prevent sodium loss, the samples may be packed in calcined powder
or in a platinum foil envelope during sintering. The B/p" ratio increases with increasing sinter-
ing temperature, resulting in a decrease in ionic conductivity [27]. In practice, sintering is
typically accomplished between 1550 and 1650"C, followed by annealing between 1400 and
1475°Cto achieve optimum grain size and conductivity. Microstructures of P-and p"-alumina
ceramics are shown in Figure 9-8.
Ceramic tubes of Li-stabilized v-alumina were also produced by sintering in an inductively
coupled plasma both in a static and in a flowing argon atmosphere using a radiofrequency

Figure 9-8.
Scanning electron micrographs
(a) of &alumina ceramic
sintered at 1540°C for 1 h in
argon atmosphere (1 bar) and
@) of commercial Mg-
stabilized Ag-B"-alUmina
ceramic (Ceramatec, Salt Lake
City, USA) (courtesy of
0. Schtif, unpublished).
264 9 Ceramics and Oxides

of 5 MHz at power levels up to about 2 kW [28]. The tubes were prepared from powder mix-
tures consisting either of Al,03, Na,CO,, and LiNO, or of calcined A1,0, and Na2C0, and
-
added Li,O 5A1,0, to yield a composition of 90.4 wt.-Yo A1,0,, 8.85 wt.-Yo Na,O and
0.75 wt.-Yo Li,O. These mixtures were calcined at 1250°C for 2 h. The tubes were prepared
by isostatic pressing at 380 MPa and presintered at 600°C in air. The plasma sintered
specimens exhibited an average Na,O loss of about 10% and low porosities of the order of
1.5%. However, the temperatures in the plasma could not be measured. Also, no electrical con-
ductivity data are reported.
An important application of sodium (3"-alumina ceramics is as solid electrolytes in
sodium/sulfur batteries. However, under the harsh service conditions, the ceramics undergo
degradational changes which may cause catastrophic failure, resulting in destruction of the
battery. Possibilities of avoiding these deleterious effects are microstructural modifications,
including grain size control and transformation toughening by dispersed zirconia particles
1291.

9.2.1.3 NASICON

Goodenough et al. [30) discovered a new sodium solid electrolyte material termed
NASICON with the chemical composition Na,Zr,Si,PO,,. In fact, this compound is a
member of the continuous NASICON solid solution series of the type Na, +,Zr,Si,P, -,O,,
with x ranging from 0 to 3. Maximum conductivity was reported for x = 1.8 - 2.4 [31].
NASICON-type materials have very promising electrical properties (superionic conduction by
Na + and Li +) as solid-state electrolytes for use in both battery and sensor applications.

Structure of NASICON

The rhombohedra1 NASICON structure [32, 331 can be broken down into basic groups of
ZrO, octahedra separated by three (P/Si)O, tetrahedra with which they share corner oxygens
(Figure 9-9). These structural units, ie, 2ZrO, - 3 (P, Si) 0,, are linked by additional corner
sharing, resulting in a three dimensional framework which can be described as infinite rib-
bons. The rather loose arrangement of these ribbons provide conduction channels correspon-
ding to infinite zig-zag lines. The N a + ions (four possible positions per formula unit) are
distributed within the channels on the two different types of sites M(1) and M(2) (in rhom-
bohedral symmetry). The diameters of the tunnels are large enough to permit rapid diffusion
of the Na+ ions. The conductivity of NASICON at 300°C is close to that of (3-alumina
(Figure 9-7).

Fabrication of NASICON Single Crystals and Ceramics

Single crystals of Si/P-NASICON compounds cannot be grown by conventional methods


owing to incongruent melting, but may be prepared by grain growth in sol/gel ceramics. Dur-
ing annealing for long time periods at 1245"C, abnormal growth of NASICON crystals of
length up to 300 pm was observed [34].
9.2 Electrical Conducting Materials 265

I I

E
Figure 9-9. Polyhedra linking in the NASICON-typestructure and detailed environment of M (1) and
M (2) sites [33].
266 9 Ceramics and Oxides

Ceramic NASICON materials consisting of a pure phase cannot be obtained by conven-


tional processing because of the persistence of unreacted ZrO, particles. However,
monophase ceramics may be obtained via the sol/gel route with conductivity characteristics
comparable to those of crystalline NASICON materials [35].

9.2.2 Semiconductors

It is well known that a semiconducting oxide can change its electrical resistance when its
surface comes into contact with a gas. Such a material is thus a potential gas sensor. In an
n-type semiconductor, the concentration of conducting electrons can be decreased by an ox-
idizing reaction or increased by a reducing reaction. In a p-type semiconductor with holes as
major charge carriers, the reverse changes in hole concentration are observed. These
phenomena are the result of the change in concentration of the adsorbed oxygen and/or the
oxygen vacancies on the surface of the semiconductor by the adsorption of the gases, eg, (361.
In the following some important semiconducting oxide ceramic materials used for gas sensing
will be considered.

9.2.2.1 Titania

TiO, crystallizes in the tetragonal rutile structure (Figure 9-10). It is a promising exhaust
gas sensor material. Loss of oxygen on heating, especially under reducing conditions, results

@Ti 0 0
Figure 9-10. Rutile structure showing escape of oxygen ion at low oxygen partial pressure and high
temperature.
9.2 Electrical Conducting Materiak 267

in the generation of oxygen vacancies, balanced by the reduction of some of the Ti4+ ions
to Ti3+, which act as electron donor states. Since the oxygen loss increases with rising
temperature, an increasing number of electrons contribute to the 'conduction process. Over the
temperature range 300-1000"C, pure TiO, is an n-type semiconductor. The mechanism of
electrical conduction and its dependence on temperature and oxygen partial pressure were
discussed by Williams and McGeehin [37] and Williams [38].
The response speed of a TiO, sensor is most influenced by the oxygen mobility in the TiO,
lattice, the rates of chemical reactions on the TiO,/F't catalyst surface and the gas transport
through the open-pore structure of the ceramic sensor pellet [39]. An important requirement
is thus a high porosity of the ceramic Mechanically stable ceramic pellets of TiO, with high
porosity can be prepared by sintering at 900°C from freeze-dried gel prepared by a chemical
precipitation method (Figure 9-11). It is worth noting that this sintering temperature is about
400°C lower than the temperature normally required for sintering of calcined TiO, powder.
The response time of this sensor equipped with surface electrodes is equivalent to that of a
ZrO, sensor at temperatures above 500°C [39].

EFgure 9-11.
Scanning electron micrographs
at two different magnifications
of porous TiO, ceramic
sintered from freeze dried gel
at 900°C (courtesy of
G. Reinhardt, unpublished).
268 9 Ceramics and Oxides

9.2.2.2 Tin Dioxide

SnO, is oxygen-deficient and therefore an n-type semiconductor. It also crystallizes in the


rutile structure (Figure 9-10). Gas sensors using ceramic SnO, materials are of significant
commercial importance for the detection of reducing gases in domestic and industrial en-
vironments (eg, the Tagushi gas sensor). The sensitivity of these sensors may be increased by
adding various dopants and catalysts of precious metals such as Pt and Pd [38,40]. Their prin-
cipal disadvantages are low selectivity, a strong dependence on atmospheric moisture, and the
high temperatures of 300-400°C required for operation. A block diagram of ceramic SnO,
sensor fabrication is shown in Figure 9-12.

r
ICalcinationl

Oxides I Gamma-aluminia I

Pd-impregnated
tin oxides Gamma- aluminia

w
t
ISintered material I Sintered material

I IMixingl
IPulverising)

I Mixed, pulverised
I
-
materials

- Binding material
Paste-making reagent
c Screen printing

Figure 942.
Block diagram of fabrication of a SnO, gas
on substrates
sensor in thick film technology [41].

9.2.2.3 Zinc Oxide

ZnO crystallizes in the hexagonal wurtzite structure. When doped with indium, ZnO
becomes an n-type semiconductor. ZnO semiconducting ceramics are of considerable interest
for the detection of hydrocarbon gases [42]. By surrounding the ZnO ceramic with a Pt layer
as a catalyst, these gas sensors are highly sensitive to isobutane and propane and are therefore
used for the detection of these gases. Using a Pd compound as the catalyst layer, the sensor
9.2 Electrical Conducting Materials 269

is sensitive to CO and H,. Humidity sensors based on ZnO ceramics containing Cr,O, have
also been developed.
Important ZnO-based ceramic semiconductor devices are varistors, which are effectively
“electric field sensors”. These may be regarded as insulators as long as the applied voltage
is below a certain breakdown value. When this value is reached, they exhibit a highly non-
linear current-voltage relation. A 5% change in voltage may result in an increase in current
flow of an order of magnitude. Ceramic ZnO varistors represent an impressive example of the
dependence of electrical behavior both on the microstructure of the device and on specific
processes occuring at the ZnO grain boundaries.
Liquid-phase sintering of pressed pellets of powder mixture of 97 mol-% ZnO and additives
of 1 mol-Yo Bi,03 and 0.5 mol% each of Sb,O,, COO,MnO, and Cr,O, at temperatures bet-
ween 1000 and 1400°C results in a ceramic consisting of semiconducting ZnO particles with
mean diameter of ca. 10 pm, surrounded by a grain boundary layer into which some of the
additives have been segregated [43]. At grain corners, the intergranular layer is several
micrometers thick, but in the region of closest grain-grain contact it is extremely thin, perhaps
only a few nanometers. In addition, electrically insulating “depletion” layers about 100 nm
thick are formed which extend from the surface of the ZnO crystals into their interior. These
grain-grain interfaces control the electrical varistor characteristics according to a model which
associates the highly nonlinear conduction in the breakdown region with electron tunneling
through the depletion layers constituting Schottky barriers to electron flow. At low voltages,
tunneling is improbable but when a critical value is reached, the barriers are abruptly thinned
by hole creation, resulting in an extremely fast increase in current flow via ZnO grain-grain
contacts. The pronounced nonlinearity of the current corresponds to the initiation of the
varistor breakdown.
The electrical properties of the varistor are determined solely by the behavior of the grain-
grain junctions with a microscopic breakdown voltage of about 3.5 V per grain barrier. This
breakdown voltage is rather insensitive to the kind and amount of the oxide additives and the
sintering parameters. The macroscopic varistor breakdown voltage, however, can be varied
either by changing the number of grain-grain junctions using appropriate fabrication
parameters or by changing the varistor thickness. The breakdown range is limited at high cur-
rents by the resistivity of the ZnO grains, resulting in ohmic intragrain conduction as indicated
by an upturn in the current-voltage characteristics. The resistivity of the ZnO grains is com-
monly decreased by doping with Al. Grain resistivities of about 0.1 Qcm are achieved.
However, SIMS depth profiles revealed that A1 tends to segregate at grain boundaries rather
than within the ZnO grains themselves. Hence doping of the grain boundary material also
takes place, resulting in an increase in leakage conduction at voltages below the breakdown
value [43].

9.2.2.4 Perovskite-type oxides

Semiconducting oxides of the perovskite structure family (Figure 9-13) generally have high
potential as gas sensor materials because their electrical properties can be modified to a large
extent by selecting appropriate combinations of the cation constituents and because of their
excellent chemical stability in reducing environments at high temperatures. Porous BaTiO,
ceramics coated with Ag or Ni as catalysts exhibit a significant sensitivity in the high negative
270 9 Ceramics and Oxides

temperature coefficient (NTC)of resistivity region with regard to CO gas. A rate of resistivity
change of more than three decades was observed at 300°C on changing from air to an at-
mosphere containing 1% CO [45].

fY
@ 02- Ti"

Figure 9-13.
Structure of cubic BaTiO, (perovskite
structure) (bottom) and T i - 0 chain in
the BaTiO, crystal (a) above the Curie
temperature (unpolarized) and
(b) below the Curie temperature (spon-
taneously polarized along the z-direc-
tion) [MI. The cubic perovskite struc-
ture has the general chemical formula

0
JY ABO,, where A is a divalent cation
Bo2+ @ 02- Ti" such as Ba2+ and B is a tetravalent
cation such as Ti4+.

SrTiO, ceramic material showed a high sensitivity to oxygen in the temperature range from
550 to 800°C at oxygen partial pressures from 10' to lo4 Pa, whereas ceramic BaTiO, is in-
sensitive to oxygen under the same conditions. The sensitivity may be increased by replacing
some of the Ba'' by Na . Perovskite ceramics based on stannates such as CaSnO,, SrSnO,,
+

and BaSnO, are also promising gas sensor materials [46].


9.3 Insulating Materials 271

Semiconducting BaTiO, ceramics doped with small amounts of donor elements such as
Ln, Sb, Bi, and Nb exhibit a pronounced positive temperature coefficient (F'TC) of electrical
resistivity [47]. The PT'C effect is a grain boundary property associated with the phase transi-
tion at the Curie temperature at 120°C in semiconducting BaTiO, ceramics (see below). The
objective of processing ceramic PTC resistors is the formation of a semiconducting grain
structure interleaved with a very narrow barrier layer system at the grain boundary [48]. The
barrier height determines the impedance to current flow through the ceramic and thus its
resistivity. Below the Curie temperature the barrier height and hence the resistance are low.
Approaching the Curie temperature, the resistivity increases drastically. Changes of more than
eight orders of magnitude may be encountered with the phase transition. The temperature
range for the PTC effect may be extended from about 50 to 300°C by forming solid solutions
of BaTiO, with PbTiO, and SffiO,.
PTC materials are used for a number of applications such as constant-temperature heating
elements, starters for motors, and degaussing girdles for color TV.

9.3 Insulating Materials

9.3.1 Piezoelectric Materials

Piezoelectric crystals develop an electric moment when they are subjected to a stress. The
magnitude of the moment is proportional to the applied stress ; this is the direct piezoelectric
effect. Conversely, application of an electric field in a piezoelectric crystal results in changes
in its shape; this is the converse piezoelectric effect. In both effects the components of electric
moment and of stress are related through the piezoelectric moduli which form a third-rank
tensor. Crystals belonging to the centrosymmetric crystal classes and also to one specific cubic
class cannot be piezoelectric. Therefore, piezoelectric crystals are restricted to 20 of the total
32 crystal classes. Comprehensive treatments of the theory of piezoelectricity of crystals were
given by Mason [49] and Haussiihl [50] and of crystals and ceramics by Berlincourt [51].
Quartz is the most important piezoelectric single crystal material owing to its excellent
mechanical properties. Other classic piezoelectric crystal materials are Rochelle salt
(NaKC,H,O, . 4H@), KDP (KH,PO,) and ADP (NH,H,PO,). A large number of other
piezoelectric crytalline materials were accumulated by Mason [49].

9.3.1.1 Perovskite-Type-Materials

The discovery of the piezoelectric behavior of polycrystalline BaTiO, ceramic in 1947 was
the beginning of increasing interest in piezoelectric ceramics. The perovskite structure of
BaTiO, (Figure 9-13) is cubic above the Curie temperature of 120°C. On cooling below this
temperature, spontaneous polarization takes place in the Ti - 0 chain system, whereby all its
Ti4+ ions are shifted towards the 02-ions and a permanent dipole moment is induced
owing to the displacement of the positive and negative charge centers (Figure 9-13). This
uniformly polarized region is ferroelectric and the symmetry of the structure is now
272 9 Ceramics and Oxides

tetragonal. Below 5 "C, a second phase transition takes place in which a second chain system
is permanently polarized and the structure becomes orthorhombic. A further distortion of the
structure occurs at - 80 "C. These transitions are reversible.
The polarization generally does not take place uniformly over the whole crystal but different
polarization directions exist in different crystal domains. Another interesting feature with
regard to the application of BaTiO, as a capacitor dielectric is the nonlinear temperature
dependence of its dielectric constant, which reaches a maximum value at the Curie
temperature at which the spontaneous polarization disappears.
A BaTiO, ceramic is composed of a multitude of randomly oriented, tiny piezoelectric
crystallites which themselves in the most cases consist of a large number of domains with dif-
ferent polarization directions. Such a ceramic exhibits as a whole no piezoelectric effect.
However, it can be polarized fairly uniformly by either orienting the crystallites during fabrica-
tion with an appropriate process such as the application of a large mechanical stress or more
conveniently by applying a DC poling field of several thousand volts across the ceramic as it
is cooled through the Curie temperature. The orientation of the field determines the orienta-
tion of the mechanical and electrical axes in the ceramic. The poling field can be applied so
that the ceramic exhibits piezoelectric responses in various directions or combination of direc-
tions. The piezoelectric properties of BaTiO, ceramics may be changed by variations in the
chemical composition. For example, on substituting Ba2+ by Pb2+ or Sr2+, the Curie point
increases or decreases.
Jaffe et al. [52] discovered a strong piezoelectric effect in compounds of the solid solution
system PbZrO, - PbTiO, (PZT), which also belongs to the family of perovskite structures.

500 -

/
-
-
0
0
300 - A,
f
E a
-8-

;200-
I-
FT
(tetragonall

100 -

20 LO 60 80 1(
PbZrO, Mole % PbTiO, PbTiO,
Figure 9-14. Sub-solidus phase diagram of the PbZr0,-PbTiO, (PZT) solid solution 151.
9.3 Insulating Materials 273

Its phase diagram is shown in Figure 9-14. In this system rhombohedra1 or tetragonal fer-
roelectric phases exist. The Curie temperature increases from 230°C for PbZrO, to 490°C for
PbTiO,. The piezoelectric properties of these compounds depend mainly on the Ti/Zr ratio.
Favored compositions are situated close to the morphotropic boundary between the rhom-
bohedra] and tetrahedral phase (Ti/Zr = 1 :1) or lie in the region of the tetragonal phase field
(Ti/Zr = 1.2) (morphotropy designates a structural change due to chemical substitution).
The PZT family represents the basis of most ceramic piezoelectric transducers. In order to
tailor properties for specific device demands and to optimize ceramic manufacturing pro-
cesses, a large number of compositional variations of the PZT materials have been made.
These chemical modifications encompass partial replacement of Pb *+ by other divalent ca-
tions and of Ti4+ and Zr4+ by tetravalent cations. Divalent substitutes are Ba, Ca, Sr, and
Cd ; the tetravalent substitute is essentially Sn. Further variations have been accomplished by
doping with elements from nearly the whole Periodic Table, as revealed by the patent literature
(531.
Incorporation of lanthanum into the PZT system yields important ceramic materials ex-
hibiting both piezoelectric behavior and electrooptic effects which result from the change in
refractive index produced by an external electric field (eg, [50, 541). Both the linear and the
quadratic electrooptic effect (linear and quadratic relationships between effective bire-
fringence and electric field, respectively) are used. Lanthanum can replace lead cations in
the A position of the perovskite structure (Figure 9-13) to an appreciable extent. In this
lanthanum-modified PZT system (PLZT), all compositions may be represented by
Pb, -,LaJZr,Ti, -J 1 - ~ / 4 0 3 . Since La3+ replaces Pb2+,electroneutrality is maintained by the
creation of vacant lattice sites, probably at both the A2+ and B4+ positions. The addition of
lanthanum has profound effects, among others maintaining an extensive solid solution
throughout the PZT system and a decrease in the stability of the ferroelectric phases, ie, lower-
ing the Curie temperatures.
The solubility of lanthanum in the PZT structure is directly related to the amount of
PbTiO, present. In the two end-members PbZrO, and PbTiO, , the solubility limits are 4 and
32 at - Vo La, respectively. For intermediate compositions, the limits are proportional to their
Ti/Zr ratios. Ceramics of the PLZT system used for piezoelectric applications are usually con-
fined to compositions containing less than 5% lanthanum, whereas the electrooptic ceramics
contain 6% lanthanum or more. The majority of compositions are located at or near the mor-
photropic phase boundary, since most properties are optimum along it.
PZT ceramics are generally manufactured from mixed-oxide powders. Sintering may be ac-
complished at normal pressure at temperatures between 1200 and 1300°C. Since PbO has a
high partial pressure, a refractory enclosure with excess PbO in the atmosphere is used to avoid
losses. The porosity of the resulting ceramic is less than 5% and the average grain size is about
20 pm. It is well known, however, that bulk density and grain size considerably influence the
electrical characteristics of sintered piezoelectric ceramics. These are therfore preferentially
manufactured by the hot-pressing technique which also allows control of the grain size. By
this technique completely dense PZT ceramics can be obtained which have superior piezoelec-
tric behavior compared with conventionally sintered ceramics [55].
For PLZT electrooptic ceramics, excellent transparency, ie, complete elimination of pores
and flaws, is an inevitable requirement. This is achieved by hot-pressing at 1250°C at a
pressure of 14 MPa for 18 h in an oxygen-rich atmosphere. Also, the sinter plus HIP treatment
of ceramics without encapsulation has been succesfully used [15]. Additional important re-
274 9 Ceramics and Oxides

quirements are a small, homogeneous grain size to achieve a uniform refractive index within
the ceramic, preventing any second phase which can lead to light scattering, and controlling
the grain size for optimization of the electrooptic properties. These requirements are difficult
to achieve by the mixed-oxide method and various coprecipitation processes have been
employed sucessfully for the fabrication of high quality PLZT electrooptic ceramics.
Piezoelectric ceramics are used in a large number of applications such as capacitors,
resonators, electroacoustic transformers, pressure sensors, and high-voltage generators.
Piezoelectric actuators based on PZT materials have received increasing attention for applica-
tions such as ultrasonic motors, deformable mirrors for control of the phase of light, and
ultra-precision guide mechanisms in mechanical systems [54, 561. The electrooptical effects in
PLZT ceramics are used in a large number of applications such as optical shutters,
modulators, color filters, displays, image storage, holographic recording, and optical
waveguides 1541.

9.3.2 Pyroelectric Materials

Pyroelectric crystals have the property of being permanently polarized within a given
temperature range. They develop the electric polarization spontaneously and form permanent
dipoles within their structure. This polarization changes when the temperature is changed.
This is the pyroelectric effect, which can exist only in non-centrosymmetric dielectric crystals
possessing a unique polar axis along which a spontaneous polarization exists. Hence only
crystals belonging to the ten so-called polar crystal classes may theoretically show pyroelec-
tricity (501. The length of the polar axis, ie, the dipole moment, varies with temperature and
changes sign on either increasing or decreasing the temperature. If a crystal has electroded
faces perpendicular to its polar axis, a temperature change produces an electrical signal pro-
portional to the rate of temperature change.
Materials with good pyroelectric properties are LiTaO, , SrBaNbO, , PbTiO, , lead ger-
manate, PZT-like ferroelectric ceramics, triglyzine sulfate (TGS)and polyvinylidine flouride
(PVDF) [57]. Most of these materials are also ferroelectric.
Single crystals of TGS have excellent pyroelectric properties. However, their applications are
limited to high-grade pyroelectric sensor devices because of water solubility and the low Curie
temperature of 49 "C. LiTaO, single crystals have good pyroelectric properties and high
stability over the temperature range -20 to 100°C.
One of the most important applications of pyroelectric materials is as infrared radiation
detectors, in which the temperature change is brought about by the absorption of time-varying
radiation. These are highly sensitive transducers which can be operated advantageously at
room temperature. These properties have prompted a number of applications such as fire
prevention, cooking and intruder alarms, thermal imaging systems, and geographical map-
ping.
Pyroelectric oxide ceramics are promising materials with a number of advantageous proper-
ties compared with single-crystal material such as chemical inertness, preparation in large
sizes, and, most important, the large range of solid solution. A number of pyroelectric ceramic
systems have been studied, many of them based on solid solutions of the system PZT and
PbZrO, - Pb,FeNbO, - PbTiO, doped with uranium [58]. These ceramics may be produced
from relevant oxide raw materials by the mixed-oxide technique using repeated milling and
9.4 Negative Temperature Coefficient (NTC) Ceramic Thermistors 215

calcination at 800 or 900°C and final hot pressing at about 1200°C. These ceramics were
found to be very suitable materials for pyroelectric infrared radiation detectors [%I. Introduc-
tion of Ca into PbTiO, ceramics improves markedly the pyroelectric properties. The material
shows a high pyroelectric response at chopping frequencies lower than 1 Hz. This material is
thus well suited for human body sensing, which requires a high response in the frequency
range below 1 Hz [42].

9.3.3 Ferroelectric Materials

Ferroelectric crystals also show spontaneous polarization like pyroelectric crystals but have
the additional property that the polarization can be reversed by applying a sufficiently large
electric field. In a strong alternating field a hysteresis in the induced change in polarization
is observed and the most important characteristic of a ferroelectric material is its hysteresis
loop. It describes the nonlinear polarization switching behavior as a function of the electric
field. Most ferroelectric crystals exhibit a Curie temperature above which they are non-polar.
Ferroelectrics are principally piezoelectric in the polar state, but a piezoelectric material need
not implicitly also be ferroelectric. The theory of ferroelectricity was treated in detail by
Mason [49].
The classical ferroelectric single crystal materials are Rochelle salt, KDP and BaTiO,, but
the number of actual ferroelectric materials including ceramics is of the order of thousands
[54]. Important ferroelectric ceramics are based on BaTiO, and PZT solid solutions.
Most ferroelectric materials are used in capacitor and transducer components. Ferroelectric
single crystals are frequently used for electrooptic applications such as optical shutters.
Strong internal space charge fields can be generated within ferroelectric crystals by the
movement and retrapping of carriers excited by photon energy produced by doping [48]. These
fields affect the refractive index and thus produce photo-ferroelectric effects which may be
used, for example, for the optical storage of holograms. However, there are few switching ap-
plications which depend directly on the ferroelectric polarization reversal at present owing to
instability and degradation of the properties in the ferroelectric ceramic materials during
operation.

9.4 Negative Temperature Coefficient (NTC) Ceramic Thermistors

Ceramic NTC thermistors, in contrast to FTC materials, exhibit a decrease in resistance with
increasing temperature. As described in Section 9.2.2.4, the PTC effect is essentially a grain
boundary property related to semiconducting barrier layer systems. On the other hand, the
NTC effect is a volume property. Depending on the nature of the dominating charge carriers,
NTC ceramic thermistors can be classified into three groups : p-type semiconductors, n-type
semiconductors, and ionic conductors [59]. The resistivity of these thermistors is inversely pro-
portional to the product of concentration and mobility of the dominating charge carriers and
is therefore affected by the microstructure of the ceramic, ie, by grain size, grain boundary
characteristics, and voids. The potential barrier formed at the grain boundaries by segregation
276 9 Ceramics and Oxides

and precipitation of impurities and additives affect to a large extent both the total resistivity
and its temperature dependence. Control of microstructure and avoidance of impurities dur-
ing manufacturing are therefore most important for obtaining thermistors with reproducible
NTC properties. Resistance-temperature characteristics of various NTC ceramic thermistors
are shown in Figure 9-15.
The NTC thermistors are cheap and fairly sensitive temperature sensors fabricated in
various shapes such as pellets, beads, disks, rods, and thick films. They are used for thermal
sensing in a large number of applications covering the temperature range from below 0 to
above 1OOO"C. An important application is medical temperature monitoring. This application
requires small sensor devices for fast thermal response, a large temperature dependence of the
resistance, a high resistance reproducibility during fabrication, chemical and mechanical
stability, and, as a fundamental requirement, biocompatibility. Since the medical application
area promises to become a profitable market, considerable research and development*efforts
are currently being made by the thermistor industry in oder to meet the required demands.

Figure 9-15.
Resistance-temperature characteristics
of various ceramic NTC thermistor
materials (TMO = transition metal
oxides) 1591.

9.5 Conclusions

In this review, only some of the basic aspects related to ceramic oxide materials for sensor
applications could be covered. However, several general conclusions can be drawn concerning
future developments in this field.
9.6 References 277

In the technological area, high-purity and well characterized powders prepared by non-con-
ventional processes will in most cases replace the calcined oxide powders. Most of the
manufacturing processes for ceramics will be automated with the use of programmable elec-
tronic controllers. Processing parameters and the control of microstructure will be optimized
by computer modelling of expert systems. New, sophisticated sintering technologies such as
hot isostatic pressing (HIP) and high frequency plasma sintering will become familiar techni-
ques and ceramics fabrication in a clean-room environment will become common practice.
Miniaturization of the ceramic components for integration into electronic devices will be
developed. These goals will probably be approached within a rather short time.
In the scientific field, much interdisciplinary effort has to be expended to elucidate the com-
plex relationships between microstructure and properties of sensor ceramics. The chemical
and structural characterization of the grains and particularly of the grain boundary phases
is of eminent importance, especially in semiconductor sensor ceramics. High-resolution
analytical TEM and sophisticated methods for surface and depth profile analysis are the most
promising approaches. Possibilities of controlled manipulation of grain boundaries and the
effects of grain size and shape on the properties of interest of the ceramics must be explored.
In addition, the knowledge of phase equilibria is currently incomplete in many important
ceramic systems. Detailed studies, especially at high temperatures and in many cases at high
pressures are necessary.
The basic mechanisms responsible for specific sensor properties in many ceramic sensor
materials are not well understood at present. This represents a serious obstacle both for the
development of new sensor materials and for the improvement of existing ceramics. An
understanding of these mechanisms is a fundamental requirement for the successful future
development of ceramics for sensor applications.

9.6 References

[l] Fisher, G., Ceram. Bull. 63 (1984) 569.


[2] Kingery, W. D.,Bowen, K. H., Uhlmann, D. R., Introduction to Cerumics, New York : Wiley 1976.
[3] Fisher, G., Ceram. Bull. 65 (1986) 1355.
[4] Yoshimura, M., Cerum. Bull. 67 (1988) 1950.
[5] Stevens, R., Zirconia and Zirconia Ceramics, Magnesium Elektron Ltd., 1986.
161 Fisher, G., Ceram. Bull. 65 (1986) 622.
[7] Nassau, K., Lapidary Journal35 (1981) 1194.
[8] Johnson, D. W., Ceram. Bull. 60 (1981) 221.
[9] Van de Graf, M. A., Burggraaf, A. J., “Science and Technology of Zirconia 11” in: Advances in
Ceramics Vol. 12, Claussen, N., Riihle, M., Heuer, A. H. (eds); Columbus, OH: The American
Ceramics Society, 1983, p 744.
[lo] Rice, R. W., J. Phys. Chem. Solids 45 (1984) 1033.
[ll] Evans, A. G., Heuer, A. H., J. Am. Ceram. SOC.63 (1980) 241.
[I21 Petzow, G., Huppmann, W. J., Z. Metallkunde 67 (1976) 579.
[13] Mountvala, A. J., J. Am. Ceram. SOC.42 (1963) 120.
[14] Oberacker, R., Kiihne, A., Thiimmler, F., Powder Metall. Int. 19 (1987) 43.
[IS] Hardtl, K. H., Ceram. Bull. 54 (1975) 201.
[16] Rankin, G. A., Merwin, H. E., J. Amer. Chem. SOC.38 (1916) 568.
[17] Yamaguchi, G., 1 Japan Electrochemical SOC. ll (1943) 260.
[I81 Bettman, M., Turner, L. L., Inorgan. Chem. 10 (1971) 1442.
278 9 Ceramics and Oxides

[19] Beevers, C. A., Ross, M. A. S., Z. Kristullogruphie 97 (1937) 59.


[20] Boilot, J. P., Collin, G., Colomban, P., Comes, Phys. Rev. B. 22 (1980) 5912.
[21] Hodge, J. D., 1 Am. Cerum. SOC. 66 (1983) 166.
[22] Dunn, B., Schwarz, B. B., Thomas, J. O., Morgan, P. E. D., in: Solid State Ionics-87, Weppner, W.,
Schulz, H. (eds.); Amsterdam: North Holland, 1988, p. 301.
[23] Farrington, G. C., Sens Actuutors 1 (1981) 329.
[24] Baughman, R. J., Lefever, R. A., Mat. Res. Bull. 10 (1975) 607.
[25] Farrington, G. C., Briant, J. L., Science 204 (1979) 1371.
[26] Song, H. I., Kim, E. S., Yoon, K. H., Physicu B 150 (1988) 148.
[27] Stevens, R., Binner, J. G. P., 1 Mat. Sci. 19 (1984) 695.
[28] Johnson, D. L., Rizzo, R. A., Cerum. Bull. 59 (1980) 467.
[29] De Jonghe, L. C., Ceram. Bull. 65 (1986) 1158.
[30] Goodenough, 3. B., Hong, H. Y. P., Kafalas, J. A., Mat. Res. Bull. 11 (1976) 203.
[31] Hong, H. Y. P., Mat. Res. Bull. 11 (1976) 173.
[32] Collin, G., Comes, R., Biolot, J.-P., Colomban, P., in: Solid State Ionics-87, Weppner, W.,
Schulz, H. (eds.); Amsterdam: North Holland, 1988, p. 427.
[33] Delmas, C., Nadiri, A., Soubeyroux, J. L., in: Solid State Zonics-87, Weppner, W., Schulz, H.
(eds.); Amsterdam: North Holland, 1988, p. 419.
[34] Colomban, P., Solid State Ionics 25 (1986) 97.
[35] Boilot, J. P., Colomban, P., Collin, G., in : Solid State Ionics-87, Weppner, W., Schulz, H. (eds.) ;
Amsterdam: North Holland, 1988, p. 403.
[36] Oyabu, T., A Appl. Phys. 53 (1982) 2785.
[37] Williams, D. E., McGeehin, P., in: Electrochemistry Vol. 9, Pletcher, D. (ed.); London: Royal
Society of Chemistry, 1984, p. 246.
[38] Williams, D. E., “Conduction and Gas Response of Semiconductor Gas Sensors”, in: Solid State
Gas Sensors, Moseley, P. T., Tofield, B. C. (eds.); Bristol: Adam Hilger, 1987, p. 71.
[39] Micheli, A. L., Cerum. Bull. 63 (1983) 694.
[40] Norris, J. 0. W., in: Solid State Gus Sensors, Moseley, P. T., Tofield, B. C. (eds.); Bristol: Adam
Hilger, 1987, p. 124.
[41] Tofield, B. C., in: Solid State Gas Sensors, Moseley, P. T., Tofield, B. C. (eds); Bristol : Adam Hilger,
1987, p. 198.
[42] Ichinose, N., Ceram. Bull. 64 (1985) 1581.
(431 Levinson, L. M., and Philipp, H. P., Ceram. Bull. 65 (1986) 639.
(441 Ricke, W., Handbuch der Keramik, Freiburg: Verlag Schmidt, 1973.
[45] Kuwabara, M., Ide, T., Ceram. Bull. 66 (1987) 1401.
[46] Shimizu, Y., Fukuyama, Y., Arai, H., Seiyama, T., “Fundamentals and Applications of Chemical
Sensors”, in: ACS Symposium Series Vol. 309, Schuetzle, D., Hammerle, R. (eds.); Washington,
DC: American Chemical Society, 1986, p. 83.
[47] Kuwabara, M., Yanagida, H., in: Fine Ceramics, Saito, S. (ed.); New York : Elsevier Applied Science
Publishers, 1985, p. 286.
[48] Cross, L. E., Cemm. Bull. 63 (1984) 586.
[49] Mason,. W. P., Piezoelectric Crystals and Their Application to Ultrasonics, New York : Van
Nostrand, 1950.
[50] Haussiihi, S., Kristullphysik, Weinheim : Verlag Chemie, 1983.
[51] Berlincourt, D., in: Ultrasonic 7jansducerMaterials, Mattiat, 0. E. (ed.); New York, Plenum Press,
1971.
[52] Jaffe, B., Roth, R. S., Marzullo, S., 1 Appl. Phys. 25 (1954) 809.
[53] Stark, K., Handbuch der Keramik, Freiburg: Verlag Schmidt GmbH, 1971.
[54] Haertling, G. H., in: Ceramic Materials for Electronics, Buchanan, R. C. (ed.); New York: Marcel
Dekker, 1986, p. 139.
[55] Patel, N. D., Nicholson, P. S., Ceram. Bull. 65 (1986) 783.
[56] Uchino, K., Ceram. Bull. 65 (1986) 647.
[57] Liu, S. T., Ferroelectrics 10 (1975) 83.
[58] Whatmore, R. W., Ferroelectrics 49 (1983) 201.
[59] Koumoto, K., Yanagida, H., in: Fine cerumics, Saito, S . (ed.) ; New York : Elsevier Applied Science
Publishers, 1985, p. 281.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

10 Signal Processing
Universitilt der Bundeswehr Miinchen. Neubiberg.
HANS-ROLF~UNKLER.
FRG

Contents
10.1 Smart Sensors in Microelectronic Systems .............. 280

10.2 Characteristics of Sensors and Signals . . . . . . . . . . . . . . . 281

10.3 Structural Limits of Analog Signal Processing . . . . . . . . . . . . 282


10.3.1 Serial Structure . . . . . . . . . . . . . . . . . . . . . . . . . . 282
10.3.2 Parallel Structure . . . . . . . . . . . . . . . . . . . . . . . . . 284
10.3.3 Loop Structure . . . . . . . . . . . . . . . . . . . . . . . . . . 286

10.4 Sensor Oscillators and Frequency-to-Digital Conversion . . . . . . . . . 288


10.4.1 Relaxation Oscillator . . . . . . . . . . . . . . . . . . . . . . . 288
10.4.2 Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . . . 289
10.4.3 Ring Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . 290
10.4.4 Frequency-to-Digital Conversion . . . . . . . . . . . . . . . . . . . 291

10.5 Physical Model Functions . . . . . . . . . . . . . . . . . . . . . 295


10.5.1 'Ifansistor Thermometer . . . . . . . . . . . . . . . . . . . . . . 295
10.5.2 Inductive Displacement Sensor . . . . . . . . . . . . . . . . . . . 2%
10.5.3 Pressure Sensor Using the Hall Effect . . . . . . . . . . . . . . . . 297

10.6 Digital Modeling of Static Sensor Characteristics with Mathematical


Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
10.6.1 Look-up Table Method . . . . . . . . . . . . . . . . . . . . . . 299
10.6.2 Polygon Interpolation . . . . . . . . . . . . . . . . . . . . . . . 299
10.6.3 Polynomial Interpolation . . . . . . . . . . . . . . . . . . . . . . 300
10.6.4 Cubic Spline Interpolation . . . . . . . . . . . . . . . . . . . . . 301

10.7 Minimum Principles for Approximations ............... 302


10.7.1 Example of Regression Method . . . . ............... 303

10.8 Correction of Cross Sensitivities . . . . . . . . . . . . . . . . . . 304

10.9 Dynamic Correction of Sensors . . . . . . . . . . . . . . . . . . . 307

10.10 Examples of Sensor Signal Processing . . . . . . . . . . . . . . . . 308


10.10.1 Inductive Sensor System for Mechanical Quantities . . . . . . . . . . 308
10.10.2 Gas Concentration Measuring System . . . . . . . . . . . . . . . . 309

10.11 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311


280 10 Signal Processing

10.1 Smart Sensors in Microelectronic Systems

Sensors with dedicated microcomputers (“smart sensors”) offer the possibility of realizing
high-quality, low-cost microelectronic systems. Sensor-specific methods of signal processing
allow the use of simpler sensors without trimming elements, improving the sensor quality with
only little additional expenditure. Therefore, the reproducibility of both the final output and
the reliability of the sensor is improved. The microcomputer can be utilized to measure and
correct the influence of interfering parameters on the sensor.
Microelectronic systems are needed, for example, to control decentralized processes such as
car engines or heating systems, which up to now had to work without adequate control
facilities.
Although in principle it may be possible to realize sensors with ideal properties by
technological means, this would most often be too expensive. Therefore, sensor-specific signal
processing is added to the technological measures. Figure 10-1 shows the structure of a
microelectronic system with integrated smart sensors. Each sensor consists of a sensor element
(se), an analog signal-processing unit (asp), an analog-digital (A/D) or a frequency-to-digital
(F/D) converter, and a microcomputer (pC). A digital bus system allows the communication
of the individual units. The signal-processing features have to be designed in a way that
minimizes the data transfer on the bus.

technology

specific
signal
processing

specific
signal
processing

technology
I
I man I
L_----------------
I
2

Figure 10-1. Structure of a future microelectronic system with smart components.

In this context, a smart sensor will be characterized by the possibility of signal processing
in combination with one or more sensor elements. In general, sensor signal processing can be
divided into analog, hybrid, and digital signal processing.
10.2 Characteristics of Sensors and Signals 281

10.2 Characteristics of Sensors and Signals

Some of the most important features of sensors are as follows (see also Chapter 2 of this
volume):
- static characteristics,
- sensitivity to influence effects and environmental conditions,
- dynamic characteristics,
- reliability and economy.
Regarding the static transmission characteristics of a sensor, its sensitivity and permissible
error limits are of special interest. A sensitivity which is too small can produce additional er-
rors because of the amplification required. The total error of a sensor should be as small as
possible if, eg, exact temperature or position control is to be effected.
Moreover, sensors should show negligible reactions to interferences and parasitic effects. An
interference such as temperature can be eliminated either by holding it at a constant value or
by correcting it in the processing unit. Parasitic effects, for instance, are mechanical vibrations
or oscillations, and electromagnetic fields which can cause disturbances to various systems.
In addition, in most cases limiting values of the environmental conditions have to be main-
tained to ensure reliable operation of a sensor. In general, maximum limits for the tolerable

Table 10-1. Comparison of the features of signal forms.

Amplitude analog Frequency analog Direct digital

Possible
static Restricted Unlimited Unlimited
accuracy
Dynamic In general very good Limited by conversion Limited by sampling
characteristics speed rate and transmission
speed
Noise immunity on Small Good at FM transmis- Good at PCM (pulse
signal transmission sion code modulation)
transmission
Possible arithmetic Limited or Division and integra- Any operation with
operations expensive tion easy feasible arithmetic processors
Connection to digital By analog-digital Eg, by frequency Direct at suitable code
computers conversion counter and signal level
Error Improvement only by On frequency counting With error-correcting
correction analog averaging natural redundancy codes
Galvanic signal separa- Very expensive Easy Easy
tion (isolation) (modulator) (transformer, (optocoupler)
optocoupler)
Human Trends easy to Acoustic signals Bad, but highest possi-
adaptation recognize possible ble resolutions
282 10 Signal Processing

mechanical and thermal stress are set, eg, a maximum acceleration value or certain tempera-
ture ranges.
Sometimes applications depend on the dynamic characteristics of sensors. The achievable
accuracy of resistance thermometers, eg, is not relevant if the dynamic characteristic of ther-
mocouples is needed for rapid temperature control. In principle, dynamic correction can be
realized with a processing circuit, but there the efficiency achievable is limited because the real
dynamic parameters are not known with sufficient accuracy.
The importance of the reliability and economy of sensors needs no further explanation.
Which kind of sensor signal should be used depends inter alia on the necessary features for
signal transmission and on the kind and size of signal processing applied. Important signal
forms can be specified as follows:
- amplitude analog,
- frequency analog (FM = frequency modulation and PDM = pulse duration modulation),
- direct digital.
The main characteristics of these signal forms are listed in Table 10-1.
For some arithmetic operations, eg, division or integration, frequency analog signals are
well suited. Also, they can be converted easily into digital signals. Today many such sensors
already exist. In addition, the importance of frequency analog signals and sensors is increasing
because in most cases their design is easier than that of comparable amplitude analog sensors.
To gain higher accuracy, the expenditure necessary with amplitude analog sensors increases
faster than that with frequency analog or direct digital sensors.

10.3 Structural Limits of Analog Signal Processing

Sensor signals are affected by non-linearities and interfering quantities, eg, temperature. In
general, the achievable characteristics of measuring components depend on the structure and
complexity of the individual devices. Mostly the following structures are used for analog com-
pensation:
- serial structure,
- parallel structure,
- loop structure.
The serial structure is the commonest, whereas the loop structure is only used when it is im-
possible to achieve the same result by other means.

10.3.1 Serial Structure

In a serial structure such as that shown in Figure 10-2, the output signal yi of one device
is the input signal x i + , of the following device. If all units have linear transmission
characteristics according to

y 1. = kI . xa. i = 1, 2, ... n , (10-1)


10.3 Structural Limits of Analog Signal Processing 283

Figure 10-2.
Serial structure of measuring -- -
device.

the resulting sensitivity, k,,, of the whole structure can be written as

k,, = k, * k, . .. . k, . (10-2)

In this way, sequences can be realized, leading in several stages from the non-electrical input
signal of a sensor element to the output of a display unit. Between input and output,
amplification and processing of the signals can be carried out. Linearization of a non-linear
sensor characteristic can also be performed by a subsequent unit which has the inverse
transmission characteristic of the sensor.
An application of a serial structure, for instance, is a piezoresistive pressure sensor with an
analog signal processing unit, realized in a multi-layer hybrid technique [l]. In general, the
analog signal processing involves components for amplification, offset correction, and com-
pensation of temperature effects on the offset and sensitivity of the sensor. Figure 10-3 shows
the processing unit for the aforementioned pressure sensor. The sensor element is followed by
a resistance network which affects the temperature compensation.

sensor-element temperature I amplification


compensation I

Figure 10-3. Analog signal processing unit for a piezoresistive pressure sensor (according to [l]).

Ideally, the correction of the temperature effects on offset and sensitivity should be ac-
complished independently of each other, but in practice this is very difficult to achieve and
depends on the circuit. Here the offset temperature coefficient is corrected by trimming the
resistors R, and R,, while temperature influences to the pressure sensitivity of the sensor are
284 10 Signal Processing

compensated by the temperature-independent resistor R,. The necessary values of the


resistors are calculated for each individual sensor according to the calibration data at different
temperatures and pressures. As in Figure 10-4, the temperature dependence of the uncompen-
sated sensor is about -2.0% per 10 K, whereas the improvement reached with the compensa-
tion circuit is approximately 0.5% per 10 K within the temperature range 10-90°C.
Finally, the temperature-corrected sensor signal is amplified by a symmetrical amplifier
stage with an amplification factor of up to 50.
Temperature compensation in a serial structure is also achieved by an industrially designed
integrated circuit (IC) for signal processing [2]. It offers the possibility of separate adjustment
of sensitivity and offset and of compensating temperature effects. In addition, this circuit is
not designed only for one special sensor element, but can also be used with different sensors
using strain gage elements. The IC is completely analog, but shows good compatibility with
microcomputers.

dUp/Up (25OC’I
in %

75 F
3 in
‘%a OC

Figure 104. Output of the pressure sensor (according to [l]): (a) without temperature compensation;
(b) with compensation unit.

10.3.2 Parallel Structure

The most significant characteristic of a parallel structure is shown in Figure 10-5. One input
signal x is sent to two or more measuring components. The resulting output signal, yres,is
10.3 Structural Limits of Analog Signal Processing 285

j _ x m y " j
Figure 10-5. Parallel structure.

obtained by arithmetic operations - usually addition or subtraction - between the single


outputs yi. Adding the signals yi leads to

Yres = Y , + Yz + + Yn . (10-3)

Assuming a transmission characteristic yi = ki . x of the single unit we obtain a total sen-


sitivity k,, of

kEs = k, + kz + . .. + k,, (10-4)

The most important form of the parallel structure is the so-called differential structure,
which gives the possibility of simple linearization and elimination of quantities that interfere
with the sensor characteristics. nhro identical displacement sensors, for example, can be posi-
tioned in such a way that a displacement x produces opposite responses of the sensors. In-
terfering quantities such as temperature, however, will cause reactions of both sensors in the
same direction. The dependence of the output yi of each measuring device at a certain
operating point xo can be expressed as a function of the generalized input t and the
temperature 19:

(10-5)

With the two sensor drive signals tl = xo - x and t2= xo + x, we obtain the following out-
put signals y , and y 2 :

(10-6)
(10-7)

The signal difference yns is then

With the above assumptions, it is exactly linear and independent of the temperature 19 (see
Figure 10-6).
286 10 Signal Processing

Figure 10-6.
Linearization with differential structure.

The total sensitivity, k,,, is constant, and its abolute value is double the individual sen-
sitivity ki of each unit at the operating point t = xo:

(10-9)

6yges
= 2 (a, + 2u2x0). (10-10)
kws = (7) x=o

Generally there is a point of inflection in the neighborhood of the operating point and
therefore local linearization of the resulting characteristic. Simultaneously the cross sen-
sitivities working in the same direction are reduced.
The differential structure can also be used in all cases, in which a sensor element can be
affected by the measurand in the opposite direction. This is possible, eg, for displacement,
force, or strain sensors. Using a bridge circuit, oppositely directed strain gauges can double
the measuring effect and severely reduce the unidirectionally acting temperature effect. Here
the linearization effect is only of second order, because the driving strains and the resulting
resistance changes are very small, usually less than 1%.

10.3.3 Loop Structure

The loop structure is realized as a negative feedback circuit as shown in Figure 10-7. Thereby
the output x, of the unit in the feedback path counteracts the measurand x and will be varied
until it is nearly equal to it. This means an inversion from the viewpoint of measurement
10.3 Structural Limits of Analog Signal Processing 287

Figure 10-7. Loop structure. -EJ-


engineering. Instead of the actual value of the measurand, the corresponding feedback quan-
tity is now measured.
If both transmission factors v and rare constant, the transfer function of the negative feed-
back circuit,
y(x) = v(x - xk) = v(x - r - y ) (10-11)

leads to the total transmission factor y / x :

Y -
_ V
(10-12)
x l + r . v '
At very large transmission factors, v + l/r, of the unit in the forward path, we obtain as the
total transmission factor y / x the reciprocal factor of the feedback path: y/x = l/r, which is
independent of v.
An example of a loop structure using the compensation principle is the high-precision
micromachined quartz accelerometer [3] shown in Figure 10-8. The seismic mass m of the ac-
celerometer is linked to a crystalline quartz wafer by two beams, which are flexible in the direc-
tion parallel to the wafer. The displacement of the mass due to the acceleration a, is detected
by a capacitor circuit, the signal of which is amplified to drive a current, Zyrthrough the coil
on the seismic mass placed in a magnetic field B,, generating a feedback force Zy . B, which
counteracts the acceleration force m . a, to bring the mass back to the original position,
ZyB, t: ma,. The current Iy = (m/B,) a, can be used to represent the acceleration a,.

- seismic mass
2 - flexible beams
3 - capacitor circuit
Figure 10-8. Closed-loop accelerometer (according to [3]). 4 - coil
288 10 Signal Processing

10.4 Sensor Oscillators and Frequency-to-Digital Conversion

The conversion of amplitude analog signals to frequency signals is advantageous concerning


signal transmission and digital conversion. The oscillator circuits described below permit the
conversion of the values of electrical components, eg, resistors, to corresponding frequency
signals.

10.4.1 Relaxation Oscillator


The output signals of sensors whose reaction to the measurand results in a change in
resistance or capacitance can easily be converted to frequency signals by a relaxation oscillator.
The simplest relaxation oscillator is realized by using an astable multivibrator as shown in
Figure 10-9. The capacitor C is charged via the series resistor R to the voltage u, (t) given by
the voltage divider and the comparator output voltage U,. When u, (t) reaches the com-
parator switching point at Urn, = Uo. R z / ( R l + R2), the comparator output changes to
- Uo.Now the capacitor will be discharged until u, (t) = Urnin= - Uo R , / ( R , + R 2 ) ,-
then the comparator switches again. Thus a square-topped pulse signal is generated at the
+
comparator output, oscillating between Uoand - U,,at a frequency which depends on the
values of R and C, respectively.

I)%(
While the capacitor is charged, the voltage u, (t) follows the equation

u,(O = [- Rl:R2 + (1 + Rz
R , + R2
) (1 - WP * uo

= [l - (1 +
Rl
Rz
+ Rz
). exp (
31 . uo. (10-13)

la1

---- -
u It1
u,l t 1
t --
Figure 10-9.
Simple relaxation oscillator:
(a) astable multivibrator;
(b) voltage characteristic.
10.4 Sensor Oscillators and Frequency-to-Digital Conversion 289

At the time t = T/2 the voltage u, (t) = Urn, = U, . R , / ( R , + R , ) is reached. The period

I)%(
T = l/f can be expressed as

T = 2 RC . In 1
+ [+ 2 . (10-14)

Setting R , / R , = (e -1)/2 = 0.859, eg, corresponding to a ratio of R , / ( R , + R , ) = 0.462,


we obtain T = 2 . RC or f = 1/2 . RC.
The saturation voltages + U,, and - Uo have an absolute value of about 1 V below the
supply voltage. They have neither the same absolute value nor the same temperature
dependence. Using two serial, oppositely directed Zener diodes (D,and D, in Figure 10-9) at
the comparator output, the saturation voltages and their temperature dependence can be made
equal.

10.4.2 Harmonic Oscillator

Today the use of precision oscillators for the conversion of the output signals of inductive
or capacitive sensors is gaining importance. In contrast to the relaxation oscillator, a harmonic
oscillator generates a sine-wave signal, the frequency of which changes with the measured
value. A modified LC-Franklin oscillator as shown in Figure 10-10 is well suited for the conver-
sion of inductive or capacitive sensor signals.

frd = - 1-R2 -
Figure 10-10.
Modified LC-Franklin
oscillator. A ,
! ---A

The Franklin oscillator produces the necessary phase shift of 180" by a second amplifier
stage, not by passive components. The oscillator frequency will nearly coincide with the reso-
nant frequency of the R-LCcircuit if the amplifiers have a high input resistance and a small
output resistance. Moreover, their input and output capacitances have to be small and in-
dependent of frequency over a wide range. Usually the shift of the operating point entailed
by frequency changes in an oscillator may cause a thermal transient effect, which can last for
several minutes. Frequency independence of the input and output impedances of the Franklin
oscillator prevents this effect.
290 10 Signal Processing

Measurements at a conical coil spring applied as a linear displacement sensor [4] led to the
following results: displacement of 100 mm caused a frequency change from 1.3 to 2.2 MHz,
which is equivalent to a resolution of 9 Hz/pm, the standard deviation of the output signal
was less than 2 Hz over the whole measuring range.
This shows the high quality obtainable with LC oscillators, additionally achieved by the
possibility of fabricating the circuit with SMD (surface-mounted design) technology. The
resolution of this measuring system is not restricted by the oscillator, but by the mechanical
hysteresis of the spring material used, leading to a resolution of about

10.4.3 Ring Oscillator

A ring oscillator circuit is mostly realized in MOS technology. It consists of an odd number
of cascaded inverters or NAND- or NOR-gates (see Figure 10-11). The output of the last gate
is fed back to the input of the first, leading to a ringconnection. The circuit is self-oscillating
if the amplification factor of each stage is larger than 1. The oscillation frequency depends
primarily on the number of gates and their delay times. It also depends on the supply voltage
and the chip temperature.

1st stage 2nd stage 3rd stage


...
* * a 8th stage 9th stage

...

...

Figure 10-ll. Circuit of a nine-stage MOS ring oscillator.

MOS-ring oscillators can be used to design pressure sensors which convert the pressure
directly to a frequency analog output signal [S]. According to the piezoresistive effect, the
channel resistance of a MOS transistor changes under mechanical stress. This affects the gate
delay time of each oscillator stage and therefore causes a change in the oscillator frequency.
Figure 10-12 shows the dependence between the mechanical stress and the frequency change
of a MOS ring oscillator, which is linear in a certain range. The sensitivity of the oscillator
10.4 Sensor Oscillators and Frequency-to-Digital Conversion 291

depends on the direction of the stress relative to the channel orientation. Influences of the
temperature and the supply voltage on the basic oscillation frequency can be compensated by
deriving the output signal from the ratio of the oscillation frequencies of two appropriately
located ring oscillators [6].

1.1

# 11iO] direction
1.0 ? (1 101 direction

0.9

-
Af
f, 0.8
in Hz
0.7

0.6

0.5

0.4

0.3

0.2’
I

0.1
//

-
0 I I I I I
0 10 20 30 40 50 60
G in-N/mmz
Figure 1042. Percentage of frequency change of a ring oscillator as a function of mechanical stress,
depending on the channel orientation.

10.4.4 Frequency-to-Digital Conversion

Signal processing with microelectronic systems requires digital sensor signals. Only a few
sensors with direct digital output are available, eg, incremental or coded displacement and
angle sensors. Therefore, it is mostly necessary to convert the output signals into digital form.
Sensors with frequency analog output signals allow easy digitization. They are usually less ex-
pensive than amplitude analog sensors with subsequent analog-digital converters. Frequency
analog sensor signals have great advantages with respect to signal transmission regarding, eg,
292 10 Signal Processing

fail-safeness and easy galvanic separation. The galvanic separation can be achieved by
transformers or optocouplers.
Some examples of sensors with frequency analog output are:
tuning-fork quartz as a frequency standard for exact digital conversion,
quartz sensors for temperature measurement allowing high accuracy after individual
calibration,
ring oscillators integrated on silicon diaphragms as pressure sensors for evaluating the
piezoresistive effect,
resistive sensors, eg, silicon resistance thermometers, using the spreading resistance princi-
ple or photoresistances for illuminance measurement, in relaxation or harmonic RC-
oscillator circuits,
capacitive and inductive sensors controlling the frequency of oscillators.
The counting technique is the easiest way to convert a frequency signal into a digital signal.
The interdependence between the frequency J the constant measuring time T and the count
N is given by

N =f . T. (10-15 )

The measuring time T determines the conversion speed and the achievable resolution,
a small T giving a low resolution and a large T a high resolution, because the quantization
error 1/Ncan be written as

_1 -- - 1
(10-16)
N f * T '

To achieve a high resolution when measuring frequencies below 1 kHz with this method,
a comparatively large measuring time T of about 10 s or more is necessary for sufficient
resolution values. As a faster response time is usually required, the multi-period measurement
is a better choice. The principle is shown in Figure 10-13a.
The measured frequency f , is divided by the factor N T . The resulting count N is

(10-17)
Jx

with a resolution 1/N of

(10-18)

This linear dependence of the resolution on the measured frequency f x is shown in a dou-
ble logarithmic plot with NT as parameter in Figure 10-13b. The measuring time N T / f xis

NT - (10-19)
fx fref
10.4 Sensor Oscillators and Frequency-to-Digital Conversion 293

I 11

- b -
I-

Counter N

Figure 1043. Multi-period measurement: (a) block diagram; (b) resolution as a function of the
measured frequency.

The product of the resolution 1/N multiplied by the measuring time NT/f, is constant; its
minimum value is fixed by the reference frequency fRr. At a permitted measuring time
NT/f, = 0.1 s and a reference frequency fRf = 10 MHz, a resolution of 1/N = can be
obtained. To measure a frequency f, = 10 kHz, eg, NT= 1000 cycles of this frequency have
to be evaluated.
In multi-period measurement, the displayed count N is proportional to the period of the
frequency signal f,. To obtain the frequency value, the reciprocal of N must be calculated.
This can easily be performed by a microcomputer, which will need less than 100 ps. The multi-
period measurement has become an important method for the measurement of frequencies
in all ranges.
A variation of this method is multi-period measurement with the input-synchronized gate
time as shown in Figure 10-14. Here three counters are needed. Counter 1 (Nl) stores the
input pulses, counter 2 (N2) measures the synchronized gate time, and counter 3 is used as
294 10 Signal Processing

Figure 10-14. Multi-period measurement with input-synchronized gate time.

a programmable divider to preset the gate time. The quantization error is now one pulse of
the reference frequency fRf,
independent of the input frequency fx,because the flip-flop syn-
chronizes the counters 1 and 2. The measured frequency f, is given by

fx = f,f * (10-20)

The frequency-digital conversion allows the very easy performance of two mathematical
operations: the division of two frequency signals and integration over time.
With the circuit shown in Figure 10-15, the division of the two frequencies f,and fi is
evaluated. The count N is given by the product of the measurement time T = N T /fiand the
frequency f,/NFat the input of the AND-gate:

(10-21)

To reach a sufficient precision, the frequency f,has to be comparatively high, whereas the
frequency f2 should be small. The division factor NTshould be much greater than 1.

Figure 10-15. Division of two frequency signals.


10.5 Physical Model Functions 295

10.5 Physical Model Functions

One of the best options for effective sensor signal processing is to know the physical model
which gives the exact mathematical relationship between the measurand and the sensor out-
put. In most instances it is very difficult to find such physical model functions because the
measuring effects of sensors - if they are known at all - are very complex and therefore not
easy to express by mathematical equations.
The following sections give three examples of how to obtain physical model functions.

10.5.1 Transistor Thermometer

The p-n junction of a silicon transistor can be used as a temperature sensor for the range
from -50°C to +120"C [7]. The base-emitter voltage VBE of the transistor at U,, = 0 V
depends on the energy A E at r9 = O"C, the collector current I, and the temperature 8, and
can be written as

UBE= A E + -.kB e. 8 In (&) '


(10-22)

where k, is the Boltzmann constant and e the elementary electronic charge (Z, can be used
instead of Zc). The constant a depends on geometrical factors and physical constants, and n
describes the temperature dependence of the diffusion constant of minority carriers in the
base (n = 1.5 for n-p-n and n = 1.3 for p-n-p junctions).
In the simplest case, the temperature dependence of the base-emitter voltage U,, at a con-
stant collector current 1, is used for temperature measurement. The sensitivity, 6 U,,/6 8, is
then

-SUB,
_-.
68
- k~
e
[I. (h) - n] # constant. (10-23)

It is not constant and varies with the manufacturing tolerances of A E , n, and a.At ambient
temperature the sensitivity of silicon transistors is about -2 mV/K.
For more exact temperature measurements, the individual values of A E , a, and n of each
transistor thermometer must be determined by calibration cycles. Subsequently the measured
temperature is calculated using the physical model. This means a comparatively large expen-
diture.
A better way is to use the difference A UBE of the two base-emitter voltages of a transistor
resulting from two different collector currents, Z,, and Zc2:

(10-24)
296 10 Signal Processing

Now the tolerances of AE, a, and n do not have any effect on the measured results
and AUBEbecomes a linear function of temperature. With Zcz/Zc, = lo00 and kB/e =
86.3 bV/K, a constant sensitivity of

(AuBE)
68
--
-kB
. In
e
(%) = 0.595 mV/K (10-25)

is obtained.

10.5.2 Inductive Displacement Sensor

Neglecting leakage fluxes, the characteristic of a choke coil system used for displacement
measurement can simply be described by a hyperbola shifted in the direction of the measured
displacement. This physical model can easily be derived from a magnetic circuit. Considering
the leakage fluxes, however, we obtain a finite non-zero inductance L, even with a very large
air gap, because the leakage fluxes can be introduced into the model as a second magnetic
circuit parallel to the first. The behavior of the resulting inductance L, depending on the
displacement x, is shown in Figure 10-16.

t
L
Lo

L"
Figure 10-16.
Dependence of the inductance L on the displace-
ment x. (The structure of the coils is shown in
0 Xm X- the insert).

Lo and L, are the inductances at the displacements x = 0 and x -+ 00, respectively. At a


displacement x = x, the mean inductance shall be L, = (Lo + L,)/2. Three model func-
tions seem to be suitable for a description:
- an exponential function,
- a fractional rational first-order polynomial,
- a fractional rational second-order polynomial.
The rational second-order polynomial has been applied for the correction of the effect of
temperature on an inductive displacement sensor [4] (see also Section 10.10.1).
The exponential function used as a model function can be written as

L (x) = L, + (Lo - L,) . exp [ -(x/x,,,) - ln21 (10-26)


10.5 Physical Model Functions 297

The fractional rational first-order polynomial gives

L(x) =
Lox, + L,x (1 0-27)
x, +x
If three points of the characteristic are known, the coefficients x,, Lo and L , can be
evaluated from the values (x, , L,), (x2, L 2 ) , and ( x 3 ,L , ) as

(1 0-28)

(10-29)

1
Lo = (x3L3 - x3 L , + x, L3) -.* (10-30)
xm

In each particular case the model should be chosen by comparing the several results of the
different model functions with the actually measured sensor data.

10.5.3 Pressure Sensor Using the Hall Effect

The sensor in Figure 10-17is able to measure both absolute and relative air pressure. For
the measurement of absolute pressure, the space behind the diaphragm has to be evacuated.
The permanent magnet attached to the diaphragm is deflected by the pressure p. The
distance x between the pole surface and the magnetic sensitive area of the Hall-effect sensor
decreases. The resulting magnetic induction B (more exactly, the component orthogonal to the
sensor area) at the Hall-effect sensor generates the Hall voltage Up. The additional

diaphragm

Hall sensor
temperature sensor

Figure 10-17. Principle of a pressure sensor utilizing the Hall effect [S].
298 10 Signal Processing

temperature sensor can be used to perform a temperature-effect compensation, which is


described in more detail in Section 10.8.
Now the physical model function of this pressure sensor has to be developed. For small
deflections x of the diaphragm, x is linearly dependent on the pressure p [9]:

x = x, - k , ‘ p , with k , > 0, (10-31)

where x, is the initial distance between the permanent magnet and the Hall-effect sensor at
the pressure p = 0.
The magnetic induction B at the Hall-effect sensor depends on the distance x and the
geometry of the permanent magnet. The cubic magnet used here leads to

R
(a/2)2
x 1/2(a/2)2 + x*
- arctan
(a/2)2
(x + c) 1/2(a/2)2 + (x + c)2 1
(10-32)
where M is the magnetization, a the side length of the square front area, and c the thickness
of the permanent magnet in the direction of x. According to [lo], it is a good approximation
to describe this relationship with an exponential function, which leads to the following equa-
tion:

, with B, = B (XO) . (10-33)

The Hall-effect sensor produces a voltage Up according to the equation

Z
U,(X) = R , * $ * B ( x )3 (10-34)

where R, is the Hall constant, d the thickness of the Hall plate, and Zpthe control current
of the Hall sensor.
Summing up the above equations, the relationship between the Hall voltage Up and the ac-
tive pressure p, neglecting all temperature effects, can be written as

(10-35)

With the substitutions U, = Bo R , - . Zp/d and po = x,/k,, we obtain the simplified ex-
pression

up@) = u, * exP @/Po) (10-36)

as a physical model of this Hall-effect pressure sensor.


10.6 Digital Modeling of Static Sensor Characteristics with Mathematical Models 299

10.6 Digital Modeling of Static Sensor Characteristics with


Mathematical Models

If no physical model of a sensor is known, or effects of manufacturing tolerances and en-


vironmental interferences degrade the sensor signal such that the sensor model becomes
unusable, the sensor characteristic should be described with a mathematical model to allow
digital signal processing. In general, reference points are needed to determine the sensor
behavior. Between these points the sensor characteristic can be estimated by different
mathematical methods, usually interpolation or approximation algorithms.
By addition and multiplication of constants, it is possible to compensate for production
derived deviations of the offset and slope of a linear sensor characteristic (scaling). In this
case, the standard mechanical and electrical calibration measures in sensor manufacture, even
laser trimming, are no longer needed. In either instance, however, it is essential to determine
the sensor responses at two different measurand inputs, eg, zero and full-scale.
Up to now in the selection of sensor principles, non-linear sensor characteristics were
avoided as far as possible because analog linearization methods (using, eg, a diode function
generator) other than digital methods require great expenditures and pose problems with
respect to reproducibility.
For digital linearization the following methods can be used:
- look-up tables,
- polygon interpolation,
- polynomial interpolation,
- interpolation with cubic splines.
In the following the main aspects of these methods will be discussed briefly.

10.6.1 Look-up Table Method

This method uses narrowly spaced reference points to describe the characteristic of the sen-
sor over the whole measurement range for linearization. These points are stored in the ROM
(Read Only Memory) of a microprocessor. Every time the sensor responds to a measured
value, the microprocessor looks up the corresponding value in the table and passes it to the
output, eg, a display unit. A disadvantage of this method is the great storage capacity needed
to keep the large number of reference points necessary to obtain a sufficient accuracy of the
sensor characteristic.

10.6.2 Polygon Interpolation

The polygon interpolation (sectional linearization) needs fewer reference points than the
look-up table method. Between the stored reference points the sensor characteristic will be in-
terpolated by straight lines (see Figure 10-18). If the curvature of the characteristic is not too
strong, it is possible to obtain a good approximation with only a few reference points. Strongly
curved functions, however, can only be described adequately by a large number of straight
lines and therefore again large storage capacity is necessary.
300 I0 Signal Processing

xo x, x2 xj X& xg x, x - Figure 10-18.


Polygon interpolation.

10.6.3 Polynomial Interpolation

The polynomial interpolation describes the functional relationship between the n measured
reference points of a sensor characteristic with a single polynomial of order s n - 1 over the
whole measuring range. The main disadvantage of this method is the oscillations of
polynomials of higher orders. In general, all n - 2 points of inflection of the interpolation
polynomial are located within the considered interval, and outside the interval the function
grows beyond all bounds. This does not coincide with the smooth shape of real sensor
characteristics. Therefore, polynomials with an order higher than 3 are rarely suitable for the
interpolation of sensor characteristics. The example in Figure 10-19 shows the unsatisfactory
result of an interpolation of a curve (for which five reference points are known) with the
following fourth-order polynomial:

c ki
4
f(x) = * xi. (10-37)
i=O

Figure 10-19.
Interpolation with a fourth-order
x - polynomial (---).
10.6 Digital Modeling of Static Sensor Characteristics with Mathematical Models 301

10.6.4 Cubic Spline Interpolation

With cubic spline interpolation it is possible to obtain a smooth curve of a sensor


characteristic which leads exactly through all given n + 1 reference points. To carry out this
interpolation, the measuring range is divided into intervals bounded by the reference points
(see Figure 10-20).

Figure 10-20.
Interpolation with cubic splines.

For each interval, separate cubic parabolas Si(x), which have to pass through the reference
points, are defined as

with x E [xi, xi+J and i = (0, 1, 2, ...,n - 1). For a simpler mathematical description the ad-
ditional polynomial

s,, (x,,) = a,, (10-38 b)

is defined, which is only valid for x = x,,.


The coefficients a,, b , ci,and di of the parabolas will be determined in such way that at
their junction points adjacent functions coincide with each other in the function values and
have the same gradient and curvature. This leads to the following conditions for the
polynomials Si(x):

Si(Xi) = Yi with i = (0, 1, ..., n), (10-39)


&(xi) = Si-,(xi) with i = (1, 2, ..., n), (10-40)
S,!(xi) = S;-l (xi) with i = (1, 2, ..., n-l), (10-41)
Sy (xi) = qLl (xi) with i = (1, 2, ..., n -1) . (10-42)

To determine all 4 n + 1 coefficients, we need two more equations, which we obtain by


specifying additional features of the two polynomials bounding the measuring range. Most
302 10 Signal Processing

commonly the curvature at the boundary points (x,,, yo) and (x,,, y,) is set to zero. The spline
function is then called a “natural cubic spline”.
With c, = c = 0 and the abbreviations hi = xi + - xi and y , = yi+ - yi = constant,
we obtain the following equations for evaluating the spline coefficients [ll]:

a.I = SI . ( xI . ) = y I.# i = (0, 1, ..., n), (10-43)

hi-i(~;-i) + 2(hi-l + hJci + hici+l = 3ym ( Li...,- - ’


i = (1, 2,
-
hll-l)
n-1), (10-44)
bi = ym/hi - (c;+ I + 2Ci) hi13 , i = (0, 1, ..., n-1), (10-45)

d; = (c;+, - c;)/(3 hi) , i = (0, 1, ..., n-1) . (10-46)

To obtain a good interpolation result for common types of sensors, typically five reference
points are needed for a cubic spline interpolation. Especially if a priori information about the
sensor behavior is missing it is recommended that cubic splines are used for interpolation.

10.7 Minimum Principles for Approximations

The efficiency of interpolation methods depends on the accuracy of the reference points
used since an interpolated characteristic coincides with these points. Considering all variations
and deviations in measuring the reference points, which have a great effect on the results of
an interpolation, the use of approximation methods may offer considerable advantages. The
coefficients of the smoothing functions are achieved by minimizing a deviation function. The
approximation function does not lead exactly through the given set of measured points. Com-
mon approximation methods are the various regression methods, including smoothing splines.
The most frequently used minimal principles, which will be discussed below, are
- the deviation function R for the L, approximation,
- the deviation function S for the L, approximation, and
- the deviation function T for the L, approximation.
Naming the measured reference points ( x k , Yk)r the coefficients of the approximation func-
tion a,, a,, ..., a, and the values obtained by the approximation function f ( X k ) , the devia-
tion functions R, S, and T can be written as

c Pk !
n
R(a1, a,, -.-,a,) = IYk,-f(aI, ..-)am, = Min. (10-47)
k= 1

k= I

!
T(a,, a,, .. ., a,) = max Pk
k
I yk - f ( a , , a,, . .., a,, xk) I = Min. (10-49)

where pk is the weighting factor of the reference point (xk, yk).


10.7 Minimum Principles for Approximations 303

The deviation function R is the weighted sum of the absolute values of the deviations and
produces the discrete L , approximation for minimal deviations. The L,approximation is par-
ticularly insensitive to mavericks. If only one point is outlying like a wild shot from an other-
wise linear relationship, the approximation line leads directly through all of the other points
and neglects the maverick.
The deviation function S is the weighted sum of the squared deviations and gives the
discrete L, approximation when minimized (least-squares method). In standard cases and in
cases of doubt, this method should be applied because at identical weighting factors primarily
large deviations are accentuated in the deviation sum.
The deviation function Tgives the maximum weighted deviation between the sensor charac-
teristic and the approximation function. This kind of approximation is also called Chebyshev
approximation. It is of particular interest in the field of sensor techniques.
For practical applications, the number of given reference points should be three to five times
greater than the number of parameters to be determined [12].

10.7.1 Example of Regression Method

A simple example of the use of an approximation method is linear regression. The


gradient m of a linear characteristic y = f (x) = m . x passing through the origin has to be
determined such that the sum of the squared deviations of n measured points (xk’ yk)
becomes a minimum (see Figure 10-21).

Figure 10-21. Regression line with slope m. x-

Setting all weighting factors identically to pk = 1, the deviation function S is

(10-50a)

(10-50b)

(10-50~)
304 10 Signal Processing

leading to the minimization condition

(10-51)

As the solution, the gradient m is obtained as

k=l
m = (10-52)
i
k=l
xz

More information about approximation algorithms can be found in Chapter 11.

10.8 Correction of Cross Sensitivities

If the basic behavior of a sensor characteristic is known and is not changed qualitatively
by manufacturing tolerances and cross sensitivities, the base function method is well suited
to describe the effects of interferences on the sensor characteristic. As seen in Figure 10-22,
the base function

coincides with the nominal characteristic of the sensor at a constant rated value xzoof the in::
terfering quantity x,. The base function method then gives the following notation for the

I
Xl - Figure 10-22.
Base function method.
10.8 Correction of Cross Sensitivities 305

sensor output signal y, which depends on the measurand x, and on variations of the in-
terference x2:

This means an expansion of y (x,, x2) in a series of the base function yo (x,, x,,). Thereby
the functions cj (xz), i = 0, 1, .. ., n, describe the effect of the interference x2 and become
zero at the rated interference value x, = x,,. Setting them up as polynomials, we obtain

cj (x,) = bi, (x2 - Xzo) + biz(x2 - x,,), + . . . (10-55)

Figure 10-23. -

The pressure sensor has a temperature dependence as shown in Figure 10-23. The nominal
temperature is Lp0 = 20°C. Restricting the general base function model to its first three terms
and setting up the functions cj (x, = 8) to second-order polynomials, we obtain the follow-
ing description of the sensor signal, y (p, 8):
Y ( P ,8 ) = [bOl (8 - 80) + b,, (8 - L90)21+ [1 + b,, (8 - 80) + b,, (L9 - 8d21
. YO (pt 80) + [b21 (8 - 80) + bz2 (8 - 80)21 Y:
1 (P, 80). (10-56)
The calculation of the coefficients bij and ci(8) by multiple linear regression gives non-
zero values for all of them. This means that it is not only the obvious zero shift of the sensor
characteristic that is caused by the effect of temperature; in addition, small changes in gradient
and curvature are also produced. The remaining maximum difference between the measured
sensor characteristic t a temperature Lo, # d,, and that which is obtained by the base func-
tion method is about 1% of the measuring range (y-) (see Figure 10-24a and b).

(a)

t
Y (P.3
in V

(b)

t
-
AY
YI#x
in %

-0.4 -
- 0.6 I 1 I 1 I 1

Flgm W24. (a) Measured and calculated characteristic of the pressure sensor at 19, = 80°C; @) re-
mainii relative deviation.
10.9 Dynamic Correction of Sensors 301

10.9 Dynamic Correction of Sensors

The use of microcomputers with suitable algorithms makes a dynamic correction of sensors
feasible. If the dynamic parameters of the sensor system are known, the dynamic correction
of linear systems can be performed by solving the convolution integral. With the terms in
Figure 10-25, the input signal x, ( t ) is given by

(10-57)

-
x,(t)
F(p1
x,(tl x,(t) xa (t1 x g ( t )

The weighting function g ( t ) results from the Laplace retransformation of the inverse
transfer function l / F ( p ) of the sensor:

g ( t ) = L - I (l/F(p)) . (10-58)

The dynamic correction becomes easier if the temporal behavior xe ( t ) of the sensor input,
expressed in the differential equation, can be written as an explicit function of the output
signal xa ( 2 ) . This is possible for many sensors. These can then be described with sufficient
accuracy as first- or second-order systems. For a second-order system the input x e ( t )
becomes

(10-59)

where 4 is the damping factor and oothe angular frequency of the undamped self-oscilla-
tion. A first-order system with a time constant r has an input signal x, ( t ) of

Both input signals x e ( t ) can be calculated from the output signals x , ( t ) and their
derivative(s). Thereby x, ( t ) is approximated with some previous sampled data. For second-
order sensor systems cubic spline polynomials are advantageous, because then the second
derivative of x, ( t ) can be at least a linear function of the time t.
A similar approach is adopted with difference equations which are usually obtained with
digital systems. There the derivatives are calculated from previous values:
308 10 Signal Processing

Simple examples of dynamic systems are temperature sensors, which can be considered as
first-order systems 1131.

10.10 Examples of Sensor Signal Processing

10.10.1 Inductive Sensor System for Mechanical Quantities

Following the base function method, the temperature effect on a microcomputer-oriented


sensor system with inductive sensors for the measurement of mechanical quantities can be cor-
rected [4]. The temperature is measured with a silicon temperature sensor. A microcomputer
performs the linearization of the sensor characteristic and the correction of temperature ef-
fects depending on the behavior of each individual sensor. This leads to a simple sensor
calibration without any trimming elements. The specific signal processing of the sensor is
done in three stages as shown in Figure 10-26.

sensor

-1
I
I
I coefficients
I determined
I at calibration
1 I
(regression)

1 1
F
1 I

j
I
correction of I
temperature, influence I
I I

I
coefficients
base function method determined
(rational polynomial) by physical
model
Figure 10-26.
I display I Influence corrections in an inductive
sensor system.

First in the calibration procedure, manufacturing tolerances and temperature effects are
determined. The manufacturing tolerances are then compensated by a low-order polynomial
giving the corrected sensor output F, :

F, = al + b, F + c, F 2 (10-63)
10.10 Examples of Sensor Signal Processing 309

Subsequently the temperature compensation is carried out utilizing the equation

F2 = + c2Fl + (8 - d o ) . (4 + d2F1+ e , F : ) , (10-64)

with do = reference temperature.


All coefficients of these two functions are evaluated from calibration data by regression
methods. Finally, the base function of the sensor, here realized by a fractional rational
polynomial, leads to the required measured value s.
A computer-corrected inductive displacement sensor with a measuring range of 2.5 mm
showed absolute deviations of the nominal value of less than 1 pm at temperatures between
25 and 50°C (see Figure 10-27).

deviation in prn

in rnrn
0
Figure 10-27. Deviations of a temperature corrected inductive displacement sensor.

10.10.2 Gas Concentration Measuring System

For the measurement of gas concentrations low-cost sensors that allow the selective deter-
mination of the particular gas components of a gas mixture have so far not been available.
In addition to its sensitivity to the desired component, each gas sensor always shows a cross-
sensitivity to other gases. Therefore, a determination of several gas components is feasible
only with several sensors showing different sensitivities to the individual components. Such
a multi-sensor system can be used for a multi-dimensional analysis if the n measuring com-
ponents x 1, x,, . .., x, are calculated from the output signals y , ,y , , .. .,y , of the rn gas sen-
sors. The coefficients of the equation system

(10-65)
310 10 Signal Processing

have to be determined by a large number of measurements on well known gas mixtures. Only
in the case of a linear equation system with m = n is an explicit determination of the gas com-
ponents possible by matrix inversion.
Assuming a non-linear equation system, which additionally contains mixed terms, the
superposition principle cannot be used. Hence the algorithms to solve the equation system
become more complex. The analysis of a binary gas mixture with two gas sensors requires the
evaluation of the following equations:
XI = a,, + ally, + any2 + a13-Y: + a14-Y: + a,,Y,Y, 9
(10-66)

x2 = + a21yl + a22Y2 + a23Y? + a24y: + a25YIy2 * (10-67)


These twelve coefficients a,, and aZican be obtained off-line from the measurement data of
the calibration cycles, eg, by multiple regression methods. It is then easy to determine the two
gas concentrations x , and x2.

Contour line of
sensor

y2=const.

Evaluation of the
gasconcentration

1 I
Figure 10-28
Determination of gas concentrations
with a multi-sensor system by
Xl graphical means.
10.11 References 311

Another possibility for obtaining the gas concentrations from the sensor outputs is given
by search strategies [14]. As before, the necessary procedures for two sensors and two gas com-
ponents will be discussed.
First we must know the whole system of characteristics of the sensors used, ie, for each com-
bination of the two gas components x1 and x, the resulting output yi of each sensor has to
be specified (Figure 10-28a). For this purpose a rectangular grid is laid on the x,x,-plane.
The values of yi at the intersection points are determined. The intermediate values can be
achieved by a suitable two-dimensional interpolation method, eg, the bicubic spline interpola-
tion. Now the sensor signals y1 and yz are analyzed as follows: every yi corresponds to a con-
tour line in the characteristic system of that sensor. The projection of that contour line on
to the x,x,-plane leads to a line of possible concentration combinations (see Figure 10-28b).
The intersection points of the two resulting lines (one for each sensor) give the possible solu-
tions. Therefore, the uniqueness and accuracy of the solution depend on the features of the
sensors used. Especially different sensitivities of the sensors to the two gas components are
necessary.
A more detailed discussion of multi-sensor systems for measuring gas concentrations and
identifying gases follows in Chapter 11.

10.11 References

[I] Binder, J., Tech. Messen 53, No. 2, (1986) 55-59.


(21 Kowalski, G., Sens. Actuators ll, No. 4, (1987) 367-376.
131 Delapierre, G., Danel, J. S., et al., “A quartz micromachined closed loop accelerometer” in:
Eurosensors ’87, Cambridge, UK, 22-24 September 8Z pp. 223-224.
(41 Kohn, D., Fortschr. Ber. VDL Reihe 8, No. 120, Diisseldorf: VDI-Verlag, 1986.
[5] Breimesser, F., Poppinger, M., Schorner, R., VDZ Ber. No. 509, (1984) 147-150.
161 Schorner, R., VDZ Ber. No. 677, (1988) 93-97.
[7] Verster, T. C., Electron. Lett. 4, No. 9, (1968) 175-176.
[8] Bottcher, J., MeJsignalverarbeitung fur Druckaufnehmer mit Hallsensoren; Diplomarbeit,
Technische Universitat Miinchen, Miinchen 1988.
[9] Rohrbach, C., Handbuch fur elektrisches Messen mechanischer GroJen; Diisseldorf: VDI-Verlag,
1967, pp. 532ff.
[lo] Heidenreich, W., Kuny, W., Magnetfeldempfindliche Halbleiter-Positionssensoren; Special print
from: Elektronik Industrie, No. 5, 1985, and No. 6, 1985.
1111 Jordan-Engeln, G., Reutter, F., Numerische Mathematik fur Zngenieure, Mannheim: BI-
Wissenschaftsverlag, 1982, pp. 281 ff.
1121 Bauer, G., Kohn, D., Trilnkler, H.-R., Measurement 2, (1984) 145-148.
1131 Woschni, E.-G., MeJdynamik; Leipzig: S. Hirzel-Verlag, 1972.
1141 Matt, K., Pointner, P., Ziemann, G., Hard and Soft, March 87, Fachbeilage Mikroperipherik,
pp. v-VII.
[I51 Trilnkler, H.-R., Tech. Messen 49, No. 10, (1982) 343-353.
[I61 Trilnkler, H.-R., Messen-Prufen-Automatisieren22, No. 6, (1986) 332-338.
1171 Trankler, H.-R., Tech. Messen 55, Nr. 3, (1988) 114-120.
[I81 Trankler, H.-R., “Signalverarbeitungskonzepte”, in: Technologietrends in der Sensorik; Berlin:
VDIIVDE-TechnologiezentrumInformationstechnik, 1988.
1191 Trankler, H.-R., Taschenbuch der MeJtechnik mit Schwerpunkt Sensortechnik; Miinchen: Olden-
bourg, 1989.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

11 Multisensor Signal Processing


RUDOLFMULLER.Technische Universitiit Miinchen. FRG

Contents
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 314
11.2 Correlation Method . . . . . . . . . . . . . . . . . . . . . . . 316
11.3 Vector Method . . . . . . . . . . . . . . . . . . . . . . . . . 320
11.4 Partial Least Squares (PLS) Method . . . . . . . . . . . . . . . . 324
11.5 Transformed Least Squares (TLS) Method . . . . . . . . . . . . . 326
11.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
11.7 References . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
314 I1 Multkensor Signal Processing

11.1 Introduction

This chapter relates mainly to chemosensors. The task of a chemosensor system is to identify
one or more substances out of a large number of possible substances and to indicate its or
their concentration. Figure 11-1 demonstrates this in more detail. On the left-hand side there
is a number N of substances with the actual concentrations [XI] to [X,] and the sensor
system should be able to deliver electrical signals ([XI]) to ([X,]) as measures of the actual
concentrations, so that ([X,]) should be proportional to [X,].

tx11-

[XNI -
sensor

system

- m,1> Figure 11-1.

- <[X,I>
Task of a sensor system. [Xi] denote
the measurands, ( [ X i ] > the correspon-
ding measured values.

The usual approach to achieving this task is to use a set of Z = N sensors or transducers
T I to TI, each of which is selectively sensitive to only one component, as shown in
Figure 11-2. Thus the upper square matrix should have non-zero elements only in the diagonal.
In this case the signals S , to S, delivered by the array of sensor elements contain the desired
information already in a well ordered form; each signal is a measure of one and only one gas
component.
If the responses of the sensors are linear, the signals Siare proportional to the actual con-
centrations [X,] so that the desired answers for the system are simply achieved by multiplica-
tion of the signals by “calibration” constants, one for each substance. If the transfer
characteristics of the sensor elements are non-linear, it is necessary to apply a correcting back-
transformation in order to obtain an output ([X,]) proportional to the measured quantity
[X,]. The methods used for this purpose are described in Chapter 10.
For a sensor system according to Figure 11-2, it is required that the sensors have zero cross
sensitivities. It is possible to reduce these cross sensitivities by technological means (as
11.1 Introduction 315

a11 0 0 0 0
0 . 0 0 0 Si = anjlX,1

!r
0 0 a,; 0 0
-1
0 0 0 . 0 <[X,1> = a,; S; =[X,l

a -1I 1

Figure ll-2.
. -I

.
Desired properties of the an;
sensor elements T, to T I .
Siare signals; the transfer
characteristics are assumed -1
"NI
to be linear.

described extensively in Volumes 7 and 8), but this is usually not achieved ideally and real sen-
sors exhibit non-zero matrix elements outside the diagonal, as shown in Figure 11-3. In addi-
tion, the transfer characteristic (signal as a function of concentration) is usually non-linear,

IX,I-

[Xn1-
4 I,,

*
-
. . . . .
fni *
. . . . .
. . . .
*

fNIl

1 F'xn
I

PARC
-
Figure ll-3.
Actual properties of the sensor elements T, to TI: <lXN1>
PARC = pattern recognition.
316 11 Multisensor Signal Processing

as indicated by the functions fni. Here it is necessary to use more elaborate computational
techniques in order to extract the required information.
In the following sections, four methods of pattern recognition (PARC, eg, [l]) are described,
which are suitable for use with chemosensors:

- correlation method (CM) [2, 31,


- vector method (VM) [4],
- partial least squares (PLS) method [l, 51,
- transformed least squares (TLS) method [6].

The first two methods serve the purpose of identification only and an evaluation of the con-
centration has to be done separately. The other two methods were developed for identification
and indication of concentration in one step.
The signal processing has been discussed above as a means of reducing the cross sen-
sitivities. The combination of sensor arrays with methods of pattern recognition brings addi-
tional advantages, however. For instance, it is possible to identify a number N of gases (eg,
100) that are not present simultaneously with an array of Z sensors, with Z c N (eg, Z = 10).
This situation is similar to the senses of smell and taste, so that the expression “electronic
nose“ can be used for a chemosensor system with non-selective sensors (in the human nose
there are seven different cells for smell detection [7] and many hundred different smells can
be distinguished).
The term sensor is usually used for a device that delivers an electrical signal as a measure
of a non-electrical quantity. If an array of non-selective chemosensors is used, the sensor array
may be termed a “multi-dimensional sensor”. In order to avoid confusion, the following nota-
tion is used in these sections: sensor array for the whole set of chemosensors and sensor ele-
ment for a single sensor in this array.

11.2 Correlation Method

The pattern generated by the signals from a sensor array is characteristic of the substance
applied to the array. The evaluation method tries to find a similarity between this pattern and
those already learned in a teaching process. The similarity index used in Reference [2] and the
correlation coefficient used in Reference [8] are both suitable quantities in this respect.
Consider as an example the signals delivered by four different MOS chemosensors such as
that shown in Figure 11-4. The sensors are modified hydrogen sensors [9], the modification
being a zeolite film on top of the palladium electrode; one sensor is uncovered and the other
three sensors have zeolites with different pore sizes (3, 4 and 9A) [lo]. Figure 11-5 shows the
signals of this sensor array. In this case the rate of change of the MOS capacitance immediate-
ly after application of the test gas is used as the signal [8]. It can be seen that each substance
has a charachteristic signal pattern.
It is essential to note that the number of distinguishable patterns can be greater than the
number of sensor elements. It is not the number of sensor elements that limits the number
of detectable substances, but rather the accuracy of the signals obtained. However, here it is
necessary to know that at a given time only one substance is applied to the sensor array.
11.2 Correlation Method 317

0 zeolite layer

I 1 I

p Si

p’ -Si

Figure ll-4.
MOS gas sensor with zeolite layer. A1

~-c-,-oHydrogen
Ammoniac
Methand
Acetone

Benzol
Figure 11-5.
Signal pattern obtained with four
different MOS sensors. The rate of
change of the capacitance after applica-
tion of the gas is used as the signal [lo].

Figure 11-6.
Change in signal pattern with gas con-
centration (---, acetylene; -, methane).
The voltage shift necessary to keep the
capacitance constant after application of 2 3 4
the gas is used as the signal [12]. i-

If all the transfer characteristics of the sensor elements used in the array are linear, the pat-
terns obtained will be independent of the concentration of the applied substance. Usually,
however, the transfer characteristics of sensor elements are non-linear and a difference in this
non-linearity between sensor elements leads to a (slight) change of the signal pattern with con-
centration. Figure 11-6 shows this change for methane and acetylene. In spite of this change,
a visual inspection enables one to distinguish between the two substances.
A convenient means of automatic inspection is to calibrate the “mean pattern” N j ksimply
as the linear average of the signals obtained for different concentrations p:
1 m
Njk = NjkpI (1 1-1)
p=l
318 I1 Multisensor Signal Processing

Here Nikp is the signal of the sensor element i for the gas or substance k with concentration p.
Hence N i k corresponds to the averaged signal (of sensor element i for the calibration gas k )
with the concentrations p distributed as uniformly as possible over the concentration range
of interest.
This shows that the calibration (teaching) process should be carried out with the attempted
application in mind. In order to remove the bias signal, which usually contains no information
suitable for identification, it is of advantage to subtract the average value Nkover the sensor
elements used:

(11-2)

Hence the quantities (Nik - Nk)contain the information necessary for identification of
gas k.
The correlation coefficient is defined as the average of the normalized product of the quan-
tities to be compared [ll]. We denote by S, the signal of sensor i for gas j with concentra-
tion p and by Sip the average value of the signal for gas j:

(11-3)

According to its definition, the correlation coefficient is written as

(11-4)

The correlation Coefficient pjkpdefined above may assume values between - 1 and + 1 ; for
identical spectra it is + 1. The question now is whether P,kp for j = k (test gas = calibration
gas) assumes values in the vicinity of + 1 irrespective of the concentration p of the test gas.
Figure 11-7 shows the evaluations for the gases hydrogen, methane, and acetylene.
The measured signal pattern was correlated with each of the three calibration spectra on
an individual basis. Figure 11-7a shows the measurements of hydrogen at different concentra-
tions, Figure 11-7b applies to methane and Figure 11-7c to acetylene. It will be noted that the
type of gas is reliably identified in each case irrespective of the concentration. If a threshold
is inserted, eg, at p = 0.8, the identified gas can be automatically indicated.
As mentioned above, it is not the non-linearity itself which makes the signal pattern change
with concentration, but rather the difference in non-linearity between the sensor elements
which limits the application of the correlation coefficient. On the other hand, the cross sen-
sitivities of the sensor elements turn out to be an advantage. In principle, all substances to
which the sensor array responds with distinguishable patterns can be identified.
It has been stated above that it is necessary to know that only one substance is present at
a time. In practice, the sensor system will respond to the application of a gas mixture with
11.2 Correlation Method 319

pJk I

Figure 11-7.
Correlation coefficient between test gas j and
calibration gas k for hydrogen, methane, and
acetylene.

the answer “this pattern is not known”, just as it will respond if a substance is applied which
has not been used in the calibration process. If, however, a mixture of fixed composition is
used in the calibration process, the system can identify it, no matter how complex this mixture
may be.
So far we have considered only the identification of a substance and not the measurement
of its concentration. If the identification has been achieved, the measurement of the concen-
tration can be done in the conventional way using one of the sensor elements, preferably that
with the highest sensitivity towards the identified substance [2]. Another way is to use the
weighted signal of all elements as described in detail in [12]. In this way the magnitude of all
the signals is used as a measure of the concentration, thus giving a better signal-to-noise ratio.
Essentially, two sensor elements would be sufficient to identify a gas, as represented for in-
stance by the signals S, and S, in Figure 11-8. The signals Siare here normalized to the signal
of the sensor element 1. The number of identifiable gases is related to the number of

Figure 11-8.
Normalized signal pattern for two different substances
A and B. The signals Si may be of the type shown in
Figure 11-6. i-
320 11 Muitisensor Signal Processing

separately distinguishable signal values S2/S,. Since the ratio S,/S, for a single substance
may change with concentration, the "identification range" for this substance must cover a
finite region of S,/S,. Hence both the difference in non-linearity of the characteristics and
the measurement accuracy will limit the number of distinguishable substances [13]. If more
than two sensor elements are provided, the measurement will contain a redundancy with
respect to the identification of the substance. This can be used to advantage, for instance, in
the following ways:
1. To detect a defective sensor element. If p does not tend towards unity for all calibration
gases, an attempt can be made to exclude the sensor elements from the evaluation one at a
time. Once p z: 1 has been reached by this method, this will signify that the excluded sensor
element is defective.
2. To improve the measurement accuracy. If, for example, the probability of error
in the identification of a gas with two sensor elements is about 30%, this can be reduced
to 0.3 . 0.3 = 10% by adding a third sensor element, if the errors of the individual sensor
elements are assumed to be of the same size and statistically independent.
The combination of signals to extract the information by simple mathematical operations
such as subtraction or division is good practice: a resistance bridge using one element sensitive
to the desired quantity and temperature and another element being sensitive to temperature
only is an example. The new aspect introduced by pattern recognition is the extension to many
sensor elements and the generalization to more complicated calculations.

11.3 Vector Method

The vector method uses essentially the same information as the correlation method but in
another sequence and sometimes also with modifications.
Let us plot the signals S,to S, in an I-dimensional measurement space; Figure 11-9 shows
this for the example of Z = 3. Each measurement thus corresponds to a vector S with I com-
ponents. The same holds for the calibration process, that is, for each measurement there is
a vector N
In the correlation method, the vectors Np for different concentrations of the same
substance are averaged so that a vector hfis obtained, which carries the information for iden-

Sensor 3

,' Sensor 2 Figure 11-9.


Average vector for the calibration gas
(averaged for different concentrations)
and vector S for a test gas in the signal
space.
11.3 Vector Method 321

tification in the direction of the vector. The correlation coefficient is simply the scalar product
of S and &'normalized after multiplication. Hence the correlation coefficient is essentially the
cosine of the angle 0 between S and &'.
The vector method described by Horner and Albertshofer [4] normalizes the vectors S and
N first and compares them afterwards.
It is evident from the discussion above that the cross sensitivity is not detrimental to the
identification of the substance. It does not matter that the vectors are not pointing in the direc-
tion of the axes, as they would do without cross sensitivities. It is further evident that many
more directions are distinguishable than there are axes (dimensions) of the array, which means
that more substances can be distinguished than there are sensor elements. The non-linearity
of the transfer characteristics is not necessarily detrimental either, but if a difference in the
non-linearity between different sensor elements leads to a movement of the directions of the
vector S with concentration, the number of distinguishable directions with finite space angle
decreases. It may be possible to reduce this movement of the vector with concentration by cer-
tain manipulations of the sensor signals, that is, by an appropriate signal transformation.
These transformed signals can again be plotted in an I-dimensional space, the so-called feature
space.
This will be explained by means of an example, as follows. The transfer characteristic of
a metal oxide sensor can be approximated by the following expression [2]:

R.. -
L - rij = (aj [Xj] + l)"ii (11-5)
4 0

where R , is the resistance of the sensor element i with the gas j applied with a concentration
[Xj]. The quantity Rio is the resistance of the sensor element i in air. If the coefficient aj
depends mainly on the type of gas and is essentially the same for all sensor elements i, the
quantity Pij is independent of the concentration of Xi but characteristic for the gas compo-
nent j [4]:

(1 1-6)
C
i=l
log rij C
i= I
mij

Figure 11-10 shows the pattern Pij for the two gases methane and carbon monoxide
generated by four different metal oxide gas sensors. Figure 11-11 shows this feature space for
the example of three sensor elements. Each measurement is represented by a vector Pi with
the components Pip According to Equation (11-6), the locus of all possible values for Pi is a
plane of order i - 1. As a consequence of a slight change of uj from sensor element to sensor

j : f i

Pjj 0.2
0.1
Figure ll-10.
Signal pattern for methane and carbon 0
SI 52 53 54
monoxide. i- I-
322 I1 Multisensor Signal Processing

P3, (Sensor 3 1

(Sensor 2 1

Figure 11-ll. Location of two substances A and B in the feature space defined by Equation (11-6). The
coordinate axis z is defined as vector from p B to pA.

element and drift or fluctuations, the vectors Pj for a given gas will cover a range of finite
extent in the plane of order i - 1. However, an average vector pA can be defined for substance
A and a similar one for substance B.
The distribution of points can be projected to the coordinate z linking the two vectors pA
and pB. Figure 11-12 shows the probability density function along z for the measurements
with methane and carbon monoxide [4] with the assumption of a Gaussian distribution. In
a Gaussian distribution the variance o describes the width of the distribution and 84% of the
measurement points are within a region ,u f o. Within p k 2 o there are 98% of the

Figure 11-12.
Distribution in the feature space for methane
and carbon monoxide along the coordinate z
of Figure 11-11.
11.3 Vector Method 323

measurements. Thus bA,pB and a ,, a, can be determined in a calibration process for the
two substances A and B.
There are two suitable ways to compare test measurements with the calibration data [I]:
1. Around pA a region can be defined which covers a certain percentage of all the results
obtained with the substance A, limited by the so-called confidence envelope. With respect to
Figure 11-12 there are, for instance, 98% of all corresponding results for substance A within
the region pA f 2 aA.If a test measurement vector now is within this region the substance
is identified as substance A with a confidence of 98%.
2. The distances between the measurement vector and the calibration vectors are calculated
and the nearest neighbor is determined. Thus the substance can be identified. The disadvan-
tage of this method is the misinterpretation of a substance not used in the calibration process,
whereas method 1 would give the result “the substance is not known”.
With the aid of Figure 11-12, it is possible to define a quantity S, representing the selec-
tivity of the sensor system with respect to the substances A and B.
The quantities S, and S, are defined as follows:

(1 1-7)

This definition of selectivity can be used for sensor arrays but it cannot be applied directly
to single sensors. The quantity S*, however, defined in Equation (11-8) can be used for com-
parison. Figure 11-13 shows schematically the signals obtained from two single sensors; sensor
a is especially sensitive to substance A and sensor b to B as indicated by the transfer
characteristics S, ([A]) and S, ([B]). Nevertheless, each sensor is sensitive to the other
substance also, as is evident from the characteristics S, ([B]) and S, ([A]). The selectivities
SPT/, and S&, are measures of these cross sentivities:

Two non-selective elements would have S * = 1, whereas ideal elements give S* - 00 (note
that this definition of selectivity is oriented to the method of pattern recognition described
here, so it does not necessarily coincide with the general definition given in Chapter 2). In
Figure 11-14, the measured selectivity of two single elements is compared with the selectivity
of an array consisting of four elements including the two single elements. All four sensors are
non-selective but the best elements are used as “single element” sensors. It can be seen that
almost no selectivity can be obtained with the single elements (ie, S* = l), but that an array
consisting of these two elements and using the vector method already gives an improvement
in the selectivity. This is further improved if three or four sensor elements are used.
If the identification of a substance is achieved, the concentration can be obtained in the
same way as described in Section 11.2.
324 I1 Multisensor Signal Processing

Figure 11-W.
Transfer characteristics of non-selective sensor elements. The
I61 [el hatched areas represent measurement inaccuracies.

2 single
Sensors I sensor array, S
15 I I

Figure U-14.
Selectivity for methane versus carbon mon-
oxide for single sensors [left-hand side, see
Equation (11-8)] and two, three or four sensors
with pattern recognition [right-hand side, see
2 sensors 2 S e m 3s. 4s. Equation (11-7)].

11.4 Partial Least Squares (PLS) Method

The PLS method [l] is a special linear regression method suitable for identification and
measurement of the concentration. The aim is to set up a linear mathematical model linking
the concentration with the sensor signals. Since the model is linear, the application to gas mix-
tures is no problem since the superposition law holds (in the model!).
Let us consider an array of Z sensor elements T, to T, to which the substances X, to X,
are applied. The system is linear if the transfer characteristics are linear and each coefficient
describing the sensitivity for a particular substance is independent of all other gas concentra-
tions. A set of Z linear equations then describes the system. If the number of substances N
is equal to the number of sensor elements Z, the set of equations is determined and the solution
11.4 Partial Least Squares (PLS) Method 325

(the concentrations as functions of the sensor signals) can be obtained by simple matrix in-
version.
Usually N will not be equal to Z and if N < I the system is overdetermined. In this case
a linear model can be found by linear regression methods. A very useful method has been
termed PLS (partial least squares) [5] and, as an example of the excellent results obtainable
for N 4 I, Figures 11-15 and 11-16 show the determination of the protein content in wheat
by means of NIR spectroscopy [14]. In Figure 11-15 the reflectance R is plotted as log (1/R)
versus the wavelength. The two dotted curves correspond to a protein content of 12.9% and
to two different particle sizes as a consequence of different grinding procedures. The two solid
curves correspond to the same particle sizes but to 17.9% protein. A visual inspection would
not encourage an evaluation of the protein content from these data. Figure 11-16, however,
shows that the agreement between the NIR prediction and the actual concentration is ex-
cellent.

Figure ll-15. -m
Spectra of wheat samples with
two different mean particle sizes
(A and B) and two different 0.0t
protein contents (. . ., 12.9% 1000 I400 1800 2200 2600
protein; -, 17.9% protein) [15]. wavelength nm

20 -

18-
‘L
z
I6 -
n
c
-
._
01
0 14 -

Figure 11-16.
Protein content evaluated from 12 -
the NIR spectra plotted versus the
protein content evaluated by / ,
chemical analysis (Kjeldahl pro-
cedure) [15].

The PLS method is very powerful if the transfer characteristics of the sensor elements are
linear and the superposition law holds. The signals of an array of electrochemical cells, for
instance, may be treated in this way.
326 I1 Multisensor Signal Processing

11.5 Transformed Least Square (TLS) Method

In certain cases it may be possible to transform the measurement data and the calibration
data so as to achieve an artificial linear superposition plane. The PLS Method may be applied
to these transformed data. This is explained for the example of metal oxide gas sensors. A
model for the function of these sensors may be a guide for the transformation [6, 151. Let us
consider the analytical expression for the signal of a metal oxide sensor as given by Clifford
and Tuma [16]:

(11-9)

If for each gas component (u, b, . ..) the exponents are the same for all sensor elements, a
transformation can lead to a set of linear equations that is to an artifical superposition plane.
The requirements are

m,, = mzO= ... m a ,


mlb = 11126 = .. . mb, (11-10)

With the substitutions

and

(11-11)

one obtains the following set of linear equations:

(11-12)

The linear PLS method can be applied to this transformed set.


For practical applications, it is first necessary to fit a proper mathematical expression [eg,
a power law, such as Equation (11-9)] to the measurement data [15] in order to find the best
transformation.
Figures 11-17 and 11-18 compare the results obtained with a single sensor element and an
array of four elements using TLS. Figure 11-17 shows the cross sensitivity of sensor No. 2 (type
TGS 812), which is a "CH, sensor". It illustrates the indicated CH, concentration (after cor-
rection for the non-linear transfer characteristic) as a function of the simultaneously applied
11.5 Transformed Least Square (TLS) Method 321

Figure ll-17.
Dependence of measured con-
centration of methane versus
additionally applied carbon 200 - [CHkl = Oppm
monoxide concentration for a 0

- -.--. -
I

500- [CHk] = 500 ppm


Figure 11-18 T
-
~

Dependence of indicated con-


centration of methane versus
200 -
additionally applied carbon
monoxide concentration for a - I I *
[CHkl
#
E Oppm
-

CO concentration. Owing to the calibration, the indicated CH, concentration ([CH,]) is


equal to the applied CH, concentration [CH,] for [CO] = 0. Without cross sensitivity, the
indication would be independent of [CO] (dashed, horizontal line). The actual signal is much
higher and the hatched area is a measure of the error due to the cross sensitivity.
If the proper transformation and the PLS method are used for the evaluation of the signals
of four sensor elements, the results in Figure 11-18 are obtained. There the error is considerably
smaller and the cross sensitivity is reduced by a factor of about 10 compared with the isolated
sensor No. 2.
Figure 11-19 shows the overall improvement achieved by using more than one sensor element
and the TLS method. The quantity Ejkis used as a measure of the relative error in the
328 11 Multisensor Signal Processing

prediction of the concentration of gas j due to the presence of gas k. Here this error (its ab-
solute value) is averaged over all the concentrations used in the measurement (test gas j and
calibration gas k). The term TLS234 is an abbreviation for a sensor array of sensors Nos. 2,
3, and 4, evaluated by the TLS method.

j:CHI,
k : CO

4
! PLS1234 TLS24 TLS 234 TLS 1234
Figure 11-19.
Comparison of the cross sen-
sitivity of carbon monoxide for
the indicated methane concen-
tration for single sensor
elements and arrays evaluated
with TLS.

Owing to the high cross sensitivity of about 60% of the CH, sensor No. 2, the TLS
method gives a considerable improvement even if applied only to two elements. Application
to three or four elements gives a further improvement with a final reduction of the cross sen-
sitivity to about 5%.
Figure 11-20 shows the corresponding results for a CO sensor. The single CO sensor used
shows a low cross sensitivity of only about 6%, but the TLS method still gives an improve-
ment.

1 54 -
% -
j : CO
k : CHI,
42 -
5 k

30 -

Figure ll-20.
Comparison of the cross sen-
sitivity of methane for the in-
dicated carbon monoxide con-
centration for single sensor
elements and arrays evaluated
Sen.4 PLS 1234 TLS 24 TLS 234 TLS 1234 with TLS.
11.6 Discussion 329

11.6 Discussion

Generally, it can be stated that sensor elements which are highly selective neither require
pattern recognition nor are suitable for it.
If sensor elements are non-selective, arrays of them can be used and pattern recognition pro-
vides a better and electronically tunable selectivity.
The four pattern recognition methods described above have been selected because they are
particularly suited for the use with chemosensor arrays. Each one has special advantages and
disadvantages. Table 11-1 shows a rough comparison of these methods.

Table 11-1. Comparison of different methods of pattern recognition with respect to chemosensor applica-
tion.

CM VM PLS TLS
Identification X X X X
Task
Measurement of concentration (X) (XI X X
Single component X X X X
Gas mixture
Arbitrary mixture X X
Linear X X X X
Transfer characteristic
Non-linear X X X

The correlation method and the vector method are primarily suited for identification of
single gas components or gas mixtures with a fixed composition. On the other hand, these
methods can be applied to sensor signals coming from elements with non-linear transfer
characteristics. The evaluation of the concentrations can be done separately in the conven-
tional way after identification as described in Section 11.2.
The PLS method is well suited for identification and measurement of the concentration and
it can be used (in principle) for the investigation of gas mixtures with arbitrary composition.
This is only possible, however, by use of the superposition law applicable to linear systems
only. Most gas sensors, however, have non-linear transfer characteristics and the signals of dif-
ferent substances cannot be superimposed linearly. In these cases it is advisable to look for
an artificial superposition plane, which can be reached by transformation of the signals (TLS).
The transformation is specific for the type of sensors used and the group of substances to be
investigated.
Beyond the above-mentioned requirement for linearity for the application of the PLS
method, there is no specific assignment of specific methods to specific sensor types. The ex-
amples used here have rather been selected arbitarily.
The field of sensor arrays with pattern recognition is fairly young and optimization criteria
have not yet been formulated. As an example, however, it should be evident from the above
that linear operation conditions are desirable, selectivity being of minor importance. Further,
the properties of sensor elements used in an array (eg, drift, aging, and temperature
330 I1 Muitisensor Signal Processing

dependence) should be as similar as possible; the only difference should be the different (non-
zero) sensitivities to the substances to be detected. It can be expected that the application of
pattern recognition will have an impact on the development of sensor elements and arrays.

11.7 References

[l] Sharaf, A., lllmann, D., Kowalski, B. R., Chemometrics, New York: Wiley, 1986.
[2] Ikegami, A., Kaneyasu, M., in: Proceedings of the 3rd International Conference on Solid-State Sen-
sors and Actuators. KO, W. H. (ed.); New York: IEEE Press, 1985, pp. 136-139.
[3] Miiller, R., Lange, E., Hinterstocker, A., in: Proceedings of the 3rd International Conference on
Solid-State Sensors and Actuators, KO, W. H. (ed.); New York: IEEE Press, 1985, pp. 81-84.
[4] Horner, G., Albertshofer, W., Sensorarrays mit nichfselektiven Chemosensoren - Analyse von
Gasgemischen und Verbesserung der Selektivittit, 1988, in: Archiv fur Elektronik und 06er-
tragungstechnik; Stuttgart : Hirzel, 42, pp.85-90.
[S] Wold, S. et al., “Multivariate Data Analysis in Chemistry” in: Chemometrics, Kowalski, B. R. (ed.);
Boston: Reidel, 1984.
[a] Hierold, C., Miiller, R., “Quantitative Analysis of Gas Mixtures with Non-Selective Gas Sensors”,
Sens. & Actuators 17 (1989) 582-687.
[7] Hearst, J. E., Ift, J., Contemporary Chemistry, San Francisco: Freemann, 1976.
[8] Albertshofer, W., Horner, G., Kress, U., Lange, E., “Influence of Zeolite Filter Layers on the
Dynamic Behaviour of PD-MOS-Sensors”, in: Proceedings of the 2nd International Meeting on
Chemical Sensors, Bordeaux 1986, Aucouturier, J.-L., et al. (eds.); Bordeaux: Imprimerie Biscaye,
1986.
[9] Landstram, J., Sens. Actuators 1 (1981) 403-426.
[lo] Miiller, R., Lange, E., Sens. Actuators 9 (1986) 39-48.
[ll] Davenport, W., Root, W., An Introduction to the Theory of Random Signals and Noise, New York:
McGraw-Hill, 1958.
[12] Muller, R., Horner, G., Siemens Forsch. Entwicklungs6e~15 (1986) 95-100.
(131 Stetter, J. R., Zaromb, S., Sens. Actuators 6 (1984) 225-243.
[14] Norris, K. H., in: Food Research and Data Anabsis: Martens, H., Russwurm, H. (eds.); Barking Ap-
plied Science, 1983.
[15] Gall, M., Mliller, R., Sens. Actuators 17 (1989) 583-586.
[la] Clifford, P. K., Tuma, D. T., Sens. Actuators 3 (1982-83) 233-254.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

12 Smart Sensors
JOHN ERNEST
BRIGNELL.
University of Southampton. UK

Contents
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
12.2 What is a Smart Sensor? ..................... 332
12.3 Primary Sensor Mechanisms . . . . . . . . . . . . . . . . . . . . 336
12.4 Enabling Technologies . . . . . . . . . . . . . . . . . . . . . . . 339
12.5 Internal Compensation . . . . . . . . . . . . . . . . . . . . . . 341
12.6 Information Encoding . . . . . . . . . . . . . . . . . . . . . . . 344
12.7 An Early Pilot Study . Smart Magnetic Sensor . . . . . . . . . . . 346
12.8 Integrated Sensors . . . . . . . . . . . . . . . . . . . . . . . . 348
12.9 Industrial Progress . . . . . . . . . . . . . . . . . . . . . . . . 349
12.10 Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
12.11 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
332 12 Smart Sensors

12.1 Introduction

The purpose of this chapter is to outline the philosophy and conceptual development of
smart or intelligent sensors. Obviously the actual implementation of a smart sensor depends
upon a number of detailed techniques, which will not be elaborated here as they are dealt with
comprehensively elsewhere in these volumes. In particular, this treatment depends heavily on
techniques discussed in the other chapters of this volume, notably in Chapter 10, and while
for the sake of intelligibility reference will be made to them, this will only be to sufficient
depth to preserve the integrity of the text. The smart sensor can be a complex system, and
there are a number of potential pitfalls in its design; so emphasis is given here to the general
design principles which will encourage the design of an efficient and reliable product.

12.2 What is a Smart Sensor?

Use of words such as smart or intelligent in application to the products of electronic


engineering is unsatisfactory in many respects, but is just one example of the way that the ex-
isting vocabulary is being stretched by the continuous arrival of new concepts in technology.
We use these terms for want of more accurate descriptions, and their employment in no way
implies that they are to be given traditional interpretations. In particular, it would be futile
to suppose that at the existing stage of technology such terms would suggest a level of achieve-
ment approaching their meanings in human terms. They simply imply that the benefits of
enclosed signal processing have been harnessed to improve the performance of particular
devices, and the consequent intellectual capability on a human scale is still very modest. As
might be imagined there are variations in definitions of a smart sensor, some consider it to
apply only to single chip realizations, while others in contrast restrict it to devices with incor-
porated signal processing.
There are two basic motivations for the development of the concept: firstly, to deal with
the non-ideal behavior of sensor mechanisms, and secondly, to accomodate communication
with the host system. It is the essence of a smart sensor that all its complexities of behavior
are dealt with internally and it presents a simple face to the host system via a digital interface,
although particular current realizations may fall short of this ideal. The need for smart sensors
has arisen from a number of causes. The growing scale of industrial measurement systems
means that the compensation and signal processing problems posed by large numbers of tradi-
tional sensors represent a sufficient load to overwhelm even the most powerful central
processing facilities. The variety of physical and chemical variables which need to be sensed
is continually increasing, so we need to exploit more mechanisms, which will tend to be less
ideal in their behavior, thus implying an even greater processing load.
When, towards the end of the 1970’s, research workers in the field of instrumentation began
to raise the topic of smart sensors, the suggestion was greeted with some scepticism, par-
ticularly in industrial circles. The main points raised by the objectors were cost and complex-
ity. These objections, however, were not entirely valid. Firstly, with regard to costs, the impor-
tant factor is not the cost of a particular sub-system but that of system ownership. Secondly,
12.2 What is a Smart Sensor? 333

insufficient account was taken of the rate of progress of developments in electronics and com-
puter science.
The costs of system ownership are influenced by factors other than the expenditure required
to acquire the various sub-systems, be they hardware or software, which go to make up the
total system. Among the important influences are the following:

- Maintenance
- Down-time
- Reliability
- Fault tolerance
- Fault recovery
- Adaptability
- Commonality of hardware.
In order to develop a methodology which will accommodate these factors in such a way that
a large number of potential systems is realizable, it is necessary to exploit fully the available
technologies of hardware and software. Not least of the recent advances which make these
developments possible is the growth of computer design aids - a process which is continuing
and which will lead to the ultimate development of a comprehensive range of smart sensors.
There is an essential rule of design which characterizes the smart sensor and differentiates
it from the traditional variety:

Sensor complexities must be concealed internally and must not leak out into the host system.
The reason for this requirement is that, as industrial measurement systems become larger,
the total processing load relating to transducer compensation problems can become excessive,
even for the more powerful central computers which are available today. Furthermore, the
transmission of raw unprocessed data will threaten to overwhelm the communication system,
and as the use of analog encoding forces the adoption of the traditional star network topology
the costs and complexity of cabling become a dominant factor.
It follows that all transducer related processing should take place within, or close to, whe
sensor housing. This means that the loading of the communication system is restricted to
what is essential for system operation. Naturally, this must be done in such a way that the
potential performance of the transducer is not downgraded significantly by the presence of
extra processing electronics. Indeed, the investment in smartness is justified by improved per-
formance, though no information can be added that is not already present in the primary sen-
sor signal.
The other key factor in the behavior of a smart sensor is the way in which it communicates
with the host system. This must be such as to enhance the efficiency and accuracy of informa-
tion distribution, while at the same time reducing costs. In this respect we have to bear in mind
the dominant role that cabling can pay in the cost of a large system. The argument therefore
leads inevitably to a design based upon serial binary signalling, which has a number of other
advantages as discussed below. The mode of communication with the outside world is perhaps
the major feature of a smart sensor. The essential sub-systems within a smart sensor are

- A primary sensing element


- Excitation control
334 12 Smart Sensors

- Amplification (possibly variable gain)

-
- Analog filtering
- Data conversion
- Compensation
- Digital information processing
- Digital communications processing.

Figure 12-1 illustrates the way in which these sub-systems relate to each other. Note that cur-
rent realizations of smart sensors may incorporate only some of these elements.
The primary sensing element, which is dealt with more fully in the Volumes 2 to 8, has an
obvious fundamental importance. It is more than simply the familiar traditional sensor incor-
porated into a more up-to-date system. Not only are new materials and mechanisms becoming
available for exploitation, but some of those that have been long known yet discarded because
of various difficulties of behavior may now be reconsidered in the light of the presence of in-
telligence to cope with these difficulties.

Sensor
Hardware
Elenent Process

Input
Var lables
Monltorlng

..........j

Conns. Process

.........................................................
4 b
Figure 12-1.
BUS The essential elements of a smart

Excitation control can take a variety of forms depending on the circumstances. Some sen-
sors, such as the thermocouple, convert energy directly from one form to another without the
need for additional excitation. Others may require fairly elaborate forms of supply. It may,
for example, be alternating or pulsed for subsequent coherent or phase-sensitive detection. In
some circumstances it may be necessary to provide extremely stable supplies to the sensing ele-
ment, while in others it may be necessary for those supplies to form part of a control loop
to maintain the operating condition of the element at some desired optimum. While this
aspect may not be thought fundamental to smart sensors there is a largely unexplored range
12.2 What is a Smart Sensor? 335

of possibilities for combining it with digital processing to produce novel instrumentation


techniques.
Amplification of the electrical output of the sensing element is almost invariably a require-
ment. This can pose design problems where high gain is needed. Noise is a particular hazard,
and a circumstance unique to the smart form of sensor is the presence of digital buses carrying
signals with sharp transitions. For this reason circuit layout is a particularly important part
of the design process.
Analog filtering is required at minimum to obviate aliazing effects in the conversion stage,
but it is also attractive where digital filtering would take up too much of the real-time process-
ing power available.
Data conversion is the stage of transition between the continuous real world and the discrete
internal world of the digital processor. It is important to bear in mind that the process of ana-
log to digital conversion is a non-linear one (in mathematical terms it is a many-one mapping)
and represents a potentially gross distortion of the incoming information. We traditionally
skirt the difficult implications of a non-linear system by dealing with it in terms of aliazing
and quantization noise theory, which in general serve the purpose. It is important, however,
for the smart sensor designer always to remember that this corruption is present, and in certain
circumstances it can assume dominating importance. Such circumstances would include the
case where the conversion process is part of a control loop or where some sort of auto-ranging,
overt or covert, is built in to the operational program.
Compensation is dealt with later in this section. Suffice it to say at this stage that it is central
to the philosophy and reason for the existence of the smart sensor: also its needs may affect
the basic design of the system, as exemplified by the presence of the monitoring line in
Figure 12-1.
Information processing is, of course, unique to the smart form of sensor. The range and
variety of techniques which can be employed are far too broad to be treated here and many
have been discussed in detail in Chapter 10, but it is worthwhile to summarize the general aims
of using it in this application. There is some overlap between compensation and information
processing, but there are also significant areas of independence.
An important aspect is the condensation of information, which is necessary to preserve the
two most precious resources of the industrial measurement system, the information bus and
the central processor. Sensor compensation will in general require the processing of incoming
information, and in some circumstances will represent the major processing task. The smart
sensor, to some degree, can be responsible for checking the integrity of its information;
whether, for example, the range and behavior of the incoming variables is physically
reasonable.
It is a trite point, but nevertheless an important one, that the information processing stage
cannot create information. It can, however, destroy information or introduce false informa-
tion. This must be regarded as a major hazard in smart sensor design, as it is so easy to insert
a process realized intuitively in software which may not be fully understood. Anyone involved
in the real time programming of discrete information should have a working knowledge of
discrete signal theory. The potential for the destruction of information is sufficiently impor-
tant to be picked up again in Section 12.6.
A final, but extremely important, element is communicationsprocessing. It is so important
that it requires a processor of its own, though this may be realized as part of the main pro-
cessor chip. The natural form of communication for the smart sensor processor is the multi-
336 12 Smart Sensors

drop bus, which can produce enormous cost savings over the traditional star network. It has,
of course, fairly obvious vulnerabilities. These can be reduced by forming the bus into a loop
which can be addressed from both ends, thereby not only increasing the integrity but also pro-
viding a means of fault location. In severe circumstances, where there is, for example an explo-
sion hazard, the bus may be duplicated with a substantial physical separation and devices con-
nected on spurs. A most important attribute of the smart sensor concept is addressability,
which is of course essential to the multi-drop principle and a powerful aid to the logical
organisation of sensor systems, but it does introduce limitations. Addressability implies some
form of polling of the devices, and though this may be prioritized in various ways, it does
imply a constraint on the response time of the system to changes at any particular sensor site.
In critical circumstances it may be necessary to provide some form of interrupt mechanism.
A major contribution of intelligence is the integrity of communication. The transmission
process can be protected by various forms of redundant coding, of which parity checking is
the simplest example. In crucial applications information can be double checked by means of
a high level handshake dialog, in which the central processor asks for the information and
then returns it to the sensor for confirmation. This deals with almost every possible fault ex-
cept where the sensing element, though behaving apparently reasonably, is wrong. In such a
case the only cure is the triplication of sensor elements, or in the extreme the triplication of
smart sensors.

12.3 Primary Sensor Mechanisms

As stated above an essential reason for the development of the smart sensor concept is deal-
ing with the problems created by the available primary mechanisms. Physical variables in the
real world tend to appear in pairs [l]; potential (or across) variables, and flux (or through)
variables; distinguished by obeying a zero path integral law or a point continuity law. Their
product tends to represent an energy or power while their ratio tends to represent an im-
pedance.
Examples of pairs of physical variables are

- voltage and current


- pressure head and fluid mass flow
- temperature and heat flow
- velocity and force,

and there are many others.


There are two reasons why this rather academic concept is important, which are familiar
from our experience with electrical devices. They are impedance matching and loading. Im-
pedance matching can be important in some sensing applications, but it is not in general as
serious a problem as loading. The presence of the sensor always to some extent alters the field
of physical variables in which it is placed. Thus a temperature sensor will have a thermal mass
of its own, and will require a flux of thermal energy each time an equilibrium is re-established
with its surroundings. For this reason temperature sensors, and indeed most other forms of
12.3 Primary Sensor Mechanisms 337

sensors, will be required to be physically small, so that the interference with the test environ-
ment is minimized. It is possible to use the processing power of the smart sensor to overcome
some of these problems; for example, a thermal element of a known response can be corrected
by means of a simply derived predictive algorithm. This must, however, be performed with
some care, as the corrective treatment is in effect a high-pass filter, which is likely to emphasize
noise.
A basic transducer is an element which transforms changes in one of these pairs of physical
variables into changes in another pair. Sometimes there are attempts to make nice distinctions
between transducers as devices which convert energy and sensors as devices which convey in-
formation, but as there are no cases in which information is transferred without the transfer
of energy, the distinction is largely artificial. Mathematically the link between two pairs of
variables is a 2 x 2 matrix. Physically this matrix is almost always realized by resort to the
properties of materials. These ideas are illustrated in Figure 12-2 where a compound sensor
is modelled as cascaded subsystems or matrices. Such representations are often gross
simplifications, especially when non-linearities are present, but they demonstrate the potential
complexity of the sensor as a compound system, and particularly that there is a complicated
inverse problem to solve if our smart sensor is to restore a good representation of the target
variable.

Figure 12-2.
Two forms of model for the compound primary
sensor element - cascaded subsystems and linear
matrices.

Materials in their basic forms (atoms, molecules, crystals etc.) change their behavior in
response to the impact of various external stimuli (fields, strain, thermal agitation, other
material etc.). Thus the primary sensor element takes the form of an assemblage of materials
with an electrical output connection, organized in such a way that the response to a specific
physical variable is maximized, and the response to all other physical variables is minimized.
A simple element may in fact embody a chain, sometimes quite complex, of transformations
between pairs of variables. For example, a wire carrying an electric current, through the
transformations of flow, thermal and electrical variables, becomes a basic anemometer.
Ideally, each of these transformations is linear and time invariant: practically it never is. Hence
even the conceptually simplest of primary sensor elements presents problems of control and
information processing. These concepts are important, as one of the basic functions for the
smartness in the sensor is the realization of inverses of the transformation characteristics.
We see that at the heart of each sensor is a block or blocks of material which provide the
gateway between the target pair of variables and the utilizable pair of electrical variables. Since
electric circuits are now largely realized in silicon, the properties of silicon assume a particular
338 12 Smart Sensors

importance in the search for primary elements. The sensitivities of silicon are many and
various [2]. Silicon will change its electrical behavior in response to

- electric and magnetic fields


- mechanical strain
- temperature
- radiation
- and the presence of other materials.

This embarras de richesses makes silicon a promising but formidable basis for a primary
sensor [3]. Its multiplicity of sensitivities make the task of designing to optimize one of them
a particularly difficult one, which accounts for the relative paucity of available silicon sensors.
Nevertheless, the prospect of combining the sensor element with the processing elements on
a single chip is a powerful motivation to pursue further research and development in this area.
Not the least of the attractions of silicon is its very favorable mechanical properties combined
with the ease of shaping it. ..

Invariably a piece oE material in a particular physical form is used to provide a bridge be-
tween the sets of physical variables. Some of these forms are relatively novel (micromachined
silicon [4], optical fiber [ 5 ] etc.), and new types of materials are continually emerging (high
temperature superconductors, organic semiconductors, supermagnets, etc.). Nevertheless, the
appearance of new principles is a comparatively rare event. One of the significant interven-
tions made by the intelligent sensor concept is that many of these principles which were once
discarded as being too difficult may now be re-addressed in the light of new techniques of
compensation.
Many sensor elements which were cumbersome when fabricated by traditional technologies
have become more attractive in the light of new technologies. A good example is the vibrating
element, which can be sensitive to such variables as strain, environmental pressure and
viscosity, temperature, etc. Constructed in older technologies these tended to be large and
cumbersome, operating at low frequencies (and therefore slow-responding) and fragile. New
techniques, particularly the micromachining of silicon, have suddenly made these available as
small, robust, high-frequency devices. This is especially convenient as timer-counters are easily
realized as electronic circuits.
Optical fibers represent another new technology which has had profound implications in
sensing. Their large degree of freedom from interaction with electromagnetic fields makes op-
tical fibers the ideal communication medium, and variation of their structure turns them into
various forms of intrinsic sensors. Semiconductors can be adapted to provide both input and
output opto-electronic devices, which means that the intelligent part of the sensor can be
physically separated from hostile environments. The development of a low-loss, cheap T-junc-
tion would provide the missing ingredient to make the optical fiber a dominant constituent
of distributed instrumentation systems.
One of the n ' .. s t important areas of sensing from a requirements point of view is chemical
sensing. Chemical hazards are increasing and legislation is trying to cope with this. There are
many basic types of chemical sensor (chemiresistors, fuel cells, ChemFETs, solid electrolytes
[6] etc.), but they all tend to suffer from common disadvantages. Important among these are
lack of specificity, short life and susceptibility to poisoning. The intelligent sensor approach
can often mitigate these problems. Sensor arrays can overcome the specificity problem, while
12.4 Enabling Technologies 339

adaptability of electronic systems can overcome ageing effects. The ChemFET suffered
somewhat from the overenthusiasm of its proponents early in its life, but it is certain even-
tually to be an important primary sensor.
The subject of primary sensing elements is large and covered elsewhere in these volumes.
In the smart sensor context it is sufficient to remember a few of the global characteristics men-
tioned above. They have a two way interaction with their environment, they exhibit a number
of imperfections of behavior, which present us with a significant inverse problem to solve, and
they are constrained by the materials and fabrication technologies available.

12.4 Enabling Technologies

The concept of the smart sensor owes its very existence to the rise of the various
technologies which support it. These are developing at such a rate that any description is
simply a snapshot which is out of date almost as soon as it is written. The basic engine of
this change is the rapid evolution of electronic hardware, particularly in its dramatically falling
size and cost. This has, in turn, generated a further phenomenal growth in software techni-
ques. The latter not only contribute by easing the task of programming the intelligent devices,
but also create a significant impact by providing design aids, which render manageable tasks
of potentially overwhelming complexity. It is quite remarkable that all of these developments
have occurred within a decade of computer-aided measurement appearing as a recognizable
discipline [7].
The basic unit of electronic circuits is the transistor, either bipolar or field-effect, which
combined with resistors and capacitors goes to make up the sub-units of our integrated cir-
cuits. The properties of the transistor (speed, power, size, noise, etc.) control the overall
characteristics of an integrated circuit, and various technologies coexist because of the variety
of requirements that occur. A point of particular relevance to the present context is the differ-
ing requirements of discrete and continuous systems which generally both occur in smart sen-
sors. The continuous part of the system requires to be highly linear with good noise immunity,
while the discrete part needs to be fast and compact. Thus at the present state of practice it
will usually be the case that chips of more than one type of technology will be required, so
the techniques of interconnecting such chips assume a great importance. The discrete part of
the system tends to be a generator of noise, while the continuous part is sensitive to noise,
so combining them can be a tricky part of the design process.
In the days before programmable electronics, all the variability of design had to be catered
for in hardware, which made the design/construction loop very cumbersome and expensive.
The appearance of the microprocessor meant that much of the variability of design could be
accommodated in software, and considerably more complicated concepts could be im-
plemented. The subsequent appearance of the ASIC has again moved the design frontier in
the hardware direction. Nevertheless, with certain exceptions, such as the motor industry, pro-
duction runs of sensors tend not to be very large, so the use of standard hardware devices is
indicated wherever possible.
At the outset the costs and complexities of software design were grossly underestimated, but
the subsequent growth of software design aids greatly improved matters, and now the in-
340 I2 Smart Sensors

troduction of similar design aids into the hardware area has a great influence on the cost and
reliability of designs. In software we have not only the advantages of assemblers, compilers,
libraries, etc, but also programs to verify the logical integrity of the programs we generate
Similarlyin hardware more and more powerful means of design automation are supported by
design rule checkers and simulators of improving accuracy, so that the overall integrity of our
designs can be maximized before the productions stage
Interconnection technology is particularly important in smart sensor realizations. The
technology is required to be compact, robust, cheap and amenable to design automation
methods. These things considered, there is a powerful case for thick film techniques to be
used, at least at the present state of technology. The design process is one of shape manipula-
tion, which can be obtained by a relatively simple adaptation of the tools which are already
available for integrated circuit design. The masks can be directly photo-plotted from the com-
puter and made cheaply, while the manufacturing process itself is the inexpensive one of silk
screen printing. The various ink layers (resistor, conductor, dielectric) are deposited onto a
substrate, usually alumina, and the end result is a structure which exhibits all of the above re-
quirements. Furthermore thick film elements can be important as the supporting structures
for primary sensor material, and even as the primary sensors themselves [8] (see also
Chapter 6).
The basic design tool for both hardware and sof€wareis the personal computer with the ad-
dition of a high resolution graphics capability and a pointing device, such as a mouse or
tracker balI. With the addition of appropriate software tools the whole smart sensor can be
designed and checked. Figure 12-3 shows the mask design of a typical sensor support structure
produced by a CAD package The availability of processes such as copy, translate, rotate,
group, etc makes the generation of such a design a relatively simple matter. The figure also
illustrates the important point thdt sensor structures can be easily realized in thick or thin film
form and produced as arrays of any required multiplicity. The importance of the availability
of sensor arrays in the smart sensor context is discussed later in this section.

Ngure 12-3.
A typical sensor structure created by
CAD and realizable in thick or thin
f i i An array of iuterdigitated
electrodes (designed for Michell
Instruments by N. White).

We cannot leave the topic of enabling technologies without a discussion of the single chip
solution. This has become something of a holy grad among research workers and its attrac-
tions an self evident. There are, however, major problems to be solved before it approaches
12.5 Internal Compensation 341

reality. One has been referred to in Section 12.3, the multiplicity of silicon sensitivities.
Another is the fact that different silicon integration technologies lend themselves to different
application areas: for example, a high performance operational amplifier would impose quite
different requirements from a fast communications processor, yet both are needed in many
smart sensor applications. However, newer processes such as BiCMOS give the opportunity
of realizing both functions on one chip. Thirdly, there are problems over the interaction of
various process stages which might be desired in combination, eg, the etching of mechanical
structures and the fabrication of circuits. Nevertheless, there is a strong motivation towards
such solutions, and some examples of progress are reviewed in Section 12.8. Generally speak-
ing, the monolithic solution will be preferred when either (a) the cost can be reduced (eg, by
on-chip automatic trimming) or (b) performance requires it (eg, in an optical sensor array).

12.5 Internal Compensation

The requirement that complexities should be dealt with by the smart sensor internally can
lead to a significant reduction in the amount of processing being carried out at the central
point. It is useful to list some of the main sensor defects which may have to be accommodated
PI.
- Non-linearity
- Noise
- Pick-up (interference, crosstalk)
- Frequency (time) response
- Drift
- Cross-sensitivy.

Each of these can be dealt with by various methods which are described elsewhere in these
volumes and in the literature [9]. Many such methods have to be tailored to a particular cir-
cumstance, but also many of them can be generalized. Occasionally the implied processing
load can be too great to be accommodated by current technology (eg, wide band frequency
response correction) but the great majority of problems met in sensor compensation imply
relatively modest requirements of processing power.
In general the compensation problems arise from the nature of the primary sensing element.
This will exhibit behavior conditioned by its structure and content which is usually describable
in systems theory terms. Non-linearity was one of the most serious defects of sensors in the
days of linear continuous electronics, but is now relatively trivial, as it can be dealt with by
look-up-table, curve fitting, etc. This capacity has led to a great enlargement of the range of
sensor mechanisms that can be utilized.
Because the universe is in a state of thermal agitation, noise is always present, but there are
important sources apart from the thermal one. The process of dealing with noise is very con-
text dependent. Noise, including pick-up, is any unwanted signal, and is thus defined with
respect to the wanted signal. Other factors which affect the choice of treatment are the nature
of the information required from the signal and the degree of manipulation which can be
342 12 Smart Sensors

tolerated. Thus we may select methods such as filtering, signal averaging, and correlation as
appropriate, but the problem can only be mitigated to a certain extent and not entirely cured.It
must always be remembered that the process of data conversion itself is effectively a form of
noise, as are other corruptions such as rounding and truncation errors.
The term noise is often considered to imply randomness, but coherent forms of unwanted
signal are also likely to be present. External electromagnetic fields will induce unwanted si-
gnals within an inadequately screened structure, and the industrial environment can be ex-
tremely hostile from this point of view. As mentioned above, the smart sensor itself can be
a source of noise, as the digital part of the system will generate fast edges which are easily
coupled to the input stages by stray mutual reactances.
The primary sensor element will exhibit a time response which is non ideal, and if such a
response occupies the same time or frequency frame as the target measurement some form of
compensation is necessary. The response arises from the existence within the element of
storage and dissipative elements analogous to resistance and reactance, though these may ap-
pear in the guise of complicated physical processes such as diffusion. Thus the coefficients
in Figure 12-2 exhibit frequency dependence, which is represented in the time domain by the
process of convolution, and an essential element of the smart sensor approach is the
availability of the inverse operation of deconvolution.
Many of the classical sensors were deliberately given a single-pole low-pass filter
characteristic (i e, damping) to overcome problems of noise and oscillatory tendencies, but this
is a relatively crude device, and the smart sensor approach offers more delicate tools.
The essence of the frequency compensation process is the realization of a filter whose
characteristic is the inverse of that of the sensor element (a pole for every zero and a zero for
every pole). One of the great benefits of digital filters is their ease of realization in comparison
with classical continuous methods. It must be remembered, however, that digital filters are
very hungry for real time processing power, so bandwidth requirements are very critical in
determining whether the methods can be used. If the incoming signal can be dealt with in
blocks in non-real time it is possible to use mathematical methods of deconvolution, such as
the Fast Fourier Transform; but as always there are concomitant obstacles which have to be
dealt with, such as the window problem.
Drift is also ever present to some extent. It may be caused by slow changes in the physical
parameters of the primary element (eg, by ageing or oxidation) or it may be due to changes
in the physical variables (eg, the leakage of charge). One of the most difficult forms of noise
to deal with, l/fnoise, also manifests itself as apparent drift. Indeed, drift is a form of noise
and the constraints on dealing with it are the same, in that they depend on how the desired
signal is defined. For example, in the counting of objects by weight the wanted signal is
discrete, so it is relatively easy to develop an algorithm to distinguish this from the continuous
drift component by updating a stored offset [17, 181.
Without doubt the most important of sensor defects which have to be countered is cross-
sensitivity. No sensor element is responsive to one physical variable only. A major component
of the art of sensor design has always been the maximizing of the desired sensitivity and the
minimizing of undesired ones. The smart sensor approach greatly increases our armoury of
techniques to combat the problem. The most important among the various potential cross-
sensitivities is that to temperature and it is safe to assume that it is always present.
It is not the purpose of this section to deal with specific compensation methods as these
have been considered in detail in Chapter 10, but it is useful at this stage to classify the various
12.5 Internal Compensation 343

compensation methods made available by digital methods. One form of classification is in


four groups [9], namely

- Structural compensation
- Monitored compensation
- Tailored compensation
- Deductive compensation.

It goes without saying that the physical design of the sensor is approached with the inten-
tion of minimizing the defects mentioned above. Nevertheless, this immediately implies a com-
promise, since any perturbation of the design to favor one constraint is likely to prove disad-
vantages in some other respect. The most basic tenet of structural compensation is embodied
in the principle of design symmetry. The objective of design symmetry is to make the desired
physical variable produce a differential output while all other variables, including interference,
produce a common mode output. The classical example is the strain gauge bridge. The princi-
ple of structural compensation is just as important in modern varieties of sensor, and
especially silicon sensors, where the significance of cross-sensitivity can be major.
When all efforts have been made to exploit the advantages of structural compensation there
is inevitably a residual effect which requires correction, and as this is in the form of an error
in the structural component it is less predictable and less well behaved than the systematic
defects that have been dealt with up to this point. Thus, if there is the need to obtain further
compensation it has to be performed on the individual sensors as they come off the produc-
tion line. An important feature differentiating this tailored component from the structural one
is that it contributes to the variable costs of production rather than the fixed (design and
development) costs. It can therefore be a major factor in determining the selling price of the
sensor, particularly if skilled labor is involved. The smart sensor philosophy, however, would
dictate that this tailoring process should itself be automated in the form of the measurement,
calculation and storage of appropriate coefficients.
The most obvious way to deal with cross-sensitivity in an intelligent sensor is to measure
the unwanted variable and remove its effect by calculation. There are two variations of princi-
ple by which this monitored compensation could be applied. The traditional method would
involve a model of the sensor behavior which would give rise to a compensation algorithm.
More attractive in many ways is the use of model-free methods, in which the system acquires
the compensation information by a process of learning or optimization [9]. There is a strong
element of tailoring introduced, which reduces the risk of preconceptions being made incor-
rect by small changes in the production process.
There are two basic forms of sensor combination which provide for monitored compensa-
tion. The first is the sensor within a sensor, while the second is the sensor array. The former
structure is often found in silicon devices, where it is relatively easy to fabricate one of a
variety of temperature sensitive structures as a minor part of the device. The second form, the
sen’sor array, is one in which a number of sensor structures are created on a common substrate.
A prime example is in gas sensing, where a number of conductive films, or the same film at
different temperatures, will have different responses to a variety of impinging gases, so that
with luck the actual gas content can be determined by solving a set of (not necessarily linear)
simultaneous equations, as outlined in Chapter 11.
344 I2 Smart Sensors

It is not always the case that the conditions at the desired sensing point are physically ac-
cessible, and the classical application area would be medical sensing. Often in this cir-
cumstance one is able to make deductions from the behavior of external variables by reference
to a physical model: eg, the internal temperature of a device may be estimated from the exter-
nal temperature and the power supplied to it. Deductive compensation is only used as a last
resort, as all models are imperfect, often to an extent which is unknown. Again, however, this
form of compensation is unique to digital methods as a certain amount of information pro-
cessing is a prerequisite.

12.6 Information Encoding

The encoding of information is obviously of great importance in the design of smart sen-
sors. Externally the serial binary method is dominant, as it offers cheapness, robustness and
versatility. Internally a wide varity of encoding methods is available.
The sensor is a link in the chain of information flow between the external physical source
and the host processor. Information, like any physical quantity, is subject to laws of con-
tinuity. The generalized law can be written by considering all the possible things which could
happen to a quantity. This is illustrated in Figure 12-4 which takes account of the possibilities
that a quantity, say Q,with respect to the volume within the closed surface, can enter, Q,,
leave, Qo, be stored, Qs, unstored, Qu, destroyed, Q, or created Qc. Many basic laws of vec-
tor fields derive from the simple statement that the algebraic sum of the rates of change of

QO
Figure 12-4.
Illustration of all the things which can
d , - d o + d, - a, + d, - d, = 0 happen to a quantity with respect to a closed
surface. Information within a smart sensor
obeys this rule with the exception that it
cannot be created.
12.6 Information Encoding 345

these quantities must be zero (eg, Gauss’ theorem), but the idea can also be usefully applied
to information.
With reference to any closed surface, which we might imagine as the physical boundary of
our sensor system, information can enter, exit, be stored or unstored and, most importantly
it can be destroyed. The algebraic sum of the rates at which these information quantities
change is zero. Note that we do not include creation in the equation. The question of whether
and how information can be created is a deeply philosophical one, but it is certain that a
limitation to the smartness or intelligence of our sensors is well below this level. However, our
sensors can and do create disinformation, or noise, and this occurs at virtually every stage of
information processing. The integrated form of the equation in Figure 12-4 when applied to
information gives an important relationship

and the constant is the amount of information contained on initialization, normally zero.
Although based on extremely simple premises this equation and its differentiated form are im-
portant design equations for the smart sensor system. The balance must always occur,
whatever the constraints on the quantities. Q, is constrained by the measurement strategy
(eg, the sampling rate), Qs - Q, by the amount of storage provided in the smart sensor,
while Qo is a stochastic variable determined by the amount of bus access available to the
device. If the algebraic sum of these quantities becomes positive then Q, becomes positive,
and information must be destroyed. This may seem an over-elaborate way of describing buffer
overflow, but thinking in terms of these equations has two favorable effects on the design
discipline. Firstly, by applying worst case values to the variations in these rates of information
flow one has a means of calculating the size of buffer store required for a particular cir-
cumstance [19]. Secondly, one is forced to take account of the possibility of buffer overflow;
and in a real time data collection system this is something that has to be built into the pro-
gramming of the sub-system. It is, of course, important not to confuse the information
capacity of a system with the information content. Although both are measured in the same
units (bits) the latter is always less than the former.
The information carrying capacity of a distributed measurement system is conditioned by
a number of factors - the bandwidth of the communication channel, the degree of
parallelism in the system, the level of integrity required etc. As in any system of flow, the rate
at which information can be handled is controlled by the narrowest bottleneck through which
it has to pass. Almost invariably this bottleneck will occur at the input to the host computer,
and there is little point in devising an instrumentation system which exceeds this rate.
In the traditional instrumentation system sensors are connected to the host in a star
topology, typically using 4-20 mA encoding. This is a highly parallel, low bandwidth system.
It is profligate in cabling, hence the need for commonality of power and signal wires, and there
is a complicated situation where all the cables arrive at the hub of the network where they
have to be scanned for input to the computer.
As suggested earlier, the network most attuned to the capabilities of a number of smart sen-
sors would comprise a multi-drop, two-wire, serial bus, formed into a loop so that it can be
addressed from either end. Power should be provided separately. Although there are demands
for the continuation of the two wire tradition, the penalties in capability and performance do
not merit its continuance, and the constraints which produced it no longer exist as a restriction
346 I2 Smart Sensors

on the overall system design. There is a case, however, for retaining the two-wire option at a
local level when physical conditions warrant it.
We see that information starts out in the form of a continuous variation of the target
physical variable in the real world and ends up as a stream of binary digits, which though cor-
rupt is an adequate representation of the required information content. Subject to the con-
tinuity law, the information may be encoded in a variety of ways in between [2]. Within the
smart sensor itself there will be two or more forms of encoding. A common form of internal
electrical encoding would be as a continuous voltage within a prescribed range (so-called ana-
log encoding). Via an A/D converter this is transformed into parallel digital encoding.
This is by no means the only form of intermediate encoding found in smart sensors. Fre-
quency, for example, can be a useful parameter upon which to impose the information, and
many interesting new silicon sensors are based on oscillating elements, while many forms of
sensor based on variations of reactance lend themselves to inclusion in oscillator circuits.
Timing and counting are particularly easy to implement in digital electronics, and there is the
added attraction of a two wire or single fiber connection. In the example discussed in the
following section frequency difference encoding is put forward as a valid solution.
Yet another form of encoding which can be useful in certain circumstances is in the form
of control of the mark-space ratio of a square wave. This has the advantages of being a binary
signal and therefore immune to non-linearity, while again taking advantage of the timing
capability of the electronics. It is not, however, immune to noise, as edges are just as prone
to corruption as amplitudes, and the trade-off between precision and time can cause dif-
ficulties (the time required increases exponentially with the number of bits of information).
Furthermore it represents a very poor utilization of the bandwidth of the channel.
The selection of encoding method at the various points in the information chain is condi-
tioned by a number of factors, and it needs to be considered with care at the outset.

12.7 An Early Pilot Study - Smart Magnetic Sensor


It is useful at this stage to examine one of the first attempts to produce a smart sensor. The
work was carried out in the author’s laboratories in the early 198Os, and is now rather
primitive, but it does illustrate very well some of the principles established above [20]. Firstly,
it was necessary to select a sensor mechanism which would enable the design problems of
smart sensors to be investigated without the introduction of any irrelevant complications. The
magnetic Hall effect was chosen, as it is a bulk effect in silicon and does not therefore suffer
from some of the difficulties associated with surface effects.
The first design problem one encounters is the structure of the primary sensing element.
A simple Hall plate is a good structure, as it is basically symmetrical, but after some con-
sideration a slightly more complicated structure was selected. This is the split drain transistor.
If a simple field effect transistor is provided with two drains side by side the imposition of
a magnetic field will upset the current balance between them. By providing suitable loads in
the form of transistors with their slope characteristics optimized at the equilibrium operating
point, we are able to achieve current to voltage conversion with a high transfer ratio. Thus we
have achieved a compact symmetrical structure converting magnetic field strength to a voltage
difference, while other physical variables produce a common mode signal.
12.7 An Early Pilot Study - Smart Magnetic Sensor 347

At this stage it is important to point out that structural compensation only gives a first order
correction for cross-sensitivity with, say, temperature. No real structure is truly symmetrical
and any variation of the transfer ratio with temperature will not be dealt with by this means.
The next essential stage is amplification. A first reaction might be to add a differential input
single amplifier. However, the amplifiers themselves in the chosen technology (standard
NMOS) are temperature sensitive, so the design symmetry can be extended by providing two
identical amplifiers.
Now we reach the problem of encoding the information in a form acceptable by the digital
part of the system. Conventional A/D converters were not available on-chip at that time, no
doubt an integrating type could have been attempted, but on consideration of the overall
system a better solution emerged. The digital part of the system had already been selected,
and it took the form of an 8751 UV erasable micro controller. This has particularly good in-
dependent timedcounter facilities, so frequency encoding was an attractive option. Trials with
various forms of voltage controlled oscillator (VCO) showed that these too were temperature-
sensitive and exhibited non-linearity. By duplicating the VCOs we extend the scope of the
design symmetry and have a convenient form of encoding the information in the form of a
frequency difference. At this stage we have exploited fully the possibilities of structural com-
pensation, but are left with a residual temperature dependence and a non-linearity. The ob-
vious next step is to resort to monitored compensation. In the trial chip various temperature
measuring devices were included, but because it includes the encoding of the information a
third VCO was found to be the most convenient. A small amount of interfacing logic and the

r
- - - - - - - - ,
I
I
custon chip
- I
I D

1
vco I 4
I I
I 1 b

I Sensor
I
I e I ene

I
I
ITrans- I
I
I
I
1
L

Figure U-5. Outline of the final design of the smart magnetic sensor. Points to note are the symmetry
and the presence of a monitoring line as in Figure 12-1.
348 12 Smart Sensors

Flgwe 126. Photograph of the complete smart magnetic sensor. The custom chip is seen at bottom left.
The microprocessor was only available in a large package at the time.

micro controller all mounted on a substrate and interconnected to the special chip by thick
film techniques completed the smart magnetic sensor (Figures 12-5 and 12-6).
Some interesting principles emerge from the calibration process. The most important of
these is that it is not necessary to measure the temperature as long as the device is taken
through the required range and allowed to equilibrate at a known field. It can then fill up its
own look-up table using internal units of temperature (in this case a frequency count). The
non-linearity is dealt with at the same time. We end up with a two dimensional look-up table,
the graphical equivalent of which is a surface plotted above orthogonal axes of frequency and
frequency difference, that surface representing the c o w e d magnetic field.
Thus we finish with a device that, once taken through an automated calibration cycle, is
capable of reporting the value of magnetic field in ASCII characters on a serial bus in
response to a request to its own particular address, and while a little primitive it illustrates
nicely the principles and characteristics of the smart sensor. Incidentally, the device proved un-
suitable for further development, as the primary sensor structure proved to be particularly
prone to l/fnoise, but in the sense that it threw light on the problems of smart sensor design
it represented a successful project.

12.8 Integrated Sensors

As stated early in this chapter, the definition of smart sensorsvaries. In the above discussion
this has largely been restricted to sensors with digital processing incorporated. The develop-
ment of silicon technology is rapid, and it will eventually be true that such a definition will
be consistent with a single chip realization. Meanwhile there has been a number of
developments in the direction of placing a sensing element and some of the support electronics
on a single chip. The special chip in the above example is typical of a class of integrated sensor
elements with frequency output. There is a variety of ways in which such an output can be
generated (VCOs, ring oscillators, flip-flops, etc. (111).
12.9 Industrial Progress 349

Even the small step of adding buffer amplifiers can considerably enhance the utility of sen-
sor elements. An example is the realization of chemical sensor electrode arrays on silicon [12].
Unity gain buffer amplifiers provide both impedance transformation and bootstrap shielded
line coupling so that isolation can be established between the electronics and the chemically
exposed areas.
In capacitive pressure transducers on-chip electronics can provide the excitation, the
capacitance bridge and the output buffer amplifier [13]. The degree of miniaturization ob-
tainable is impressive, and a micro-diaphragm pressure sensor with incorporated encoding
electronics can be made small enough for insertion into a 0.5 mm catheter [14].
Another important application of on-chip electronics is to the scanning of sensor arrays.
This has been applied to both photodiode arrays [15] and microdiaphragm pressure sensor ar-
rays. In the latter case 32 x 32 arrays have been realized in the construction of a tactile sensor
for robotic applications [16].

12.9 Industrial Progress

So far in this section we have examined the smart sensor concept from the point of view
of the research and development engineer, ie, we have been concerned with what is scien-
tifically desirable and technically feasible. The industrial situation is very different. There is
a massive investment in existing technology, which is of proven reliability, and the major
customers for sensors tend to be highly price conscious (like the motor industry) or highly con-
servative (like the process industry). The other problem is that industry needs standards. It
is not appropriate here to discuss the ramifications of the debate about fieldbus standards.
Suffice it to say that it is a highly political issue and there exist a number of possible can-
didates all with excellent characteristics. Thus the industrial approach to the problem tends
to take the form of cautious small steps out from the established technology with which there
is an attempt to preserve compatibility.
One approach is to establish a technology which is independent of the individual sensing
problem, but provides a step towards the establishment of standards which can be absorbed
into smart sensors in the future. This approach is exemplified by a system developed in the
author’s laboratories. The approach here is to provide a number of powerful system com-
ponents, realized as circuit boards or complete software entities, which can be combined in
various ways to produce a powerful distributed instrumentation system. The hardware com-
ponents are an intelligent bus driver, which resides in a personal computer, a versatile remote
intelligence unit with a wide variety of I/O facilities and an intelligent bus repeater, which
enables extension of the network over several km. The software components allow easy access
to the hardware facilities so that remote units can be simply addressed with all communica-
tions problems transparent to the user. The bus standard is a variation of the well established
SDLC protocol. Figure 12-7 shows the skeleton of a typical remote intelligence unit. Important
points to note that all analog functions (signal selection, gain, and offset) are under control
of the digital program, and through that of the bus: also offset correction is applied before
data conversion, so that the offset does not occupy the valuable resource of the dynamic range
of the A/D converter [18]. An equally important component of an overall system is the soft-
350 12 Smart Sensors

V 8

-
out In
v)
W

TL"+
8 blt 0
ref DA * 0
a
CL

U
I I I ?
# -

m I I
I

Systan
ROM

= R A M

User

BUS

Figure U-7. Example of a comprehensive remote intelligence unit, which enables the processing func-
tions of a smart sensor to be carried out.

ware, and it is imperative that the user is protected from the complications of low-level pro-
gramming of a relatively complex system such as that shown in Figure 12-7. The means of
doing this is a local operating system residing in the system ROM of Figure 12-7. Figure 12-8
shows in a conceptual form how this is realized so that the internal complexities of the hard-
ware are transparent to the user, who simply makes high level calls to the operating system
for whichever hardware facility he wishes to employ.
A good example of the cautious step forward, and by no means the only one is the HART
protocol. It preserves compatibility with existing systems by superposing a digital transmission
protocol on the standard 4-20 mA loop. Each device adopting the standard can be read
remotely either from its continuous representation in the current loop or from the superposed
digital signal. A typical example is the 3044 temperature transmitter shown in Figure 12-9. This
accepts signals from a variety of primary sensors and converts the information to frequency
so that it may be passed to the digital sub-system through a transformer isolation barrier. The
microprocessor then provides compensation and communication facilities of the type
discussed above. HART has been made available as an open protocol, and it could well serve
a useful purpose in enabling circumspect industrial engineers to dip their toes into the digital
waters.
The industrial situation of smart sensors was analyzed by Favennec in 1987 [21], but actual
credible devices began to emerge from industry in 1988 (notably at the CongrCs Mesucora of
12.9 Industrial Progress 351

Figure 12-8. A system such as that shown in Figure 12-7 needs an internal operating system if it is to be
usable. Shown conceptually is the hiding of the complexities of hardware operation from the

nL
n -
3 c

Jie-
0 - 0
0

0 0 -
n 0

4 -20 nh
Current

Figure 12-9. An example of the implementation of the HART protocol in the form of the 3044
temperature transmitter from Rosemount.
352 I2 Smart Sensors

that year when some 15 different industrial devices were described). It is evident that the
variety of approaches is almost as wide as the number of manufacturers. They range from the
HART type of variation on 4-20 mA to the full adoption of one of the newer protocols (such
as FIP). While this is a little discouraging from the point of view of the establishment of
industrial standards, it is encouraging in the comprehensive way that manufacturers have
adopted the methods of digital compensation and information processing that research
workers have proposed over recent years.
A review of smart sensor development in one particular industry, the motor industry, reveals
a wide range of application of microprocessors and microelectronics to the measurement pro-
blems [22].

12.10 Trends

The evolution of smart sensors will continue to be controlled by the natural tension between
what is scientifically feasible and what is industrially acceptable. The needs of industry are
dominated by the requirement for standards, and while some industries can support elaborate
and advanced standards (such as the CAMAC system in the nuclear industry) these will
generally require to combine low cost with reliability. It is probable that industrial practice will
proceed in small advances like that represented by the HART system until some agreement is
reached on a more common fieldbus standard. These constraints, however, apply mainly to
the communications aspects of smart sensors, and the internal operation of industrial sensor
systems will be more open to free development.
One of the most important aspects of trends in electronic technique is the appearance of
the ASIC and its supporting technology. When the various elements can be selected from a
library and implemented quickly and cheaply by such methods as direct electron beam
writing, the variety and capability of smart sensors become greatly enlarged. Some of these
library sub-systems can be quite elaborate, so that, for example, alternative communications
processors can be selected and a design may be xlapted to a variety of bus standards.
The greatest danger for the whole field of smart sensing is that too much activity in too
many directions, without long term commitment and attention to the fundamentals, will pro-
duce a rapid development of the technology from the inchoate to the chaotic.

12.11 References

[l] Brignell, J. E., Rhodes, G. M., Laboratory On-line Computing, London: Intertext, 1975.
[2] Brignell, J. E., J. Phys. E: Sci. Instrum. 20 (1985) 559-565.
[3] Middlehoek, S., Audet, S. A., 1 Phys. E: Sci. Instrum. 20 (1987) 1080-1086.
[4] Greenwood, C. J., J. Phys. E: Sci. Instrum. 18 (1984) 650-652.
[5] Gambling, W. A., J. Phys. E: Sci. Instrum. 20 (1987) 1091-1096.
[6] Nylander, C., J. Phys. E: Sci. Instrum. l8 (1985) 735-751.
I2.lI References 353

(71 Brignell, J. E., Young, R. A., J. Phys. E: Sci. Instrum. 12 (1979) 455-463.
[8] Brignell, J. E., White, N. M., Cranny, A. W. J., Proc. IEE, Pt I 4 (1988) 77-84.
[9] Brignell, J. E., J. Phys. E: Sci. Instrum. 20 (1987) 1097-1102.
[lo] Press, W. H., Flannery, B. H., Teukolsky, S. A., Veterling, W. T., Numerical Recipes: the art of scien-
tific computing, Cambridge: University Press, 1986.
1111 Middlehoek, S., French, P. J., Huijsing., J. H., Lian, W. J., Proc. Transducers 87, p. 17.
(121 Lauks, I, Van der Spiegel, J, Sansen, W, Steyart, M., Proc. Transducers 85, p. 122.
1131 KO,W. H., Shao, B, X., Fung, C. D., Yeh, G. J., Sens. Actuators 4 (1983) 403-411.
1141 Chau, H. L., Wise, K. D., Proc. Transducers 87, p. 344.
[15] Erb, K. J., Proc. Transducers 87, p. 159.
[16] Tsukada, F., Sebata, M., Maruizumi, T., Miyahara, Y., Miyagi, H., Proc. Tranducers 87, p. 155.
[17] Brignell, J. E., Dorey, A. P., J. Phys. E: Sci. Instrum. 16 (1983) 952-958.
(181 Brignell, J. E., Sens. Actuators 10 (1986) 249-261.
[19] Brignell, J. E., J. Phys. E: Sci. Instrum. 18 (1985) 759-765.
[20] Cooper, R. A., Brignell, J. E., Sens. Actuators 7 (1985) 189-198.
[21] Favennec, J. M., J. Phys. E: Sci. Instrum. 20 (1987) 1087-1090.
[22] Wolber, W. G., SAE Special Publications 536 (1983) 1-10.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

13 Interface Systems
.
University of Southampton. Southampton. UK
JOHN ATKINSON

Contents
13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
13.2 Analog Transducer Interfacing . . . . . . . . . . . . . . . . . . . 356
13.3 Data Acquisition Systems . . . . . . . . . . . . . . . . . . . . . 359
13.4 Digital Transducer Interfacing . . . . . . . . . . . . . . . . . . . 361
13.5 Transducer Networks . . . . . . . . . . . . . . . . . . . . . . . 362
13.5.1 Network Topology . . . . . . . . . . . . . . . . . . . . . . . . 363
13.5.2 Serial Digital Data Transmission . . . . . . . . . . . . . . . . . . 363
13.5.3 Parallel Digital Data Transmission . . . . . . . . . . . . . . . . . 366
13.5.4 Propriety Bus Based Systems . . . . . . . . . . .......... 367
13.5.5 Multivendor Bus Based Systems . . . . . . . . . . . . . . . . . . 367
13.6 Communication Protocol Models . . . . . . . . . . . . . . . . . . 368
13.6.1 Physical Layer . . . . . . . . . . . . . . . . . . . . . . . . . . 369
13.6.2 Link Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
13.6.3 Application Layer . . . . . . . . . . . . . . . . . . . . . . . . . 369

13.7 Physical Interfaces . . . . . . . . . . . . . . . . . . . . . . . . 370

13.8 Link Transport Mechanisms .................... 373


13.8.1 Media Access Control . . . . . . . . . . . . . . . . . . . . . . . 374
13.8.2 Frame Formats . . . . . . . . . . . . . . . . . . . . . . . . . . 374
13.8.3 Error Checking . . . . . . . . . . . . . . . . . . . . . . . . . . 376
13.9 Application Data Protocols . . . . . . . . . . . . . . . . . . . . 377

13.10 The Fieldbus Standard . . . . . . . . . . . . . . . . . . . . . 378

13.11 Power Distribution . . . . . . . . . . . . . . . . . . . . . . . . 379

13.12 Design Considerations . . . . . . . . . . . . . . . . . . . . . . . 380

13.13 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381


356 13 Interface Systems

13.1 Introduction

Modern instrumentation systems often employ some form of distributed processing which
inevitably involves the use of high speed digital communication links [l]. Due to the existence
of a mulitplicity of design parameters, however, the correct choice for any given system is not
always readily apparent. As a consequence, instrumentation system designers have tended to
work to so called standards which often offer little more than a previously demonstrated mode
of operation. The main objective of a true standard, however, is to guarantee interoperability
between diverse equipment conforming to that standard. Unfortunately, the achievement of
such a goal often entails a significant sacrifice in terms of efficiency and cost.
The problems concerning standardization of instrumentation communication interfaces
have, however, been recognized by the international standards bodies with the result that a
concerted attempt is being made to define a common standard. Unfortunately, this process
is likely to be extremely lengthy with the distinct possibility of a multiplicity of new standards
evolving in the interim period rather than the one common standard that was the original goal.
The proliferation of digital communication interface techniques is not an impossible situation,
however, and design methods to mitigate against the difficulties do exist.
In many respects the problems faced in designing a distributed instrumentation system
employing digital communications are not new. Telemetry systems have always had to over-
come the problems associated with guaranteeing reliable transmission of transducer data over
large distances. The difference between the communication requirements of a distributed in-
strumentation system and those of an analog signal transmission system are most apparent
when considering the nature of the transducer data itself. In the case of the latter, the data
is usually no more than an analog signal, a voltage or current, which corresponds to the value
of a real world variable.
In the case of the distributed system, however, a much higher information content is usually
to be found with typically the type, units, and status of the variable being transmitted along
with the value. This concept of a system variable being self identifying leads to perhaps one
of the most attractive aspects of distributed processing in instrumentation namely the use of
multiplexed wiring. It is no accident that virtually all digital communication systems currently
in use in instrumentation systems employ multiplexing, being testimony to the fact that wiring
costs are often the dominant factor in many application areas.

13.2 Analog Transducer Interfacing

The principle difficulty with most transducer signals is that they are low level, this is best
illustrated with reference to a typical sensor. Consider a metal foil strain gage connected to
a resistance bridge as shown in Figure 13-1.
With a typical gage factor (unit change in resistance per unit strain) of approximately 2 the
maximum output signal from the circuit shown is of the order of 1 mV/V of excitation at the
yield stress of steel. In many transducer applications this figure would, for reasons of
reliability, represent at least twice the full scale output signal and hence, for only a 1% resolu-
13.2 Analog Transducer Interfacing 357

R +

Figure W-1. Strain gage interface via resistance bridge.

tion, it would be necessary to resolve down to 5 pV. Clearly it is less than ideal to transmit
such a low level signal over any distance and some form of local signal conditioning would
consequently be advantageous.
In its simplest form local signal conditioning consists of some means of voltage amplifica-
tion to magnify the signal levels to a point where they are no longer susceptible to extraneous
noise. The level at which this becomes the case is dependent on the medium over which the
analog signal is to be transmitted and the characteristics of the signal itself.
Figure 13-2 shows a typical arrangement where an amplifier is incorporated into the
transducer to provide both signal magnification and impedance matching. One obvious
benefit of this technique is that the interface is now reduced to three wires, additionally the
output signal can often be made to correspond to some standard range (eg, 0 to 5 volts) to
represent zero to full scale reading.
When operating in electrically noisy environments, or over long distances, even these
relatively high signal levels can be very prone to interference. This fact, together with a desire

Figure w-2.
Local signal conditioning of sensor
reading as employed in a typical
transmitter.
358 I3 Interface Systems

Figure W-3.
4 to 20 mA transmission system.

to reduce the wiring and power supply requirements still further, gives rise to what has subse-
quently become a de facto standard in the process control industry, namely the 4 to 20 mA
current loop [2]. The principle of operation is illustrated in Figure 13-3.
The transducer is powered from an arbitrary supply voltage, in the range 10 to 30 volts for
example, and the output current is constrained within the range 4 to 20 mA, where 4 mA cor-
responds to zero and 20 mA corresponds to full scale.
The principle advantages of this technique include the fact that as a current the signal is
less liable to interference and the interface is reduced to just two wires. This latter point is also
noteworthy in that the 4 mA base current, which is used to power the transmitter electronics
and bridge excitation, is a good indication that the transmitter is functioning correctly with
a zero input level. Conversely an open circuit condition in the transmitter wiring is im-
mediately apparent due to the absence of current. Hence the introduction of the 4 mA live

Figure W-4. 4 to 20 mA transmitter functional diagram.


13.3 Data Acquisition Systems 359

zero current allows powering of the transmitter and failure detection using only the two wires.
The ability to operate from a wide range of supply voltages is also particularly attractive in
a process control environment where many different power sources are encountered and power
supply regulation can sometimes be a problem.
Figure 13-4 shows the basic functional schematic of a 4 to 20 mA transmitter. The sum of
the excitation and supply current (Zs), the floating, Zener stabilized, supply current (Z,) and
the offset adjust current (IA) is arranged to be exactly 4 mA. The gain of the differential
amplifier and U to Z converter is then adjusted to give a measurement current (IM)in the
range 0 to 16 mA. Hence, the output current (Z,,,,) is confined to the range 4 to 20 mA und
U,,, = I,,[ R , volts. The constant current supply (Z,) ensures that variations in the supply
voltage do not affect Z,, , and hence U,,,is also independent of Us,,,,, . The use of a 100 SZ
load resistor with a current loop interface results in a 0.4 to 2 volt signal which has in itself
become something of a standard in instrumentation equipments.

13.3 Data Acquisition Systems

Nowadays the information presented by an input transducer (sensor) is almost always pro-
cessed in a digital computer. This information can then be acted upon, either by human inter-
pretation or in the case of an automatic control system by the digital computer via an output
transducer (actuator). The use of digital controllers in almost all aspects of supervisory con-
trol and data acquisition (SCADA) systems results in the requirement for the digitization of
the transducer data. The point at which the analog to digital, and conversely the digital to
analog, conversion is carried out significantly affects the philosophy of the system design.
Traditionally the high cost of analog to digital converters (ADCs) prevented their use
anywhere other than at the central point or system controller. A large plug-in printed circuit
board with an ADC and analog multiplexing capability residing within the central controlling
computer being the typical hardware configuration (Figure 13-5). This method of operation
also owed much of its popularity to the ease with which it could be retrofitted to existing in-
strumentation where a separate cable is used to connect each and every transducer to the cen-

ADC
DIQltal 8
Computer- A n n 1o g
Mux

L
Trensducers

Figure w-5.
Centralized data acquisition.
360 13 Interface Systems

tral point. The ability to use existing plant cabling when installing computer control was thus
a very strong argument for retaining analog transmission techniques.
The reductions in size, cost and power consumption experienced with the continuing ad-
vances in microelectronics, however, have resulted in the viability of situating the ADC in the
field at the transducer. Hence, signal conditioning can be carried out locally in digital form
with the transducer becoming a digital transmitter. In terms of the communication of
transducer data this enables the use of multiplexed wiring and the two way transfer of infor-
mation such that transducers can both talk and listen. Naturally, having the transducer data
in digital form at the transducer itself opens up many other possibilities in terms of distributed
processing and complex signal conditioning. The most significant possibility is perhaps that
of reducing the amount of transmitted data as it is often possible to implement a relatively
high level of local autonomy. The use of a changes only protocol is a good example, whereby
data is not transmitted unless it has changed by a predetermined amount from that transmit-
ted previously.
The interconnection strategy of a distributed instrumentation system can take many forms
but perhaps the most common is that of a multidrop bus. Figure 13-6shows a schematic
diagram of this method of networking where transducers may be located at some considerable
distance from the system controller. The concept of a plant wide data highway, to which
transducers and controllers may be simply connected, offers the opportunity for significant
savings in wiring costs.

Dlgltal
Si gna I s
Host
Computer

Figure u-6.
Transducers Multidrop transducer network.

Whilst the multidrop bus is the most popular method of networking, being employed on
all the currently emerging contenders for a future standard, there are other possibilities. The
use of a star shaped network, where, for example, the analog signals of Figure 13-5 are
replaced by digital signals, has sometimes been employed. This is particularly likely in an ap-
plication where existing analog wiring can be reused in a retrofit of digital communications.
Here the benefit of employing digital communications not only avoids the problems
associated with the transmission of analog data in terms of the quality of the data, but can
also, by increasing the quantity of information available, enhance the performance of the
overall system.
13.4 Digital Transducer Interfacing 361

13.4 Digital Transducer Interfacing

Extending the concept of carrying out the nalog to digital conversion locally at th
transducer leads to the incorporation of signal conditioning at the source and hence to
distributed instrumentation systems.
Local signal conditioning inevitably implies the compensation of transducer characteristics
such as linearization and offset correction. Increasingly this also takes the form of dynamic
response compensation in the form of signal filtering often implemented using software on
a microprocessor. Cross-sensitivity correction can also be realized by using a secondary sensor
to measure an offending environmental parameter such as temperature. The incorporation of
a single chip microprocessor into a transducer is an extremely cost effective method of realiz-
ing the concept of an autonomous transducer often referred to as an intelligent transducer or
smart sensor [3,4]. Figure 13-7 shows such an arrangement where the microprocessor controls
the analog to digital front end in such a way as to select between primary and secondary sen-
sors. The incorporation of gain and offset control not only facilitates the transducer calibra-
tion procedure, but can also allow range changing, or span turn down as it is sometimes called,
to take place. More details on smart sensors have been given in the previous chapter.
Perhaps the greatest benefit offered by the exploitation of autonomy in transducers is the
ability for self identification. It now becomes theoretically possible for a sensor to be con-
nected to the system at any point and to be incorporated into the overall scheme via a start-up
dialog with a controlling element or computer. If, for example, a temperature sensor and a

MUX ADC
Micro-
processor
Secondary A
Sensor

‘rr
p,
Control llnes

D r Iv e r

Data Bus

Figure W-7. Incorporation of a signal conditioning microprocessor, resulting in an autonomous


transducer.
362 13 Interface Systems

pressure sensor were added to a system but connected each to the wrong input channel the
mistake would be immediately apparent when the sensors identified themselves. Often the
system hardware and software do not even require modification to take account of the new
configuration. In the example given above the computer would automatically know which
channel was temperature and which was pressure from the information transmitted by each
sensor. This is a highly significant development given the continuing shortage of skilled man-
power experienced by the instrumentation industry and the consequential high costs of system
ownership. In many applications, however, the cost of a system failure is the dominant factor
and any improvement in the ability to maintain the transducers in a reliable operational state
is justifiable. Hence, the increased functionality of an autonomous transducer can often be
employed so as to enhance the operational integrity of the overall system. Some typical func-
tions of an autonomous transducer are listed below.

- self compensation (eg, linearity and offset correction)


- cross sensitivity correction
- frequency compensation (filtering)
- self identification
- auto ranging (resolution maintaining)
- self diagnostics
- auto calibration
- local display
- remote transmissionheception of data.

13.5 Transducer Networks

It would be a little unfair, though none the less true, to describe most of the current com-
mercially available transducer networks as being the result of a rather chaotic evolution. This
has been largely due to the absence of any common standard but also to the gradual evolution
of microelectronics technology which results in systems being continually superseded with
each new generation of products. As will be seen in later sections of this chapter some
semblance of order is beginning to emerge from the standards debate. Functional requirements
have now been defined independently of the implementing technologies themselves. For a long
time, however, most of the functionality of transducers containing autonomous processing
capabilities were dictated by the physical realization of the communications link. A typical ex-
ample is the adoption of the modem standard EIA RS232 [5] (CCITT V24) for point to point
connections which many manufactures have erroneously referred to as a protocol. Certainly
RS232, as it is usually designated, defines the electrical signalling to represent the logic levels
of the data and does include a definition of a handshaking style of data flow control in the
form of additional interface circuits such as “Request to Send” and “Clear to Send”. Unfor-
tunately, this standard is very rarely adhered to by the implementors with the result that in-
compatibilities abound. Some attempt has been made to standardize in the computer industry
and BS6514 [6] defines a common subset of the EIA RS232 standard which allows serial
13.5 Transducer Networks 363

digital interfaces to interoperate without the usual confusion resulting from an attempt to in-
terconnect EIA RS 232 interfaces.

13.5.1 Network Topology

As has already been mentioned, the most popular interconnection strategy for transducer
networks is based on the multidrop bus (Figure 13-6). Other topologies include the star net-
work such as that employed in a centralized data acquisition system (Figure 13-5) and the ring
network. The latter is really only a special case of the multidrop bus with active repeaters
replacing the passive taps (junction boxes).
Ring networks tend to suffer from reliability problems in that a fault in any one node (active
tap) usually renders the network inoperable.
In terms of overall reliability, the star network obviously performs better than the bus since
a single cable fault should, in theory, only disable one transducer. The major perceived
weakness of a bus system is the vulnerability of the bus itself to a cable fault, especially a short
circuit. In practice, it is possible to guard against system failure in the event of a bus fault in
several ways. Dual redundancy is one method where, for example, each autonomous
transducer can be connected to more than one bus, possibly with the data flow being in op-
posite directions for added security (Figure 13-8). Additionally, the use of galvanic isolation
(eg, transformer or optical coupling) together with resistive wiring stubs can offer a high level
of protection to the bus particularly in the event of transceiver faults.

Autonomous

Figure 33-8. Transducer


Example of multidrop bus fault protection.

13.5.2 Serial Digital Data Transmission

Despite the problems associated with defining the physical connection and controlling the
flow of data over digital communication interfaces, their use has spread significantly. Bit serial
digital data transmission is perhaps the most common technique in use for remote data
transmission. This is largely a result of the ease with which this method can be implemented
and the cost savings in wiring as compared to parallel digital transmission. Many single chip
microprocessors incorporate a serial digital interface as standard and the well tried universal
asynchronous receiverhransrnitter (UART) format shown in Figure 13-9 can be readily
adopted for the transmission of serial 8 bit data packets.
364 13 InteMace Systems

I I
line idle I or 2

Start b i t
8 data b l t s
1'
O p t i one 1

p a r I ty bit

Figure W-9. UART serial data format.

The adoption of an 8 bit quantum for the transmission of digital data was a significant con-
tribution towards the development of the character oriented communication protocols which
distinguish most of the current commercially available implementations. Certainly the use of
ASCII (American Standard Code for Information Interchange) encoded data has much to
offer due to the ease with which the data can be interpreted by a human operator. Addi-
tionally, it allows the use of unique framing characters which do not then appear among the
characters of the data string, hence simplifying the generation and detection of complete
messages.
A typical example of a simple message passing protocol is that of the Southampton
Transducer Protocol (STP) [7] developed by the author in 1983. The physical realization of
the transducer network took the form of a twin twisted pair cable carrying EIA RS485 [8]
differential data on one pair and power supplies on the other. Autonomous transducers (both
sensors and actuators) were then multi-dropped onto this transducer bus which had a controll-
ing microcomputer as the source of bus access arbitration and control.
The method of working adopted for the communication of transducer data within STP
relied on the host computer polling the outlying transducers with a message of the following
form:

* nnn cmd ddd CR LF

where the characters have the following meanings:

* - start of master message


nnn - three byte slave address
cmd - one byte command character
ddd - optional data field
CR LF - carriage return, line feed

A transducer might then respond to a request for data with a message typically of the form:

nnn tt ff ddd ddd CR LF


13.5 Transducer Networks 365

where the characters have the following interpretation:

- start of slave message


nnn - three byte slave address
tt - two byte transducer type code
ff - two byte data format code
ddd - data field
ddd - data field (repeated)
CRLF - carriage return, line feed.

This masterlslave mode of operation ensures that transducers only “speak” when “spoken
to” thus avoiding the problems of transmission media access contention. The use of carriage
return, line feed (CR, LF) as the message terminating characters was designed to allow easy
display of the message strings using conventional equipment such as visual displays units
(VDUs) and printers. Although succesfully employed with actuators [9] STP was mainly
designed for use with environmental sensors and as such has been successfully implemented
commercially by Sieger Ltd. [lo], a UK manufacturer and supplier of gas detection systems.
Whilst STP is a simple message passing protocol with centralized media access control onto
a multidrop bus, distributed processing has been realized using a peer to peer communication
mechanism among slaves under the control of the master. Hence, the autonomous transducers
are able to act upon data they obtain directly from the other transducers without using the
hostlmaster computer as an intermediary information buffer. Examples of this technique in-
clude the compensation of gas sensor cross-sensitivities resulting from environmental
parameters which are monitored by pressure and temperature sensors distributed along the
bus. A similar system to the Sieger Bus is marketed by Rhopoint [ll]. The Rosemount HART
protocol [12] also makes use of a simple character based message passing protocol which is
transmitted using the Bell 202 modem standard as a frequency shift keyed (FSK) signal
superimposed on the transducer power supply lines. Although capable of being used in a
multidrop configuration, HART is preferably used for point to point connections where it
replaces a 4 to 20 mA signal.

x - w

x w
w

>

Power X -
rn
w
x =-
a
>
>
366 13 Integace Systems

The major appeal of these types of transducer interface must be the low cost with which
they can be implemented. Their greatest limitation though is undoubtedly the lack of func-
tionality in a simple message passing protocol and hence the difficulty in extending or
upgrading the capability of any system employing character based message passing. In terms
of the interface to the physical medium for autonomous transducers, however, the serial digital
transmission format serves very well provided the access control methodology does not imp-
inge on the message format. This technique is demonstrated well in the PROFIBUS [13] where
the standard 8 bit UART format is used together with EIA RS485 to form the physical
medium of a multilayered, multivendor communications protocol for industrial automation
purposes.

13.5.3 Parallel Digital Data Transmission

Character oriented message passing protocols are not restricted to the serial formats
discussed in the previous section and can also be implemented as parallel interfaces. Perhaps
the best known is the general purpose instrument bus (GPIB) IEC625, also known as the
Hewlett Packard Interface Bus or IEEE488 bus [14]. This international standard has rapidly
become the mainstay of most laboratory experiments involving computer controlled in-
strumentation. The bus consists of 8 data lines and 8 control lines which permit the high speed
transfer of parallel data bytes over the short distances typically encountered between
laboratory equipments (eg, up to 20 meters). The names given to the handshake lines
employed with the IEC625 bus generally serve to describe their function adequately and are
shown in Figure 13-11.
Whilst suitable for the interconnection of microcomputers, oscilloscopes, digital voltmeters,
power supplies etc., the GPIB is generally regarded as being too expensive for transducer inter-
facing, the main cost element being due to the duplication of circuits across the byte wide par-
allel interface and the cabling between units. The fact that the circuits to implement the GPIB
interface are available as a relatively low cost chip set serves to further illustrate that wiring
costs still tend to dominate in industrial instrumentation systems.

OAV d m t m vmltd
W D not rmmdg f o r dd.0
WAC not data mcs.pt.d
ATN dt.nt,on
SRO ..?".C. r.qu..t
REN r.nr.t. .nabl.
EO I .nd ~r nd.ntify
I FC rnt.rf.s. s1.w

Figure Sll. GPIB IEC625 signal names.


13.5 Tkansducer Networks 367

13.5.4 Propriety Bus Based Systems

In addition to the previously mentioned Sieger bus, Rosemount HART and Rhopoint
systems employing character based message passing protocols there are a significant number
of digitally interfaced transducer network products on the market. Schlumberger have S-NET
[15], a typical system employing mixed power and data over a transformer coupled bus. Fox-
boro have employed an improved version of the HDLC protocol [16] which opens up many
interesting possibilities in the context of wide area networking since HDLC forms the basis
of many public service telephone network protocols. Another company exploiting HDLC in
an interesting manner is Newmark Technology whose OMNIBUS product range implements
a slave to slave communication capability whilst maintaining a centralized (master) access con-
trol to the bus [17].

13.5.5 Multivendor Bus Based Systems

The proprietary systems discussed above are only a few examples of the many currently
available, in most cases their use is generally determined by the purchase of particular
transducers and instrumentation equipments that employ the protocol concerned. There are,
however, several multivendor systems on the market such as those generated by state owned
corporations, eg, British Coal’s BS6556 [IS] protocol, or those resulting from collaborative
projects such as ERA 1553, EUREKA, FIP and PROFIBUS.
The ERA 1553 initiative is the result of a UK consortium headed by the Electrical Research
Association (ERA) establishing a committment to use a modified version of MIL.STD.1553
[19] for instrumentation purposes. MIL.STD.1553 was originally established as an avionics
data bus standard which the ERA project has extended to give increased distance operation
and to meet the requirements for intrinsic safety.
The EUREKA fieldbus project initially involved 13 companies from 6 different countries
throughout the European Community, who are committed to produce a working demonstra-
tion of a fieldbus. The intention is that each participating company will produce at least one
product capable of being used over the fieldbus at this demonstration.
FIP (Factory Instrumentation Protocol) [20] is a French initiative with the objective of pro-
ducing a fieldbus which can be presented to the International Electrotechnical Commission
(IEC) for possible adoption as an international standard. The FIP consortium was established
in 1986 with some 25 member companies and societies and appears to have a fair degree of
overlap with the EUREKA project.
PROFIBUS [21], from PROcess FIeldBUS, is a German fieldbus standard derived from the
efforts of a group of 15 companies and 5 institutes with government support. In order to meet
short timescales the standard was based on well established physical techniques, such as those
documented in Section 13.5.2, and similarly well established protocols.
With such a proliferation of initiatives and potential standards the main objective of achiev-
ing uniformity at the interface would appear to be in danger of being defeated. What is re-
quired is one common standard and to this end a set of functional requirements for an interna-
tional standard fieldbus have been defined [22]. Before examining the fieldbus issue, however,
it is important to consider how a communication protocol for instrumentation purposes might
be designed and what choices are available to the designer.
368 13 Interface Systems

13.6 Communication Protocol Models

The difficulties encountered in designing communication protocols are not new, par-
ticularly to telecommunication engineers who have had to deal with the ever increasing load
of international traffic in digital communications. In an effort to facilitate the process of
designing internationally compatible communication systems the International Standards
Organisation (ISO) has defined a multi-level communications protocol model [23]. This
model is designed to be used as a guideline for the development of actual protocols, employing
a strategy known as Open Systems Interconnect (OSI). Figure 13-12 shows a diagrammatic
representation of the communication protocol model as a 7 layer stack. This ISO/OSI model,
as it is known, describes the flow of data across a network as a downwards progression
through the layers from the application to the physical, across the physical medium (eg, co-
axial cable, twisted pair) and back up the stack of the receiving station. The translation of data
format and the addition of error checking and routing information is effectively transparent
in this mode of operation so that a peer to peer exchange appears to take place at each level.
The overhead involved in achieving the apparent transparency between the levels is, however,
unacceptable in a real time response situation such as that encountered in industrial in-
strumentation systems. As a result, a reduced model has tended to gain favor and has been
described as an enhanced performance architecture (EPA) in the manufacturing automation
protocol (MAP) [24] for example. MAP in fact, uses the full 7 layer concept but. .for . ,real time

- _ _ _ _ _ 3 APPL ICAT ION I


PRESENTATION - _ _ _ _ _ 3 PRESENTATI ON I
I SESS I ON k--____
--____
- _ _ _ _ _
DATA LINK ---___ DATA LINK I

Y
I I
PHYS ICAL

Figure U-l2. ISO/OSI 7 layer communications protocol model.

~~

TRANSMISSION MEDIUM
[ e . g . t w i s t e d p a i r . cooxial c a b l e 1 .

Figure W-W. Communications protocol model for instrumentation.


13.6 Communication Protocol Models 369

response the EPA is expected to be employed on lower level instrumentation buses such as the
fieldbus. Exactly which layers are adopted in a reduced model is open to debate but their
classification according to the function performed reveals a “collapsed” version of the full
OSI implementation. Figure 13-13 shows a diagrammatic representation of an instrumentation
communications protocol model.
In attempting to explain the operation of the model, it is generally easiest to start with the
bottom layer and work upwards defining the function of each layer in turn.

13.6.1 Physical Layer

The lowest level of the protocol model defines the interface to the interconnecting medium
to be employed. Included in this definition is the type of medium (coaxial cable, twisted pair,
fiber optic) and the method of connection (transformer coupling, direct tap at junction box).
The bit encoding of the data is also specified at this stage together with a complete electrical
specification regarding the transmitted and received signal levels. Ideally, a physical layer
definition should also contain information relating to the loading of the bus and any timing
constraints that have to be met. In practice, it is difficult to produce a standard that covers
these parameters without placing undue restrictions on the installation of the wiring itself. The
usual method adopted to allow the physical layer to be defined in isolation is to consider the
input and output parameters of each connected device and, as long as these parameters are
not violated, disregard the actual interconnection. Thus the length of a bus, its cable
characteristics and the number of connected devices and repegiers are unimportant so long
as the effective impedance at the transducer, for example, is within specified limits.

13.6.2 Link Layer

The middle level of the 3 layer stack is responsible for the reliable transmission of data from
one connected node to another. To this end the addition of connection information and error
checking and the acknowledgement and re-transmission of data are all duties that have to be
carried out by the link layer transport mechanism. With the emphasis being on transparency
of operation, the link layer should be completely independent of the physical layer by pro-
viding the “logical link control”. The assumption is that a message received from the applica-
tion level of one device will be transmitted to its destination and an acknowledgement of
receipt, or a notification that the message was not transferred correctly, returned from the link
layer of the destination device to the application layer of the sending device.

13.6.3 Application Layer

Typically, the application layer defines the services that are provided by the network com-
munication protocol (eg, the ability to request a reading from an input variable or a
transducer’s identification message). Also defined is the sequence of events, of transmitted
and received messages, to implement each of the services offered. In its detail, an application
layer protocol should thus specify the byte ordering of the messages and hence the context
310 13 Interface Systems

of the message when interpreted at both ends of a communication link. At the top of the pro-
tocol model the application data protocol effectively defines a descriptive language with which
the functionality of the transducer or instrument is described.

13.7 Physical Interfaces

At the lowest level of an autonomous transducer interface to a network is the physical


medium forming the interconnect. Whilst implementations employing radio transmission or
optical fiber do exist, the vast majority of these interfaces employ copper wire, usually in the
form of a twisted pair or coaxial cable. A communication standard consequently needs to
define the exact method of interconnection to the medium (eg, direct tap, transformer coupl-
ing, etc.). Typically some limits are placed on the level of drive employed (ie, transmitter out-
put levels) to represent the binary logic levels. This can most usefully be done by defining the
acceptable maximum and minimum levels at the receiver which consequentially sets the levels
at the transmitter taking into account any losses due to the transmission path.
The factors influencing the choice of signal levels typically include the trade-off between
noise rejection and power consumption, any considerations for intrinsic safety and, perhaps
most importantly, the availability of integrated circuit transceivers. This latter consideration
has influenced many transducer network designs where the use of signalling levels such as EIA
RS232 or RS422 have been adopted for just such a reason. Increasingly differential line
drivers, such as the RS422 type devices, are being preferred to single ended line drivers such
as the RS232 type devices. This is due to the significant improvement in noise immunity and
transmission rate offered by the differential line drivers. RS 232 typically employs a voltage
in the range - 3 to - 12 volts to signify a logic 1 and a voltage in the range + 3 to + 12 volts
for a logic 0. As any voltage outside of these two ranges is undefined there is an obvious limit
to the amount of signal degradation that can be tolerated due to noise or signal attenuation.
A differential line driver conforming to RS422 (Figure 13-14) signifies a logic 1 with a
+ 5 volt signal on the H-data line and a 0 volt signal on the Ldata line. A logic 0 simply

+---+I
R5232 Slngie ended

0 01

R5422 Differential
I data data

data

Figure W-14. Line driver comparisons.


13.7 Physical Interfaces 371

reverses these two signals to give 0 volts on the H-data and +5 volts on the Ldata line. Hence
the polarity of the line indicates the transmitted logic level which allows reliable detection
down as low as 300 millivolts at the receiver. Hence significant attenuation of the signal can
be tolerated allowing operation over long distances at very high data rates. The differential
nature of the signal is also well suited to transmission over twisted pair cabling further reduc-
ing the susceptibility to interference.
An extension of the differential line drive standard to a multidrop capability allowing up
to 32 simply connected transceivers is defined in EIA RS485 [9]. Transceiver chips meeting
this specification have a thermal shut down capability to prevent device damage in the event
of bus contention.
In considering the transmission of binary data to line the actual encoding format of the data
is an important consideration. By simply representing the logic states with voltage levels it is
possible to achieve long periods where the line is idle, for example, when a run of binary 1’s
is transmitted. This situation can be seen in Figure 13-15. With non-return to zero (NRZ) data
a run of logic 1’s (or logic 0’s) results in a loss of synchronizing edges in the data waveform.
This loss of a synchronizing reference can give rise to bit errors if any isochronous phase
distortion (jitter) is present in the incoming data stream. An alternative encoding strategy is
to use edges in the data to indicate the logic level as in the biphase (Manchester) coding shown
in Figure 13-15. Here a low to high transition indicates a logic 0 while a high to low transition
a logic 1, thus ensuring that a resynchronization with the incoming data stream occurs in each
and every bit period.
As well as benefitting from being a self-clocking waveform, biphase encoding also offers
the additional benefit of removing the DC component from the signal consequently permit-
ting the use of transformer coupling. This latter consideration is important both from the
point of allowing galvanic isolation for intrinsically safe circuits and permitting the use of im-
pedance matching techniques. The differential Manchester encoding format of Figure 13-15
is important in this context in that it is polarity independent and can still be decoded should
the line signals become reversed. Differential Manchester encoding is generated according to
the following rules. A logic 0 is transmitted as TT and a logic 1 as NT where T stands for
toggle and N means no toggle. In other words the logic level of the line is changed in both
halves of a bit period representing a logic 0. A logic 1 however retains the previous logic level

I I / I A
I I I I I
NRZ
I - I I I I

Figure W-15.
Binary encoding formats.
312 13 Interface Systems

of the line during the first half bit period and toggles the 'logic level during the second half
of the bit period.
A further aspect of both Manchester and differential Manchester encoding is that it is possi-
ble to define non data patterns. As a consequence violations of the encoding rules can be used
to signify message preambles and the line idle condition. This technique is often used with
Manchester encoded protocols to afford an additional degree of error protection. The main
disadvantage however is that out of band frequencies are produced as a result of these changes
in signal patterns giving a significantly wider spectral occupancy of the digital transmission
than would be the case if only valid logic level patterns were transmitted.
The NRZ waveform is often referred to as a baseband signal as its frequency spectrum is
mostly contained in the range DC to Rb/2, where Rb is the bit rate of the data. Biphase en-
coding has the effect of moving the baseband signal along the frequency spectrum and center-
ing it about a carrier frequency numerically equal to the bit rate. The term carrierband,
however, describes the general case when a sinusoidal carrier signal is modulated in some way
by the baseband binary signal. This frequency translation is usually performed to allow the
binary data to be transmitted over a frequency sensitive medium such as a public service
telephone network (PSTN),a radio link or a fiber optic cable for example. The carrier signal
can have its amplitude, frequency or phase angle modulated by the baseband signal as shown
in Figure 13-16 below.

I I
I
I
I - I

Baseband I NRZ 1 I I I I I
- -
I I I I I

I
Amplitude Shift
Key I ng I

Frequency Shift
Key I ng

I I I I I I I I I I I

Phase Shlft
Key i ng

I I I I I I I I I I I

Figure U-16. Digital modulation techniques.


13.8 Link Transport Mechanisms 373

Of the modulation techniques shown here, FSK is perhaps the most common and in par-
ticular the case where one cycle of carrier is used to represent a logic 1 and two cycles a logic
0. This phase continuous signal forms the basis of most of the local area network standards
(eg, I S 0 8802.4AEEE 802.4) that have been considered for adoption in industrial instrumen-
tation systems in particular the Manufacturing Automation Protocol (MAP). Additionally,
MAP goes a stage further and uses a concept known as broadband where several carrierband
signals, at different carrier frequencies, are combined onto one cable. This results in a plant
wide data highway, or backbone as it is often called, capable of carrying not only instrumenta-
tion data but also, for example, closed circuit television, telephone and other computer data.
To allow sufficient bandwidth for each channel on the link the carrier frequencies of a broad-
band system are generally in the VHF range (eg, around 200 MHz). Naturally, this makes
MAP, and similar broadband networks, too expensive for the direct connection of in-
struments. However, it is quite likely that these networks will carry transducer data and hence
compatability is required. This can be achieved via an interface or gateway to the broadband
network from a carrierband fieldbus as shown in Figure 13-17. The function of the gateway
is to convert, or translate, from the protocol and physical medium of one network to those
of another.
BROADBAND i MAP 1

GATEWAY

w CARR 1 ERBAND I F IELDBUS I

Figure W-17.
Fieldbus connection TRANSDUCERS
to broadband network.

13.8 Link Transport Mechanisms

In a distributed instrumentation system the transfer of transducer data should be achieved


in a reliable manner with the complexities of any error detection and correction or retransmis-
sion of the data being transparent to the source and destination. In keeping with the OSI con-
314 I3 Interface Systems

cept the link transport mechanism provides this transparency by appending the necessary in-
formation to allow routing and error correction to take place and subsequently removing it
from the application data when this has been achieved. At a more detailed level a link trans-
port mechanism must typically provide the physical medium access control, the framing for-
mat for the data, the calculation of any redundancy check or parity codes and the source and
destination device addressing. It may also be required to perform the sequencing of the com-
ponents of any application data message communicated across the network in separate
frames.

13.8.1 Media Access Control

The media access control method can take many forms but the two most popular would
appear to be that of a multimaster, or token passing, access method and a subset employing
a single master with centralized media access control (CMAC). Essentially both methods rely
on there being only one node on a network at any time which can initiate the transmission
of messages or data packets. With CMAC the master station is fixed (usually situated within
a controlling master computer) with all other connected nodes being slaves, or outstations,
which respond only when spoken to by the master. With token passing any one of the con-
nected nodes can become a master station and initiate data transfers, unless of course it does
not have the capability to become a master and is permanently configured as a slave. To avoid
contention, however, only the station possessing a notional token can assume the role of the
master station. Hence, some mechanism is required to enable the token to rotate among the
potential masters in order that they can gain access to the physical communication media. This
is usually done by requiring each station to pass the token on after some predetermined time
interval, naturally a mechanism must exist to regenerate the token if it should become lost due
to a communications error. Of the two methods CMAC is generally the easier to understand
and implement and does have the advantage of being the more deterministic. This stems from
the fact that with a CMAC system the rate at which outstations are polled can be predeter-
mined and hence the response time of a particular transducer can be easily calculated. Token
passing systems are less suited to low level transducer communications since polling rates are
not so easily defined and hence response times are difficult to calculate. Certainly token pass-
ing has its advantages in the area of computer to computer communications and consequently
forms the basis of several standard protocol implementations. This is also true of the carrier
sense multiple access collision detect (CSMA-CD) method where multi master operation
results in lost data packets due to random access contention. Again the non-deterministic
nature of these protocols renders them unsuitable for most instrumentation purposes where
there is a requirement for time scheduled, as well as aperiodic, data transfer.

13.8.2 Frame Formats

The framing format adopted by the link transport mechanism either allows for the transmis-
sion of whole messages (or in some cases more than one message) within a packet, or alter-
natively breaks messages down into a succession of smaller data packets. Figure 13-18 shows
the various frame formats for the MIL. STD. 1553 protocol [19] which segments messages into
13.8 Link Transport Mechanisms 315

Sync Tx/Rx Blt Par I t y


Pattern Bit

Corn ma nd
Word
5 111 5
Remote Sub Data
Terrnlnal Address Word
Address Count

Data
Word
16
Data B l t S

Status
Word
11
Remote Status B l t s
Terminal
Address
Figure U-18.MIL.STD.1553 frame formats.

16 bit quanta. This type of transport mechanism implements a packet assembler/disassembler


(PAD) at each node which is responsible for the reconstitution of whole messages.
Another noteworthy aspect of MIL. STD. 1553 is its use of a non-data, packet framing
character. Each packet commences with a 3 bit synchronization pattern which violates the
usual Manchester encoding format employed with the data field and hence delimits the packet
in a unique manner. Figure 13-19 by contrast shows the HDLC frame format [25] which ac-
commodates the transmission of whole messages within the information field. The frame
delimiter is a binary “flag” which may well be matched by a binary pattern occurring in the
data field. In order to guarantee the uniqueness of the frame delimiter, HDLC employs a
technique known as bit-stuffing whereby every run of five consecutive binary 1’s has a zero
appended by the transmitter. This zero is automatically deleted by the receiver following a run
of five binary 1’s. Hence if the pattern 01111110 is ever detected by the receiver it must denote
the frame delimiter flag. It is generally held that the use of bit stuffing significantly reduces
the Hamming distance which is widely accepted as being an indication of the number of bit
errors that can occur in a transmission and still be detected. This weakness of bit stuffing

011111110 8 B IT S 8 BITS VARIABLE LENGTH 16 B I T S 01111110

FLAG ADDRESS CONTROL INFORMATION FRAME FLAG


FIELD F IE L D FIELD CHECK
SEQUENCE

Figure W-19. HDLC frame format.


316 I3 Interface Systems

stems from the fact that the encoding scheme involves memory of the preceding data patterns
thus, for example, allowing a bit error to be erroneously detected as a flag.
The typical frame or packet format of a link transport mechanism is also exhibited by
BS6556 [18] or Simple Asynchronous Protocol (SAP) as it is also known. Figure 13-20 shows
the presence of a frame delimiter, address and control fields, variable length data field and
a 16 bit check code for this byte oriented protocol. The uniqueness of the packet framing
character is guaranteed using byte stuffing where any data byte looking like the start of
message byte is preceded by an escaping character (binary loooOOOO). For obvious reasons the
binary pattern corresponding to the escape character must also be escaped in the manner of
a double negative.

SENT FIRST SMB START OF MESSAGE BYTE


AB ADDRESS BYTE
APPLICATION DATA
ADD
DESCRIPTION BYTE

ADF
LENGTH APPLICATION DATA
1 TO 128 F I EL0
BYTES

SENT LAST +I CHECK F I E L D BYTES


I 1 6 B I T CRCI
Figure 13-20.
BS6556 (SAP) packet format.

The need for bit or byte stuffing can be obviated if the frame format is either fixed or
preceded by a suitably unique (ie, non data) sequence. Both these techniques are employed in
the PROFIBUS frame format which takes as its basis the IEC TC57 FT1.2 frame formats [26]
shown in Figure 13-21. Use of these standard frame formats gives a Hamming distance of 4
which appears to have become established as something of a standard target figure for accep-
table residual error rate probabilities.

13.8.3 Error Checking

The degree of error checking present in link transport mechanisms varies considerably from
a simply parity check on the transmitted packet to the use of a cyclic redundancy check (CRC)
code of varying lengths. The detection of an error in an incoming data packet should initiate
13.9 Application Data Protocols 377

TOKEN

F I XED
LENGTH

FIXED
LENGTH
I 53 I DA I SA I FC I DATA I 8 BYTES I I CH I E 1

VAR I ABLE IS2IL / L 152 DAISA/FC( OATA


LENGTH
I

S = START DELIMITER FC = FUNCTION CODE


E = END DELIMITER CH = CHECK BYTE
DA = DESTINATION ADDRESS L = LENGTH
Figure W-21.
TC57 frame formats. SA = SOURCE ADDRESS

some form of error recovery, usually a request for re-transmission, to maintain the
transparency of the communications link as far as the application layer is concerned. Should
it prove impossible to correct the error (eg, due to complete loss of the link and a receiver
timeout) the transport mechanism has the task of providing as much information as possible
to the application layer to facilitate the diagnosis of the problem. From this point onwards
the autonomous nature of the transducer itself must be exploited to provide for some con-
tingency plan in the event of a communications failure.

13.9 Application Data Protocols

Above the application layer of the hierarchical model of a communications protocol is the
application data protocol. In the context of instrumentation this essentially consists of
transducer data in the form of sensor readings, actuator commands and general identification
and status information. Apart from defining the format in which these various data entities
are represented, the applications data protocol must also define the sequences of events
necessary to invoke the services offered by the autonomous transducers and any related con-
trolling elements. In terms of standardization this is perhaps the area that has seen the least
progress to date, which is not so surprising given the complexity of the situation. The defini-
tion of a standard set of services, and the mechanisms by which they are accessed, that will
cater for the instrumentation industry as a whole, is a daunting task. An initial attempt has
been made within MAP which defines a manufacturing message service (MMS) as the ap-
plication layer. For low level transducer use, however, MMS appears rather inefficient and
378 I3 Interface Systems

somewhat open to interpretation, factors which appear to have combined to make it relatively
unacceptable for instrumentation purposes. These shortcomings have been recognized
however and a presentation [27] based on a subset of MMS has been made to the fieldbus
working group in an effort to achieve efficient coding and timing of messages.
This work is expected to progress with an eventual standard for the application data pro-
tocol emerging. Early attempts have been made to define “function blocks” for a variety of
typical devices. The role of companion standards in this context is likely to be very important.
It is anticipated that once standards have been defined for process control use (eg, PID con-
trollers), a mapping between these function blocks and the application layer services will be
possible.
A good example of the capabilities of an existing applications data protocol is provided by
an examination of British Coal’s Application Protocol - Version 7 [28]. This well docu-
mented protocol was originally defined for use over British Coal’s extensive instrumentation
and control networks and was developed in conjunction with their major instrumentation
equipment suppliers. The protocol is particularly noteworthy in that it is used at all stages of
the instrumentation and control network. Low speed working underground with BS6556,
medium speed operations using HDLC and high speed computer to computer communica-
tions employing an Ethernet (CSMACD) type network all utilize the same applications pro-
tocol. The wide variety of transducers (both sensors and actuators) employed in the mining
industry has resulted in this application protocol being well suited to most other application
areas. The protocol defines both the transactions that are possible with particular transducers
and the format of the data messages necessary to implement the services provided. All
transducers are self-identifying and also notify precisely which services and data formats are
supported.
A key feature of the British Coal Applications Protocol is the changes only mode of opera-
tion whereby sensors typically only transmit readings in response to a poll when their data
values have changed by more than some predetermined amount. The protocol does, however,
support a request mode where a transducer can be forced into transmitting data. This may
be undertaken periodically, together with requests for configuration data which normally are
only required on system start up, for example, as part of a system integrity check. Transducers
generally also transmit status information, such as under or over range and alarm conditions,
and will usually be configured to notify any changes in status without being specifically re-
quested. The typical method of operation is’that once a transducer notifies a significant
change in status it is interrogated with a sequence of requests for data to provide more infor-
mation on the cause.

13.10 The Fieldbus Standard

As stated in an earlier section, the biggest drawback to the use of digital communications
in distributed instrumentation systems is the proliferation of different techniques and stan-
dards. In an attempt at resolving this difficulty the Instrument Society of America SP50
fieldbus committee invited the submission of proposals for consideration as an international
standard. In keeping with the spirit of this initiative most suppliers and users of fieldbus
equipment immediately expressed a committment to using the eventual standard.
13.11 Power Distribution 379

Submissions were received from Foxboro (modified HDLC), ERA (modified


MIL. STD. 1553), Rosemount (IEEE 802.4), FIP and the PROFIBUS group. Foxboro
withdrew their proposal at an early stage when it became apparent that much difficulty was
being experienced in reaching an agreement. This lack of agreement apparently stemmed from
the unwillingness of any of the interested parties to accept the commercial advantage clearly
to be enjoyed by any of the other parties whose submission might eventually be accepted as
the standard for fieldbus. A period of stalemate followed before a joint proposal was received
from Rosemount and Siemens setting out a strategy for a unified fieldbus [29].
It would appear that in order to achieve an agreement a compromise was required whereby
the final standard involved a significant amount of development effort on the part of all the
potential suppliers. Whilst possibly being fairest to the suppliers this additional effort
translates ultimately into increased cost and timescales with the result that the fieldbus is by
no means guaranteed as being a success. Previous attempts at defining industry wide stan-
dards would appear to have run into difficulties, for example PROWAY [30] which involved
significant effort on the part of the various standards bodies to produce, yet is not particulary
in evidence in industry.

13.11 Power Distribution

It is difficult to generalize on the subject of power supply within distributed instrumenta-


tion systems. The problem is that the amount of power required varies enormously with the
application. This is not just a function of the number of transducers employed but also
depends on the individual power requirements of the sensors and actuators themselves. For
example, a network of self generating sensors such as thermocouples requires very little power
for the purposes of excitation when compared with catalytic gas detectors, each employing a
heating element requiring up to 1 Ampere of current. In fact, the power budget of a system
is often the limiting factor as regards the number of transducers connected to a network since,
in theory, the only limit to the number that can share a communications medium is the size
of the address field used in the link level frame. Despite the difficulties associated, however,
the fieldbus specification (and also many of the commercially available transducer networks)
caters for multiplexed power and data over the bus. This is perhaps a legacy from the wide
usage of the self-powered 4 to 20 mA interface and a reluctance on the part of users to install
other than two core cabling.
Not withstanding the difficulty of deciding exactly how much power to make available it
is relatively easy to modulate a data transmission onto a DC power supply bus. In order to
retain galvanic isolation, however, many implementations use AC power and separate the
power and data with low pass and high pass filters. For certain applications it becomes imprac-
tical to provide power over the bus and local power supplies (including batteries) are used in-
stead. One application area that does pose severe constraints on the provision of power is
where intrinsic safety (IS) is required. IS is essentially a methodology whereby equipment
design techniques are employed to guarantee that, even in the presence of faults, insufficient
energy is released to ignite any flammable gases that may be present. Whilst the exact design
principles of IS are many and varied, a first order pTinciple is that large voltages should not
380 I3 intet$ace$ystems

be allowed to exist on equipment or wiring in the hazardous area (where flammable gas is pre-
sent). To ensure that a dangerousvoltage is not passed from the safe area (flammable gas not
present) along a bus, a zener barrier is used. Figure 13-22 shows the circuit arrangement of
a simple zener barrier, the main point to note is the series mistauce used to limit the current
in the pmeam of a fault condition. This is done to ensure that the energy capable of being
supplied to the hazardous area is limited to a level where ignition of flammable gases cannot
occur. Duplication of the zener diode, resistor combiiation is a precaution to protect against
a fault in either component.

0 I I 0
Elgum l3-22. IS Zener barrier.

The presence of the current limiting resistors and the relatively low voltage levels employed
in IS circuits generally places a significant limitation on the speed with which an IS distributed
instrumentation system can communicate data.

13.12 Design Considerations

Faced with the multiplicity of communications interfaces employed in distributed in-


strumentation fiystems it would appear rather daunting to have to choose which of them to
adopt for a t ~ product
w range Iadeed, the problem is likely to get worse for quite some time
until industry standards are established and commercial competition “weeds out” some of the
candidates. By adopting a modular design strategy, however, it should be possible for
manufacturers to alleviate the situation. There is no reason why once a transducer has been
developed it cannot be offered with a choice of communications interfaces. Thus the develop-
ment costs of a communicationsinterface can be recouped over a range of transducer pro-
ducts and similarlya new transducer can have the appropriatecommunications interface fitted
to offer it to all the various application areas. Figure 13-23 shows a block schematic of an
autonomous transducer employing both an instnunentation processor and a communications
processor. These two processors may well reside within the same integrated circuit, which is
increasingly the case with the ever escalatinglevel of integmtian and use of application specific
integrated circuits (ASICs).
Whether part of the same chip or not, the interface between the instrumentation processor
and the communications processor is the key issue. It is along this internal interface that the
13.13 References 381

INTERNAL
INTERFACE

INSTRUMENTATION COMMUNICATIONS COMMS . I / F


PROCESSOR PROCESSOR

APPLICATIONS
PROTOCOL

Figure 13-23. Modularity of autonomous transducer interface design.

split is made to change the back end communications module of a given transducer. The
physical implementation of the internal interface is of less concern than the actual data struc-
tures employed. This is because the physical implementation is usually a relatively trivial deci-
sion between shared memory and a direct (memoryless) parallel or serial communications
channel. The solution is to utilize a mapping of the users application data protocol to the ap-
plication layer protocol as the internal interface since it is at this level that the functionality
of the transducer is defined. Hence, communication between instrumentation processor and
communications processor takes place in the language of the users standard applications pro-
tocol, even if the output from the communications processor is nothing more than a digital
encoding of an analog voltage or current.

13.13 References

(11 Atkinson, J. K., “Communication Protocols in Instrumentation”, J Phys. E.: Sci. Insfrum., 20,
No. 5 (1987).
[2] “Analogue Signals for Process Control Systems - Part I Specification for direct current signals”,
ZEC381/1, 1984, International Electrotechnical Commission.
[3] Attwood, M. J., Wheable, D., Bond, D. F., “An Intelligent Digital Transmitter for Analogue
Transducers”, Proceedings Transducer Tempcon, London, 1983.
[4] Brignell, J. E., “Sensors in Distributed Instrumentation Systems”, Sens. and Actuators, 10, (1986).
[S] “Interface Between Data Terminal Equipment and Data Communication Equipment Employing
Serial Binary Data Interchange”, RS-232-C, 1969, Electrical Industries of America.
[6] “Implementation of V24 or RS232 as an asynchronous local interface”, BS6514, 1984, British Stan-
dards Institute.
[7] Atkinson, J. K., “Addressable Transducer Protocols”, Proceedings Capfeurs ’84, Paris, France,
1984.
[8] “Standard for Electrical Characteristics of Generators and Receivers for Use in Balanced Digital
Multipoint Systems”, EIA-485, 1983, Electrical Industries of America.
[9] Atkinson, J. K., Collins, R. P., Shahi, S. S., “Control of Air Powered Motors Using Distributed Pro-
cessing”, Proceedings Zncemads ’86, Romania, 1986.
[lo] “Interfacing with GDACS Protocol”, Technical Manual 05700-M-5009, 1989, Sieger Ltd.
[ll] Kimber, T., “Harnessing the PC to Data Acquisition”, Process Industry Journal, 3, No. 1 (1988).
[12] “HART Smart Communications Protocol”, Documentation Package No. :08700046. 1988, Rose-
mount Inc.
[13] “PROFIBUS proposal”, (DZN V19245), 1988, Instrument Society of America SP 50.
[14] “An Interface System for Programmable Measuring Instruments (byte serial, bit parallel)”, ZEC 625,
1988, International Electrotechnical Commission.
382 I3 Interface Systems

[15] “S-NET The Network for Distributed Industrial Data Acquisition”, Technical Report No. 015/84,
1985, Schlumberger.
[16] “Fieldbus proposal”, ISA SP50 Committee, 1988, Foxboro.
[17] Atkinson, J. K., “PC Based Instrumentation Systems”, IOP Short Meetings Series No. 5, (Trends
in Instrumentation) (1987).
I181 “Low speed digital signals for use in coal mines”, BS6556, 1985, British Standards Institute.
1191 “Avionic data transmission interface systems”, US MIL-STD-I553B/DEFSTAN 00- 18, US Depart-
ment of Defence.
[20] “FIP bus for exchange of information between sensors, actuators and controllers”, 1986, Le Club
FIP, 54011 Nancy, France.
[21] Schwaier, A., The Way to Open Systems Interconnection in Industrial Automation, Hardware and
Software for Real Time Process Control, Amsterdam: Elsevier Science, 1989.
[22] “Fieldbus, Functional Requirements Document”, IEC/E65C/ WG6, 1987, International Elec-
trotechnical Commission.
[23] “Reference Model of Open Systems Interconnection”, ISO/E97/SC 16, DIS 7498, 1983, Interna-
tional Standards Organisation.
[24] “Manufacturing Automation Protocol Reference Specification”, 1988, Dearborn, MI, Society of
Manufacturing Engineers, USA.
[25] “Data Communication - high level data link control procedures - frame structure”, IS0 3309,
1976, International Standards Organisation.
1261 “Telecontrol Equipment and Systems - Part 5.1 : Transmission Frame Formats”, Technical Commit-
tee N0.57,1987, International Electrotechnical Commission.
[27] Doebrich, U.,& Warrior, J., Fieldbus Application h y e c Presentation to ISA SP 50 working group,
Sept. 1988.
[28] Applications Protocol - Version 7, British Coal, HQ and Technical Department, Bretby, UK, 1987.
[29] Instrument Society of America SP 50 & International Electrotechnical Commission SC65C/WG6,
Proposal for Unified Centralized and Decentralized Fieldbus, Jan 1989.
[30] “Process Data Highway for Distributed Process Control Systems”, IEC 955, 1987, International
Electrotechnical Commission.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

14 Automotive: Onboard Sensors


ISEMI IGARASHI. Toyota Central Research and Development Laboratories.
Aichi. Japan

Contents
14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
14.2 Special Requirements of Onboard Sensors . . . . . . . . . . . . . . 384
14.3 Sensors for Engine Control . . . . . . . . . . . . . . . . . . . . 385
14.3.1 Flow-Rate Sensors . . . . . . . . . . . . . . . . . . . . . . . . . 385
14.3.1.1 Ultrasonic Flow Sensors . . . . . . . . . . . . . . . . . . . . . . 385
14.3.1.2 Solid-state Flow Sensors . . . . . . . . . . . . . . . . . . . . . 386
14.3.2 Pressure Sensors . . . . . . . . . . . . . . . . . . . . . . . . . 387
14.3.2.1 Semiconductor Pressure Sensors . . . . . . . . . . . . . . . . . . . 387
14.3.2.2 Capacitive Pressure Sensors . . . . . . . . . . . . . . . . . . . . 390
14.3.2.3 Negative-Pressure Sensors for Intake Manifold . . . . . . . . . . . . 390
14.3.3 Gas Sensors (Oxygen) . . . . . . . . . . . . . . . . . . . . . . . 391
14.3.3.1 Zirconia Oxygen Sensors . . . . . . . . . . . . . . . . . . . . . . 392
14.3.3.2 Titania Oxygen Sensors . . . . . . . . . . . . . . . . . . . . . . 392
14.3.3.3 Limiting Current Oxygen Sensors . . . . . . . . . . . . . . . . . . 392
14.3.3.4 Porous Coating 5 p e Sensors . . . . . . . . . . . . . . . . . . . . 394
14.3.3.5 Thin-Film Type Sensors . . . . . . . . . . . . . . . . . . . . . . 395
14.3.4 Engine Control Systems . . . . . . . . . . . . . . . . . . . . . . 397
14.3.5 Non-Contact Position Sensors . . . . . . . . . . . . . . . . . . . . 398
14.3.6 Torque Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . 398
14.3.6.1 Magnetoresistive Torque Sensors . . . . . . . . . . . . . . . . . . 400
14.3.6.2 Additive and Substractive Signal Processor . . . . . . . . . . . . . . 402
14.3.6.3 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . 402
14.4 Future Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . 403
14.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
384 14 Automotive: Onboard Sensors

14.1 Introduction

There is no doubt that the utilization of electronic technology is the key to the development
of automobile control systems. Among a number of electronic components, the capability of
sensors is a decisive factor in determining whether a system is of practical use or not.
Therefore, much attention has been directed towards the improvement of their durability and
reliability since cooperation between the fields of electronics and automobile mechanics has
been promoted and strengthened. Sensor technology plays a very important role whenever
electronics and automobile technology interact.
The substantial recent developments in semiconductor technology allow many applications
of microcomputers in nearly all fields, and a variety of microcomputers have also been in-
troduced to automotive equipment. However, there is still incompatibility among the elec-
tronics, sensor, and automobile technologies, probably owing to the differences in the environ-
ment in which each technology has been developed. Ruggedness is most important for
automobile systems, whereas electronic devices are appreciated for their precision and effi-
ciency under a limited range of environmental conditions. In sensor technology, ruggedness
has been improved and precision has been maintained. After all, the objective is to establish
more rational and reliable control systems by combining these three differing technologies and
extracting the best features from them.
So far, we have briefly considered the recent progress in this new field produced by merging
automobile technology with electronics. The greatest success of new automobile technology
combined with sensors and electronics is probably the development of efficient exhaust emis-
sion control systems and the discussion of new automotive sensors in this chapter will
therefore focus on this field. A perspective of other sensor applications relevant to automobile
technology will be given at the end of this chapter.

14.2 Special Requirements of Onboard Sensors

Sensors for automobiles, home appliances, and laboratory instruments are usually very dif-
ferent from each other in shape, specifications, and cost. The special features of automotive
on-board sensors are as follows:
Environment
A typical sensor used in an engine compartment is exposed to a temperature of -40 to
+150°C and has a maximum vibrational acceleration of 5-30 g. Moreover, the sensor may
suffer exposure to rain water, mud, salt, oil, and electromagnetic interference.
Reliability
Automotive sensors, like other automotive components, are required to be trouble-free and
to maintain their accuracy for 5-10 years.
cost
The cost of automotive sensors should be two orders of magnitude lower than those used
for other instruments.
14.3 Sensors for Engine Control 385

Resources
If an automobile is equipped with a sensor that uses 0.1 g of noble metal per sensor, the
noble metal consumed would be about one ton per year for lo7 sensors, which is the amount
for today’s U.S. domestic use alone. Therefore, the material of these sensors should be chosen
with due consideration of resource conservation.
The above requirements for automotive sensors are not easily satisfied, making their
development difficult. A sensor with the expected performance does not guarantee its prac-
tical success, but only signifies the beginning of its practical development.

14.3 Sensors for Engine Control

Almost three quarters of the total energy involved in an automobile, from its manufacture
to its scrapping, is consumed in traveling (fuel and oil). Thus, engine control has a decisive
influence on the valuation of automobiles. The automobile engine is directly exposed to a
variety of severe conditions such as temperature extremes, vibration, dust, and electrical noise.
Needless to say, engine control sensors are required to work under such conditions without
lowering their performance. The desired automobile sensors are listed in Table 14-1. Typical
sensors for the measurement of flow-rate, pressure, and exhaust gas concentrations are
described in this section as examples.

14.3.1 Flow-Rate Sensors

For the conventional carburetor-type engine there is no need for a flow-rate sensor because
the air-to-fuel ratio is automatically self-adjusted. On the other hand, an engine equipped with
an electronic fuel injection system employs a flow-rate method or pressure measurement
method for estimating the air volume taken into the engine. In the latter method the air
volume is calculated from the engine revolution and the negative pressure, measured at its in-
take manifold.
In order to achieve more accurate control of the carburetor-type engine, a feedback control
system with a precise fuel meter is under consideration. In this case, a fuel flow-rate sensor
is indispensable. Unfortunately, acceptable sensors for air and fuel flow-rate measurements
have not yet been developed, although a number of designs have been proposed and tested.
Some examples are described in the following subsections.

14.3.1.1 Ultrasonic Flow Sensors

Flow-rate is determined by measuring the difference between the sound speeds in both
upstream and downstream directions of the flow. A piezoelectric device is available for this
purpose [l]. Although it provides relatively accurate values of flow-rates, temperature correc-
tion is required. Another problem is the instability of the output at either low or high
flow-rates.
386 14 Automotive: Onboard Sensors

Table 14-1. Sensors required for automobiles in the future.

OBJECTS OF TYPE OF SENSOR


PURPOSE FUNCTION
DETECTION TO BE USED
Reliability Self-diagnosis Change with time Standard (time) sensors,
Automatic detection Deterioration Memory sensors
or correction
Safety Detection of abnor- Impossible to drive Engine
overall
mality (functional deteriora- Drive train
Self-diagnosis tion) Tire pressure’
Detection of obstacles (breakdown of major Brake pressure
parts) Intensity of electrolysis
Electromagnetic Ultrasound, Laser, In-
interference frared, Electromagnetic
Road condition wave
Object recognition
Anti-lock brakes Rotational speed of Hall plates
wheels Magnetoresistors
Crash Deceleration Silicon Accelerometers
Active Suspensions Acceleration Silicon Accelerometers
Lowering fuel Optimum A/F ratio Theoretical A/F ratio A/F, 0,, NO,, CO
consumption control (ignition tim- A/F ratio (realtime) Flowrate (air, fuel)
ing, exhaust gas) Knocking Pressure (absolute, gauge,
Lean combustion Drive train atmospheric)
Knock, Vibration, Ac-
celeration, Torque,
Displacement, Angle,
Revolution
Comfort Qualitative display of Vibration and noise Speed, Acceleration,
ride comfort around seats Hardness of seats
Sun-visor Width of vision Amount of light
Amount of total Skin temperature
sunlight
Room temperature Room environment Mean room temperature
(multipoints)
Humidity control Humidity, Dew-point
Fatigue detection Relation between Odor, Pulse, Blood-
fatigue and pressure, Breathing, Ox-
physiological ygen concentration in
parameters blood

14.3.1.2 Solid-state Flow Sensors

This is basically the same as the heated wire-type used for conventional flow-rate measure-
ment, except that the wire is replaced by a silicon micro-element [2, 31. The solid-state sensor
has better mechanical strength and higher sensitivity than the heated wire-type. Solid-state
14.3 Sensors for Engine Control 387

sensors can be manufactured by semiconductor technology, discussed in Chapter 5, allowing


large-scale production at lower cost.
This sensor can be used for the measurement of both air and fuel flow. The application to
fuel flow measurement is more advantageous than that to air flow measurement. Recently,
new types combining the heating and detecting elements in a chip have been reported [4,51.
Two possible structures are shown in Figure 14-1.Although there are some unsolved problems
with these schemes, it is expected that acceptable sensors of this type will be manufactured
by using advanced micromachining technology (see Chapter 5, Section 3).

Sensing
transistor Heating transistor
Air flow

la)

( b ) Air f l o w Resistor Contact

Figure 14-1.
Two implementations of a flowmeter.
(A) Transistors used for sensing
temperature. (B) Temperature sensed
through a Wheatstone bridge arrangement
of temperature-sensitive resistors.

14.3.2 Pressure Sensors


For automobiles it is required to measure the pressure of the intake manifold, atmosphere,
engine oil, brake oil, fuel, and tires, among others. Usually, the pressure is measured by detec-
ting the small change in the displacement or shape of the diaphragm and bellows. A differen-
tal transformer, capacitive gage, or strain gage is used to convert the displacement to electrical
signals.
A sensor consisting of a bellows and differential transformer has been used to measure the
pressure at the intake manifold. However, semiconductor and capacitive sensors are gradually
replacing this type because of its complicated structure and relatively high cost. Recently, a
new type of sensor which uses a surface elastic wave has attracted attention owing to its unique
feature of high-frequency signal output. Other sensors have been proposed and are still under
investigation. A representative one is based on the principle that a crystal oscillator changes
its frequency with applied stress. Another type uses a PZT piezoelectric device.

14.3.2.1 Semiconductor Pressure Sensors

The principle of this sensor [6, 71 is based on the fact that when a strong force is applied
to a silicon crystal, its electrical resistance changes significantly owing to the well-known
388 I4 Automotive: Onboard Sensors

piezoresistive effect. It is fabricated by forming a thin-layer semiconductor strain gage on a


silicon wafer using the diffusion technique. Here the single crystal silicon diaphragm itself
works as a pressure plate. Thus, the structure of this unit reduces the effect of creep and
hysteresis and permits the fabrication of a high-precision device. Since the semiconductor sen-
sor is produced by the same process as that for ordinary ICs, it has some advantages in quality,
productivity, reliability, and production cost per unit. Further, it is easy to integrate some ICs
with it since it is made of the same silicon material as an IC. Sophisticated technology now
makes it possible to combine ICs directly with this pressure sensor on the same silicon
substrate.
A typical structure of the basic semiconductor diaphragm is shown in Figure 14-2; Figure
14-2a shows the top view and Figure 14-2b the side view. In this example, the silicon substrate
is n-type with a (110) face. A thin-layer section which functions as a diaphragm is formed in
the middle of the substrate by photoetching from the reverse side. A desired thickness of the
diaphragm is easily obtained by controlled etching so that various sensors working in different
pressure ranges can be fabricated by the same process. Figure 14-2 shows four diffused gages
(resistors) connected to one another by low-resistance lead wires to form an electric bridge.
The diffused lead wires are extended to the outside of the diaphragm area and connected to

la1

Oiffused gauge Lead Ilayer \


Al terminal

(b)

Pressure in
Figure 14-2. Structure of a semiconductor pressure sensor.
14.3 Sensors for Engine Control 389

the aluminium electrodes which are made by the vacuum evaporation and located around the
edge of the substrate. All other parts except the electrodes of the diaphragm are covered with
a silicon oxide (SO,) or silicon nitride (Si3N4)layer to protect them from exposure to the at-
mosphere, enabling the device to be used in direct contact with a fluid, unless it impairs the
diaphragm surface.
To facilitate handling of the silicon substrate, it must be mounted on a base plate with the
proper thickness. The type and shape of this base material are important factors for perfor-

*
I Etched diaphragm I
I
I boundary I
I
I I
I I
I I
I I
I I
I Transverse voltage I
1 I
I strain gauge I

la)

Diffused
Enhancement strain gauge
Thermal oxide
passivation

Glass f r i t
Optional pressure p o r t
for differential sensors

(b)
Figure 14-3. Two possible structures of a Motorola “X-ducer” silicon pressure sensor.
390 14 Automotive: Onboard Sensors

mance. An example is shown in Figure 14-2b where a special crystallized glass is used as the
base plate. Its thickness is more than ten times that of the silicon substrate. This plate is fixed
to the substrate with low-melting-point glass by pressing it at high temperature. If a base plate
with no hole is fixed to the substrate in a vacuum, the etched hollow behind the substrate
forms a vacuum chamber which works as an absolute-pressure sensor based on the vacuum
reference. Figure 14-3 shows the top and cross-sectional views of another piezoresistive
pressure sensor on silicon.

14.3.2.2 Capacitive Pressure Sensors

The displacement due to pressure change results in change in the electrical capacitance. This
is the principle of the capacitive sensor. An example is given in Figure 14-4 [8, 91. Semiconduc-
tor sensors may also use the capacitive principle. A pair of quartz crystal plates with thin-layer
electrodes are placed in parallel 12 pm apart from each other and are sealed with a glass
material. The main electrode is located in the central part on one crystal plate and the
reference electrode is positioned around the edge of the other crystal plate. Both amplifying
and compensating circuits adjoining the quartz crystal capsule are incorporated into the
device. An accuracy of better than 1% has been reported in the temperature range from -40
to 125"C. This sensor is reliable and inexpensive because of the form of the enclosed quartz
crystal capsule, but certain problems remain such as its high impedance and instability in pro-
duct quality.

2.54 crn
lr-----l

Figure 14-4.
Cross-sectional view and electrode
pattern of MAP quartz capsule.

14.3.2.3 Negative-Pressure Sensors for Intake Manifold

For engine control, the negative-pressure sensor is one of the most important, as is the flow
sensor described above. In fact, much effort has been made to develop it, and the demand
for automotive semiconductor pressure sensors is rapidly increasing. To date, the number pro-
duced has reached more than one million per year. As mentioned in Section 14.3.2.1, this type
14.3 Sensors for Engine Control 391

of sensor excels in sensitivity, compactness, and response time, in addition to its high produc-
tivity, reliability, and low unit cost because it is produced by using conventional IC technology.
Figure 14-5 shows an example of these sensors for automotive use [lo]. A constant-current
electric bridge with a weak temperature dependence is formed on a silicon substrate by ad-
justing the amount of impurity added to the diffused gage (resistance).

Silicon chip
Sealed by solderin Case
Sealed by Vacuum chamber
electric welding
Hermetic s Stem

/
Sensing unit package
Hybrid IC

Wire harness

\ Pressure port
Figure 14-5. Negative-pressure sensor for intake manifold.

14.3.3 Gas Sensors (Oxygen)

Gas sensors for automotive use are classified into two types: one is used for the measure-
ment of oxygen concentration in exhaust gases and the other for the measurement of gas con-
centration inside the automobile. The former is installed in the emission control system to
reduce the amount of toxic exhaust gases and at the same time to improve fuel consumption.
The zirconia and titania sensors have already been used in automobiles to measure the
stoichiometric air-to-fuel ratio. A niobium oxide sensor is under development. Recently, lean
air-to-fuel ratio sensors have also been attracting attention for lean fuel combustion control.
392 14 Automotive: Onboard Sensors

Some of these are already in use. Other gas sensors such as smoke, humidity, and odor sensors
are required for the detection of the atmosphere inside automobiles. Exhaust gas sensors are
described in this section.

14.3.3.1 Zirconia Oxygen Sensors

Zirconia oxygen sensors are used in engine control systems with three-way catalytic con-
verters to detect and control the air-to-fuel ratio of engine exhaust gas [ll]. The output of
these sensors changes at an air-to-fuel ratio near the stoichiometric composition. The
characteristics of the sensor are influenced in a complicated manner by various parameters
such as gas composition, temperature, flow-rate, catalytic activity of the sensor electrode, and
its structure. Analyses of this sensor mechanism are still being made, although its development
is near completion [12-141.

14.3.3.2 Titania Oxygen Sensors

The principle of this sensor is based on the electrical resistance change of titania (TiO,)
through its oxidation-reduction reaction when exposed to exhaust gas. The resistance of the
sensor changes significantly near the stoichiometric air-to-fuel ratio [15]. The advantages of
this sensor are its simple structure, small size, and low cost. It does not require a reference
electrode as do zirconia oxygen sensors. Practical application of the titania sensor was delayed
when it was compared with zirconia oxygen sensors because of its inferior accuracy and
durability. Recently, however, development of the titania sensor has progressed and some
engine systems are now using it.

14.3.3.3 Limiting Current Oxygen Sensors

When voltage (current) is applied to a zirconia electrolyte cell, oxygen is pumped through
the zirconia electrolyte from the cathode side to the anode side because the carriers of the cur-
rent flowing through the zirconia electrolyte are oxygen ions. At low voltage, the current is
nearly proportional to the voltage. However, as the voltage is increased, the current shows
saturation due to a rate-limiting step in the transfer of oxygen “at the cathode”. The satura-
tion current is called the limiting current and is nearly proportional to the ambient oxygen
concentration. This phenomenon can be used for oxygen measurement [16]. The limiting cur-
rent of this type of sensor, however, is not stable because it is directly affected by the degrada-
tion of the cathode.
To overcome this disadvantage, a gas diffusion limiting device was attached to the outside
of the cathode [17, 181. An example is shown in Figure 14-6, where a cover with a pinhole is
mounted on the cathode side of a zirconia cell [18]. The cross-sectional area of the pinhole
is so small that gas diffusion through the pinhole becomes a rate-limiting step. As a result,
a saturated limiting current characteristic is observed, as shown in Figure 14-6b, where the
pumping current (the output of the sensor) is plotted against applied voltage for various ox-
ygen concentration in 0,-N, gas mixtures. In the measurement, the sensor was heated to
14.3 Sensors for Engine Control 393

Cathode 02 3 Oxygen
0 Concentration
1 7
\
QE I
\
4
.
a
C
L
-
L
L

.-
L

+
L
L
/ Anode
-a
U
Zr0230Z'-

Voltage / V
(a) (b)
Figure 14-6. Limiting current type oxygen sensor with pin hole. (a) Structure. (b) Current vs. voltage
characteristic.

700°C with a small heater surrounding the sensor. The limiting current of this sensor depends
only on the dimension of the pinhole and the ambient oxygen concentration. Therefore, the
value of the limiting current is not influenced by a slight degradation of the cathode.
In typical pinhole-type oxygen sensors, the cross-sectional area of the pinhole is small
enough for the gas diffusion in the pinhole to be a rate-limiting step, but the diameter of the
pinhole is so much larger than the mean free path of oxygen that ordinary diffusion dominates
and the Knudsen diffusion can be neglected. In this case, the limiting current, Zl,is given by
the relation

Zl= 4FDsct
L
4 1
1
1 - (C/C,) '
(14-1)

where F is the Faraday constant, L is the total length of the pinhole, C, is the total molar
concentration of gases, C is the molar concentration of oxygen, and D is the diffusion coeffi-
cient. This equation shows that the limiting current varies logarithmically with oxygen concen-
tration, but if C / C , is much less than 1, the limiting current varies almost linearly with the
molar concentration of oxygen as

Z1=-.
4FDS c. (14-2)
L

This almost linear oxygen concentration dependence of the limiting current type oxygen
sensor is more advantageous for measuring high oxygen concentration than conventional con-
centration cell type zirconia oxygen sensors which are insensitive at high concentrations,
although sensitive at very low concentrations of oxygen. For example, at atmospheric pressure,
the limiting current type oxygen sensor is more sensitive than the concentration cell type sen-
sors above 1 v01.-% of oxygen.
394 I4 Automotive: Onboard Sensors

14.3.3.4 Porous Coating 7Jpe Sensors

The cover with a pinhole which acts as an oxygen diffusion rate-limiting device, described
in the previous section, may be replaced by a porous layer. An example of such a sensor is
shown in Figure 14-7. Thin-film platinum electrodes were formed by sputtering on both sides
of a zirconia electrolyte disk 3 mm in diameter. After platinum wire had been bonded to the
electrodes, the cathode was coated with a porous spinel layer, which limits oxygen diffusion.
The anode was also coated with a thin, porous spinel layer to protect against toxicity and ther-
mal shock.

Porous Layer
\
Pt E'ectrode
/ Dense Ceramic
/ Coating

Figure
- 14-7.
PO0rO"S Layer Porous limiting current type oxygen sensor (prototype).

The limiting current of this porous coating type sensor can be represented as

(14-3)

by analogy with Equation (14-2) for pinhole-type sensors, where Deffis the effective diffusion
coefficient of oxygen, S is the area of the cathode, and L is the thickness of the porous layer
on the cathode. The current against voltage characteristic of the porous coating type sensor
is similar to that of the pinhole-type sensor shown in Figure 14-6b.
The primary advantage of this sensor is its rapid response, which is due to the inner space
on the cathode side being very small compared with the pinhole-type sensor. It also has other
advantages, such as small size, simple structure, and resistance to toxicity. These advantages
indicate its suitability for automotive applications. The sensor has been further improved and
is now in practical use in a lean combustion system.
The schematic structure of this sensor is shown in Figure 14-8 [20]. The electric current is
also limited by a porous coating layer on the outside of a closed-end tube-like zirconia cell.
A cylindrical ceramic heater is installed inside the zirconia tube to heat the sensor to above
650°C. The outside of the tube is exposed to the exhaust gas of an engine whereas the inside
is exposed to air. The construction of the sensor is similar to that of conventional zirconia
oxygen sensors for stoichiometric air-to-fuel ratio detection, hence both sensors can be
manufactured by common facilities using a similar process.
The current vs. voltage characteristic of the sensor is shown in Figure 14-8b for values of d,
where d is the air-to-fuel (A/F) ratio normalized by the stoichiometric A/F ratio. The limiting
current is roughly linear to the A/F ratio and reverses its sign when the A/F ratio changes
from lean (d > 1 :fuel deficient) to rich (d c 1 : fuel excess). In the lean combustion system,
the sensor is not used in the rich A/F region but only in the lean A/F region.
14.3 Sensors for Engine Control 395

Output -+l Ph
current
I IJ

Exhaust

+I Exhaust gas 7 0 0 O C

\
c
C

I I I I J
--0.5 0 0.5 1.0 1.5 2.0
Voltage / V
Figure 14-8. Porous limiting current type oxygen sensor (production type for lean burn engine control
system). (a) Structure. (b) Current vs. voltage characteristic at 700°C for various normalized
air-to-fuel ratios, h , in engine exhaust gases.

14.3.3.5 Thin-Film Tjpe Sensors

Recently, thin-film and micro-machining technologies have been actively applied in the
fabrication of various sensors. A trial has also been made for limiting current type sensors.
The structure of a thin-film limiting current type oxygen sensor is shown in Figure 14-9a [21].
The fundamental construction is the same as that of the limiting current type sensors already
396 14 Automotive: Onboard Sensors

Pt wires

Thin film Zr0,+Y20,

la)

Porous AI,O, substrate

Figure 14-9. Thin-film limiting current type oxygen sensor. (a) Structure. (b) Current vs. voltage
characteristic at 690°C for various oxygen concentrations (v01.-% 0,)in 0,-N, gas mix-
tures.

described. A thin-film platinum cathode electrode, zirconia electrolyte, and platinum anode
electrode are deposited in turn by sputtering on a porous A1,0, substrate. On the other side
of the Al,03 substrate, a thin-film platinum heater is deposited by sputtering. The porous
structure and porosity of the A1,03 substrate are designed so that oxygen transport is limited
by the gas diffusion process through the porous Al,O, substrate.
The current vs. voltage characteristic of the thin-film sensor in an 0,-N, mixture is shown
in Figure 14-9 b. The slope of the current-voltage curve is very steep. As a result, the measuring
range of oxygen concentration at a constant applied voltage is extended over 70 vo1.-070 O,,
as shown. The steep slope of the current-voltage curve is caused by the low resistance of the
zirconia thin film.
14.3 Sensors for Engine Control 391

Other advantages include the small size of the sensor chip (as small as 1.7 x 1.7 mm), low
electricity consumption of the heater, and excellent linearity of the output (limiting current)
against oxygen concentration.

14.3.4 Engine Control Systems

There are two kinds of feedback engine control systems which use oxygen sensors. One is
a three-way catalytic converter system in which the A/F ratio of engine exhaust gas is main-
tained at the stoichiometric A/F ratio, where the three-way catalyst has optimum efficiency
[22]. The other is a lean combustion system which uses a limiting current type oxygen sensor.
The purpose of this system is to improve fuel economy while keeping the emission level below
an allowable limit.
The concept of the system is illustrated in Figure 14-10 in comparison with the three-way
catalytic converter system in which the high NO, level of exhaust gas is reduced by the three-
way catalyst. In the lean combustion system, the NO, level is reduced by increasing the A/F
ratio over 20. An extreme increase in A/F ratio, however, causes misfiring in engines, which

II
leads to appreciable fluctuations of the engine torque and deteriorates the drivability.
Therefore, it is necessary to keep the A/F ratio within a limited range.

NOx (Before 3-way catalyst)

NOx (After 3-way catalyst)


NO,, CO

Allowable limit

Torque
Variation
Allowable limit

Fuel
Consumption
t
3-way catalyst Lean combustion
system system
Figure 14-10. Illustration of the concept of the lean combustion system in comparison with the three-way
catalyst system.

For this purpose, the limiting current type oxygen sensor shown in Figure 14-8 is used to
control the A/F ratio of engine exhaust gas in the lean combustion system. The system is
shown schematically in Figure 14-11 [23]. The sensor is installed in the exhaust manifold of
the lean combustion engine. The output signal corresponding to the limiting current of the
sensor is sent to the CPU and is compared with a target A/F ratio according to driving condi-
398 14 Automotive: Onboard Sensors

i I
Sequential injection

Swirl control valve

‘Manifold convene!

Ignition signal

‘i
Top dead center signal:
Ignition fail signal
Crank angle signal

F F ’
signa’=l
ECU

,
Speed signal
*- I Circuit opening relay

Figure 144. Lean combustion system.

tions to compensate the A/F ratio by fuel injection. Fuel economy was improved about 18%
from 14.4 to 17 km/l in a Japanese 10-mode driving cycle by introducing the lean A/F control
in a 1.6 1 car.

14.3.5 Non-Contact Position Sensors

There is much need for non-contact position sensing for automotive uses such as crankshaft
position, throttle position, engine speed, and potentiometer position. Hall elements, semicon-
ductor magnetoresistive elements, ferromagnetic magnetoresistive elements, electromagnetic
pickup, capacitive modules, optical modules, and Wiegand wires are usually used as non-con-
tact position-sensing elements. Of these elements, the electromagnetic pickup is the most
widely used. The sensor is low in cost, but the disadvantages are impossibility of static
measurement and variation of the output with rotational speed.
Semiconductor technology is also being applied to position sensors. A magnetic field sen-
sitive transistor is made by a standard analog IC process [24]. The device is essentially a lateral
p-n-p silicon transistor with two collectors. The difference in the collector currents depends
on the magnetic field. Its sensitivity is 100 times greater than that of the Hall element.
The ferromagnetic magnetoresistive element has the advantage of a small temperature
dependence and high sensitivity at low magnetic field. An example [25] of such a sensor com-
bined with integrated circuits is shown in Figure 14-12. It consists of a sensing part and an
IC part on the same silicon chip. The sensing part is composed of NiCo thin films deposited
by the electron-beam evaporation method. The IC part contains a differential amplifier and
signal-processing circuits for digital output.
14.3 Sensors for Engine Control 399

la)

co

Figure 14-12.
Integrated magnetic sensor. (a) Photograph
of top view. (b) Cross-sectional view.

The performance of the sensor is shown in Figure 14-13a, which shows the output waveform
of the sensor located 3 mm away from a small rotating magnet, as shown in Figure 14-13b.
The switching levels of the sensor output are 0.8 and 5.0 V source voltage.

14.3.6 Torque Sensors

Torque measurements by torque sensors will provide the appropriate control of power trains
and advancement of efficiency, energy-saving, and diagnosis in vehicles and industrial
machines [26].
Automobiles are run by means of power trains consisting of an engine, transmission, link,
differential gear, axle, and wheels. The torque generated in the engine is increased or decreased
and distributed to each wheel through the power trains. If torque sensors are installed at each
position of the power trains, they provide a precise and rapid response to power controls and
attain the ultimate performance potential of automobiles.
Measurement of the engine output torque involves a torque feedback control system, eg,
an appropriate MBT (minimum spark advance for best torque) system which effectively deters
the generation of NO, gases and knocking by choosing the best torque. Therefore, running
performance, specific fuel consumption, and drivability are improved by this sensor. A torque
sensor installed in an engine was reported by Fleming and Wood [27], who detected the engine
torque at a specially fabricated crankshaft of 22 mm extension.
The development of the torque sensor for practical use, however, has encountered some dif-
ficult problems in technology: (1) in compact front-wheel-drive and transverse-engine power
trains, it is unrealistic to expect that additional space can easily be made available for installa-
400 14 Automotive: Onboard Sensors

F’igure 1443.
(a) Output waveform in a magnetic field
induced by rotating a small magnet.
(b) Experimental arrangement for changing
the direction of a magnetic field.

tion of the sensor; and (2) ruggedeness and durability for heat, vibration and pressure are
essential for automotive use. However, the torque sensor has been developed with these pro-
blems taken into consideration. The features of this sensor are as follows:
1. miniature non-contact torque,
2. installation of the sensor in a last main bearing without elongation of the engine,
3. detection of mean output rorque,of the multi-cylindm engine by only one sensor.
A new signal processing method has been developed to obtain excellent linearity of output
signals in spite of using the commercial crankshaft as the measuring material.

14.3.6.1 Magnetomistive lbrque Sensors

The intra-bearing torque sensor is shown in Figure 14-14. The sensor consists of both
exciting core (laminated silicon steel) and pickup core (laminated permalloy), as shown in
F- 14-14a. The exciting and pickup cores are wound with 200 and 400 turns, respectively,
and mounted at right-angles to each other in a ceramic holder with heat-proof epoxy resin.
The sensor is as small as 12 x 8 x 16mm, as shown in Figure 14-14b. It should be emphasized
that the area of the sensor exposed to the shaft is 12 x 3 mm, where the length along the
thrust of the crankshaft is only 3 mm. The bearing cap as the sensor housing, shown in
Figure 14c, is 102 x 23 x 55 mm, which is larger than a commercial cap made of cast iron,
and is made of austenitic stainless steel SUS304 (in JJS code, equivalent to AISI304) non
magnetic material.
143 Sensorsfor Engine Control 401

(a

F@m 1414.
Intra-bearing torque sensor.
(a) Sensor head consisting of
exciting coil (large) and pickup
coil (small)(b) Form of the
sensor. Size: 12 x 8 x 16 mm.
(c) Torque sensor integrated
into bearing. Bearing cap size:
102 x 23 x 55 mm.

A commercial bearing insert consisting of cold rolled carbon steel strip SPCC (in JIS code)
and white metal, which has a cutout of 16 x 5 mm for sensor exposure, is used with the bear-
ing cap. A commercial crankshaft, 55 mm in diameter, made of cast nodular iron FCD7O (in
JIS code), which is approximately equivalent to ASTM A536-80grade 100-70-03 nodular iron
(spheroidal graphite cast iron), is used as the measurement shaft of ferromagnetic material.
402 14 Automotive: Onboard Sensors

The sensor and bearing cap are placed adjacent to the crankshaft with an air gap of
0.2 mm between the shaft and the poles of the cores. The exciting core of the present torque
sensor is perpendicular to the axis of the shaft because it is larger than the pickup core. The
durability of the bearing can and bearing insert was verified by installing it in an engine and
subjecting it to a full road test.

14.3.6.2 Additive and Subtractive Signal Processor

A block diagram of an additive signal processor is shown in Figure 14-15. A new signal pro-
cessor, called “the additive and subtractive signal processor”, consists of the basic parts and
the additive, subtractive, and compensating signal parts.

Figure 14-15.
Block diagram of additive and
subtractive signal processor. Amp.
Phase Adj.: amplitude and phase
adjuster of the additive signal.
Offset & Sens. Comp.: offset and
sensitivity compensator for
temperature and revolutions. Temp.:
TEMP. REV. temperature, Rev.: revolutions.

The additive signal part consists of an amplitude and phase adjuster and an adder. The ad-
ditive signal, derived from the exciting signal by adjusting both the amplitude and phase, is
added in AC mode to the sensor signal. As a result, the amplitude of the synthesized signal
only varies proportionally with the applied torque.
The subtractive signal part consists of a rectifier, LPF, subtractor, and DC voltage generator.
The rectified signal of the synthesized signal includes a DC offset voltage corresponding to
the additive signal. If the stability of the additive signal is not sufficient, the synthesized signal
drifts by an amount equivalent to the drift of the additive signal. Therefore, the rectified signal
of the additive signal is subtracted by the subtractor from the rectified signal of the syn-
thesized signal in DC mode, and the stabilized torque signal to obtained. At the same time,
an offset voltage generated in an electrical circuit is subtracted from the rectified signal of the
synthesized signal by the DC voltage generator and the subtractor.
The influence of temperature and revolution on the sensitivity and residual offset is com-
pensated by both the electromotive force of a thermocouple installed in the torque sensor and
the output voltage of a revolution meter.

14.3.6.3 Experimental Results

Sensor outputs are shown in Figure 14-16, where the revolutions are in the range
1500-4000 rpm in 500 rpm increments. At each revolution the torques were applied from the
14.4 Future Sensors 403

Figure 14-16.
Sensor outputs in the range 1500-4000 rpm in
500 rpm increments.

ECDY (eddy current dynamometer) controllable lowest value to the highest value at WOT
(wide open throttle).
Good linearity with an error factor of less than f 2 % of the full-scale value of 150 Nm
and hysteresis of 2% of full-scale were obtained, although the torque sensor was integrated
in an engine with considerable vibration and the magnetic properties were not uniform within
the commercial crankshaft.
Here, sensor sensitivity is defined as the value of the inclination of the best fitted line to
the sensor outputs and is expressed in mV/Nm. An offset output, zero torque signal, is defined
as the output extrapolated to zero torque along the best fitted line. The extrapolated value
divided by the sensitivity is the offset output, which is expressed in Nm. The sensor sensitivity
and offset output error have been estimated to be 0.16 mV/Nm and f15 Nm, respectively,
over wide ranges of temperature and revolutions.

14.4 Future Sensors

Considering future trends in automobile sensor technology, it seems essential to develop so-
called optimized sensor systems which allow the systematic control of safety, economy,
resource saving, and comfort. Sensors that will be required for automobiles in the future are
listed in Table 14-1 [28], as mentioned in the Introduction.
Some of the specifications for these sensors are described below:
1. Sensor with learning capability - The intelligent sensor must have the capability to
detect temperature effects, their changes with time, and data fluctuations, and to correct
automatically deviations from built-in standards. Figure 14-17 shows an example which is in
the initial stage of development. A pressure sensor and an arithmetic circuit are fabricated on
the same silicon substrate and the pressure-receiving section is located in its rear side [29]. This
is the first step towards realizing an intelligent sensor which allows both sensing and data pro-
cessing functions with one chip. Figure 14-17a shows the top view of the sensor and Figure
14-17b shows a cross-sectional view. The sizes 6f the chip and of the pressure-receiving section
are 3 x 3.8 and 1.5 x 2 mm2, respectively.
404 14 Automotive: Onboard Sensors

la)

PIEZO-
S i0; ,RESISTOR I nPn -Tr -
PnP Tr

I
SUBSTRATE

Ib) Si3N.l

Figure 14-17. Integrated piezoresistive pressure sensor. (a) Microphotograph of sensor. (b) Schematic
cross-section.

2. Sensors for detection of dynamicparameters - There is also a need for sensors that can
detect transient states of driving on a real-time basis. When an automobile weighing about
1 ton repeats starting, acceleration, cruising, deceleration, and stopping, the most economical
way to drive it is to select the most efficient range of engine revolutions. To achieve this, a
great deal of information, such as speed, road surface condition, driving torque, transferred
power, water temperature, and exhaust gases, must be collected continuously for overall engine
control. Sensors to detect these parameters will one day be mounted on every automobile.
3. Obstacle detection and environmental sensors - In order to improve safety and comfort
while driving, some of the sensing systems of the future will include a radar system capable
14.5 References 405

of recognizing the shapes of obstacles, a highly reliable humidity sensor, a dew-point sensor,
an odor sensor, and a total-sunlight measuring sensor.
New technology, new materials, and new principles for reliable sensors are expected to be
developed to meet the present needs of automobiles, while increased demands for advanced
automobile performance will, in turn, create new sensor needs.

14.5 References

[l]Rinolfi, R., SAE Tech. Pap. 770855, 1977.


[2]Rehn, L. A.. Tarpley, R. W., Wiemer, K. C., Duirkam, K. M., SAE Tech. Pap. 800128, 1980.
[3]Middelhoek, S., Angell, J. B., Moorlag, D. J. W.. IEEE Spectrum 17, No. 2 (1980)42-46.
[4]Johnson, R. G., Higashi, R. E., Sens. Actuators 11, No. 1. (1987)63-72.
[5] Nuijsing, J. H.,Schuddemat, J. P., Verhoef, W., IEEE i”hns. Electron Devices ED-29. No. 1, (1982)
133-136.
[a]Igarashi, I., Electron. Mag. 29, No. 6, (1984)17-22.
[7]Igarashi, I., ToyofaPhysical and Chemical Researrch Institute 30th Anniversary Report, 1970.
[8]Lee, C. Y., Pfeifer, J. L., SAE k h . Pap. 810374, 1981.
[9]Ko, W. H., Sens. Actuators 10, No. 3/4, (1986)303-320.
[lo]Kobashi, M., Kawamura, T., Nakano, J., Kawashima, Y.,Sugiura, J., J. SOC.Automot. Eng. Jpn.
36,NO. 10, (1982)1070-073.
1111 Gruber, H. U., Wiedenmann, H. M., SAE Tech. Pap. 800017, 1980.
[12]Saji, K.. Kondo, H., Takeuchi, T., Igarashi, I., in: Proceedings of the 1st Sensor Symposium,
Tsukuba, Japan, 1981, pp. 103-107.
[13]Anderson, J. E., Graves, Y. B., J. Electrochem. SOC.128 (1981)294-300.
[14]Cook, J. A., Hamburg, D. R.. Kaiser, W. J., Logothetis, E. M., SAE k h . Pap. 830985, 1983.
[U]Gibbons, E. F.,Meitzler, A. H., Foote, L. R., Zacmanidis, P. H., Beaudoin, G. L., SAE Tech. Pap.
750224, 1975.
[la] Ruka, R. J., Panson, A. J., US Pat. 3 691 023, 1972.
[17] Isenberg, A. 0..Jpn. Laying-Open Pat. 52-69650, 1977.
(181 Saji, K., ‘hkeuchi, T., Igarashi, I., Jpn. Laying-Open Pat. 52-7228e and Saji, K., J. Elektrochem.
SOC.W (1987)2430-2435.
1191 Saji, K., Takahashi, H., Kondo, H., m u c h i , T., Igarashi, I., in: Proceedings of 4th Sensor Sym-
posium, Tsukuba, Japan, 1984,p. 147.
[20]Kamq T., Chujo, Y., Akatsuka,T., Nakano, J., Suzuki, M.. SAE Tech. Pap. 850380 1985.
[21]Kondo, H.,Takahashi, H.,Saji, K.. Takeuchi, T., Igarashi, I., in: Proceedings 6th Sensor Sym-
posium, Tsukuba, Japan 1986, p. 251.
[22]Zechnall, R., Baumann, G., Eisele, H.. SAE l2ch. Pap. 730566, 1973.
1231 Matzushita, S.. Inoue, T., Nakanishi, K., Kato, K., Kobayashi, N., SAE Tech. Pap. 850044, 1985.
[24]Halbo, L., Haraldsen, J., SAE k h Pap. 8WJ122, 1980.
[25]Usuki, T.. Sugiyama, S., ’Lgkeuchi, M., Takeuchi, T., Igarashi, I., in: Proceedings of 2nd Sensor
Symposium, 1982, p. 215.
[26] Nonomura, Y., Sugiyama, J., Tsukada, K., ’Lgkeuchi, M., Itoh, K., Konomi, T., SAE l k h . Pap.
870472, 1987.
[27]Heming, W., Wood, P., SAE k h . Pap. 820206. 1982.
[a]Sugiyama. S., Takigawa, M., Igarashi, I., Sens. Actuators 4 (1983)113-120.
[29]A Handbook of Automotive Engineering 3, Society of Automotive Engineers of Japan, 1983,
pp. 1-5.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

15 Traffic Surveillance and Control


PETERDREBINGER.
Siemens AG. Munchen. FRG

Contents
15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 408

15.2 Traffic Measurement and Application . . . . . . . . . . . . . . . . 408

15.3 Vehicle Detectors . . . . . . . . . . . . . . . . . . . . . . . . . 409


15.3.1 Inductive Loop Detector . . . . . . . . . . . . . . . . . . . . . . 410
15.3.1.1 Design and Principle of Operation . . . . . . . . . . . . . . . . . 410
15.3.1.2 Characteristics and Applications . . . . . . . . . . . . . . . . . . 411
15.3.2 Ultrasonic Detector . . . . . . . . . . . . . . . . . . . . . . . . 413
15.3.2.1 Design and Principle of Operation . . . . . . . . . . . . . . . . . 413
15.3.2.2 Characteristics and Applications . . . . . . . . . . . . . . . . . . 414
15.3.3 Radar Detector . . . . . . . . . . . . . . . . . . . . . . . . . . 415
15.3.3.1 Design and Principle of Operation . . . . . . . . . . . . . . . . . 415
15.3.3.2 Characteristics and Applications . . . . . . . . . . . . . . . . . . 415
15.3.4 Passive Infrared Detector . . . . . . . . . . . . . . . . . . . . . 415
15.3.4.1 Design and Principle of Operation . . . . . . . . . . . . . . . . . 415
15.3.4.2 Characteristics and Applications . . . . . . . . . . . . . . . . . . 417
15.3.5 Magnetic Field Detector . . . . . . . . . . . . . . . . . . . . . . 418
15.3.5.1 Design and Principle of Operation . . . . . . . . . . . . . . . . . 418
15.3.5.2 Characteristics and Applications . . . . . . . . . . . . . . . . . . 418
15.3.6 Optical Detector . . . . . . . . . . . . . . . . . . . . . . . . . 418
15.3.6.1 Design and Principle of Operation . . . . . . . . . . . . . . . . . 418
15.3.6.2 Characteristics and Applications . . . . . . . . . . . . . . . . . . 419

15.4 Detectors for Priority Vehicles .................... 419

15.5 Outlook . ............................ 419

15.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423


408 I5 Traffic Surveillance and Control

15.1 Introduction

In view of the demand for a safe and smooth flow of traffic on our roads in the face of
ever increasing traffic volumes and a limited number of available roads, it is clear that
measures taken towards effective traffic management are indispensable.
Already within cities nowadays, management of road traffic is an absolute necessity in order
to ensure safe and effective movement of vehicles on the roads. Management of the traffic
on roads outside of conurbations, particularly on trunk roads, is without doubt a challenge
for the future not only to guarantee the safety and efficiency of road traffic but also in order
to meet pollution control requirements.
The most important prerequisite for traffic management is information about the traffic
flow and one way to supply the pertinent data is through the use of vehicle detectors.

15.2 Traffic Measurement Engineering and Application

Traffic measurement engineering supplies characteristic traffic parameters which describe


the prevailing traffic conditions. These parameters are then employable not only as a source
for collecting traffic data but also as a means for controlling traffic in accordance with
specific established methods. Thus, depending on the functional requirements and intended
application, a number of different types of characteristic traffic parameters have to be con-
sidered.
Vehicle detectors differ with respect to their physical principles of operation and technical
realization and thus, by their nature, supply differing characteristic traffic parameters. Before
dealing in detail with the various kinds of vehicle detectors available (see Section 15.3)’ the
nature of the numerous characteristic quantities which can be derived are first discussed
below.
Many vehicle detectors generate a vehicle occupancy signal as long as a vehicle is located
within the detector’s measurement range, whilst other detectors supply only a vehicle pulse
of predetermined duration for each vehicle that passes through the detector’s measurement
range. In turn, other verhicle detectors generate a train of pulses of undefined length when
the measurement range is passed over, the nature of this train being dependant upon specific
characteristics of the vehicle such as its base shape. As a result of their principle of operation,
some vehicle detectors can directly measure the vehicle speed and if this is only possible above
a minimum speed threshold, the detector will thus not detect stopped vehicles. The detector’s
evaluation unit mostly offers vehicle occupancy and vehicle pulse signals and occasionally also
the direction of travel as a binary signal (ie, states 0 and l), but whilst the vehicle’s speed may
be depicted by the duration of a binary signal, it may also be provided as an analog or digital
value.
The vehicle pulse is ideal as a parameter for actuation by vehicles, eg, for requesting a green
signal at a street crossing, and at the same time it can provide a suitable means for apportion-
ing the green light phase of a traffic signal based on the count of vehicles travelling in the
different directions. (N. b., for counting vehicles, the detector must issue only one vehicle pulse
per vehicle.)
15.3 Vehicle Detectors 409

The vehicle occupancy signal is slightly more versatile since it is suitable both for the ap-
plications described above but also as a means for generating further characteristic traffic
parameters, a few examples of which are:
- Traffic volume:
number of vehicles divided by the measurement time,
- Occupancy/time ratio:
total occupancy time divided by the total measurement time (the occupancy time is defined
as the time interval for which the signal of a vehicle detector exceeds a preset threshold
level when a vehicle passes by, thus the total occupancy time is obtained by summing the
individual occupancy times),
- Gap period ratio:
100% minus the occupancy/time ratio.
Vehicle speed, direction, cadtruck distinction and queue detection are frequently deter-
mined from the signals of two detectors located at intervals of a few meters in one direction
of travel.
Thus, vehicle speed is the reciprocal value of the travel time elapsing from entry into the
measurement zone of the first detector up to entry into that of the second detector and the
vehicle’s direction is determined from the time sequence of the detector signals.
In order to distinguish between cars and trucks, the vehicle length is determined by means
of the vehicle speed and the length of time the vehicle occupies a detector measurement zone,
and this is then compared with a preset car length limit. Hence, vehicles shorter than the preset
limit are indicated as cars, whilst longer vehicles are indicated as trucks.
To enable traffic queue detection, the time a vehicle occupies a detector measurement zone
is repeatedly compared with a preset time threshold. If the duration of the vehicle occupancy
signal is shorter than the threshold time, the traffic is detected as flowing. If it is longer than
the threshold time, a queue is detected and a corresponding message is issued.
Since the flow of traffic follows a random distribution it follows that the individual
measurements are frequently subject to extreme fluctuations. As such, it is often necessary to
smooth these measurements mathematically to provide a more accurate picture of mean traf-
fic flow, and in this the exponential smoothing method has proven itself to be very effective.
By this means, the previously obtained measurements are evaluated exponentially and in a
decaying form, so providing an up-to-date floating mean at all times:

Y,, = Yn-1 + a @ , -Y,,-*) (15-1)

Where:
jj,, new floating mean
jjnw1last floating mean
yn new current value
a smoothing factor (0 < a < 1)

15.3 Vehicle Detectors

Vehicle detectors exploit the typical characteristics of well-known physical effects in order to
detect vehicles. A sensor is used for measuring and determining any changes in this physical
410 I5 m f i c Surveillance and Control

quantity and then converting it to an electrical signal which subsequently the evaluation unit
is capable of interpreting as indicating the presence of a vehicle [8, 91. The most common
detectors are described briefly below, but more detailed descriptions of physical principles can
be found in the corresponding volumes of this series.
It is not always possible to make a clear distinction between the sensor and the signal pro-
cessing or evaluation unit. The term vehicle detector or, in short, detector is frequently
understood to mean the sensor with the evaluation unit.

15.3.1 Inductive Loop Detector

15.3.1.1 Design and Principle of Operation

The inductive loop detector consists of a wire loop laid in the road, which represents the
actual sensor, and an evaluation unit. An AC current, usually sinusoidal, in the freqency range
from approx. 20 to 150 kHz flows through the wire loop producing an alternating magnetic
field in the loop area. When a vehicle is located in the loop area, the alternating magnetic field
produces eddy currents in the vehicle’s metal parts and these eddy currents then cause an at-
tenuation in the alternating magnetic field with the result that the loop’s inductance drops and
its magnetic loss resistance increases.
The change in the loop inductance A L / L caused by the vehicle causes a frequency change
A f in comparison with the basic frequency f o :

Af -0.5 fo A L / L . (15-2)

For example, at a basic frequency f o of 50 kHz and with a loop inductance change A L / L
of 10% corresponding to A L / L = 0.1, a frequency change A f of -2.5 kHz will occur.
Figure 15-1 shows a typical loop inductance change as a result of cars and trucks passing over
the detector.
With the aid of these quantities, the evaluation unit detects a vehicle in the loop area by
means of frequency or phase shift measurement in the manner explained below.

-DISTANCE

I -DISTANCE

Figure 15-1.
Typical induction variation (AL) as a result of a
car and a truck passing over the detector.
15.3 Vehicle Detectors 411

When using the frequency shift mode (Figure 15-2), the inductive loop constitutes the induc-
tance of the generator’s LC tuned circuit which defines the frequency. The frequency change
caused by the vehicle only results in an indication that a vehicle is present when a preset
threshold is exceeded.
-
- -- -- e

Figure 15-2.
Inductive loop detector; fre-
quency shift (Anmode;
fT
nf-CALC. - OUTPUT
SIGNAL
-
When measuring the phase shift (Figure 15-3), the inductive loop is supplied with a fixed
frequency by the generator with a series resistor. The presence of a vehicle in the loop area
causes a phase shift between the generator voltage and the voltage through the inductive loop,
and it is this which is used to detect the vehicle.

-
-- -- 4

15.3.1.2 Characteristics and Applications

The inductive loop detector is the vehicle detector put to use most frequently. Most induc-
tive loop detectors generate a vehicle occupancy signal, but in some cases only a vehicle pulse
is issued.
As described in the previous section, two inductive loop detectors arranged in the direction
of travel frequently serve to detect speed and direction. One common arrangement to have
proven itself to be effective is to have inductive loops with a length of 2 to 3 m and a width
of approx. 1.75 m, whose front edges are 4 m away from each other. When a vehicle passes
over the two loops, the reciprocal value of the time t taken to travel the 4 m distance from
entry into the first loop to entry into the second loop corresponds to the vehicle’s speed, and
this is determined using the vehicle occupancy signals from both loops’ evaluation units.
When the distance between the front edges of the two loops is 4 m, the following relation
holds:
u (km/h) = 14.4/t (s) . (15-3)
412 15 Traffic Surveillance and Control

At the same time, the direction of travel is determined from the time sequence of the respec-
tive vehicle occupancy signals from the first and second inductive loops. Car/truck distinction,
however, is mostly based on a vehicle length measurement, in which case comparison with a
preset length limit serves to distinguish between vehicles similar to cars or trucks. According
to experience, however, characteristic structural features of cars and trucks, for example, their
differing ground clearances, are also taken into account in the decision in order to distinguish
adequately between them. The inductance change measured by the evaluation unit is well
suited to this purpose.
The dimensions of the inductive loop (sensor) essentially determine the range of a detector’s
zone of measurement, thus making planning and design quite straightforward from the traffic
engineering point of view. To detect all motor vehicles with the exception of motorcycles,
mostly rectangular inductive loops are used with a length of 2 to 3 m (in vehicle direction)
and a width of approx 1.75 m. These loops are also suitable for measuring most other
characteristic traffic parameters.
If shorter inductive loops are used, vehicle detection sensitivity is increased such that trucks
with trailers, for example, might be detected as separate vehicles; similarly, if loops of more
than 3 m in length are used there is a risk that two cars passing at only a slight distance from
each other will be detected as only one vehicle, thus a truck would be incorrectly indicated
if one were trying to distinguish between cars and trucks. Longer inductive loops of more than
5 m in length are rarely used but have been incorporated into some systems, for example, when
trying to detect a minimum distance between two vehicles on the basis of an unoccupied loop.
Inductive loop detectors are particularly suited to detecting vehicles in separate lanes as long
as the inductive loops of neighboring lanes are no closer to the lane separation line than ap-
prox. 60 cm.
Owing to the inherent design of inductive loops, their modus operandi, and the smaller sur-
faces compared to cars and trucks that are involved, shorter inductive loops are needed to
detect motorcycles and bicycles and loop lengths of about 1 m have proven themselves to be
effective here. Inductive loops with other dimensions and other shapes are, however, also used
for this purpose.
The inductive loop mostly consists of 3 to 6 turns of insulated copper cable (1.5 mm2) laid
in a groove 6 to 10 mm wide and approx. 6 to 8 cm deep that has been cut into the road surface.
The groove must then be cast watertight and the wire ends of the inductive loop twisted and
routed to the evaluation unit. If the evaluation unit is located at a distance of more than ap-
prox. 20 m away from the loop, then the connection must be made via a balanced telephone
line.
Steel mat braiding (road substructure) or other metal parts in the immediate environments
of the inductive loop lead to a reduction in the inductivity change caused by a vehicle and
this in turn requires a higher sensitivity in the evaluation unit, which may result in more
measuring errors occuring.
Inductive loop detectors have established themselves as reliable vehicle detectors. The most
frequent causes of malfunctions are interruptions or short circuits of the inductive loop or
supply line and defective insulation of both to earth. The evaluation unit should be protected
against voltage overloads caused by lightning.
15.3 Vehicle Detectors 413

15.3.2 Ultrasonic Detector

15.3.2.1 Design and Principle of Operation

The ultrasonic detector consists of an ultrasonic transmitter and receiver, most often in the
form of a jointly used ultrasonic transducer, and the evaluation unit. The transducer and the
evaluation unit are built into a common housing.
The ultrasonic transducer emits ultrasonic waves in a frequency range from around 30 kHz
to 100 kHz and these are reflected either off the background or off a vehicle in the detector’s
measuring area and then received again by the ultrasonic transducer. Two methods of vehicle
detection are used, either the pulse mode or the frequency shift (Doppler) mode.
When operating in the pulse mode (Figure 15-4), the detector evaluates the time A t elapsing
between transmission and reception of an ultrasonic signal and thus is able to determine the
distance d between the transmitterheceiver and the reflecting object according to the following
relation:

d = 1/2 c At, c = velocity of sound in air,


c = 331 m/s . (15-4)

When a vehicle is located in the measuring area, a distance shorter than that between the
transducer and the background is measured, thus the vehicle is detected.

-
TRANSMITTER/
GENERATOR SWITCH RECEIVER

~T-~FI w
CALC. OUTPUT
SIGNAL
at-CALCULATION LOGIC

Figure 15-4.Ultrasonic detector; pulse ( A t ) mode; T : transmitter, R :receiver.

The frequency shift (Doppler) mode (Figure 15-5) makes use of the Doppler effect, ie, the
frequencyf, of the received ultrasonic signal changes in relation to the emitted frequency f,
depending on the velocity u at which the vehicle reflecting the ultrasonic signal approaches
the detector or travels away from it. When a vehicle is approaching the detector, the frequency
f , of the received ultrasonic signal increases in relation to the emitted signalf,, whilst it drops
when the vehicle is moving away from the detector. This effect is described by the relation:

(15-5)

where up is the velocity component parallel to the line between detector and reflecting object.
up is positive when the object travels towards the detector.
414 15 TMffic Surveillance and Control

U
n
c
- ULTRASONIC
TRANSMITTER/
GENERATOR
fT

d-CALCULATION LOGIC

Figure 15-5. Ultrasonic detector; frequency shift (Anmode; T :transmitter, R :receiver.


This frequency shift in the received ultrasonic signal is used to detect vehicles and occa-
sionally also used to measure their velocity. The speed of sound in air depends mainly on the
ambient temperature, but there is also a dependency on the barometric pressure fluctuations
which can influence the measuring accuracy of both the pulse and Doppler modes [7].

15.3.2.2 Characteristics and Applications

Ultrasonic detectors operating in accordance with the pulse mode, are usually installed
above the road or by the roadside to detect vehicles in different lanes. They measure the
distance between the detector and the vehicle or background, and if the measured distance
changes, they perceive a vehicle and thus generate a vehicle occupancy signal.
Basically there are two ways of measuring the velocity of vehicles with an ultrasonic
detector.
In the first method, the detector would be of the pulse mode variety and velocity is com-
puted from the change in the distance between detector and vehicle over a given time interval.
With the second method, the ultrasonic detector operates in the Doppler mode and a
minimum speed of, eg, 10 km/h is required to detect a vehicle. Vehicles at rest are not detected,
ie, the detector supplies only a vehicle pulse signal.
The measured velocity u, is related to the actual velocity u of the vehicle by:

-
up = u cos a (15-6)

where a is the angle between the ultrasonic waves and the motion of the vehicle. In order to
maximize the frequency change made apparent in Equation (15-5), the ultrasonic waves should
be emitted at an acute angle a with respect to the direction of travel.
Modern ultrasonic transducers frequently consist of piezo-electric transducers which
generate very narrow ultrasonic lobes of only a few degrees and are thus capable of detecting
vehicles at distances of up to approx. 10 m.
Measuring errors with ultrasonic detectors can arise from a number of sources. Ultrasonic
signals generated by secondary sources may lead to measuring errors as well as disturbing
reflections of the emitted ultrasonic wave. Strong winds and thermal convection of air in-
fluence the propagation of ultrasonic waves and may thus cause measuring errors and lastly,
changes in the speed of sound caused by physical interferences also have an important bearing
on measuring accuracy.
15.3 Vehicle Detectors 415

15.3.3 Radar Detector

15.3.3.1 Design and Principle of Operation

The radar detector consists of a microwave transmitter, a microwave receiver, a transmitting


and receiving antenna and the evaluation unit. The antenna emits electromagnetic waves
mostly in the GHz range and these waves are then received by the antenna after they have been
reflected from the background or from a vehicle within the measuring area. Most radar detec-
tors operate in accordance with the frequency shift (Doppler) mode.
Equations (15-4) to (15-6) described for the ultrasonic detectors in Section 15.3.2.1 also
apply here, except in this case the constant c now represents the velocity of light (c, =
300000 km/s).
In a similar way to the ultrasonic devices, the presence of a vehicle is detected by measuring
the frequency shift between the emitted frequency f, and the received frequency f, and its
velocity is assessed using Equations (15-5) and (15-6). The pulse mode is also used with this
type of detector, but less frequently.

15.3.3.2 Characteristics and Applications

Radar detectors operating in Doppler mode measure the speed at which vehicles approach
the detector or move away from it. A minimum speed, usually of about 10 km/h, is needed
to detect vehicles in order to enable the detector to issue a vehicle pulse in addition to the vehi-
cle speed.
To guarantee a high degree of measuring accuracy, the microwave signal should be focused
as sharply as possible and be emitted at an acute angle with respect to the direction of travel.
Radar detectors operating in pulse mode measure the distance between the detector and
vehicle or background. The vehicle speed is determined on the basis of the change in distance
over a given time period and preset distance limits allow detection of a vehicle in the measure-
ment zone. In addition to vehicle speed, pulse mode radar detectors also supply a vehicle oc-
cupancy signal and in contrast to the Doppler mode detector can detect vehicles at rest.
Measuring errors can occur through disturbance of reflections and by extreme attenuation
of the microwave signal as a result of rain or snowfall but usually, radar detectors measure
the vehicle speed with inaccuracies of only a few percent.

15.3.4 Passive Infrared Detector

15.3.4.1 Design and Principle of Operation

Any warm subject emits energy in the infrared range (so-called thermal emission). The fre-
quency and intensity of this IR radiation depend on a large number of influencing parameters.
(Passive) dielectric infrared detectors like the one sketched in Figure 15-6 consist of an infrared
sensor for receiving this IR radiation coupled to an evaluation unit (amplifier, logic).
Modern dielectric infrared detectors mostly detect IR radiation with pyroelectric IR sensors
consisting of a dielectric material (eg, lithium tantalate LiTaO,) or polymer plastic (eg,
416 I5 mfic Surveillance and Control

--'B"'lTS
IR-RADIATION IR-SENSOR

AMPLIFIER/
LOOK:
-156
Sketch of a (passive) dielectric
infrared detector.

PVDF). Pyroelectric IR sensors evaluate the Variation of the intensity of the IR radiation,
which results in a temperaturechange A Ton the sensor. A surface charge is produced on the
sensor which is then tapped as the measuring voltage Au:

A u = c * A i7 c = pyroelectric constant [l] . (15-7)

The measurement zone of the IR detector is defined with the aid of an optical system. An
optical filter which only allows radiation within the range of approx. 8 to 14 pm to pass
through to the sensor serves to suppress interferences resulting from extraneous light. This
wavelength band represents an "atmospheric window" with minimum absorption of radiation
by water vapor and carbon dioxide in the air. Quation (15-8) below shows the relationship
that exists between the temperature T of the emitter and the wavelength with maximum
specific radiation, I:,

A, = 2889/ T [2]. (15-8)

Thus, the sensor's range of measurement from approximately 8 to 14 pm corresponds to


a temperature range of 361 K to 206 K or 88°Cto -67°C.
Figure 15-7 shows a modern design for an dielectric (passive) infrared detector and Figure
15-8 shows the geometry of the measurement field created when this infrared detector is
mounted on a roadside mast.

Figure l5-7. Design for a modem (passive) dielectric infrared detector.


15.3 Vehicle Detectors 417

Measuring field geometry of a typical arrangement


IDP detector

3
Installation
height
approximate11
* ‘
,/

7.
Figure 15-8. Measuring field geometry of a typical arrangement of an infrared detector. Upper part :side
view, Lower part : plan view.

15.3.4.2 Characteristics and Applications

The passive infrared detector only detects changes in the intensity of received IR radiation.
To guarantee this intensity change, the vehicle must pass the infrared detector’s measurement
zone at a minimum speed of, for example, about 6 km/h and usually the signal is in the form
of a train of vehicle pulses. The potential uses of this detector are limited, partly because of
the pulse train nature of the signal, and also partly due to the response and decay delays
caused by the sensor and evaluation method which usually fall within the range of a few
seconds. The passive IR detector is predominantly used to demand and terminate green traffic
light times since it is not suitable for counting vehicles and does not detect vehicles at rest.
As human beings also emit infrared radiation, passive dielectric IR detectors are suitable
for detecting the presence of people within a measuring zone which can be useful for re-
questing or extending the green traffic light phase at signalled pedestrian crossings, for ex-
ample.
Infrared detectors are installed above the road or at the roadside and are aligned to the lane
of a road to be detected and owing to the fact that the acceptance angle of the IR optics can
be limited to a few degrees, vehicles on multi-lane roads can in most cases be detected
separately for each lane.
418 I5 Traffic Surveillance and Control

15.3.5 Magnetic Field Detector

15.3.5.1 Design and Principle of Operation

The magnetic field detector consists of a magnetic field sensor which converts changes in
the intensity of the earth’s magnetic field or in an artificially produced magnetic field into
an electrical parameter. When a vehicle crosses over the magnetic field detector installed in
the road surface, the iron parts of the vehicle influence the earth‘s magnetic field in the detec-
tor’s measurement zone. The detector measures this change’in intensity of the earth’s magnetic
field, and thus the evaluation unit registers a vehicle as being present.
Modern magnetic field sensors operate with resistors or semiconductors sensitive to
magnetic fields whereas older magnetic field sensors mostly consisted of a rod-shaped coil
with several windings.

15.3.5.2 Characteristics and Applications

The magnetic field sensor is installed in the road at the measured location. This type of
detector has a number of potential drawbacks. Firstly, only ferromagnetic objects can be
detected and secondly, operation of the magnetic field detector may be influenced by exter-
nally induced interferences in the earth’s field or in the artificial magnetic field, caused for
example by iron masses or extraneous magnetic fields.
In addition, the highly localized registration zone of magnetic field sensors does not allow
the reliable detection of motorcycles and bicycles.
Depending on the evaluation mode, magnetic field sensors supply either a train of vehicle
pulses when the vehicle crosses over them, only a single vehicle pulse or even a vehicle oc-
cupancy signal, the signal type being defined by the intended application. Higher-level
characteristic traffic parameters can also be derived using several magnetic field detectors ar-
ranged like a chain in the direction of travel.

15.3.6 Optical Detector

15.3.6.1 Design and Principle of Operation

The optical detector (Figure 15-9) consists of a light source which emits light (mostly in-
frared) and the sensor which detects the transmitted light, as well as the evaluation unit
(amplifier, logic). When a vehicle is located in the path of light between transmitter and

LENS
-
SIGNAL LENS

-- -
- OUTPUT
SIGNAL
TRANSMITTER RECEIVER
PULSE- AMPL.1
GENERATOR LOGIC

Figure 15-9. Sketch of an optical detector .using diodes for infrared light emission and detection.
15.5 Outlook 419

receiver, the light fluxes are interrupted, the received signal disappears, and the evaluation unit
registers a vehicle. Usually pulsed or modulated light is transmitted in order to suppress
spurious light interferences and the optical detector’s measuring area can be easily defined
with the aid of optical systems to enable concentration and guidance of the light.
Optical sensors evaluating the light reflected from vehicles are not practicable because the
color and surface conditions of vehicles vary over such a diverse range.

15.3.6.2 Characteristics and Applications

Optical detectors generate a vehicle occupancy signal and therefore also detect vehicles at
rest. Using two optical detectors arranged in the direction of travel, vehicle velocity can be
measured very precisely and at the same time direction of travel be determined. Beyond that,
optical detectors are used to fulfill special tasks such as to check the permissible vehicle height
ahead of tunnel entrances.
Difficulties in achieving an effective configuration of optical emitters and receivers and the
risk of soiling of the optical systems pose important restrictions on their potential applica-
tions.

15.4 Detectors for Priority Vehicles

Constant developments in city traffic and ever increasing traffic volumes on limited road
space more and more strengthen the case for granting priorities to specific groups of vehicles,
(eg, police, ambulance) in road traffic; that is to say, these vehicles must be brought to their
destinations more quickly by means of particular traffic management measures, mostly by
achieving fewer stops and shorter stop times. This aim is fulfilled by equipment installed in
the priority vehicles themselves as well as with equipment installed at selected roadside loca-
tions along the route. Depending on the method employed, they are used to detect a vehicle’s
position and then to allow an exchange of data between that vehicle and a control center.
Usually, a transmitter installed in the vehicle transmits an identifying signal to the stationary
equipment (roadside beacon) along the route in order to allow detection of the special vehicle
and assessment of its current position. This is frequently followed by an exchange of data be-
tween the vehicle and the beacon.
An inductive transmission principle employing an inductive loop in the road in the fre-
quency range up to approx. 150 kHz frequently serves this purpose. Modern systems make use
of microwaves or also, to an increasing extent, of infrared radiation for transmitting the identi-
fying signal and data between the vehicle and stationary road marker.

15.5 Outlook

Thanks to the use of modern technologies, newer, more sophisticated sensors are coming
into being which exploit the possibilities offered by data processing to supply “intelligent”
statements concerning traffic flow. For example, image sensors coupled to image processing
420 I5 Traffic Surveillance and Control

circuitries ultimately will not only be capable of detecting vehicles, but will also be able to
see their characteristics and their drivers’ road behavior, thus supplying far more information
about traffic flow than is currently possible.
In addition, information and guidance systems for drivers are currently under development:
The information system informs the driver, for example, about traffic disturbances ahead or
about environmental influences along his route. The purpose of the guidance system is to in-
dicate to the driver the best route by which to reach his destination. Through an exchange of
data between the vehicle and a control center, the current position and intended destination
of each vehicle on a specific road network will be known. With this data, the optimum route
for each vehicle is calculated, taking the prevailing traffic situation into account, and is
transmitted to the driver’s vehicle as a suggested route. Navigation equipment belonging to
the vehicle is part of these guidance systems.
The intention of all these systems is to pursue the aim of taking the “workload” off the
driver in order to bring him more safely and quickly to his destination. At the same time, traf-
fic safety, pollution control, and economic feasibility criteria all play and important role. A
description of a few relevant European research and development projects is given below in
order to help to convey an insight into the nature of these ideas.
ARIAM
ARIAM is an information system for drivers which conveys information by means of
broadcast announcements (traffic radio) about the current traffic conditions on motorways

Figure 15-10. Components of the LISB.


15.5 Outlook 421

and?runk roads. The current traffic data is generated by means of inductive loop detectors
along the routes concerned and is transmitted to central computers which determine the traffic
situation data USBd as the basis for broadcast announcements 13).
LIS8 Guidance and Idormation System, Berlin
Beacons to enable the exchange of data between vehicles and the traffic guidance computer
are located on traffic signal equipment poles at selected locations in the road network. In-
frared signals serve as the transmission medium (Figure 15-10).
Before commencing the journey the driver informs the in-vehicle unit of his destination by
entering it on a keyboard. During the journey, a data exchange takes place between the vehicle
and traffic guidance computer in the beacons’ acquisition zones. The vehicle communicates
its destination to the traffic guidance computer and then receives guidance information con-
sisting of suggested optimum routes to the destination and data about the road network. After
this information has been processed, the driver receives recommendations on the display unit
about the best route to his destination and also other additional information. An example of
a display is shown in Figure 15-11.

Figure 15-ll. Display examples in a car using the LISB.


- The direction and distance to the destination are displayed at the start of the journey (A)
- Alternative routes are displayed during the journey (B)
- ‘‘’Run ahead” message with direction arrow and distance information from a bat graph (C, D)
- Lane recommendation with direction arrow and distance by a bar graph @)
- Depiction of an intersection with several approaches or a roundabout (F, G)
- Change to autonomous mode in the vicinity of the destination (H)

The limited number of beacons in the city area demands vehicle-integrated position detec-
tion which is implemented with the aid of a magnetic field sensor for measuring the direction
of travel, a wheel pulser for determining the distance, a position finding device and a naviga-
tion device.
Practical trials of LISB began in 1988 [3, 4, 51.
422 15 li-affic Surveillance and Control

A UTOGUIDE
AUTOGUIDE, a British system, has much in common with the LISB system described
above. The vehicle transmits data about its destination and the course of its journey to IR
beacons along the route and, from these IR beacons, receives route recommendations leading
to the destination which are indicated on a display in the vehicle.
A demonstration project comprising only five IR beacons and a few appropriately equipped
vehicles is under way in London. A pilot system involving 200 to 300 IR beacons and approx.
1000 specially equipped vehicles is planned. The project is to commence operation at the start
of the 1990’s and is to be expanded as a system financed by the drivers themselves.
PROMETHEUS
PROMETHEUS is the name of a research project undertaken by the European automotive
industry. It is intended to employ the most up-to-date technologies, electronics and telecom-
munications to make road traffic of the future safer, to take the workload off the driver and
to protect the environment.
Essential elements of the project are:
PRO-CAR
A computer-assisted on-board system for supporting and relieving the driver particularly
in critical situations.
PRO-CHIP
Development of technologies, computer architectures, and equipment for the vehicle for the
purposes of data acquisition and data processing.
PRO-NET
Systems for communication from car to car which will provide the driver with information
about the traffic situation in the direct proximity of his route, allowing him to obtain an over-
view of the situation himself.
PRO-ROAD
Development of communication and information systems between the vehicle and com-
puter at the roadside or in a control center with the aim of informing the driver about traffic
situations and guiding him to the destination with individual items of information [3, 51.
Travelpilot, Citypilot, EVA, CARIN
These are self-supporting destination guidance systems which require no equipment outside
the vehicle. An electronic compass, mostly a magnetic field sensor, determines the vehicle’s
direction of travel, wheel pulsers measure the distance covered and, from this data, the in-vehi-
cle computer calculates the vehicle’s position. A section of the road map (the data being stored
in a CD-ROM) or route recommendations maneuvering the driver to the destination appear
on a graphic display. At the start of the journey the driver will have communicated his destina-
tion to the in-vehicle computer on an input keyboard [3, 61.
15.6 References 423

15.6 References

[l] Siemens Forschungs- und Entwicklungsberichte, 15, No. 3, (1986) 105-114; Heidelberg: Springer
1986.
[2] Funkschau 4 (1982).
[ 3 ] Magazin 75 Jahre Automation, Wurzburg: Vogel, 1989, pp. 184-193.
[4] Strajenverkehrstechnik, 2 (1987), 35-37.
[5] Elektronikpraxis, No. 4, (1989), 92-95.
[6] Strujen und Verkehr 2000, Berichte 4, Berlin, 1988.
[7] Sorge, G., Hauptmann, P., Ultraschall in Wissenschaft und Technik, FrankfurtIM.: Harri Deutsch,
1985.
[8] Heywang, W., Sensorik, Heidelberg: Springer, 1988.
[9] Merkblatt Detektoren fur den Strajenverkehr, 1972, Forschungsgesellschaft fur das Stranenwesen,
Kdn, FRG.
[lo] Autoguide, Pilot stage proposds, 1988, Department of Transport, UK.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

16 Home Appliances
TETSUJIKOBAYASHI.International Superconductivity Technology Center
(ISTEC). Tokyo. Japan

Contents
16.1 Introduction ........................... 426
16.2 Sensors for Home Appliances . . . . . . . . . . . . . . . . . . . 426
16.2.1 Purposes of Electronic Control for Home Appliances . . . . . . . . . 426
16.2.2 Sensor ‘Ijrpes for Home Appliances . . . . . . . . . . . . . . . . . 421
16.2.3 Requirements for Home Appliance Sensors . . . . . . . . . . . . . 429
16.3 Sensor Applications in Home Appliances . . . . . . . . . . . . . . 429
16.3.1 Microwave Ovens . . . . . . . . . . . . . . . . . . . . . . . . . 429
16.3.2 Automatic Washing Machine . . . . . . . . . . . . . . . . . . . . 432
16.3.3 Electric Refrigerators . . . . . . . . . . . . . . . . . . . . . . . 434
16.3.4 Rice Cooker . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
16.3.5 Audio/Video Equipment . . . . . . . . . . . . . . . . . . . . . . 436
16.3.6 Miscellaneous Appliances . . . . . . . . . . . . . . . . . . . . . 439
16.4 Specific Considerations and Problems for Sensors Used in Home
Appliances . . . . . . . . . . . . . . . . . . . . . . . . . .
.. 440
16.5 Future Sensor Needs ....................... 440
16.5.1 New Technology Requirements for Sensors in Computer-Based Home
Appliances . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
16.5.2 Sensors in Home Automation . . . . . . . . . . . . . . . . . . . 441
16.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
426 16 Home Appliances

16.1 Introduction

The rapid progress in semiconductor technology, including microcomputers and VLSIs, has
increased the capabilities of electronic equipment used in industry. Microcomputers used in
electronic equipment have played a very important role in making equipment automatic, safe,
and more efficient. Microcomputer-based equipment now requires various kinds of sensors
capable of interfacing the microcomputers with external conditions such as temperature and
pressure, among others [l].
The categories of sensors, the world market, and the many applications of sensors are
discussed in Chapter 1. About 30% of all sensors produced are used in consumer products,
where the use of temperature sensors is dominant. Recent data reported by JEIDA (Japanese
Electronic Industrial Development Agency) shows that the production value of sensors in
Japan reached about 34 billion Yen in 1986 and the share of consumer sensors was about 20%
of sensor production [2].
Figure 16-1 shows the trends of adoption rates (the ratio of the number of families installing
consumer equipment to total number of families) of major consumer electronic products in
Japan. The adoption rate of refrigerators, washing machines, color TVs and carpet cleaners
is nearly 100%, and that for microwave ovens, VCRs and room air conditioners is rapidly in-
creasing.

Refr igerato
90 -
80-
70 -
60 -
50 -
40 -
30 -
20 -
10- Figure 16-1.
0- Trends of adoption rates of major
1960 1965 1970 1975 1980 1984(March)
consumer electronic uroducts in
(by Economic Planning Agency) Japan.

In 1981 microcomputers providing simple and smart controls were used in over 80% of
home appliances produced. These home appliances need various kinds of sensors to realize
many new requirements, such as comfort, convenience, automation, safety, energy and
resource conservation.

16.2 Sensors for Home Appliances

16.2.1 Purposes of Electronic Control for Home Appliances

Requirements for home appliances are comfort and convenience, high performance
(automatic), energy and resource savings, and safety [4]. Various kinds of sensors
16.2 Sensors for Home Appliances 427

and microcomputers have been introduced into home appliances in order to meet these
requirements.

Comfort and Convenience


Home appliances are used daily. Comfort and convenience are therefore demanded as im-
portant sales features for home appliances. For instance, a smart switch or remote control
channel selector for T V sets can be used by a bed-ridden invalid without the help of a nurse
or other person when he or she wants to watch TV.
Consumers want kitchen appliances to help make food tastier and preparation easier. A
smart microwave oven operated by sensors and microcomputers can cook a large frozen meal,
and the rice cooker can prepare rice during the night.

Automation (Intelligence)
It is necessary for homemakers to turn off the anti-sweat heat manually on conventional
refrigerators in order to maintain efficient operation. Consumers want high-performance
refrigeration where this function is done automatically and efficiently.

Energy Conservation
The energy shortages of 1973 and 1978 demanded more efficient products from the con-
sumer appliance market, as well as reductions in size and weight of these appliances. The new
home appliances are considerably more resource-efficient, using less electrical energy and
water. Sensors and microcomputers have been introduced into advanced home appliances like
refrigerators and washing machines in order to reduce the waste of electric power and water.

Safety
Obviously, consumers want to prevent accidents. Gas and humidity sensors are required for
improved environment protection [26]. The role of sensors for general environmental protec-
tion and monitoring is discussed in Chapter 20.
Safety sensors to detect carbon monoxide and fuel gas have been produced. When en-
vironmental conditions are dangerous, these sensors can detect unsafe conditions, and ventila-
tion fans will turn on automatically. Gas sensors can be integrated with appliances to shut
down the gas flow in an emergency [S]. These gas and humidity sensors can serve general at-
home needs to alert the occupants to unsafe conditions and can be sold separately to monitor
indoor air quality.

16.2.2 Sensor Types for Home Appliances

Table 16-1 shows sensors used in present day home appliances. Mechanical, radiation (op-
tical), temperature, chemical and magnetic sensors are used.
Temperature sensors, especially those of thermistors, are the ones most often used in home
appliances. The demand for chemical sensors;such as humidity and gas sensors, is increasing
in advanced electronic home appliances. The actual application of sophisticated sensors, such
I
CdS I
Photodiode
Phototransistor gg
gg
CCD image sensor
8'
G! a
MOS imaee sensor
Bellow
c
Humidity sensor
Gas sensor
Ion sensor
" C
Hall sensor
MR device
Hall IC
Thermistor
Bimetal
4I
Thermocouple d
Thermoferrite 33
*B
Shape memory alloy P2
"C
Infrared sensor a
Si transistor
16.3 Sensor Applications in Home Appliances 429

as CCD or MOS imaging sensors and IC sensors, is quite limited at present because of their
cost. However, as mentioned in the previous section, the demand for sophisticated sensors is
increasing.

16.2.3 Requirements for Home Appliance Sensors

Sensors used in home appliances should meet the requirements of low cost, reliability, small
size, low weight and easy handling. The cost of the sensor is the most important factor because
the price of home appliances must be kept within the range of other equipment sold by com-
petitors. To maintain sales levels, new features must be added to the appliances. The sensor
is an important means for realizing this requirement. Thus, since the sales price must be kept
low, the sensor cost must be very low.
Reliability is also important for electronic equipment used in home appliances because such
equipment is often used under demanding conditions. The sensor must maintain its original
characteristics under severe conditions.
Sensors used in home appliances must be small and light. Light weight and small size have
become a prevailing trend in modern electronic equipment, so the amount of space alloted
to the sensor is decreasing as well.
Easy handling is another requirement of computer-based home appliances. As mentioned
above, equipment costs must be low to compete in the market, production costs must be kept
down, and ease in sensor handling with respect to repairs is also an important factor in main-
taining competitive market prices.

16.3 Sensor Applications in Home Appliances

Sensor applications for microcomputer-based home appliances are growing. Microwave


ovens, fully automatic washing machines, electric refrigerators, small appliances and
audiohideo equipment are examples described as follows.

16.3.1 Microwave Ovens

Microwave ovens have become more automatic, using microcomputers and sensors. Various
kinds of sensors have been introduced to microcomputer-based microwave ovens (smart
microwave ovens) for controlling food cooking processes. In the early models of microwave
ovens, a needle type thermometer inserted into meat or other foods was used to control
cooking time, but it was plagued with various problems. With the development of sensor
technology from contact measurement to non-contact measurement, non-contacting sensors
were introduced in more advanced microwave ovens. Table 16-2 shows sensors used in
microwave ovens.
430 16 HomeAppliances

'IlIMe 162. Sensors used in smart microwave ovens.


Sensor purpose
Temperature sensor Food cooking control
Humidity sensor Food cooking control
Gas sensor Food cooking control
Infrared sensor Food cooking control
Weight sensor Food weight measurement

Cemmic Humidity Sensor


The ceramic humidity sensor is used primarily for nonantact food cooking control [a].
A photograph of a ceramic humidity sensor is shown in Figure 16-2. The sensing principle
is to measure the profile of the relative humidity versus the heating time of cooking food in
the microwave oven using a ceramic humidity sensor. The relative humidity rises rapidly when
the water contained in food reaches boiling temperature Using this boiling time data, a
microcomputer-based microwave oven can automatically control food cooking time
A ceramic humiditiy sensor used in microwave ovens requiresthe followingcharachteristics:
- high-temperature and humidity resistance,
- high sensitivity in a lower humiditiy atmosphere, and
- ability to withstand high temperature

Flgm 16-2. Photograph of a ceramic humidity sensor wed for microcomputer-based microwave
ovens. (courtesy of Matsushita Electrical Induetrial Company Ltd.)

Idmted Sensor
Another useful sensor used in microcomputer-basedmicrowave ovens is the infrared sensor.
Figure 16-3 shows a block diagram of a microwave oven using an infrared sensor [7].
16.3 Sensor Applications in Home Appliances 431

lempcraturc
Sensing Unit

Figure 16-3. Block diagram of a smart microwave oven using an infrared sensor.

A pyroelectric infrared (IR) sensor detects the infrared radiation emitted from cooked food,
then the signal from the IR sensor is fed to the electronic circuits and to the microcomputer.
The microcomputer processes those electronic signals and controls the food cooking condi-
tions according to preset cooking programs. The structure of the pyroelectric infrared sensor
is shown in Figure 16-4.
Advantages of the IR sensor are as follows:
- the surface temperature of food can be detected without touching the food,
- progress in the food defrosting process can be measured.

Infrared radiation

Figure 16-4.
Structure of the pyroelectric
infrared sensor used in a smart
microwave oven.

Typical characteristics of the IR sensor made from LiTaO, pyroelectric single crystal are
shown in Table 16-3. The sensor element is mounted into a TO-5 metal casing with a FET
amplifier chip, an output resistor and a load resistor. The sensor is very sensitive and has low
noise characteristics at room temperature.
432 16 Home Appliances

Table 16-3. Characteristics of pyroelectric infrared sensors used in smart microwave ovens.

Responsivity :Rv 200 - 300 V/W


Noise Equivalent Power : NEP 2 - 5 x 1 0 - ~W/HZ
Response time : T~ 0.25 sec
Operating temperature range -20 - 100°C
Temperature coefficient of Rv < 0.1%/ "C
Response wavelength 2 - 15 pm (Si window)
Sensitive area 3.14 mm2

A modulation-type pyroelectric IR sensor for use in a microcomputer-based microwave


oven was developed [8]. The sensor consists of an IR sensor chip and a slit plate vibrated by
a bimorph-vibrator. This sensor is considered to be more reliable than the conventional IR sen-
sor used for microwave ovens because it is less prone to drift effects.

16.3.2 Automatic Washing Machine

Washing machines are found in most homes. Housework such as washing clothes and clean-
ing rooms make up much of the time of a houswife's day. In order to save some of this time
and to minimize electric power and water consumption, a fully automatic washing machine
has been marketed.
Most electromechanical washing machine models require the user to manually select
washing conditions such as the degree of agitation, washing time, water level and weight of
clothes.
The microcomputer-based washing machine is operated by sensors in combination with a
computer program in which conditions such as washing time, degree of agitation, water level
and rinsing conditions are selected automatically.
Table 16-4 shows sensors used in fully automatic washing machines.

Table 16-4. Sensors used in fully automatic washing machines.

Sensor Purpose
Optical sensor Detection of water dispersion, water level
Pressure sensor Water level detection
Magnetic sensor Detection of rotation speed of agitator and/or induction motor
Rinsing sensor Water transparency measurement
Load sensor Detection of weight of clothes
Humidity sensor Detection of water dispersion of washed clothes

Water Level Sensor


In the microcomputer-based washing machine a water level sensor is used consisting of
LEDs (light-emitting diodes), a photosensor, and a light-slit that is movable by the water level.
Figure 16-5 shows a schematic drawing of a typical water level sensor [9].
16.3 Sensor Applications in Home Appliances 433

output
signal Power supply

Housing
Figure 16-5.
Schematic drawing of a typical
water level sensor used in a
microcomputer-based washing
machine. Water level can be
determined by photo-diodes,
LED and a movable light-slit
configuration. Air pressure pressure

When the washing drum is empty the slit rests at zero level. As water is poured into the
drum, the light-slit is raised by the water pressure to reduce the detected light. The new water
level is selected by the light-slit signal and is detected with the optical sensor. The photosensor
output is fed into the microcomputer, and the water is automatically stopped according to
the preset computer program. A mechanical diaphragm-type pressure sensor is also used
to control water level by measuring the hydrostatic pressure at the bottom of the water con-
tainer.

Rinsing Sensor
A rinsing sensor is used to detect the rinsing degree, that is, the concentration of residual
detergents in the rinse water. A schematic drawing of a rinsing sensor consisting of two
phototransistors (PT) and two LEDs is shown in Figure 16-6 [lo, 111. The sensor has two
optical paths of different lengths: IS and IR, namely the sample light-path and reference light-
path, respectively. The amount of incident light emitted from the LED into the photo-
transistor is proportional to the transparency of the rinse water. The rinsing condition can
therefore be monitored by comparing the phototransistor PTs and the reference phototran-
sistor FTr.
Another type of rinsing sensor is simpler, consisting of a photodiode and an LED [9]. In
this case, the rinsing condition is determined by the difference in water transparency before
and after rinsing.

Load Sensor
The washing process consists of two functions, the physico-chemical action of detergent and
the mechanical action of the washing machine. Detergent activity is proportional to the
mechanical action and washing time. Washing time must be reduced in proportion to the
amount of clothes to reduce the electric power comsumption of the washing machine. A load
sensor has been introduced for power control. A load sensor consisting of a rotation sensor,
434 16 Home Appliances

TOP View

Sample Ilght-pafh Bybbles Detector-caslng


/

Outer tub Figure 16-6.


Schematic drawing of a rinsing
sensor consisting of two photo-
Side View transistors and two LEDs.

a proximity switch and a rotating tooth disk directly connected to an agitator has been
reported [ll].
The speed of the agitator decreases as the wash load increases because an induction motor
drive is connected to the pulsator. The load torque of the induction motor or agitator varies
with the speed of the agitator. The washing load can be estimated from the measurement of
the agitator speed.

Other Sensors
A spin-dry sensor is also used in the smart washing machine [12]. The principle of the spin-
dry sensor is based on the fact that when water from clothes drips onto the surface of a
piezoelectric sensor the electrical voltage drops proportional to the impingement power of the
water. The sensing element of spin-dry sensors is usually made of piezoelectric elements such
as PZT ceramics.

16.3.3 Electric Refrigerators

In order to save electric energy and improve the performance of a refrigerator, sensors and
microcomputers have been introduced. The demand for energy-saving refrigerators is a major
problem in the home because the electric power consumption of the refrigerator is large.
To realize these demands, a refrigerator with electronic monitoring, self diagnosis, and
defrost-on-demand function has been introduced using sensors and microcomputers.
Table 16-5 shows sensors used in microcomputer-based refrigerators. The most important
sensor used in smart refrigerators is the frost sensor. Many types of frost sensors such as
temperature sensors, optical sensors and ultrasonic sensors have been used to detect frost for-
mation [13, 141. Smart refrigerators can monitor temperature, humidity and other conditions
leading to frost build-up and automatically defrost only when necessary.
16.3 Sensor Applications in Home Appliances 435

Table 16-5. Sensors used for computer-based refrigerators.

Sensors Purposes
Temperature sensor Measurement of atmosphere temperature, temperature of freezer and
refrigerator box
Frost sensor Detection of frost formation
Door check sensor Detection of opening or closing of refrigerator door

A conventional refrigerator has a switch to turn the ice heater off manually. Evaporator
coils are defrosted at regular intervals whether defrosting is needed or not.
In a computer-based refrigerator, the frost sensor detects frost formation and automatically
defrosts.
An optical frost sensor has been developed for computer-based refrigerators [13]. It consists
of a photodiode and an LED. Frost detection is carried out by measuring the change in
transmission light intensity between the photodiode and an LED.
A piezoelectric vibrator-type frost sensor consisting of a piezoelectric vibrator and FTC
thermistor has been developed [15]. The piezoelectric vibrator remains at the self-oscillating
stage using electronic circuits. When frost condenses the surface of a piezoelectric vibrator,
the resonance frequency of the vibrator changes. Frost formation can therefore be detected
by this resonance frequency change. The frost sensor is reversed to the initial stage by heating
with a FTC thermistor.

16.3.4 Rice Cooker

The traditional way to cook the rice in Japan requires five steps: first, the rice is placed in
water and heated over a low flame; second, the flame is increased until the water boils; third,
the heat is reduced to a point which will keep the water just boiling for several minutes until
most of the water is absorbed; fourth, the remaining liquid is boiled off with a quick applica-
tion of high heat, and finally, the pot is set on very low heat to keep it warm.
Now rice can be cooked using a smart rice cooker operating with sensors and a microcom-
puter. The temperature is monitored by a thermistor and the rice weight is automatically deter-
mined by measuring the speed at which the water temperature rises. In this case, there is no
need to measure the absolute value of the water temperature, but only its incremental value.
A block diagram of the control systems of a smart rice cooker is shown in Figure 16-7 [16].
The sensors used in these systems are very simple, such as thermistors or thermal read-swit-
ches. Precise cooking control is operated by software in which the knowledge of the cooker
is stored in the computer.

16.3.5 Audio/Video Equipment

Audio and video equipment such as color TVs, video casette recorders, compact disk players
and high-fidelity sound stereo radios are commonly found in today’s living room. The
436 16 Home Appliances

a
Constant output Temperature
voltage interlace sensing circuit

t
AC lOOV- magnetic Thermal
50 I 6 0 Hr Trans.
read-switch

heater 1[-? heater

Figure 16-7. Block diagram of the control systems of a smart rice cooker.

technical trends of audiohide0 equipment have progressed from analog technology to digital
technology. Digital equipment has excellent features such as low hysteresis, a wide dynamic
range and a high S/N ratio, which analog equipment cannot realize. Sensors and microcom-
puters play a very important part in making high-performance audiohide0 equipment.
Table 16-6 shows sensors used in audiohide0 equipment.

Table 16-6. Sensors used in audiohide0 equipment.

Sensor Purposes
Temperature sensor Cylinder temperature sensing
Magnetic sensor Motor revolution measurement, tape end detection, rotor position detection
Ultrasonic sensor Remote control
Position sensor Loading condition detection
Dew sensor Cylinder dew detection
Optical sensor Photo pickup, remote control
Electric sensor Capacitive pickup

Video Casette Recorders


Magnetic sensors and photosensors have been used to control the cylinder head and to
detect the start and end of the video tape for the auto-stop mechanism. In order to precisely
16.3 Sensor Applications in Home Appliances 437

control the servo motor which drives recording and play-back heads, magnetic sensors such
as Hall sensors and magnetoresistance sensors have been used.
It is very important to detect the dew drop formation on the cylinder head of a VCR because
a video tape is wound onto the cylinder head drum and the video tape can be damaged when
dew drops collect on the cylinder head. In order to protect the video tape, an electronic
conduction-type humidity sensor was developed as a dew detector for the VCR cylinder in
1978 [17]. The humidity-sensitive film is a cross-linked hydrophilic acrylic polymer in which
carbon particles are dispersed. Figure 16-8 shows the basic structure of a dew sensor [17].
The features of this dew sensor are as follows:
- high sensitivity in high humidity range,
- operable on a DC circuit, and
- simple electric construction.

sensitive
film
r
A

jll
i L------ I I- lnsurator
I I------- I

.-
A-- --.t
.

1
1

9 A

Figure 16-8.
Basic structure of a dew sensor used in a
VCR. - Leads -
Electrodes

Compact Disk Players


The sales of compact disk players (CD players) are growing rapidly in Japan and other
countries. Many kinds of advanced models have been introduced into the consumer market
[18]. The need for small, light, and high-precision optical pickups is increasing, since light
weight and small size have become prevailing trends in CD players. An optical pickup for an
advanced CD player consists of a low noise semiconductor laser diode, a silicon PIN
photodiodes and low-hysteresis actuator.
Figure 16-9 shows a schematic drawing of an optical pickup developed for a CD player. A
quadrant photodiode and a monitor photodiode are used as optical sensors. An optical pickup
has several servo systems for controlling the focusing of the laser beam on the CD disk and
for keeping the beam on track.
438 16 Home Appliances

Photosensor

Figure 16-9. Schematic diagram of the


optical pickup system
developed for a CD player.

High-Fidelity Stereo Sound Equipment


Magnetic sensors and photo-sensors have been used to make high-fidelity stereo equipment
intelligent.
To control the rotation rate of the turn table directly connected to the driving motor, Hall
sensors or photocouplers have been used. Sensors are used to measure the rotating speed and
phase signals of the motor.
A multi photosensor system is used to automatically select the record size. LEDs send light
onto the record disk through tiny holes underneath the record disk tray. The light is reflected
by the record disk onto photosensors. The sensors send signals to a microcomputer that
determines how the record disk is to be placed and activates the swing mechanism motor to
put the record disk in place.
The recorded music selection is also done by reflective type photocouplers. Photosensors
can detect the differences in recording groove density and select the desired music from the
music disk.

Remote Control Systems


Remote control systems have become more popular in consumer electronics and home ap-
pliances. T V sets, VCRs, room air conditioners, CD players, all have multifunction infrared
16.3 Sensor Applications in Home Appliances 439

remote control systems for error-free easy operation. 'Silicon photodiodes are used as sensing
elements.
Infrared remote control systems use about- 800nm infrared emitted from GaAs-LED as the
optical source. Infrared beams are modulated at about 40 kHz to eliminate operating errors.
At the receiver side, the signal received by the PIN photodiode is amplified and fed into the
microcomputer. The microcomputer analyzes the code pulses and operates the on-off power
switch, channel selection and sound volume control. They are controlled by one signal selec-
tion push button.

16.3.6 Miscellaneous Appliances

Microcomputers provide simple and smart controls for electric carpet cleaners, electric
clothes dryers, and other electrical equipment.

Carpet Cleaners

To control the motor revolution when the cleaner is unloaded and to freely select and con-
trol the sucking power of home vaccum cleaners, sensors and computers have been introduced.
Principal among sensors used for computer-based carpet cleaners are pressure sensors and
air massflow sensors. Silicon pressure sensors and conventional metal diaphragm pressure sen-
sors are also employed.
Si pressure sensors have been introduced in home cleaners in order to control the motor
power by measuring the difference between the atmospheric pressure P, and the air intake
tube pressure P, [19]. Si pressure sensors developed for this purpose have good linearity, easy
combination with IC circuits, long life, and high reliability. Figure 16-10 shows a schematic
diagram of a home cleaner using a Si pressure sensor [19].
An air mass flow sensor in an air vane and potentiometer is reported. The principle of this
sensor is very simple. A vane directly connects to the potentiometer shaft. The angle change
of the air vane is proportional to the air mass flow. Because the air vane is directly connected
to the potentiometer shaft, the change of the air vane angle adapts to the change of resistance

Dust filter Motor

sensor

Figure 16-10. Schematic diagram of a home cleaner using a Si pressure sensor.


40 16 Home Appliances

of potentiometer and an electrical signal proportional to the air mass flow can be detected.
This type of sensor is low in cost and has a high-level output. However, this sensing device
has moving parts, resulting in hysteresis. Therefore, a solid-state air flow sensor might be used
in an advanced home cleaner.

Electric Clothes Dryers


Electric clothes dryers have some built-in intelligence using sensors and microcomputers.
The clothes dryer controls the small amount of moisture in the drum and the temperature of
the exhaust air. The humidity sensor and the temperature sensor, combined with the
microcomputer, control the drum speed, temperature and flow of the drying air.

16.4 Specific Considerations and Problems for Sensors Used in


Home Appliances

In Section 16.3 some typical sensor applications in home appliances were described, but
many other automatic devices and appliances for use in the home have been created by the
use of sensors and mircrocomputers. The important features of sensors used in home ap-
pliances are good cost/performance ratio, high reliability, and a long maintenance-free period.
These sensors must be fabricated from safe, non-toxic materials which will remain non-hazar-
dous after the sensors are destroyed or exposed. However, in order to reduce cost, the sensors
must be made from materials appropriate for mass production that can be processed with
established technology. These problems must be solved with new materials, technology, and
design in the near future to assure safety, reliability, and low cost.

16.5 Future Sensor Needs

16.5.1 New Technology Requirements for Sensors in Computer-Based Home


Appliances

Sensors like those described in the previous sections are used in commercial home ap-
pliances with microcomputers as control units.
To improve computer-based home appliances, the demand for low-cost and high-perfor-
mance (intelligent) sensors is increasing. Research and development on semiconductor sensors
used in home appliances has moved into more sophisticated high-technology areas. Table 16-7
shows the technical requirements for sensor development in future home appliances.
A microchip FET humidity sensor is an example of sophisticated semiconductor integrated
sensors [20, 291. Using IC technology, a humidity sensor element and a transistor-type
16.5 Future Sensor Needs 441

Table 16-7. Technology requirements for sensor development in future home appliances.

Heavy duty environmental protection : High reliability sensors


Improvement of sensor accuracy and linearity
Integration of sensors and related electronic circuits : IC sensors, low cost sensors
Standardization of sensors : Easy handling
Development of new sensors (torque sensor, accelerometer, acoustic sensor, pattern information sensing
device, smell sensor, taste sensor etc.)

temperature sensor are fabricated on a silicon chip, and high sensitivity and long-term stability
of sensors are realized.
An olfactory sensor, which consists of multichannel ceramic gas sensors and a micro-
processor, is another example of intelligent sensors [21, 301. It is considered that this approch
to intelligent sensors is very important to obtain sophisticated smell and taste sensors which
could be used in advanced microwave cooking ovens and other home appliances in the future.
The sensor technology of chemical sensors such as humidity, gas, smell, and taste sensors
lags behind the technology of physical sensors such as thermal, optical, magnetic and pressure
sensors. Research and development of chemical sensors must be accelerated to meet future de-
mand.
Home appliances could be improved by simplified operation so they can be operated by
even the least experienced users. A voice recognition sensor is expected to be used in advanced
home appliances. Synthetic speech technology has been introduced to confirm commands or
set the operating processes of microcomputer-based equipment. Voice recognition sensors
combined with synthetic speech technology may soon be available for remote-control
microcomputer-based home appliances.

16.5.2 Sensors in Home Automation

Home automation has been introduced to modern households [22-241. The computer,
which functions as the brain and control device, and sensors imitating the functions of the
five human senses are the most important elements in any home automation system.
Figure 16-11 shows a block diagram of a home automation system [23]. It consists of house
control function, energy control function, security function, and communication function.
The system has the capacity to turn lights on and off, adjust heating and cooling for max-
imum comfort and energy conservation, and it can also fill the bath if signalled by telephone.
Supervised wireless systems, in which sensors communicate with controllers via radio, offer
still more convenience, since they can be installed without expensive wiring. For example,
house surveillance, both outside and inside the home, can be achieved with microwave sensors,
ultrasonic sensors, photoelectric sensors, and passive infrared motion detectors.
Various kinds of sensors are needed in order to operate home automation systems effi-
ciently. These are listed in Table 16-8.
442 16 Home Appliances

Main
controler

Sensors

Air conditioning

..
I
Home telephone Sub- telephone
Main
telephone 1: Bath sensor, 2:Therrno sensor.3:Srnoke sensor.
&:Ventilation sensor, 5: Security sensor,
6:Gas sensor ,l:Earthquake sensor, 8: Electric key

Figure 16-ll. Block diagram of a home automation system. Main controller receives information from
various sensors and messages are given by displays or synthetic voices.

Table 16-8. Sensors required for home automation.

I Functions
~~
Sensors
temperature, humidity, wind flow, taste, cooking condition, defrost,
House control frost, water level, dust, rinse, weight, pressure, vision, gas, flame,
gas leak

temperature, infrared, smoke, flammable gas, earth quake, electric


Home security leakage,
over heating, vibration, window/door ultrasonic, voice recognition,

electric power, voltage/current, flow (gadwater), temperature, water


Energy control
level, water freezing, sun, illumination

The required characteristics of sensors used in home automation systems are as follows:

- low cost (the sensor itself and related electronic circuits),


- high reliability and long life (maintenance-free),
- small size and light weight,
- low electric power dissipation.

The home automation system will provide home security, including, eg, the ability to dial
fire or police stations when accidents occur.
16.6 References 443

16.6 References

[I] Fuji Communication Co., Report on Sensors and Their Applicationsfor Electronic Equipment, 1981
(in Japanese).
(21 Report of Electronic Industries Association of Japan, 1986 (in Japanese).
131 Adria, P., Sensors 3 (1986) 16.
141 Kobayashi, T., Sens. Actuators 9 (1986) 235-248.
[5] Kawase, A., Nut/. Tech. Rep. 30 (1984) 111-118 (in Japanese).
161 Nitta, T., Proc. 1st Sensor Symp. (1981) 75-84.
[7] Fukuda, N., J. Microwave Power 17 (1982) 267.
[8] Yokoo, T., Proc. 4th Sensor Symp. (1984) 92 (in Japanese).
[9] Nakano, H. et al., Nut/. Tech. Rep. 30 (1984) 4-10 (in Japanese).
[lo] Sato, H. et al., Toshiba Rev. 34 (1979) 983 (in Japanese).
1111 Matsuo, K. et al., IEEE Trans. Ind. Appl. IA-20 (1984) 1171.
1121 Matsumoto, K. et al., Sharp Tech. J. 30 (1984) 185-189 (in Japanese).
[13] Suzuki, K., Digest of Technical Papers, Transducers '87 (1987) 69-72.
1141 IEEE Spectrum 22 (1985) 81-85.
1151 Isogai, H., Nut/. Tech. Rep. 29 (1983) 2 (in Japanese).
1161 Toshiba Rev. 40 (1985) 2499.
1171 Ishida, T. et al., Natl. Tech. Rep. 24 (1978) 436.
[I81 Nabeshima, D. et al., Toshiba Rev. 40 (1985) 97-100 (in Japanese).
1191 Takahama, K., Denshi Zairyo 21 (1982) 92 (in Japanese).
1201 Hijikigawa, M. et al., Tech. Digest of Transducer's I985 (1985) 220.
1211 Ikegami, A. et al., ibid p. 136.
1221 Yata, K., Control and Measurement 23 (1984) 913 (in Japanese).
1231 Matsukura, T. et al., Toshiba Rev. 39 (1984) 689 (in Japanese).
1241 Fishetti, M. A. et al., IEEE Spectrum 25 (1985) 36-50.
[25] Joseph M. Giachino, Sens. Actuators 10 (1986) 239-248.
1261 Solid State Gas Sensors, Moseley, P. T., Tofield, B. C. (eds.); Bristol, UK: Adam Hilger, 1987.
1271 Miyoshi, S. et al., Digest of Technical Papers, Trunsducers '87 (1987) 309-311.
1281 Matsumura, S. et al., Proc. Ist Sensor Symp. (1981) 205-208.
1291 Yamamoto, T. et al., Digest of Technical Papers, Transducers '87 (1987) 658-660.
1301 Bott, B. et al., Digest of TechnicalPupers, Transducers '85 (1985) 128-131.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

17 Automated Manufacturing
INPRO. Berlin. FRG
J ~ G E ROCOS.
N

Contents
17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 446

17.2 Sensors and Sensor Systems in Automated Production . . . . . . . . . 447


17.2.1 General Classification . . . . . . . . . . . . . . . . . . . . . . 447
17.2.2 Distance Measurement . . . . . . . . . . . . . . . . . . . . . . 447
17.2.3 Contour Tracking . . . . . . . . . . . . . . . . . . . . . . . . . 451
17.2.4 Machine Vision . . . . . . . . . . . . . . . . . . . . . . . . . . 454
17.2.5 Measurement of Process Parameters and Machine Diagnosis . . . . . . 458
17.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
17.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
17.3.2 Fettling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
17.3.3 Spot Welding/Arc Welding . . . . . . . . . . . . . . . . . . . . . 462
17.3.4 Manipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
17.3.5 Inspection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470

17.4 Current Problems of Sensor Applications . . . . . . . . . . . . . . 471


17.5 Sensor Market Trends . . . . . . . . . . . . . . . . . . . . . . . 473

17.6 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473


446 17 Automated Manufacturing

17.1 Introduction

Competition among manufacturers on the international market puts pressure on produc-


tion costs. The needs of customers lead to a steady growth in the variety of product-types and
special options causing a decrease in the number of units issued per model. Rapid progress
in terms of innovations results in a shorter life time of the products. Because of these in-
fluences, there will be strong demands on the factory of the future, first concerning higher
levels of productivity with an increasing degree of automation and availability and second
regarding greater flexibility of output and model types.
In order to cope with these challenges, the factory of the future has to link the flows of
material, energy and information together more effectively. This interconnection has been
described by the concept of Computer Integrated Manufacturing (CIM).
Sensor data from all areas of the plant provide an essential support or in some cases act
as a fundamental requirement for Computer Integrated Manufacturing. This data flow cannot
be provided directly by man in either the time required or with the required complexity.
Therefore, the most important engineering task in coming years will be to link the sensor level
with the different levels of data processing, as illustrated in Figure 17-1 [l].
The main tasks of sensors will be to facilitate automated production processes by compen-
sating for unexpected disturbances, unavoidable tolerances of the workpieces, and per-
manently changing environmental conditions. Another wide field of application for sensors
is in automatic quality assurance. Errors should be recognized in the shortest possible time.
The goal of further research is to eliminate errors by means of adaptive process control.

Plant coordinating
level

Control level

Sensor level
Swl,cch U;pr DIsmnc*
unmr
Torquo
-or
Irnsps
pmrmslng

Processes
Figure 17-1. Interconnection between sensors and various automation levels in a plant.
17.2 Sensors and Sensor Systems in Automated Production 447

The following sections discuss these previously mentioned sensors in greater detail. Sec-
tion 17.2 gives a general view of sensor principles, relevant to production technologies. In par-
ticular principles for distance measurement, contour tracking and machine vision will be ex-
plained. This is followed by examples of applications in the production process in Sec-
tion 17.3. Here the emphasis is on fettling, arc welding, and manipulation. Essential for cur-
rent applications are sensor solutions with respect to measurement technologies, process
strategies, dynamic response and interfacing. These topics are summarized in Section 17.4.
Particularly important in this context are the so called Intelligent Sensor Systems. They in-
terface with each other via computer-based components that recognize the sensor signals and
prepare, process, and interpret the electronic signals. The structure of these systems is men-
tioned in Chapters 10 and 11.
In addition to production processes, many chemical processes take place in a plant, eg, in
the paint shop or in the galvanic shop. The sensors for chemical process control and measure-
ment will be described in further detail in Chapter 18.
Measuring devices for workpiece tolerances and material quality will not be discussed. To
include such information would be beyond the scope of this chapter.

17.2 Sensors and Sensor Systems in Automated Production

17.2.1 General Classification

In various industrial branches, many different production technologies will be applied, so


there is a broad range of conceivable sensor tasks. These can be grouped into distance
measurement, contour tracking, machine vision, measuring of process parameters, and
machine diagnosis. In many applications these tasks are approached not only alone, but also
in combination. Additionally, a large number of special sensor tasks cannot be taken into ac-
count here, although they are of interest for sensor application.
Figure 17-2 illustrates the sensor principles used most often. These will be discussed in the
following Sections 17.2.2 to 17.2.5. Other sensor principles, such as infrared measurement,
have become important only in special cases. Some new principles, such as 3D laser ranging
sensors with time measurement, holographic methods, or moire fringes 12, 31, have not yet
been fully tested with regard to shop-floor conditions. Therefore, these principles will not be
described here.
The control of production processes and machine diagnosis requires the measurement of
various parameters, as illustrated in Figure 17-2. However, many of these sensor principles are
common to other applications, so the reader is referred to other chapters of this book for a
more detailed description. An exception is represented by tactile force-torque sensors and
oscillating arc sensors, which have been developed specially for robot guidance. A short
description of these is given in Section 17.2.5.

17.2.2 Distance Measurement

Distance measurement is one of the most frequent tasks in the production process. In order
to avoid collisions between the workpiece and the approaching tool during processing, the
448 17 Automated Manufacturing

sensor systems for

I
dlstance measurement
I I
pattern recognition
I I
machlne diagnosis
contour-tracking process parameters
- tactile - inductive - optical - force - force
scanning binary picture
- inductive - capacitive - optical - torque - torque
scanning gray level picture
- capacitive - optical - stereo - pressure - pressure
triangulation systems
scanner
- optical - optical solution - tactile arrays - current - speed
with structured
light
- acoustic - acoustic - power - rotation speec
scanning
- frequency - temperature

- frequency

- amplitude

- electrical
parameters

Figure 17-2. Classification of sensor systems for production technology.

planning engineer has to take gripping, fixturing, and workpiece tolerances into consideration.
Often one particular point on a workpiece surface becomes the starting point for various
operations to be performed on it. In this context, it is helpful to look for edges which can
be detected through a discontinuity within the distance signal. In certain production
technologies, a constant distance between tool and workpiece is necessary, eg, in laser cutting.
Contact sensors were some of the first sensors to be developed and are still the most widely
used today for distance measurement. Besides switches, key-buttons with potentiometric or in-
ductive pick-up are most often used. Additionally, non-contact inductive, capacitive, optical,
and acoustic principles have been developed. Their advantage with regard to contact sensors
is that they are free from wear. On the other hand, some of these principles yield disturbances,
which limit their application to certain tasks.
Inductive sensors evaluate the damping of a high-frequency electromagnetic field, caused
by induced eddy currents in the workpiece. Therefore the measurement of proximity and
distance is possible with a one-coil construction. Multi-coil sensors allow the measurement of
distances in two coordinates. In addition to recognizing the real distance from the workpiece
surface, it is also possible to recognize holes and edges, and to measure the angular orientation
of the workpiece.
An example of the schematic diagram of a multi-coil inductive distance sensor is shown in
Figure 17-3.A high-frequency alternating current is generated into coil 1, which causes an
alternating magnetic field in proximity of this coil. Coils 2 and 3 are symmetrically posi-
17.2 Sensors and Sensor Systems in Automated Production 449

tioned, but linked electrically to coil 1 in opposite phase so that their induced voltages cancel
each other at the output. In case a metallic body approaches toward coil 2 and 3, the distribu-
tion of the magnetic field to these coils will change. As a result, a difference signal is
generated, which is proportional to the distance and angle between the metallic body and the
coils.

Figure 17-3.
Schematic structure of a multi-coil inductive distance sensor.

Capacitive sensors measure the capacitance between the workpiece and a conductive plate
of small radius. A requirement is that the dielectric constant between workpiece and plate re-
mains constant. If the capacitive sensor with a metallic surface touches a metallic workpiece,
the capacitance is short-circuited. By using this particular signal, collision hazard signals can
be generated.
As inductive and capacitive sensors are sensitive in a large field of measurement, it is possi-
ble that geometrical changes of the workpiece cause disturbances within the field. Those
geometrical changes could be fixture tools, large variations in cross section, specially
neighboring edges and inhomogeneities of material. In the special case of capacitive sensors,
steam and moisture additionally influence the dielectric constant in the free space between the
sensor and workpiece, thus changing the output signal.
Opto-electronic sensors record the diffuse reflection of the spotlit workpiece and analyze
this information in an electronic circuit. Generally, laser or infrared diodes are used as light
source. The application of laser light leads to well-focused spots, resulting in high accuracy
of the distance measurement between workpiece and sensor. The larger luminescent spot of
infrared diodes can be of advantage when measuring the distance of rough surfaces.
Two measuring principles are especially widely used : the measurement of light intensity and
the triangulation principle. In the first case, the light intensity I is dependent on the distance
-
z according to I 1/z2. By means of phototransistors or photodiodes, the light intensity
will be transformed proportionally into an electrical current. An advantage of this principle
is that the optical axis of the emitted light beam is identical with the axis of the reflected beam.
On the other hand, the triangulation principle needs, as the name implies, an angle between
the emitted and reflected beams.
This priciple is explained by Figure 17-4. The distance z is related to point P, the point at
which the reflected beam hits the diode array. Diode arrays generally consist of 1000 or more
linear positioned single diodes and allow resolutions to the pm-range. An advantage of this
principle is the linear relationship between the distance and the diode position.
Errors in distance measurements mostly arise because of the different intensities of reflec-
tions from different places on the surface. In addition to metallic, polished, or painted sur-
faces with nearly total reflection, one has to consider sooty or dirty sufaces as the other ex-
treme. This could mean nearly total absorption. Transducers therefore need a very high
dynamic range in order to be useful in applications.
450 17 Automated Manufacturing

sensor

1 I I / ‘ I

Figure 17-4.
Distance measurement by
means of optical triangulation.

1 A

Figure 17-5. Mode of operation of an ultrasonic sensor, integrated in a robot gripper, sensing different
distances of a workpiece (taken from [4]).

Ultrasonic distance or proximity sensors emit focused ultrasonic impulses produced by


piezoelectric ceramic transducers. When the emitted impulse hits an object, the return echo
is directed back to the transducer and is converted into an electrical signal. Stray and foreign
17.2 Sensors and Sensor Systems in Automated Production 45 1

sound reflections are distinguished and suppressed from the regular echoes by use of the
special control unit.
Figure 17-5 shows the tracing of the amplitude of the echoes reflected by a step-like object.
As shown schematically, the ultrasonic sensor is integrated into the gripper of a robot. The
different distances between the sensor and workpiece produce a characteristic echo profile,
which can also be used for pattern recognition [4].
Moisture and changes in temperature causing variations in air density change the propaga-
tion velocity of the sound waves and produce measurement errors. One way of avoiding this
is by measuring a reference length. In this way ultrasonic sensors reach a resolution of 1 mm
or less. Owing to its finite lobe diameter, an ultrasonic beam produces a relatively large
measurement area, which is a nuisance in some manufacturing processes. lfrpical beam angles
of the lobe vary between 5 and 10".

17.2.3 Contour Tracking

In various production processes, tools have to operate along contours. Examples are
workpiece edges used in seam-welding processes, burr-contours in fettling processes, and pro-
files of surfaces in seaming and coating operations. Contour tracking with robot-guided tools
has significant advantages, because of the robots' flexibility of movement. Owing to
workpiece tolerances, heat expansion, or tolerances in fixtures or robot grippers, the real path
of the contour can differ from the planned one. One alternative is to increase the accuracy
of the production process. Another, and in some cases a more economical alternative, is to
recognize arbitrary deviations in order to give correction signals to the control unit. To com-
plete this task with a sensor system, one needs a suitable strategy and accordingly the right
sensor device in each special case. In principle, off-line measurement before processing of the
workpiece or on-line measurement during processing is conceivable.

- Off-line measurement :
This strategy must be chosen if the computation capacity of the robot control or of the sen-
sor computer is not large enough for real time processing. It can be caused by a high speed
of the robot along the path or a complex structure of the workpiece surface. Another con-
straint for on-line measurement can be bad visibility of the relevant contours during the pro-
duction process, caused by fixtures or unfavorable tool orientation. Basically two methods can
be applied : 1. Measurement of the position of the workpiece. 2. Measurement of the actual
contour.
In the first case the programmed reference coordinate system of the robot will be adjusted
to the real position of the workpiece. This position can be recognized, eg, by detecting relevant
edges or surfaces of the workpiece. After the correction the programmed path can be applied.
This method allows merely a compensation of location and gripping tolerances. However, it
is not possible to collect any information about local workpiece tolerances or tracking errors
that occur during the manufacturing process.
A better strategy is the off-line measurement of the individual contour itself (second case).
The sensor system can be used to find the beginning of the contour automatically and collect
data about the deviation in contour tracking. These data are transferred into the robot pro-
gram.
452 17 Automated Manufacturing

This strategy avoids static tracking errors caused by the signal processing time of the robot
control. It cannot cope with any dynamic tracking errors caused by the electro-mechanical
structure of the robot and the robot control. A general disadvantage of off-line methods is
the additional time needed for teaching the contour, in comparison with on-line methods.
- On-line measurement during processing :
There are two possible strategies, depending on the performance of the sensor system. The
first strategy is to find the starting point of the programmed nominal contour automatically
and then to measure the deviation of the tracked contour in relation to the programmed one.
A more efficient strategy is to recognize the whole course of the real contour by the sensor
system, so that only some intermediate points of the nominal path are programmed in the
planning phase. In this way, time-consuming programming will be avoided, especially if the
path is three-dimensional and curved.
For processing the data and making decisions, the robot control needs a certain processing
time. This undesirable delay can be compensated by measuring at some distance ahead of the
tool center point. Therefore a sensor configuration looking ahead is called preview sensor. For
path correction, the robot control has to take into account the robot’s coordinates and its path
speed in addition to the data provided by the sensor. Today, this strategy seems to be the best
to compensate for nearly all workpiece and gripping tolerances, excluding dynamic tracking
errors of the robot. For contour tracking, essential sensor principles are distance measurement
with laser scanners and pattern recognition with vision systems.
Figure 17-6shows the principle of a preview laser scanner [ 5 ] basically using the triangula-
tion principle described in the previous section. The sensor is a device for measuring the sur-
face profile of a workpiece in a plane perpendicular to the direction of robot motion (hatched
area). It obtains that profile by scanning the workpiece with a narrow laser beam and sensing

CCD-line
detector F

beam deflector

Figure 17-6.
Preview sensor, optical ar-
rangement of the camera,
laser scanning principle (taken
from [6]).
17.2 Sensors and Sensor Systems in Automated Production 453

the reflection from the workpiece surface with an array of detectors. The index number of the
detector element (see also Figure 17-4), which receives a maximum of reflected laser radiation,
indicates the distance to the workpiece surface. The scanning operation is done by two
oscillating mirrors that are positioned on the same shaft. In practice, the sensor measures 10
profiles per second, each consisting of 200 distance measurements. It covers a fan-shaped area
of about 60 x 60 mm2 with a resolution of 0.2-0.3 mm. The profile data are interpreted by
an associated computer.
For data processing, a number of seperate tasks are implemented in software modules, some
of them being related to specific hardware such as the processor or the memory. One module
is responsible for template matching. For this purpose, the processor receives a so-called
template from the robot controller, describing the type of the nominal contour and addi-
tionally permissible tolerances. The module reduces the original 200 coordinates, describing
the contour geometry by approximately six parameters.
In contrast to the principle of laser scanning by mirrors and template matching, the sensor,
shown in Figure 17-7, uses stripes generated by two solid-state lasers [7].These stripes provide
structured light illumination of the surface of the workpiece. The video camera detects line

Figure 17-7.
Optical preview sensor with in-
tegrated welding torch, structured
light illumination by laser generated
stripes (taken from [8]).
454 17 Automated Manufacturing

sequences in the field of vision, produced by scattering of the laser light from the metal sur-
face. Using a video camera, there is enhanced potential for searching for complex features
such as location of start and stop points for the contour. By increasing the optical magnifica-
tion of the camera system, tracking accuracy down to less than 1 mm can be reached for a
smaller total field of view. This feature is ideal, eg, for precision welding of accurately
machined parts.
Another significant difference of this strategy from the laser-scanner sensor described in
Figure 17-6 is the method of contour detection by means of pattern recognition. In the case
of Figure 17-7 a so-called vision teach is used. This consists of a semiautomatic phase, follow-
ing manual teaching of the contour trajectory on the nominal three-dimensional terrain at all
points along the path. It enables the sensor to track accurately during processing, particularly
along sharp curves, say 40 mm, or less, radii of curvature.
Today the main application for contour tracking sensor systems is in arc welding (see Sec-
tion 17.3.3).Derived from these solutions, future applications may be the recognition of burrs,
pasted seams and quality control of sealings. To solve these problems, special software has
to be developed. The recognition strategies of complex structures, such as that of the burrs
shown in Figure 17-16, are not yet entirely understood. Furthermore, the power of current pro-
cessors is not sufficient for the desired high path speed of a robot in such technologies (see
Table 17-1 in Section 17.4).

17.2.4 Machine Vision

Vision systems can be considered intelligent sensor systems for complex measuring or in-
spection tasks. Figure 17-8 shows the main components of a vision system. The obvious im-
portance of machine vision has been pointed out in a number of books describing the basic
research and development in this field [lo, 131.

models

image feature classification


processing computing inspection inspection
result

Figure 17-8 Simplified robot vision system (taken from [9]).

The task can be performed by common hardware components such as TV cameras (141 and
computers and therefore the hardware for optical machine vision will not be explained here.
The difficulty of image processing is selecting the methods and algorithms which fit the pro-
blem. Figure 17-9 gives a summary of essential processing strategies. The single processing
steps will be explained and illustrated by the example of processing an automobile wheel rim
(see the four pictures in Figure 17-10).

- Image Acquisition 1151


The aim of image acquisition is to obtain an appropriate digital image of the object under
investigation. The equipment for image acquisition must be selected very carefully. This in-
17.2 Sensors and Sensor Systems in Automated Production 455

image scanning process quantization

"
local
image enhancement

feature extraction

Figure 17-9.
Methods of image processing and pattern
recognition.

cludes the selection of mechanical components, illumination, optics, camera, analog to digital
converter, and image memory.
Attention must be given to the lighting and viewing techniques, the image representation
(square or hexagonal raster), the resolution (number of picture elements, called pixels), and
the number of gray levels. Interpretation of color images is generally at the beginning of in-
dustrial application. A special difficulty in the selection process is that the best picture for
automatic image processing and pattern recognition is not neccessarily identical with that of
human perception. The best image is the one from which the system can compute the desired
parameters as fast and as accurately as possible. Figure 17-10a shows an image of an
automobile wheel rim digitized to 256 x 256 pixels with 256 gray levels (corresponding to
8 bits/pixel).
Images can be obtained by many other devices, such as X-ray scanners or ultrasonic
transducers [16]. Special interest is given to tactile arrays [17], which transduce pressure into
electrical signals. The main physical effects used for transducing are conducting rubber, which
changes the resistance by pressure, capacitors, which change the capacitance by touching and
arrays of piezoresistive elements. Figure 17-11 shows an array of conducting rubber elements.
The resistance of the elements decreases under pressure and conversely the current through
an element increases. The current through the elements is scanned by a multiplexer. The
456 17 Automated Manyfacturing

Figure 17-20. (a) Image of an automobile wheel rim digitized to 256 x 256 pixels with 6 bitslpixek
@) Image processing of an automobile w h d rim,using sobel operator as special edge filter;
(c) Image segmentation and data reduction for an automobil wheel rim; (d) Internal object
model of the automobile wheel rim, used for measuring the similarity with the features in
6).

preprocessed signals are stored for serial data communication. The cycle time for a tactile
array image of % pixels, for example, is less than 1 ms.
- Image Processing (Image E n h a n c e m e n t / R o n ) (18, 201
Image enhancement involves the manipulation of an image in such a way that the resulting
image interpretation is made as simple as possible. Image enhancement methods can be
divided into global (eg, Fourier transformation), local (eg, convolution), linear (eg, high pass
filter), and non-linear (eg, median filter) methods. Image enhancement is usually supported
by dedicated hardwarej for instance bit-slice pmcessors, pipeline processors, or array pro-
cessors.
17.2 Sensors and Sensor Systems in Automated Production 457

conducting
tactile pressure rubber

current depends upon


pressure

Figure 174.
Operating principle of a tactile sensor array and serial
data transmission to a microprocessor board (taken from serial data
~71). communication

Figure 17-lob shows the image from Figure 17-10a after it has been processed using the
Sobel-operator, a special two-dimensional edge filterI201. Most edges show up fairly well.
- Image Segmentation and Data Reduction
Image segmentation means the partition of the image into regions which obtain pixels
belonging to the same detail. One region could, for example, include all pixels belonging to
a specific object. The two basic approaches to the problem are boundary segmentation (eg,
edge following) and area segmentation (eg, binarization by a gray-level threshold). Image
segmentation can work by means of deterministic and statistical features, or can be supported
by image models.
Another aim of segmentation is to reduce the amount of data. The relevant regions of the
image can be coded in a way well suited for further processing. The automobile wheel rim was
segmented using an edge-following algorithm (Figure 17-1Oc). The lines are chain coded.
Chain code means first noting the coordinates of the beginning and then noting the direction
of the neighboring line pixel, and so on [21].
- Pattern Recognition [22]
Pattern recognition means the measurement of similarities between stored object models
and the parts (regions) of the image. The parts (regions) acquire the names of the correspond-
ing objects. The pattern recognition system usually consists of an analyzer that extracts rele-
vant features from the segmented or coded image, a database containing the features of the
object models, and a classifier for similarity measurement and decision.
The concepts for a pattern-recognition system can be divided into three main groups with
respect to object models : vector concept (statistical pattern recognition), semantic nets (rela-
tion models), and grammars. The design of the classifier is sufficiently understood and sup-
458 17 Automated Manufacturing

ported by mathematical methods. For example, the coefficients of a statistical classifier can
be learned by showing the system classified samples of the object. Still unsolved is the problem
of systematic design of the analyzer. What features describe a class of objects, eg, cars, in-
dependent of the color, size, shape, or visual angle?
Pattern recognition is usually implemented as software. The low compatibility between dif-
ferent solutions is caused by the use of different hardware architectures and implementation
in different programming languages. Most implementations use standard processors, however.
In our example, the analyzer extracted straight lines, parts of circles, and corners from the
chain coded lines. Figure 17-10d shows the used object model of the automobile wheel rim.
The internal model consists of a tree, describing the circles and their relations. The classifier
uses a tree search algorithm in order to match the circles in the image with the model. If the
similarity is higher than the threshold, say 8O%, the wheel rim is considered to be recognized.

17.2.5 Measurement of Process Parameters and Machine Diagnosis

Undesired process states can be detected very quickly by monitoring process parameters
such as the power consumption of tool machines. In some cases it is also possible to determine
the position or edges of workpieces from changes in the monitored parameters. Power sensors
are suitable, because of the step change in the power signal, when the tool contacts the
workpiece surface.
Further, forceltorque sensors are becoming important process sensors for production ap-
plications. The sensor market offers a large variety of devices, ranging from simple sensors
with one or two forceltorque axes, to six-parameter devices to measure forces and torques in
three coordinates. In most cases the measuring sensor body consists of elastic spokes, fixed
at a hub. The elastic deformation of the spokes is measured by strain-gauges, and the strain-
gauge signals are transformed into forces and torques in the corresponding coordinate’s direc-
tion by means of a decoupling matrix in an appropriate electronic circuit. In some fields of
production, especially assembly, a defined sensor compliance seems to be an advantage. A
suitable example is shear pads [23].
A very cost-effective principle, which uses process parameters, is the tracking of seam welds
by an oscillating motion of the welding electrode wire. The welding arc itself supplies the sen-
sor signal [24].
The principle of a sensing element is shown in Figure 17-12a. If the position of the center
of rotation is just above the middle of the seam, the arc lengths on the right and left sides
are equal. In the case of wire deviation, the arc length will be extended on one side and
shortened on the other. Depending on this effect, the welding current changes proportionally.
To locate the center of the seam, a difference voltage signal is derived that corresponds to the
welding current (see the functional scheme of the components in Figure 17-12b). Proportional
to this difference signal the path of the welding torch is corrected laterally relative to the seam.
This oscillating welding torch needs materials greater than about 1.5 mm thick. It cannot
measure gaps or cross sections. The oscillating time is additional to the welding operation
time. Nevertheless, this principle is of increasing importance because of its simple design and
ruggedness. In the future, applications in the welding of thin sheets (less than 1.5 mm) are ex-
pected.
17.2 Sensors and Sensor Systems in Automated Production 459

source (welding torch)

+ - shunt

D
-
c

Is us

/41
" amplifier evaluation

process (welding groove)

Figure 17-12. (a) Scanning of a welding groove by the arc itsel., guided by an oscilliating welding torch
(taken from 1241); (b) Functional scheme of an arc oscillating sensor system (adapted from
1241).

In order to insure quality in production, processes in addition to machines and robots have
to be monitored by sensors. Advanced microprocessor systems are able to use the sensor sig-
nals to detect adverse conditions of machines at an early stage. These fault diagnoses make
it possible to avoid down-times by protective maintenance. Machine servicing can be reduced
from regularly scheduled check-ups to servicing only when conditions make it necessary.
Because of the many different machine types, a large number of different sensors are used
[25]. To explain these in detail is beyond the scope of this chapter, and therefore only some
general remarks will be made. Important sensor principles are distance sensors and force/
torque sensors. Another group used for a long time is vibration and noise sensors. A major
application of these sensors is in the inspection of bearings. The evolution of cheap sensors,
which is within sight, makes increasing use of fault detection probable. In the future, the
economic application even to small production systems will be possible.
460 17 Automated Manufacturing

17.3 Applications

17.3.1 Introduction

Profitable use of advanced sensor systems seems most likely in modern mass production
since here reliable machinery is essential. Therefore, the automobile industry and the electric
and electronic industry are playing a leading part in evaluating novel sensors-controlled pro-
duction processes, especially in combination with industrial robots. The idea is to improve the
production processes and to automate manual tasks that are usually unhealthy and tiresome.

Figure 1743. Increase in robot population in Japan, USA and Europe (taken from [26]).
17.3 Applications 461

4000

3000

2000

1000

coating pressing research, test


spot welding forging and training
track welding pressureldie casting
deburring machine tools
assembly other work piece
handling
others
I I I I
tool handling work piece handling

fields of application

Figure 17-14. Fields of application of industrial robots in the Federal Republic of Germany in 1987 (taken
from 1261).

Figure 17-13 shows the population of industrial robots in some leading industrial nations,
and Figure 17-14 gives statistical information about what these robots are doing in Germany.
These data are representative of other industrialized nations also.
Welding, spray painting, assembly, and machine-loading are the usual tasks for robots. Ac-
cording to many experts, the number of robots used for assembly will increase significantly
in the coming years. Progress in the field of fettling seems likely only if economic problems
can be solved.

17.3.2 Fettling

Health hazards arising from harsh environments and economics are the reasons for
automating the fettling process. For fettling castings that require low surface quality using
preprogrammed robots is the state of the art. If castings with large geometric tolerances need
to be deburred to yield high quality surfaces, sensor control is essential (see Figure 17-15).
Several strategies using off-line measurement (discussed in Section 17.2.3) are implemented in
pilot installations.
Measurement of complex burrs according to Figure 17-16 is still under development. Using
fast computers, the measurement of the workpiece including the burrs is possible. In
Figure 17-16, the profiles can be seen which are scanned by a laser sensor (distance measure-
462 17 Automated Manyfacturing

suitability
for
automatic
fettling

dimensional tolerance in (“A)

Figure 17-15. Dimensional tolerances of important casting technologies and suitabilityfor automatic fett-
ling (taken from [27]).

ment). Owing to these profiles the robot will be guided along the workpiece. In some cases
parameters for path and speed adaptation can be derived from power- and force sensors.
Figure 17-17 shows an experimental robot fettling cell with a robot handling a cast. Global
path adaptation is derived from off-line distance measurement at several points in the surface
of the cast. The robot’s speed is controlled by measuring the power of the grinding machine.

17.3.3 Spot Welding/Arc Welding

The application of arc and spot welding robots represents the largest part of the world
robot-market today. It has grown steadily in recent years, even if not quite up to the more op-
timistic forecasts. The automobile industry is today’s main user welding robots.
The first large application was the spot welding process, but until now all significant tasks
could be done without sensors. However, no form of quality control during the spot welding
process is possible without sensors. Therefore, quality assurance had to be done as a post-pro-
duction step, in an expensive and old fashioned way with a special chisel test to prove the
strengths of the welded junction. For this reason in the future greater process control by sen-
sors is required, in order that the automatic quality control problem can be solved in an
economical way. Devices offered commercially so far have not improved the production pro-
17.3 Applications 463

Figure 17-16. Example of a workpiece with burrs scanned by a laser sensor.

cess in the long run. For the most part they have been devices which can be used for measuring
the weld current, power, and, at stationary machines, the welding gun acceleration.
Although spot welding was the primary application, arc welding has recently gained nearly
the same importance. This approach has a high potential for sensors and seam tracking.
By far the largest number of arc welding robots are used for metal-inert gas (MIG) process
with argon or the metal-active gas (MAG) process with carbon dioxide. In addition use is also
made of tungsten-inert gas (TIG) and plasma arc processes, especially for precision work in
the aerospace and related industries. In the future, arc welding seems likely to become a larger
scale application that will benefit from continuous adaptive control by sensors.
464 17 Automated Manufacturing

Figure 17-17.
Robot fettling cell with sensor-guided grin-
ding tool, fettling the intake manifold of
an automobile motor.

According to the large number of robots used, today’s most significant application of sen-
sor-guided contour tracking is arc-welding. Figure 17-18 gives a schematic overview of the sen-
sor principles for fully mechanized arc welding. Currently, the most promising applications
of geometry-oriented optical sensors in manufacturing involve the use of preview laser sensors,
which have been described in Section 17.2.3. In addition, the welding process itself creates the
possibility of putting certain process signals into the seam tracking process of the welding
robot. As explained in Section 17.2.5, mechanical oscillating torches have a particular advan-
tage, which has already been demonstrated in practice. Procedures that exploit secondary pro-
cess data do not apear to be applicable so far. The opto-electronic analysis of the melting bath
area especially permits, in addition to geometrical tracking, an on-line evalution of the quality
of the weld. Taking that into consideration, this procedure can be expected to be an important
application in the future.
In Japan, in contrast to the developments in Europe and the USA, broad experience has
been gained with simple sensors, for example, arc and proximity sensors, as described in Sec-
tion 17.2.2. and 17.2.5. Nevertheless, sensor-controlled arc welding processes of a complex
nature cannot be solved today. Figure 17-19 shows an example of a difficult welding task.
Often it is not possible to process a gray-level picture of such a tube-flange joint with a
variable gap and different plate position.
In addition to the problem of pattern recognition, there is the second problem of sensor
accessibility, caused by flat angles between the tube and plate. Another problem is the ac-
17.3 Applications 465

electric capacitive - arc deviation radiation

optical - two-wire
mm
-
acoustic -
pneumatic -

Figure 1749.
Example of a difficult welding
task: tube-flange joint with
orbital fillet weld, variable gap
and different plate position.

curacy of the sensor measuring thin-sheet workpieces as used in the automobile industry (up
to 5 mm) and irregular surface reflections.
To program a complicated three-dimensional path is a very time-consuming procedure
which ties up robot capacity and requires the time of specialists. It is possible to rationalize
this process by automatically teaching the path with sensors. Figure 17-20 shows an example
466 17 Automated Manyfactwing

FDgare 17-20.
'LBctile inductive sensor in operation on a
part of a wheel suspension teaching the
seam center of an arbitrary 3D curve for
a welding robot (taken from 1291).

Ftslye l7-21. Inductive sensor application on an automobile underbody, to enable the welding robot to
fiid the initial gap of each seam, before the arc is switched on (taken from, 1301).
17.3 Applications 467

Figure 17-22. Laser sensor, tracking seam center line of a part of a moving gear during welding, cor-
responding to Figure 17-7 (taken from 181).

Fipre 17-23.
Distance-controlled laser-cut-
ting system by a capacitive
sensor, positioned in the laser
nozzle tip (taken from [31]).

of a tactile sensor, which can be attached to the torch instead of the gas nozzle. With its
touching tip, it records the track in terms of width and height.
In Figure 17-21, a shop floor application of an inductive sensor is illustrated, as explained
schematically in Section 17.2.2. In some cases, especially when short seams are to be welded,
468 17 Automated Manufacturing

the detection of start-point locations is sufficient. In this example, the dimensional tolerances
of an automobile underbody require the search for an initial gap.
Figure 17-22 shows the laser stripe sensor (Figure 17-13) tracking the seam center line of
part of moving gear, in contrast to the previous figure online, ie, during welding.
In laser cutting, sensor tasks similar to arc welding can be found. In many cases, distance-
controlled guidance of the laser cutting system by a capacitive sensor is used. The sensor is
represented by the tip of the laser nozzle itself (see Figure 17-23). The capacitance between
the nozzle and workpiece varies with their distance. Changes in distances cause frequency
variations in a high-frequency oscillator. These variations are transformed into analog DC
signals in a following component of the device.

17.3.4 Manipulation

Manipulation tasks are divided into a large number of special production processes.
Especially assembly processes, which are mainly carried out manually, represent a field of
rationalization with large potential. The change to automation requires sensor guidance of a
robot or assembly devices in many cases.

Machine vision applications

I I
I

Manipulation of Manipulation in
separated Bin - picking manufacturing Assembly

I
workpieces
on convevors processes
d

Workpieces Finishing In-process


completely sealing inspection
lying random deburring
stably spatial cutting
on belt organization process
control fastening
flash spot welding
I removal riveting
Workpiece
hung on
hooks
Workpiece
highly
organized
flquid
gasketting - arc welding
bolting
screwing
partially
separated naiIi ng
glueing
stapling

Workpieces fitting
-
partially
organized parts
spatially presentation
mating of
Darts

Figure 17-24. Machine vision applications, sensor controlled manipulation (taken from [32]).
173 Applications 469

Apart from the application of a large number of special sensor types, machine vision is the
dominant method. Figure 17-24 gives a classification into the main groups of application.
Today there are no production-tested solutions of bin picking if the workpieces have a com-
pletely random spatial distribution. This area is one of the focal points of research in univer-
sities and scientific institutes. On the other hand, the problem of picking partially organized
and unseparated workpieces is solved under favorable conditions. In addition to stationary
and robot-guided gray-level vision systems, multi sensor robot grippers will also be applied.
These grippers include, eg, ultrasonic or laser sensors for distance measurement and inductive
sensors, to avoid collision.
In the future, to have better control of gripping operations, gripper integrated tactile sensor
arrays will be available Figure 17-11 shows an example for this technology, Figure 17-25 shows
a test version of a three-finger gripper with tactile arrays in each finger, gripping a shaft. The
different gripping points in the fingers, measured by their pressure on the single pixels are
monitored (see the right-hand part of Figure 17-25).

Figure 17-25. Three-finger gripper with tactile sensor arrays in each finger, gripping a shaft. Monitor pic-
ture of the sensed distribution of pressure in the gripper fingers (taken from [17]).

Manipulation of workpieces, laying stably and separated on a belt, in bins, or hung on


hooks, controlled by vision systems, is in many applications state of the art.
Manual gripping operations are still advantageous to gripping guided by force-torque sen-
sors, owing to the limited computation speed of present robot control and sensor processors.
Installed in production is the assembly of windscreens into automobile bodies and the
assembly of wheel rims. By means of laser stripe or laser scanner sensors, the recognition of
the automobile-body location and tolerances is possible Also in the assembly of household
devices, automobile-engines and electric motors, sensor-guided robots are in operation.
470 17 Automated Manufacturing

17.3.5 Inspection

Today and probably in the near future, the main area of vision system applications is in in-
spection. A detailed classification is given in Figure 17-26 [32]. Many small enterprises are
busy in this field, and a number of solutions have been developed for shop floor operations.
Nevertheless, strategies for pattern recognition of complex patterns such as completeness of
motor units or automobile underbodies with many different components are just at the begin-
ning. Further, surface inspection just after machining is of increasing importance.
For qualitative and semiquantitative measurement of discontinuities of painted bodies,
several methods have been tested in the laboratory or in production processes. The costs of
these systems, including software development, are very high, so their use is suitable only in
the context of large numbers of pieces per series.

MACHINE VISION APPLICATIONS


INSPECTION

HIGHLY QUALITATIVE A N D

Y
QUANTITATIVE
MEASUREMENT SEMIQUANTITATIVE

I
I
LABEL READING INTEGRITY COSMETIC SAFETY
AND S 0RTIN G AND AND AND
REGISTRATION COMPLETNESS SURFACE FINISH VONITORING
PROPERTIES

-
STAINS AND

-
CRITICAL EXTERIOR ALL PARTS AND
-
SMEARS
A N D INTERIOR FEATURES PRESENT COLORS
DIMENSIONS O F BLEMISHES
KEY FEATURES RIGHT PARTS SURFACE DISCON-
O F WORKPIECES HANDEDNESS TlNUlTlES

I- I
I I%!
IELJRRS
SACKS
WARPING
DEFECTS
I

APPROXIMATE SIZE
A N D LOCATION OF
KEY FEATURES

Figure 17-26 Machine vision applications, inspection tasks (taken from (321).
17.4 Current Problems of Sensor Applications 471

17.4 Current Problems of Sensor Applications

- Measurement
Sensors are impaired during the application process by disturbances in their proximity.
These are generally electromagnetic interference, dirt, and thermal disturbances. The constant
change in light in plants is a serious problem for optoelectric sensors and vision systems. In
addition, sensors have to cope with surface variations to be recognized or measured. Dirt, soot
particles, oil, and other contaminants make the system believe it is recognizing, eg, edges and
holes, which do not exist in reality. The different surface reflections also influence the ac-
curacy of many optoelectronic sensors measuring distance.
Sensors which were exclusively developed for laboratory applications, but in spite of that
have been used in manufacturing, do not have the necessary robustness to tolerate mechanical
over-range during non-stop production processing. Insufficient electronic reliability of the
sensor devices is a major factor in many cases.

- Process Strategy
Another problem for sensor applications is that the task structure is too complex, which
makes it more difficult to develop overall processing strategies. Often the important parts of
the workpieces are partially inaccessible for sensor scanning. Further, the development of pro-
cess strategies has to take into account the requirement of necessary collision space, especially
if using preview sensors in edge tracking.

- Dynamic Response
In many cases, the processing power of microchips which are built into the sensor systems
is inadequate, especially for optoelectronic systems. In addition the processing power for
today’s robots is also partly insufficient. The result is that the tracking time for completing
many tasks is too long. Time periods of 30- 100 ms are normal. The calculations needed for
transforming the coordinates into the movement of the robot’s arms requires a large amount
of processing time. Neural networks and combined analog-digital signal processing may be a
step towards solving the processing power of the robot system [35]. An overview of the re-
quired path speed of robots for some important production technologies is given in Table 17-1.

Table 17-1. Some examples of typical path speeds of robots, dependent on the technological processes.

Technology Path speed m/min


Milling 2.4
Grinding 6.4
Water-jet cutting 15.0
Plasma cutting 1 .o
Laser cutting 15.0
Arc welding 1.5
Laser arc welding 10.0
Seaming 36.0
Spray prainting 90.0
472 17 Automated Manufacturing

A sensor-guided system with closed feedback control results in a dynamic response which
can be described with the Nyquist plot of sensor-guided path control. Figure 17-27 shows ac-
tual measurement results from a robot.
It is possible to ignore amplitude errors at the frequency of 1 Hz,but one can already
recognize a phase shift of 90"between the sensor signal and the reaction of the robot. At 2 Hz
and a 180" phase shift the robot would even drive in the opposite direction to the desired path
correction, which means that the error would become even larger. By recording test curves the
user is able to analyze the particular dynamic errors of the robot. Executing a 90" comer is
often part of the test path.
Figure 17-28 demonstrates the results obtained with most robots :the higher the speed, the
less precisely the comer is driven.
- Interface
In addition to the dynamic behavior of the robot-sensor system, interfacing is a major prob-
lem in shop floor applications. The sensor information can be transferred as an analog,
binary, or digital electric signal. Producers use a variety of hardware and software for the most
common digital interfaces in intelligent sensor systems, and IEC standards exist only for ana-
log and binary interfaces. The implementation of a useful interface in a compatible robot con-
trol system may require several months of special adaptation. In order to improve this situa-

Im .

R~~IIN!17-27.
Nyquist plot of a sensor guided robot (taken
f = 0,9 Hr from [33]).

v -\
17 mmls
v==mm/s

- 1 cm 17-28
Test curves of a robot, executing a 90" angle
with various path speeds (taken from (341).
17.6 References 473

tion, research has been and will continue to be done within national and international com-
mittees.
The latest development in the Federal Republic of Germany is the publication of prestan-
dard DIN V66 311 [36]. Further activities are outlined in Chapter 13.
- User Interface
A large number of intelligent sensor systems cannot be programmed by a formal language.
The program languages of different systems used by the same company often differ con-
siderably. Therefore the assistance of external specialists is needed to ensure perfect operation
of the system or to adjust programs to new requirements; often the correction of the software
has to be done by the sensor producer himself. Here we find a potential field for im-
provements, development, and growth in the future.

17.5 Sensor Market Trends

The world sensor market today is rapidly growing, see Chapter 1. Nevertheless, in this
market the application of sensors in production technologies remains behind the optimistic
prognoses of the 1970s. Though the number of installed industrial robots shows an increasing
trend (see Figure 17-13), this situation does not support the essential increase in the number
of sensor devices operating in production processes.
Apart from the technical and technological problems, the main reason for this situation is
often the poor economy of sensor-guided robots. This affects particularly the comparison be-
tween manual and automated operations.
In the future, decreases in the cost of robots and electronic devices and additionally the
development of new measuring principles, should lead to the expansion of the application of
sensor systems in production. Finally, it should be pointed out that the importance of these
sensor systems should be assessed not only on the basis of their sale, but rather on the basis
of their innovative contributions to the field of production technologies.

17.6 References

[l] Spur, G . , Manufacturing System Colloquium Berlin 1986, Abstracts, pp. 5-19.
[2] Schwarte, R., “Implementation of an Advanced Laser Ranging Sensor Concept”, in: Proc. ZAF
Conference, Stockholm, 1985.
[3] Tiziani, H. I., “Computer Aided Laser Measurement Techniques”, Technisches Messen 54, No. 6
(1987).
[4] Bernst, T., Liischberger, J., Magori, V., NE-Fachber. 93 (1986) 253-263.
[5] Verbeck, W. J. P. A., The Industrial Robot, June, (1984) 86-88.
[6] Werkfoto by Optische lndustrie De Oude Delft, (Oldelft), Delft.
[7] Davey, P. G . , Barratt, J. W., Morris, J. L., Met. Constr. 19, No. 12 (1987) 688-691.
[8] Werkfoto Meta Machines Ltd. Oxford
414 I 7 Automated Manufacturing

[9] Geisler, W., in :Proceedings of the 3rd International Coderenee on Robot Visionand Sensory Con-
trols (RoViSec),London: IFS (Conferences), 1983, pp. 107-111.
[lo] Braggins, D., Hollimgum, J., The Machine VisionSourcebook, London: IFS (Publications), 1986.
[ll] Zuech, N., Machine Vision. Capabilities for Industry, SME-Publication, 1986.
(121 Computer Graphics and Image Processing, New York : Academic Press.
(131 IEEE Trans. Pattern Anal. Machine Intell.
[14] Geisler, W., Weitz, G . , “Untersuchung des Einflusses von Bildinhaltsveranderungen und Helligkeits-
schwankungen auf die Abbildungseigenschaften von Fernsehkameras”, Opt0 Elektronik Magazin 4,
No 2, (1988).
[15] Automated VisualInspections, Batchelor, B. G., Hill, D. A., Hodgson, D. C. (eds.); London: IFS
(Publications), 1985.
[I61 Ballard, D. M., Brown, CH. M., Computer Vision,Englewood Cliffs, NJ: Prentice Hall, 1980.
[17] Schmid, D., Hardter, H., Michalak, E., Robotersysteme 4.3, Berlin : Springer, 1988, pp. 157- 160.
[18] Castleman, K. R., Digital Image Processing, Englewood Cliffs, NJ: Prentice Hall, 1979.
[19] Praht, W. K., Digital Image Processing, New York: Wiley, 1978.
[20] Hall, E. L., Computer Image Processing and Recognition, New York : Academic Press, 1979.
[21] Freeman, H., “Boundary Encoding” in : Picture Processing and Psychopictorics, Lipkion,
Rosenfeld (eds.); New York : Academic Press, 1970.
[22] Rosenfeld, A., A. C. Kak, Digital Picture Processing, New York : Academic Press, 1967.
[23] Hirzinger, G., Dietrich, J., Schott, J., Advanced Robotics Programme Workshopon Manipulators,
Sensors and Steps towards Mobility, Karlsruhe 11-13 May 1987.
[24] Eichhorn, F., Platz, J., DVS-Berichte 65, No. 1, (1980) 125-131.
[25] Tschulena, G. R., “Sensoren fur Fehlerfriihdiagnose”, in : Sensoren 86/87: Trends, obersichten,
Anwendungsbeispiele,Shah, R. (ed.); Dusseldorf : VDI-Verlag, 1986, pp. 86-90.
(261 Fraunhofer-Institut fur Produktionstechnik und Automatisierung (IPA), Stuttgart 1987.
(271 Sturz, W., in : IPA-IAO- Reihe Forschung und Praxis Vol. 92, Berlin : Springer, 1986.
(281 Merkblatt DVS 0927, Teil I , Diisseldorf: DVS-Verlag, 1988.
[29] Werkfoto Fachhochschule Aalen, Aalen.
[30] Werkfoto BMW AG, Miinchen und Precitec GmbH & Co, Baden-Baden.
[31] Werkfoto Messer Griesheim GmbH, Frankfurt/Main, Sensor C. A. Weidmiiller GmbH & Co, Det-
mold.
[32] Rosen, C. A., in: Computer Vision and Sensor - Based Robots, New York: Plenum Press, 1979.
1331 Schmid, D., VDI-Berichte551, Diisseldorf: VDI Verlag, 1985.
[34] Schmid, D., Nowak, H., Aymanns, H., Hardter, H., Thum, E., Internal Report INPRO, 1987,
INPRO, Berlin.
[35] Proceedings of the International Workshop on Neural Networks and their Applications, Nimes,
France I988 EC 2, Nanterre, France (ISBN 2-906899-14-3).
[36] Sensor Interfaces for Industrial Robots, DIN V 66311, Berlin: Beuth Verlag.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

18 Process Control
KLAUSHARTMA".EBERHARD NICKLAUS.
WOLFGANG Bayer AG
NOERPEL.
Leverkusen/Wuppertal. FRG

Contents
18.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 476

18.2 Structure of Sensor Systems . . . . . . . . . . . . . . . . . . . . 479

18.3 Applications and Methods . . . . . . . . . . . . . . . . . . . . . 481

18.4 Requirements for Sensor Systems . . . . . . . . . . . . . . . . . . 485

18.5 Indirect Methods of Measurement . . . . . . . . . . . . . . . . . 487

18.6 Trends of Further Development . . . . . . . . . . . . . . . . . . . 490

18.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 492

18.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492


476 18 Process Control

18.1 Introduction

Chemical processes are usually performed in a closed apparatus environment, so the status
and dynamics are not as evident as they are in manufacturing processes. In order to be able
to control a chemical process, a priori knowledge of its properties and dynamics is required,
together with information on the process quantities that characterize the actual status of the
process. The less complete the a priori knowledge is, the more urgent and important is the
need for detailed information which is mainly provided by sensor systems. This need for infor-
mation is as old as the chemical industries themselves. Sensors for controlling chemical pro-
cesses therefore have a long tradition, and many methods and instruments have been
developed with the purpose of controlling and improving chemical processes.
This chapter deals with the application of sensors in chemical processes. In practice, in-
struments are usually applied to the process as they are provided by the manufacturers.
However, for special and important tasks for which instrumentation has not yet been
developed, it is necessary for chemical industries to develop new solutions of their own. The
successful application of instruments needs a comprehensive knowledge of the processes con-
cerned. The functioning of measuring instruments is mainly based on the measuring method
involved and on the sensors and components used.
In this chapter we survey the principal aspects of chemical processes and of sensor systems,
including fields of application and the methods and requirements for instrumentation, but we
shall not discuss these topics for specific sensors. Owing to the wide variety of possible
methods, instruments and applications, a discussion of specific aspects would fill several
volumes [eg, 11.
A chemical process can be described as a sequence of products and operations [2, 31. The
latter transform the properties of the products (Figure 18-1, left). Corresponding to this flow
of material and energy is the flow of information in the process control system (Figure 18-1,
right), In this system, a priori information is handled together with the information provided
by sensor systems, so that by means of actuator systems the chemical process is kept in the
desired state. The information may be processed manually or by computer. Sensor and ac-
tuator systems serve as an interface between the process with its flow of materials and the
human or machine intelligence that controls the process.
A process operation transforms the properties of products. Similarly, a sensor system con-
verts the properties of a mixture of material into information. From this point of view, sensor
systems can be defined as [4]: the subsystems of process control systems which provide infor-
mation on process or product properties which can be processed in the information processing
subsystems. In addition, sensor systems are used to provide information on plant safety, work
place hygiene, or environmental protection.
This “black box”-like definition is based on the function of sensor systems and not on their
internal operation or structure. In this understanding, sensor systems may consist of com-
paratively simple temperature transmitters or complex measurement systems consisting of a
sample handling system and several process analyzers. Even plant laboratories in which
analyses are performed on samples that are taken manually from the process may be regarded
as sensor systems (note that this is a more general definition than that adopted for other ap-
plication areas; see also Chapter 1). The term “sensor system” expresses its basic function and
its position within the process control system.
18.1 Introduction 477

Figure l8-1. Structure of a chemical process with a p m s s control system.

As pointed out above, the main purpose of sensor systems in the chemical industries is to
provide information on process and product properties. Process properties [4] are either
- thermodynamic variables, eg, temperature, pressure, or concentration, which characterize
the process, or
- process parameters, eg, catalyst activity, which characterizethe conditions under which the
process is run, or
- control quantities, which characterize the settings of the process as, eg, the position of a
valve or the status of a pump.
Theoretically the process can be described exactly with a complete set of these quantities.
In practice, however, only a few of the thermodynamic variables and control quantities can
be measured or set, respectively. Frequently one has to use substitutional information:
478 18 ProctamControI

- process indicators [5] are empirically correlated with one or more thermodynamic
variables; product properties may also be used as process indicators; and
- control parameters are correlated with control quantities.
Control quantities and control parameters, b e i i set by actuators, characterize the input to
the process from the outside The measurement of thermodynamic variables and process in-
dicators has to be performed by sensor systems.
Product properties are either
- physical quantities,
- chemical quantities,
- technological properties, or
- product indicators [5].
Technological properties and product indicators arc substitutional information. Techno-
logical properties as, eg, the viscoelastic properties of a polymer principally depend on the
structure and compositionof the material. If these microscopicproperties are not measurable
or if their correlation with the macroscopic properties is unknown, the latter have to be
measured with special methods and apparatuses. product indicators arc empiricallycorrelated
with physical or chemical properties. So, eg, the melting point of a material may be used as
a measure for its purity. Process properties may also be used as product indicators, eg, the
temperature in a drying stove may be an indicator for the quality of the dry material. All of

Flgure 182. Classification of sensor systems in chemical production processes.


18.2 Structure of Sensor Systems 419

the product properties represent a kind of information which possibly is to be provided by


means of sensors.
The whole variety of sensor systems used in the chemical process industry can be classified
on the basis of the purpose of the information provided by the sensor systems, as displayed
in the upper half of Figure 18-2. From another point of view, sensor systems may be classified
by the quantity measured (lower half of Figure 18-2) [6].First, there are “classical” quantities
such as temperature, pressure, flow or level which can be measured by in-line sensors. Second
are collective properties of a mixture such as viscosity, density or pH, which depend on the
composition of the source material. A third group consists of specific properties such as the
concentration of a component or molecular weight distribution.
On going from the first to the third group, the variety of measuring methods and available
instruments increases, the sensor systems become more complex, the expenditure on invest-
ment and maintenance rise, and, of great importance, the need for specific application-
oriented engineering increases rapidly.

18.2 Structure of Sensor Systems

Despite the diversity of sensor systems, a basic structure is common to all of them (Figure
18-3) [6].The main input is a property Pi of a source material, and the desired output is a
signal i which represents the information on that property. The fundamental function of the
sensor system is to measure the property Pi, ie, to transform that property to’information for
the information processing system. This information may be just the signal itself or may con-
sist of additional items including measuring units, identifications of the respective sensor
system and other additional information which may be relevant to the processing system (see
Chapter 13 for sensor interface systems).

Figure l8-3.Structure of a simple sensor system. E,, quantity of the source fluid to be measured, i. out-
put signal.
480 18 Procesr Control

With regard to the internal structure of sensor systems, three types are usually distinguished
in process engineering: in-line or in situ, on-line or extractive, and off-line
Most measunments of temperature and pressure are performed with in-line or in situ sen-
sors, which are frequently designated transmitters. In this type of sensor system, the measur-
ing device is in close contact with the source material and at least the primary signal is
generated directly. It possibly has to be conditioned for further processing. Only electrical
power is needed as input to operate such a sensor system.
On-line or extmctive methods are typical of sensor systems in chemical processes. A sample
stream is extracted from a process stream or a volume and is conditioned and transported to
the measuring device. In sensor systems of this type the measuring device, eg, a process
analyzer, is embedded in a samplehandling system which may fulfil the following functions
(Figure 184) [7]:
- sample extraction,
- sample transport,
- sample conditioning,

Figure l84. Structure of a complex sensor system with on-line analyzer and device controller. S, Binary
status signal for indication of failures.
18.3 Applications and Methods 481

- exhaust stream disposal,


- supply of utilities (electrical power, fluids as, eg, calibration gas, etc.)
- sample stream switching and signal processing,
- performance monitoring and control.
The most important function of sample handling systems is sample conditioning, by means
of which the sample stream is adapted to the sensitivity of the measuring device, eg, a
photometer or a chromatograph. The design of a complete sample-handling system depends
on the properties of the source material, on the requirements of the measuring device, and
on the conditions at the disposal points. Further, the design depends on the specifications to
be met by the complete sensor system, eg, time constants or reliability.
Whereas in-line and on-line sensor systems work automatically, for off-line methods at least
sample extraction and sample transport are performed manually. Table 18-1 shows a com-
parison of these types of sensor systems.

Table 18-1. Characterization of in-line, on-line, and off-line measuring methods.

I 1 In-line, in situ I On-line, extractive I Off-line I


Sample extraction - Not necessary - Automatic Manual
Sample transport - Not necessary - Automatic Manual
Sample conditioning - Not necessary - Automatic ManuaVAutomatic
Measurement Automatic Automatic Manual/Automatic
Evaluation Automatic Automatic Manual/Automatic
Automatic Automatic ManuaVAutomatic

Availability of results Continuous Continuous, quasi-continuous Discontinuous

18.3 Applications and Methods

The sensing principles employed in sensor systems for chemical processes are numerous.
Some are widely used and have large, exclusive fields of standard application, eg, resistance
thermometers. Others are relatively common because they can be adapted to very different
problems, eg, photometers. Some are applicable only for very special problems, eg, col-
orimeters for water hardness.
Table 18-2 gives a rough overview over the properties of materials that need to be measured,
and Table 18-3 shows the sensing principles or devices that are most frequently used in
chemical process engineering. The principles and devices are discussed in specific chapters.
The necessary measuring range may vary widely, depending on the purpose of the measur-
ing system. If, eg, carbon monoxide (CO) is to be measured as a process or product property
it is mostly measured in percentage ranges and a non-dispersive infrared photometer might
meet the requirements. For workplace hygiene, however, it has to be measured at parts per
million levels and an electrochemical method might be suitable. Only very few methods and
482 18 Process Control

Table 18-2. Properties of source materials that need to be measured in chemical processes.

‘Classical”
lroperties
which do Non-specific pro-
rot depend perties which de-
Group of com-
m the com- pend on the com-
ponents Indicators Components
losition of position of the
he source source fluid
hid

remperature Density Organic carbon Chemical or Inorganic com-


’ressure Viscosity Flammable organic biological oxygen ponents:
xvel Electrical and inorganic corn- demand in waste - oxygen
conductivity ponents water - water
’low
Total organic car- Toxicity - hydrogen
PH
bon (TOC), eg, in Corrosiveness - SO,, NO,, CO,
Redox potential
CO,, and others
Turbidity
Dew point Organic com-
Color ponents:
Particle size - special compo-
Absorbance nents, eg,
benzene and
Heating value
many others
Mean molecular - molecular weight
weight distribution
Dielectric constant
Optical activitiy

instruments are equally suitable for all possibly required ranges, so that the appropriate
method depends also on the measuring range required.
Only a few methods transform the property to be measured into a signal directly. Examples
are most temperature and pressure sensors mounted in-line or in situ. The same sensors may
be used to measure other quantities indirectly. With a pressure sensor mounted at the bottom
of a vessel, the level of a liquid can be measured when the specific gravity of the liquid and
the pressure above it are known and constant. The flow of a liquid can be determined by
means of differential pressure sensors which are connected to a pipe upstream and
downstream of a flow resistance with known flow characteristics.
For on-line or extractive sensor systems, mostly indirect methods are used. Typically a
physical or physico-chemical effect is generated and then measured [8]. The signal is amplified
and conditioned for further processing. Its correlation with the quantity to be measured is
based on the physical or chemical effect used and on what is known about the source fluid.
An example is absorption photometers. If, eg, chlorine is to be measured, its optical absor-
bance with the maximum at 332 nm can be used for generating a measurable effect, which
depends on the chlorine concentration. The conditioned gaseous sample stream is transported
18.3 Applications and Methods 483

Table 18-3. Methods and groups of methods most frequently used in chemical industries.

Methods for
“classical” quantities Non-specific methods Group-specific methods Specific methods
to be measured

Temperature: Density-measuring Electric conductivity Paramagnetism


- resistance ther- systems Flame ionization Photometry
mometry - radiometric Methods with group- - absorption
- thermocouples - others specific auxiliary reac- (ultraviolet, visible,
- radiation pyrometers Viscosimetry tions near-infrared, infrared)
Pressure: Thermal conductivity Calorimetry - remission
- elastic elements with Light scattering and Gas chromatography
Dielectric constant
capacitive, inductive, extinction Liquid chromatograph)
piezoresistive or Photoionization
Acoustic methods Mass spectrometry
piezoelectric signal
conversion Electrochemical
methods
Flow and mass flow
measurements:
- potentiometry (eg,
pH electrodes,
- magnetic-inductive
ZrO, cells)
- differential pressure
- ultrasonic methods
- amperometry (eg,
fuel cells)
- others
Titration methods
Level:
- displacement Analysis methods with
methods auxiliary reactions
- hydrostatic methods - chemoluminescence
- electric and elec- - absorption in
tronic methods aerosols or liquids
- radiometric methods Special methods for
measuring humidity

through an optical cell and, by means of an appropriate light source, an optical filter, and a
detector, and the optical absorbance of the cell is measured. The signal is correlated with the
chlorine concentration if the characteristics of light source and detector are stable, if the win-
dows are clean, if the sample stream does not scatter light, and if there are not other com-
ponents present which may absorb light at the measurement wavelength. Only under all these
conditions is the absorbance a measure of the chlorine concentration. It may be impossible
to measure chlorine with such an optical method if the source fluid contains components with
a comparable absorbance at 332 nm.
Some specific methods are dedicated to special quantities to be measured, eg, paramagnetic
oxygen analyzers. Others cover a wide field of application, eg, infrared photometers and
chromatographs, as they can be adapted to different tasks. It depends on the composition of
the source fluid whether a specific method has to be applied in order to obtain specific infor-
mation. For simple mixtures, eg, hydrogen in air, a non-specific method such as, in this exam-
ple, thermal conductivity measurement may be appropriate to obtain specific information on
the hydrogen concentration.
484 18 Pmms Control

A simple example of an on-line sensor system is shown in Figure 18-5 161. The sample is
extracted from a source fluid stream and transported to an analyzer house through a heated
sample line After conditioning with a coalescer, cooler, and filter, it is transferred to an
analyzer in which the measurement is performed. Valves and flow meters are installed for sup-
porting maintenance

plant
T control
LQQm

1
4 zero gas

-mw

ra7! source fluid stream

Figure 185. Simple example for an on-line sensor system with sample extraction, sample transport, sam-
ple conditioning, maintenance facilities, and a process analyzer.

On-line sensor systems are usually much more expensive than in-line systems, and especially
sample conditioning requires much maintenance effort. Moreover, because the sample stream
carries the information to be acquired from the sampling point to the analyzer, errors specific
to on-line systems are possible if the properties of the sample stream are changed in an uncon-
trolled way so that it loses an unkown part of the information to be acquired [7,91.
If a property can be measured with either in-line or on-line instruments, the in-line method
is generally much more advantageous. There are, however, many methods as, eg,
chromatography which can only be used on-line and which need sample handling systems.
Besides, special methods of sample conditioning may essentially extend the applicability of
a measuring method. Examples are phase exchange, vaporization, chemical conversion of
components, and dilution. The possibilities in designing a sample handling system are an im-
portant tool in solving measuring problems of chemical processes.
The examples are intended to show that it is far from a trivial task to select the optimum
sensor system for a specific chemical process. A detailed pre-knowledge of the process and
the source fluid combined with a knowledge of methods and instruments and their advantages
and limitations are the basis for a possibly successful solution to a problem. The final success
of a field installation depends on the reliability of the systems installed, its performance, the
18.4 Requirements for Sensor Systems 485

maintainability with regard to available maintenance resources, and on an adequate relation-


ship between the expenditure on investment and maintenance on the one hand and the advan-
tages for the process and the products on the other hand.
Despite the variety of methods and instruments, frequently not all the information required
can be provided by on-line or in-line process sensor systems. For technical or economic
reasons, information gathering may only be possible oSf-line by analytical laboratories with
their more sophisticated methods and instrumentation. In that case, samples have to be taken
manually from the process and the measurement is performed off-line in a plant or central
laboratory. The time delay is much greater than in process sensor systems, and therefore off-
line measuring is not usable if the value of a quantity is needed immediately, eg, for fast con-
trol loops or for safety instrumentation.
The sample has to be accompanied by information on the place and time where it was taken
and on the measurements to be performed. The results of measurements have to be transferred
to the process control system as rapidly as possible. This special flow of information is best
handled by interfacing the process control system directly with the laboratory information and
management system.

18.4 Requirements for Sensor Systems

The more general requirements which are shared with sensors for other applications include
the following features:
- measuring range,
- accuracy,
- time constants,
- reliability,
- maintainability,
- life cycle cost (investment expense plus maintenance expense).

These features partially overlap. If appropriate equipment reduces the maintenance re-
quirements, both life cycle cost and reliability are affected, and reliability normally depends
on the maintenance expense, which is part of the life cycle cost.
Measuring range, accuracy, and also reliability are frequently understood as intrinsic
features of a sensor system. However, in a chemical process environment they additionally de-
pend more or less strongly on the process and the properties of the source material, and of
course on ambient conditions such as temperature, pressure, vibrations, or relative humidity.
All these special conditions can be only partly simulated in laboratories. Hence the develop-
ment of a new sensor system is not finished after testing the equipment in the laboratory. Field
tests are necessary to prove its practical suitability and, above all, to assure sufficient reliability
and maintainability.
The special requirements for process instrumentation make the transition from laboratory
methods and laboratory instrumentation to process sensor systems much more difficult than
is generally supposed. Laboratory instruments are usually not constructed for permanent
operation, as process instruments are. An additional problem is that the requirements for
486 18 Process Control

reliability are much more stringent for process instrumentation than for laboratory in-
struments. The failure of a process sensor system may cause considerable cost because it is
an integrated part of the control system, on which production and product quality depend.
In laboratories, the result of an analysis is validated by the personnel, and a failure usually
leads only to a time delay. Thus, applying laboratory instrument technology to process in-
strumentation normally requires a reconstruction or even new developments of the critical
components.
Sensor systems as used in chemical industries typically have to provide information on a
property of any source material. This may be gaseous, liquid, or solid or it may be a mixture
of these physical states. It may flow through a process line or fill a volume. Ambient air may
also be the source material.
The composition of a certain source material may vary widely, and in fact almost every
possible combination of components may occur in practice. This should be considered when
sensors for chemical industries arc to be selected. A sensor that is sensitive to a certain compo-
nent, may prove useless in practice because it can measure that component only in air as
source fluid.
With regard to the application of sensor systems, the components of the source material
may be classified as follows [7]:
- Irmlevant components do not affect the performance of the sensor system.
- Obstructive components adversely affect the performance. The effect may be physical, eg,
by dirtying windows in optical systems, or chemical, eg, by corrosion, or by causing other
unacceptable errors, eg, bubbles in a liquid sample. Obstructive components can be solid,
liquid or gaseous.
- Interferingcomponents: if a component of the source material is to be measured, the in-
terfering components are those which give rise to unacceptable errors in the system by cross
sensitivity (interference errors).
The classification of the source material in a certain case depends not only on the composi-
tion of the source material itself but also on the measuring principle and instrumentation.
Which components of the source fluid are obstructive owing to corrosiveness depends, for ex-
ample, on the material used in the sensor system components and coming into contact with
the source material. It is an advantage of methods and instruments operating at high
temperatures that the number of obstructive components is reduced because the dew point of
the sample gas is exceeded, so dew drops will not be formed. When measuring oxygen with
a paramagnetic analyser chlorine will not be an interfering component, whereas for elec-
trochemical analysers it is because of its oxidizing property which causes interference errors.
It is a matter of optimization to find the most appropriate method and instrumentation for
a particular problem with its source material. Each measuring principle and instrument has
its own field of possible applications. The less they depend on the composition of the source
fluid, the more versatile they may pnwe in practice.
On the other hand, when a certain measurement task with a certain source fluid has to be
solved, a measuring method and components have to be chosen that are most suitable for this
special problem. This needs not only experience and a comprehensiveand detailed knowledge
of the commercially available sensor equipment, but also a full knowledge of the composition
of the source material for the normal states of the process and for extreme states, in which
the need for information may be even more urgent than in the normal state. This knowledge
18.5 Indirect Methods of Measurement 481

is not always simple to acquire. For new tasks of measurement, especially concerning specific
properties of the source fluid, the process should be reviewed in terms of sensor technology
in order to find out which requirements of the source fluid will have to be met. Even if applica-
tion tests with carefully chosen equipment have been performed, there may still be some
uncertainty about performance and reliability when the sensor system starts its operation in
the process, and improvements may be necessary. During ongoing operations it is a
maintenance task to find out weak points which reduce the reliability or maintainability.
Sometimes apparently small changes in the process adversely affect the performance of a sen-
sor system, which then has to be modified.
There is a wide range of measurement tasks that can be fulfilled with standard instrumenta-
tion without risks and with established performance. However, the need for a profound
knowledge of the chemical process often leads to new and demanding tasks. The necessary
efforts in development and application are justified by the importance and benefits of the in-
formation provided.

18.5 Indirect Methods of Measurement

There may be important process and product properties in chemical processes that are not
accessible to direct measurement. A combination of measurable variables may sometimes be
a useful substitute provided that some additional information is available.
In this context a distinction can be made between two fundamentally different cases:
- Although most or all of the parameters to be measured are measurable, the interference
errors inherent in the measuring method make each sensor signal dependent on several of
these parameters to different extents.
- The parameters cannot be measured directly, but may be derived from the primary sensor
signals by means of a suitable algorithm and appropriate signal processing on the basis
of known physical or chemical laws.
In both cases, a number of primary variables must be determined, and this can involve con-
siderable effort unless the measuring method involved offers the possibility of efficient multi-
component measurement by the parallel use of several sensors. Separation methods prin-
cipally offer such a possibility by using an appropriate number of primary sensors in the case
of spatial seperation (eg, in process mass spectrometry) or by using a corresponding number
of measurement memories controlled on a time basis in the case of temporal separation (eg,
in chromatography).
In the following, these considerations are applied to an important chemical process. In this
case, the information obtained from the individually measured variables is expanded by com-
bining the individual items of information to reach a higher decision level for control systems.
The recovery of sulfur is, like the desulfurization of natural gas or coke oven gas, carried
out using processes that in many respects are derived from the original Claus process and
adapted to suit the prevailing conditions in each specific case. If such processes are to be con-
trolled optimally, it is important to have a knowledge of the hydrogen sulfide (H2S) content,
488 18 Process Control

gas throughput and, for stoichiometric reasons, the links between H,S, sulfur dioxide (SO,),
and oxygen (02). Before going into detail, however, a few comments will be made on the
Claus process itself.
In a typical Claus plant, the H,S-containing gas is passed into a thermal reaction stage -
a combustion chamber with a temperature > 1000°C - together with air and fuel gas. There,
exothermic combustion causes part of the hydrogen sulfide to be oxidized to sulfur dioxide
according to the equation (Figure 18-6) .

3 H,S + 3/2 0, = 2 H2S + SO, + H,O.


Additionally, sulfur vapor is produced in this reaction stage according to

2 H,S + 0, = 2 S + 2 H,O.
This vapor is condensed and drained into a collector vessel.
The remaining gas mixture, which contains H,S and SO,, is converted catalytically in the
second reaction stage at a lower temperature according to the equation

2 H,S + SO, = 3 S + 2 H,O.


This process yields optimum results at a stoichiometric ratio of H,S : SO, = 2 : 1.
In order to avoid H,S or SO, in the off-gas in the second stage, the air supply to the first
reaction stage is set so that an H,S/SO, ratio of 2 is obtained in the inlet to the second reac-
tion stage. In practice, however, difficulties arise owing to fluctuations, which must be ex-
pected, and also to the accompanying substances that often occur, such as hydrocarbons, car-
bon dioxide, and ammonia gas, which are also involved in the chemical reactions. The concen-
trations of these substances also fluctuate with time.

@
I
0 I
l o be set

I \
I

I I I

Figure 18-6.

I I Scheme of the reaction stages of a


desulfurization process (Claus pro-
cess) with the corresponding chemical
reaction equations. The stoichiometric
ratio H,S/SO, = 2 required for the
f second stage is set by the air input to
the first stage.
18.5 Indirect Methods of Measuremeni 489

In order to control this process, continuous information o n the H,S and SO, concentra-
tions between the reaction stages is necessary, so that an H,S/SO, ratio of 2 between the two
reaction stages can be established as an initial approximation. The continuous analytical
determination of these two components is possible, although it involves all the difficulties that
occur in on-line gas analysis in chemical processes. Photometers operating in the infrared and
ultraviolet wavelength ranges are used for this purpose. Naturally, the unavoidable time delays
resulting from supply of the sample gas to the analyzers and the conditioning of the sample
gas, and also the time constants of the analyzers, must be taken into account. Combination
of the two measured signals to give the H,S/SO, ratio yields the parameter for controlling
the air supply to the Claus process.
The disadvantage of this method of controlling the air supply is the non-linearity of the
H,S/SO, ratio with respect to the volume of air added (Figure 18-7).

H2S - 2 SO2

I , , S " , " , I

100% stoichiometric air

Figure 18-7. Characteristic curves of a desulfurization process. The dashed curve shows the H,S/SO,
ratio as a function of the air input. The solid curve represents the H,S surplus with respect
to the stoichiometric ratio.

Investigations have shown [lo] that the profiles of other combinations of the H,S and SO,
concentrations as a function of the volume of air supplied are linear, ie, H,S-2 SO,. O n the
basis of this knowledge, it is now possible to use the process analysis systems in an improved
arrangement, in which their output signals are combined according to this function. The 0,
supply to the Claus process can be controlled in an optimized manner via the corresponding
electrical output signal of the computing element.
Another means of indirect access to process o r product properties which are not directly
measurable are model-based measuring techniques. If the a priori knowledge about a process
is so comprehensive that the process can be described by a mathematical model with sufficient
accuracy, it may be possible to derive specific and important information from easily
measurable quantities [eg, 4, 111.
490 18 Process Conlml

18.6 Rends of Further Development

In conventional instrumentation of control equipment, some of the sensor signals are used
as input for controllers, but most of them are only indicated and registered (Figure 18-8). Most
information is judged and valued by the operator in the control room before activities result.
This type of information processing is relatively tolerant to failures because they may be
recognized by a plausibility check of the indicated values. On the other hand, however, it gives
rise to mistakes because at least for complex sensor systems a failure is often suspected when
an unexpected signal is indicated, and the appropriate reaction is delayed by an unnecessary
failure check.

Fngure E M Processing of information and communication with conventional process control equip-
ment.

In modern process control techniques the sensor signals are processed by “intelligent
systems”, which perform the routine part of human control activities (Figure 18-1). The reac-
tion of the system to sensor signals is part of the control system program. Hence failure of
the sensor system leads to unwanted reactions of the process control system.
Therefore, sensor systems in chemical processes with modem control systems have to meet
a new spectrum of requirements. Above all, more reliability is necessary. For simple sensor
systems this may be achievable with redundant instrumentation. In many cases, however, it is
appropriate and sufficient to configure self-checking sensor systems which provide a binary
18.6 Trends of Further Development 491

failure signal indicating their non-availability to the control system, so that misinterpretation
of the signal can be prevented (Figure 18-4).
Automated processing of information requires an improved quality of the information pro-
vided by sensor systems. Therefore, much work still has to be done to improve the func-
tionality of existing and well known sensor systems. This concerns especially complex on-line
sensor systems with their wide range of possible causes of failure. Not only has their reliability
to be improved, but also their maintenance requirements (Figure 18-9).

Figure 18-9.
Further development of sen-
sors for chemical processes.

The higher degree of automation with process control systems also leads to new tasks for
measurement. When a well known process is equipped with modern control systems, usually
the need for information is reviewed and often new tasks for measurement arise, or known
but previously unsolved tasks gain more importance. Sometimes the concept of automation
depends on the possibilities and limitations of sensor systems.
Many efforts are being made to expand the fields of application of sensor systems. Methods
used exclusively in laboratories up to now are being transferred to field measurement devices.
Much work is needed in order to make all the components applicable to continuous operation
and to solve the problems of automatic sample conditioning. Examples for such methods are
liquid chromatography [121 and flow-injection analysis.
Microelectronics and control systems are resulting in new requirements for sensor systems.
Microelectronics are also an important tool in increasing the functionality of sensor systems,
eg, as device controllers. Sometimes microelectronics pave the way for essentially new
methods, eg, for correlation techniques in flow measurement [13].
New materials help to improve the properties of existing instruments, eg, their resistance to
corrosion. New instruments and technologies, eg, optical fibers, are being used in special ap-
plications, and new types of sensors are still being developed, eg, ChemFETs.
492 18 Process Control

As pointed out above, the practical importance of a new sensor can only be established
under the demanding conditions of a chemical process. Especially in the field of sensors for
the chemical industries the efforts of manufacturers and users have to be combined to make
succesful innovation possible.

18.7 Conclusions

In the field of information processing, current developments are rapid and revolutionary.
In the field of sensors for chemical process engineering the development is evolutionary,
despite all efforts and the wide scale of sources for innovation. It is a long path from an initial
good idea to a reliable and established sensor system which can be maintained with existing
resources. However, further progress in sensor properties and performance is a necessary con-
dition for automation, better quality of products and improvements in plant safety, en-
vironmental protection, and workplace hygiene. The advantages of new solutions of measure-
ment problems can be considerable.

18.8 References

[l] “Betriebsmesstechnik”, in: Messen, Steuern, Regeln in der Chemischen Technik, Vols. I and 11,
Hengstenberg, J., Sturm, B., Winkler, 0. (eds.); Heidelberg: Springer 1980.
[2].Perne, R., Polke, M., Regelungstechnik 30 (1982) 147-156.
[3] Polke, M., AutomatisierungstechnischePraxis 27 (1985) 214-223.
[4] Gilles, E. D., Nicklaus, E., Polke, M., Automatisierungstechnische Praxis 26 (1986) 423-432,
479-484.
[5] Ecker, R., Kramer, H., Miiller, K. H., Polke, M., Kunststoffe (1972) 5-10.
[6] Nicklaus, E., Chem.-hg.-Tech.59 (1987) 615-621; and: Nicklaus, E., Messen Priifen Automatisieren
(1987) 518-522.
[7] Expression of Performance of Sample Handling Systems for Process Analysers, IEC-Publication, in
press.
[8] Warncke, H., Tech. Messen 52 (1985) 135-144.
[9] Houser, E. A., Principles of Sample Handling and Sampling System Design for Process Analyzers,
Pittsburgh: Instrument Society of America, 1972.
[lo] Mathews, M., Ger. Pat. 3321 165, 1984.
[ll] Gilles, E. D.. Tech. Messen 46 (1979) 225-232, 271-274.
[12] Fuller, E. N., Porter, G. T., Roof, L. B., J. Chromatogr. Sci. 20 (1982) 120-125.
[13] Schneider, H.J., Chem.-ZngXech.55 (1983) 767-774.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

19 Energy Production
RAJIVSACHDEVA.
Siemens India Ltd. New Delhi. India

Contents
19.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . 495
19.2 Temperature Measurements . . . . . . . . . . . . . . . . . . . . 498
19.2.1 Thermocouples . . . . . . . . . . . . . . . . . . . . . . . . . . 498
19.2.1.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498
19.2.1.2 Mechanical Design . . . . . . . . . . . . . . . . . . . . . . . . 499
19.2.1.3 Special Connection Arrangements . . . . . . . . . . . . . . . . . 500
19.2.2 Resistance Thermometers . . . . . . . . . . . . . . . . . . . . . . 502
19.2.2.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502
19.2.2.2 Mechanical Design . . . . . . . . . . . . . . . . . . . . . . . . 502
19.2.2.3 Connection Arrangements . . . . . . . . . . . . . . . . . . . . . 502
19.3 Pressure. Flow. and Level Measurements . . . . . . . . . . . . . . . 503
19.3.1 General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
19.3.2 Instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
19.3.3 Sensing Elements . . . . . . . . . . . . . . . . . . . . . . . . . 505
19.3.4 Examples of Transmitters . . . . . . . . . . . . . . . . . . . . . . 506
19.3.5 Float-Type Transmitters for Level Measurements . . . . . . . . . . . . 507
19.3.6 Transmitter Connection . . . . . . . . . . . . . . . . . . . . . . 508
19.3.7 Future Trends . . . . . . . . . . . . . . . . . . . . . . . . . . 508
19.4 Expansion and Vibration Measurements . . . . . . . . . . . . . . . 510
19.4.1 Expansion Measurements . . . . . . . . . . . . . . . . . . . . . . 511
19.4.2 Vibration Measurements . . . . . . . . . . . . . . . . . . . . . . 511
19.5 Water and Steam Analysis . . . . . . . . . . . . . . . . . . . . . 512
19.5.1 Conductivity Measurements . . . . . . . . . . . . . . . . . . . . 513
19.5.2 pH Measurements . . . . . . . . . . . . . . . . . . . . . . . . 514
19.5.3 Silica Measurements . . . . . . . . . . . . . . . . . . . . . . . . 515
19.5.4 Sodium Measurements . . . . . . . . . . . . . . . . . . . . . . . 515
19.5.5 Dissolved Oxygen Content . . . . . . . . . . . . . . . . . . . . . 516
19.5.6 Hydrazine Measurements . . . . . . . . . . . . . . . . . . . . . . 517
19.5.7 Turbidity Measurements . . . . . . . . . . . . . . . . . . . . . . 518
19.6 Flue Gas Analysis . . . . . . . . . . . . . . . . . . . . . . . . 518
19.6.1 Sample Extraction Systems . . . . . . . . ............. 519
19.6.1.1 Emission Monitoring . . . . . . . . . . . . . . . . . . . . . . . 519
494 19 Energy Production

19.6.1.2 Ammonia Measurements . . . . . . . . . . . . . . . . . . . . . . 521


19.6.2 Direct Measurement ‘In Situ’ Analyzers . . . . . . . . . . . . . . 522
19.6.2.1 Oxygen Analyzer . . . . . . . . . . . . . . . . . . . . . . . . . 522
19.6.2.2 Carbon Monoxide Analyzer . . . . . . . . . . . . . . . . . . . . . 523
19.6.2.3 Particulate Measurements . . . . . . . . . . . . . . . . . . . . . . 524
19.7 Availability Considerations . . . . . . . . . . . . . . . . . . . . . 524
19.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
19.9 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 526
19.1 Introduction 495

19.1 Introduction

Power plants are subject to demands of maximum possible availability, operational safety,
and environmental protection, and at the same time, profitability with regard to energy pro-
duction has to be ensured. These factors require careful design and optimum plant manage-
ment. A human operator alone is not capable of acquiring the large quantity of information
needed, processing it logically in the available time, making decisions, and taking correspond-
ing actions. For this reason, instrumentation and control (I + C) systems are required. The
principal tasks of the I + C system are measurement, open- and close-loop controls, protec-
tion, process operation, and monitoring. The measurement system in turn is subdivided into
the sensors and the signal conditioning equipment. The sensors are usually mounted close to
the process or equipment and transmit suitable signals, mostly electric, to the signal condition-
ing equipment, located remotely.
The basic demands on the measurement system in a power plant are:

(a) measurement must be as accurate as possible,


(b) the measuring equipment must be reliable,
(c) measurement delays should be as short as possible,
(d) signal conditioning must be designed specifically to match the requirements of the
process and the system,
(e) for all critical measurements, redundancy should be provided,
(f) upon failure of a control system (or a part of it), it must be ensured that the
measurement results can be used for manual operation or override.

The measurement signals can be classified either in accordance with the sensors (physical
classification) or in accordance with the application (functional classification).
Restricting ourselves to physical parameters which are of interest in a power plant, the
physical classification divides the measurement into:
- temperature
- pressure
- flow
- level
- expansion
- vibration
- water and steam analysis
- gas analysis.

The functional division divides the measurements according to the use of the sensor signals
for:
- process monitoring
- closed-loop controls
- open-loop controls
- alarm annunciation
- safety and protection circuits.
496 19 Energy Production

Table 19-1. Overview of the various measurements in a typical 500 MW coal-fired power plant.

ppical
Physical quantitiy
No. number of Example of important measuring locations
(sensor)
such sensors

1 TEMPERATURE
la Thermocouple 500 Main steam at superheater inlet and outlet.
H P bypass steam.
Primary and secondary air.
Flue gas at superheater and reheater inlet.
Tbrbine and generator bearings.
lb Resistance ther- 250 Auxiliary steam main header.
mometer (RTD) Feedwater at economizer inlet.
Circulating water at condenser inlet.
H P heater drain to deaerator.
Bearings of various plant equipment (such as
F. D. fans, I. D. fans, P. A. fans, boiler feed
pumps, condensate pumps, air preheaters, and
mills).
2 PRESSURE 400 Turbine first stage.
Turbine throttle.
Boiler drum.
Furnace.
Primary air header.
Deaerator.
3 FLOW 80 H P bypass steam.
Primary air.
Secondary air.
Feedwater.
Condensate.
Heater drains.
4 LEVEL 80 Boiler drum.
Deaerator.
Heaters.
Condensate storage tanks.
5 EXPANSION 6 Turbine shaft
Turbine casing.
6 VIBRATION 40 Turbine shaft and bearings.
Bearings of F. D. fans, 1. D. fans, P. A. fans,
boiler feed pumps, condensate pumps, etc.
I WATER AND
STEAM ANALYSIS
la Conductivity 12 Condenser.
Boiler feed pump suction.
Feedwater at economizer inlet.
Boiler drum water.
Saturated steam.
Main steam.
19.1 Introduction 497

a b l e b-1. continued

'Ijrpical
Physical quantity
No. number of Examples of important measuring locations
(sensor)
such sensors

7b PH 2 Feedwater at economizer inlet.


Boiler drum water.
7c Silica 4 Condensate pump discharge.
Feedwater at economizer inlet.
Boiler drum water.
Main steam.
7d Sodium 4 Condensate polishing unit inlet.
Make-up water.
Demineralization plant outlet.
Steam at superheater inlet.
7e Dissolved Oxygen 2 Condensate pump discharge.
Boiler feed pump suction.
7f Hydrazine 1 Economizer inlet.
7g Turbidity 1 Condenser polishing unit outlets.
8 FLUE GAS
ANALYSIS
8a Oxygen 2 Gas duct between economizer and air heaters.
8b Carbon monoxide 2 Stack.
8c Carbon dioxide 2 Air heaters inlet and outlet.
8d Sulfur dioxide 1 Stack.
8e Nitrogen oxides 1 Stack.
8f Ammonia 1 Exhaust duct after DeNO,.
8g Dust concentration 2 Stack.

CONDENSER

Figure 19-1. The power plant cycle.


498 19 Energy Production

The description of the various types of sensors in this chapter is based more on the physical
classification given above. The subject is restricted to their application and selection in conven-
tional energy production. Physical principles will be treated only briefly as they are outlined
in more detail in other volumes of this series.
For easy reference for the reader, a simplified power plant cycle is depicted in Figure 19-1.
An overview of the various measurements in a typical conventional power plant of 500 MW
rating is given in Table 19-1. Some important application points are also included therein.

19.2 Temperature Measurements

In power plants, temperature is mainly measured with thermocouples and resistance ther-
mometers.

19.2.1 Thermocouples

19.2.1.1 General

A thermocouple consists of a thermocouple element (sensor) and protecting and connecting


parts [l-61. The sensor is formed by two wires of different metals or metal alloys which are
soldered or welded together at one end (measuring junction). If the measuring junction is at
a temperature different from that at the free ends (reference junction) of the sensor, a voltage
(termed the thermoelectric voltage) is produced at these free ends (Seebeck effect). The
magnitude of this thermoelectric voltage is a function of the temperature difference between
the measuring junction and the reference junction and also the combination of metals in the
sensor.
To determine the temperature at the measuring point, the reference junction of the sensor
must be maintained at a known and constant temperature. This is achieved by extending the
thermocouple by compensating cables up to a point of constant temperature.

Balancing

Figure 19-2.
Thermocouple connection.

The usual way to form a reference junction is to connect a second thermocouple element
(reference element) of opposite polarity to the measuring junction (Figure 19-2). The reference
element is inserted in a heated metal block, the temperature of which is controlled at 50, 60
or 70°C.
19.2 TemperatureMeasurements 499

Expressed mathematically, the output voltage is

(19-1)

where

t, = measurement temperature
t, = temperature of connection point
t, = temperature of reference element (heated metal block)

or

This means that the output voltage is independent of the temperature of the connection
point. By maintaining a constant temperature t, an exclusive dependence of the output
voltage on the temperature at the measuring junction is achieved.
The calibration data and permissible error limits for commonly used thermocouples are
listed in various international standards (ISA-ANSI MC 96.1, ASTME.230-77, IEC Publica-
tion 584-1, etc.). They state the generated thermoelectric voltages at different temperatures
based on a reference junction temperature of 0°C. Different thermocouples and the recom-
mended measuring temperature ranges are listed in Table 19-2. The thermocouple of type K
is characterized by high sensitivity and an almost linear dependence of the thermoelectric
voltage on the temperature, and is widely applied in power plants. For instance, all main steam
temperature measurements (typically 540°C) use type K thermocouples. Type S or R ther-
mocouples are used in the upper temperature ranges, ie, above 1000°C. Typical applications
are flue gas temperatures at superheater and reheater inlets (measuring range 0- 1200°C).

'liable 19-2. Important thermocouple types.

Sensor Materials Type Measuring range/ "C

Copper-constantan Cu-CuN i T -200 to 400


Iron-constantan Fe-CuNi J -200 to 700
Nickel chromium-nickel NiCr-Ni
K -200 to 1000
(chromel-alumel) (NiCr-NiAl)
Platinum rhodium-platinum PtRh (10%)-Pt S 0 to 1400
Platinum rhodium-platinum PtRh (13l70)-Pt R 0 to 1400
Nickel chromium-constantan
NiCr-CuNi E -100 to 1200
(chromel-constantan)

19.2.1.2 Mechanical Design

The sensor is mounted in a measuring insert, in which it is usually insulated by surrounding


it with pulverized mineral compounds (eg, magnesium oxide). It is protected against external
mechanical damage and chemical attack by a ceramic or metal protection tube [ 5 , 61, which
500 19 Energy Production

is mounted using flanges, screwed connections, or by welding into the pipeline or tank. The
sensor terminates in the connection head. A typical assembly drawing is shown in Figure 19-3.
One or two sensors can be provided in one thermocouple assembly. The latter is commonly
referred to as a duplex assembly. Duplex-type thermocouples are preferred in power plants.
It is common to use one sensor for control purposes and the second for pure monitoring (in-
dication, recording, or display on visual display units). This ensures that temperature measure-
ment is possible even with the failure of one sensor.

Figure L9-3.
Thermocouple, mechanical design. 1) Ceramic protective tube;
0 2) locating flange; 3) support tube; 4) retaining ring; 5) terminal;
6) connection head; 7) terminal block; 8) temperature sensor.

19.2.1.3 Special Connection Arrangements

Figure 19-4 shows three typical thermocouple connections [I]. Since a thermocouple is a DC
voltage source, these circuits are similar to those commonly employed for battery cells. For
averaging the temperatures of several thermocouples, either series (Figure 19-4a) or parallel
circuits (Figure 19-4b) may be used. Series connection suffers from the disadvantage that an
open circuit in any one of the sensors results in loss of output. This drawback does not exist
in a parallel connection because, if one or several sensors are open, the average of the remain-
ing “healthy” sensors is still available. The balancing resistors are meant to adjust the
resistances of the individual sensor lines to equal values.
Examples of applications using such average formation circuits are flue gas temperatures
at primary and secondary air heater inlets and outlets.
The temperature difference between two points is obtained with two oppositely connected
thermocouples (Figure 19-4~).For difference measurements, an additional error can arise if
the calibration data of the sensors in the acquisition range are not linear. Q p e K ther-
mocouples are therefore most suitable for difference measurements because of their almost
linear characteristics. Examples of these measurements include metal temperature differences
between specific points on plant machinery. High metal-temperature differences are alarmed
because they signify operating stress.
19.2 Temperature Measurements 501

Meas. junction Ref. junction


- - - --- - - - --- .
,

Meas. junction

Figure 19-4. (a) Series connection of thermocouples; (b) parallel connection of thermocouples; (c) dif-
ference connection of thermocouples.
502 19 Energy Production

19.2.2 Resistance Thermometers

19.2.2.1 General

Temperature measurements with these thermometers [l-61 are based on the change in the
electrical resistance of metals with temperature. The resistance thermometer is also referred
to as a resistance temperature detector (RTD). It consists of the basic sensor (ie, the measuring
resistor) and the protecting and connecting parts.
Platinum measuring resistors are mostly used. The temperature measuring range is from
- 220 to +750"C. The resistors are balanced at 0°C to 100 ohm (Pt 100). In a few cases, nickel
and copper RTDs are also used for temperatures up to 150°C.
The basic calibration data of RTDs (ie, the dependence of resistance on temperature) and
permissible error limits are specified in various international standards (IEC Publication 751,
DIN 43760, etc.). It is important to note that, for temperatures up to 500"C, the error limits
of RTDs are less than those of thermocouples. In general, the RTD has a response time ex-
ceeding 5 s (compared with 2 1 s for thermocouples). This relatively poor response is due
primarily to the slowness of thermal conductivity in bringing the device into thermal
equilibrium with its environment.
RTDs exhibit excellent time stability. The drift of a Pt 100 resistor is less than f0.1 "C/a,
the comparable figure for a thermocouple being 5"C/a. On the other hand, thermocouples
are generally more reliable and cheaper than platinum RTDs.
Some examples of applications of RTDs are the measurement of:

- auxiliary steam main header temperature


- boiler feed pump suction temperatures
- feedwater temperature at economizer inlet
- fuel oil temperature.

19.2.2.2 Mechanical Design

The constructional features of RTDs [5, 61 are similar to those of thermocouples described
in Section 19.2.1. In this instance also, the use of duplex RTDs is preferred. Suitable protective
fittings are used for installation in pipes, tanks, etc., depending on the mechanical and
chemical requirements. The protective tube material must be carefully selected to meet the re-
quirements of static pressure, flow and temperature.

19.2.2.3 Connection Arrangements

Depending on the accuracy required, the thermometers are connected in two-, three-, or
four-wire systems [4, 61. Three- and four-wire systems are usually employed in power plants.
Figure 19-5 shows the essential features of a four-wire system. This symmetrical arrangement
ensures that all line losses are completely compensated. The voltage drop, AU across the RTD
is directly proportional to its resistance, and thereby to the temperature.
19.3 Pressure, Flow, and Level Measurements 503

I
-
FIELD SIGNAL PROCESSING SECTION
I

Const. current source

Figure 19-5.
Four-wire connection arrange- Const. current sink
ment of an RTD.

Since the RTD is an electrical resistance, there is an 12R power dissipation in it which
causes a slight self-heating effect. This causes an erroneous reading and is termed the self-
heating error. Apart from the magnitude of the current and resistance, this error depends on
the design of the thermometer and on the thermal transfer between the protective tube and
the medium. To keep this self-heating error to the minimum, the current through the RTD is
kept constant and as low as possible (typically 2 mA).

19.3 Pressure, Flow, and Level Measurements

19.3.1 General

Pressure, flow, and level measurements generally use similar types of sensors and instrument
designs [2-4, 7- 181. For this reason, these measurements are treated together in this section,
any distinctions being specifically mentioned.
Pressure is an extensively measured variable in a power plant. The range of application ex-
tends from measurements in the vacuum ranges (eg, furnace pressure) to over 200 bar for the
main steam pressure.
Flow is a measure of the quantity of fluid transported per unit time. To achieve optimum
plant efficiency, flows of various media (fuel, air, feedwater, steam, etc.) have to be monitored
and controlled within defined limits. The differential-pressure method is predominantly used
for flow measurement. The well known constriction devices (orifice plate, venturi, nozzle, etc.)
are used to generate the differential pressure (Figure 19-6). Measurement of flow is then per-
formed by employing a differential-pressure measurement system between the two pressure
tapping points. Expressed mathematically, the relationship is

(1 9-2)
504 19 Energy Production

where

q = flow
K = constant, depending mainly on the dimensions of the pipe and the shape of the
constriction
p = density of medium
A p = differential pressure.

Figure 19-6
Pressure around a constriction
device. Internal diameter of the
pipe: D; opening diameter of the
constriction device: d; pressure in
the pipe: p ; pressure immediately
upstream of the constriction device:
p , ; pressure immediately
downstream of the constriction
device: p z ; differential pressure:
Ap; remaining differential pressure:
AP".

Changes in the density of the fluid can lead to errors in the flow derived from the differen-
tial pressure. The density is generally dependent on the pressure and temperature. For example,
the density @) of gases (primary and secondary air) is expressed as

p = P/RT, (19-3)

where

R = gas constant,
T = absolute temperature,
P = absolute pressure.

Therefore, for all important measurements, the pressure and temperature are additionally
measured. These variables are then used to correct the differential-pressure value in a separate
flow-correction computer.
Certain flow measurements, however, do not use the differential-pressure method. These in-
clude the main steam flow, where the permanent pressure loss of a constriction device cannot
be tolerated because it results in a significant reduction in the efficiency of the power plant
cycle. Instead, the turbine first-stage pressure is measured. This first-stage pressure is linearly
proportional to the main steam flow.
The level of boiling water in various pressure vessels (boiler drum, deaerator, etc.) of a power
plant is also measured by employing the differential-pressure method. Density variations are
19.3 Pressure, Flow, and Level Measurements 505

compensated by using a pressure signal to correct the differential-pressure signal in a separate


level computer.

19.3.2 Instruments

The instruments used for pressure, differential-pressure, flow, and level measurements are:
a. gages for local indication of the measured pressure
b. switches for remote binary transmission of limits, usually in the form of potential-free
contacts
c. transmitters for remote analog transmission of the measured variable, usually in the form
of standardized-current signals in the range 0-20 mA (so-called “dead zero”) or 4-20 mA
(“live zero”).

19.3.3 Sensing Elements

A wide variety of sensing elements are used for pressure and differential-pressure
measurements [3, 4, 131. These include rotary piston, Bourdon tube, diaphragm, capsule
spring, bellows, ring balance manometer, and floats. All these elements are based on the com-
mon principle of converting the pressure information into a physical displacement. In a
transmitter, for instance, this displacement is further converted into a linearly proportional
current signal by means of appropriate electronic circuitry.
The Bourdon tube, shown in Figure 19-7a, is the most common pressure-displacement con-
version element. Pressure variations in the tube cause it to coil or uncoil, resulting in the
desired displacement. The Bourdon tube is a robust element and, except for very low pressures,
can be used over practically the entire pressure measuring range of a power plant.
Another common pressure-sensing element is the diaphragm (Figure 19-7b). With varia-
tions in pressure, a diaphragm extends or contracts like a spring, the displacement being pro-
portional to the force. Diaphragms can be suitably designed to cover a very wide pressure-
measuring range. They are also used for differential-pressure, flow, and level measurements.
A bellows, shown in Figure 19-7c, is similar to a diaphragm and is also based on the force-
balancing principle.

Figure 19-7. Sensor elements for pressure/differential pressure measurements. a) Bourdon tube;
b) diaphragm; c) bellows.
506 19 Energy Production

19.3.4 Examples of Transmitters

Figure 19-8 depicts the basic layout of a pressure transmitter with a Bourdon tube sensing
element. A linear variable differential transformer (LVIYT) is connected to the Bourdon tube.
While the Bourdon tube converts the pressure variations to mechanical displacements, the
LVDT transduces these displacements to electrical signals.

Figure 19-8.
Pressure transmitter with Bourdon tube. I,: output
+ signal. 1) Bourdon tube: 2) differential transformer.

Another common means of transducing a mechanical displacement into an electrical signal


is the use of a variable-capacitance system. The basic principle is the dependence of the capa-
citance between two plates on the common area of the plates and the distance between them
(CaA/ d). Thus, by bringing the plates closer the capacitance is increased, or by reducing
the common area the capacitance is decreased. This variable capacitance is then used in elec-
tronic circuitry to generate a proportional current signal.
For highly accurate and stable measurements, a differential-capacitor system is often em-
ployed, based on the variable-capacitance principle described above. A three-plate configura-
tion, with two fixed and one movable plates, forms two variable capacitors. Displacement of
the movable plate increases the capacitance of one capacitor and, at the same time, decreases
the capacitance of the other capacitor.
A differential-pressure transmitter with a diaphragm sensor and capacitive pick-up is shown
in Figure 19-9. The measuring cell of the transmitter is designed as a flat, cylindrical housing
sealed on both sides by two isolating diaphragms. The process value reaches the cell via the
two pressure covers. A hollow ceramic chamber with an annular diaphragm floats inside the

Figure 19-9.
Differential pressure transmitter with
a diaphragm sensor and capacitive
pickup. I,: output signal: 1) ceramic
bed; 2) annular diaphragm; 3) cell
body; 4) measuring diaphragm;
- I I \ + 5 ) isolating diaphragm; 6) pressure
1 2 3 cover (flange).
19.3 Pressure, Flow, and Level Measurements 507

cell, ie, it moves with respect to the cell body. The measuring diaphragm and the annular
diaphragm divide the cell into two symmetrical halves, thereby forming a sealed double
chamber the interior spaces of which are filled with oil. The inside walls of the ceramic
chamber are covered with metal plates. They constitute the fixed plates of a differential
capacitor, the movable plate of which is the measuring diaphragm itself. In a separate elec-
tronic section, this differential-capacitor configuration is connected to an AC bridge circuit.
When the measuring cell is exposed to the varying process pressures, the measuring diaphragm
is displaced by the isolating diaphragms via the filled oil. The resulting change in capacitance
leads to a proportional change in the AC bridge voltage. This voltage is rectified, amplified,
and converted into a load-independent direct current. For flow measurements, the signal is
square-root extracted so that the output signal is linearly proportional to the flow [see Equa-
tion (19-2)]. During overload, the annular diaphragm is displaced and the isolating diaphragm
clings to the housing. Damaging overload effects are thereby eliminated.
This measuring principle (diaphragm sensor and differential capacitor) it also employed for
pressure and absolute-pressure measurements. Pressure is measured by exposing one port of
the differential-pressure cell to the atmosphere while absolute pressure is measured with refer-
ence to a vacuum. The measuring cell for differential pressure is also used for level measure-
ments.
Transmitters based on the above principle are characterized by extreme ruggedness, high ac-
curacy, excellent long-term stability and ability to operate under severe ambient conditions.
They have therefore found extensive acceptance in power plants [4, 7-10, 181.

19.3.5 Float-Type Transmitters for Level Measurements

For level measurements of tanks (condenser hot-well, HP/LP heaters, etc.), displacement
float-type transmitters (Figure 19-10) are often employed [ll, 121. A change in the liquid level
displaces the float position. This mechanical displacement is transmitted by a lever assembly
to a strain gage, which transduces the mechanical movement into a proportional change in

4 5

Figure 19-10. Displacement float-type transmitter. 1) Float; 2) magnet Hall sensor; 3) connection
terminals; 4) amplifier section; 5) indicator.
508 19 Energy Production

resistance. By connecting the strain gage in a temperature-compensated Wheatstone bridge


circuit, this resistive change is reflected in a proportional voltage change. A differential
amplifier across the bridge diagonal converts the voltage into a load-independent direct cur-
rent.
Several variations of the above principle are available. These include converting the displace-
ment of the float into a mechanical rotation which, in turn, varies the angular position of the
core of a rotary variable displacement transformer (RVDT). The RVDT is similar to the LVDT
except that, instead of a linear displacement of the core, an angular movement of the
transformer core takes place. This alters the number of lines of magnetic flux around the
secondary coils, thereby resulting in an increase in output of one secondary coil and a decrease
in output of the other.
In another variation of these float-type level transmitters, magnets are attached to the lever
assembly. These move past a Hall-effect position sensor that converts the magnetic field to
an electronic signal.

19.3.6 Transmitter Connection

Depending on the plant philosophy, two- or four-wire transmitters are used. In the four-wire
connection, separate leads are provided for the signal (0-20or 4-20 mA) and power supply.
The transmitter is connected to four conductors. The signal and power supply sections are
galvanically isolated from each other. In the two-wire connection both the signal (4-20mA)
and power supply (24 V DC) are fed on the same two conductors.

19.3.7 Future Trends

With the increasing use of digital technology, the latest trend is towards microprocessor-
based transmitters with digital outputs [13-171. These can be connected directly to a field data
bus which transmits the information to superimposed automation systems. The major advan-
tages of digital (so-called “smart”) transmitters over their conventional analog counterparts
are:
1. Higher accuracy. A typical value is 0.1% of span as compared with 0.25% of span for
analog transmitters.
2. Increased rangeability. It is possible to have a rangeability as high as 400 : 1, the com-
parable figure for analog transmitters being only about 6 : 1.
3. Better noise immunity. Digital signals are intrinsically less prone to distortions caused by
noise.
4. Almost no drift with time. The sensor data are stored in digital form.
5. Ambient temperature compensation. The ambient temperature is additionally sensed and
fed to the microprocessor for compensation.
6. Self-diagnostic facilities. The ability to monitor themselves is a major advantage of
digital-based systems.
7. Remote adjustability of range, damping, polarity, etc. (for example, from the control
room). This makes the commissioning of the entire system simpler.
8. Economical, because of improved overall performance.
9. The use of a fieldbus cuts down cabling cost.
19.3 Pressure, Flow, and Level Measurements 509

“Smart” transmitters are today available from a number of manufacturers (Rosemount,


Toshiba, Honeywell, etc.).
Figure 19-11 shows the functional block diagram of the Rosemount Model 3051 pressure
transmitter. The sensor is based on the proven capacitance cell technology (Section 19.3.4).
In addition, the sensor incorporates a temperature measurement to compensate for thermal
effect. During the cell characterization process at the factory, it is run through pressure and
temperature cycles. The data from these cycles are then stored in each transmitter’s
characterization PROM to ensure precise signal correction during operation. The input signal
from the sensor is converted to a digital signal for further processing by a mircroprocessor.
The sensed signal is thereby stored as digital data, permitting precise corrections and engineer-
ing unit conversion. The corrected output signal is made available to the user both as a digital
signal and as an analog 4-20 mA current signal. A hand-held, battery-powered unit (remote
transmitter interface) can perform diagnostics, configuration, and interrogation on these
transmitters. It can be connected at any termination point in the 4-20 mA signal loop.

Bur

I I I I I I 1 4-20mA Signal
To Control
Annlog To 3051 Sensor

Figure l9-ll. Smart pressure transmitter block diagram (Rosemount Model 3051).

A more detailed discussion of smart sensors and transmitters is given in Chapter 12. Owing
to their numerous advantages, it is expected that they will gain wide acceptance in the near
future. So far, however, their application has been restricted because of the absence of a
universal protocol for digital data communication between different field-mounted sensors
and the higher level automation systems from various manufacturers. To overcome this pro-
blem, manufacturers of digital transmitters are continuing to provide 4-20 mA current out-
puts.
Fieldbuses of universal protocols are currently under development (see also Chapter 13).
With these, it should be possible to connect all the field-mounted digital devices with each
other and to the automation systems (Figure 19-12).
Local consortiums are already working on the development of protocols of broad-based
field buses. These include:

- PROFIBUS (PROcess FIeld BUS) project; a West German consortium of 14 manufac-


turers and 5 research institutes
510 19 Energy Production

- EUREKA project; a european initiative


- FIP (Flux d’hformations Processus) project; being undertaken mainly by French
manufacturers
- ISA consortium SP50 in the USA.
All these projects are being executed in close contact with each another. Although in-

c
dividual projects are likely to produce results earlier, an international fieldbus protocol is not
expected before 1991. Suitable products with direct interfaces to such a bus will follow
thereafter.

Backbone bus

Cell bus
or
Process bus)

I Automation
systems 1 Automation

controller

I
I 8
=-I ransmitter

I
I I I
r----* #I
I
I I I

Field bus Field bus Field bus


(PROFIBUS) (PROFIBUS) (PROFIBUS)
PLC = Programmable
logic controller

Figure 1942. Fieldbus concept (PROFIBUS).

19.4 Expansion and Vibration Measurements

Mechanical displacements and vibrations are among the major causes of excessive wear,
thermal stress, and failure of machines. Their early detection is therefore of utmost impor-
tance for the safety and availability of the plant. Sensors are installed to monitor these
19.4 Expansion and Vibration Measurements 511

phenomena continuously and feed appropriate signals to indicators, control systems, com-
puters, and other signal analysis equipment [4, 19-22].

19.4.1 Expansion Measurements

During operation of steam turbines, relative displacements occur between the various parts
of the turbine owing to their different heat capacities and hence different temperatures. This
is particularly the case during start-up and when major load changes occur. These
displacements are measured inductively, ie, in a contactless manner [4, 191.
Figure 19-13 shows the sensor principle for the measurement of shaft expansion, relative to
the casing. If there is a relative displacement of the shaft, the air gaps d, and d, between the
disk at the shaft and the two coil cores change. This changes the coil inductances inversely.
For small displacements, the inductance change is almost proportional to the change in air
gap.

Figure 1943.
Schematic measuring setup for relative
expansion. d , , d2: air gaps; L,,L,:
measuring coils. 1) Tbrbine shaft;
2) turbine casing; 3) transmitter;
4) power pack; 5) servo recorder;
6) indicator.

19.4.2 Vibration Measurements

Turbine rotors are manufactured and balanced with a high degree of precision. However,
some residual imbalance is unavoidable. During operation this results in vibrations which are
transmitted through the bearing on to the casing. In order to analyze the load on the bearings,
these vibrations must be measured. Any change from quiet running means a deterioration of
the balanced state of the rotor. An abrupt change in quiet running may be caused by a blade
rupture, whereas a gradual deterioration may be due to bearing problems. Vibrations are
measured on shaft, bearing shells, and bearing casings. In addition to steam turbines, vibra-
tion measurements are also carried out on other plant equipment such as F. D. fans, I. D. fans,
P.A. fans, boiler feed pumps, condensate pumps, and cooling water pumps.
Figure 19-14 is a schematic diagram of an absolute vibrations sensor based on the common
principle of the spring/mass system. In this method, a coil is suspended in an air gap by means
of a spring. The permanent magnet is firmly connected to the housing. Above the natural fre-
quency of this spring/mass arrangement, the coil remains steady in space and the vibrations
move the housing about it, a relative motion thereby being generated between the permanent
magnet and the coil.
512 19 Energy Production

Figure 19-14.
Schematic diagram of an absolute vibration sensor. 1) Generator
of vibrations; 2) permanent magnet; 3) plunger coil; 4) magnetic
return path; 5) spring; 6) sensor housing.

A voltage that is proportional to the vibration rate is induced in the coil:

e = Be I . v, (19-4)

where

e = induced voltage
B = magnetic flux density
I = length of conductor in coil
v = vibration rate.

If the sensor output voltage is integrated, a measure of the amplitude of vibrations is ob-
tained.
Various other sensors have been developed for vibration measurements, including poten-
tiometric, inductive, capacitive, eddy-current, piezoelectric and optical types (4, 19-22]. The
piezoelectric sensor is based on the phenomenon that a voltage is generated across certain
crystals when stressed. A typical piezoelectric sensor consists of a crystal in contact with a
mass, which is spring loaded. When subjected to an acceleration (a), the mass ( m ) stresses
the crystal by a force (F = ma). A voltage is thereby generated across the crystal and is a
measure of the acceleration. Electronic circuitry converts this voltage signal into a propor-
tional load-independent direct current.
Fiber-optic equipment, which has recently been introduced in power plants, is particularly
suitable for difficult operating conditions. It can monitor vibrations produced by mechanical
parts in high-voltage environments, under conditions of severe electromagnetic interference or
in the presence of highly explosive gases. Successful test installations include vibration
monitoring of generator terminals, stator coil ends, and stator core teeth in turbogenerators.
It is expected that fiber-optic equipment will gain wide acceptance in the near future for these
and similar applications.

19.5 Water and Steam Analysis

High efficiency and availability of energy production demand a constant and excellent
quality of water and steam. Moreover, the presence of impurities in high-pressure boiler water
and sub- or supercritical steam leads to corrosive attacks on the main plant equipment
19.5 Water and Steam Analysis 513

(pumps, superheaters, reheaters, turbine blades, etc.). Various points in the power plant cycle
are therefore monitored to determine the presence and concentration of salts, alkalis, acids,
dissolved gases, minerals, and contamination. Samples of steam and water are extracted, con-
ditioned and then analyzed. Conditioning fundamentally entails reducing the temperature and
pressure of samples to values that can be accepted by the analytical equipment. Steam, for
example, is condensed to water at room temperature. The sampling points and their analysis
vary from plant to plant, depending on the size, type, quality, and source of fuel and cooling
water. The important types of analysis, together with typical sampling points, are explained
below.

19.5.1 Conductivity Measurements

Conductivity measurements [6, 231 are used to determine residual salt contents in boiler
feedwater and for leakage tests on condensers. The measuring equipment usually consists of
a conductivity sensor, a measuring attachment, and a temperature sensor for temperature
compensation. The output signal of the measuring attachment is a load-independent direct
current.
The measuring principle is based on the fact that the conductivity of electrolytes (aqueous
solutions of acids, alkalis, or salts) depends on the dissociation of molecules into positive and
negative ions. These ions act as current carriers when an electric voltage is applied. Since
almost all salts form ions in dilute solutions, the measurement of conductivity is a measure
of the total ion content and thereby the concentration.
The conductivity K of a solution can be measured by determining the resistance of the elec-
trolyte between two electrodes having an area A and a separation 1. The resistance R is given by

(19-5)

or

where I/A (cm-') is defined as the cell constant. The cell constant depends purely on the
geometry of the cell. The resistance is obtained by measuring the current flow between the
electrodes when a constant AC voltage is applied across them.
Measurements in power plants fall in the range of low to very low concentrations. A typical
range for the specific conductivity, for instance, is 0-2.5 pS/cm. The measurement accuracy
is usually less than f 2% of the full-scale value. Qpical conductivity measurements include
those for:

- condenser
- make-up water
- condensate pump discharge
- boiler feed pump suction
- feedwater at economizer inlet
514 I9 Energy Production

- boiler drum water


- saturated steam
- main steam.

19.5.2 pH Measurements

The pH values of feedwater at the economizer inlet and of boiler drum water are monitored
to ascertain the alkali content. This measurement is based on the determination of the cell
voltage of a galvanic cell (Figure 19-15). It consists of two electrodes (measuring and reference
electrodes) dipped into the unknown solution. The measuring electrode is provided with a
glass membrane, which responds preferentially to hydrogen ions. A buffer solution (pH 7) is
filled into this electrode. A potential difference occurs at the two boundary surfaces of the
membrane (measured solution/glass membrane/buffer solution) and is a measure of the dif-
ference between the pH values of the measured solution and the buffer solution. It cannot
be measured with wires because they can form different electrochemical potentials with the
solutions. Tapping electrodes of thalamide, calomel or silver chloride are therefore provided.
One tapping electrode (together with the buffer solution) is contained in the measuring elec-
trode. The second tapping electrode is immersed in saturated KC1 solution and connected
through a membrane to the measured solution. This forms the reference electrode. Output in-
struments are connected to this galvanic cell via an amplifier.

Figure 19-15.
Sensor arrangements for pH measurements.
1) Measuring electrode (glass electrode);
2) output instrument; 3) amplifier (measuring
attachment); 4) reference electrode;
5) membrane; 6) measured solution.

For most applications, it is usual to house the measuring and reference electrodes in the
same sensor stem. Such sensors are commonly referred to as combination electrodes. The
measuring range of pH monitors is typically 2- 12. The measurement accuracy is better than
+1% of span [6, 231.
19.5 Water and Steam Anabsis 515

19.5.3 Silica Measurements

Silica, contained in water and steam, has a tendency to form insoluble deposits in pipes and
on equipment surfaces (eg, turbine blades) [23-271. This is detrimental not only to the equip-
ment performance but also to the efficiency of the entire power plant cycle. Continuous
monitoring of silica contents is therefore essential. Qpical measurement points are the con-
densate pump discharge, feedwater at the economizer inlet, boiler drum water, and main
steam.
Silica analyzers are mainly based on the successive addition of chemicals (molybdate,
sodium citrate, amino acids) to the sample. These chemicals react with the silica to give a reac-
tion product of a specific color. The absorbance of this product is measured with a photo-
meter and is proportional to the concentration of silica.
Typical specifications of the silica analyzer are:
- measuring ranges : 0-50/0- 100/0-500 ppb
- accuracy: k 5 070 full-scale
- reproducibility : f2% full-scale
- sensitivity : 2 PPb.

19.5.4 Sodium Measurements

The presence of sodium in water and steam indicates the coexistence of potentially corrosive
anions (eg, hydroxides, sulfates, chlorides). Sodium salts, such as sodium chloride and sodium
hydroxide, become particularly aggressive when present in steam at high pressure and
temperature. They are known to cause stress corrosion cracking of boiler and superheater
tubes. The continuous monitoring of sodium, as an effective means of avoiding these pro-
blems, has therefore recently gained rapid acceptance [23, 24, 28-30].
Sodium measurements are also useful for detecting leaks in the condenser. Normally, the
concentrations of sodium in the steam and condensate should be equal. A difference can mean
either a condenser leakage or deposits of sodium salts in the steam cycle.
Typical application points are the condensate polishing unit inlets, make-up water,
demineralization output, and steam at superheater inlet.
The operating principle (Figure 19-16) involves conditioning the sample to a pH in excess
of 10. The conditioning is accomplished by mixing the sample with air saturated with am-
monia. Interference from other ions, particularly hydrogen, is thereby largely eliminated.
The conditioned sample is passed through a measuring cell provided with a sodium-specific
glass electrode, a reference electrode and a temperature sensor for automatic temperature com-
pensation. The electrode potential is a logarithmic function of the sodium concentration ac-
cording to the Nernst equation:
RT
emf = -. In ANa++ C , (1 9-6)
F
where

emf = cell output (volts)


R = gas constant
516 19 Energy Production

F = Faraday constant
T = absolute temperature
AN=+ = concentration of sodium ions
C = constant.

This cell voltage is picked up, amplified, and processed in electronic circuitry and output
as sodium concentration. Qpical specifications of the sodium analyzer are:

- measuring ranges: 0-10/1000 ppb (log)


- accuracy: f 5Yo full-scale
- reproducibility: +2% full-scale
- sensitivity : 0.1 ppb.

No electrode
I Temperature sensor
Venturi Reference
or dusopropylamine electrode

I - 1

J
?
Measuring cell
In-line f i l t e r
dusopropylamine or
calibration

A
cartridges

Sample
Pressur
controller meter
Y
Drain

in
Shutoff Needle
valve valve

Figure 19-16. Operating principle of the sodium analyzer (courtesy Polymetron).

19.5.5 Dissolved Oxygen Content

The presence of dissolved oxygen in feedwater is the main cause of corrosion of the main
plant equipment. Dissolved oxygen reacts with iron to form iron oxides that flake off metal
surfaces and produce sludge. Its amount is therefore monitored at various points, which are
typically the condensate pump discharge and boiler feed pump suction. One of the common
measuring principles is similar to that for conductivity measurement [6,23, 31, 321. It is based
on the fact that oxygen-free water does not attack thallium (a metal similar to lead), whereas
oxygen dissolved in water oxidizes thallium. The thallium oxide formed in the process com-
19.5 Water and Steam Analysis 517

bines with water to give thallium hydroxide. The latter is highly soluble in water and is a very
strong electrolyte. Thus, when water with dissolved oxygen is passed through a cartridge filled
with thallium chips, the conductivity of the water is increased. The increase in conductivity
is a measure of the content of dissolved oxygen. Therefore, the measurement basically entails
measuring the conductivity of water before and after the thallium chips cartridge. The dif-
ference in the measured conductivities is a measure of the content of dissolved oxygen.
Typical specifications of the dissolved oxygen analyzer are:

- measuring ranges : 0- 10/0- 100 ppb


- accuracy: f 5 @lo full-scale
- reproducibility: +2% full-scale
- sensitivity: 0.1 ppb.

19.5.6 Hydrazine Measurements

Steam is injected into the deaerator to extract dissolved air from the boiler feedwater. This
mechanical means of deaeration, however, does not suffice alone to keep the dissolved oxygen
levels below harmful limits. For this reason, almost all modern power plants also use chemical
means. The most common and successful method is the addition of hydrazine (N2H4) to the
feedwater. Hydrazine dosing possesses a number of advantages:
a. Its chemical reaction with the dissolved oxygen results in nitrogen and water. No solid
products, which can lead to sludge formation, are left behind.
b. It combines with iron oxides to form a protective layer of magnetite on iron surfaces. The
formation of magnetite hinders the further attack of oxygen on iron parts.
c. It increases the pH of the feedwater such that it becomes alkaline. This prevents acidic
corrosion. In general, pH values around 9 are maintained.
In view of the above, overdosing with hydrazine is usually used. The factors against overdos-
ing are the high cost and toxic nature of hydrazine. The plant management therefore decides
the optimum extent of overdosing with hydrazine and an automatic control system is then used
to regulate the dosing to maintain the hydrazine level at the desired value (23, 31, 321.
Hydrazine is normally injected at the condensate pump outlet and the residual value is
measured at the economizer inlet. One of the popular measurement methods involves an
amperometric cell consisting of an outer platinum anode and an inner silver/silver oxide
cathode, separated by a porous porcelain tube. The cell voltage generated by the electrodes
causes a current to flow, and the following reactions occur:

Anode:
N,H4 + 4 0 H - -, N, + 4H,O + 4 e -
Cathode:
4e- + 2Ag20 + 2 H 2 0 --t 4Ag + 40H-
The passage of a current creates hydroxyl ions and silver at the cathode and depletion of
hydrazine and hydroxyl ions at the anode. The current is limited by the rate at which hydrazine
diffuses to the anode, and the diffusion rate is directly proportional to the concentration of
hydrazine. Since electrode reactions are temperature dependent, the cell is also provided with
518 19 Energy Production

a thermistor for automatic temperature compensation. The output of the cell is processed by
electronic circuitry to provide user-specific interfaces such as a load-independent direct cur-
rent (0/4 to 20 mA), limit value outputs, etc.
Typcial specifications of the hydrazine analyzer are:

- measuring ranges : 0-5010- 100/0-200 ppb


- accuracy: f4% full-scale
- reproducibility: f 2 % full-scale
- sensitivity : 1 PPb.

19.5.7 Turbidity Measurements

Turbidity measurements monitor the contamination of water at condensate polishing unit


outlets. The sensors operate on the principle of scattered light [6]. A light source of constant
power shines into the water and the light scattered by the solid particles is detected by a
photocell, the output of which is a measure of the turbidity. A measuring attachment receives
the signal output of the photocell and converts it into user-specific signals.
Typical specifications of the turbidity measurement are:

- measuring ranges: 0-2/0-5 TE/F


- accuracy: f 1.5% full-scale
- reproducibility : f0.5% full-scale
- sensitivity : 0.04 TE/F
where TE/F = formazine turbidity units, defined to DIN 38404, Part 2, by the
turbidity of standard formazine suspensions.

19.6 Flue Gas Analysis

With increasing environmental awareness and stricter regulations for pollution control, flue
gas analysis equipment now occupies a very significant place in the instrumentation and con-
trol system of a conventional power plant. The following components of flue gas are usually
monitored:

- oxygen
- carbon monoxide
- carbon dioxide
- sulfur dioxide
- nitrogen oxides
- ammonia in the denitrification (DeNO,) process;
- dust concentration (particulate measurement).

The measurements of oxygen and/or carbon monoxide are also used as inputs for control
of the combustion air. They serve as a measure of excess air. By maintaining a specific excess-
19.6 Flue Gas Analysis 519

air ratio, the combustion process is optimized. Carbon dioxide is monitored in the flue gas
at the air heater inlet and outlet.
In most countries, it is regulatory to provide stack emission monitoring. This must include
0,, SO,, NO, (NO + NO,) and particulate measurements at a height where a near-uniform
mixing of the flue gas contents can be assumed. A homogeneous mixture is necessary for
reliable measurements. It is therefore not uncommon for these measurements to be undertaken
at stack elevations of around 100 m for a power plant rating of 500 MW.
Ammonia is measured in the flue gas after the DeNO, process. The details are given in
Section 19.6.1.2.
The gas analysis equipment can be classified into two broad categories: (a) sample extract-
ion type and (b) direct measurement type, also referred to as 'in-situ' type.
The sampling method is an indirect way of ascertaining the gas contents. It involves extract-
ing the sample by means of suitable probes, conditioning it (filtering, drying, etc.), and then
feeding it to analyzers. This method becomes cumbersome, the maintenance intensive and ex-
pensive if the extraction probes are located far away from the analyzer cabinets. Moreover, the
distance itself can lead to considerable delays before the sample reaches the analyzer.
On the other hand, direct measurement systems integrate the sensor in the probe itself. An
electrical signal proportional to the measured variable is obtained directly from the probe.
Since no sample system is required, the installation and maintenance costs are low. The high
speed of response makes them ideal for closed-loop control. Reliable and proven direct-
measurement systems are available today for 0,, CO and particulate measurements.

19.6.1 Sample Extraction Systems

19.6.1.1 Emission Monitoring

Figure 19-17 shows a typical gas flow diagram for flue gas measurements. The flue gas is
extracted by a sampling device designed specifically for such conditions. The most suitable
location should be agreed upon with the user and the responsible technical authorities. The
flue gas is aspirated through a cooler and other gas preparation devices by a sample pump.
It is sucked through the gas analyzer via a flowmeter with needle valve.

Figure b-17.
Typical gas flow diagram for flue gas
measurements by the extraction method.
520 19 Energy Production

Most gas analyzers operate in accordance with the non-dispersive infrared absorption
(NDIR) principle. This measuring principle is based on the molecular-specific absorption of
bands of infrared radiation [6,331.
Figure 19-18 shows the function diagram of the Siemens Ultramat 32 analyzer, which can
be used for measuring SO,, NO, CO,, and CO. An adjustable infrared (IR) source (la) is
heated to about 700°C. The radiation is directed into the gas-filled filter chamber (3), where
it is split into two beams. The right-hand (reference) beam passes through the reference
chamber (6) filled with N, and enters the right half of the receiver chamber (8) unweakened.
The left-hand (measuring) beam first passes through the sample chamber (7)through which
the sample gas flows. The sample gas components with infrared absorption capacity weaken
this beam. It then enters the left half of the receiver chamber (8).
Both sections of the receiver chamber are filled with gas (the same gas as the component
to be measured). As the measuring beam has been subjected to infrared absorption in the sam-
ple chamber, the gas in the right-hand part of the receiver chamber heats up more than that
in the left-hand part, which causes a difference in pressure and a corresponding equalizing
flow in the connecting duct. A rotating chopper (2) between the IR source and the filter

I
I
I
I
I
I
I
I
I
$p' 22Hz
\

I
-6 i
E- I
I
A- I
I
I
_--.J

---

A Sample gas inlet 11 Zero adjuster for receiver chamber 2


E Sample gas outlet 12 Receiver chamber 2 (shock compensated)
1a IR source, shiftable 13 A C voltage amplifier
1b Reflector 14 A C intermediate voltage amplifier
2 Chopper 15 Rectifier
3 Filter chamber 16 DC voltage output amplifier
4 Reflection diaphragm 17 Sensitivity adjuster
5 Window 18 Chopper motor
6 Reference chamber 19 Indicator
7 Sample chamber (200 m long) 20 Correction factor adjuster for interfering component
8 Receiver chamber 1 (shock compensated) (internal cross sensitivity compensation)
9 Compensation volume 21 Ring modulator
10 Microflow sensor

Figure 19-18. Function diagram of the gas analyzer Siemens Ultramat 32.
19.6 Flue Gas Analysis 521

chamber interrupts both beams simultaneously and periodically, which causes a pulsed flow.
A microflow sensor (10) in the connecting duct on the right- and left-hand sides of the receiver
chamber (8) converts the flow pulses into electrical signals, which are amplified, rectified, and
applied to the indicator and output in the form of a load-independent current.
To increase the selectivity of the analyzer, the main receiver chamber (8) is followed by a
second receiver chamber (12) with a second microflow sensor (10). This second chamber is sen-
sitive to interfering components in the sample gas, especially water vapor. The DC output
voltage of the second receiver chamber is used as a negative feedback for the measuring signal.
Typical specifications of this analyzer are:

- accuracy: f 2% full-scale
- sensitivity : 1% full-scale
- response time (90070 value): 7 s.

19.6.1.2 Ammonia Measurements

Ammonia is injected into the flue gas to reduce the amounts of harmful nitrogen oxides
that escape into the atmosphere. The reaction takes place in the presence of a catalyst
(titanium dioxide and vanadium oxide), the final products being harmless nitrogen and water.
4N0 + 4NH, + 0, 4N, + 6H,O
+ 8NH3 + 12H,O.
-+

6NO2 -+ 7N,

This is known as the selective catalytic reduction (SCR) process for DeNO,.
Residual ammonia in the flue gas after DeNO, is measured and serves as an indication of
the efficiency of the reduction process [34-371. The measurement is also used to control the
amount of ammonia injected into the flue gas in the SCR process.
Figure 19-19 shows the design of an ammonia measurement system. The design is based on
the ability of NH, to absorb certain frequencies in the infrared spectrum. The DeNO, flue-
gas probe extracts a sample of flue gas. A built-in SO, absorber removes the SO, and SO,
from the sample, the absorber and the path before it being heated to about 300°C to avoid
the reaction between NH, and S02/S0, to form ammonium sulfate. The sample (with SO,
and SO, removed) is led to the gas analyzer cabinet, where it is first filtered and then divided
into two equal paths. One path leads directly into the measuring chamber. In the other path,
NH, is removed from the sample gas by passing it through an NH, absorber. The gas is then
led to the reference chamber. Both chambers are irradiated with infrared light and a detector
measures the intensity of the incoming light after each chamber. Since the only difference in
the gas constituents in the two paths is the NH, content, the difference in the two detected
intensities is a measure of the latter. This difference is picked up, amplified, and converted
into user-specific signals. The cross-sensitivity to interfering components (water vapor, CO,,
etc.) is largely eliminated here because they affect both chambers equally.
Typical specifications of the NH, analyzer are:
- measuring ranges: 0-10/0-50 ppm
- accuracy: +5% full-scale
- detection limit: 5% full-scale
- response time (90Vo value): 7 s.
522 19 Energy Production

=
Ceramic
€I-=
dust filter

Gas preparation
Wastegar

Water separalw

Condensate
Calibration
Connedion

Figure 19-19. Design of an ammonia measurement system after DeNO,.

19.6.2 Direct Measurement ‘In Situ’ Analyzers

19.6.2.1 Oxygen Analyzer

The oxygen analyzer is based on a design pioneered by Westinghouse Electric Corporation


[38-401. The measuring principle involves a zirconium oxide cell, which develops a voltage
across the two sides when each side is exposed to a different oxygen concentration. The
phenomenon is described by the Nernst equation:

(19-7)
4F

where

emf = cell output (volts)


R = gas constant
F = Faraday constant
p 1 (0,) = reference air partial pressure
p , (0,) = sample gas partial pressure
C = cell constant
T = absolute temperature.

This means that if the temperature is held constant and air (20.95%oxygen) is used as the
reference gas, the output is the inverse logarithm of the partial pressure of oxygen in the
19.6 Flue Gas Analysis 523

measured gas. As the oxygen concentration decreases, the voltage increases, giving an in-
creased sensitivity at low oxygen concentrations.
The voltage is tapped by means of electrodes, which are formed by a thin platinum coating
on each side of the cell. Precious metal brazing anchors the cell in a stainless-steel tube. The
probe is provided with a diffusion element which keeps it clean. A heater and thermocouple
are provided for maintaining a constant temperature of 850°C.
Figure 19-20 illustrates the oxygen analyzer system. The probe fits directly into the flue gas
stream, thereby eliminating the necessity for any sampling system. Separate electronic circuitry
converts the cell emf to user-specific signals (current/voltage outputs, high/low alarms, etc.).

ier Output 4- 20 madc


Instrument Air

Reference Air Set

Te mpe ra t ure Control le r I Amplif ier

Figure b-20.Oxygen analyzer system.


4.J
Typical specifications of the oxygen analyzer are:

- measuring range: 0-10% 0, linear


- overall static system accuracy: + 5 % of acutal oxygen reading
- resolution sensitivity: 0.01% 0,
- system speed of response (amplifier output): 3 s.

19.6.2.2 Carbon Monoxide Analyzer

This is based on the property of carbon monoxide to absorb a specific wavelength of in-
frared radiation. It consists of two separate units, a source and a receiver, mounted on op-
posite sides of the stack or duct. The source transmits a beam of infrared light across the stack
or duct on to the receiver and part of the radiation is absorbed by the carbon monoxide. The
524 19 Energy Production

degree of absorption (ie, the difference between the transmitted and received intensities) is a
measure of the concentration and is output as user-specific signals [41-43].
vpical specifications of the CO analyzer are:

- measuring range: 0-1000 ppm


- accuracy: +3% full-scale
- reproducibility : +2% full-scale
- response time: adjustable (up to 250 s).

19.6.2.3 Particulate Measurements

This is a measurement [6, 441 to determine the dust concentration in flue gas. It uses an
optical system consisting of a measuring head and a reflector. These two units are mounted
diametrically across the stack and are aligned precisely with one another.
The measuring head consists of a light source and a receiver. The light from this source is
beamed by a lens assembly through the stack and on the other side it falls on the reflector,
which returns it back across the stack to the receiver (principle of autocollimation). The beam
of light thereby traverses the measurement path twice. The attenuation due to the dust content
of the flue gas is measured and evaluated (absorption principle). User-specific signals, propor-
tional to the dust content, are made available.
Typical specifications of the particulate measurement are:

- measuring ranges : Extinction 0 to 0.1/0.2/0.4/0.8/1.6


- accuracy: f 2% full-scale
- reproducibility: f 2 % full-scale
- response time: 8 s.

19.7 Availability Considerations

Certain measurements are classified as critical when they have a direct bearing on the plant
safety and availability. These include, for example, turbine pressure and furnace pressure.
Critical measurements are therefore usually provided with redundant sensors. Depending on
the plant capacity (110, 200, 500 MW, etc.), redundancy can be two- or three-fold. The redun-
dant sensor signals are evaluated and processed in one-out-of-two, two-out-of-two, or two-
out-of-three logics.
A simple arithmetic method for calculating the availability is as follows. The following
values are defined:

MTBF = mean time between failures


MDT = mean down-time

MTBF
Availability = V = (19-8)
MTBF + MIYT
19.8 Conclusion 525

F= 1 - v= MDT
= failure probability (19-9)
MTBF + MDT

For a system dependent on the availability of n components, the failure rate (F,) is

F, = Fl + F2 + ... + F n - , + F,, , (19-10)

where F,, F,, etc., are failure probabilities of individual components. If each component is
given a weight ( W) according to its importance, then

F, = W,F, + W2F2+ ... + W,,-lFn-l + W,,F,,. (19-11)

If the function provided in F, (for example) is redundant in another component with


failure rate F,, then

(19-12)

The system availability is given by

V, = 1 - F , . (1 9-13)

It can be seen that, as the number of components increases (see Table 19-l),the availability
calculations become more and more cumbersome and a computer is therefore required for
such calculations. The computer can be additionally used for guiding the plant management.
For instance, if the targetted system availability is 99.5%, the computer can suggest redun-
dancy measures to be undertaken to achieve this figure. A cost analysis can be simultaneously
made to check the technoeconomic feasibility of redundancy.

19.8 Conclusion

In addition to stringent accuracy stipulations, safety and availability are the primary factors
which influence the design and selection of sensors in power plants. A natural outcome of
these considerations is the emphasis of users on proven reliability. New makes, types and
designs seldom find their way directly into power plants. These must, almost invariably, prove
themselves in process industries, industrial power plants, and pilot plants before being ac-
cepted in utility power plants. Even after such acceptance, redundancy or backup may be
demanded in the first few installations to take care of unforeseeable failures.
In view of the above, power plant users are generally known for being conservative.
“Operating experience” often becomes their deciding criterion for the selection of equipment.
This is amply reflected in numerous project-purchase enquiries which specify not only the type
but even the make of a particular sensor.
526 19 Energy Production

19.9 References

[l] Schaller, A., Messen Prufen, No. 2 (1972) 81-84.


[2] Profos, P., Handbuch der Industriellen Messtechnik, Essen: Vulkan Verlag, 1987.
[3] Messen in der Prozesstechnik, Berlin: Siemens, 1972.
[4] Johnson, C. D., Process Control Instrumentation Technology, New York: Wiley, 1977.
[5] Kennedy, R. H., “Selecting Temperature Sensors”, Chem. Eng, August 8 (1983).
(61 Industrial Analytical Instruments and Temperature Measuring Instruments, Catalog MP 11.1985,
Berlin, Munich: Siemens, 1985.
[7] Product Data Sheet 2256 of Model 1151 DP Alphaline DifferentialPressure Transmitter, Rosemount.
[8] Product Literature on FC Series Electronic Transmitters, Fuji Electric.
191 Teleperm Transmitter K. Publication: E 86060T6017A301-A 1-7600, Siemens AG, 1000 Berlin, FRG.
[lo] Transmitters, Pressure Gauges and Primary Differential Devices, Siemens Catalog MP 17, 1987,
Siemens AG, 1000 Berlin, FRG.
[Ill 12 120 Electronic Level Transmitter, Instruction No. EU 500E, Masoneilan.
112) 2390-249 Series Level Transmitters and Level Sensors, Bulletin 11.2 : 2390-249, Fisher Controls.
1131 Bradshaw, A. T., Meas. Control 17, October (1984).
1141 Bradshaw, A., “ST3000 Intelligente Druckmessumformer”, Regelungstech. Praxis 25, No. 12,
(1983).
1151 The Rosemount Smart Family: Product Data Sheet 2561. Model 3051: Differential Pressure
Transmitter, Rosemount.
[16] The Rosemount Smart Family: Product Data Sheet 2560. Model 268: Remote Transmitter Interface,
Rosemount.
1171 Hoffmann, E., “Das BMFT Verbundprojekt Fieldbus”, Automatisierungstech. Praxis, 30, No. 5,
(1988).
1181 Electric Engineering Handbook, Siemens AG; New York: Wiley, 1985.
1191 Instrumentsfor Measuring Expansion, Position and Vibrationson Large Machines, Siemens Catalog
MP 15, 1986, Siemens AG, 1000 Berlin, FRG.
1201 Henze, M., “Fiber Optic Vibration Monitoring in HV-Generators”, in: Asea Fiber Optic Sensors,
CF23-1051E Ed2, 26.09.86, Asea, 1986.
1211 Product Literature on Accelerometer 2010, Asea.
1221 Product Literature on Piezo-electric Accelerometer n p e CA900-6, Vibro-Meter Corp.
1231 ChemischeMess- und Analysentechnikfur thermische Kraftwerke, Publication PD 818/8.87/3000,
Polymetron.
1241 On-Line Analyzing Systems, Publication BLA2 E0587, Bran und Lubbe.
[25] Automatic Silica Analyzer Model 12344 Bulletin 1234D-2E0, Hach.
[26] Silikostat, Model M, Publication TD 8861 M, Polymetron.
[27] On-lineAnalyzer Systems, SilikometerAC570, Computer Controlled, Catalog 2.1, Bran und Lubbe.
[28] Model 1811 Low Level Sodium Monitor, Form 1811 LL DS/6870, Orion.
[29] On-line Analyzer Systems, Ionometer AC200, Computer Controlled, Catalog 3.1, Bran und Lubbe.
[30] Sodimat, Publication TD 8853, Polymetron.
[31] Series 5000 Analyzers, Dissolved Oxygen, Hydrazine, Application/Data Bulletin 700, Cambridge
Instruments, Cambridge.
1321 Oxyflux 4, Oxygen Trace/Hydrazine Analyzer, Data Sheet 22-7.20EN, Hartmann und Braun,
6000 Frankfurt, FRG.
1331 Abshagen, J., Liebenow, D., Stahl, H., “Automatisierte Messsysteme zur Erfassung von Luft-
schadstoffen”, Autom. Messsysteme, March, (1988).
1341 Neumann, U., “Drei Trockene >pen: Das DENOX-Verfahren von Hitachi”, Energie 38, No. 4
(1986).
1351 Sajonz, D., Envir. Protect., Energy Autom. 9, No. 3 (1987).
1361 NH3-Schlupf-Messanlagefur DeN0,-Reaktoren, Bestell-Nr. A96000-S 1972, 1987, Siemens AG,
1000 Berlin, FRG.
1371 Weber, E., Huebner, K., ,,uberblick uber Denox-Verfahren“, Energie 38, No. 4 (1986).
19.9 References 521

[38] Probe Type Oxygen Analyzer Package, Model 218, Descriptive Bulletin 106-101, Westinghouse Elec-
tric Corporation.
[39] Probe Type Oxygen Analyzer with Digital Electronics Package, Model 218A, Descriptive Bulletin
106-101A, Westinghouse Electric Corporation.
[40] Basic Economical In Situ Oxygen Monitoring System Model 401, Data Sheet: 401, Dynatron.
[41] Carbon Monoxide Analyzer Package, Model 620, Descriptive Bulletin 106-620, Westinghouse Elec-
tric Corporation.
[42] Microprocessor Based In Situ CO Monitoring System, Mode13IOOM, Data Sheet: 3100M, Dynatron.
[43] IIOCO Monitor for Optimum Boiler Efficiency, Publication PA203 P, Pentomag.
[44]RM41 Dust Density Monitor, Publication Order No.: 8002 612.0586, Erwin Sick GmbH.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

20 Environmental Monitoring
RICHARDB. BROWN.EDWARD
T. ZELLERS.
University of Michigan. Ann
Arbor. MI. USA

Contents
20.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 530
20.2 General Considerations . . . . . . . . . . . . . . . . . . . . . . 530
20.3 Radiation Hazards . . . . . . . . . . . . . . . . . . . . . . . . 532
20.3.1 Ionizing Radiation . . . . . . . . . . . . . . . . . . . . . . . . 532
20.3.2 Non-ionizing Radiation ...................... 535
20.4 Biological Hazards . . . . . . . . . . . . . . . . . . . . . . . . 536
20.5 Chemical Hazards . . . . . . . . . . . . . . . . . . . . . . . . . 538
20.5.1 Water Monitoring . . . . . . . . . . . . . . . . . . . . . . . . . 538
20.5.2 Air Monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . 540
20.5.2.1 Outdoor Air Monitoring . . . . . . . . . . . . . . . . . . . . . . 540
20.5.2.2 Indoor Air Monitoring . . . . . . . . . . . . . . . . . . . . . . 542
20.5.2.3 Occupational Air Monitoring . . . . . . . . . . . . . . . . . . . . 543
20.5.3 Biological Monitoring . . . . . . . . . . . . . . . . . . . . . . . 545
20.6 Remote Sensing . . . . . . . . . . . . . . . . . . . . . . . 541
20.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
20.8 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 549
530 20 Environmental Monitoring

20.1 Introduction

Many natural events and human activities once thought to be of only local or regional con-
sequence are now the subjects of global concern; acid rain, stratospheric ozone depletion,
radioactive contamination, and global climate changes are just a few examples. The emergence
of efforts such as the Global Environment Monitoring System (GEMS) for the coordinated
review of international data on climate, health, natural resources, the oceans, and long-range
transport of pollutants, is evidence of the trend toward a world view of the environment [l].
The risks to human health and environmental quality from the byproducts of technological
development, while ostensibly controllable, are even today matters of uncertainty and con-
troversy [2, 31. Assessing these risks for the purpose of establishing rational limits on pollution
levels requires detailed knowledge of the identities, quantities, and sources of contamination;
the transport, transformation, and fate of contaminants in the environment; mechanisms of
toxic action, dose-response relationships, and possible synergistic or antagonistic interactions
in the target organisms; and the extent of human exposure. While considerable progress has
been made in our understanding of these issues for a few agents, our overall knowledge can
still be considered rudimentary. The development of new sensors for monitoring the complex
matrix of physical, chemical, and biological entities is critical to our ability to deal with evolv-
ing environmental problems.
In this chapter we discuss existing and emerging applications of sensors in environmental
monitoring. Detailed descriptions of specific sensor technologies and data processing methods
have been avoided; these topics are covered in cited references and other chapters in this
volume (Chapters 3-11) and Vols. 7 and 8. Our goal is to provide a general overview of impor-
tant environmental issues, rationales for monitoring, and the advantages and limitations of
current sensors. We hope that by providing insight into environmental sensing needs, we will
stimulate sensor development to aid in their resolution. The chapter is divided into sections
dealing with radiological, biological, and chemical hazards, with remote sensing being treated
separately as it represents a distinct approach that cuts across the other classifications.

20.2 General Considerations

Figure 20-1 shows the major pathways between a given hazard source and individual health
effects. Deriving quantitative descriptions of the various transfer processes is an extremely
complex task: the myriad toxic agents potentially released into the environment (there are cur-
rently about 10 million known chemicals) far outweigh those for which valid analytical pro-
cedures have been developed [4]; uncertainties remain regarding appropriate models for
predicting transport and dispersion through the environment [ 5 ] ; and the data base describing
the distributions of human exposures is far from complete.
Exposure to exogenous toxic agents can occur through inhalation, absorption through the
skin, or ingestion. Inhalation is the primary route of exposure for most toxic chemical agents
owing to the high surface area and gas-exchange capacity of the lungs as well as the large
volume of air breathed by an individual (about 1 m3/h for the average, moderately active
person). Dermal absorption of certain industrial chemicals, in particular, organophosphorus
20.2 General Considerations 531

SURFACE k
c SEDIMENT
DEPOSITION

BIOACCUMULATION
c IN FOOD
PRODUCTS

+
c INDIVIDUAL
EXPOSURE
*

METABOLIC

HEALTH

Figure 20-1. Generalized map showing pathways between hazardous sources and human health effects
(after 151).

pesticides, nitro compounds, and glycol-ether derivatives, can be significant if direct contact
with the liquid occurs [6]. Estimating dermal exposure is very difficult and relatively few
methods have been developed for this purpose [7,81. Ingestion of contaminants also con-
tributes to overall exposure. The presence of increasing levels of residual pesticides and in-
dustrial pollutants in food and water supplies has prompted intensified efforts to determine
the significance of this route of exposure. The accumulated burden from all exposure routes
must be accounted for in determining the overall impact on human health.
The most common approach to environmental monitoring involves three basic steps: collec-
tion of a representative environmental sample; extraction, separation, and/or other sample
pretreatment; and finally, chemical analysis to determine the identity and quantity of material
present in the sample (from which the environmental concentration can be calculated). The
sampling time may vary from nearly instantaneous (grab sampling) to several hours or days
(time-averaged sampling). Often the sample collection device is used to concentrate the
analyte(s) to permit detection of the minute quantities present in the environment. Collection
532 20 Environmental Monitoring

media such as solid adsorbents and polymeric resins are used for concentrating gases, vapors,
or dissolved species, while filters, precipitators, or impaction devices are used for dusts, fumes,
microorganisms, or other suspended particulate matter. Where reactive or volatile species are
involved, chemical trapping agents may be helpful in preventing sample loss or degradation.
Collected samples can be analyzed directly on the collection medium, as in microscopic
asbestos-fiber counting, bacterial cultures, and certain radiation measurements, or extracted
from the collection medium by thermal or solvent treatment. Separation of the components
of a sample mixture either by wet chemical methods or, more commonly, by chromatography,
is often necessary prior to analysis of the individual species. The types of analytical detectors
employed vary considerably with the nature of samples. Several of these are discussed in the
sections below.
The principal advantage of methods employing discrete sampling and analytical steps is
their applicability to a broad range of analytes, alone or in complex mixtures. The principal
limitations to this approach are the need to transport samples from the field to the laboratory,
the delay in obtaining analytical results, and the considerable labor costs of collecting and
analyzing large numbers of samples. Examples of situations where it is desirable to have an
immediate or continuous indication of environmental concentrations include monitoring an
acutely toxic material (eg, arsine or hydrogen cyanide), detecting leaks in critical process
vessels or containment facilities, monitoring the status of emission-control systems (eg, ven-
tilation systems, and air-cleaning equipment), obtaining continuous or periodic time-concen-
tration profiles, or collecting measurements from inaccessible locations. When large numbers
of routine measurements are needed, continuous monitoring can dramatically reduce the cost
per sample.
The performance requirements of an instrument in terms of sensitivity, selectivity, response
time, size, etc., will depend on variables specific to the intended application: the sampled
medium (air, water, food, or soil), the type of hazard (chemical, biological, physical, acute,
or chronic), the physical and chemical properties of the analyte(s), the sampling location
(residential, office, industrial, laboratory, field, or upper atmosphere) and the type of infor-
mation sought (stack emissions, leak detection, ambient concentrations, personal exposures,
or accumulated doses). Consideration of such application-specific variables can often reduce
sensor constraints.
Direct-reading instruments have been developed for many environmental applications. At
this point, however, use of such instruments is generally restricted to situations where the
monitored environment is well-defined, ie, where the nature of both the target analyte(s) and
potential interferences are known. Where such information is not available, qualitative data
are often all that can be obtained. Expanding the capabilities of field instrumentation will de-
pend strongly on improving sensor technology.

20.3 Radiation Hazards

20.3.1 Ionizing Radiation


Ionizing radiation (eg, alpha and beta particles, neutrons, X rays and gamma rays) arises
from both natural and man-made sources [9]. The bulk of the radiation to which the general
20.3 Radiation Hazards 533

population is exposed emanates from natural sources such as cosmic rays and radioactive
minerals. Man-made, or so-called technology enhanced, sources from mining, processing and
disposal of nuclear fuels, routine and accidental releases from reactors, diagnostic and
therapeutic procedures (eg, chest X rays), and industrial and research applications can con-
tribute measurably to exposure levels in certain segments of the population as well as to the
long-term environmental radiation load.
The biological effects of inhaled or ingested radionuclides or external radiation fields are
a function of the type, intensity, and rate of exposure, with certain tissues being more affected
than others [lo]. Maximum permissible exposure levels have been established for whole-body
and specific-organ exposures [ll, 121. Damage can result directly via the ionization of
biological molecules or indirectly via oxygen-mediated free-radical chain reactions [13, 141. Of
primary concern are the known carcinogenic and reproductive effects of radiation exposure
thought to stem from alterations, respectively, in the structure of somatic and germinal DNA.
Technology-enhanced sources considered to have the highest potential for significant en-
vironmental impact are the decay products from uranium mine tailings, gaseous and liquid
effluents from fuel reprocessing plants, and long-term storage of high-level radioactive waste
[15]. The largest volumes of nuclear waste are generated in the energy, military, and medical
sectors [17]. Certain wastes, such as tritium and krypton-85, are routinely vented into the air
and water during fuel reprocessing operations, although containment and storage options are
now being considered [16]. Most other high-level or long-lived waste is concentrated and
stored either in subsurface tanks or in deep underground sites. For tank storage the potential
release of radioactive gases and aerosols from the self-heating waste slurries, as well as the
possible rupture of the tank walls after prolonged storage, require extensive monitoring
systems at exhaust ports and around the periphery of storage sites. Certain deep geological
disposal options are limited because it is not possible to effectively monitor the status of the
stored waste [16].
Another problem of growing concern is indoor radon contamination [20, 211. Radon-222
("'Rn) is a naturally occurring radioactive gas produced from the decay of uranium-238
found in rocks and soil. Principal exposures occur in homes and offices by infiltration of
radon gas from construction materials or the surrounding soil. While 222Rnitself is a noble
gas of minimal intrinsic hazard, its decay products (radon daughters) such as Polonium-214
and -218, are short-lived charged species that emit high-energy alpha radiation. These charged
elements tend to adhere to the surfaces of small dust particles and tissues. Their deposition
in the lung is thought to be a major contributor to lung cancer in the general population [22].
It is estimated that over half of the average radiation dose-equivalent received by individuals
in the U.S. is from radon and radon daughters [17].
Occupational radiation exposure has traditionally been associated with workers in the
nuclear fuel cycle industries, radiologists, and military personnel [18]. However, advances in
the use of nuclear and radiochemical analytical methods (eg, neutron activation analyses) for
medical, pharmaceutical, geological, environmental, industrial, and materials science applica-
tions [I91 promise a continuing expansion in the population of potentially exposed workers.
Currently, an estimated one million workers routinely wear radiation dosimeter badges in the
U.S. alone [17].
The capabilities of radiation measurement equipment vary in their sensitivity to different
types of radiation, dynamic energy range, and resolution [9, 231. Some detectors provide only
gross radiation measurements while others provide detailed spectral information. No single
534 20 Environmental Monitoring

detector is suitable for all measurement applications and the proper choice requires knowledge
of the source composition and the relevant decay products and half-lives. Table 20-1 lists the
most common detectors for measuring ionizing and non-ionizing (see Section 20.3.2) radia-
tion. Film badges, etched-track detectors, and thermoluminescent detectors (TLDs) are used
for measuring gross integrated radiation exposure, generally over periods of days to weeks. For
low radiation levels the time-averaged exposure obtained from these devices is considered an
adequate indicator of risk. The most common direct-reading instruments are based on radia-
tion-induced gas ionization (eg, ionization chambers, proportional counters, and Geiger-
Muller tubes), fluorescent emissions from solid organic and inorganic phosphors (scintillation

'Pable 20-1. Types of ionizing and non-ionizing radiation and environmental monitoring instrumentation
19, 24, 30, 33, 381.

Radiation Sources/Characteristics Instrumentation


Ionizing Radiation
Alpha particles Natural radionuclides (e.g., Rn); highly Ionization chambers; proportional
( H e + + , a) ionizing, positively charged particles; counters; ZnS scintillation counters;
discrete energies; low penetration semiconductors
Beta particles Electrons / positrons from fission and Proportional counters; Geiger-Mul-
(e-. e + , P ) neutron activation products; contin- ler tubes; liquid and solid scintilla-
uous energy distribution; more pene- tion counters; photographic film;
trating than a particles semiconductors
Gamma and X rays Penetrating electromagnetic radiation Photon spectrometer; Nal scintilla-
(YW from natural minerals, cosmic rays, tion counters; proportional coun-
nuclear fuel cycle, weapons testing, ters; Geiger-Muller tubes; Ge(Li)
medical diagnostics semiconductors; thermoluminescent
detectors
Neutrons Neutral fission product from cosmic 'He, BF, filled counters; 6Li-loaded
(n) rays, nuclear reactors and high voltage thermoluminescent detectors; pn
accelerators; indirectly ionizing; fast / junction diode; etched track film
slow (thermal) neutrons differentiated detectors; activation techniques

Non-ionizing Radiation
Ultraviolet and Visible Important exposure sources include Schottky-barrier detectors; photo-
(UV / vis) sunlight, arc welding, discharge lamps, diodes; thermocouples; photovoltaic
germicidal lamps, lasers; thermal and cells
photochemical effects important
Sources include drying and baking Pyroelectric detectors; bolometers;
ovens, glassblowing, lasers; thermal ef- photoconductors; Schottky-barrier
fects most important detectors
Extremely Low Fre- Sources include transmission lines Thermocouples; diodes; bolometers
quency, Radio Fre- (ELF), communication antennae,
quency, and Microwave ovens, heaters, plasma generators
(ELF / RF/ MW) (RF/ MW); acute thermal effects;
chronic systemic effects possible
20.3 Radiation Hazards 535

counters), or changes in charge transport properties of semiconductor materials (eg, Ge,


Ge(Li), Si and Si(Li) semiconductor detectors). Handheld and pocket-sized instruments
employing these detectors are available for field surveys or real-time measurement of personal
exposures. Data-logging capabilities permit collection of individual historical exposure pro-
files. Low radionuclide concentrations coupled with high natural background radiation re-
quire preconcentration on adsorbents (gases) or polymer-membrane filters (aerosols) to
achieve acceptable counting precision and accuracy. In this case only intermittent output is
possible.
Multichannel analyzers or radiation spectrometers are used to identify specific ra-
dionuclides by their respective radiated-energy spectra. Preseparation using wet-chemical
methods may be necessary with certain mixtures to obtain component resolution, and stan-
dard chemical instrumentation (eg, laser fluorescence) can be employed to aid in the analyses.
Successful spectral analyses of radionuclides have been performed using direct-counting
methods in samples of air, water, plants, soil, and sediments [19]. Recent work on radiation
sensors has focused on new inorganic scintillator materials, photodiode scintillation detectors,
passive planar silicon detectors, and high-resolution pulse-mode cryogenic bolometric detec-
tors [25].

20.3.2 Non-ionizing Radiation

Non-ionizing radiation, which includes radiation in the ultraviolet and longer wavelength
regions, has been associated with a range of biological effects that depend on both the inten-
sity and frequency [26]. Thermal and photochemical effects are of concern in the IR-UV spec-
tral region. High-intensity IR sources can cause thermal injury to the cornea, retina, and more
rarely, the skin, particularly for wavelengths of 760- 1400 nm where tissue penetration is most
efficient. UV-radiation exposure below 320 nm is associated with skin cancer, erythema (sun-
burn), and ocular damage. Concern over stratospheric ozone depletion stems from the resul-
tant increase of radiation in this spectral range reaching the Earth’s surface. Electric-arc
welding, cutting, and foundry furnaces are important sources of occupational exposure to UV
radiation [27]. Recommended limits for occupational exposure have been defined according
to the relative physiological spectral effectiveness of the radiation [28, 291. For example, maxi-
mum erythema1 and carcinogenic effects are produced at 260 nm with a secondary peak at
295 nm, while keratitis effects peak at 280 nm [30]. A variety of photosensors have been
adapted for direct UV-exposure monitoring (Table 20-l), however instruments incorporating
these physiological spectral weightings have not been developed, to date [26, 311.
Acute effects of microwave (MW) and radiofrequency (RF) radiation arise mainly from
heating of exposed tissue, which is a function of incident power density [27, 32, 331.
Wavelengths from 1 mm to about 10 m are efficiently absorbed by the human body. Chronic
RF/MW exposure has been associated with subtle effects on nervous system, cardiovascular
system, and hematopoietic system function [26, 34, 351. However, the causal relationship bet-
ween exposure and clinical findings is tenuous due, in part, to incomplete exposure char-
acterization. Recent studies of exposure to extremely low-frequency radiation (ELF, 0-300 Hz)
suggest an association with cancer and reproductive effects, specifically in electrical workers
and people living near high-voltage sources (eg, transmission lines, transformers, etc.)
[361.
536 20 Environmental Monitoring

Higher frequency RF/MW exposures are most likely to occur in occupational settings [27].
Residential microwave ovens are another potential source, though improved shielding has vir-
tually eliminated MW leakage from properly maintained appliances. Inductive and dielectric
RF heat sources are used extensively in manufacturing industries and the use of RF plasmas
in semiconductor processing (eg, dry etching and plasma-enhanced deposition) is steadily in-
creasing. High-power RF/MW communication antennae may pose a threat to maintenance
workers near transmission sources.
Measurements of MW and RF fields are subject to a number of factors that can influence
the measurement accuracy and impose special requirements on sensor performance [37, 381.
In the far-field (generally defined as greater than one wavelength from the source) the electric
and magnetic fields are proportional and measurement of either field is sufficient to
characterize the total field strength. In the near-field (ie, in relatively close proximity to the
source), which is the region of most concern from the standpoint of hazard evaluation, the
electromagnetic field structure may be highly inhomogeneous resulting in sharp spatial
oscillations in the field strength and significant variations in the electric and magnetic com-
ponents. Other factors that must be accounted for include the presence of multiple sources
or multiple frequencies (eg, harmonics), reflected or scattered signals from nearby structures,
and the polarization and modulation properties of emitted signals. Perturbations of the field
by the instrument itself must also be considered in this region.
Hazard surveys usually entail measurements of broadband field intensity. Narrowband
spectrum analyzers or field strength meters can provide frequency discrimination where it is
necessary. The two most common sensors used for ELF/RF/MW electric-field monitoring are
thermocouples and diodes (Table 20-1). Connecting a small loop antenna to either of these
sensors imparts sensitivity to magnetic fields. In most instruments, three sensors are arranged
in mutually orthogonal directions in order to obtain isotropic response characteristics. While
diode sensors are generally more sensitive than thermocouples, they suffer from nonlinearities
at higher intensities and are less suitable for situations involving multiple frequency sources
or modulated fields. Personal dosimeters have been developed for measuring ELF exposure
[39], but similar devices designed for measuring the higher frequency fields suffer interference
from interactions with the body [38].

20.4 Biological Hazards

Human health effects from biological agents, such as bacteria, viruses, protozoa, fungi, and
pollens, include acute and chronic infections, and toxic and allergic reactions [40, 411. Many
of these organisms are transmitted only through intimate contact with an infected host and
require clinical detection. Other biological agents can be carried through food, water and/or
air and are theoretically amenable to detection with environmental sensors. Foodborne agents
such as Salmonella, Staphylococcus and aflatoxins are usually monitored as part of quality
control efforts during food processing and packaging operations [42].
Due largely to improvements in sanitary practices (eg, food sterilization and treatment of
wastewater and drinking water), disease rates from food and waterborne pathogenic
organisms have been virtually eliminated in most industrialized nations. However, in many
20.4 Biological Hazards 537

developing countries adequate treatment facilities are not available. The prevalence of high
viral and bacterial water pollution levels in these countries contribute to the elevated rates of
morbidity and mortality from pathogenic infections [l].
The ubiquity of microorganisms in the environment makes the detection of specific
biological agents difficult. Special culturing procedures and/or microscopic inspection are
generally required for positive identification of a given species [43]. Because of the need for
specialized techniques, as well as the low concentrations of most waterborne organisms, in-
dividual pathogens are rarely monitored directly in water supplies. Rather, one class of
organisms, coliforms, is used as a surrogate indicator of pathogenic contamination. Coliforms
(specifically E. coli) are normal bacterial constituents of the gut flora of warm-blooded
animals. They are easily detected and hardy, surviving longer than most known pathogens.
Since human and animal fecal waste is the most common source of waterborne pathogens,
a low coliform count is indicative of uninfected water.
The risk of exposure to airborne microorganisms is of particular concern in agricultural
workers, microbiologists, and individuals working with laboratory animals [41]. Fungal spores
found in hay, sugar cane, and various wood dusts, for example, can cause acute respiratory
infection [40]. Repeated exposure may lead to allergic sensitization where subsequent exposure
to even minute doses can elicit severe asthmatic attacks. Lab animal allergy (LAA) caused by
inhalation of aerosolized allergens shed from the fur or excreta of animals has become a
significant problem in animal research laboratories [44]. Outbreaks of microbial disease in of-
fice buildings (eg, legionellosis and hypersensitivity pneumonitis) have been linked to potable
water supplies, ventilation heat-exchange systems and other humid environments within the
buildings [45, 441.
Collection of bioaerosols is typically done by filtration or inertial impaction [46]. In some
cases liquid or gel culture media are employed directly in the sampler. Problems can arise with
dessication or disruption of cells which prevent or inhibit culture growth, precluding an ac-
curate assessment of air concentration levels. Certain organisms are inherently difficult to
culture and must be idevtified microscopically by size, shape and other morphological
features.
Recent advances in recombinant DNA research and molecular biology have spawned the
biotechnology industry which promises great benefits for agriculture, pollution control and
medicine, but has created some unique environmental sensing challenges. Realization of the
benefits will require the deliberate release of genetically modified organisms into the environ-
ment. While the consequences of human and ecological exposure to bioengineered organisms
are not known with certainty, opponents have predicted disaster and called for a moratorium
on genetic engineering [47]. The challenges in biotechnology-related sensing derive from both
the general difficulty in directly sensing microorganisms and the need to differentiate between
slightly modified organisms and the natural strain. These capabilities, with samples from
water, soil, potential animal hosts, and plants, are central to monitoring transport, prolifera-
tion and fate of engineered mircoorganisms.
Traditional detection methods involve counting colonies after culturing in a specific
medium or monitoring solid or gaseous byproducts of metabolism. These processes are time-
consuming and labor-intensive. Furthermore, various genetic markers, such as resistance to a
particular antibody or production of a chromogenic substance, must be employed after cultur-
ing in order to identify the bioengineered strains. Restriction enzyme mapping of DNA, or
fluorescent- or radioactive-labeled DNA probes can also be used to confirm the presence of
538 20 Environmental Monitoring

recombinant organisms [48]. Immunoassays, which are amenable to microsensor implementa-


tion, have become an important segment of microbiological as well as clinical analysis [49].
These methods involve binding between specially designed antibodies and specific sites on an
organism. Where monoclonal antibodies are used specificity can be excellent, but these ap-
proaches can be limited by a lack of sensitivity, interference by bacterial slimes, and the in-
ability to distinguish living and dead cells [48].

20.5 Chemical Hazards

20.5.1 Water Monitoring

Approximately 97% of the water on Earth is saltwater; of the 3% that is freshwater, about
77% is in the polar ice caps, 22% is groundwater, and 1% is surface water (lakes, rivers, snow,
soil moisture, and water vapor) [50]. In addition to biological contamination, discussed in the
previous section, water quality is also compromised by chemical pollution. The adverse effects
of chemical impurities in water range from aesthetic problems and corrosion of plumbing to
disruption of ecosystems and toxicity in humans. Interest in water pollution has previously
centered on surface waters, but now includes groundwater and the oceans, since pollution in
these has risen in many places to unacceptable levels.
Regulatory agencies attempt to maintain water purity by controlling sanitary-sewer
wastewater and industrial effluents returned to rivers and aquifers, and by enforcing hygienic
standards for drinking water. As with biological water pollution, runoff is a major source of
chemical contamination of surface water, carrying pesticides, herbicides, and fertilizer from
agricultural land, as well as salt, oil, and consumer chemicals from municipal sewers. Heavy
industry located near surface water sources has contributed significantly to the pollution of
inland and coastal waters, often rendering fish and other seafood unsafe for human consump-
tion. Heat from power-plant Qr industrial discharge can also be a form of water pollution:
a reduction in dissolved oxygen, combined with increases in metabolism of cold blooded
animals accompanying even small increases in water temperature, can prove lethal. Airborne
sulfur and nitrogen oxides from industrial emissions, along with sulfur released from volcanic
eruptions, return to the surface as acidic species and lower the pH of lakes.
Once thought to have inexhaustible capacity to neutralize chemical waste, the oceans are
now fouled along many coasts with heavy metals from industrial sources. These are concen-
trated in bottom-dwelling sea life and passed up through the food chain. Other ocean
phenomena suspected of being related to pollution are the algal blooms which cause red and
brown tides. Runoff from surface water containing nitrogen and phosphorus fertilizers is
thought to stimulate this explosive growth of algae, which blocks sunlight from reaching
submerged plants [51]. The ensuing decay depletes local oxygen levels, creating so-called
“dead zones” of anoxic water. Large-scale fish and dolphin kills have resulted both from
toxins produced by algae and from oxygen depletion of the water [52]. Remote sensing
(discussed in Section 20.6) is invaluable in monitoring the oceans where only sparse sampling
would otherwise be possible.
- .
Groundwater naturally contains dissolved or suspended minerals. Selenium, vanadium, and
zinc, are beneficial to human health in trace quantities. Calcium and magnesium, the cause
20.5 Chemical Hazards 539

of water hardness, are troublesome but not usually dangerous. Others, such as mercury, lead,
barium, beryllium, and cadmium are highly toxic. Groundwater quality has become a topic
of much greater concern since chemicals leaking from hazard.ous waste dumps and industrial
storage tanks have been found in drinking water supplies [53, 541. Cleanup of such leaks in-
volves drilling of test wells, soil sampling to define the extent of the contamination plume,
removal of contaminated soil at the source, and reclamation of the aquifer by extracting the
polluted water [55].

Table 20-2. Typical water quality parameters monitored in public drinking water systems. Maximum
Concentration Limits (MCL) listed are based on the U.S. Safe Drinking Water Act [after 561.

Parameter MCL Effects Sources


Turbidity 1-5 tu Interferes with disinfection Erosion, runoff
Radium-226, -228 5 pCi/L Bone cancer Geological, radioactive waste
Gross Alpha 15 pCi/L Cancer Uranium deposits, radioac-
tive waste
Total Coliform 1/100 mL Indicator of contamination Human and animal fecal
that may cause dysentery, matter
hepatitis, typhoid fever,
cholera, etc.
Giardia ___ Giardiasis (intestinal parasite)Mammal fecal matter
Fluoride 4.0 mg/L Skeletal damage Geological, water additive
Nitrate 10 mg/L Methemoglobinemia (blue- Fertilizer, sewage, feedlots,
baby syndrome) geological
Arsenic 0.05 mg/L Dermal and nervous system Geological, pesticides, in-
toxicity dustrial waste
Cadmium 0.01 mg/L Kidney effects Geological, mining, smelting
Lead 0,05 mg/L Nervous system damage, Lead pipes, lead-based
kidney effects solder pipe joints
Mercury 0.002 mg/L Central nervous system Industrial discharges,
disorders, kidney effects fungicides, geological
Selenium 0.01 mg/L Gastrointestinal effects Geological, mining
Benzene 0.005 mg/L Cancer Fuel and industrial solvent
leaks
Carbon Tetrachloride 0.005 mg/L Possible cancer Cleaning agents, coolant
manufacturing
1,2-Dichloroethane 0.005 mg/L Possible cancer Insecticides, gasoline
Trichloroethylene 0.005 mg/L Possible cancer Dry-cleaning materials,
pesticides, paints, degreasers
Vinyl Chloride 0.002 mg/L Cancer risk Polyvinylchloride (PVC)
pipes, manufacture of PVC
Endrin 0.2 pg/L Nervous system/ kidney effects Insecticide
Linda n e 4 Pg/L Nervous system/ liver effects Insecticide
Methoxychlor 100 pg/L Nervous system / kidney effects Insecticide
Toxaphene 5 Pg/L Cancer risk Insecticide
2.4-D 100 pg/L Liver / kidney effects Herbicide

- - - No MCL specified.
540 20 Environmental Monitoring

Methods for monitoring chemical contaminants in water employ a wide range of analytical
instrumentation, the preferred approach in a given application being dictated by
characteristics of the analyte, concentration levels, and the nature of potential interferents. A
variety of physical, radiological, microbiological and chemical parameters are routinely
monitored in potable water. Table 20-2 is a sample of these parameters showing typical sources
and maximum concentration limits mandated by laws for public drinking water systems. Stan-
dard methods of analysis are predominantly reagent-based, though chromatographic, spec-
troscopic, and electrochemical instruments are prescribed for some measurements [43, 57, 581.
The need for continuous monitoring of water quality for closed-loop control of water
softeners and water treatment facilities is stimulating development of small direct-reading sen-
sors for hardness, disinfection byproducts, and other ions [59].
Where complex mixtures of organic contaminants are present or the identities of the
analytes are unknown, analysis by gas chromatography/mass spectrometry ( G U M S ) is
usually required. Several imaginative approaches have been applied to tracking dispersion of
pollution from waste sites in lieu of costly GC/MS analyses. For example, certain chemical
compounds in waste leachate have been used as tracers to identify contaminants in fish and
sediment as far as 300 km from the source [60]. Leaks from hydrocarbon fuel tanks containing
well-characterized chemicals can be analyzed by simple gas chromatography with photo-
ionization detection [61]. Another example of lower cost groundwater monitoring involves the
use of soil-gas analysis to define the plume boundaries in the vicinity of a solvent leak. In
one report, volatile organics in the soil atmosphere were sampled and then analyzed using gas
chromatography with electron-capture detection [62].
Immunoassay techniques, mentioned in the previous section, have also been applied to the
detection of pesticides and carcinogens [63]. The formats most used in these measurements
are the traditional enzyme-linked immunosorbent assay, radioimmunoassay, and fluorescence
immunoassay. Sensors combining the unparalleled specificity of antibodies and antigens with
optical fibers [64], electrochemical detectors [65] and piezoelectric devices [66, 671 have been
reported.

20.5.2 Air Monitoring

20.5.2.1 Outdoor Air Monitoring

On the basis of volume, the primary sources of airborne chemicals are industrial manufac-
turing and production facilities, fossil-fuel burning power plants, automotive exhaust, and in-
cinerators. Accordingly, a majo: focus of air pollution monitoring and regulatory activity has
been on these sources and their primary emissions: nitrogen oxides, sulfur dioxide, carbon
monoxide, lead, total hydrocarbons, and suspended particulates [68]. Ozone and other
photochemical oxidants are “secondary pollutants” formed by the action of sunlight on
nitrogen dioxide and other atmospheric gases. The majority of these chemicals are respiratory
health hazards in humans [69]. Lead can affect the central nervous system and blood-forming
organs [70]and carbon monoxide is an acute chemical asphyxiant [71]. Hydrocarbons have
a range of health effects [72] as well as contributing to the formation of ozone and
photochemical smog in the lower atmosphere. Ozone and oxides of nitrogen and sulfur can
also have deleterious effects on vegetation [73]. In addition to these seven major pollutants,
20.5 Chemical Hazards 541

increased efforts have been mounted recently to characterize and regulate ambient levels of
volatile organic compounds (VOCs), particularly those suspected of causing cancer and other
specific health effects [74].

Table 20-3. Common direct-reading instrumentation for major air pollutants [102].

Operating Principle Applications / Comments Sensitivity / Selectivity


Chemiluminescence Light produced upon reaction of Low-ppb; excellent selectivity
analyte with reagent; 0, by reaction
with ethylene; NO by reaction with
0 3
Conductivity Aqueous-phase ions detected; ap- Low-ppb (SO,); low-ppm for most
plications include SO,, acidic and other species; poor selectivity
basic gases, CI,, freons, halogenated
organics; pyrolysis pretreatment used
for nonionic species
Electrochemical Amperometric or potentiometric Low-ppb to ppm; poor to good selec-
Oxidation / Reduction detection of many electroactive tivity depending on cell voltage, elec-
pollutants, including CO, 0,. CI,, trode catalysts, membrane permea-
SO,, NO, NO,, H,S, HCHO, bility
hydrides
Catalytic Oxidation CO, H,S, hydrocarbons oxidized on Low-ppm (metal-oxide semiconduc-
metal-oxide semiconductor (e.g., tor); high-ppm to percent (Pt fila-
SnO,) or on Pt filament; measure ment); poor selectivity
change in current
Flame- and Photo- General detectors for organic com- High-ppb to low-ppm; can be used as
Ionization pounds; measure current from ions gas-chromatographic detectors to
produced by burning in flame or achieve selectivity; photo-ionization
ultraviolet photo-ionization detector insensitive to most alkanes
and other compounds with high
ionization potentials
Colorimetry Reaction with chromogenic reagents Low-ppb to ppm, compound-depen-
in solution or impregnated on paper dent; selectivity depends on reagents
tape followed by photometric detec- employed
tion; SO,, NO,, NH,, CI,, 0,, other
reactive inorganic and organic
gases / vapors
Spectrophotometry Light absorption is proportional to Infrared: low-ppm, moderate selec-
gas phase concentration; infrared tivity; ultraviolet and visible: ppb to
used for organic and dipolar in- ppm, variable selectivity
organic gases; ultraviolet used for
SO,, 0,, Hg; visible used for NO,
Aerosol Photometry ,;
Scattering or attenuation of incident Low-pg/m poor selectivity; particle
light by aerosol particles; forward size and composition can affect ac-
scattered light used most often curacy
542 20 Environmental Monitoring

Monitoring outdoor levels of these chemicals is performed using networks of fixed-site


monitoring stations located on building tops and various other sites around population
centers [68, 751. Emission-stack sampling for major pollutants is also performed, though
regulations are often targeted only at the largest-volume producers [68]. A wide variety of
automated monitoring instruments have been developed for ozone, nitrogen oxides, sulfur
dioxide, and carbon monoxide, most of which take advantage of the reactivity of these
chemicals [76]. Total hydrocarbon measurements can be made using catalytic-filament or
ionization-based detectors, and suspended particulates can be monitored using optical scatter-
ing or transmission instruments [77]. Table 20-3 lists the most common direct-reading in-
struments currently used for monitoring stack emissions and ambient pollutant concentra-
tions.
Where continuous monitoring is not feasible, air samples can be collected and subsequently
analyzed by any of a range of laboratory instruments. For identifying unknown mixtures of
organic compounds GC/MS is the most common approach [78], though more traditional
flame-ionization, photo-ionization and electron-capture GC detectors are still widely used.
Concern over carcinogenic organic particulates has led to methods that analyze certain soluble
fractions for their mutagenic activity rather than, or in addition to, detailed structural analysis
[79]. For metals and other elemental analyses, either inductively-coupled plasma atomic emis-
sion spectroscopy (ICP-AES) or graphite-furnace atomic absorption spectrophotometry (GF-
AAS) can be used [4]. In response to the complexity of many environmental samples, there
is a trend toward coupling detectors together to create novel hybrid instruments. Examples in-
clude ICP-AES/MS and triple-quadrupole mass spectrometers (MUMS).
There is an increasing trend toward adapting sophisticated laboratory instruments to field
use. Portable gas-chromatographs have been available for many years, benefitting recently
from advances in silicon micromachining and column and detector technologies [80,811. The
emergence of on-site direct-inlet mass spectrometers [82, 831, Fourier-transform infrared spec-
trophotometers [84, 851, portable ion-mobility spectrometers [86], and tunable atomic-line
molecular spectrometers [87] for organic gas and vapor monitoring holds great promise for
identification of contaminants in the field; however the improved analytical capabilities of
these instruments are partially offset by their complexity and high cost.

20.5.2.2 Indoor Air Monitoring

Several studies have indicated that the airborne contaminant levels arising from sources in
homes, office buildings, schools, and other indoor environments can be much higher than
those found outdoors [88-961. Since most people spend about 90% of their time indoors, the
impact on health from low-volume indoor pollution sources may be much greater than that
from major outdoor sources. Some of the chemicals measured in these studies and their
respective indoor sources include benzene and particulates from cigarette smoke, nitrogen
dioxide and carbon monoxide from gas stoves, and chlorinated hydrocarbons from dry-
cleaned clothes (tetrachloroethylene), air deodorizers (p-dichlorobenzene), and hot-water
showers (chloroform). Building surveys have revealed low levels of hundreds of aliphatic,
aromatic, and chlorinated organic compounds emanating from a variety of sources, including
paints, carpeting, particle board, cleaning solvents, and pesticides [96]. These findings are
prompting a shift in regulatory thinking: the concept of total exposure assessment (TEA),
20.5 Chemical Hazards 543

where a focus is placed on individuals and their local environments, has been suggested as a
more rational approach to hazard assessment than the traditional emphasis on high-volume
pollutants in the general ambient environment [88, 951. Implementation of the TEA concept
will have a dramatic influence on current approaches to pollution monitoring, creating in-
creased demand for compact, inexpensive instruments to be placed in the home or worn on
the body. The development of new sensors for such instruments will be critical to these efforts.

20.5.2.3 Occupational Air Monitoring

The value of personal monitoring has been recognized in the field of industrial hygiene for
many years. Virtually all standards for occupational exposure to toxic chemicals require per-
sonal, breathing-zone exposure measurements [97]. Fixed-site (area) measurements do not cor-
relate well with individual exposures due to a combination of worker mobility and spatial
variations in contaminant concentrations. Temporal variability is also important: intra-day
and inter-day fluctuation in contaminant levels gives rise to distributions of individual ex-
posures ranging over several orders of magnitude even for well-controlled industrial processes
[981.
Most personal sampling methods employ a battery-operated sampling pump to pass air
through a collector, such as a small sorbent-filled tube, attached to the lapel of the worker
[99]. Samples are collected in the worker’s breathing zone for periods of a few minutes to
several hours, and subsequently analyzed to yield a time-weighted-average exposure value. The
inconvenience of operating and maintaining the sampling pumps has led to the development
of badge-type passive monitors which rely on diffusion through a stagnant air layer or
permeation through a thin polymer membrane to control transport of gases and vapors to the
surface of a collector [loo, 1011. The effective sampling rate is determined by the diffusion or
permeation coefficient of the contaminant and the dimensions of the sampling aperture.
While these devices are unobtrusive and simple to use, they are generally limited by their low
sampling rates to monitoring long-term (ie, several hours) exposures. As with pump-based
methods, chemical analysis is necessary to obtain the time-averaged contaminant concentra-
tion.
In recent years, there has been a marked increase in the number and sophistication of direct-
reading instruments designed for monitoring toxic chemicals, oxygen deficiency, and flam-
mable or explosive atmospheres in the workplace [102-1041. These range from large multi-
point systems for process control and leak detection to portable survey instruments. In most
cases, the operating principles of these instruments are similar to those listed in Table 20-3.
Several hand-held (and fewer pocket-sized) instruments now operate continuously and have
alarms to warn of excessive exposure levels. Some accommodate data loggers for obtaining
individual exposure histories over a given sampling interval. Direct computer interfacing also
eases the data management burden. While such features facilitate monitoring worker ex-
posures, most of these instruments suffer from limitations associated with the types of sensors
employed. For example, the majority of hand-held instruments used for monitoring organic
gases and vapors employ catalytic combustion, semiconductor charge-transport, or photo-
ionization detection mechanisms. These sensors generally provide adequate sensitivity, but
selective measurement of specific organic compounds or classes of compounds is not possible:
the presence of even simple mixtures of organic contaminants can preclude an accurate deter-
544 20 Environmental Monitoring

mination of the exposure hazard due to any of the mixture components. Pocket-sized in-
struments employing electrochemical detection or reagent-impregnated paper-tape col-
orimetry provide improved selectivity, but their use is limited chiefly to reactive inorganic
gases. Furthermore, with these techniques, only one type of contaminant can be detected with
a given instrument.
Monitoring exposures to aerosols must not only take account of the nature and concentra-
tion of the contaminant but also the size distribution, since this is the principal determinant
of where in the respiratory tract the aerosol will be deposited. For certain mineral dusts, such
as crystalline silica, one is primarily interested in the fraction reaching the alveolar region (ie,
the respirable mass) since this is the site of toxic action (fibrotic scarring in the case of silica).
For certain allergens, the aerosol fraction reaching the tracheo-bronchial region (thoracic
mass) is thought to be most important in eliciting respiratory asthma. For soluble systemic
toxicants one is interested in the entire inspired mass since absorption can occur at any point
in the respiratory tract. These size-selective sampling criteria have recently been incorporated
into recommended exposure limits for aerosols (see Table 20-4) [105, 1061.

"able 20-4. Recommended size-selective sampling criteria for aerosol samplers [lo61

Mass Fraction Collection efficiency criteria


Inspirable Particulate Mass E = 50 (1 + e-O."%) 3z 10 for 0 < d,, < 100 pm
E = collection efficiency (To)
d,, = aerodynamic diameter (pm)
Thoracic Particulate Mass lognormal distribution with median aerodynamic diameter =
10 * 1 pm and geometric standard deviation = 1.5
Respirable Particulate Mass lognormal distribution with median aerodynamic diameter =
3.5 * 0.3 pm and geometric standard deviation = 1.5

Portable direct-reading aerosol monitors employ one of the following detection


mechanisms: beta-attenuation, piezoelectric-crystal mass loading, or light scattering [107].
Light scattering photometers are the most common aerosol detectors. Unfortunately, their
response is affected by changes in the size distribution and composition (refractive index) of
the aerosol, making calibration more difficult [108]. Detectors based on beta-attenuation and
piezoelectric crystals are less sensitive to these variables. One type of photometer-based instru-
ment is designed specifically for fibrous aerosol detection: an oscillating voltage aligns fibers
perpendicular to the path of a laser, permitting differentiation of fibrous and non-fibrous par-
ticles [109]. Most available models can be equipped with inertial preselectors (e.g., a 10-mm
cyclone) to remove non-respirable particles upstream from the detector. The principal limita-
tion common to these instruments is the inability to provide information on aerosol composi-
tion since detection is based solely on mass. Additional difficulties are encountered with gas
and vapor adsorption on aerosol surfaces and with reactive or unstable aerosols.
Fortunately, in most industrial worksites the nature of possible hazards can be determined
from chemical feedstocks, reactor processes, endproducts, byproducts and work practices.
This knowledge simplifies the selectivity requirements of sensors since the identity and likely
physical state of the contaminant(s) and potential interferents can be determined before
20.5 Chemical Hazards 545

sampling. Arguably, similar a priori hazard identifications could be made in offices and
homes. The small size and low power requirements of microfabricated chemical sensors lend
themselves to incorporation into personal direct-reading instruments. For gas and vapor
monitoring, sensor arrays might be used to provide enhanced selectivity, particularly for
organic vapors, and to expand the range of inorganic gases detectable with a single instrument
[110, 1111. For aerosols, there is a need to augment current sensors with additional capabilities
for surface and elemental analysis.

20.5.3 Biological Monitoring

An alternative, or complement, to air monitoring for assessing exposure to chemicals is


biological monitoring, entailing the analysis of human biological media for the presence of
a chemical, its metabolite, or some other biochemical index of exposure [112]. The most com-
mon media employed are exhaled breath, urine, and blood, although hair, nails, saliva, and
fat have also been used for certain chemicals.
One advantage of biological monitoring is that it provides the best measure of the absorbed
dose of a chemical and, thereby, yields a more accurate estimate of risk to an individual.
Monitoring air concentrations, even in an individual’s breathing zone, cannot account for in-
terpersonal differences in ventilation rates and retention factors, or other physiological factors
such as age, weight, and health status, which can affect susceptibility to adverse chemical ef-
fects. In addition, biological monitoring can account for exposure via all routes (ie, inhalation,
dermal absorption, or ingestion) and from all sources of exposure.
Monitoring blood and urine levels of chemicals has been used for many years in medical
surveillance programs, chiefly for heavy metals [113]. Only recently have efforts been mounted
to incorporate such methods into routine exposure monitoring programs, and to expand the
list of target chemicals to include, for example, a variety of organic solvents [114]. A steadily
increasing number of recommended biological reference values are being adopted by occupa-
tional health organizations [115, 1161. Examples of several biological indices of exposure are
shown in Table 20-5. The use of biomarkers to document exposures in the general public has
also been suggested [117].
The collection and interpretation of biological monitoring data require an understanding
of the uptake, distribution, metabolism, and excretion of the chemicals of interest. Knowledge
of the factors affecting these toxicokinetic parameters for the target chemicals is also necessary
to accurately correlate the measured quantity with the dose received. Since the concentrations
of chemicals and metabolites are in flux, timing and frequency of sample collection are
critical.
Many organic solvents can be measured directly in exhaled breath during, or up to several
hours after, exposure [118, 1191. Given the short half-lives of many solvents in the body,
however, the breath concentration is affected strongly by the timing of sample collection and
by the exposure profile. Thus, for samples collected at the end of the day, a person exposed
only in the morning will have a much lower breath concentration than a person exposed in
the afternoon even though their daily total exposures are equivalent. Furthermore, for highly
lipid-soluble solvents (eg, chlorinated hydrocarbons) two individuals with different amounts
of body fat will have different breath concentrations even for identical exposure profiles. Tox-
icokinetic models are useful in these situations to aid in interpreting the data [120]. Repetitive
546 20 Environmental Monitoring

Table 20-5. Examples of recommended biological indices of occupational exposure to industrial


chemicals [115].

Index Sample Collection Maximum Concentration

Styrene
Mandelic acid in urine end of shift 1 g/L; 0.8 g/g creatinine
Styrene in blood end of shift 0.55 mg/L
Styrene in mixed-exhaled air prior to next shift 40 PPb
Trichloroethylene
Trichloroacetic acid in urine end of work week 100 mg/L
Trichloroethylene in end-exhaled air prior to last shift of week 0.5 ppm
Carbon Monoxide
Carboxyhemoglobin in blood end of shift 8 To
CO in end-exhaled air end of shift 40 PPm
Lead
Lead in blood timing not critical 0.5 mg/L
Lead in urine timing not critical 0.15 mg/g creatinine
Zinc protoporphyrin in blood after 1 month exposure 1 mg/L

sampling may be necessary in order to monitor the decay in chemical concentration over time.
Concentrations of urinary metabolites have also been recommended as measures of ab-
sorbed dose for many chemicals [115, 1211. For organic solvents, urinary metabolite concentra-
tions are less dependent upon exposure profile than are exhaled breath concentrations.
However, individual differences in metabolism will affect the concentration of these products
present in the urine over time. Co-exposure to chemicals that alter the activity of critical
metabolic enzymes can also affect metabolite concentrations: ethanol is one example of a
potential metabolic inhibitor of several organic solvents [122]. Standardization to urinary
creatinine levels or total dissolved solids is used to account for variations in urinary volumes
(Table 20-5).
The amount of the chemical or its metabolite reaching the ultimate site of toxic action in
the body, ie, the biologically effective dose, is the most relevant quantity in determining risk.
The biologically effective dose is rarely measured directly but can be inferred, in some cases,
from surrogate measures. For example, alkylation of DNA by ethylene oxide (EtO) is thought
to be the mechanism by which EtO causes cancer. Measuring alkylated DNA is impractical.
However, hemoglobin alkylation by EtO is well-correlated with DNA alkylation, permitting
use of the former as a surrogate measure of the latter [123, 1241. Carboxyhemoglobin can be
used to indicate the amount of carbon monoxide reaching the red blood cells, and
methemoglobin levels are useful indicators of effective doses of certain amines and nitro com-
pounds 1711.
Biological monitoring can also be used to measure the effect that the chemical has exerted
at the target site. As an example, blood-cholinesterase levels indicate cholinesterase inhibition
caused by exposure to organophosphorus pesticides [125]. Blood levels of delta-aminolevulinic
acid or zinc protoporphyrin are relevant indicators of interference by inorganic lead in the syn-
thesis of heme [126]. Increased sister-chromatid exchanges in peripheral lymphocytes are in-
dicative of somatic mutations, though the relationship between these phenomena and the risk
20.6 Remote Sensing 547

of cancer has not been established [127]. Levels of certain enzymes can also be used as (non-
specific) indices of early damage to organs such as the liver and kidney [l28].
Despite increased research efforts to develop and validate biological indices of exposure, the
use of biological monitoring in the field is still limited. The additional analytical burden
associated with a biological monitoring program undoubtedly contributes to this situation.
Another complicating factor in blood and urine testing is the need to refrigerate samples until
analysis to prevent degradation. Simple field instruments that could be used by untrained per-
sonnel are needed to provide rapid analysis of biological samples. The development of sensors
capable of selectively measuring chemicals or their metabolites in breath, urine, and blood
would be of considerable value in promoting the use of biological monitoring in the workplace
and eventually in the home.

20.6 Remote Sensing

Viewed in a global sense, our environment includes all of the atmosphere, land, and water
on Earth. Earth scientists are most interested in elements which are important to life and have
complex biogeochemical cycles, such as oxygen, hydrogen, carbon, nitrogen, phosphorus, and
sulfur. Research since the 1950s has established many interactions between land, oceans and
the atmosphere that influence climate and weather. For example, an anomalous rise in
temperature that occurs periodically in the equatorial Pacific Ocean (known as “El Niiio”)
has been correlated with modifications in weather patterns throughout the world [129]. Ac-
curate weather prediction requires knowledge of air and ocean speeds and temperatures,
humidity levels, cloud cover, solar activity, land surface temperatures, and snow and ice cover.
The Earth’s biota also affect the climate. Human activity, for example, has increased the
atmospheric content of carbon dioxide by 25% since the industrial revolution. An increase of
more than 10% occurred between 1958 and 1986 [130]. This if of concern because carbon diox-
ide, while transparent to sunlight, reflects energy radiated outward from the Earth, causing
a net warming trend through the “greenhouse effect” [131]. Other gases, such as methane,
chlorofluorocarbons, and tropospheric ozone, though at lower concentrations than carbon
dioxide, are increasing at a faster rate, and have greenhouse effects per molecule as much as
10000 times that of carbon dioxide [130]. Another global environmental issue is the depletion
of ozone in the stratosphere [132]. Ozone absorbs ultraviolet light from the Sun, influencing
weather and shielding plants and animals on Earth from damaging levels of ultraviolet radia-
tion. An alarming decrease in Antarctic ozone levels in the mid-1970s raised the possibility that
chlorofluoromethanes, used as propellants in aerosol products and as refrigerants, would
cause a major decrease in the ozone layer worldwide.
These are a few familiar examples of environmental parameters earth scientists wish to
monitor globally. There are many more in the areas of agriculture and forestry, ocean-resource
management, upper-atmosphere chemistry, pollution monitoring, and prediction of earth-
quakes and volcanic eruptions. When regional or global measurements of these kinds are to
be made, satellite-based remote sensing is often the only practical approach.
Remote sensing is currently performed from numerous geostationary, Sun-synchronous,
and polar-orbiting satellites operated by many nations. Instrumentation used for satellite
548 20 Environmental Monitoring

remote sensing can be categorized as passive (sensors only) or active (providing its own radia-
tion source), and employs electromagnetic energy in the ultraviolet to microwave spectral
range [133].Passive instruments are used for visible imaging, radiometry, chemical analysis,
and sounding. The applications of passive imagers include monitoring cloud formation and
motion, soil erosion, deforestation, land and water surface temperature, and vegetation type
and condition. Broad-band radiometers are used to track the balance between energy received
from the Sun and that reflected and thermally emitted from the Earth. This radiation budget
is of key interest to long-term climate change. Passive chemical analyses are made by sensing
radiation emitted, scattered, or absorbed by the atmosphere. Aerosols, ozone, and other trace
gases have been measured by viewing the Sun through the stratosphere and detecting absorp-
tion of certain wavelengths by molecules of interest [137].Sounders are used to create three-
dimensional atmospheric maps of temperature to a resolution of 2°C. With sensors tuned to
specific frequencies, sounders measure electromagnetic energy emitted by certain atmospheric
constituents. Depending upon temperature, the atmosphere will either be transparent at these
wavelengths or will absorb and emit energy. Molecular oxygen is responsible for this
phenomenon at microwave wavelengths; carbon dioxide can be used at infrared wavelengths.
Similarly, water vapor molecules can be used by sounders to remotely probe atmospheric
moisture.
Active remote sensing systems include radar for microwave altimetry, scatterometry, and
imaging, and Light Detection and Ranging (LIDAR) for measuring atmospheric chemicals,
temperature and pressure profiles, and wind speeds. Altimeters measure altitude from the time
required for reflection of a transmitted microwave pulse. This approach is used for mapping
ocean topography (a resolution of 5 cm has been achieved), from which surface currents can
be inferred [134].Scatterometers are used for marine forecasting; using the amplitude of
energy reflected from the ocean surface, they measure near-surface wind speed and direction.
Microwave radar has the advantage (over visible imagers) of being able to image independent
of sunlight, and through cloud cover. Synthetic aperture radar is a powerful tool capable of
mapping, for example, geological features in three dimensions, subsurface bedrock in arid
regions, soil moisture, and vegetational composition. Light Detection and Ranging (LIDAR)
systems are similar to radar, but use a laser source at visible or infrared wavelengths combined
with a telescope for collecting energy scattered by aerosols and gas molecules. An integral
detector quantifies back-scattered radiation as a function of time. This technology has been
demonstrated in ground-based and airborne systems, and will be applied in space [135].In ad-
dition to identifying and quantifying atmospheric constituents, LIDAR is used with multiple
wavelength differential absorption techniques to map atmospheric temperature and pressure
profiles. Improvements in laser stability have made possible detection of the Doppler shift of
energy backscattered from the atmosphere, providing direct wind-profiling capability [136].
Satellite-based remote sensors must be calibrated using data from instruments on the sur-
face or in aircraft or balloons. Calibration stability will become increasingly important as
earth scientists attempt to track more subtle changes in atmospheric chemistry and
geophysical variables. Demands for ever greater precision and resolution continue to drive the
development of new remote sensors. Ozone-related studies of the upper atmosphere, for exam-
ple, need sensitivity to chemicals in the parts-per-trillion and -quadrillion range. There is a
trend in remote sensing toward the use of spectrometers, which are more powerful and flexible
than the single-frequencysystems. An example of this trend is the ATMOS infrared absorption
spectrometer, which has produced data on atmospheric gases from altitudes of 5 to 140 km,
20.8 References 549

and on nitrogen and halogen compounds in the stratosphere [137]. Future imaging in-
struments will also operate in spectrometer mode; with amplitude information for each pixel
in as many as 100 spectral bands, they will provide much more information than their single-
frequency predecessors [138, 1391. An immediate goal of this multiband, or color, imaging
capability will be to map ocean chlorophyll. The use of solid-state detectors in place of vidicon
tubes has reduced weight, extended lifetimes, and lowered power requirements. With over 4
million pixels, solid-state imaging sensors now lead conventional technology in resolution
[140].

20.7 Summary

Clearly, important challenges remain in the area of environmental sensing. The scope of
these challenges ranges from monitoring global atmospheric phenomena to detecting altera-
tions in the structure of DNA. In the foregoing discussion we have attempted to provide some
insight into the nature of environmental pollution problems and the difficulties encountered
in their evaluation. This overview of current approaches to sensing radiological, biological and
chemical hazards in various environmental media points out the importance of application-
specific variables in the design of sensors. Still, a few general environmental sensing trends
can be identified. In assessing the risks to human health from chemical hazards, increasing
emphasis is being placed on measuring individual exposures, with personal monitors or
through analysis of biological fluids. Tighter regulatory restrictions on volatile organic
chemicals in air and water will create a greater demand for inexpensive field monitors for
tracking the dispersion of these contaminants in the environment. Requirements for multi-
species detection and enhanced selectivity in these applications will be addressed by portable
instruments equipped with miniaturized sensors or sensor arrays. Preliminary reports on the
use of immunochemical interactions in sensors suggest some intriguing possibilities for selec-
tive detection of biological and chemical agents in the environment. Evaluation of the health
effects of chronic non-ionizing radiation exposure is currently limited by a lack of human ex-
posure data; the development of radio-frequency and microwave dosimeters, and UV detectors
that incorporate physiologically relevant spectral weightings, would aid collection of this in-
formation. Finally, continued improvements in spectrometers for imaging and chemical
analysis will provide increased sensitivity in remote-sensing systems for environmental applica-
tions.

20.8 References

[l] Global Pollution bnd Health, World Health Organization; London: Yale Press, 1987.
[2] Deisler, P. F., Environ, Sci. Tech. 22, No. 1, (1988) 15-19.
131 McDowall, M. E., The Identification of Man-made Environmental Hazards to Health, London:
MacMillan, 1987.
550 20 Environmental Monitoring

[4] Gurka, D. F. Betowski, L. D., Hinners, T. A., Heithmar, E. M., Titus, R., Henshaw, J. M., Anal.
Chem. 60 (1988) 454A-467A.
(51 NCRP Report No. 76, (1984), National Council on Radiation Protection and Measurements,
Bethesda, MD, USA.
[6] Flynn, G. L., in: Percutaneous Absorption, Bronaugh, R. L., Maibach, H. I. (eds.); New York:
Marcel Dekker, 1985, p. 17.
[7] Vo-Dinh, T., Am. Ind. Hyg. Assoc. J. 4 4 No. 6, (1987) 594-598.
[8] Keenan, R. R., Cole, S. B., Am. Ind. Hyg. Assoc. J. 43 (1982) 473-476.
[9] Instrumentation for Environmental Monitoring; New York: Wiley, 1983, Lawrence Berkeley
Laboratory; Vol. 1, Ch. 3.
[lo] Reference (91, Chapter 2.
(111 International Commission on Radiological Protection, Ann. ICRR 1, No. 3, (1977), ICRP Pub. No.
26.
(121 NCRP Report No. 39, 1971, National Commission on Radiological Protection and Measurements,
Bethesda, MD, USA.
[13] Petkau, A., Acta Physiol. Scan. Suppl. 492 (1980) 81-90.
[14] Sternglass, E. J., Int. J. Biosocial Res. S, (1986), 7-36.
[I51 Eichholz, G. G., EnvironmentalAspects of Nuclear Power, Chelsea, MI: Lewis, 1985, pp. 501 -554.
[16] Reference [15], Chapter 12.
[I71 Young, A. L., Pix, G. D., Environ. Sci. Tech. 22, (1988) 733-738.
[IS] Barnes, E. C., in: The Industrial Environment - Its Evaluation and Control, 1973,
USDHEW(NIOSH), U. S. Govt. Printing Office, Washington, D. C., Stock No. 017-001-00396-4,
Ch. 29.
1191 Ehmann, W. D., Yates, S. W., Anal. Chem. 60,(1988) 42R-62R.
[20] Indoor Radon and Its Hazards, Bodansky, D., Robkin, M. A., Stadler, D. R. (eds.); Seattle: Univ.
of Washington Press, 1987.
[21] Hanson, D. J. Chem. Eng. News 67, No. 6, (1989) 7-13.
[22] Boyle, M., Environ, Sci. Tech. 22, No. 12, (1988) 1397-1399.
[23] Cooper, P. N., Introduction to Nuclear Radiation Detectors, New York: Cambridge, 1986, Ch. 1.
[24] Instrumentation and Monitoring Methods for Radiation Protection, 1978, The National Council
on Radiation Protection and Measurements, Washington, D. C.
[25] Knoll, G. F., Nucl. Instrum. Meth. Phys. Res. B24-25, (1987) 1021-1027.
[26] Occupational Hazards from Non-IonizingElectromagnetic Radiation, 1985, International Labour
Office, Geneva.
[27] Largent, E. J., Olishifski, J., in: Fundamentals of Industrial Hygiene, Plog, B. A. (ed.); Chicago:
National Safety Council, 1988; Ch. 11.
[28] TLVs, Threshold Limit Values and Biological Exposure Indices for 1988-1989, (1988), American
Conference of Governmental Industrial Hygienists, Cincinnati, pp. 104- 106.
[29] Environmental Health Criteria 14, UltravioletRadiation, (1979), World Health Organization, Inter-
national Radiation Protection Association, Geneva.
[30] Wilkening, G. M., in: The Industrial Environment - Its Evaluation and Control, 1973,
USDHEW(NIOSH), U.S. Govt. Printing Office, Washington, D. C., Stock No. 017-001-00396-4,
Ch. 28.
[31] Sliney, D. H., Wolbarsht, M., Safety with Lasers and Other Optical Sources, New York: Plenum,
1980.
[32] Biological wfects and Dosimetry of Nonionizing Radiation, Grandolfo, M., Michaelson, S. M.,
and Alessandro, R. (eds.); New York: Plenum, 1983.
[33] CRC Handbook of Biological Effects of ElectromagneticFields, Polk, C., Postow, E. (eds.); Boca
Raton, FL: CRC Press, 1986.
[34] Environmental Health Criteria 16, Radiofrequency and Microwaves, 1981, World Health
Organization, International Radiation Protection Association, Geneva.
20.8 References 551

Assenheim, H. M., Hill, D. A., Preston, E., Carinie, A. B., NRCCReport No. 16448, 1979, Na-
tional Research Council of Canada No. ISSN 0316-0114.
Aldrich, T. E., Easterly, C. E., Environ. Health Perspec. 75, (1987). 159-171.
Stuchly, M. A., Stuchly, S. S., in: CRC Handbook of Biological Effects of ElectromagneticFields,
Polk, C., Postow, E. (eds.); Boca Raton, FL: CRC Press, 1986, pp. 233-240.
Tell, R.A., in: Biological Effects and Dosimetry of Nonionizing Radiation, Grandolfo, M.,
Michaelson, S. M., Alessandro, R. (eds.); New York: Plenum, 1983, pp. 116-119.
Wigdor, M., USEPA Technical Report ORPIEAD 80-1, 1980, Washington, D. C., USA.
Barber, T. E., Husting, E. L., in: OccupationalDiseases - A Guide to Their Recognition, Cincin-
nati: USDHEW(NIOSH), 1977, pp. 45-78.
Miller, A. L., Volk, C. S., in: Fundamentals of Industrial Hygiene, Plog, B. A. (ed.); Chicago:
National Safety Council, 1988, Ch. 14.
S. Segall, Environmental Health, Purdom, P. W. (ed.); New York: Academic Press, 1971, pp.
76- 149.
Standard Methods for the Examination of Water and Wastewater, 1985, Greenberg, A. E.,
Trussell, R. R., Clesceri, L. S. (eds.); American Public Health Association, American Water
Works Association, and Water Pollution Control Federation, Washington, DC.
Morey, P. R., Woods, J. E., in: Occupational Medicine - State of the Art Reviews: Health Pro-
blems of Health Care Workers, Vol. 2, Emmett, E. A. (ed.); Philadelphia: Hanley and Belfus,
1987, NO. 3, pp. 547-564.
Murca, P. W., Stout, J. E., Yu, V. L., Yee, Y. C., Am. Ind. Hyg. Assoc. J 49, No. 11, (1988)
584- 590.
Muilenberg, M., Immun., All. Clinics N. Am,, in press.
Berg, P., Baltimore, D., Boyer, H. W., Cohen, S. N., David, R. W., Hogness, D. S., Nathans, D.
Roblin, R., Watson, J. D., Weissmann, S., Zinder, D., Nature 250, (1974) 175.
Biotechnology Risk Assessment, Fiksel, J., Covello, V. T. (eds.); Elmsford, NY: Pergamon Press,
1986, pp. 144-153.
Janata, J., Bezegh, A., Anal. Chem. 60, No.12, (1988) 62R-74R.
Space-Based Remote Sensing of the Earth, 1987, Oceanic and Atmospheric Administration and
National Aeronautics and Space Administration, Washington, DC.
Murakami, A., Information Bulletin 19, No. 2, (1973) 131 - 132, Nansei Regional Fisheries
Research Lab., Maruishi, Hiroshima, Japan.
Loughry, T., Surveyor 10, No. 4, (1976) 11-16.
Ground-WaterContamination:FieldMethods, Collins, A. G., Johnson, A. I. (eds.); Philadelphia:
ASTM, 1986.
Olivieri, A., Eisenberg, D., Kurtovich, M., Pettegrew, L., Journal of Water Resources Planning
and Management lll, No. 3, (1985) 346-358.
Iscoff, R., Semicon. Int. 8, No. 7, (1985) 112-119.
Fact Sheet: National Primary Drinking Water Standards, 1988, United States Environmental Pro-
tection Agency, Office of Water; U. S. Govt. Printing Office, Washington, D. C.
MacCarthy, P., Klusman, R. W., Anal. Chem. 59, (1987) 308R-337R.
Van Loon, J. C., Chemical Analysis of Inorganic Constituents of Water, Baca Raton, FL: CRC
Press, 1982.
Brown, R. B., Huang, T. C., Report FLECK-887 (1988). University of Michigan, Ann Arbor, MI,
USA.
Hites, R. A., Anal. Chem. 60, No.10, (1988) 647A-650A.
[61] Koehn, J. W., Stanko Jr., G. H., Environ. Sci. Tech. 22, No. 11, (1988) 1262-1264.
[62] Marrin, D. L., Kerfoot, H. B. Environ. Sci. Tech. 22, No. 7, (1988) 740-745.
[63] Vanderlaan, M., Watkins, B. E., Stanker, L., Environ. Sci. Tech. 22, No. 3, (1988) 247-254.
[64] Ives, J. T., Lin, J. N., Andrade, J. D., Am. Biotech. Lab. 7, No. 3, (1989) 10-18.
552 20 Environmental Monitoring

[65] Heineman, W. R., Hasall, H. B., Anal. Chem. 57, No. 12, (1985) 1321A-1331A.
[66] Ngeh-Ngwainbi, J., Foley, P. H., Kuan, S. S., Guilbault, G. G., J. Am. Chem. Soc. 108, No. 18,
(1986) 5444-5447.
[67] Roederer, J. E., Bastiaans, G. J., Anal. Chem., 55, (1983) 2333-2336.
[68] Cannon, J., J. Air Poll. Control Assoc., 36, No. 5, (1986) 562-573.
[69] Amdur, M. 0. in: Casarett and Doull’s Toxicology,Doull, J., Klaassen, C. D., Amdur, M. 0. (eds.);
New York: MacMillan, 1980, pp. 608-631.
[70] Landrigan, P. J., West. J. Med. W7, No. 6, (1982) 531-539.
[71] Smith, R. P., in: Casarett and Doull’s Toxicology, Doull, J., Klaassen, C. D., Amdur, M. 0. (eds.);
New York: MacMillan, 1980, pp. 319-321.
[72] Cornish, H. H., in: Casarett and Doull’s Toxicology, Doull, J., Klaassen, C. D., Amdur, M. 0.
(eds.); New York: MacMillan, 1980, pp. 468-496.
[73] Air Quality Guidelines for Europe, WHO Regional Publications, European Series No. 23, 1987,
World Health Organization for Europe, Copenhagen pp. 373-404.
[74] Berry, D. K., Environ. Sci. Tech. 20, No. 7, (1986) 647-651.
[75] Hinton, D. O., in: Air Sampling Instrumentsfor Evaluation of Atmospheric Contaminants, Lioy,
P. J., Lioy, M. J. Y., (eds.); American Conference of Governmental Industrial Hygienists, Cincin-
nati, 1983, Ch. C.
[76] Fox, D. L., Anal. Chem. 59 (1987) 280R-290R.
[77] Swift, D. L., in: Air Sampling Instrumentsfor Evaluation of Atmospheric Contaminants, Lioy, P.
J., Lioy, M. J. Y. (eds.); American Conference of Governmental Industrial Hygienists, Cincinnati,
1983, Ch. U.
[78] Swallow, K. C., Shifrin, N. S., Doherty, P. J., Environ. Sci. Tech. 22, (1988) 136-142.
[79] ToxicAir Pollution: A ComprehensiveStudy of Non-Criteria Pollutions, Lioy, P. J., Daisey, J. M.,
(eds.); Chelsea, MI: Lewis Publishers, 1987.
[80] Angell, J. B., Jerman, J., Barth, P., Sci. Amer. 248, No. 4, (1983) 44-55.
[81] Daisey, J. M., in: Advances in Air Sampling, Chelsea, MI: Lewis, 1988, pp. 225-242.
[82] ICAMS - Industrial Central AtmosphereMonitoring System, Perkin-Elmer, Applied Science Divi-
sion, Pomona, CA.
[83] Rotheram, M., Semicon. Safety Assoc. J., (1988) Sept., 51-64.
[84] Strang, C. S., Levine, S. P., Herget, W., Amer. Ind. Hyg. Assoc. J. 50, No. 2, (1989) 70-77.
[85] Strang, C. S., Levine, S. P., Amer. Ind. Hyg. Assoc. J. 50, No. 2, (1989) 78-84.
[86] Spangler, G. E., Carrico, J. P., Campbell, D. N., J. Test Eval. W, No. 3, (1985) 234-240.
[87] Hadeishi, T., Pollard, M., McGlaughin, R., Koga, M., NTIS Pub. No. PB85-225027, 1985, Na-
tional Technical Informaton Service, Washington, D. C.
I881 Smith, K., Environment 30, (1988) 10-35.
[89] Budiansky, S., Environ. Sci. Tech. 14, (1980) 1023-1026.
[90] Nero, A., Scientific Amer., 258, (1988) 42-48.
[91] Spengler, J. D., Soczek, M., Environ. Sci. Tech. Is, (1984) 268A-280A.
[92] Rowe, M. J., Transport. Engin. 111, (1985) 679-691.
[93] Akland, G. G., Hartwell, T. D., Johnson, T. R., Whitmore, R. W., Environ. Sci. Tech. 19, (1985)
911-918.
[94] Ware, J. H., Dockery, D. W., Spiro 111, A, Speizer, F. E., Ferris, B. G., Am. Rev. Respir. Dis. 129,
(1984) 366-374.
(951 Wallace, L. A., Total Exposure AssessmentMethodology (TEAM)Study: Summary and Analysis,
Vol. 1, 1987, USEPA, Washington, D.C.
[96] Ember, L. R., Chem. Eng. News 66, No. 49, (1988) 23-25.
[97] Cook, W., Occupational Exposure Limits - Worldwide, 1987, American Industrial Hygiene
Association, Akron, OH.
[98] Leidel, N., Busch, K., Lynch, J., Occupational Exposure Sampling Strategy Manual, Cincinnati:
USDHEW(NIOSH), 1977, App. I.
20.8 References 553

1991 Eller, P., NZOSH Manual of Analytical Methods Vols. 1 and 2, 1984, DHHS/CDC/NIOSH, Cin-
cinnati.
[loo] Harper, M., Purnell, C. L., Am. Ind. Hyg. Assoc. J , &No. 3, (1987) 214-218.
11011 Rose, V. E., Perkins, J. L., Am. Znd. Hyg. Assoc. J , 43, (1982) 605-621.
11021 Nader, J. S., Lauderdale, J. F., McCammon, C. S. in: Air Sampling Instruments for Evaluation of
AtmosphericContaminants, Lioy, P. J., Lioy, M. J. Y. (eds.); American Conference of Governmen-
tal Industrial Hygienists, Cincinnati, 1983 Ch. V.
11031 Firth, J. G., in: Detection and Measurement of Hazardous Gases; Cullis, C. F., Firth, J. G. (eds.);
London: Heinemann, 1981; pp. 29-68.
11041 Langhorst, M. L., Coyne, L., Anal. Chem. 59, (1987) 1R-17R.
11051 Air Sampling Procedures Committee, Ann. Am. Con$ Govt. Znd. Hyg. 11, (1984) 23-102.
11061 TLVs - Threshold Limit Valuesand Biological Exposure Indicesfor 1988-1989, American Con-
ference of Governmental Industrial Hygienists, Cincinnati, 1988, pp. 45.
11071 Baron, P. A,, in: Advances in Air Sampling, Chelsea, MI: Lewis Publishers, 1988, pp. 205-223.
11081 Smith, J. P., Baron, P. A., Murdock, D. J., Am. Znd. Hyg. Assoc. J 48 (1987) 219-229.
[I091 Iles, P. J., Shenton-Taylor, T., Ann. Occ. Hyg., 77, (1986) 111-114.
[110] Rose-Pehrsson, S. L., Grate, J. W., Ballantine, Jr., D. S., Jurs, P. C., Anal. Chem. 60, (1988)
2801-2811.
[111] Stetter, J. R., Jurs, P. C., Rose, S. L., Anal. Chem., 58, (1986) 860-866.
11121 Bernard, A., Lauwerys, R., in: Biological Monitoring of Exposure to Chemicals - Organic Com-
pounds, Ho, M. H., Dillon, H. K. (eds.); New York: Wiley, 1987, Ch. 1.
11131 Cooper, W. C., in: Patty’s Industrial Hygiene and Toxicology Vol. 3A, Cralley, L. J., Cralley, L.
V. (eds.); New York: Wiley, 1985, Ch. 5.
11141 Fisherova-Bergerova, V., Appl. Znd. Hyg., 2, No. 2, (1987) 87-92.
11151 TLVs - Threshold Limit Values and Biological Exposure Indices for 1988-1989, American Con-
ference of Governmental Industrial Hygienists, Cincinnati, 1988, pp. 50-62.
[I 161 MaximumConcentrations at the Workplaceand Biological Tolerance Valuesfor WorkingMaterials,
Report No. 31, Commission for the Investigation of Health Hazards of Chemical Compounds in
the Workplace; Weinheim, FRG: VCH, 1986.
[117] Pereras, F. P., Health and Environment Digest 2, No. 9 (1988) 1-8.
1118) Droz, P. O., Krebs, Y., Nicole, C., Guillemin, M., Am. Znd. Hyg. Assoc. J. 49, No. 7 (1988) 319-324.
11191 Droz, P. O., Guillemin, M., J. Occ. Med. 28, No. 8, (1986) 593-602.
[I201 Fisherova-Bergerova, V., in: Biological Monitoring of Exposure to Chemicals - Organic Com-
pounds, Ho, M. H., Dillon, H. K. (eds.); New York: Wiley, 1987, Ch. 3.
1121) Guillemin, M. P., Berode, M., Am. Znd. Hyg. Assoc. 49, No. 10, (1988) 497-505.
11221 Waldron, H. A., Cherry, N., Johnston, J. D., Znt. Arch. Occup. Environ. Health 51, (1985) 365-369.
11231 Hemminki, K., Vainio, H., in: Occupational and Environmental Chemical Hazards - Cellular and
Biochemical Indices for Monitoring Toxicity, Foa, V., Emmett, E. A., Maroni, M., Colombi, A.
(eds.); New York: Ellis Horword, 1987, Ch. 46.
11241 Calleman, C. J., Ehrenberg L., Jansson, B., Osterman-Gulkar, S., Segerback, D., Svensson, K.,
Wachtmeister, C. A., J. Environ. Pathol. Toxicol. 2, (1988) 427-442.
11251 Dillon, H. K., Ho, M. H., in Biological Monitoring of Exposure to Chemicals - Organic Com-
pounds, Ho, M. H., Dillon, H. K., (eds.); New York: Wiley, 1987, pp. 227-288.
11261 Hesley, K. L., Wimbish, G., Am. Znd. Hyg. Assoc. J 42, (1981) 42-46.
[127] Berlin, A., Draper, M., Hemmiki, K., Vainio, H., IARC Scientific Publications No. 59, Lyon
France.
[128] Franco, G., Fonte, R., Tempini, G., Candura, F., in: Occupational and Environmental Chemical
Hazards - Cellular and Biochemical Indices for Monitoring Toxicity, Foa, V., Emmett, E. A.,
Maroni, M., Colombi, A. (eds.); New York: Ellis Horward, 1987, Ch. 17.
11291 Palmer, T. N., Mansfield, D. A., Nature 310, (1984) 483-485.
554 20 Environmental Monitoring

[130] Earth System Science Overview, 1986. National Aeronautics and Space Administration, Washing-
ton, D. C.
[131] Jager, J., Environment 30, No. 7, (1988) 12-15.
[132] Earth SystemScience, A Closer View,1988. National Aeronautics and Space Administration, Wash-
ington, D. C.
[133] 1987 IEEE International Geoscience & Remote Sensing Symposium, Ann Arbor, MI, May 1987.
Digest, Piscataway, NJ: IEEE, 1987.
(1341 1987 IEEE International Geoscience & Remote Sensing Symposium, Ann Arbor, MI, May 1987,
Digest, Piscataway, NJ: IEEE, 1987, pp. 141- 192, 237-262.
[135] Curran, R. J., 1987 IEEE International Geoscience & Remote Sensing Symposium, Ann Arbor, MI,
May 1987, Digest, Piscataway, NJ: IEEE, 1987, pp. 379-380.
(1361 Fitzjarrald, D., 1987 IEEE International Geoscience & Remote Sensing Symposium,Ann Arbor,
MI, May 1987, Digest, Piscataway, N k IEEE, 1987, pp. 599-600.
[137] NASAEarth Science and ApplicationsDivision: The Program and Plans for FY 1987-1988-1989;
1987, National Aeronautics and Space Administration, Washington, p. 87.
[138] 1987 IEEE International Geoscience & Remote Sensing Symposium, Ann Arbor, MI, May 1987.
Digest, Piscataway, NJ: IEEE, 1987, Session MP-4, pp. 89-146.
[I391 Schanda, E., IEEE Antennas and Propagation Society International Symposium, Seattle, WA,
1979, Digest, Piscataway, NJ: IEEE, 1979, p. 762.
[I401 McCreary, M. D., KAF-4200 Specification, 1989, Eastman Kodak Company, Rochester, NY, USA.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

21 Medical Diagnostics and Patient Monitoring


R . NEUMAN.
MICHAEL MetroHealth Medical Center. Cleveland. OH. USA

Contents
21.1 Medical Sensor Requirements . . . . . . . . . . . . . . . . . . 556
21.2 Special Considerations for Sensors in Medical Application . . . . . . . 558
21.2.1 Size and Shape . . . . . . . . . . . . . . . . . . . . . . . . . . 558
21.2.2 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
21.2.3 Packaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
21.2.4 Sterilization . . . . . . . . . . . . . . . . . . . . . . . . . . . 559
21.2.5 Reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 559
21.2.6 Lead Wires and Connectors . . . . . . . . . . . . . . . . . . . . 560
21.2.7 Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560
21.3 Medical Applications . . . . . . . . . . . . . . . . . . . . . . . 560
21.3.1 Biomedical Research . . . . . . . . . . . . . . . . . . . . . . . 560
21.3.2 The Clinical Laboratory . . . . . . . . . . . . . . . . . . . . . . 560
21.3.3 Clinical Monitoring . . . . . . . . . . . . . . . . . . . . . . . . 562
21.3.4 Implantable Closed Loop Control System . . . . . . . . . . . . . . 563
21.3.5 Applications in the Home . . . . . . . . . . . . . . . . . . . . . 564
21.4 Examples of Sensors in Medical Applications . . . . . . . . . . . . 564
21.4.1 Geometric Variables . . . . . . . . . . . . . . . . . . . . . . . 564
21.4.2 Kinematic Variables . . . . . . . . . . . . . . . . . . . . . . . . 565
21.4.3 Radiation and Sound . . . . . . . . . . . . . . . . . . . . . . . 567
21.4.4 Force and Tactile Sense . . . . . . . . . . . . . . . . . . . . . . 568
21.4.5 Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . 569
21.4.6 Hemodynamic Variables . . . . . . . . . . . . . . . . . . . . . . 569
21.4.7 Electromagnetic Variables . . . . . . . . . . . . . . . . . . . . . 571
21.4.8 Chemical Sensors . . . . . . . . . . . . . . . . . . . . . . . . . 572
21.4.8.1 Electrochemical Sensors . . . . . . . . . . . . . . . . . . . . . . 572
21.4.8.2 Bioanalytical Sensors . . . . . . . . . . . . . . . . . . . . . . . 574
21.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575
21.6 References . . . :. . . . . . . . . . . . . . . . . . . . . . . . 576
556 21 Medical Diagnostics and Patient Monitoring

Sensors play an important role in a medical instrument since they serve as the interface
between the biological system and the hardware of the instrument. A sensor must be able to
transmit physiological information across this interface without adversely affecting the condi-
tions existing on either side, and in order to be suitable for application in most medical in-
strumentation systems it must be able to do this in a reproducible fashion over time and for
a range of different organisms. Thus, there must be a number of stringent requirements layed
down for sensors that are to be used in clinical medicine or in basic medical science research.

21.1 Medical Sensor Requirements

The concept of a sensor as an interface is so important, one should begin by considering


the various forms that this interface can take. These are summarized in Table 21-1 where the
left hand column indicates the degree of biological invasion in relation to the actual location
of the sensor about the subject’s body, which is shown in the right hand column. At one ex-
treme there is the non-contacting, non-invasive sensor that picks up biological information
from the host without physically touching them. Such sensors and their associated instrumen-
tation are useful for the rapid assessment of large numbers of subjects as is often the require-
ment in public health screening programs. For example, a radiation temperature sensor can
be used to assess the individual surface temperatures of a group of patients as they walk past
the instrument one by one.
When considering medical sensor types at the other extreme there are sensors that are con-
sidered invasive; these are placed in surgically created cavities in tissues that would not nor-
mally have such structures. Continuing with the use of temperature measurement as an exam-
ple, an invasive measurement might involve determining the temperature of a skeletal muscle
by percutaneously injecting into the muscle, a needle containing a miniature thermal sensor
at its tip. The cutting edge of the needle creates the cavity in which the temperature is
measured.
Minimally invasive and non-invasive sensors fall between these two extremes. In the former
case, the sensor is placed within an existing body cavity in a way so as not to significantly
distort the anatomy of this cavity. An example of minimally invasive temperature measure-
ment is the determination of core body temperature using a temperature sensor on the tip of
a probe inserted into the rectum or a temperature sensor that is placed in the ear and measures
the tympanic membrane (ear drum) temperature.
Non-invasive sensors do not enter the body at all but can be placed on its surface. Again,
using temperature measurement as an example, probes that are placed on the skin to monitor
body temperature can be included in this category.
The nature of the interface between sensor and body as described in Table 21-1, is closely
related to one of the principal requirements for a biomedical sensor, namely the ability to ef-
ficaciously measure the variable being sensed whilst having a minimal effect on the biological
system being measured. Table 21-2 lists this and other requirements specifically intended for
sensors used in biomedical applications. Sensors used in clinical medicine must be manufac-
tured from materials that are non-toxic in the systems for which they are intended. This is most
crucial for implantable sensors since toxic substances can leach out of the sensor and its
packaging, and these can not only affect the local tissue surrounding the sensor, but may have
21.1 Medical Sensor Requirements 557

systemic effects as well. This is an important consideration even for non-invasive sensors on
the skin surface since toxins can leach out of these also and be transported through the skin
into the body.

Table 21-1. Classification of biomedical sensors according to their interface with the body.
~ ~~~

Classification Location

Invasive Surgical implant


Minimally invasive Indwelling
Non-invasive Skin surface
Non-contacting Remote from body

Table 21-2. Requirements of biomedical sensors.

1. Classical general sensor requirements


2. Minimal effect on biological system
3. Non-toxic
4. Mechanically matched to biological system
5. Stable in an aqueous environment
6. Capable of withstanding sterilization

The sensor and its packaging should be mechanically matched to the tissue which it is in
contact with, regardless of whether it is an invasive or a non-invasive sensor. Mechanical
mismatch can lead to irritation and inflammation which besides being uncomfortable for the
patients, can locally change the variables being sensed. A common example of this problem
is found with the attachment of biopotential electrodes to the skin surface. This non-invasive
sensor is usually held in place using a tape material possessing a strong adhesive tack. Unfor-
tunately, although the tape is flexible, as is the skin, it has a much lower compliance than skin.
Thus as the skin stretches, the tape cannot which leads to skin irritation at the edge of the
tape. Frequently when an electrode is removed, one sees a red outline from the edges of the
tape on the skin due to this irritation resulting from a poor mechanical match.
Biomedical sensors that are implanted or indwelling have additional requirements. Firstly,
they must be stable in an aqueous environment. Not only must they withstand water, but the
body's internal environment includes corrosive ions such as chloride and various enzymes and
cells that can attack polymers such as used for sensor packaging as well. Sensors placed in
the blood stream not only have to contend with the above problems, but they also must not
promote the clotting mechanisms of blood which could result in the deposition of fibrin and
other proteins on the sensor's surface. Secondly, it is generally impossible to calibrate im-
planted sensors since access is limited, and whilst indwelling sensors make the problem a little
easier, they still require either complete removal or coupling of the cavity to an additional
calibrated sensor for comparison. These sensors must also be sterile when they are placed in
tissue or body cavities. Thus the sensors must be constructed from materials that can be
sterilized, although the sensors themselves must not be affected by the sterilization process.
558 21 Medical Diagnostics and Patient Monitoring

21.2 Special Considerations for Sensors in Medical Application

In any instrumentation system, it is necessary to use sensors that are appropriate for the
particular measurement required. Some of the special considerations that are necessary for
optimal interface between a sensor and the system being measured are quite critical when con-
sidering biomedical applications. Since sensors can have a strong effect on the biological
system being measured and the biological system itself can influence the behavior of the sen-
sor, these special considerations can be especially critical in the use of invasive sensors.

21.2.1 Size and Shape

When used in biomedical applications, the physical dimensions of a sensor can assume a
greater degree of importance. Obviously, indwelling sensors must be of a size and shape that
allows them to be introduced into, and then to remain in, the cavity being measured without
interfering with the normal biological function of that cavity. Implantable sensors must also
not appreciably distort the tissues in which they are located. In general, this means that
biomedical sensors need to be small, although there are other factors and exceptions to this
rule that should be considered as well. Sharp corners on a sensor can create high local stress
in a tissue which can then induce a response that interferes with the sensor operation. Surface
texture of the sensor and its packaging can also affect the biological interaction of an im-
planted or indwelling biomedical sensor with its tissue, and this can not only affect the sen-
sor’s performance but also whether it can remain in place for longer periods of time.

21.2.2 Materials

If at all possible, sensors should be constructed from materials that are compatible with
biological systems. Toxic materials, even though they may not be in direct contact with a tissue,
can leach out and cause a local or systemic toxic reaction. Thus, in situations when it is
unavoidable but to use toxic materials in the construction of a biomedical sensor, it is crucial
that these materials be packaged in such a way that they will not be introduced into the living
system.

21.2.3 Packaging

One of the major problems in developing biomedical sensors is finding appropriate


materials to package and protect the actual transducer. The package must allow the sensor
to make contact with the organism as far as the measurement of the quantity being sensed
is concerned, but at the same time it must protect other parts of the sensor from the hostile
environments such as those that are found in implantable and indwelling sensor applications.
The packaging must do this whilst remaining relatively small so that an appreciable portion
of the sensor is not taken up by packaging. It must also not significantly contribute to
degradation of sensor characteristics. A typical example of this problem might be a protective
21.2 Special Considerations for Sensors in Medical Application 559

polymeric
. ~.
package over a pressure sensor used for implantable or indwelling applications. If
the packaging takes up water from the aqueous environment in the host biological system as
most polymeric materials do, this may result in swelling of the materials which can generate
a back force against the sensitive membrane of the pressure sensor. This will cause the pressure
sensor to drift even though there is no direct change in the pressure being measured in the
cavity.

21.2.4 Sterilization

There is a strong tendency for infection to occur at the site of an implanted foreign material
such as a sensor [l]. It is thus very important to ensure that implantable sensors and indwelling
sensors in most applications are sterile before they are placed. If there is any chance of defects
occuring in the sensor package, the contents of that package must also be sterile. Various
sterilization procedures can be used to prepare a sensor for implanted or indwelling
applicators. Devices made of materials unaffected by elevated temperatures may be sterilized
by autoclaving. This technique is simple and is used in most hospitals and clinics for general
sterilization. Since many sensors are temperature sensitive, sterilization at room temperature
is preferred. This can be achieved by soaking in a solution such as glutaraldehyde and then
rinsing with sterile water. This is often used for indwelling sensors. Implanted sensors are
frequently sterilized in ethylene oxide gas. Ethylene oxide is a toxic gas that can pollute the
environment and present a health hazard. Stringent regulations have, therefore, been
established to control the use of this substance. Sensors must be degassed after exposure to
this toxic agent before they can be placed in the body, since trace quantities of the gas can
cause severe tissue reactions. This results in a much longer time required for the sterilization
process before the sensor can be used. Once infection occurs at the site of an implantable
sensor, there is little that can be done to fight it. The sensor usually has to be removed as a
part of the treatment. Thus, it is important to make every precaution in sensor preparation
and use aseptic technique in implantation to avoid this problem.

21.2.5 Reliability

Long-term reliability is a crucial consideration in the development of implantable sensors.


It is not easy to replace sensors that malfunction, so it is necessary to keep the need for such
changes to an absolute minimum. Implantable biomedical sensors must be carefully manufac-
tured using the best manufacturing techniques, and devices must be tested throughout their
production in order to ensure and also document their safety and reliability. Even so, conven-
tional sensors can fail, and this can have serious medical consequences. For this reason, new
approaches are being developed to improve the overall reliability of such sensing systems.
Multiple, redundant sensors with on-sensor electronics to eliminate individual elements which
fail are under development. Another important research area in sensor development, is the use
of special on-sensor signal processing and testing procedures to identify sensor elements that
have failed or are at risk of failing so that the contribution from these can then be eliminated
from the output signal.
560 21 Medical Diagnostics and Patient Monitoring

21.2.6 Lead Wires and Connectors

A recurrent problem with both invasive and non-invasive biomedical sensors stems from the
lead wires and connectors used to link the sensor to the remainder of the instrumentation
system. Frequently, failures in such sensors involve wire breakage either at the sensor or at the
connector. Although highly sophisticated sensing systems have been developed, adequate lead
wire designs with strain relief at points where these wires attach to the sensor or to the connec-
tor still need to be perfected. This must be achieved whilst still retaining all of the desired
characteristics of the lead wire systems including small size, high flexibility, axial compliance
and good electrical insulation.

21.2.7 Costs

An important consideration in any medical system today is the cost associated with its ap-
plication. If biomedical sensors are to be more widely applied, their costs need to be con-
tained. The use of mass production techniques similar to those employed in the microelec-
tronics industry and the construction of sensors that require a minimum amount of individual
calibration can greatly help to reduce costs. By increasing the number of applications for sen-
sors produced using these technologies, per unit costs will drop just as they have done
previously for other widely used semi-conductor devices.

21.3 Medical Applications

Applications for biomedical sensors can be found in research laboratories, in clinical


laboratories, in clinical care and as components of special therapeutic devices. Each of these
applications puts unique demands on the sensors involved. In some cases these special re-
quirements can be met by existing sensors whilst the sensors for others have yet to be op-
timized and so provide areas for current and future research.

21.3.1 Biomedical Research

Medical electronic sensing systems have an important role in medical research. Sensors of
physical and chemical variables that can be used for acute and chronic experimentation are
commonly employed in medical research 1aboratories.Here implanted or indwelling sensors
are frequently used to make measurements in experimental animal models that could not be
routinely performed in clinical medicine. The development of new sensors that will allow in-
novative medical research is an important research area in itself.

21.3.2 The Clinical Laboratory

The clinical laboratory is concerned with taking measurements from biological specimens
such as blood, urine or cerebro-spinal fluid. Measurements involve assessing both chemical
21.3 Medical Applications 561

quantities such as the amount of glucose in the blood and physical quantities such as the
specific gravity of urine or the portion of a volume of blood taken up by cells. The instruments
used to measure these quantities involve the use of chemical and physical sensors, and fre-
quently they also employ a sensing chamber or cuvette into which the sample is injected. In
most of these instruments the measurement is made, and then the chamber is flushed out to
prepare it for the next measurement. Automatic analyzers have been developed which allow
this process to be carried out for a large number of specimens with minimal technician sup-
port. Thus, the clinical laboratory makes many types of measurements using sensors and in-
strumentation and the major classifications of these measurements are listed in Table 21-3.
This list is meant to cover general areas with a few examples given for each area, but it is by
no means complete. Usually, different laboratories specialize in different clinical assays, and
new assays and devices are always appearing on the market.

Table 21-3. Measurements made by the clinical laboratory.


~~ ~ ~

General Classification Examples

1. Chemical determinations in blood and other 1. Serum electrolytes


biological specimens Serum glucose
Urine creatinine
2. Immunological determinations 2. Serum estrogen
3. Microbiological determinations 3. Bacterial culture and sensitivity to drugs
Viral cultures
4. Histological analyses 4. Tumor identification
Cytology
5 . Analysis of cellular components of blood 5. Complete blood count
White cell count
6 . Blood gas and acid-base balance analysis 6. Arterial blood oxygenation, tension and
hemoglobin saturation
Whole blood pH
7. Special studies including:
a. Trace element analysis I a. Serum zinc
Serum cadmium
b. Determination of drug levels b. Drug metabolites in urine
Metabolites from drugs of abuse
c. Genetic analysis c. Chromosome analysis

A novel approach gaining popularity in the clinical laboratory involves the use of disposable
sensors designed for just a single measurement. These single shot sensors take the form of a
disposable cartridge that not only serves as a sample chamber but contains the sensor as well
and frequently these devices also include a means of self calibration. An example of such a
system is the fetal pH microblood analyzer developed by Enzer and Diamond [2]. This instru-
ment is built around a small disposable cartridge of the type illustrated in Figure 21-1. The
sensing cartridge consists of a glass pH sensor and a silver chloride coated silver wire to serve
as the reference electrode. A glass ampule filled with a buffer system is used to calibrate the
sensor in the following way. When the cartridge is snapped into the electronic system, the am-
562 21 Medical Diagnostics and Patient Monitoring

pule is broken and the buffer solution runs into the channel between the pH and reference
electrodes. The potential of this electrochemical cell is then measured by the electronic in-
strumentation which automatically performs a calibration at the approximate blood pH value
likely to be encountered. Once this microprocessor controlled automatic calibration is com-
plete, the instrument is ready to accept a microblood sample which can be as small as twenty
microliters. This sample displaces the buffer solution around the pH electrode, and the buffer
now serves as a salt bridge to the reference electrode.

BUFFER

REFERENCE
ELECTRODE

ELECTRICAL ’
CONTACTS

Figure 21-1. Sensor cartridge used by Enzer and Diamond’s fetal microblood pH analyzer [2].

The principal advantage of this system is its portability. Namely that through the use of
relatively inexpensive disposable sensors and microprocessor controlled automated electronic
instrumentation, it is possible to make reliable measurements of blood pH at the site where
the blood sample is obtained thereby getting the information back to the physician as quickly
as possible. In addition, the use of disposable cartridges and automated electronics makes it
possible for reliable measurements to be made by clinical staff who are not trained as clinical
laboratory technicians. By using the sensor cartridge only once and then discarding it, the pro-
blem of contamination with proteins and other materials from one measurement to the next
is eradicated. The self calibration routine minimizes the need to calibrate the system against
a standard. Thus, in the case of this example, a measurement that normally would be done
in the clinical laboratory can now be done in the patient area for more rapid and effective feed-
back of information to the clinical staff managing the patient’s care.

21.3.3 Clinical Monitoring

Clinical care is becoming more and more complex, and the development of new types of
instrumentation and therapeutic devices is helping to provide higher quality patient care all
the time. Instrumentation that continuously monitors a number of physiological variables on
critically ill patients has been developed and is routinely applied both in the operating room
21.3 Medical Applications 563

and in intensive care units. Such devices enable clinicians to continuously monitor physical
variables such as blood pressure [3], temperature [4], electrical potentials [ 5 ] , respiration [6]
and movement [7]. Devices for monitoring chemical variables such as pH [8], oxygen and car-
bon dioxide tensions, partial pressure [9, 101, and various gases in the airway are also available
[ll]. A critical component of each of these instruments is the sensor. Because this monitoring
is often carried out over relatively long periods of time lasting from a few hours to a few weeks,
sensors must be stable and capable of continuous operation over such periods of time.As this
instrumentation moves from critical care areas of the hospital to less skilled facilities or even
the home, the instrumentation must be more highly automated and capable of being operated
by individuals with less training. In the following section, some examples of sensors that can
be used for patient monitoring will be examined more closely.

21.3.4 Implantable Closed Loop Control System

One of the recent advances in biomedical electronics is the development of closed loop con-
trol systems to replace the function of diseased or damaged organs [12]. In many cases, these
systems are totally implantable and therefore require the highest degree of reliability. Sensors
are an important part of such systems. Not only must they be reliable, but they must be stable
over long periods of time since the system generally will not be accessible for recalibration.
An example of such a closed loop control system is the so-called “artificial pancreas” [13].
A block diagram of this system is illustrated in Figure 21-2. Its fundamental operating princi-
ple is quite straightforward. The artificial pancreas carries out the function of the pancreatic
beta cells, that is it senses the blood glucose level in the patient, and when this becomes too
high, a small infusion pump injects insulin into the circulation to bring glucose levels back
down. The technology for such an implantable system is already well developed with the ex-
ception of a reliable sensor of blood glucose concentrations. Open loop insulin infusion
pumps that are preprogrammed or can be programmed by the patient are currently available
and widely used by diabetics [14], and whilst several types of sensors are under development
[15, 16, 171, none have proven to be reliable enough for closed loop control to be incorporated
into devices that could be routinely used in the treatment of diabetes. Thus, here is a problem
whose solution, the development of an appropriate implantable sensor, could potentially have
a huge beneficial impact on the treatment of a major disease state.

Figure 21-2. CIRCULATING GLUCOSE


Simplified block diagram of an artificial pancreas. BLOOD SENSOR
The glucose sensor measures the glucose concen-

insulin pump. The controller can have a fixed set


point about which it regulates the serum glucose I NSUL I N - CONTROLLER
concentration or a set point that is adjustable PUMP
according to factors such as the time of day, the
activity of the patient or when meals are taken.
564 21 Medical Diagnostics and Patient Monitoring

21.3.5 Applications in the Home


Medical instrumentation has recently been seen to be moving outside of the hospital and
clinic and into the home in an effort to help keep health care costs down whilst providing a
better level of care. Sensors are an important part of the new technology that is allowing this
to happen. Home monitoring devices at the present time are limited to instruments for
monitoring clinical variables such as blood pressure, heart rate and respiration and there are
also a number of home devices for assessing some blood and urine chemistries. The most
notable of these are the home glucose monitors where a patient is trained to take a small
droplet of peripheral blood, place it on a reagent strip and analyze the colorometric indicators
on this strip using a photometric instrument [HI. Devices such as this have allowed diabetic
patients to more accurately control their serum glucose levels through diet and the administra-
tion of insulin. The development of low cost, reliable optical sensing systems helped to make
this instrument practical.

21.4 Examples of Sensors in Medical Applications

The biomedical applications of sensors are both numerous and diverse, and it would be dif-
ficult to cover all of the possible applications in this chapter. Instead, it is more helpful to
consider a few examples of medical applications of sensors in order to illustrate some of the
major types of sensors in use today.

21.4.1 Geometric Variables

Sensors of linear and angular displacement, surface area and volume fall into this general
category. These sensors and their associated instrumentation systems are involved in the
measurement of the size and positions of anatomical structures. Quantities such as organ
dimensions and the dynamic variations of these dimensions, joint position, growth, cross-sec-
tional area of vessels, seating contact area, volume capacity of various internal organs such
as the urinary bladder, and tumor size assessment are examples of the kinds of measurement
assessed by this type of sensor.
A joint position sensor under development at Case Western Reserve University can be used
to illustrate an angular displacement transducer based upon thin film strain gauge technology
[19]. The sensor consists of a thin gold or palladium film deposited on a flexible 75 pm thick
polyimide substrate as illustrated in Figure 21-3 a. When this sensor is flexed around an angle
as shown in Figure 21-3 b, it can be demonstrated using beam theory that if the substrate is
thin and the strain gauge is on the convex surface of the substrate, the change in resistance
of the film will be given by
AR = kyR,O (21-1)

where R, is the unflexed resistance of the strain gauge,


0 is the angle through which the sensor is flexed in degrees,
y is the gauge factor of the strain gauge,
K is a constant depending on the length and thickness of the device.
21.4 Examples of Sensors in Medical Applications 565

NEUTRAL
AXIS \

Figure 21-3. (a) The layout of a thin-film joint angle sensor [19]; (b) side view of the joint angle sensor
with definitions of the geometrical terms used in Equation (21-1).

It is seen that the change in resistance of the strain gauge is dependent only upon angle for
a particular structure and material, and since radius of curvature does not enter into Equation
(21-1), the shape that the sensor adopts between its ends should not affect the measurement
of an angle.
This sensor is used to measure the angle of the proximal interphalangeal joint of the index
or middle finger to provide proprioceptive feedback to a patient with a high level spinal cord
injury. It can also be used to provide a feedback signal to the controller of a functional elec-
trical stimulation system used to help this type of patient to regain some grasping function.
The sensor is attached to the finger by a strap which fixes it to the finger segment between
the proximal interphalangeal joint and the interphalangeal-metacarpaljoint of the finger.
The free end of the sensor slides through a strap attached to the segment between the two
interphalangeal joints as the finger flexes. An example of a calibration curve for such a sensor
is shown in Figure 21-4.

21.4.2 Kinematic Variables

Velocity and acceleration sensors have also found applications in the field of biomedical
measurement although these sensing systems are not as widely used in medicine as they are
in industry. When studying the cardiovascular system it is desirable to be able to measure
566 21 Medial Diagnostics and Patient Monitoring

0.7555
7
0.755

?z
L
0.7545

w
a
5
0.754

$
t
3
0-7535
0.753
0
re
Q5
0

0*7525
cn
0.752

0.751 5

0.751
0 15.5 30 45 60 72 90

JOINT ANGLE (Degrees)


curve (1) + curve (2)

Fpgnre 214. 'ILpical calibration curve for the joint angle sensor shown in Figure 21-3 when attached to
the middle finger of a human subject.

blood flow in major vessels and capillaries, but unfortunately this usually is not possible
especially when using non-invasive instrumentation. Recently however, ultrasonic sensors and
instrumentation have been utilized to measure the velocity of flowing blood in a vessel using
the Doppler effect. This is aceomplished by reflecting ultrasound off the cellular components
of the blood. One can then measure the Doppler shift in frequency of the reflected ultrasound
which can be related back to the velocity if the angle between the ultrasonic beam and the
flowing blood is known.Both invasive sensors that attach to the blood vessel and non-invasive
Doppler flow measurement systems have been developed 1201.
Kinematic sensors are also useful in studying limb movement and in analyzing the gait of
a patient. Miniature accelerometers are primarily used for this purpose and can be attached
to various points on a limb to measure accelerations of the limb under normal movement. This
acceleration information can then be integrated to obtain velocity and displacement data, and
through creative signal processing the movements can be recreated on a computer screen and
the information used to analyze the patient's motion [21].
Miniature accelerometers have also been used to non-invasively measure the apical impulse
of the heart in a modem-day version of the ballistocardiograph. Such measurements assist in
the diagnosis of heart disease and are important in that they use relatively inexpensive non-
21.4 Examples of Sensors in Medical Applications 561

invasive procedures to obtain information that would otherwise have to come from invasive
studies that put the patient at increased risk [22].

21.4.3 Radiation and Sound


Sensors of visible or invisible light, nuclear radiation, sound and ultrasound have found
many different applications in medicine. Sensors that measure light intensity can be used to
determine for individuals sensitive to it, the amount of exposure to light radiation such as
ultraviolet light. In the care of newborn infants, it is sometimes necessary to remove bilirubin,
a degradation product of hemoglobin, which collects as the result of the breakdown of blood
cells. One way this material can be removed is to convert it to a form that can be naturally
excreted by the body. This conversion process can be promoted by placing the patient under
fluorescent lights [23]. This phototherapy technique is widely used in nurseries, but it is im-
perative to monitor the total dose of radiation that the patients receive since too much of it
can lead to complications. Hence, photometric instruments based upon light sensors have
been developed to meet this need [24].
Sensors to measure infrared radiation from the body have been developed to allow rapid
measurements of body temperature without actually physically touching the patient. Infrared
sensors that can scan the body surface can be used with other instruments to provide a thermal
image of the body. This technique, known as thermography, is often useful in detecting
peripheral circulation abnormalities or tumors located near the surface of the skin [25].
Sensors of X-rays are particularly important in various medical imaging instruments based
upon ionizing radiation [26]. Although X-ray equipment has traditionally employed photo-
graphic film as the X-ray detector, this has changed in recent years. Computerized
tomography, digital subtraction radiography, low intensity fluoroscopy, etc, all require elec-
tronic sensors of radiation so that electrical signals to be analyzed by specialized computers
can be used. In the future, such sensors will be even more important as radiology moves away
from the use of X-ray film and into the age of electronics and computers.
Nuclear radiation sensors are important in imaging the location of radioactive tracers in-
troduced into patients. Nuclear cameras are complex instruments based upon these sensors
and are useful in detecting abnormalities in tissues, such as tumors. Nuclear sensors are also
important for dosimetry devices that monitor the exposure of an individual to ionizing radia-
tion, and they are also used in carrying out tracer studies where biochemical reactions are in-
vestigated.
One application of ultrasonic sensors for measuring the velocity of flowing blood has
already been discussed above. Ultrasound is also important for imaging soft tissue and for
detecting the heart beat of the fetus. There are several ultrasonic instruments that measure the
function of the heart. Ultrasonic sensors are important in all of these applications, and the
sensitivity and resolution of all these instruments has been improved in recent years as a result
of new ultrasonic sensor developments.
The measurement of audible sound is also used in medicine to electronically detect what
physicians listen for with a stethoscope. Not only can normal and abnormal heart sounds be
detected and documented with these sensors [27], but breathing sounds, sounds emanating
from the gastrointestinal system and sounds generated by turbulence in the peripheral vascular
system. Such measurements are often very useful both in diagnosing disease and in the
monitoring of its progress.
568 21 Medical Diagnosrics and Patient Monitoring

21.4.4 Force and ’kctile Sense

When studying biomechanics, the clinician and researcher are concerned about measuring
forces applied to various parts of the body and seeing how the body responds to these forces.
Thus it is important to have sensors that can measure the applied force as well as a host of
variables that are affected by this force. Force sensors frequently take the form of a load cell
and can either be used for laboratory measurements on specimens or for direct measurements
on human subjects or animal models. In the former case+ this includes various procedures for
fiding the material properties of tissues such as bone, tendon, muscle, skin, etc Whereas the
l a m case involves measuring forces associated with movement activities such as walking,
sports activities or obstetrical labor.
A commonly applied form of a force sensor is a patient weighing instrument. In critical
care situations where a patient will be confined to his bed, it is necessary to detect changes
in a patient’s weight since these can reflect excessive storage or loss of water in the patient.
A straightforward way to make such a measurement is to place a load cell under each leg of
the patient’s bed and sum the forces measured by each of the load cells [28]. As long as addi-
tional objects are not placed on the bed, this technique can give the weight of the patient and
the bed. One can then look at the changes in this weight during procedures such as in-
trapcritoncal dialysis to determine changes in the patient’s total water content.
’Pactilesensors consist of arrays of individual force transducers. These sensors are of interest
in the field of robotics since they can give a robot gripper the ability to sense shape, position
and texture of an object. Recently the author and coworkers have undertaken the development
of a multi-element force sensor for use on the digit tips of patients with insensate hands secon-
dary to high level spinal cord injury [29].These sensors do not need as high a resolution as
do the sensors used on robot grippers, but they must be small and thin enough so that they
can be placed over the subject’s hand without appreciably distorting its anatomy or function.
Thus, thin film, multielement force sc~lsorsare under development for this purpose.
Figure 21-5 illustrates such a sensor which consists of an array of thin rectaagular gold elec-

Figure 2l-5. A 64-element capacitativc force sensor (20 x 20 mm)constructed using thin- and thick-fii
microelectronic technology. Reprinted from [29]. with permission.
21.4 Examples of Sensors in Medical Applications 569

trodes deposited upon a flexible polyimide substrate. Two of these substrates are used to make
up a sandwich with a thick film printed silicon elastomer as the filling. As is seen in
Figure 21-5, the electrodes on one substrate are orientated such that they are at right angles
to the electrodes on the other. The intersections of these electrodes form force sensitive
capacitors that are 2 mm square. As a force is applied to one of these elements, the elastomer
is compressed, and the capacitance therefore increases. Signal processing can determine the
capacitance of each element and hence the applied force needed to produce this capacitance.
This can then be used to generate a signal proportional to the total force on the sensor or a
display of the distribution of this force. Figure 21-6 illustrates such a pattern for a sensor that
has a metal washer pressed against it.

Figure 21-6.
Perspective plot of the output voltage
from the multi-element force sensor
array shown in Fig. 21-5 when a cir-
cular washer is pressed against its
surface.

21.4.5 Temperature

Temperature is one of the vital signs that is always measured in the care of hopsitalized pa-
tients and the assessment of others. Ideally, clinicians desire to measure core body temperature
rather than surface temperature, and so indwelling rather than non-invasive sensors are usually
used. These sensors are most often placed in the mouth or rectum. Rapid response time is an
important factor for sensors of temperature. A clinician who has to measure the temperature
of every patient on a ward can complete this task faster using rapid responding electronic ther-
mometers than he could using the traditional mercury in glass thermometers. Non-invasive
methods of measuring core temperature have been developed where the sensor determines the
temperature of the skin under conditions of zero heat flux from the body [30].Also, as men-
tioned previously, infrared sensors can be used as non-contacting rapid responding
temperature sensors.

21.4.6 Hemodynamic Variables

Measurements of blood pressure and flow are some of the most important in clinical
medicine and physiological research. Nevertheless, these measurements are among the most
570 21 Medical Diagnostics and Patient Monitoring

difficult to perform. Miniature pressure sensors manufactured by semiconductor technology


(see Chapter 5 ) have been placed on probes that can be introduced into the circulation to
measure central arterial, venous and intracardiac pressures. These sensors provide efficacious
measurements in acute preparations since their calibration and baseline can be routinely
checked between measurements. Using such sensors for chronic measurements, however, is
more difficult. Baseline drift caused by the sensor packaging makes chronic measurements of
low pressures such as those found in the venous circulation somewhat difficult. With the ex-
ception of some research devices, a high quality, low drift pressure sensor suitable for
chronically implanted applications has yet to be developed.
The non-invasive measurement of blood pressure is an important theme in clinical medicine
[31]. Blood pressure is a vital sign and should be checked every time a patient is examined and
indeed it is checked on a regular basis for hospitalized patients. Yet if the only way to do this
was by direct measurement of blood pressure by introducing a sensor into an artery it would
not be possible to use the technique on a very widespread basis. Therefore, non-invasive
methods need to be developed. Although there are several such methods, each has its own
limitation, however, many of these, if not 100% accurate, are able to provide reproducible
results on individual subjects. This makes reproducible measurements possible even though
there might be a systematic error. Thus, devices have been developed that allow patients to
monitor their own blood pressure at home and report the results to their physician.
Flow measurements in all systems of the body are difficult to perform either invasively or
non-invasively. The measurement of blood flow to vital organs is important in assessing the
adequacy of the circulation, and the determination of total cardiac output is an important way
of evaluating the condition of the heart. Several types of invasive sensors have been developed
in an attempt to measure blood flow, but each has its limitations [32]. Some of these sensors
depend upon the injection of an indicator into the flowing blood which is then detected at
a point downstream or as it is recirculated after making one complete circuit through the
periphery, heart and lungs. An alternative method utilizes an electromagnetic flow sensor that
can either be surgically placed around the vessel to be measured or introduced into the lumen
of that vessel [33]. This sensor establishes a magnetic field in a region through which blood
is flowing and measures electrical potential differences normal to the direction of the flow and
the direction of the magnetic field in order to determine flow velocity. The sensor itself when
placed around a vessel establishes a fixed cross-sectional area for the vessel so that volume
flow can also be determined. Ultrasonic Doppler methods can also be used to determine flow
as indicated previously.
Pressure sensors are not only useful for measurements in the cardiovascular system, they
can be used to determine pressures in other hollow viscera as well. Pressure sensors for
measurements in the gastrointestinal tract, urinary tract, pregnant uterus, eyes and central ner-
vous system have all been reported [33]. A common feature of all of these is their miniature
size and the necessity for stable characteristics.
A specific aspect of blood flow that is important in medicine is the determination of the
profusion of the capillaries of vital organs. At present there is no reliable sensor either implan-
table or non-invasive that can make such a measurement. Most studies today utilize an in-
dicator (usually a radioactive isotope of a gas such as Xenon) and measure the washout of
this indicator from an organ that has been previously saturated with it. Alternatively, thermal
profusion sensors have been developed in which heat serves as the indicator. This method ap-
pears to have the most promise at the present time, however, sensors optimized for this type
21.4 Examples of Sensors in Medical Applications 571

of measurement still need to be developed. Non-invasive measurement of profusion of some


organs such as the brain can be carried out using very complex instrumentation systems such
as magnetic resonance imaging [35] although this type of measurement is not practical at the
present time for routine clinical monitoring.

21.4.7 Electromagnetic Variables

Electrical signals in the body are important and can yield a variety of useful information
about a patient’s condition. Measurements of electrical potentials, currents, impedances and
magnetic fields help to determine the status of electrically excitable tissues and to make elec-
trical measurements of the properties of other tissues. These electrical measurements all re-
quire electrodes to couple the electrical signals of the body to the electronic instrumentation.
These electrodes can be considered to be sensors since they convert the electrical signals in the
body which are based on chemical gradients and membrane potentials to electrical signals in
instruments that are based upon energy and transport of electrons.
Electrodes for biopotential measurements have taken on many different forms over the years
[24]. Basically they consist of an electronic conductor that is coupled to the body through an
electrolyte layer. The electrical interaction is in the form of oxidationheduction reactions be-
tween the electrode and the electrolyte. Even when the electronic instrumentation presents a
very high load impedance to the electrodes, a small amount of current must flow to measure
the potential and this must cross the electrode/electrolyte interface by means of oxidation/
reduction reactions. Ideally, biopotential electrodes should be non-polarizable, that is there
should be no charge build-up at the electrode/electrolyte interface due to charge crossing this
interface. In practice, this is not usually possible. Although electrodes such as the silver-silver
chloride system can approach this ideal situation, they do not completely meet the re-
quirements, furthermore, silver ions can actually be toxic in the body under certain cir-
cumstances.
Electrodes are used to measure electrical potentials in acute and chronic situations as well
as invasive and non-invasive situations. The most common approach is to measure potentials
on the body surface using electrodes placed on the skin. Electrodes can also be implanted in
a tissue using needles, or surgically secured to the tissue being measured. Once in place these
electrode systems are used to measure bioelectrical potentials such as the electrocardiogram
(arising from the heart), electromyogram (arising from muscle), electroencephalogram (arising
from the brain), nerve action potentials (arising from nerves), electrogastrogram (arising from
the stomach), electro-optogram (arising from the position of the eye), etc [36]. These poten-
tials can also be related to currents in the body such as injury currents in the heart during
myocardial infarction and volume currents arising through many different soft tissues in
response to biopotentials.
By passing currents through the body via a set of electrodes and measuring the resulting
potential differences, or by applying a constant voltage across the electrodes and measuring
the current, one can determine the electrical impedance of different tissues. This technique
has been useful in monitoring breathing [37], in the plethysmographic measurement of blood
volumes in tissue [38] and in measuring the presence of other fluids in the extracellular space,
for example, in tissue edema [39].
572 21 Medical Diagnostics and Patient Monitoring

In recent years, the magnetic fields associated with electrically excitable tissues have been
measured. The sensors needed to achieve this are complex because the magnetic fields involved
are very small. The field strength of the magnetoencephalogram is of the order of lo-**T as
compared to the earth's magnetic field at the surface of 5 x T. Thus, special sensors
that require the use of superconducting materials at very low temperatures must be used in
specially designed magnetically shielded rooms to detect these magnetic fields of biologic
origin. The sensors, collectively known as SQUID (superconducting quantum interference
devices), operate in a liquid helium cryostat and although large and difficult to maneuver, they
are exquisitely sensitive to the very small magnetic fields encountered in living tissues [40].It
is thought that the new high temperature superconducting materials that are becoming
available will contribute to the simplification of this apparatus. The primary application of
these sensors is in measuring the magnetic fields associated with the heart, a technique known
as magnetocardiography, and magnetic fields associated with electrical activity of the brain,
known as magnetoencephalography. At the present time, however, these sensors are primarily
limited to research applications.

21.4.8 Chemical Sensors

Since physiological systems are based upon chemical interactions, there are many important
applications for chemical sensors in medicine. These sensors can be classified with regard to
their modus operandi or according to the types of materials that they sense. Table 21-4 lists
some of the types of chemical sensors that are important in medicine and gives some of their
principal applications.

Table 21-4. Chemical sensors and their medical applications.

s s
v)

i2Y
M

=
-0
0
--
Electrical conductivity
Electrochemical + +
Enzyme electrodes
Immunosensor
Ion-sensitive field effect transistor +
Polarography +

21.4.8.1 Electrochemical Sensors

Body chemistries can be assessed using electroanalytical techniques similar to those applied
in the laboratory. Miniature electrochemical sensors for conductiometric, potentiometric,
21.4 Examples of Sensors in Medical Applications 573

amperometric and voltammetric determinations have been used for these purposes in medical
instrumentation [41].
Perhaps one of the simplest electrochemical sensors that has found application in the field
of biological measurements is the electrical conductance sensor. A basic electrical conduc-
tance probe for measuring the electrical conductivity and hence the electrolyte content of
uterine cervical mucous is illustrated in Figure 21-7. This sensor consists of two parallel gold
conductors deposited on an alumina substrate with a separation of 0.5 mm. When a 22 kHz
constant amplitude current is passed between the electrodes, the resulting voltage across the
electrodes is proportional to the resistivity of the solution that the electrodes are in contact
with. This proportionality is fixed provided the extent of the solution is large with respect to
the spacing between the electrodes. By using microelectronic photolithographic techniques to
form the electrodes, reproducible sensors can be produced in quantity. This sensor has been
placed on the tip of a probe to allow clinicians to quantitatively evaluate the uterine cervical
mucous of women in situ during pelvic examinations in an effort to determine optimal times
for artificial insemination [42].

INSULATING
GOLD FILMS

Figure 21-7. Sensor for determining the electrical conductivity of uterine cervical mucous in situ. The
dimensions are in millimeters.

One of the major biological applications for electrochemical sensors is the determination
of the partial pressure of oxygen in body fluids and tissues. The widely applied Clark electrode
is a two-electrode amperometric cell with an oxygen permeable membrane [43]. This sensor
can be used either for determining gaseous oxygen partial pressures for use in pulmonary
measurements or it can be used in determining the partial pressure of dissolved oxygen in body
solutions such as blood or extra-cellular fluid. In either case, the oxygen to be measured dif-
fuses through the oxygen permeable membrane into the electrochemical cell where it is
reduced according to the reaction

0, + 2H,O + 4e- == 4 0 H -

It is seen that four electrons are required for each oxygen molecule that is reduced. It is
possible to determine the oxygen availability at the cathode by measuring the cathode current
and this is found to be proportional to the partial pressure of oxygen at the sensor membrane.
574 21 Medical Diagnostics and Patient Monitoring

There are similar potentiometric techniques which are used for determining oxygen tension.
In this case, an electrochemical reaction between molecular oxygen and a metal electrode
determines a potential that is logarithmically related to the amount of dissolved oxygen in the
cell [44].
Amperometric and potentiometric sensors can be developed for other chemical substances
found in the body including ions, dissolved gases and some biochemical materials.
Optical techniques have also been important for the development of chemical sensors both
in the laboratory and in their medical applications. Optical sensors made possible the
automatic analyses of chemical specimens in the clinical laboratory, and now these optical
systems are moving out of the laboratory into the patient monitoring area as well [45].The
pulse oximeter is an instrument that transilluminates peripheral tissue such as a finger or toe
and by spectrophotometrically analyzing the transmitted light is able to determine the percent
oxygen saturation of the hemoglobin in arterial blood [46].Chemical indicators have been
fixed to the tips of fiber-optic probes that can be introduced into the vasculature [47].These
indicators respond to oxygen, carbon dioxide and hydrogen ions in the arterial blood, and
these probes can be used for both acute and chronic measurements in patients.

21.4.8.2 Bioanalytical Sensors

One of the most exciting areas in biomedical sensors in recent years has been the develop-
ment of bioanalytical sensors [48].These devices are used for detecting small amounts of
biochemical substances in specimens, and whilst currently they are used in the clinical
laboratory, perhaps eventually they will also be used in the body itself. These sensors make
use of some sort of biologically specific material to form a selective transducer for a specific
analyte. One form of this sensor that has been extensively investigated is the enzyme electrode.
In this case, a particular substrate such as glucose enters a biochemical reaction that either
generates or consumes a substance that can be measured with a chemical sensor of the type
described in the previous paragraphs. For example, if it is desired to detect glucose, the enzyme
glucose oxidase immobilized near an electrochemical oxygen sensor will oxidize the glucose
to gluconic acid and consume oxygen in the process. Thus, the local oxygen partial pressure
will decrease as oxygen in the vicinity of the oxygen sensor is consumed. The amount of ox-
ygen consumed will be dependent on the amount of glucose available, thus the oxygen partial
pressure which can be measured by the oxygen sensor will be directly proportional to the
glucose concentration. The important aspect in the design of this type of enzyme electrode
is the process whereby the enzyme is immobilized in the sensor. Since enzymes are complex
proteins, they must be immobilized in such a way that they retain their biochemical activity.
It must also be possible to maintain this immobilized enzyme for a long enough period of time
to make the sensor useful. This remains a significant stumbling block which limits the
development of enzyme electrodes for applications requiring a more protracted enzyme-
substrate contact time.
Another type of bioanalytical sensor is based upon the highly specific and very sensitive
antigen-antibody reaction. A general sensor that employs this technique is illustrated in
Figure 21-8. Again, an important part of the sensor is a biospecific membrane that selectively
interacts with an analyte. A physical sensor coupled to the membrane detects this interaction
21.5 Summary 575

and generates the resulting electrical signal. Four different possibilities for this physical sensor
are indicated in Figure 21-8.

0 r

-
ELECTRICAL ELECTRODES
n Potent 1.1
O D Currant

THERMAL THERMISTORS

SON I C SAW DEVICES ELECTRICAL


D SIGNAL
OPT I GAL PHOTODETECTORS
\

BIOSPECIFIC
MEMBRANE
Figure 21-8. General example of a bioanalytical sensor.

The biospecific membrane can be made up of a material that undergoes electron exchange
reactions with the analyte so that there is a direct electrical output signal consisting of a poten-
tial or a current that can be detected by electrodes. The interaction can be a thermal interac-
tion in which case sensitive temperature transducers such as thermistors can be used to detect
the analyte. Since mass changes are involved in the interaction between the biospecific mem-
brane and the analyte, sensors that are capable of detecting small changes in mass can be
bound to the biospecific membrane. One such sensor involves the use of ultrasound surface
acoustic waves (SAW). The propagation of surface acoustic waves is affected by the mass of
materials on the surface of the propagating medium. Thus, surface acoustic wave devices that
have a biospecific membrane bound on their surface can be used to detect the analyte. Optical
interactions can also be used in this way. When the analyte is bound to the biospecific mem-
brane, the optical properties of this membrane can change, and this change may be detectable
with optical sensors such as photo-detectors.
Bioanalytical sensors are under development to facilitate the measurement of such diverse
materials as glucose, lactate, various hormones, neural transmitters, and creatinine.

21.5 Summary

The applications of sensors in medicine represent a broad field in which just about every
type of sensor developed can find some use. Although currently the markets for these sensors
are not as large as they might be in industry or consumer products these markets are expan-
ding rapidly. As medical procedures move out of the hospital and into the physician’s office
or even the home, the need for sensors and their associated instrumentation will continue to
increase.
516 21 Medical Diagnostics and Patient Monitoring

21.6 References

[l] Hanker, J. S., Giammara, B. L., Science 242 (1988) 885-892.


[2] Enzer, S., Diamond, H., Workshopon Theory, Design and Biomedical Application of Solid State
Chemical Sensors, March 28-30, 1977, Cleveland, OH: Case Western Reserve University.
[3] Geddes, L. A., The Direct and Indirect Measurement of Blood Pressure, Chicago: Year Book, 1970.
[4] Vaughan, M. S. in: Encyclopedia of Medical Devices and Instrumentation Vol. 4, Webster, J. G.
(ed.); New York: J. Wiley 1988, pp. 2723-2730.
151 Geddes, L. A., Electrodes and the Measurement of Bioelectric Events, New York: J . Wiley, 1972.
[6] Bartels, H., Bucherl, E., Herty, C. W., Rodewald, G., Schwab, M., Methods in Pulmonary
Physiology, New York: Hafner, 1963.
[7] Montoye, H. J. in: Encyclopedia of Medical Devices and Instrumentation Vol. 1, Webster, J. G. (ed.);
New York: J. Wiley, 1988, pp. 1-15.
[8] Lauersen, N. H., Hochberg, H. M., Clinical Perinatal Biochemical Monitoring, Baltimore: Williams
& Wilkins, 1981.
191 Huch, R., Huch, A., Lubbers, D. W., Transcutaneous PO,, New York: Thieme-Stratton, 1981.
[lo] Astrup, P., Severinghaus, J. W., The History of Blood Gases, Acids and Bases, Copenhagen:
Munksgaard, 1986.
[ll] Fowler, K. T., Phys. Med. Biol. 14 (1969) 185-199.
[12] Katona, P. G., CRC Crit. Rev. Bioengrg. 8 (1982) 281-310.
(131 Nalecz, M., Lewandowski, J., Werynski, A., Zawicki, I., ArtiJ Organs 2 (1978) 305-309.
[14] Brink, S. J., Stewart, C., 1 Am. Med. Assoc. 255 (1986) 617-621.
[15] Clark, L. C., Duggan, C. A., Diabetes Care 5 (1982) 174-180.
[16] Yao, S. J., Chan, L.-T.,Wolfson, S. K., Krupper, M. A., Zhou, H. F., IEEE Trans. Biomed. Engrg.
33 (1986) 139- 146.
1171 Cough, D. A., Leypoldt, J. K., Armour, J. C., Diabetes Care 5 (1982) 190-198.
[18] Haffner, M. E., Gottesman, R. A., in: Encyclopedia of Medical Devices and Instrumentation:
Webster, J. G. (ed.), New York: J. Wiley, 1988; Vol. 3, pp. 1509-1516.
[19] Jespersen, E., Neuman, M. R., Fortieth Annual Coderenre on Engineering in Medicine and
Biology, Proceedings, Niagara Falls: Alliance for Engineering in Medicine and Biology, 1987, p. 104.
[20] Fri, F. J. (ed.), Ultrosound: Its Applications in Medicine and Biology, Amsterdam: Elsevier, 1978.
[21] Morris, J. R. W., J. Biomech. 6 (1973) 729.
(221 Starr, I., Rawson, A. J., Schroeder, H. A., Joseph, N. R., Am. J. Physiol. 127 (1939) 1-28.
[23] Ennever, J. F., Knox, I., Denne, S. C., Speck, W. T., Pediatr. Res. 19 (1985) 205-208.
[24] Stobbart, D., Diffey, B. L., Clin. Phys. Physiol. Meas. 1 (1980) 267-273.
[25] Gershon-Cohen, J., Sci. Am. 216, No. 2, (1967) 94-102.
[26] Curry, 111, T. S., Dowdey, J. E., Murry Jr., R. C., Christensen’s Introduction to the Physics of
Diagnostic Radiology, Philadelphia: Lea & Febiger, 1984.
[27] Yartak, J., Phonocardiology: Integrated Study of Heart Sounds and Murmurs, New York: Harper
& Row, 1972.
[28] Neuman, M. R. in: Bronzino, J. D., Biomedical Engineering and Instrumentation: Basic Concepts
and Applications, Boston: PWS Engineering, 1986, pp. 43-44.
[29] Neuman, M. R., Liu, C. C., Proc. IEEE 76 (1988) 1218-1225.
[30] Fox, R. H., Solman, A. J., J. Physiol. 212 (1971) 8-10.
[31] Greatorex, C. A., in: IEE Medical Electronics Monographs, Watson, B. W. (ed.); London: Peter
Peregrinus, 1971.
[32] Roberts, V. C., Blood Flow Measurements, Baltimore: Williams & Wilkins, 1972.
[33] Bergel, D. H., Gessner, U., in: Methods in Medical Research Vol. 11, Rushmer, R. F. (ed.); Chicago:
Yearbook, 1966, pp. 70-82.
[34] Indwelling and Implantable Pressure Transducers, Fleming, D. G., KO, W. H., Neuman, M. R.,
(eds.); Cleveland: CRC Press, 1977.
[35] Riederer, S. J., Proc. ZEEE 76 (1988) 1095-1105.
[36] Geddes, L. A., Electrodes and the measurement of Bioelectric Events, New York: J. Wiley, 1972.
[37] Olson, T., Daily, W., Victorin, L., ACTA Paediatr. Scand. (Suppl.) 207 (1970) 1-27.
21.6 References 577

1381 Nyboer, J., Electrical Impedance Plethysmography, 2nd ed., Springfield: Thomas, 1970.
[39] Khan, M. R., Guha, S. K., Tandon, S., Roy, S. B., Med. Biol. Engrg. Compul. 15 (1977) 627-633.
1401 Biomagnetism: Applications and Theory, Weinberg, H., Stroink, G., Katila, T., (eds.); New York:
Pergamon, 1985.
1411 Medical and Biological Applications of Electrochemical Devices, Koryta, J., (ed.); Chichester :
J. Wiley, 1980.
142) Neuman, M. R., Medical Progress Through Technology 9 (1982) 95-104.
(431 Clark, L. C., Trans. Am. Soc. Artif: Intern. Organs 2 (1956) 41.
1441 Towell, M. E., Lysak, I., Layne, E. C., Bessman, S. P., . IAppl. Physiol. 41 (1976) 245.
1451 Peterson, J. I., Vurek, G. G., Science 224, No. 4645, (1984) 123-127.
1461 Payne, J. P., Severinghaus, J. W., Pulse Oximetry, Berlin: Springer-Verlag, 1986.
[47] Gehrich, J. L., Lubbers, D. W., Opitz, N., Hansmann, D. R., Miller, W. W., Tusa, J. K., Yafuso,
M., IEEE Trans. Biomed. Engrg. 33, 117-132.
1481 Cobbold, R. S. C., Transducers for Biomedical Measurements, New York: J. Wiley, 1974.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

22 Aerospace Instrumentation
ISIDORE WARSHAWSKY.
NASA Lewis Research Center. Cleveland. OH. USA

Contents
22.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 580
22.2 Total and Static Pressure Measurement . . . . . . . . . . . . . . . 581
22.2.1 Total Pressure Probes . . . . . . . . . . . . . . . . . . . . . . . 582
22.2.2 Static Pressure Measurement . . . . . . . . . . . . . . . . . . . . 584
22.2.3 Average Pressure Measurement . . . . . . . . . . . . . . . . . . . 586
22.2.4 Dynamic Pressure Measurement . . . . . . . . . . . . . . . . . . 586
22.2.5 Impact Pressure Measurement . . . . . . . . . . . . . . . . . . . 587
22.3 Flow Direction Measurement . . . . . . . . . . . . . . . . . . . . 587
22.3.1 Probes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587
22.3.2 Impact Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . 588
22.3.3 Angle of Attack Sensors . . . . . . . . . . . . . . . . . . . . . . 588
22.4 Temperature Measurement . . . . . . . . . . . . . . . . . . . . . 588
22.4.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 588
22.4.2 Total Temperature of a Flowing Gas . . . . . . . . . . . . . . . . 589
22.4.3 Static Temperature of a Hot Gas . . . . . . . . . . . . . . . . . . 590
22.4.4 Temperature of Cryogens . . . . . . . . . . . . . . . . . . . . . 591
22.4.5 Temperature of Solid Objects . . . . . . . . . . . . . . . . . . . . 591
22.5 Probe Cooling . . . . . . . . . . . . . . . . . . . . . . . . . . 592
22.6 Measurement of Aircraft Speed . . . . . . . . . . . . . . . . . . . 593
22.7 Fluid Velocity Measurement . . . . . . . . . . . . . . . . . . . . 594
22.7.1 Local Linear Velocity . . . . . . . . . . . . . . . . . . . . . . . 594
22.7.2 Bulk Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . 595
22.8 Strain and Force Measurement . . . . . . . . . . . . . . . . . . . 596
22.8.1 Strain Gages . . . . . . . . . . . . . . . . . . . . . . . . . . . 596
22.8.2 Load Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . 597
22.8.3 Thrust Measurement . . . . . . . . . . . . . . . . . . . . . . . 597
22.9 Acceleration Measurement . . . . . . . . . . . . . . . . . . . . . 598
22.10 Aircraft Icing . . . . . . . . . . . . . . . . . . . . . . . . . . 598
22.11 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 599
580 22 Aerospace Instrumentation

22.1 Introduction

Sensors and associated instrumentation may be chosen for any of the following purposes:
1. To provide an accurate indication of the value of some physical quantity.
2. To provide a reproducible, but not necessarily accurate, indication of some physical
quantity.
3. To provide a sensitive indication of changes in some physical quantity whose absolute
magnitude is not accurately known.
Requirement 1. may be needed if the measurement is to be used, together with other
measurements, in the computation of some derived parameter (eg, the computation of aircraft
speed from measurements of total pressure, static pressure, and total temperature). Require-
ment 2. may be needed to set a nominal operating point (eg, to maintain an indicated jet
engine exhaust gas temperature). Requirement 3. may be needed in research or development
to determine the effect of varying just one of several independent parameters (eg, to observe
the effect of a change of load distribution on the strain in some structural element of an air-
craft).
Usually, the principal problem in measurement is to establish the relation between the
magnitude of the quantity to be measured and the indication of the sensor used to make that
measurement; the sensor’s indication itself can usually be measured without difficulty. For ex-
ample, although the temperature of a thermometer bulb immersed in a fluid can be deter-
mined with negligible error, the relation between bulb temperature and fluid temperature
depends on many properties of the bulb, the fluid, and the installation.
This chapter will be devoted principally to indicating the special considerations involved in
establishing the relation between some common sensors and the physical quantities those sen-
sors are intended to measure in order to serve the purposes indicated above. Very frequently,
only an indirect measurement can yield the quantity of principal interest: it must be deduced
from an appropriate combination of actual measurements provided by a variety of sensors.
(Aircraft velocity is a common example.)
A distinctive feature of aerospace application is the wide range of physical conditions under
which the sensors must operate. Air density covers all regimes from the free-molecule to the
continuum (although only the latter regime, wherein molecular mean free path is small com-
pared to the dimensions of mechanical parts, will be covered here). At supersonic speeds,
deviations from perfect-gas behavior become significant. Fluid flow must be measured at
temperatures ranging from 20 K to 300 K. In aircraft flight, a sensor may be subjected to ap-
preciable acceleration in any direction; and the deleterious effects of rain and ice must be
prevented.
The subjects selected for treatment here are the measurement of gas pressure and
temperature, of fluid flow velocity, of aircraft velocity and aircraft icing, of rocket thrust, and
of structural acceleration, load, and strain. In some cases, the measurement technique will be
emphasized, rather than the sensor. (An example is gas temperature measurement by spec-
troscopic methods, using optical sensors; such methods supplement thermocouples and
pneumatic probes in solving gas-dynamics problems.)
22.2 Total and Static Pressure Measurement 581

22.2 Total and Static Pressure Measurement

The total pressure p t of a flowing gas is the pressure that would be produced if the gas
were brought to rest isentropically. The static pressure p s of a flowing gas is the pressure in
a direction normal to a streamline. Knowledge of these two pressures and of the isentropic
exponent y (the specific-heat ratio, for a perfect gas) is sufficient to determine the Mach No.
of the gas stream. However, determination of the gas velocity u also requires knowledge of
static gas temperature T,. The relations among these variables are presented in [l]. For a
perfect gas, these relations, in terms of measured or assumed parameters, are

M 2 = (rZ6’Y - l ) / b u =Ma
(22-1 a)
a’ = yR,T,/m, T,= Tt/(l + bM2)
where
r = Pt/Ps b = ( y - 1)/2 (22-1 b)

and a is the local speed of sound, Tt is total gas temperature, m, is the relative molar mass
(“molecular weight”), and R , is the universal gas constant (8314 J/kmol K).
For the supersonic flow of a perfect gas, if Md represents the Mach No. downstream of a
normal shock, as computed from measured quantities by Equation (22-l), and if these
measured quantities are assigned the second subscript “ind”, then the upstream parameters
M, p s ,p t are given by

M Z = 4b/[4y - ( y + 1)2r-26’yl (22-2a)

Pr Ips,ind = 2c/(y + 1) (22-2b)


= ( b + l)Y+’/(CdY)
(pt/pt,ind)Y-’ (22-2c)

where

r = p t,ind
. /Ps,ind b (7 - 1 ) / 2 ;
(22-2d)
c Y M $- b d =b + (l/M$).

If the shock is oblique, at acute angle 0 with the direction of upstream flow (at Mach
No. M), then, in Equation (22-2), M and Md should be replaced, respectively, by M sine and
Md sin@ - (p), where (p is the acute angle between the directions of flow upstream and
downstream of the shock ((p could be the half-angle of a symmetrical wedge-shaped surface
to the tip of which the shock is attached, as in Figure 22-1). The value of 0 is given by

sin68 - g sin4e + h sin2e - (1/M4)


cos2(p = 0 (22-3a)
582 22 Aerospace Instrumentation

where

g =1+ 2 M - 2 + ysin2p (22-3 b)


h = 2 M - 2 + M - 4 + [(1/4)(y + + ( 2 b / M 2 ) ]s i n 2 p . (22-3 C)

(The smallest of the three roots should be disregarded.)

Figure 22-1.
Definition of angles in an oblique shock wave.

22.2.1 Total Pressure Probes

These are pitot tubes in the form of a slender tube whose axis is aligned with the direction
of flow. The stagnation pressure that occurs at the nose of the probe is transmitted by means
tubing to a transducer that measures the pressure. Common uses are:

in aircraft flight, as a means of establishing Mach No. and airspeed,


in rakes transverse to a duct or conduit, as a means of establishing velocity and density
distributions, or the bulk (integrated) fluid velocity, or the thrust per unit area,
to study boundary layer pressure distributions,
to determine the average value of a fluctuating pressure,
to determine the instantaneous value of a fluctuating pressure.

A review of the pitot-tube literature up to 1956 is presented in [ 2 ] . Pitot tubes for subsonic
and supersonic speeds are treated in [3]. In subsonic flow, a pitot tube will provide a pressure
indication &,ind that is a direct measure of pf. In supersonic flow, &ind is the total pressure
behind a normal shock, and P t , i n d / p f is a sensitive function of Mach No. M. However, the
quantitypt,ind/@u2)is not sensitive to Mach No., provides a good measure of stream thrust
per unit area, and has been used for this purpose [4]. Figure 22-2, derived from the equations
in [ I ] ,illustrates these relationships. The ratio Pt,ind/@u2)is relatively insensitive to gas com-
position (it changes 2 percent when the isentropic exponent y changes from 1.4 to 1.3), whereas
the ratio is quite sensitive to such changes (it changes 30 percent at M = 4 for the
same change in y).
A necessary feature of a pitot tube is that it be tolerant of misalignment between the tube’s
axis and the direction of flow. In the laboratory, a simple square-ended tube with internal
22.2 Total and Static Pressure Measurement 583

Total pressure
Drobe

Shock wave

Mach no. M
Figure 22-2. Relations between a pitot probe's pressure indication and free stream parameters, in super-
sonic flow.

bevel and almost sharp leading edge (lip) can tolerate a misalignment of 20". Figure 22-3 is
a corrected version of data from [ 5 ] and indicates the misalignment tolerances possible with
various commonly used designs. Additional data appear in [3].

Angle 8 (Deg)
Figure 22-3. Effect of misalignment on the indication of total-pressure probes.
584 22 Aerospace Instrumentation

22.2.2 Static Pressure Measurement

In aircraft flight, static pressure provides the most common means of determining altitude
and vertical speed (time rate of change of altitude). In wind tunnels and in flight, it determines
the forces on structures like wings and fuselages. In conduits, together with total pressure, it
is used to deduce bulk velocity or local linear velocity. The means of sensing static pressure
are :
1. Probes (often combined as pitot-static probes), pointed into the stream, with small holes
in the surface at a location where the gas flow is substantially parallel to the surface. Tubing
leads from these holes to a suitable transducer or gauge. Apitot-staticprobe will have an addi-
tional hole at the nose, with separate tubing leading to the total-pressure transducer.
2. In conduits, holes in the wall of the conduit (or of the throat of a venturi used to measure
the speed of the flow), drilled so that the axis of the hole is accurately normal to the surface.
In aircraft, a similar installation may be made on the fuselage (this subject will be treated in
Section 22.7) or on an airfoil surface.
Static-pressure measurement is more difficult than total-pressure measurement because, for
probes, the pressure at the sensing holes is influenced by:
- the shape of the probe - its length, its cross section, and its nose shape,
- the size and location of the sensing holes,
- the shape and location of the support,
- the proximity of other probes or structures,
- misalignment between the axis of the probe and the direction of the flow; the angle bet-
ween these vectors may change with Reynolds No. and Mach No. To minimize the effect
of misalignment, several holes are usually placed around the periphery of the probe,
leading to a common plenum that provides the average pressure.
Reference [3] treats these factors for a variety of probe designs that are in the form of a
circular cylinder. Probes have also been built that are in the form of wedges or cones, with
the orifices on the inclined surfaces. Some of these are treated briefly in [5]. The wedge is
preferred when the probe is mounted on a strut that is perpendicular to the probe; the cone
is preferred when its support is a sting that extends in the same direction as the long dimension
of the probe.
The cylinder, wedge, and cone can be used in supersonic flow. In such flow, a bow wave
stands in front of a blunt-nosed probe like the Prandtl design, and a conical shock wave is
almost attached to the tip of a cone or wedge. In either case, the pressures sensed by the probe
are those of the subsonic flow behind a shock wave. Figure 22-4 shows how a particular pitot-
static probe (in this case, a f 15O cone) may be used in supersonic flow; by use of the lower
curve, the Mach No. M is first determined from the ratio ps,ind/pt,ind of the two measured
pressures; then the static pressure p s is determined from M and ps,ind by use of the upper
curve.
Curves of this type depend on the isentropic exponent y. For the example of Figure 22-4,
a change of y from 1.4 to 1.3 lowers the upper curve about 5% and raises the lower curve 1%.
These sensitivities are deducible from the equations of [l].
Figure 22-5, from [S], shows the effect of misalignment for three designs of probes; the
probe shapes and sensing-hole locations are shown to scale in this figure. However, more
22.2 Total and Static Pressure Measurement 585

serious effects arise from the proximity of the support, of the conduit wall (when present),
and of adjacent probes. These effects are summarized in [6].

ISo Half-angle cone

(rOA-
I
1- I 0

+ Q
z
u *
m .=
c
.o ZlQ
o
c
-
-
II
II
I
Flow
+ --
0 Ps.ind to
ptSindtransducers
0
- \

Shock wave
LL 0.1
- I '
- I Ps.ind
- I -
I hind
I
I Static- and total-
-
II pressure probe
I
0.01 II

-m
0)

C
0

8 O Wedge Prondtl tube

I
loo Cone

-20 -10 -5 0 5 15 20
-v
-15 10
Angle 8 (Deg)
Figure 22-5. Effect of misalignment on the indication of static-pressure probes.
586 22 Aerospace Instrumentation

22.2.3 Average Pressure Measurement

When the correct average of a pressure fluctuating at frequency f is desired, the time con-
stant T of the pneumatic connection to the transducer must be larger than 1 /(2nf ). For tubing
of length t and cross-sectional area A leading to a transducer volume V, and if t A Q the
time constant is

T = 8xvt V/(a2A2) (22-4)

where v is the kinematic viscosity of the fluid and a is the speed of sound. Some other situa-
tions are treated in 171.
The pressure transducer indication will not represent a perfect average, if the pressure-fluc-
tuation amplitude is a large fraction of the mean pressure, because the flow coefficient of the
pitot-tube opening varies with the flow direction. An indication of the possible magnitude of
the error due to this nonlinear averaging is given in [8].

22.2.4 Dynamic Pressure Measurement

In the laboratory, it is often necessary to measure rapidly changing pressures in locations


where limitations of space or of the environment prohibit the use of flush-mounted pressure
transducers (where the pressure-sensitive diaphragm of a transducer is coplanar with the sur-
face sustaining the pressure). Typical fields of application are shock tubes, jet- and rocket-
engine combustion chambers, and jet-engine turbine test facilities. A probe or surface hole is
then used to sense the pressure, which is transmitted by appropriate connecting tubing to a
pressure transducer. Fidelity of response is then sought. The techniques that may be used are:
- The probe microphone. A transducer is located at the far end of tubing of appropriate
geometry [9].
- The infinite line method. A transducer is attached to the side of the tubing at a location
near to the pressure source, without disturbing the flow along the tubing, which is so long
as to act like a tube of infinite length.
When the probe microphone is used, some of the factors that may require consideration are
the pneumatic delay time, [/a, the acoustic damping and resonances of the combination of
transducer and connecting tubing, the wave shape distortion when the waves are of high
amplitude, and the change of transducer volume with pressure.
Design factors are treated in an elementary way in [7], and more thoroughly in [lo] and [ll].
Usually, fidelity of response is achieved if 7 is much less than 1/2xJ and if

A’ = 32x2 v 2 t V / a 2 , (22-5)

assuming t A 4 K
When the infinite line technique is used, the most important requirement is to minimize any
reflections or perturbations of the flow at the transducer location, by preserving the constancy
of cross-sectional area. This method was initially used by [12]; it is summarized in [13] and
illustrated in greater detail in [14].
22.3 Flow Direction Measurement 5 87

22.2.5 Impact Pressure Measurement

Impact pressure is the difference between total pressure pt and static pressure p s . In incom-
pressible flow, it is identical with the dynamic pressure (1/2)p u 2 , where p is density and u is
linear velocity. Ordinarily, impact pressure is obtained from measurements of p , and p s , or by
a direct measurement of their difference. However, a direct determination also is obtainable
from measuring, with resistance strain gages, the strain in a small, cantilevered impact plate
upon which the fluid stream impinges [15]. This device has been used in studies of turbulence
and of the local flow velocities in jet-engine compressors and fans, where frequencies of
several kilohertz occur. The successful construction of this device depends on the proper in-
stallation of the strain gages, a topic treated in the references given in Section 22.8.

22.3 Flow Direction Measurement

22.3.1 Probes

The direction of airflow in a single plane is often determined by using a probe resembling
the wedge or cone used for static pressure measurement. Two diametrically opposite holes, in
the inclined surfaces, lie in the plane of interest. The difference between the two pressures is
measured, rather than the sum; this difference is approximately proportional to the angle bet-
ween the airflow direction and the probe's axis. A probe with hemispherical or ogival nose
may also be used. Flow direction in two orthogonal directions may be determined by using
four holes, 90" apart. The wedge then becomes pyramidal. The pyramidal wedge may be
preferable to the cone because of reduced sensitivity to local pressure gradients and to changes
in Reynolds No.
Such probes may also have a fifth central hole that permits measurement of total pressure
(Figure 22-6a). If the pressure at each hole is measured separately, it is possible to deduce
total, static, and dynamic pressure, as well as flow direction. Such a 5-hole combination is par-
ticularly desirable, because the accurate computation of flow direction in one plane requires

(a) (b)

Pitch angle=(p,, - p s 3 ) / ( p s 4 + p s 3 ) Yaw angle~(Psz-Psl)/(Psz+psi)

Pso = (Ps, Psz


+ + P53 + PSL 1
Figure 22-6. 'Ik.0 designs of five-hole probes for total pressure, static pressure, and flow direction.
588 22 Aerospace Instrumentation

a correction for any nonzero flow angle in the orthogonal plane, and may also depend on the
total pressure.
Figure 22-6 also shows the dimensionless functions of measured pressures that best serve
to yield the parameters desired. Static pressure p s is equal to ps,oin Figure 22-6 only when the
pitch and yaw angles are zero. At all other times, the equations for these angles and for p s
must be multiplied by empirically determined correction factors that depend on the measured
pressures. Corrections are usually best described as multiples of pt - ps,o.
When a wedge or cone is used, the included angle may be increased to obtain greater sen-
sitivity to angle, at the cost of reduced accuracy in static-pressure measurement.
Alternative to the pyramidal construction is a 5-tube cluster consisting of four circular
tubes, 90" apart, around a central tube (Figure 22-6b). The latter is square ended; the others
have 45" chamfers, so that the pressures sensed are like those of the 5-tube pyramidal design.
Reference [16] illustrates the interactions among the various pressures that are sensed.
It is necessary that, if any pressure difference between two holes is to be measured or com-
puted, there be equal pneumatic time lag in transmitting each pressure to its respective
transducer. This subject was discussed in Section 22.2.4. Transducers of high linearity and high
accuracy may be required.
Direction-sensing probes are described in [17], which is a tutorial on the subject. Reference
[18] has described a 4-hole probe, in the form of a truncated triangular pyramid, that serves
all the purposes of a 5-hole probe, but without the redundancy that can provide cross checks.

22.3.2 Impact Plate

This device, described in Section 22.2.5, can also be used to determine flow direction if it
is turned 90" so that the flow is against the edge of the plate [15].

22.3.3 Angle of Attack Sensors

For angle of attack measurement in aircraft flight, an alternative approach is to provide a


pivoted vane that will align itself with the local wind direction. The angle assumed by the vane
relative to the aircraft structure is measure9 by an appropriate position transducer, like a
rotary differential transformer. The location of the device on the aircraft must be chosen so
that there is an unequivocal correlation between vane position and the angle of attack of in-
terest. Empirical calibration in a wind tunnel is essential. Heating of the device is required if
there is a possibility of icing. Such devices have been used at supersonic speeds [19].

22.4 Temperature Measurement

22.4.1 Definitions

In a flowing gas, total temperature is the temperature that the gas would acquire if it
were brought to rest (to stagnation) isentropically. Total temperature governs the rate of heat
transfer between a gas and a solid surface upon which the gas impinges. Temperature sensors
22.4 Temperature Measurement 589

that depend on this heat transfer often yield the heat transfer rate more reliably than they yield
the total temperature.
Static temperature T, is the temperature that would be measured by an observer traveling
with the gas, along a streamline. Static temperature is needed to establish the speed of sound
in a gas and thereby to establish the gas velocity from a knowledge of the Mach No. M.The
static temperature also serves to define the gas composition and the partition of energy among
the various energy states: the external energy of molecular translation, and the internal energy
of molecular vibration, molecular rotation, electronic excitation, ionization, dissociation, or
recombination [20]. The understanding of a combustion process is aided by knowledge of
these energies, each of which is associated with its own “temperature”. The process of propul-
sion depends on the translational energy; so does the process of heat transfer, unless there is
chemical action at the surface. The quantity T, conventionally is intended to represent the ex-
ternal (translational) energy and is the temperature that appears in the perfect-gas law. The
value of y, also, depends on the partition of energy that exists.
When there is a high gas acceleration or deceleration such as produced by ignition, shock,
or rapid expansion, equilibrium among the various energy states is attained only after suffi-
cient time has elapsed. This time may range from nanoseconds to milliseconds, depending on
the gas composition. A measure of this time for any energy state is the relaxation time, which
is the time constant of the exponential response of that state to a step change in energy level
such as may be associated with a change in T,. The vibrational energy usually is the internal
energy state that has the longest relaxation time.
Reference [21] presents comprehensive bibliographies on temperature measurement from
1930 to 1969. Reference [22] describes newer developments.

22.4.2 Total Temperature of a Flowing Gas

Probes that use a thermocouple or resistance thermometer as the temperature-sensitive ele-


ment (TSE) are designed to measure T . (Static temperature may be deduced from T , M , and
y.) Any probe design represents a balance among several requirements:
1. sufficient mechanical strength to withstand the impact pressure of the flowing stream,
2. sufficiently low thermal time constant to provide fidelity of response to rapid changes
in gas temperature. Time constant is proportional to the quotient of thermal heat capacity of
the TSE by the coefficient of heat transfer between the fluid and the TSE,
3. sufficient stagnation of flow at the TSE to approximate T . The recovery ratio T,ind/T
is the best practical measure of success; the recovery factor (T,ind- T,)/(T, - T,) is a com-
mon academic measure of success. Here, T,indis the TSE temperature in the absence of con-
duction and radiation errors,
4. sufficiently high convective heat transfer between the fluid and the TSE,
5 . sufficiently low heat transfer by conduction between the TSE and the probe support,
6. sufficiently low heat transfer by radiation between the TSE and its surroundings.
The balance among these criteria is illustrated in [23] and [24]. The latter compares various
designs of probes; some of these designs use aspiration of the gas to create sonic velocity at
the TSE and, thereby, a predictable recovery ratio and a short time constant. All designs in-
tended for high-temperature service permit water cooling of the exterior of the probe.
590 22 Aerospace Instrumentation

Thermocouples are commonly used TSE’s for high temperature service. Chromel/alumel
(type K) is useful up to about 1100 K. Pt-Rh alloy/Pt (type B, R, o r S) is useful up to about
1500 K if there is no concern about the catalytic action of Pt in completing a combustion reac-
tion. Standard calibrations for these wires are in [25]; convenient equations for electromotive
force as a function of temperature are in [26]; commercial tolerances are in [27]. The
W5Re/W26Re combination [28] is recommended for use u p to 2600 K in nonoxidizing at-
mospheres.
Reference [29] gives details of the design and performance of some practical probes.
Probes that have no fundamental upper temperature limit are the pneumatic probe and
various heat transfer devices. The pneumatic probe is a water-cooled aspirating device that has
been used extensively [30] so that its characteristics are well understood. It actually provides
a measure of gas density if the gas velocity at the entrance is low and a measure of the product
of density and pressure if the velocity is nearly sonic. Total temperature may be deduced if
p, is measured after stopping the aspiration.
Three heat-transfer devices have been reviewed in [31]. They are particularly well suited to
surveying the local enthalpy distribution across a hot gas stream.
In supersonic flow, there is usually a normal shock in front of the probe. The correlation
between free-stream conditions and the gas conditions actually measured by the probe
depends on a knowledge of the isentropic exponent y. If the location where the measurements
are made is very near to the shock front, the measurement may be influenced by whether o r
not vibrational modes are relaxed (ie, whether there has been time for the vibrational energy
to come into equilibrium with the translational energy). At very high temperatures, relaxation
of dissociation must also be considered. As a rule of thumb, one may assume that equilibrium
has been reached if the relaxation time is less than one half of the time of transit from the
shock front to the location where the gas property is being measured. Relaxation phenomena
have been treated by [32] and [33]. Application to the pneumatic probe is treated in [34].

22.4.3 Static Temperature of a Hot Gas

This temperature is usually deduced by optical methods. These “non-intrusive” methods


are particularly useful when a flame or other chemically active gas is being studied, because
the presence of a probe may affect or cause a chemical reaction. Sometimes several optical
methods are used because each contributes to a better understanding of what is happening
in the gas. The use of optical emission from the gas is also particularly useful because the ra-
diant flux is several times more sensitive to temperature than to gas density or emittance. This
increased sensitivity to temperature arises from the exponential character of Planck’s Law and
becomes more pronounced as the wave length of the band o r line becomes shorter.
If the optical method is one that uses the integral over an optical path through the gas, the
integration is usually nonlinear. For example, if gas emission is measured, the emission will
come principally from the nearer end of the gas path unless the gas emissivity is very low. An
indication of the errors obtainable is presented in [35]. Nonlinear averaging is also treated in
[36]. Popular optical methods are the fine- and band-reversal methods [37-401, and the ab-
sorption-emission method [41, 421. These have been shown to yield correctly the translational
temperature of a gas in equilibrium. The line-reversal method may be localized and thereby
improved in accuracy by local injection of the emitting material whose resonance line is being
22.4 Temperature Measurement 591

used [40]. The accuracy of the absorption-emission method (eg, one using the radiation from
H 2 0 to measure the temperature of the exhaust of a jet engine or of a H2-02 rocket) may be
improved by substantially simultaneous measurements at several wavelengths, using the spec-
tral scanning method described in [43, 441. Typical applications are described in [45, 461.
Other optical methods use a high-resolution spectrometer to obtain spectral intensity
distributions within a molecular band. These methods, and others, are enumerated briefly in
[47], which provides references to more intensive and authoritative treatments. More recent
and comprehensive expositions are [48-511.
In the same gas stream of known M, y, and p s , at about 2000 K, the indications of a bare-
wire thermocouple, a pneumatic probe, and a line-reversal pyrometer deviated less than 2%
from the mean of the three indications, in more than 95% of the intercomparisons [35]. Other
comparisons among pyrometers are reported in [52].
The availability of the laser as a source of high-intensity radiation has made practical the
use of Raman spectroscopy, wherein scattered radiation from the focal point of one or two
laser beams injected into the gas is observed. The temperature is correlated principally with
the spectral distribution pattern of the scattered radiation, which involves the measurement
only of relative intensities; however, the determination of concentration requires the measure-
ment of absolute intensities. Practical measurements are possible only for a gas whose optical
parameters are known. The measurements are impaired by any absorption or emission along
the path of observation, by the presence of aerosols, particulates, or severe density gradients,
and by the presence of molecular species that produce interfering spectra of comparable inten-
sity.
When the gas is in equilibrium, the rotational or vibrational temperatures of the Raman
emission also represent the translational temperature. However, Raman spectroscopy has been
most effective in identifying molecular species when there is chemical activity, as in flames.
The foundations of Raman spectroscopy are presented in [53]. A review of one modern
development, coherent anti-Stokes Raman spectroscopy (CARS), is in [54], which lists advan-
tages and disadvantages of this technique. Recent advances and applications of this method,
of others that use Raman radiation, and of still others that use lasers as sources, are described
in Section VII of [55], Section IV of [56], and Session I of [57]. References [54-571 serve as
bibliographic sources on these techniques.

22.4.4 Temperature of Cryogens

Platinum resistance thermometers are usually used to measure temperatures in the range 20
to 400 K. A review of other thermometric elements particularly suitable for the liquid-
hydrogen range (20 K) is given by [%]. Reference 1221 Vol. 4 contains descriptions of many
techniques of cryogenic thermometry. Conduction errors and dynamic response are discussed
in [59]. References [60]and [61] are reviews and comprehensive bibliographies on cryogenic
thermometry.

22.4.5 Temperature of Solid Objects


The following two aerospace applications are of considerable importance.
To measure the temperature of a surface that is at or near the temperature of liquid
hydrogen or oxygen, any suitable device listed in [58] may be attached to the surface with the
592 22 Aerospace Instrumentation

following precautions: (i) the size of the device must be very small compared to the size of
the surface element that extends in the direction of the temperature gradient and that has the
same cross-sectional area as the area of contact; (ii) electrical wires extending from the ther-
mometer must lie along an isotherm that passes through the point of contact, for a distance
of at least 20 wire diameters. Requirement (i) is intended to assure adequate speed of response;
requirement (ii) is intended to minimize conduction errors.
For operational purposes, a measurement of the average and the peak temperatures of the
turbine of a jet engine provides a more effective means of control of engine operation, to
achieve maximum efficiency and fuel economy, than does the use of an exhaust-duct ther-
mocouple. The latter device is always built so massively, for practical reasons, that its conduc-
tion and radiation errors are large. An engine operating point based on the indication of such
a thermocouple, though fully reproducible, is more conservative (and therefore less
economical of fuel) than an operating point based on turbine blade temperature. A narrow-
band (monochromatic) radiation pyrometer sighted upon the turbine blade has been used to
measure blade temperature (621. A narrow radiation band is used to reduce the uncertainty
caused by uncertain blade-surface emittance. Further reduction is possible by use of two-
wave-length (2-color) pyrometry if emittance is likely to be the same at both wavelengths. The
associated uncertainties are discussed in [47]; eg, at 1200 K, the percentage uncertainty in
temperature is about 1/4 of the percentage uncertainty in the ratio between the two emittances.
For research purposes, radiation pyrometry has been used to establish the profile of
temperature along a turbine-blade surface [63], [64].
The practical concerns in such pyrometry are:

- emission from or absorption by the gas in the optical path,


- reflection, by the surface being viewed, of radiation from other sources, such as com-
bustors or other blades,
- maintenance of the cleanliness of windows or other optical surfaces in contact with the
gas, by use of a clean-air purge.

22.5 Probe Cooling

Probes to be inserted into hot gas streams must be water cooled. Very short probes exten-
ding from a water-cooled strut may be made of copper, with reliance on conduction to keep
the probe cool. Longer probes are usually encased in an outer shield that is cooled by cir-
culating water. Another coaxial tube may provide a return path for the water. Alternatively,

High conductivity material


\ Coolant overboard

la) Conventional cooling (b) Conduction cooled tip (c) Single-pass cooling

Figure 22-7. Methods of cooling a probe tip.


22.6 Measurement of Aircraft Speed 593

the coolant may be ejected into the gas stream through holes near the tip of the probe, if the
ejected coolant will not affect any downstream measurements or operations. Some designs are
illustrated in (30, 31, 471 and in Figure 22-7.

22.6 Measurement of Aircraft Speed

On aircraft, the airspeed is deduced from measurements of total pressure, static pressure,
and total temperature. At altitudes up to 90 km, the isentropic exponent y may be taken as
1.4. The basic equations for an ideal gas are in [l]. If
..
(22-6 a)

then the velocity v, in terms of measured or assumed parameters, is given by

v 2 = (1 - r-2b’Y) y T R,/(m, b) (22-6b)

for subsonic flow and by

u2 = BT,(R,/m,) [I - r-2b’y]-’ (22-6~)

for the velocity upstream of a normal shock wave. Deviations from the ideal-gas conditions
may occur at high Mach Nos. because
1. at high gas densities, correction must be made for the size of the gas molecules by using
an equation of state that is more exact than the perfect-gas law,
2. at high temperatures, specific heats change with temperature because there are ap-
preciable amounts of vibrational and rotational energy,
3. at still higher temperatures, there may be dissociation and ionization of molecular
species, and electronic excitation energies also become appreciable,
4. at locations near a shock front, there may be a time lag in achieving equilibrium between
the translational and the vibrational energies, thereby affecting the total-pressure measure-
ment.
Effects 1 and 2 are treated briefly in [l] and more completely in [65]. At altitudes below
90 km, only effect 2 is significant. At higher altitudes, effects 3 and 4 may require considera-
tion; they are treated in [66] and [67]; applications are illustrated in [68]. Although the time
lag and dissociation problems are known, the solutions may not be, especially for gas mix-
tures.
Total pressure may be sensed by a probe (as in Section 22.2.1) extending from the nose of
the aircraft or from an airfoil suface like the wing or the vertical tail fin; in the latter cases,
the probe is usually located near the tip of the airfoil, in order to minimize the effects of
fuselage proximity. However, the location may not be one which, during different portions of
the aircraft’s speed range, will be on different sides of an oblique shock originating from some
upstream structural element. Pitot tubes may also be attached to the side of the fuselage by
594 22 Aerospace Instrumentation

means of a streamlined strut extending 10 to 30 cm from the fuselage. The strut may be heated
to prevent icing.
Static pressure may also be sensed by probes (as in Section 22.2.2) in similar locations, in
which case both total and static pressure may be sensed by a single pitot-static probe. Alter-
natively, and especially for routine aircraft operations, static pressure may be sensed by
orifices located on the side of the fuselage. These orifices are usually used in pairs, in
diametrically opposite locations, so that the average of the two pressures will be more likely
to represent the correct value in the presence of yaw or pitch. A more refined computation
of airspeed will include correction for pitch angle or angle of attack, determined by use of
a flow-direction sensor (Section 22.3) (which also may be incorporated into the pitot-static
probe).
When ruggedness is essential, the lip of a pitot tube is rounded, and the internal contour
resembles a convergent nozzle; such a design also reduces the sensitivity to misalignment with
the local direction of airflow.
Temperature is usually sensed by a platinum resistance thermometer (PRT) element in a pro-
tective housing that resembles a fuselage-mounted pitot tube. If the housing is heated to pre-
vent icing, the PRT element may be surrounded by one or more concentric tubes that act as
radiation shields; incoming air passes through these tubes. To provide for inertial separation
of water droplets, the air entering the assembly is directed along two paths. One path is in
the direction of the entering air and terminates in one or more small bleed holes; entering
water follows this path and drains through the rearmost hole. The other path is at right angles
to the first path and leads to the PRT winding. With proper sensor design, only the smallest
droplets can follow this path; a drain hole is provided for them - its presence is also required
in order to provide adequate speed of response to temperature changes. Air impinging on the
front end of the PRT winding is brought to virtual stagnation, with a known recovery ratio.
Pitot probes may also have such bleed holes for removal of rain. The holes may never be
so large as to affect appreciably the process of stagnation that is needed for total-pressure and
total-temperature measurement. The exterior of all probes may be electrically heated to pre-
vent ice formation.
Sensors for supersonic and hypersonic flight may be the same as those for subsonic flight.
However, pitot-static probes extending from the fuselage nose are most common.
A wind tunnel calibration is required for all airspeed probes used on a given model of air-
craft. In operations like takeoff and landing, the presence of the ground may alter these
calibrations. However, in such operations, it is more important to have a reproducible “in-
dicated airspeed” than to have an accurate value.

22.7 Fluid Velocity Measurement

22.7.1 Local Linear Velocity

The velocity of a fluid at a particular location in a gas stream may be determined by any
of the following methods:
1. A measurement of pitot and static pressures and of the direction of the flow, by using
sensors covered in Section 22.2.1, Section 22.2.2, and Section 22.3.1.
22.7 Fluid Velocity Measurement 595

2. A measurement of impact pressure and of flow direction, using two orientations, 90”
apart, of the probe described in Section 22.2.5.
3. Use of a hot-wire anemometer, wherein the relation between heat transfer rate and mass
flow rate per unit area is known. When two or three wires, at various angles, are used on a
single probe, both flow amplitude and flow direction may be deduced. Such probes, because
of their small size, are particularly useful in studies of boundary layers. A related device is
the hot-film anemometer, which is more rugged but more bulky. A bibliography on this sub-
ject is contained in [69].
4. Laser doppler velocimetry (LDV) and related techniques using lasers, which are “nonin-
vasive” methods requiring no interference with the gas stream other than the injection of par-
ticles that can reflect incident laser radiation to a sensitive detector. One popular technique
uses a pair of beams that intersect at the locality of interest. Two velocity components may
be measured by using two pairs of beams that have orthogonal polarizations or, preferably,
two pairs with different wavelengths. A monodispersion of particle diameters is usually
desirable. The choice of particle diameter and concentration depends on the desired frequency
response and the available detector sensitivity. Particles of 1 pm diameter usually permit
following 10 kHz fluctuations in velocity although observations in the vicinity of a shock wave
may require 0.1 pm particles. Principles are described in [70-731. Bibliographies appear in
[71-731 and, to 1974, in [74-751. Advances since 1982 have been reported biennially in [76].
Applications of LDV and related techniques have been described in Session I1 of [57].

22.7.2 Bulk Velocity

In many aerospace applications, the bulk mass flow rate of a fluid in a pipe is of principal
interest. However, flowmeters may actually provide a measure of
1. mass flow rate m (eg, gyroscopic, Coriolis, momentum, or Thomas-type calorimetric
meters),
2. volume flow rate v (eg, turbine, vortex-precession, or electromagnetic meters),
whereupon density p must also be determined if m is sought,
3. some intermediate function of m and p (eg, pitot-static, flow nozzle or orifice, impact
plate devices), whereupon p must also be determined if m or is sought.
Flowmeters may also be classified by whether they use
(i) bulk (mass or volume) velocity sensing, wherein the measured signal represents the in-
tegral of some function of local velocity over the cross section of a passageway, or
(ii) local linear-velocity sensing, wherein bulk velocity is deduced from one or two local
velocity measurements, by assuming knowledge of the velocity profile in the pipe.
Examples of (i) are all meters in l., above, and the turbine, electromagnetic, vortex preces-
sion, and nozzle or orifice-type meters. Examples of (ii) are the ultrasonic and the boundary-
layer calorimeter types, and devices using probes like those of the pitot-static and vortex-
generating types.
Reference [77] is a tutorial on flow metering. Reference [78] describes many varieties of
flowmeters.
For gases, density p can be deduced from measurements of pressure and temperature. For
liquids of known chemical composition (eg, H,, O,, C,H,OH), a temperature measurement
596 22 Aerospace Instrumentation

is usually sufficient to establish the density, although a pressure measurement may serve to
improve accuracy. For liquids of uncertain composition, one may use

- a densitometer, such as one that measures the frequency of vibration of a spring-supported


fixed volume of liquid,
- a meter that uses a combination of two different operating principles, eg, a design that
measures the torque on the rotor of a turbine-type meter as well as its speed,
- a meter that permits measurement of more than one physical parameter (eg, an elec-
tromagnetic flowmeter that provides a measurement of the linear-velocity induced emf and
of the density-proportional capacitance [79]).

The only velocity profiles which are unequivocally defined are the fully-developed laminar
and turbulent profiles in a circular pipe. The length of straight, smooth pipe, of diameter D,
required for full development of these profiles is on the order of 0.1 . DN, and 20 DN,, -
respectively. Alternatively, a flowmeter which is permanently attached to a “straightening”
assembly, consisting of a succession of screens or vanes and mixing chambers, may be
calibrated and used as a unit that constitutes the “meter”. Some straightening assemblies
merely remove swirl; others also act to flatten the profile. All are intended to create a velocity
distribution that is independent of the profile at the meter’s entrance. Such techniques are
treated in depth in [77]; one is described in [80].
Only the gyroscopic, Coriolis, and momentum-type flowmeters require no straightening
assemblies; all other types usually do require them in order to achieve maximum accuracy. In
aircraft, there is little room for straightening assemblies; momentum-type flowmeters have
therefore been popular.
lbrbine-type flowmeters are popular, in the laboratory, for measuring liquid hydrogen and
oxygen [80]. However, on rocket engines, where space is limited and vibration is severe, a
measurement of liquid temperature and pressure and of fuel- and oxidant-pump speeds may
be sufficient to yield the mass flow rates, without recourse to flowmeters.
A singular consideration in metering the flow of cryogenic propellants is the fact that these
fluids are often used near to their boiling points, so that local pressure in a flow sensor must
not be permitted to drop to the point where undesired two-phase flow may occur.
Reference [81] is a bibliography on flow measurement to 1970. Some more recent
developments appear in [78].

22.8 Strain and Force Measurement

22.8.1 Strain Gages

Modern resistance strain gages, used for structural analysis, are usually of the foil type.
Distinctively among sensors, the foil gage that is used cannot be the gage that is calibrated.
The use of these gages relies on the statistical reproducibility of a given batch of gages. If one
or several of a single batch of gages has been calibrated on a specimen of the material on
which it will be used, one assumes that all gages of that batch, that are subsequently used on
the material to be tested, will have the same sensitivity (gagefactor). The root-mean-square
22.8 Strain and Force Measurement 5 97

uncertainty (standard deviation) of this assumption is usually of the order of a few percent.
Any one batch is designed to match, as nearly as possible, the temperature coefficient of ex-
pansion of the material on which it will be used. These gages and the techniques of their use
are treated in [82] and [83].
The following distinctive considerations are applicable.
1. To the extent possible, the number of gages in a single batch should be large enough to
serve a complete testing program, so that a minimum number of calibrations will be needed.
2. For static strain measurements at low temperatures (< 500 K), on a given material, the
gage material can usually be selected and processed so that temperature changes will have
minimal effect on the strain indication.
3. If static strains are to be measured, the effect of the temperature coefficient of resistance
of the gages will be minimized if the gages are used in pairs, mechanically placed and elec-
trically connected so that the output signal is proportional to the difference between
- tensile and compressive strains at locations on equal and opposite sides of the neutral axis,
or
- strains in orthogonal directions. (Assemblies with this configuration are commercially
available).
4. Gages from a batch to be used for static-strain measurements at elevated temperatures
(> 600 K), should be mounted on a specimen that is representative of the material to be tested,
and subjected to the same pattern of temperature, strain, and time as planned for the test, in
order to evaluate stability, reproducibility, and probable accuracy of the projected
measurements.
5 . Gages intended solely to measure the alternating component of strain, at several hertz,
may be used in a single-gage configuration, provided gage temperature remains substantially
constant over several cycles. The gage material chosen then should have high fatigue strength.

22.8.2 Load Cells

The load cell, a spring whose deflection is usually sensed by resistance strain gages, is the
most common means of measuring force. Commercial cells measure a unidirectional force
and, usually through symmetrical construction, are relatively insensitive to transverse forces.
Such cells may be calibrated with dead weights or other primary standards, and can have ac-
ceptably small errors due to hysteresis, nonlinearity, and elastic creep. They are usually com-
pensated for the effects of temperature, but are not often compensated for the effects of
temperature gradients.
Load cells are used for in situ weighing of aircraft, as well as for measurement of thrust
and other forces.

22.8.3 Thrust Measurement

In the laboratory, the thrust of a rocket or jet engine may be determined from:
1. The integral of dynamic pressure as measured by an array of total-head tubes at the
engine exhaust (see Section 22.2.1). If the exhaust is axisymmetric, only a linear radial array
is needed. Optimum locations for such probes are given in [7] and in [77].
598 22 Aerospace Instrumentation

2. Measurement, with load cells, of the force on a tank into which the engine exhausts in
such manner that the gases leave the tank in a direction orthogonal to the direction in which
they enter, so that the area-integrated dynamic pressure of the exhaust is converted into a force
on the tank.
3. Measurement, with a load cell, of the force between the engine and its attachment to the
earth. Practical considerations then require that
- auxiliary restraints be provided against any moments and transverse forces that are present
because the line of action of thrust does not coincide with the load cell axis,
- these restraints do not affect the thrust measurement, or else that their effects are corrected
for,
- the stiffness of other mechanical connections, like fuel and coolant lines, between the
engine and the earth be known and corrected for, or else be included in the load cell
calibration. The internal hydraulic pressure in such lines also may affect both line stiffness
and the forces on the engine. When this technique is used, it is necessary to define these
tare effects quantitatively by a “cold run” in which as many normal operating conditions
are duplicated as possible, without actually firing the engine.

22.9 Acceleration Measurement

The lift-related stresses in the structures of aircraft during gusts or acrobatic maneuvers can
be correlated with the vertical acceleration of the craft’s center of gravity (c.g.), as measured
by an appropriate accelerometer. The measurement is more likely to be valid if
- the sensor is mounted near the c.g. of the aircraft,
- the sensor is attached to an extremely rigid portion of the airframe (like a main spar, if
one exists), rather than being mounted on some resilient structural element,
- the sensor is relatively insensitive to higher-frequency vibrations that may be due to such
an element. A properly damped sensor with a natural frequency of a few hertz is adequate
to emphasize the significant accelerations that represent the motion of the aircraft as a
whole.
The study of the stresses in small, discrete structural elements can be aided by use of vibra-
tion sensors (seismometers, velocimeters, or accelerometers) that, combined with appropriate
differentiating or integrating electrical networks, can yield the local displacement, velocity, or
acceleration. Such information supplements that provided by strain gages.

22.10 Aircraft Icing

Icing of jet engines with axial flow compressors and of fixed- and rotary-wing aircraft sur-
faces is particularly likely when the static air temperature ( T , ) is slightly below 0°C and
when the liquid water content (LWC) exceeds 0.1 g/m3 or when the product of LWC and for-
ward velocity exceeds 7 g/(m2 s). A qualitative discussion of this subject is presented in [84].
22.11 References 599

The detection of ice already forming is usually best accomplished through the pilot’s visual
observation. A special probe, more likely to accumulate ice than other structures, may be
mounted in the pilot’s view. This probe can then be manually de-iced by use of a built-in elec-
trical heater.
The severity of the threat posed by a given rate dy/dt of ice accumulation depends on T,,
the LWC, and the droplet size distribution, and on the particular model of aircraft. However,
loosely speaking, a rate of ice accumulation of an aircraft surface of the order of 3 cm/h may
warrant precautionary action.
A more serious and important problem is the detection of the incipient stage of icing. Sen-
sors for this purpose also are designed to be more likely to accumulate ice than the aircraft’s
surfaces. Such sensors are described in [84]. They effectively measure the thickness Ay of ice
that is accumulated in time At. Details of the rotating disk, pressure-sensitive pitot probe,
vibrating probe, and light-beam occluder, that are listed in [84], appear in [85-881, respec-
tively. The latter two devices use aspirators to induce airflow past the ice-accumulating sur-
face; they can therefore be used on helicopters, whose blades can accumulate ice even when
the aircraft is hovering. These instruments indicate LWC rather than Ay/At. The LWC and
T, are the principal measurable determinants of the severity of the threat of icing.
An economic advantage of such detectors is that, in addition to warning when icing preven-
tion will be needed, they also indicate when such prevention is not needed. The result is a
substantial reduction in power (and therefore, of fuel) that would otherwise be expended as
a precautionary measure.

22.11 References

Abbreviations used in this list:


AGARD Advisory Group for Aerospace Research and Development, Neuilly sur Seine, France
ANSI American National Standards Institute, New York, NY
ASME American Society of Mechanical Engineers, New York, NY
ASTM American Society for Testing and Materials, Philadelphia, PA
BHRA British Hydromechanics Research Association, Cranfield, Bedford, UK
FDS Federal Duplicating Service, Library of Congress, Washington, DC
HMSO Her Majesty’s Stationery Office, London
IEEE Institute of Electrical and Electronic Engineers, New York, NY
ISA Instrument Society of America, Research Triangle Park, NC, USA
NACA National Advisory Committee for Aeronautics, (documents from NTIS or FDS)
NASA National Aeronautics and Space Administration, Washington, DC, USA (documents from
NTIS or FDS)
NBS National Bureau of Standards, Gaithersburg, MD, USA (documents from NTIS, older ones
from FDS)
NLR Nationaal Lucht- en Ruimtevaartlaboratorium, Amsterdam
NPL National Physical Laboratory, Teddington, Middlesex, England
NTIS National Technical Information Service, Springfield, VA, USA
RAE Royal Aircraft Establishment, London
SAE Society of Automotive Engineers, Warrendale, PA, USA
600 22 Aerospace Instrumentation

[l] Ames Research Staff, NACA Report 1135, 1953, NTIS, Springfield, VA.
[2] Folsom, R. G., ASME Trans. 78 (1956) 1447-1460.
[3] Gracey, W., Measurement of Aircraft Speed and Altitude, New York Wiley, 1980.
141 Davidson, T. W., AGARDograph No. 103, Part 1, (1965) 217-244.
[5] Gettelman, C. C., Krause, L. N., ZSA Proc. 7 (1952) 134-137.
[6] Krause, L. N., Gettelman, C. C., ZSA Proc. 7 (1952) 138-141.
[7] Warshawsky, I., Foundations of Measurement and Instrumentation, NASA Ref. Publ. 1222, in press.
[8] Krause, L. N., ISA Trans. l3 (1974) 142-148.
[9] Leonard, R. W., 1 Acoust. SOC.Am. 36 (1964) 1867-1871.
[lo] Iberall, A. S., J. Res. NBS 45 (1950) 85-108.
[ll] Bergh, H., Tijdeman, H., NLR - TR R238. 1965, NLR, Amsterdam.
[I21 Blackshear, P. L., Rayle, W. D., Tower, L. K., NACA TN 3567, 1955, NTIS, Springfield, VA.
[13] Warshawsky, I., SAE Trans. 87, Paper 780076, (1978).
[14] Englund, D. R., Richards, W. B., ZSA Pans., 24, No. 2, (1985) 11-19.
[15] Krause, L. N., Fralick, G. C., ZSA Trans. 21, No. I, (1982) 37-44.
[16] Dudzinski, T. J., Krause, L. N., NASA TM X-1904,1969.
1171 Bryer, D. W., Pankhurst, R. C., Pressure-Probe Methods for Determining Wind Speed and Flow
Direction, London: HMSO, 1971.
[IS] Shepherd, I. C., ASME Pans. 1 Fluids Engng. 103 (1981) 590-594.
(191 Richardson, N. R., Pearson, A. O., NASA TND-122, 1959.
1201 Bennett, J. G., Pirani, M., 1 Znst. Fuel 12, No. 64, (1939) Sl-S4.
1211 Freeze, P. D., et al, NBS Circular 513 (from NTIS), 1951, and NBS SP 373 (from NTIS), 1972.
[22] Temperature, its Measurement and Control in Science and Industry,
Vol. 3, Herzfeld, C. M. (ed.); New York: Reinhold, 1962;
Vol. 4, Plumb, H. H. (ed.); New York: Amer. Inst. of Phys., 1972;
Vol. 5 , Schooley, J. F. (ed.); New York: Amer. Inst. of Phys., 1982.
[23] Moffat. R. J., in [22] Vol. 3, Part 2, pp. 553-571.
[24] Scadron, M. D., Warshawsky, I., Gettelman, C. C., ZSA Proc. 7 (1952) 142-148.
[25] NBS Staff, NBS Monograph 125, 1974, NTIS, Springfield, VA.
1261 Coates, P. B., Smith, A. C. K., NPL Report QU36, 1977, NPL, Teddington, UK.
[27] ANSI Std. MC 96.1-1982, ANSI, New York, NY.
[28] ASTM Std. E-988-84, 1984, ASTM, Philadelphia, PA.
[29] Glawe, G. E., Holanda, R., Krause, L. N., NASA TP 1099, 1978.
[30] Warshawsky, I., Kuhns, P. W., in [22] Vol. 3, Part 2, pp. 573-585.
[31] Krause, L. N., Glawe, G. E., Johnson, R. C., in [22] Vol. 3, Part 2, pp. 587-593.
[32] Herzfeld, K. F., in: High Speed Aerodynamics and Jet Propulsion Vol. 1, Rossini, F. D. (ed.);
Princeton, NJ: Princeton University Press, 1955, pp. 646-735.
[33] Gaydon, A. G., Hurle, I. R., in: Eighth Symposium (International) on Combustion, New York:
Williams & Wilkins, 1960, pp. 309-318.
1341 Kuhns, P. W., NACA TN 4026, 1957, NTIS, Springfield, VA.
[35] Warshawsky, I., in: Sixth Symposium (International) on Combustion, New York: Reinhold, 1956,
pp. 739-750.
[36] Buchele, D. R., NASA TN 0-2406, 1964.
[37] Fery, C., Comptes Rendus ... 137 (1903) 909-912.
[38] Strong, H. M., Bundy, F. P., 1 Appl. Phys. 25 (1954) 1521-1526.
1391 Kurlbaum, F., Phys. Zeit. 3 (1902) 187-188 and 332-334.
[40] Buchele, D. R., in [22] Vol. 3, Part 2, pp. 879-887.
[41] Schmidt, H., Ann. Physik 29 (1909) 971- 1028.
[42] Silverman, S., J. Opt. Soc. Am. 39 (1949) 275-277.
[43] Tourin, R. H., Spectroscopic Gas Temperature Measurement, New York: Elsevier, 1966.
22.11 References 601

[44]Buchele, D. R., in [22] Vol. 4, pp. 645-656.


(451 Buchele, D. R., NASA TN 0-2405, 1964.
[46] Lezberg, E. A., Buchele, D. R., NASA TN 0-2441, 1964.
[47] Warshawsky, I., ISA Journal 5 (1958) 91 -97.
[48] Penner, S. S., Quantitative Molecular Spectroscopy and Gas Emissivities, Reading, MA: Addison-
Wesley, 1959.
[49] Penner, S. S., in [22] Vol. 3, Part 1, pp. 561-574.
[50] Simmons, F. S., ISA Trans. 2, No. 2, (1963) 168-189.
[51] Gaydon, A. G., Wolfhard, H. G., Flames, Their Structure, Radiation, and Temperature, 4. ed., New
York: (Chapman & Hall) Methuen, 1979.
[52] Glawe, G. E., Johnson, R. C., Krause, L. N., in [22] Vol. 3, Part 2, pp. 601-605.
[53] Herzberg, G., Molecular Spectra and Molecular Structure, New York: van Nostrand (Vol. 1, 1950;
Vol. 2, 1945; Vol. 3, 1966).
[54] Tolles, W. M., Nibler, J. W., McDonald, J. R., Harvey, A. B., Applied Spectroscopy 31 (1977)
253-271.
[55] Characterization of High Temperature Gases, NBS SP-561, Hastie, J. W, (ed.); 1979, NBS,
Gaithersburg, MD.
[56] [22] Vol 5, Part 1.
[57] Advanced Instrumentation for Aero Engine Components,AGARD-CP-399,1986, AGARD, Neuilly
sur Seine, France.
[58] Sinclair, D. H., CIG, Cryogenics and Industrial Gases 5, No. 7, (1970) 15-22.
[59] Warshawsky, I., ISA Trans. 13 (1974) 337-346.
[60] Rubin, L. G., Cryogenics 10 (1970) 14-22.
(611 Rubin, L. G., Brandt, B. L., Sample, H. H., in [22], Vol. 5 with addenda, in Cryogenics22 (1982)
491 - 503.
[62] Curwen, K. R., Aircraft Engineering 44, No. 12 (1972) 16-21.
[63] Buchele, D. R., Lesco, D. J., in: Progress in Aeronautics and Astronautics Vol. 34, Cambridge, MA:
MIT Press, 1974, pp. 347-354.
[64] Uguccini, E. W., Pollack, F. G., NASA TN 0-8213, 1976.
[65] Eggers, A. J., NACA Report 959. 1950, NTIS, Springfield, VA.
[66] Truitt, R. W., in: Hypersonic Aerodynamics, New York: Ronald Press, 1959 Chapter 10.
1671 Zel’dovich, Ya. P., Raizer, Yu. P., Physics of Shock Wavesand High Temperature Hydrodynamic
Phenomena Vol. 1, Hayes, W. D., Probstein, R. F. (eds.); New York: Academic Press, 1961.
[68] Dorrance, W. H., ViscousHypersonic Flow, New York: McGraw-Hill, 1962.
[69] Freymuth, P., Bibliography of Thermal Anemometry, St. Paul, MN: TSI, 1982 and 1983.
[70] Drain, L. E., The Laser Doppler Technique, New York: Wiley, 1980.
[71] Watrasiewicz, B. M., Rudd, M. J., Laser Doppler Measurements, London: Butterworth, 1976.
[72] Durst, F., Melling, A., Whitelaw, J. H., Principles and Practice of Laser-Doppler Anemometry,New
York: Academic Press, 1976.
[73] Durrani, T. S., Created, C. A., Laser Systemsin Flow Measurement, New York: Plenum Press, 1977.
[74] Durst, F., Zare, M., Bibliography of Laser Doppler AnemometryLiterature, Skovlunde, Denmark:
DISA Information Dept., 1974.
[75] Stevenson, W. H., Pedigo, M. K., Zammit, R. K., Bibliography on Laser Doppler Velocimeters,U S.
Army Missile Command Report No. RD-TR-72-8, 1972, NTIS, Springfield, VA.
1761 Laser Anemometry in Fluid Mechanics, Lisbon: LADOAN-Institute Superior Tecnico, Vol. 1, 1982;
Vol. 2, 1984; Vol. 3, 1986; Vol. 4, 1988.
[77] Fluid Meters, 6th ed., Bean, H. W., (ed.); ASME, 1971.
[78] Flow, its Measurement and Control in Science and Industry, Dowdell, R. B., (ed.); Research Triangle
Park, NC: ISA Vol. 1. 1974; Vol. 2. 1984.
[79] Cushing, V., in [78] Vol. 1, pp. 723-733.
602 22 Aerospace Instrumentation

[80] Warshawsky, I., Hobart, H. F., Minkin, H. L., in [78] Vol. 1, pp. 709-719.
[81] Dowden, R. R., Fluid Flow Measurement, BHRA, Fluid Engineering, 1972.
(821 Handbook on Experimental Mechanics, Kobayashi, A. S. (ed.); Englewood Cliffs, NJ: Prentice-Hall,
1987.
[83] Perry, C. C., Lissner, H. R., The Strain Gage Primer, 2nd ed., New York: McGraw-Hill, 1962.
[84] Perkins, P. J., Sensors 4, No. 3, (1987) 33-43.
[85] Vonnegut, B., Cunningham, R. H., Katz, R. E., Army Air Force Technical Report No. 5519. 1946.
[86] Perkins, P. J., McCullough, S., Lewis, R. D., NACA RM E5IEI6, 1951, NTIS, Springfield, VA.
(871 Schwartz, B., Sheldon, R., in: Proc. 1982 Natl. Aerospace and Electronics ConJ, New York: IEEE,
1982, pp. 100-104.
[88] Barlow, G. F., RAE Tech. Memo. EP 641, HMSO, London, 1980.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989
Index

absorption alarm annunciation 495


- dermal 530 allergy, lab animal (LAA) 537
- photometer 482 alpha particles 532
absorption coefficient of a waveguide 30 aluminium oxide, CVD of 154
absorption-emission method 590 American Standard Code for Information Inter-
accelerated aging test 36ff. change (ASCII) 348, 364
acceleration ammonia measurements 521
- algorithms 62 amphiphilic materials 184
- measurements 598 amplification of electrical output 335
- sensors 565 analog signals see: signals, analog
accelerometers analog to digital converter (ADC) 359
- airborne applications 53 anemometer
- closed-loop 287 - basic 337
- micromachined quartz 287 - hot-film 595
- miniature 566 - hot-wire 595
- silicon cantilever 72 annealing 131, 182
accuracy 18f. - of microsensor structures 89
- of fiber optic sensors 232 annealing point 156
- of measurements 320 anodic bonding see: bonding, anodic
acoustic oscillator, liquid based 93 ANSI MC6.1 3
active transduction principles 9 antibodylhapten interaction 231
adsorbents, solid 532 antigedantibody
aerosol monitoring 544 - interaction 231, 540, 574
aerospace instrumentation 579-602 - sensors 94
agents approximation, Chebyshev 303
- biological 536 approximations for sensor characteristics, mini-
- foodborne 536 mum principle 302ff.
air-to-fuel (A/F) ratio 394, 397 ARIAM 420
- sensing 257 Arrhenius model 37f.
aircraft icing 598 f. ASCII (American Standard Code for Informa-
air monitoring 540 tion Interchange) 348, 364
- indoor 542f. ASIC (application specific integrated cir-
- occupational 543 ff. cuits) 339, 352, 380
- outdoor 540 see also: integrated circuits
air pollution monitoring 540 assembly techniques in IC technologies 136ff.
604 Index

ATMOS 548 biomedical research 560


atmospheric window for IR wavelengths 416 biomedical sensors
audiohide0 equipment 436ff. - classification of 557
Auger process 52 - connectors for 560
autoclaving 559 - Costs of 560
AUTOGUIDE 422 - implantable 556
automated manufacturing 445 ff. - indwelling 557
automated production, sensors in 447 ff. - materials for 556, 558
automation systems, home 442 - medical applications of 560
automotive onboard sensors 383-405 - packaging of 558f.
- engine control 385ff. - reliability of 559
- environment 384 - requirements of 557
- features of 384 - shape of 558
- gas 391 - size of 558
- reliability of 384 - sterilization of 559
- requirements of 384f. biphase encoding 371 f.
- resources for 385 bit-stuffing 375
autonomous sensor interfaces 370 bit error 376
autonomous sensors 362, 365, 380 blood analyzer, fetal pH micro- 561 f.
availability, arithmetic calculation of blood monitoring 545
524 ff. blood pressure monitoring 563
Bode plots 24
Boltzmann transport equation, diffusion approx-
imation of 197
ballistocardiograph 566 bonding
bandgap 203 - anodic 85, 91, 100, 155
Barkhausen noise 196 - - laser-assisted 158
barrier potential 234 - electrostatic 155 f.
baseband signal 372 - epoxy die 137
base function method 304f. - eutectic 162
base function series, expansion in 305 - field-assisted 91, 156, 158f.
batch assembly techniques 90 - fusion, silicon-to-silicon 160
bellows 387, 505 - glass, low temperature 161
best-fit straight line 20 - glass-to-silicon 156ff.
beta”-alumina 260ff. - glass frit 161
- Li-stabilized 261, 263 - layers of materials in IC technologies
- structure of 260ff. 155ff.
beta-alumina 259 ff. - metallic 162
- fabrication of 262 ff. - phosphosilicate glass 161
- structure of 260ff. - processes, thermocompression 138
beta particles 532 - process parameters 157f.
BICMOS processes 207, 341 - silicon-to-silicon 158f.
binary flag 375 - silicon fusion 92
bioaerosols, collection of 537 - spin on glass (SOG) 161f.
biological agents, detection methods for - tape automated (TAB) 85
537 - techniques 91
biologically effective dose 546 - thermal fusion 160
biological monitoring 545 ff. - thermosonic 138
biomedical applications 53, 197 - ultrasonic 138
biomedical microsensors 80 Bourdon tube 505
Index 605

breakdown - measures
- current-induced 27 - - electrical 299
- extrinsic 28 - - mechanical 299
- intrinsic 28 - processes for sensor arrays 318
- voltage 269 - self 561f.
Brewster angle 31 - sensors 32, 308
bridges, micromechanical 59 - spectra 318
broadband 373 - wind tunnel, for airspeed probes 594
buffer amplifiers 83 CAMAC system 352
bulk epoxy 93 capacitance
burr-contours 451 - electrical 390
burrs, measurement of 461 - parasitic 55, 82
burst noise 22 capacitance cell technology 509
bus capacitive effects, model equations for 55
- backbone 373 capacitive sensors 292, 387, 449, 468
- CAMAC 352 capacitor dielectric materials 272
- DC power supply 379 capacitors 288
- digital 335 - metal-insulator-semiconductor 184
- ERA1553 367 capsule spring 505
- field- see: fieldbus carbon monoxide analyzers 523 f.
- general purpose instruments, (GPIB) - specifications of 524
IEC625 366 carburetor-type engine 385
- Hewlett Packard Interface (HPIB) 366 CARIN 422
- IEEE488 366 carpet cleaners 439f.
- information 335 carrier
- intelligent driver of 349
- band 372
- lower level instrumentation 369
- - concentration, intrinsic 52
- multidrop 335f., 360, 363 - - deflection 197, 199
- multidrop serial 345
- - analytical model for 63
- OMNIBUS 367 - - effects 200, 210
- PROFIBUS 366f., 379, 510 - sense multiple access collision detect method
- serial 348
374
- Sieger 365 - transport, galvanomagnetic 47
bus based systems, multivendor 367
CARS (Raman spectroscopy, coherent anti-
- propriety 367
Stokes) 591
bus systems, digital 280
cartridges, sensing 561
butterfly specification 27
catalytic converters, three-way 391
byte stuffing 376
catastrophic failures 34
cell 3
- electrochemical 562, 573
centralized media access control (CMAC)
calibration 374
- automatic, microprocessor controlled 562 ceramic oxide materials 247 ff.
- capability, in situ 82 - perovskite-type 269
- curves 32 ceramic oxide sensors
- data 323 - functional properties of 250f.
- - for thermocouples 499f. - materials for 250f.
- - of resistance temperature detectors 502 ceramics 247-278
- foil gages 5% ceramic thermistors, NTC 275 ff.
- gas 318 chain code for image processing 457
606 Index

chambers, sensing 561 chemiresistors 190, 338


channeling, ion 131 chemosensors see: chemical sensors
characteristics of sensors 18ff., 281 f. chemosensor systems, task of 314
- approximation methods for 302ff. chi-squared distribution 35
- curvature of 299 chip layout 83 ff.
- degradation of 558 chips, GaAs Hall IC 207
- dynamic 23f., 282 chips, very large scale integration (VLSI) 109
- effects of interference on 304 choke coil system 296
- interpolation of 300 chromatography 481, 487
- linearization of 308 chromeValurnel 590
- mathematical model of 299ff. circuitry
- static 18ff. - active 80
- - digital modeling of 299ff. - incorporate 83
- transmission, linear 282 circuits, magnetic 296
- - static 281 Citypilot 422
- zero shift of 305 cladding propagation mode of light 229
characterization of sensors classification of measurement signals 495
- biological methods 32 classification of sensors 8
- chemical methods 32 Claus process 487f.
- documentation of results 33 cleaving 233
- electrical methods 27ff. clinical analysis 538
- mechanical methods 31 clinical laboratory 560ff.
- optical methods 30f. clinical monitoring 562 f.
- thermal methods 32 closed-loop control 495, 519, 540
charge densities, nonlinearity in 63 - systems, implantable 563
check code, 16 bit 376 clothes dryers, electric 440
ChemFET see: field effect transistors, chem- CMAC (centralized media access control) 374
ical CMOS (complementary metal oxide semiconduc-
chemical methods for thin film deposition 177 tor) 87 f., 207, 209ff.
chemical process engineering, sensing principles coating of polyimide, spin 102f.
in 481 ff. coatings of microsensors 89
chemical processes 476 coding, redundant 336
chemical sensor electrode arrays 349 coliforms 537
chemical sensors 230f.. 314, 338, 561 collective properties of materials 479
- electro- colorimeter 481
- - medical applications of 572f. colorimetric indicators 564
- - miniature 572 combustion chamber 488
- - oxygene 574 combustion control 255
- liquid-based acoustic 94 communications
- medical applications of 572ff. - interfaces, design of 380f.
- microfabricated 545 - processing 335, 380
- MOS 316f. - systems 333
- optical fiber fluorescent 231 - - optical components of 241
chemical signals 5 f. compact disk players 437f.
chemical vapor deposition (CVD) 89, 136, compatibility of microfabrication techniques
151ff., 177, 256 95 ff.
- low pressure (LPCVD) 152f. compensation
- metallo-organic (MOCVD) 111f. - circuits 284
- plasma-assisted 152f. - deductive 344
- plasma enhanced (PECVD) 177 - methods, classification of 343
Index 607

- monitored 343 corrections of sensors, dynamic 307


- principle 287 correlation method for pattern recognition
- problems 341 316ff.
- structural 343 counting techniques 292
components of source material, classification creep 34
of 486 critical measurements 524
computer-aided design (CAD) 207 cross-sensitivities 314, 323, 326, 342f., 347
- package 340 - correction of 304ff., 361
computer integrated manufacturing (CIM) - effects 212
446 - of gas sensors 309
concave lens 236 - of sensor elements 318
concentration limits, maximum 539 crosstalk 29
concentration measurements 314, 319 - error 487
- of substances 324 cryogenic thermometry 591
concentrations, metabolite 546 crystal growth 1lOff.
conducting rubber elements 455 - technique, Czochralski 262
conduction band 54, 235 crystalline structure 252 f.
conductivity measurements cubic splines
confidence envelope 323 - interpolation method for digital linearization
conformance 21 301
conformity 21 - natural 302
contact migration 41 - polynomials 307
continuity current loop, 4 to 20 mA 358
- generalized laws of 344 custom cable 85
- equation, carrier 47 cuvette 561
- - discretization of 61 CVD see: chemical vapor deposition
contour cyclic redundancy check (CRC) 376
- detection 454
- measurements, off-line 451
- - on-line 451
- tracking 451ff. data
- - sensor guided 464 - conversion 335, 349
control-region approximation 60 - - analog to digital 335
control equipment 490 - packet 374
control quantities 477 - processing 453
control systems - reduction in image processing 457
- closed-loop, implantable 563 - transmission, parallel digital 366
- failure of 495 - - serial digital 363
- torque feedback 399 deglaze 126
- of a smart rice cooker 436 degradation
- program 490 - performance 72
conversion - signal 370
- analog to digital 361 - of deposited films 135
- frequency to digital 288 - of sensor characteristics 558
converters 280 - of the cathode 393
- analog to digital 347, 349 delineation techniques for thin film processes
convex lens 236 179ff.
convolution integral 307 DeNOx 521
corekladding interfaces 221, 238 densitometer 596
core propagation mode of light 229 density measurements 595
608 Index

deposition 126 - profiles 127ff.


- chemical vapor see: chemical vapor deposi- - - phosphorous 129
tion - theory, thermionic 50
- membrane 96 digital communication 356
- metallo-organic (MOD) 177f. digital devices, field-mounted 509
- sputter- 176 digital linearization
- methods for thin films 176ff. - cubic spline interpolation 301 f.
- of monolayers 189 - look-up table method 299
- temperature 152f. - polygon interpolation 299f.
- - for encapsulation layers 90 - polynomial interpolation 300
design of sensors 19-89 digital signal 360
detectors 3 digital transmitters 360, 508
- inductive loop 410ff. digitization of sensor signals 291
- infrared, passive 415 diodes 536
- magnetic field 418 - infrared 449
- optical 418f. - laser 232
- passive infrared (IR) 415 - light-emitting (LED) 232, 432ff., 435
- radar 415 - magneto- 52, 197, 200
- ultrasonic 413 - p-n junction 53
- vehicle 408, 409ff. - photo- see: photodiodes
- - priority 419 - Zener 289
deviation function 303 DIP (packages, dual-in-line) 85, 139
diagnosis, machine 458 see also: packages
diagnostics, in vivo, optical fibers for 227 dipping
diamagnetic materials 196 - apparatus 188
diaphragm-type pressure sensor 433 - methods 189f.
diaphragm deflection 55, 68, 71 direct-reading instruments 543
diaphragms 7, 505f. direction-sensing probes 588
- capacitive 48 direct measurement analyzers 522 ff.
- n-type silicon 147 Dirichlet boundary
- piezoresistive 48 - conditions 50
- rectangular 67ff. - regions 60
- semiconductor 388 discretization procedures for numerical modeling
- silicon 292 methods 60f.
- thin 56 displacement
die attachment 136f. - float-type transmitters 507
die separation 136 - sensors 285
dielectric coating 233 - - angular 564
dielectric strength 27 - - inductive 296f.
differential-capacitor system 506 - - computer-corrected 309
differential-pressure methode 503 - transformers, rotary variable (RVDT) 508
differential line driver 370 dissociation of molecules
differential structure of measuring devices dissolved oxygen analyzers, specifications
285 of 517
differential transformers 387 distance measurements, off-line 462
- linear variable (LVDT) 506 distance sensors
- rotary 588 - laser 469
diffusion - multicoil inductive 448
- in IC technologies 125ff. - ultrasonic 450, 469
- mask 128 distortion 21, 23
Index 609

distributed processing 365 electronic phenomena, model equations for


distribution functions 35 49ff.
dopant elements in silicon 127 electronic sensors of radiation 567
Doppler effect 413, 566 electrooptic effects 273
- ultrasonic 570 emission monitoring 519ff.
Doppler flow measurement systems 566 encapsulants, solid state 90
Doppler mode encapsulation
- of radar detectors 415 - failures 93
- of ultrasonic detectors 413 - of active components 89
Doppler shift 548 - of sensor packages 93
dosimeters, personal 536 encoding
drift 22, 342 - 4-20 mA 345
- difussion equations, classical 52 - analog 333, 346
- mobilities 51, 198 - frequency 347
drive-in 126 - frequency difference 346
dual in-line packages (DIP) 85, 139 - Manchester 371f.
see also: packages - parallel digital 346
duplex assembly 500 - formats, binary 371
dynamic corrections of sensors 307 end instrument 3
dynamic pressure 587 energy-dispersive X-ray analysis (EDX) 259
dynamic response in automated production pro- energy production 493-527
cesses 471 engine control system 397
enhanced performance architecture (EPA) 368
environmental monitoring 529-554
- global 547
e-beam lithography 114, 117 - system, global (GEMS) 530
echo profile 451 environmental parameters 25 ff.
eddy currents 403, 410 environmental protection 476
Einstein relations 51 epitaxial region 213
electrical overstress 36 epitaxy 151
electrical signals 5 f. epoxy 93
electric bridge, constant-current 391 error band 19, 23
electric polarization, spontaneous 274 error bar 19
electrochemical cells, array of 325 error checking in link transport mecha-
electrodes nisms 376
- biopotential 571 error function, complementary (erfc) 126
- Clark 573 error limits
- enzyme 574 - for resistance temperature detectors 502
electrolyte cell, zirconia 392 - for thermocouples 499
electrolytes, solid 259, 338 error measurements
electromagnetic far-field 536 - dynamic 19
electromagnetic interference (EMI) 239, 471 - static 19
electromagnetic near-field 536 errors
electromigration 41 - amplitude 472
electromotive force (emf) 402 - in distance measurements 449
electron-hole pair creation 236 - in signal digitization, quantization 292
electron energy loss spectroscopy (EELS) - linearity, in Hall elements 210
259 - measuring, of radar detectors 415
electron gas, two-dimensional (2DEG) 203 f. - measuring, of ultrasonic detectors 414, 451
electronic control for home appliances 426 f. - of robots, dynamic 472
610 Index

- of sensors, total 281 exhaust stream disposal 481


- overall 22 expansion measurements %Of.
- probability of 320 expectation values 35
- quantization 294 exponential distribution 35
- radiation 592 exposure assessment, total (TEA) 542
- relative, in gas concentration measurements exposure monitoring, ultraviolet 535
327 exposure to toxic agents, human 530
- temperature span- 26 extractive methods 480
- temperature zero- 26
etchants
- for dielectrics 119ff.
- for metals 119ff. Factory Instrumentation Protocol (FIP) 367,
- for noncrystalline films 119ff. 379, 510
- for silicon, anisotropic 145 failure
etch-back methods 155 - check 490
etched-through holes 163 - mechanism 40f.
etched hole geometries 141 - of ceramics, mechanical 257
etchers - probability 525
- barrel 123f. - signal, binary 490
- plasma 123f. Faraday constant 393
- reaction-rate limited 140 Faraday rotation 196
- reactive ion 123f. Fast Fourier Transform 342
etching 118ff., 210 fatty acid films 187
- anisotropic 67, 85, 140, 143ff., 163 feature space 321 f.
- diffusion limited 140 feedback circuits 197
- directional 140 - negative 286f.
- dry 122ff., 179 ferrimagnetic materials 196
- - etch rates for 122 ferroelectric ceramic materials 275
- electrochemically 146 ferroelectric effect 275
- isotropic 140ff. ferromagnetic materials 196
- nondirectional 140 FET see: field effect transistors
- plasma 89, 122ff.. 180 fettling 451, 461
- reactive ion (RIE) 122, 181 fiber optic couplers 228, 238f.
- selectivity of 125 - four port 238
- silicon 139ff. - fused biconical 239
- wet 118ff., 179, 182 - passive 242f.
etching profiles 140f. - polished silica block (PSB) 238
- anisotropic 144 fiber optic ring resonator 225
etch rates 119, 144 fiber optic sensors 10, 218, 512
etch stop 146ff. - accuracy of 232
- dopant dependent 146 - components of 236ff.
- electrochemical 146ff. - definition and classification 218 f.
- p-n junction 149 - extrinsic 218, 227, 232
EUREKA 510 - gyroscope 225, 244
EVA 422 - intensity modulated 218
evanescent field 238, 242 - intensity type 232
excitation control of smart sensors - interferometric 218, 223 ff., 244
333 f. - intrinsic 218, 221 ff., 232
exciting core 400 - lithium niobate components 242f.
exhaust gas 394 - magnetostrictive 9
Index 61 1

- non-interferometric 218, 221 ff., 232 flow direction measurements 594


- phase modulated 218 - sensors for 587
- sources of 232ff. flow-injection analysis 491
Fick’s Law 126 flow measurements 503 ff.
fieldbus flowmeters, turbine-type 596
- project, EUREKA 367 flow-rate measurements
- specification 379 - mass 595
- standard 379 - volume 595
field effect transistors (FET) flow-rate sensors
- chemical 338 - air 385
- - implementation 93 - automotive onboard 385
- - ion selectivity 93 - fuel 385
- - pH sensitivity 93 - ultrasonic 385
- - with micromachined package 98f. flow sensors
- fabrication 87 - air mass 439f.
- ion-sensitive (ISFET) 163, 190 - electromagnetic 570
- MOS 48, 93, 99, 101, 208f. - solid state 386f.
- - dual-drain 204, 208f., 212 flue gas, components of 518
- - split-drain 205, 208 flue gas analysis 518ff.
film cracking 91 fluorescence spectrum 231
film formation 109 fluoroscopy, low intensity 567
films forbidden band 234
- dielectric 152 force measurements 596 f.
- metal 132 force sensors 458, 469
- noncrystalline, etchants for 119ff. - array of 568f.
- thick see: thick films - for medical applications 568ff.
- thin see: thin films - thin film of 568
filtering formats, universal asynchronous receiverkans-
- analog 335 mitter (UART) 363
- signal 361 four-wire systems of thermometers 502
filters four-wire transmitter 508
- digital 342 Fourier transform 24, 342
- edge 457 fractional rational polynomial 309
- high-pass 337, 379 frame formats 374
- low-pass 342, 379 - BS6556 376
- polymer-membrane 535 - HDLC 375, 379
finite-box discretization 60 - PROFIBUS 376
- procedures 60 - TC 57 FT1,2 376
finite difference techniques 47 freak data 40
finite element codes 47 frequency compensation process 342
see also: numerical codes frequency response 24
finite element techniques 60 frequency shift mode
FIP (Factory Instrumentation Protocol) 367, - of inductive loop detectors 411
379, 510 - of radar detectors 415
see also: protocols - of ultrasonic detectors 413
flicker noise 22 frequency-to-digital conversion 291 ff.
flip-flops 294 frost sensors 435
float-type transmitter 507 f. fuel cells 338
floating element 102f. fuel injection 385, 398
floats 505 full-scale output (FSO) 19
612 Index

full-scale range 22 glass as IOC material 241


fused biconical coupler 239 graded index (GRIN) lens 236f.
see also: fiber optic couplers - quarter pitch 237
grain boundary phases 259
graphite-furnace atomic absorption spectropho-
tometry 542
gages 3 greenhouse effect 547
- diffused 388 grid triangulation
- strain; see: strain gages - Delaunay 58
gallium arsenide IlOff., 144 - regular 58
- as IOC material 241 grooves
galvanomagnetic effects 7, 200 - U-shaped 144
- longitudinal 198 - V-shaped 100, 144
- transverse 198 grounding 22
gamma distribution 35 guidance systems for vehicle drivers 420
gamma rays 532 gyro, resonant fiber optic (RFOG) 225
gyroscope 244
gas
- analysis
- - direct measurement type 519
- - in situ type 519
- - multi-dimensional 309 Hall cross 63f.
- analyzers 520 Hall devices see: Hall elements
- - specifications of 521 Hall effect 198f., 346
- chromatography (GC) 540 - analytical model of 63f.
- components, identification of 310, 329 - in bulk silicon plate 200
- concentration determination by search strate- - pressure sensors 297ff.
gies 311 Hall elements 63f., 197, 213, 398
- diffusion 392 - active layer of 204
- mixtures - GaAs 202
- - binary, analysis of 310 - integrated 47f.,210
- - identification of 329 - lateral 213
gas sensors (oxygen) 391 - silicon 207
- automotive engines exhaust 257 Hall field 198
- ceramic materials for 268 Hall mobility 61, 198, 204, 208
- cross-sensitivities of 309 Hall plate analysis 60
- exhaust, materials for 266 Hall plates 204, 298, 346
- Langmuir-Blodgett based 190f. see also: Hall elements
- materials for 268f. Hall scattering coefficient 198
- metal oxide 321, 326 Hall sensors 48, 199, 297ff.
- metal oxide semiconductor 316f. - comparison of materials for 203
- semiconducting oxide ceramic materials for 266 - MOSFET 74
- surface acoustic wave 191 Hall voltage 48, 197, 297f.
- thick film tin oxide based 173 - distribution 66
gates Hamming distance 375
- NAND- 290 handshake dialog 336
- NOR- 290 harmonic oscillators 289f.
gateway 373 hazard function 35
Gauss’ theorem 345 hazardous area 380
Gaussian distribution 35, 129, 322 hazards
germanium 110 - biological 536f.
Index 613

- chemical 538ff. impact plate 588


- type of 532 impact pressure 587
HDUJ frame formats; see: frame formats impedance matching 336
height measurements, vehicle 419 impedances of sensors 27
Hewlett Packard Interface Bus (HPIB) 366 implementation of sensing element 97
high-temperature burn-in 36 impulse response 24
high-temperature storage bake 36 impurity scattering, ionized 52
Hooge parameter 204 indirect methods of measurement 487 ff.
host system 332f. inductively-coupled plasma atomic emission
hot electron effects 40 spectroscopy 542
hot pressing (HP) 258 inductive sensors 292, 448, 469
- isostatic (HIP) 258 - multi-coil 448
- techniques 273f. - one coil 448
hydrazine 143, 146 - in welding technology 467
- overdosing with 517 industrial robots 460
- analyzers, specifications of 518 infant mortality 34, 36
- measurements 517f. information
hydrothermal oxidation 256 - biological 556
hygiene, industrial 543 - capacity 345
hysteresis 21, 23, 275, 403 - condensation of 335
- content 345
- destruction of 335
- encoding 347
IC see: integrated circuits - - serial binary 344
identification - physiological 556
- of single gas components 326 - processing 335
- of substances 319 - - optical 3
identification range for substances 320 - systems for vehicle drivers 420
IEC625 bus 366 infrared chromatographs 483
IEC TC57 FT1.2 frame format 376 infrared detectors 416
see also: frame formats infrared diodes 449
IEEE 802.4 network standard 379 infrared photometer 483
image acquisition 454 f. infrared sensors 415, 430f., 567
image binarization 457 - dielectric materials for 415
image enhancement 456 - pyroelectric 415, 431 f.
image processing 454, 456 ingestion 531
- chain code for 457 injection, percutaneous 556
- data reduction in 456 in-line sensors 480ff.
- incorrect interpretations in 471 input transducer 359
- circuitries 419f. in situ analyzers 522ff.
image representation 455 inspection
image segmentation 457 - automatic 317
image sensors 419 - workpiece surface 470
image translator, optical fiber bundle 228 instability 22
image transport in medical diagnosis, optical 227 instrumentation and control (I+C) 495
imaging, magnetic resonance 571 instrumentation processor 380
immunoassay techniques 540 instrumentation systems
immunochemical reagents 231 - distributed 338, 349, 356, 373
immunosensors, surface acoustic wave - medical 556
(SAW) 191 - traditional 345
614 Index

insulator breakdown 40 ionic conductors 252 ff.


integrated circuit (IC) 81 f., 339, 380 ionic contamination 40
application specific (ASIC) 339, 352, 380 ion implantation 125, 129ff.
compatible processing 86 ff. ion milling 122
fabrication 109ff. irrelevant components of source material
packages 85, 138 486
packages, standard 84 ISA consortium SP 50 510
processing isentropic exponent 581 f., 587f., 593
- post- 86ff. ISFET (field effect transistors, ion-sensi-
- pre- 86ff. tive) 163, 190
signal processing 284 see also: field effect transistors
technologies 107-139 ISO/OSI protocol models 368
integrated electronics, conditions of usefulness IS0 8802.4 network standards 373
83 isochronous phase distortion (jitter) 371
integrated magnetic sensors 175-215 isolation 22
integrated optics (10) 217, 239-245 - galvanic 363
- applications 243 f.
- materials 240ff.
- technology 239
integrated sensors 175-215, 239-245, 348f. jitter 371
intelligence unit, remote 349 Johnson noise 22
intelligent system 491 junction field effect 202
interconnection of on-chip circuitry 83
interfaces 355-382
- communications, design of 380f. Knudsen diffusion 393
- core/cladding 221, 238
- for sensors in production processes 472
- parallel 366
- physical 370ff. laminar velocity profiles 596
- RS232 363 Langmuir-Blodgett films 95, 183ff.
- RS422 370 - balances 187f.
- RS485 364, 371 - forming apparatus 187f.
- serial digital 362f. - materials for 184ff.
interfering components of source material - types of 190
486 Langmuir-Blodgett techniques 177
interferometer Langmuir style float pressure sensors 187f.
- Mach-Zehnder 223, 226, 242 Laplace transformation 23, 307
- Michelson 224, 226 laser
- optical fiber 197 - cutting 448, 468
- resonant ring 225 - diodes 232
- Sagnac 224, 242 - distance measurements 449
intermediate signals 11 - Doppler velocimetry (LDV) 595
intrinsic safety (IS) 379 - drilled holes 163f.
- distributed instrumentation 380 - heating techniques 158
inverters, cascaded 290 - scanners, preview 451
IOC see: integrated optical circuits - scanning 461
ion-selective electrode (ISE) membrane 94 - sensors, preview 464
- technology 96 - - scanner 469
ion beam technologies, focused 205 - - stripe 467ff.
ionic conductivity of zirconia ceramics 254 - trimming 210, 299
Index 615

layers, connection between 162f. long-term drifts 34


LB see: Langmuir-Blodgett long-term stability 34 ff.
lead wires 560 look-up table method for digital linearization
leakage 29 299
lean combustion system 394 loop inductance change 410
least squares method for pattern recognition loop structure of measuring devices 286f.
- partial 324f. Lorentz deflection 199
- transformed 326ff. Lorentz force 52, 196
LED see: diodes, light-emitting low level signals 357
k i t - und Informationssystem Berlin (LISB) low-pass filters 342, 379
421 low pressure reactors, hot wall 152
length measurements, vehicle 412 LPCVD (low pressure chemical vapor deposi-
level measurements 503 ff. tion) 152f.
level sensors see also: chemical vapor deposition, low pres-
- liquid 229ff. sure
- - multi-discrete 230f. LVDT see: differential transformer, linear vari-
- water 432f. able
light, linearly polarized 196
light channel 229
light detection and ranging (LIDAR) 548
light emitter 232 Mach-Zehnder interferometer 220, 223 f., 242
light propagation, bidirectional 226 machine diagnosis 458f.
limb movement studies 566 machine vision 454ff., 469
limiting current 393 ff. - applications 470
- oxygen sensors 392f. Mach number 581 ff.
line- and band-reversal method 590 magnetic detector, three-dimensional 212 f.
he-reversal pyrometer 591 magnetic effects, pertinent semiconduc-
linearization 285 f. tor 197ff.
- methods 299 magnetic field effects, model eqations for 52f.
linear regression method 324 magnetic field sensors (MFS) 10, 196-215, 418,
linear systems, dynamic corrections of 307 437
link layer transport mechanism 369 - bipolar 48
link level frame 379 - design of 207f.
link transport mechanism 373 f. - design selection criteria for 202
liquid chromatography 491 - integrated 52, 195ff.
liquid phase epitaxy 111 - magnetooptic 196
liquid water content (LWC) 598 - magnetostrictive fiber optic 10
LISB, k i t - und Informationssytem Berlin 421 - numerical modeling of 47f., 63ff.
lithium niobate - optoelectronic 196
- electro-optic modulation coefficient 241 - semiconductor 197
- as IOC material 240f. - - examples of 204ff.
- components in I 0 technology 242f. - - materials for 202ff.
lithography 114ff., 210 - - technologies 202ff.
- pattern transfer 116 - silicon
- photo- 80, 87, 241 - - examples of 208ff.
live zero current 358f. - - technologies 207ff.
load cell 568, 597f. - - technologie limitations 210f.
log-normal distribution 35 - smart 346ff.
- model 37, 39f. - technologies 196f.
logic link control 369 - thin-metal film 196
616 Index

- three-dimensional vector 212 - for integrated optical circuits 240ff.


- wire 196 - for Langmuir-Blodgett films 184f.
magnetic sensor elements, arrays of 212 - for semiconductor magnetic field sen-
magnetic signals 5 f. sors 202ff.
magnetocardiogram 572 - for semiconductor photodetectors 234
magnetoconcentration 197 - for sensors, ceramic oxide 247ff.
- effect 201 - for sensors, electrical conducting 252ff.
magnetodiodes 197, 200 - for thick films 171
magnetoencephalogram 572 - for thin films 172
magnetoresistance 197, 199, 206 - high-permeability 196
magnetoresistive (MR) switching 196 - insulating 271ff.
magnetostrictive jacketing materials 197 - insulating, perovskite-type 271 ff.
magnetotransistors 52, 197 - low-permeability 196
- bipolar 200 - magnetic 196
- - mesh generation for 58 - magnetostrictive jacketing 197
- dual collector 65 - packaging 84
- - simulation geometry of 66 - piezoelectric 271 ff.
- noise in 211f. - radiation sensitive 114
- simulation geometry of 65 - raw, for beta-alumina 263
- suppressed-sidewall injection (SSIMT) 209 f. mavericks 303
Manchester encoding 371 f., 375 Maxwell-Boltzmann statistics 49
manipulation, sensor controlled 468f. MBE (molecular beam epitaxy) 111f., 204,
manufacturing 241
- automated 445-474 measurand range 19
- message service (MMS) 377 measurands 3, 9, 18
- tolerances 308 measurement
MAP (manufacturing automation proto- - acceleration 598
col) 368, 373 - air flow 387
see also: protocol, manufacturing automa- - anatomical structures 564
tion - angle of proximal interphalangeal joint 565
martensitic phase transformation 253 - apical impulses of the heart 566
mask - biopotential 571
- fabrication 114f. - blood flow 566, 569
- master 115 - blood pH 562
- reticle 115 - blood pressure 569f.
mass, seismic 287 - body organ pressure 570
mass spectrometry (MS) 540 - central arterial pressure 570
master message 364 - computer aided 339
master/slave mode of operation 365 - contact 429
master station 374 - contour see: contour measurements
materials - distance 447ff.
- 111-V 197, 199, 234 - electromagnetic variables 571 f.
- amphiphilic 184 - flow direction 587
- ceramic - force 596f.
- - ferroelectric 275 - fuel flow 387
- - for gas sensors 266ff. - geometric variables 564
- - for humidity sensors 269 - hemodynamic variables 569
- - pyroelectric 274 - hydrocarbon 542
- for biomedical sensors 556ff. - illuminance 292
- for Hall sensors, comparison of 203 - in clinical laboratories 561
Index 617

- intracardiac pressure 570 - oxygen permeable 573


- kinematic variables 565 f. - polymer based 96
- light intensity 449 - solid state 95
- multi-period 292f. meshes, nonrectangular 60
- non-contact 429 mesh generation for numerical sensor model-
- oxygen 392 ing 57ff.
- patient weight 568 metal-insulator-metal structure (MIM) 27
- position 451 metal-insulator-semiconductor (MIS) 27
- pressure see: pressure measurements metalization
- production process parameters 458 f. - chemical vapor deposition 136
- profusion, non-invasive 571 - multilevel 153
- radiation 533, 567 - in IC technologies 132ff.
- sound 567 metallo-organic deposition (MOD) 178f.
- spatial seperation of 487 metal oxide chemical vapor deposition 241
- speed see: speed measurements metal oxide semiconductor
- strain 596f. - inversion layer 208
- temperature see : temperature measurements - ring oscillators 290
- temporal seperation of 487 - technologies 208, 290
- test 323 metal TO-5header 85
- thrust 597f. Michelson interferometer 220, 224f.
- time domain 211 microbend sensors 218, 221 f.
- velocity see: velocity measurements microbiological analysis 538
- venous pressure 570 microbridges, polySi 182f.
measurement space 320 microelectronic systems 280, 291, 299, 459
measurement techniques, problems of 471 microfabrication technologies 80
measure of stream thrust 582 micromachining 80, 89, 107ff.
measuring junction of thermocouples 498 micromechanical structures, integrated sil-
measuring systems, gas concentration 309 ff. icon 56
measuring technique, model-based 489 microprocessors 299
mechanical effects, model equations for 53 microprocessor systems 459
mechanical mismatch 557 microsensors
mechanical quantities, measurements of 308 - biomedical 80
mechanical shock 37 - calibration of 81f.
mechanical signals 5 f. - chemical, design of 97
mechanical structures, etching of 341 - coatings of 89
- model eqations for 56ff. - corrosive environment of 83
media access control 374 - design of 80
medical sensors 227, 555-577 - encapsulation of 81
- considerations for 558ff. - examples of 97ff.
- home applications of 564 - floating element, package fabrication flow
- non-contacting 556 of 103
- non-invasive 556 - implementation of 538
- requirements of 556ff. - in-vivo pressure 99f.
see also: biomedical sensors - layout of 84
membrane deposition 96 - magnetic 205
membranes - modeling of 46ff., 57ff., 74, 199
- adhesion of 96 - packaging constraints of 89ff.
- biospecific 575 - packaging of 80
- liquid ion exchange 98 - semiconductor 46
- n-type silicon 150 - semiconductor magnetic 196
618 Index

- shear force, floating element 101ff. multivibrators 288


- thin film tin oxide 181f. NASICON 264
microsensor systems, design of 80 f. - ceramics, fabrication of 264
microstructures, silicon-based 139 - electrical conductivity of 264
microwave ovens 429ff., 536 - single crystals, fabrication of 264
microwave signals 415 - structure of 264
MIL.STD.1553 bus standard 367, 379 navigation equipment for vehicles 420
minimum detectable signal (MDS) 20, 23 negative temperature coefficient (NTC) 269 f.,
minority carriers. 275
- mobilities 53f. Nernst equation 250, 522
- diffusion constant of 295 network, star shaped 360
MOCVD (chemical vapor deposition, metallo- network topology 363
organic) 111f. Neumann boundary conditions, homogeneous
see also: chemical vapor deposition 50
model equations neural networks 471
- for electronic phenomena 49ff. neutrons 532
- for mechanical structures 56ff. noise 20, 22, 335, 341f.
- for sensor signal processing, physical - I/f 22, 204, 211, 342, 348
295 ff. - Barkhausen 196
modeling of sensors 45-77 - coherent forms of 342
modem standard EIA RS232 362 - external 22
modifiers 7 - immunity of sensor systems 339
modularity, lack of 82 - industrial 29f.
modulated light 419 - internal 22
modulating principle 9 - intrinsic 211
molecular beam epitaxy (MBE) lllf., 204, - magnetotransistors 211 f.
241 - recombination-generation (r-g) 22
monitoring, - spectra of magnetotransistors 211
- biological 545ff. - thermal 22
- environmental see: environmental moni- noise theory, quantization 335
toring non-dispersive infrared absorption (NDIR) prin-
- of chemical variables 563 ciple 520
- of physical variables 563 non-linearity 20 f.
- patient 555ff. - of sensors 341
- personal 543 - of substance characteristics, differences
monoclonal antibodies 538 in 320
monolayers 184 non-return to zero (NRZ) 371
MOSFET see: field effect transistors nose, electronic 316
MOS see: metal oxide semiconductor NRZ waveform 372
motion sensors, translational 227 ff., 232 nuclear magnetic resonance magnetometry 202
multi-component measurement 487 nuclear waste, storage of 533
multi-period measurements 292 numerical codes
multi-sensor systems 15, 309, 313-330 - ALBERTINA 48, 63
multidrop bus see: bus, multidrop - ANSYS 48, 12
multifunction integrated optical circuits - PC-ID 47, 63
(IOC) 240 - SENSIM 48
multimaster access method 374 - SPICE 74
multiplexed wiring 356, 360 numerical microsensor modeling 46ff., 57ff.,
multiplexing capability, analog 359 74
multiplexing circuits 82 Nyquist plot 472
Index 619

obstructive components of source material - ring 290f.


486 - voltage controlled (VCO) 347
occupancy signals, vehicle 408, 418 f. OSI (open system interconnect) 368, 373
off-line contour measurements 451 output 3
on-line contour measurements 451 - buffer amplifier 349
on-line sensors 480ff. - impedance 22
open-loop control 495 - transducer 359
open circuit potentials (OCP) 149 outstation 374
open system interconnect (OSI) concept 368, overall performance 22
373 over relaxation, successive 62
operating range 22 oxides 247-278
optical detectors 418 oximeter, pulse 574
optical fiber 217-245, 338 oxygen, dissolved, measurements of 516f.
- bundels 227f. oxygen analyzers 522f.
- coatings 220f. - specifications of 523
- coils 196 oxygen sensors
- detectors 234ff. - limiting current 392f.
- forms 219 - thin-film limiting current 395
- image translator 227 - titania 392
- interferometer 197 - zirconia based 254f.
- loop 225
- losses 219
- modes 219f., 234
- plastic 219 p-n junctions 232, 234f., 295
- polarization preserving 219 packages 89- 103
- sensitivity 226 - biomedical sensors 558f.
- sensors 220ff. - ceramic 83, 138
- T-junction 338 - dual-in-line (DIP) 85, 139
- waveguides 219, 240 - fabrication 103
see also: fiber optic - hermetic 99, 137
optical image transport 227 - in IC technologies 83f., 138f., 155
optical losses in IOCs 240 - sensor, encapsulation of 93
optical radiation, model equations for 51 f. packet assembler/disassembler (PAD) 375
optical sensors 419 packet formats see: frame formats
- frost 435 pancreas, artificial 563
- medical applications of 574 parabolas, cubic 301
optical time domain reflectometers (CYTDR) parallel structure of measuring devices 284
223 paramagnetic materials 196
optics, integrated 239-245 paramagnetic oxygen analyzer 483
opto-electronic integrated circuits 241 parameters of sensors 17-43
opto-electronic sensors 449, 471 - environmental 25 ff.
orifices 594 - standard 13
oscillating elements 346 parasitic carriers 210
oscillator circuits, ring 290 parasitic effects of sensors 281
oscillators PARC see: pattern recognition
- Franklin 289 parity checking 336
- harmonic 289f. particulate measurements 524
- magnetically controlled (MCO) 208 - specifications of 524
- precision 289 partitioning 82f.
- relaxation 288 passivating potentials (PP) 149
620 Index

passive transduction principles 9 phthalocyanine ring 187


patient monitoring 555 ff. pickup 3, 341
pattern delineation 125 - electromagnetic 398
pattern recognition (PARC) 315f., 454, 457 - core 400
- comparison of methods 329 piezoelectric
- correlation method 316ff. - ceramics 274
- partial least squares (PLS)method - device 385
324f. - effect 271
- statistical 457 - sensors 414, 435, 512
- transformed least squares (TLS) method piezoresistive effect 53 ff., 72, 388
326ff. PIN photodiodes 234f., 439
- ultrasonic 451 see also: photodiodes
- vector method 320ff. piping 40
- with chemosensors, methods for 316 pitot tubes 582, 584
pattern transfer procedure 118 pivoted vane 588
PECVD (chemical vapor deposition, plasma- pixels 227, 455
assisted) 177 plant safety 476
see also: chemical vapor deposition pltisma etching 89, 122ff., 180
peer to peer exchange 368 plausibility check 490
performance monitoring 481 pneumatic probes 590
performance sensitivity 47 Poisson equation 47
permeability materials polarization, spontaneous 271 f.
- high- 196 pollutants, air 540
- low- 196 - instrumentation for 541
perovskite-type oxides 269f., 271 ff. pollution control 518
phase equilibria in solid solutions of zirconia polygon interpolation method 299 f.
254 polyimide ribbon cables 85
phase modulator, integrated optical 243 polymer matrix 95
phase shift mode of inductive loop detectors polynomial interpolation method 300
411 polyvinylidene fluoride (PVDF) 274
pH measurements 515 positive temperature coefficient (PTC) 271
phosphosilicate glass (PSG) 161, 179f. powder preparation 256
photodetectors power
- p-n junction 235 - distribution in instrumentation systems
- semiconductor 234f. 379f.
- vacuum tube 234 - plant, measurements in a 496ff.
photodiodes 51 f. - sensors 458
- analysis of 60 - spectral density (PSD) 211 f.
- avalanche (APD) 234, 236 - train 399
- arrays of 349 Poynting vector 238
- GaAs 47 Prandtl design 584
- numerical modeling of 47, 62 precision of sensors 19
- PIN 234f., 439 pressure measurements 387, 503 ff., 594f.
- silicon 47 - absolute 507
photolithography 80, 87, 241 - average 586
photometers 481 - dynamic 586, 597
photoresists 116, 182 - impact 587
- sensitivity of 116f. - static 581, 584, 594
phototherapy techniques 567 - total 581, 593
phototransistors (PT) 433 pressure profiles, atmospheric 548
Index 621

pressure sensors 290, 387, 586 - host 344


- absolute- 390 - pipeline 456
- biomedical 559 - smart sensor 335
- capacitive 68, 390 product indicator 478
- - numerical results for 71 production processes
- diaphragm-type 433 - sensor-controlled 460
- flush-mounted 586 - parameters, measurements of 458 f.
- Hall-effect 297f. PROFIBUS 366f., 379, 509f.
- - physical model of 298 see also: bus
- - temperature dependence of 305 - frame formats 376
- Langmuir style float 187f. see also: frame formats
- micro-diaphragm 349 programmable read only memory
- miniature 162, 570 (PROM) 84
- negative- 390 PROMETHEUS 422
- numerical modeling of 48 protocol
- physical model function of 298 - application data 377f.
- piezoresistive 9f., 68ff., 85, 91, 99, 283 - British Coal’s Application 378
- - sensitivity of 94 - BS6556 367, 378
- semiconductor 387 ff. - changes only 360
- silicon 439, 170 - communications 368 f.
- - capacitive 164 - - multi level 368
- total- 582 - HART 350, 365
- Wilhelmy plate 187f. - HDLC 367
pressure transmitter 509 - ISO/OSI 368
- differential 506 - manufacturing automation (MAP) 368,
pressure vessel 504 373
preview sensors, optical 453 - message passing 364
primary signals 11 - SDLC 349
printing, projection 116 - Simple Asynchronous (SAP) 376
printing, shadow 116 - Southampton Transducer (STP) 364
PRO-CAR 422 Pt-Rh alloy 590
PRO-CHIP 422 pulsed light 419
PRO-NET 422 pulse mode of ultrasonic detectors 413
PRO-ROAD 422 pulses, vehicle 408
probability density function 34, 322 purple plague 137
probability function 35 pyroelectric
probe cooling 592f. - ceramic materials 274f.
process - effect 274
- analyzer 480 pyrometer 592
- control 475-492 PZT (PbZr03 - PbTiO3) solid solution
- indicator 478 272 f.
- instrumentation 485f. quality assurance, automatic for welded junc-
- process mass spectrometry 487 tions 462
- monitoring 495 quarter pitch lens see: graded index lens, quarter
- parameters 477 pitch
processes, decentralized 280 quartz 271
processing of signals 279-330 - tuning-fork 292
processors - accelerometers, micromachined 287
- bit-slice 456 - quartz crystal 390
- central 335 see also: piezoelectric
622 Index

R-LC circuits 289 resonant fiber optic gyro (RFOG) 225f.


radar, microwave 548 respiration monitoring 563
radiant signals 5f. response characteristics, isotropic 536
radiation responses of sensors, linear 314
- ionizing 532ff. RFOG see: resonant fiber optic gyro
- non-ionizing 535 f. rice cookers 435
- hazards 532ff. ring balance manometers 505
- nuclear 533, 567 ring network 363
radiography, digital subtraction 567 Riston polymer, encapsulating 96
radionuclides, inhaled, biological effects of Riston wells 96
533 Ritz strain-energy method 67
radon contamination, indoor 533 road traffic management 408
Raman scattering 222 robot grippers 450, 568
Raman spectroscopy 591 - multi sensor 469
reactants, organometallic 154 robotic applications 53
receivers, microwave 415 - tactile sensors for 349
recombinant DNA research 537 robot systems 471
recombination-generation (r-g) noise 22 robot vision systems 454
rectified signal 402 rochelle salt 271
refractive index 236 ROM 299
- cladding of fibers 219 root-mean-square 20
- core of fibers 219 rotary piston 505
- core to cladding 227 rotary variable displacement transformer (RVDT)
- graded 237 508
refrigerators, electric 434 f. RS232 interfaces 362, 370
regression methods 302 ff. RS422 interfaces 370
relaxation oscillators 288 f. RS485 interfaces 364, 371
reliability 'RTD see: resistance temperature detectors
- models 35f. Rubylith 114
- of implantable sensors 559
remote control system 439
remote operator 80
remote sensing 547 ff. S-NET 367
repeatability of sensors 22 safe area 380
resins, polymeric 532 Sagnac interferometer 224, 242
resistance sample handling 480f., 519
- bridge 320 sampling
- change, electrical 392 - emission-stack 542
- strain gage, foil type 596 - environmental 531
- thermometers 481, 498, 502f., 589 - - discrete 532
-
- silicon 292 - criteria for aerosols 544
resistive wiring stubs 363 - methods, personal 543
resistivity, grain 269 SAP see: protocol, Simple Asynchronous
resistivity of thick film cermets 174 316
resistors SCADA (supervisory control and data acquisi-
- ceramic PTC 271 tion system) 359
- magnetic field sensitive 418 scaling of miniaturized sensing elements 94 f.
- thick film cermet type 173 Scharfetter-Gummel approach 61
resists 114, 116ff. Schottky barrier 50, 269
resolution of sensors 19 screening 36f.
Index 623

screen printing - definition 4


- silk 340 - failure 559
- of thick films 175 - ideal 323
scribe-and-break techniques 136 - non selective 323
sealing - oscillating 346
- compression 91 - primary 333f., 337f.. 341, 346
- eutectic 91 - - time response 342
- hermetically 91 - vibrating 338
- hermetic low temperature 161 sensors
seam tracking 463 - aerospace application 579-602
seam welding processes 451 - amperometric 574
secondary-ion mass spectrometry (SIMS) - amplitude analog 291
259 - angle of attack 588
Seebeck coefficients 51 - antigenlantibody see: antigen/antibody sen-
Seebeck effect 499 sors
selective catalytic reduction (SCR) pro- - arrays of 314, 316, 324f., 329, 343, 349, 545
cesses 521 - - tactile 455
selectivity - automotive onboard see: automotive onboard
- of sensors 20 sensors
- of sensor systems 323 - autonomous see: autonomous sensors
self-generating transduction principle 9 - bioanalytical 574f.
self identification 361 - biomedical see: biomedical sensors
semiconductors 266, 560 - calibration of see: calibration of sensors
- diaphragms 388 - capacitive see: capacitive sensors
- integrated circuit technologies 107ff. - carbon monoxide 328
- light sources 234 - ceramic humidity 430
- magnetic effects, pertinent 197ff. - ceramic oxide see: ceramic oxide sensors
- magnetic field sensitive 418 - ceramic piezoelectric 450
- n-type 266 - characteristics of see: characteristics of
- p-type 266 sensors
- processing of 536 - characterization of see: characterization of
- single crystal 110 sensors
- sensors 387, 390 - chemical see: chemical sensors
-
- integrated 440 - classification of 8 ff.
- - modeling of 46 - compound 337
sensing element compatibility, technologi- - concentration cell type 393
cal 9Sff. - consumer equipment 428
sensing principles in chemical process engineer- - contact 448
ing 481ff. - corrections of, dynamic 307
sensing systems, optical 564 - defects 341
sensing technologies, chemical 92 - definitions and standards 2ff.
sensing windows 89 - design 79-89
sensitivity 19, 295 - dew 437
- magnetic 208 - dew-point 405
- power-related 203 - direct-reading 540
- of resists 117 - displacement see: displacement sensors
- supply-current-related 203 - distance see: distance sensors
- total 285f. - electrical conductance 573
sensor elements 92ff., 316 - electrochemical see: chemical sensors
- defective, detection of 320 - elementary 4
624 Index

- encapsulation of 90 - modeling of 45-77


- energy production 493-527 - motion 227ff., 232
- environmental monitoring 529-554 - multi- 82, 313-330
- fabrication 86ff. - multi-dimensional 316
- features of 281 - non-contacting 429
- fiber optic see: fiber optic sensors - non-contact position 398
- flow-rate see: flow-rate sensors - non-selective 316
- flow see: flow sensors - - transfer characteristics of 324
- force see: force sensors - olfactory 441
- frequency analog 291 f. - optical see: fiber optic sensors and optical
- frost 435 sensors
- gas see: gas sensors - oscillators 288 ff.
- glucose 563 - packages see: packages
- Hall see: Hall sensors - parameters see: parameters of sensors
- heated wire-type 386 - photo 432
- historical remarks 2 - porous coating type 394
- home appliances 425 ff. - position 588
- humidity 405 - potentiometric 574
- - ceramic materials for 269 - pressure see: pressure sensors
- - FET 440 - preview 451
- hydrogen 316 - process control 475-492
- hypersonic flight 594 - production processes
- imaging, solid state 549 - - market trends of 473
- immuno- 190 - - applications of 473
- inductive see: inductive sensors - - automated 447-474
- infrared see: infrared sensors - - current problems of 471 ff.
- in-line 480ff. - - classification of 447ff.
- in situ 480 - profusion, thermal 570
- integrated 175-215, 239-245, 348f. - proximity 450
- intelligent 332, 361, 403, 440f. - radiation see: radiation sensors
see also: smart sensors - redundant 524
- interfaces for 355-382 - resistance of 321
- intrinsic 338 - rinsing 433
- ion 190 - rotation 224
- joint angle 564f. - sales of 11
- kinematic 566 - semiconductor see: semiconductor sensors
- laser see: laser sensors - sensitivity of 403
- level see: level sensors - silicon 338
- light intensity 567 - single shot 561
- linear responses of 314 - smart see: smart sensors
- load 433f. - sound 567
- magnetic see: magnetic field sensors - supersonic flight 594
- market for 12ff. see also: detectors
- mechanical, numerical modeling of 48 f., 67 ff. - tactile see: tactile sensors
- mechanisms, primary 336ff. - temperature see: temperature sensors
- medical see: medical sensors - thin-film type 395f.
- metal oxide, transfer characteristics of 321 - torque 399ff.
- methan 326 - - intra-bearing 400
- micro- see: microsensors - - magnetoresistive 400ff.
- microbend 218, 221f. see also: force sensors
Index 625

-total-sunlight measuring 405 - pattern of substances, characteristic 316


-ultrasonic see: ultrasonic sensors - processing 279-330, 481, 566
-vehicle see: detectors, vehicle - - additive 402f.
- velocity 565f. - - analog 280
-vibration 511, 598 - - - structural limits of 282ff.
- voice recognition 441 - - circuitry 207
- X-ray 567 - - digital 280
- zirconia, oxygen 392 - - examples of 308ff.
sensor systems - - hybrid 280
- basic structure of 479 - - multisensor 313-330
- definition of 4 - - on-sensor 559
- digital 307 - - sensor-specific 280
- dynamic parameters of 307 - - subtractive 402f.
- extractive 482 - - units, analog 283
- first order 307 - synthesized 402
- for mechanical quantities, inductive 308 f. - to-noise ratio 204, 319
- in automated production 447ff. - transmission system, analog 356
- in chemical industries 486 - zero torque 403
- inductive, influence corrections in 308 silica analyzers 515
- internal structure 480 silicate glass (SG) 179
- microcomputer-oriented 308 silicon IlOff., 337
- networks 362 - CVD technologies 152f., 155
- requirements for 485ff. - diaphragms 292
- second-order 307 - dioxide 93
- selectivity of 323 - dopants, diffusion coefficient for 128
- self-checking 490 - elastomer 569
- smart, design equations for 345 - epitaxy 151
- structure of 479ff. - etchants for 143
Shockley-Read-Hall process 52 - formation, porous 88
shock wave 584 - integrated optical circuit material 241
short-term drifts 34 - inversion layers 208
shot noise 22 - isoetch curves for 142
Sieger bus 365 - membrane 87
Siemens Ultramat 32 analyzer 520 - micro-element 386
signal(s) - micromachined 338
- amplification, on-chip 99 - nitride 93, 153f.
- analog 282, 360 - - coating 99
- conditioning 361 - - layer 389
- conversion - oxide layer 389
- - analog-to-digital (A/D) 280 - POIY- 112, 151
- - frequency-to-digital (F/D) 291 ff. - sensitivities 341
- degradation 370 - sensors, modeling of 46
- digital 282, 360 - transistors 295
- domains 6 similarity index 316
- filtering 361 single chip solution 340
- forms 281 f., 288f. single crystal growing 110
- - characteristics of 282 sinter-HIP technique 258
- frequency 294 sintering 175, 258
- modulation technique 373 - liquid-phase 269
- pattern, normalized 319 skull melting process 255
626 Index

slave address 364f. standard deviation 35


slave to slave communication 367 star network 363, 336
smart sensors 83, 331-353, 361 - topology 333
- design 344ff. static temperature
- enabling technologies for 339ff. status information 378
- evolution of 352 steam analysis 512ff.
- in microelectronic systems 280 step response 24f.
- industrial progress of 349ff. sterilization 559
- magnetic 346ff. stiffness matrix 60f.
- sub-systems of 333f. STP (Southampton Transducer Protocol) 364
smart transmitter 508 see also: protocol
Sobel-operator 457 strain gages 356, 458, 564f., 596f.
sodium analyzers - bridge 343
- operating principle 515 - elements 284
- specifications 516 - resistance 587
sodium measurements 515 f. strain measurements 596f.
sol-gel ceramics 264 stress
sol-gel processes 256 f. - in thin films, residual 89
solar cells 51, 155 - thermal 56
- analysis of 60 - distribution 69
- numerical modeling of 47 structure of measuring devices 282 ff.
solid state membranes, scaling of 95 superconductor quantum interference devices
solution, solid 254 (SQUID) 572
solution procedures for numerical sensor model- - magnetometers 197
ing methods 62 superlattice structure AlAsIGaAs 204
span of sensors 22f. superposition law 324
spatial resolution of 3-D detectors 212 superposition plane, artificial linear 326
specificity of sensors 20 supersonic flow of a perfect gas 581
spectral scanning method 591 supervisory control and data acquisition
spectrometers, radiation 535 (SCADA) system 359
spectroscopy suppressed-sidewall injection magnetotransistors
- NIR 325 (SSIMT) 209f.
- Raman 591 surface-mounted design (SMD) technology
spectrum analyzer, narrowband 536 290
speed measurements surface acoustic wave (SAW) sensors 190f.
- aircraft 593 surface charge spreading 40
- vehicle 408, 411 surface silanization techniques 96
spin-on coatings for microsensor encapsulation system controller 359
91 system integrity check 378
spinal cord injury 565, 568
spline functions 302
springlmass system 511
sputtering 122, 133ff., 176 tactile sensors 568
- DC 134 - arrays of 469
- magnetron 135f. - for medical applications 568ff.
- reactive 134 - for robotic applications 349
- RF 134f. tailoring processes 343
SQUID see: superconductor quantum inter- tape automated bonding (TAB); see: bonding,
ference devices tape automated
stagnation pressure 582 technologies, 111-V 203
Index 621

technology incompatibility 96 - barewire 591


telephone line, balanced, application of 412 - dynamic characteristics of 282
temperature - exhaust-duct 592
- coefficient 283 - types 499f.
- compensation 309 thermodynamic variables 477
- drift 210 thermoelectric effect 50
- total, of flowing gas 589 thermography 567
temperature effects thermometers 502
- correction of 283, 296, 305, 308 - platinum resistance 591
- model equations for 50f. - transistor 295f.
- simulation of 72 - resistance see: resistance thermometers
temperature measurements 2, 295, 498 ff., see also: temperature sensors
594 thermomigration of aluminium colmn
- body 569 164f.
- optical method 590 thick films 169-194
- static, of hot gas 590f. - cermets 174
- in flowing gas 588f. - devices design 182
- of cryogens 591 - fabrication processes for 170ff.
- of solid objects 591 f. - flow chart for preparation of 173f.
temperature monitoring, body 563 - materials for 171 f.
- medical 276 - screen printed 175
temperature profiles, atmospheric 548 - shape manipulation of 340
temperature-sensitive element (TSE) 589 thin films 169-194, 395
temperature sensors 295, 336, 588 - characterization methods for 178
- criteria for 589 - delineation techniques for 179ff.
- distributed 221 - deposition methods for 176f.
- integrated 51 - fabrication processes for 170, 176ff.
- miniature 556 - - compatibility issues 182f.
- radiation 556 - ferromagnetic 207
- silicon 308 - materials for 172
see also: thermometers - tin oxide microsensors 181f.
temperature stabilization of APDs 236 thin-plate theory, thermoelastic 56 f.
temperature transmitter, 3044 350f. threshold of sensors 20
terminal-based straight line 20 thrust measurements 597f.
TGS812, Figaro, single sensor element 326f. time occupancy 409
thermal tin dioxide 268
- emission 415 titania 266f.
- evaporation 176 TLS234 sensor array 328
- noise 22 token passing system 374
- shock 37 tolerances, manufacturing 308
- signals 5f. torque sensors see: sensors, torque
- stress analysis 46 tomography, computerized 567
- transient effect 289 total exposure assessment (TEA) 542
thermistors 575 traffic
- ceramic see: ceramic thermistors - control 407-423
- resistance-temperature characteristics of - flow, mean 409
276 - light phase 417
thermocouples 334, 498ff., 536, 589 - measurement engineering 408 f.
- assembly 500 - parameters
- connection 500 - - characteristic 408f.
628 Index

- - higher level 418 UAF3 (universal asynchr. receiver/transmitter


- - measurements of 412 formats) 363
- queue detection 409 ultrasonic sensors 414, 566f.
- surveillance 407-423 ultrasound surface acoustic waves (SAW) 575
- volume 409 undercutting in IC fabrication 119
transceiver chip 371
transducers 8
- basic 337
- definition of 3 vacuum evaporation 132f.
valence band 54, 234f.
see also: sensors
vapor phase epitaxy 111
transduction mechanism in microsensor design
92ff. variance 35, 322
varistors
transduction principles 9 f.
- chemical 5ff. - breakdown 269
- zinc oxide based 259, 269
- physical 5ff.
vector method for pattern recognition
transfer characteristics
- non linear 314, 317, 329
320ff.
- of chemosensors 315
vehicle detectors see: detectors, vehicle
velocimetry, laser doppler (LDV) 595
transfer function 23
- of negative feedback circuits 287
velocity measurements
- bulk 595f.
transfer of parallel data bytes 366
transformation toughening, stress induced
- fluid 594f.
- light 415
258
transistors
- local linear 594f.
- bipolar 53, 339 - sound in air 413
velocity sensors 565
- field-effect see: field effect transistors venturi 584
- magnetic field sensitive 199 vibrating element 338
- magneto- see: magnetotransistors
- p-n-p silicon 398f.
vibration measurements 5lOff.
- split drain 346
viscoelastic properties 478
viscous damping of a micro-mechanical element
- thermometers 295 f. 94
transmission electron microscopy 259
vision systems
transmission factor, total, for loop
structures 287
- problems for 471
transmitter connection 508
- robot 454
vision teach 454
transmitters 3
- 4 to 20 mA 359
visual display unit (VDU) 365
volatile organic compounds (VOC) 541
- microwave 415
- two wire 508
travel direction measurements, vehicle 411
Travelpilot 422 wafers 111
triangulation principle, optical 449, 451 - bonding 99, 160
triglyzine sulfate (TGS) 274 - level encapsulation 89
troughs - preparation of 1lOff.
- constant perimeter 188 - quartz, crystalline 287
- Langmuir 188 - silicon 210
tuning-fork quartz 292 - - flat locations on 111
turbidity measurements 518 - - material properties for 112
- specifications of 518 - - oxidation of 112ff.
turbulent velocity profiles 596 - - specifications for 112
Index 629

washing machine, automatic 432 ff. world market for sensors 12ff.
water wraping, substrate 91
- analysis 512ff.
- monitoring 538ff.
- pollution 538
waveguides 239f. X-ray lithography 114, 117
see also: optical fibers X-rays 532
weak field limit 52
wear-out 27f.. 34
weather prediction 547 Y-junction 242 f.
Weibull distribution 35
welding
- arc 462ff.
- - electric- 535 Zener barrier 380
- - fully mechanized 464 Zener diodes 289
- cold 162 zeolites in hydrogensensors 316
- grooves, scanning of 458 zero shift 34
- metal-active gas (MAG) 463 zero torque signal 403
- metal-inert gas (MIG) 463 zinc dioxide, wurtzite structure of 268
- robots 462ff. zinc oxide 268
- spot 462ff. zirconia 252ff.
- torch, oscillating 458 - crystalline structure of 252
- tungsten-inert gas (TIC) 463 - - cell 394
well formation 99 - ceramics
Wilhelmy plate pressure sensors 187f. - - fabrication of 256f.
window technologies for chemical sens- - - ionic conductivity of 254
ing 84f. - crystals, cubic, applications of 256
wind tunnel wall, shear stress sensing at - partially and fully stabilized 253 f.
101 - polycrystals, tetragonal (TZP) 258
wire bonding 137f. - single crystals, growth of 255
work place hygiene 476, 481 - yttria-stabilized 253 f.
Sensors
Edited by T. Grandke, W. H. KO
Copyright 0 VCH Verlagsgesellschaft mbH,1989

List of Symbols and Abbreviations


The following list contains the symbols most frequently used in this book. To avoid redundancy, subscripts
are only noted in exceptional cases. References to chapters (where the quantities are explained in more '

detail) are only given for symbols with special meanings or in cases of uncommon use.

Symbol Designation Chapter

acceleration
diaphragm half length 3
lateral length of the square front area 10
velocity of sound 22
coefficients (i = 1, . ..,n ) 10
coefficient 5
cross-sectional area 22
electrode area 19
concentration of sodium ions 19

coefficient 17, 22
coefficients (i = 1,. . .,n ) 10
coefficient 22
magnetic induction, magnetic flux density

pyro-electric constant 15
specific heat
thickness of the permanent magnet 10
velocity of light
velocity of sound in air 15
velocity of light 15
coefficients (i = 1,. .., n ) 10
capacitance
cell constant 19
concentration

coefficient 22
distance
thickness of Hall plate 10
aerodynamic diameter 20
coefficients (i = 1,. . ., n ) 10
thickness of oxide 5
632 List of Symbols

Symbol Designation Chapter

diameter
diffusion constant (coefficient)

elementary charge
induced voltage 19
collection efficiency 20
electric field ( E x ,E y , E,: components)
activation energy 9
gap energy 3, 8
quantity of the source fluid to be measured 19
relative error measure 11
Young’s modulus for silicon 3

frequency
scalar field 3
probability density function 2
failure probability 19
Faraday constant
force
stress 3
general vector field
Lorentz force 7
sensor output 10
probability function 2

edge factor
gap between mask and wafer
generation rate
shear modulus

abbreviation for the sum of variables 10


response to a unit impulse 2
hazard function 2
magnetic field
transfer function 2
Fourier transform of impulse response h (t) 2

current
intensity of light 2, 17
set of N sensors or T transducers 11
1, minimum line width 5

J current density

k constant 19, 21
segregation coefficient of impurity 5
kB Boltzmann constant
List of Symbols 633

Symbol Designation Chapter

coefficients in PLS-method 5
sensitivity of device i 10
geometric sensitivity factor for current imbalance 1

etched depth 5
length
inductance
length
thickness of porons layer on cathode 14

M figure of merit for power-related Hall sensitivity 7


Mach number 22
magnetization
M T bending moment 3
rn coefficient in PLS-method 11
mass
relative molar mass (’molecular weight’) 22

n concentration (eg of negative charge carriers) 3, 5 , 7


number 10
refractive index
normal vector
coefficient in PLS-method 11
Gauss’ distribution 5
apparent refractive index 2
impurity concentration 3
number
number of counts (or of cycles) 10
numerical aperture 5
density of states in the conduction bands 3
signal of sensor element i 11
Reynolds number 22
number of cycles for time T 10
thermal load 3
density of states in the valence bands 3

P concentration of positive charge carriers 5


pressure
Pk weighting factor 10
P pressure
P polarization 9
Pi property of source material to be measured 18

4 effective loading 3
elementary charge 3, 7
flow 19
Q quantity 12
quality 2
634 List of Symbols

Symbol Designation Chapter

r Hall scattering coefficient I


ratio of total to static pressure 22
transmission factor 10
R coil radius 8
gas constant
recombination rate 3
resistance
resolution 2
deviation function 10
reliability function 8
Hall coefficient
projected range 5
gas constant 22

S ion dose per unit area 5


zero pressure separation 3
S area of cathode 14
deviation function 10
selectivity 11
signal of sensor element i 11
spline function 10
spectral density 1

t (effective) thickness of Hall plate 1


temperature
time
T (absolute) temperature
deviation function 10
time

U voltage
U voltage

U velocity
carrier velocity 7
V availability 19
contrast ratio of the resist 5
voltage 2, 3, 7
volume

W deflection 3
weighting of component 19
W width 5

X displacement, thickness, spatial coordinate


input/output signal 10
X measurand (eg concentration) 2, 11
spatial coordinate
List of Symbols 635

Symbol Designation Chapter

Y floating 15
output signal 10
spatial coordinate
Y spatial coordinate

z distance, spatial coordinate


Z spatial coordinate

a absorption coefficient 2, 3
angle
Hooge parameter I
smoothing factor 15
as Seebeck coefficient 3
Y coefficient in PLS-method 11
gage factor of the strain gage 21
isotropic exponent 22
r front surface reflection coefficient 3
A difference
& inaccuracy 2
permittivity
&O dielectric constant
t9 temperature
e, o angle
x conductivity of a solution 19
thermal conductivity 3
1 air-to-fuel ratio 14
wavelength
A selectivity 2
P magnetic permeability
mean value 2, 11
mobility of charge carriers 3, 7
PO magnetic permeability of vacuum
V kinetic viscosity 22
transmission factor 10
Poisson’s ratio 3
vibration rate 19
r damping factor 10
n tensor of piezoresistive coefficients 3
n specificity 2
e correlation coefficient 11
density
resistivity I
(3 coefficient in PLS-method 11
conductivity
standard deviation 2
stress 3. 9
636 List of Symbols

Symbol Designation Chapter

variance 2
carrier lifetime 3
function of oxide thickness 5
response time 16
time constant
angle
Fermi potential 3
photon flux 3
phonon-drag contribution 3
electrostatic potential 3
angular frequency
rate of rotation 8
volume element 3
nabla operator

~ ~~

Abbreviation Explanation
AC alternating current
ADC analogue to digital converter
A/D analog/digital
ADP NHdHzPO,
AES Auger electron spectroscopy
A/F ratio air-to-fuel ratio
ANSI American National Standards Institute
AMA Arbeitsgemeinschaft Messwert-Aufnahmer
AMT automated manufacturing technology
APD avalanche photodiode
ASCII American standard code for information interchange
ASIC application specific integrated circuit
asp analog signal processing

BE beacon electronic (device)


BR beacon receiver
BSI British Standards Institute
BT beacon transmitter

CAD computer aided design


CARS coherent anti-stokes Raman spectroscopy
CCD charge coupled device
CCITT International Consultative Committee on Telegraphy and Telephony
CD compact disc
c.g. center of gravity
chemFET chemically sensitive field effect transistor
CIM computer integrated manufacturing
CM correlation method
CMAC centralised media access control
CMOS complementary-symmetric metal oxide semiconductor
List of Symbols 637

Abbreviation Explanation

CR carriage return (ASCII)


CRC cyclic redundancy check
CSMA/CD carrier sense multiple access with collision detection
CVD chemical vapor deposition

DC direct current
DIN Deutsche Industrie-Norm
DIP dual in-line programmable
dual in-line package
DOD Department of Defence (USA)
DS destination store

EDX energy-dispersive X-ray analysis


EGS electronic grade silicon
EIA Electrical Industries of America
ELF extremely low frequency
emf electromotive force
EM1 electromagnetic interference
EP environmental parameter
EPA enhanced performance architecture
EPM electron probe microanalysis
ERA Electrical Research Association
erfc complementary error function
EtO ethylene oxide

FIP factory instrumentation protocol


FM frequency modulation
FOG fiber optic gyro(scope)
FSK frequency shift keying
FSO full-scale output

GC gas chromatography
GEMS global environment monitoring system
GF-AAS graphite-furnace atomic absorption spectrometry
GPIB general purpose instrument bus
GRIN graded index (lens)

HDUJ high-level data link control


HF high frequency
HIP hot isostatic pressing
HP hot pressing
HP/LP high pressure/low pressure

I + C instrumentation and control (system)


IC integrated circuit
ICP-AES inductively-coupled plasma atomic emission spectroscopy
IEC International Electrotechnical Commission
IEEE Institute of Electrical and Electronics Engineers
638 List of Symbols

Abbreviation Exulanation

IMU inertial measurement unit


i/o input/output
I0 integrated optic(s)
IOC integrated optical circuit
IOP Institute of Physics
IR infrared
IS intrinsic safety
ISA Instrument Society of America
ISE ion selective electrode
ISFET ion sensitive field effect transistor
IS0 International Standards Organisation
IT infrared transmitter

JEIDA Japanese Electronic Industrial Development Agency

KDP KH2POd

LAA lab animal allergy


LB Langmuir-Blodgett
LDV laser Doppler velocimetry
LED light emitting diode
LEED low energy electron diffraction
LF linefeed (ASCII)
LIDAR light detection and ranging
LISB guidance and information system Berlin
LPCVD low-pressure chemical vapor deposition
LVDT linear variable differential transformer
LWC liquid water content

MAG metal-active gas


MAP manufacturing automation protocol
MBE molecular beam epitaxy
PC micro computer
MCL maximum concentration limits
MCO magnetic-field controlled oscillator
MDS minimum detectable signal
MDT mean down-time
MFS magnetic field sensor
MIG metal-inert gas
MIM metal-insulator-metal
MIOC multifunction integrated optical circuit
MIS metal-insulator-semiconductor
MMS manufacturing message service
MOCVD metalorganic chemical vapor deposition
MOD metalorganic deposition
MOS metal oxide semiconductor
MOSFET metal oxide semiconductor field effect transistor
MR measured range
List of Symbols 639

~ ~~~~~~

Abbreviation Explanation

MS mass spectroscopy
MT measuring dwice for travel time
MTBF meantime between failures
MUX multiplexes
MW microwave

N navigation (device)
NASICON Na,Zr,Si,PO,,
NBS National Bureau of Standards
NRZ non return to zero
NTC negative temperature coefficient

0 operation (board)
OCP open circuit potential
OEIC opt0 electronic integrated circuit
OSI open systems interconnection

P position (finding device)


PAD packet assembler/disambler
PC personal computer
PCM pulse code modulation
PDM pulse duration modulation
PECVD energy enhanced chemical vapor deposition
PID proportional integral derivate (controller)
PIN p and n region, separated by an intrinsic semiconductor
PLC programmable logic controller
PLS partial least squares (method)
PLZT lanthanum-modified PZT
PP passivating potential
PROFIBUS process field bus
PROM programmable read only memory
PROWAY process data highway
PRT platinum resistance thermometer
PSB polished silica block
PSD power spectral density
PSG photo silicate glass
PSTN public service telephone network
PSZ partially stabilized zirconia
PT photo transistor
PTC positive temperature coefficient
PVC polyvinyl chloride
PVDF polyvinylidene fluoride
PZT PbZrO, - PbTi03

RBS Rutherford back-squattering spectroscopy


R&D research & development
rf, RF radio frequency
RIE reactive ion etching
640 List of Symbols

Abbreviation Explanation

rms, RMS root mean square


ROM read only memory
RTD resistance temperature detector
RVDT rotary variable displacement transformer

SAW surface acoustic wave


SAP simple asynchronous protocol
SCADA supervisory control and data acquisition
SRC selective catalytic reduction
SDLC synchronous data link control
se sensor element
SEM scanning electron microscopy
SG silicate glass
SIMS secondary ion mass spectroscopy
SMD surface mounted design
SNR signal-to-noise ratio
SOG spin-on-glass
so1 silicon-on-insulator
SPCC cold rolled carbon steel strip
SQUID superconductor quantum interference device
SSIMT suppressed-sidewall injection magneto-transistor
STP Southampton transducer protocol

TAB tape automated bonding


n: traffic signal controller
TEA total exposure assessment
TEM transmission electron microscopy
TGC traffic guidance computer
TGS triglyzerine sulfate
TIG tungsten-inert gas
TLD thermoluminiscent detector
TLS transformed least squares
TOC total organic carbon
TSE temperature-sensitive element
TV television
TZP tetragonal zirconia polycrystal

UART universal asynchronous receiverltransmitter


uv ultraviolett

vco voltage controlled oscillator


VCR video casette recorder
VDIlVDE Verein Deutscher IngenieurelVerein Deutscher Elektrotechniker
VDU visual display unit
VHF very high frequency
VLSI very large scale integration
VM vector method
List of Symbols 641

Abbreviation Explanation

voc volatile organic compound


VTR video tape recorder

WP wheel pulser
XRD X-ray diffraction
XPS X-ray photoelectron spectroscopy

You might also like