Ultra Low Power Capacitive Sensor Interfaces

ULTRA LOW POWER CAPACITIVE
SENSOR INTERFACES
ANALOG CIRCUITS AND SIGNAL PROCESSING SERIES
Consulting Editor: Mohammed Ismail. Ohio State University
Titles in Series:
ULTRA LOW POWER CAPACITIVE SENSOR INTERFACES
Bracke, Wouter, Puers, Robert, Van Hoof, Chris
ISBN: 978-1-4020-6231-5
LOW-FREQUENCY NOISE IN ADVANCED MOS DEVICES
Haartman, Martin v., stling, Mikael
ISBN-10: 1-4020-5909-4
CMOS SINGLE CHIP FAST FREQUENCY HOPPING SYNTHESIZERS FOR WIRELESS
MULTI-GIGAHERTZ APPLICATIONS
Bourdi, Taoufik, Kale, Izzet
ISBN: 978-14020-5927-8
ANALOG CIRCUIT DESIGN TECHNIQUES AT 0.5V
Chatterjee, S., Kinget, P., Tsividis, Y., Pun, K.P.
ISBN-10: 0-387-69953-8
IQ CALIBRATION TECHNIQUES FOR CMOS RADIO TRANCEIVERS
Chen, Sao-Jie, Hsieh, Yong-Hsiang
ISBN-10: 1-4020-5082-8
FULL-CHIP NANOMETER ROUTING TECHNIQUES
Ho, Tsung-Yi, Chang, Yao-Wen, Chen, Sao-Jie
ISBN: 978-1-4020-6194-3
THE GM/ID DESIGN METHODOLOGY FOR CMOS ANALOG LOW POWER
INTEGRATED CIRCUITS
Jespers, Paul G.A.
ISBN-10: 0-387-47100-6
PRECISION TEMPERATURE SENSORS IN CMOS TECHNOLOGY
Pertijs, Michiel A.P., Huijsing, Johan H.
ISBN-10: 1-4020-5257-X
CMOS CURRENT-MODE CIRCUITS FOR DATA COMMUNICATIONS
Yuan, Fei
ISBN: 0-387-29758-8
RF POWER AMPLIFIERS FOR MOBILE COMMUNICATIONS
Reynaert, Patrick, Steyaert, Michiel
ISBN: 1-4020-5116-6
ADVANCED DESIGN TECHNIQUES FOR RF POWER AMPLIFIERS
Rudiakova, A.N., Krizhanovski, V.
ISBN 1-4020-4638-3
CMOS CASCADE SIGMA-DELTA MODULATORS FOR SENSORS AND TELECOM
del Ro, R., Medeiro, F., Prez-Verd, B., de la Rosa, J.M., Rodrguez-Vzquez, A.
ISBN 1-4020-4775-4
SIGMA DELTA A/D CONVERSION FOR SIGNAL CONDITIONING
Philips, K., van Roermund, A.H.M.
Vol. 874, ISBN 1-4020-4679-0
CALIBRATION TECHNIQUES IN NYQUIST AD CONVERTERS
van der Ploeg, H., Nauta, B.
Vol. 873, ISBN 1-4020-4634-0
ADAPTIVE TECHNIQUES FOR MIXED SIGNAL SYSTEM ON CHIP
Fayed, A., Ismail, M.
Vol. 872, ISBN 0-387-32154-3
WIDE-BANDWIDTH HIGH-DYNAMIC RANGE D/A CONVERTERS
Doris, Konstantinos, van Roermund, Arthur, Leenaerts, Domine
Vol. 871 ISBN: 0-387-30415-0
Ultra Low Power
Capacitive Sensor Interfaces
by
WOUTER BRACKE
Catholic University of Leuven, Belgium
ROBERT PUERS
Catholic University of Leuven, Belgium
and
CHRIS VAN HOOF

IMEC vzw, Belgium
A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-1-4020-6231-5 (HB)

ISBN 978-1-4020-6232-2 (e-book)
Published by Springer,
P.O. Box 17, 3300 AA Dordrecht, The Netherlands.
www.springer.com
Printed on acid-free paper
All Rights Reserved

c 2007 Springer
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming,
recording or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered and executed on a
computer system, for exclusive use by the purchaser of the work.
Contents
Foreword ix
1. INTRODUCTION 1
2. GENERIC ARCHITECTURES
FOR AUTONOMOUS SENSORS 5
1 Introduction 5
2 Multisensor microsystem 6
2.1 Sensors 6
2.2 Sensor interface chip 7
2.3 Microcontroller 8
2.4 Wireless link 9
2.5 Power management 9
3 Modular design methodology 10
3.1 Programming flow 11
3.2 Operational flow 13
4 Conclusion 14
3. GENERIC SENSOR INTERFACE CHIP 17
1 Introduction 17
2 Capacitive sensors 18
3 Generic Sensor Interface Chip for capacitive sensors 20
3.1 Front-end architecture 21
3.2 Capacitance-to-Voltage convertor 24
3.3 Chopping scheme 28
3.4 SC amplifier 29
3.5 modulator 33
vi Contents
3.6 Bandgap reference, bias system and buffered

reference voltage 42
3.7 Main clock, clock generation circuits and LF clock 48
4 Configuration settings 51
5 Noise 53
5.1 Bennet model 53
5.2 Noise calculations 54
5.3 Effective number of bits 59
6 Experimental results 60
6.1 Pressure monitoring system 62
6.2 Inclination monitoring system 65
7 Performance comparison 69
8 Conclusion 70
4. ALGORITHM FOR OPTIMAL CONFIGURATION
SETTINGS 73
1 Introduction 73
2 Programmability 73
2.1 Full-scale loss 74
2.2 Programmability of Cre f 75
2.3 Programmability of C f 76
2.4 Programmability of ASC 80
2.5 Noise 82
3 Optimal settings 83
4 Results 84
5 Conclusion 85
5. PHYSICAL ACTIVITY MONITORING SYSTEM 87
1 Introduction 87
2 Background and motivation 87
3 Implementation 88
4 Conclusion 90
Contents vii
6. CONCLUSION 91
1 Realized developments 91
2 Suggestions for future work 93
References 95
Index 103
Foreword
The increasing performance of smart microsystems merging sensors, signal

processing and wireless communication promises to have a pervasive impact
during the coming decade. These autonomous microsystems find applications
in sport evaluation, health care, environmental monitoring and automotive sys-
tems. They gather data from the physical world, convert them to electrical
form, compensate for interfering variables or non-linearities, and either act di-
rectly on them or transfer it to other systems. Most often, these sensor systems
are developed for a specific application. This approach leads to a high recur-
rent design cost. A generic front-end architecture, where only the sensors and
the microcontroller software are customized to the selected application, would
reduce the costs significantly.
This work presents a new generic architecture for autonomous sensor nodes.
The modular design methodology provides a flexible way to build a complete
sensor interface out of configurable blocks. The settings of these blocks can
be optimized according to the varying needs of the application. Furthermore,
the system can easily be expanded with new building blocks. The modular
system is illustrated in a Generic Sensor Interface Chip (GSIC) for capaci-
tive sensors. Many configuration settings adapt the interface to a broad range
of applications. The GSIC is optimized for ultra low power consumption. It
achieves an ON-state current consumption of 40 A. The system maintains a
smart energy management by adapting the bias currents, measurement time
and duty cycle according to the needs of the application (parasitic element re-
duction, accuracy and speed). This results in an averaged current consumption
of 16 A in a physical activity monitoring system. The activity monitoring sys-
tem is implemented in a miniaturized cube. It consists of a sensor layer (GSIC
and accelerometer), a microcontroller layer and a wireless layer. The bidirec-
tional wireless link (from the sensor node to the computer) makes it possible to
display the data in real time and to change the interface settings remotely. So,
the smart autonomous sensor node can adapt at any moment to environmental
x Foreword
changes. The GSIC is also successfully tested with other accelerometers and
pressure sensors. Hence, the developed GSIC is a significant step towards a
generic platform for low cost autonomous sensor nodes.
Wouter Bracke
KULeuven, ESAT-MICAS/INSYS
now with ICsense NV
Leuven,
January 2007
Chapter 1
INTRODUCTION
The drive towards an intelligent environment has lead to an increased need

for intelligent and independent sensors. Possible forecasts predict these au-
tonomous sensors to work as small distributed units, that can collect data over
a longer period of time [War01, Asa98, Rab02]. According to this vision, they
should meet the following challenging criteria:
highly miniaturized. So, they can be worn or implanted without any dis-
comfort for the user.
versatile. The sensors will be able to operate without any intervention of
the user.
maintenance free. The nodes can supply their own energy. Hence, they need
to combine Ultra Low Power (ULP) electronics (sensor interfaces, micro-
controller and communication front-end) with an efficient energy generation
and storage.
low cost.
The emerging opportunities of these autonomous sensors will give the first
impulse to several new applications such as intelligent prosthesis, sport eval-
uation, observation of livestock and measurement of weather patterns. The
last decade, a tremendous progress has already been made in such smart mi-
crosystems. Some impressive realizations are a wireless multisensor medical
microsystem [Tan02] and a very low power pacemaker system [Won04].
The multisensor medical microsystem integrates a microsensor array, the sig-
nal processing electronics, a wireless transmitter and batteries in a miniaturized
capsule of 16 mm (diameter) by 55 mm (length). The sensor array contains
a dissolved oxygen sensor, a pH-sensitive Ion-Selected Field Effect Transistor
2 Ultra Low Power Capacitive Sensor Interfaces
(ISFET), a standard PN-junction silicon temperature sensor and a dual elec-

trode direct contact conductivity sensor. The complete system dissipates only
6.3 mW for a minimal life cycle of 12 h.
The implantable pacemaker system monitors the hearts rate (how fast it beats)
and rhythm (the pattern in which it beats), and provides electrical stimulation
when the heart does not beat or beats too slowly. The pacemaker IC contains
amplifiers, filters, ADCs, battery management system, voltage multipliers, high
voltage pulse generators, programmable logic and timing control. The IC has
200 k transistors, occupies 49 mm2 and consumes 8W. This enables a lifetime
of 10 years on a lithium iodine battery.
Most of these sensor systems were tailored towards the requirements of one
specific application. This design approach is inflexible and requires several
iteration steps for new sensor applications. It usually results in an intolerable
high design cost for low and medium quantity market products. An ULP generic
multisensor interface would reduce the costs significantly, since one can use the
same interface chip for several applications. Hence, the recurrent design costs
are eliminated and the time to market is shorter. Furthermore, the front-end
can be adapted during operation. Hence, we can adjust the system to changes
in the environment (e.g. enter a low power mode, when the available supply
energy is getting low). Moreover, a generic interface is capable of reading out
several sensors in different time intervals. So, we can combine the information
from different sensors to compensate for cross-sensitivities (e.g. compensation
of the temperature dependency).
Several research groups have already developed generic sensor interface
architectures [Yaz00, Mas98, VDG96]. Most of these systems were designed
for industrial applications, which do not need the lowest power consumption.
As a consequence, their power dissipation is still too high (in the order of mWs
instead of the required tens of W) to permit autonomous functioning over a
longer period of time.
This work aims to combine the flexibility of generic sensor interfacing with
ultra low power consumption. The developed modular design methodology
provides a flexible way to build a complete sensor interface out of configurable
blocks. The settings of these blocks can be changed according to the varying
needs of the application. Furthermore, the system can easily be expanded with
new building blocks. The modular system is illustrated in a Generic Sensor
Interface Chip (GSIC) for capacitive sensors. The GSIC is tested with several
micromachined pressure sensors and accelerometers. Moreover, the GSIC is
used in a miniaturized demonstrator for physical activity monitoring.
The outline of the presented work is as follows:
In chapter 2 an overview of the most important design aspects for au-

tonomous sensor nodes is given. The different building blocks are discussed
Introduction 3
and the new modular architecture for the smart sensor interface chip is de-
veloped.
Chapter 3 describes the Generic Sensor Interface Chip for capacitive sen-
sors. Firstly, the front-end architecture and the design of the analog blocks
are discussed. Secondly, the configuration settings and noise calculations
are presented. Finally, experimental results are given in state-of-the-art
pressure sensor and accelerometer applications. The performance of the
implemented systems is compared with other generic sensor interfaces and
dedicated (U)LP capacitive sensor interfaces.
Chapter 4 studies the effect of programmability on generic (capacitive) sen-
sor interfaces. It also provides an algorithm, which calculates the optimal
configuration settings for each application. These settings enable a maxi-
mal accuracy of the sensor data for a given power consumption of the GSIC
(sample frequency and measurement time).
Chapter 5 presents a 1 cm3 physical activity monitoring system. This physi-
cal activity monitoring system has been implemented as a 3D stacked sensor
node, which contains a sensor (accelerometer and GSIC), a microcontroller
and a wireless layer. The sensor node communicates with a remote station,
which is implemented on the PC (USB stick). A Labview computer interface
displays the data in real time and allows to change the settings remotely.
Finally, chapter 6 presents some general conclusions.
Chapter 2
GENERIC ARCHITECTURES
FOR AUTONOMOUS SENSORS
1. Introduction
During the past two decades, several smart sensor systems have been pre-
sented. In most of the cases, these systems contain one or more sensors, a
sensor interface and signal conditioning circuits, a microcontroller and/or a
dedicated digital signal processing unit and a display and/or a wireless core
for the communication. Most often, these sensor systems were developed for
a specific purpose. In the literature, one can find systems for a hugh vari-
ety of applications, such as intelligent prosthesis monitoring systems [Cla03],
tire pressure monitoring systems [Kol04], intelligent weather observation sys-
tems [Hua03], etc. Dependent on the application, one uses a different type of
amplifier, filter bank, analog-to-digital convertor (ADC) and digital signal pro-
cessing. In spite of these differences, there is still a common system framework
between most of the applications. Hence, one could benefit from a common
front-end architecture, where only the sensors and the microcontroller software
are customized to the selected application. Such a generic architecture would
provide a low cost, flexible and easy to use environment to create autonomous
microsystems.
In this chapter we present a generic architecture, which allows to create a
sensor interface out of configurable blocks. The configuration settings and
the combination of the blocks can be changed according to the needs of the
application. Furthermore, the modular system can be easily expanded with
new building blocks to provide extra features. Such a generic system can be
used as the core of a smart (e.g. human body) sensor network, which connects
several sensor nodes. By its generic nature, it opens opportunities for mass
production, which allows to lower the price.
2. Multisensor microsystem
In the eighties, W. Sansen has developed the first conceptual view on a generic
Internal Human Conditioning System (IHCS) [San82]. When we combine
these insights with more recent work [Mas98, Pue99], we can define a general
smart microsystem that contains a multisensor array, a sensor interface chip, a
microcontroller, a wireless link and a power management (Fig. 2.1).
2.1 Sensors
Sensors are used in several commercial markets such as automotive indus-
try, consumer electronics and medical equipment. The function of the sensor
element is to convert energy from any energy domain (magnetic, chemical, op-
tical, mechanical or thermal) into the electrical domain [Mid89]. The obtained
electrical signal can be conditioned further by the interface electronics. Ideally,
the output of the sensor is proportional to its input signal and remains the same
over time. Unfortunately, real sensors are subject to spread in the production,
non-linearities, cross-sentivities and drift.
The variations due to the manufacturing processes cause a spread in the
sensor sensitivity and offset. These effects can be dealt with by calibration
after fabrication. During this calibration, reference signals are applied on the
system. This provides the necessary correction parameters to adjust the sensor
output signal, so that its input-output relation is well defined. This correction
can be implemented in the microcontroller or in the remote station. Furthemore,
the calibration parameters are also used to compensate for the non-linearities
and cross-sentivities, like temperature dependency, in the sensor system. For
this purpose, extra temperature measurements are performed, resulting in a
multisensor microsystem. In such a system, the measured temperature value is
used to correct the output.
Sensor interface chip
Sensor 1
ADC Memory
Sensor 2
Clocking and Microcontroller
Interface Timing circuits
Sensor 3
electronics Control settings
Sensor 4 Operational settings

Calibration settings Transmitter
Sensor 5 Monitoring windows Receiver
Remote
Transmitter
Receiver
Figure 2.1. General multisensor microsystem for autonomous sensor applications.

Generic Architectures for Autonomous Sensors 7
The most difficult non-ideality to compensate is drift. Knowledge of the

sensor signal in combination with appropriate signal processing algorithms,
like correlation techniques, can reduce these effects significantly [Hos97].
2.2 Sensor interface chip

The sensor interface chip performs the amplification, the filtering and the
analog-to-digital conversion of the sensor signals. It also contains a local con-
figuration memory, a finite state machine, several timing and clock circuits and
a microcontroller interface. This makes the sensor interface chip a versatile
component, which can be programmed at any time. Hence, we can adjust the
sensitivity and compensate for the offset of the sensor to ensure that the am-
plifiers are not saturated and the dynamic range of the sensor system is not
degraded. The sensor interface chip also offers options for intelligent power
management. The duty cycle operation makes the averaged power consumption
adaptable to the accuracy and speed requirements of the selected sensor appli-
cation. Furthermore, all the channels, which are not in use, can be switched off
individually.
The low cost, Ultra Low Power (ULP) sensor interface chip will be imple-
mented in CMOS technology. In this technology, it is important to compensate
for the reduced matching (offset and drift) and 1/f noise of the transistors in
the signal conditioning chain. These problems can be reduced by Correlated
Double Sampling (CDS) and/or chopping [Enz96].
In the CDS technique, the offset compensation is performed in two phases.
During one phase, the offset is sampled and stored and during the next phase the
sampled offset is subtracted from the present one. These successive values are
strongly correlated, which results in a significant offset reduction. Moreover,
the CDS principle decreases the low-frequent 1/f noise.
In the chopping technique, the input signal is multiplied by a square wave
signal at a frequency fchop . The modulated input signal is then amplified and
demodulated back to the baseband by a second chopper. The offset is, however,
modulated only once and appears as frequency components around the odd
harmonics of fchop . These offset and 1/f noise components are removed by a
low-pass filter.
In the chopping technique, the white noise of the amplifier is not aliased into
the baseband, contrary to the CDS technique. This suggests that the chopper
technique is more appropriate for continuous time applications, whereas the
CDS technique is more suitable for sampled data applications, where aliasing
is unavoidable.
The sensor interface chip should give a flexible and easy communication
to different types of microcontrollers. For this purpose, a microcontroller in-
terface is included, that is able to perform an efficient data transfer and fast
reconfiguration of the sensor front-end with a reduced complexity (limited num-

ber of IO pins, die area and power consumption).
2.3 Microcontroller
The microcontroller has several important tasks. First of all, it controls the
sensor interface chip. It provides the settings of the sensor interface chip, such
as the configuration of the readout electronics, the application mode and the
duty cycle. Secondly, it gathers the data coming from the sensor interface chip
and stores it in a memory. Furthermore, it can perform the digital linearization
and cross-sensitivity compensation. For this purpose, we can use look-up ta-
bles or implement polynomial evaluation. Look-up table algorithms offer good
accuracy but are very demanding on system memory. An attractive alternative
is polynomial evaluation, which uses significantly less memory than look-up-
table methods but is generally slower [Cra90, Yos97] . The microcontroller
can also implement smart compression algorithms to extract the relevant data
from the sensor signals. Hence, the amount of data, that needs to be transmit-
ted is decreased. This reduces the power consumption significantly, since the
telemetry link has a relatively large power consumption in the sensor node.
The microcontroller needs to be energy efficient to enable a large amount of
signal processing with a minimum of energy. Table 2.1 lists several low power
microcontrollers and their energy consumption per instruction normalized by
the number of bits in the datapath (source [War03]). The table contains both
general purpose microcontrollers (such as the TI MSP430 and the CoolRisc) and
digital signal processing units (MIT Sensor DSP). The latter have the advantage
that their architecture is more dedicated towards wireless sensor nodes. For
these applications, this results in more efficient implementations.
Table 2.1. Energy consumption of various microcontrollers.
Microcontroller Energy (pJ/instruction/bit)

Dallas DS80C320 High Speed 8051 1100
SICAN RISC 4b (0.25m) 75
TI MSP430C1111 (2.2V) 45
Punch Multitask RISC core (2m, 1.5V) [Per94] 25
CoolRisc 81 cont. (1m, 1.5V) 5.7
CoolRisc 81 core (0.25m, 1.05V) [Arm00] 1.25
MIT Sensor DSP (0.6m, 1.5V) [Ami00] 2.2
2.4 Wireless link

The wireless transceiver eliminates the need for costly wired networks. The
bidirectional wireless link enhances the flexibility, since the system is adaptable
during operation. This bidirectional communication sends the sensor data to
a remote transceiver and provides the microsystem with new programming in-
structions. Hence, the accuracy, sensitivity, sample frequency, data processing,
etc. can eventually be changed during operation. This is necessary to adapt the
system to environment changes or to compensate for drift phenomena.
Implantable biotelemetry systems, generally use low-frequency (< 135 kHz)
signals, whereas other autonomous sensor nodes often use a high-frequency
(e.g. 433 MHz/916 MHz) communication front-end.
Low-frequency systems are most often based on inductive coupling [Cat04].
Such systems have a limited communication speed and a short communication
range. An advantage of low-frequency radio signals is their ability to propagate
through water and body tissue. This makes them very suitable for implantable
devices.
High-frequency systems on the other hand offer long communication ranges
and a high communication speed. Moreover, they allow for the use of smaller
antennas. The main disadvantage of high-frequency radio signals is their at-
tenuation by many (water containing) materials. High-frequent communication
front-ends are implemented with narrowband [Nor, Jac03, AMI] and Ultra Wide
Band (UWB) [Ryc05] solutions. Compared to narrowband implementations,
sensor nodes with UWB communication have a lower power consumption for
good channels, since they benefit from a simpler transmitter front-end. For av-
erage channels, narrowband solutions become better, since the transmit power
dominates the front-end power consumption.
2.5 Power management

Autonomous sensor systems can be divided into active or passive powered
devices.
The active devices do not require any interaction with the outside world
regarding their powering. Hence, they need ULP electronics to operate au-
tonomously over a longer period of time. These devices have a projected power
budget of 100W, which is divided into 20W for the sensing part (sensors
and readout), 40W for the digital data processing and 40W for the wire-
less transceiver. These sensor nodes can be powered by batteries or energy
scavengers. Energy scavengers extract power from the environment, such as
vibrations [Ste05, Des05] and body warmth [Leo05]. These power sources can
vary strongly. So, specialized electronics need to convert the available power
into a reliable supply voltage. Furthermore, the analog read-out electronics
should have a high power supply rejection to cope with drift in the supply
voltage.
The passive devices derive their power from an external radiofrequent (RF)
powering field. They only operate when this RF field is active and in the prox-
imity. Hence, they can only be used in non-continuous monitoring applications
where the external powering system is in close proximity to the monitoring
device.
3. Modular design methodology

From the above derived system framework, a new generic architecture will
be elaborated to create low-cost and flexible autonomous sensor nodes. Such
a generic sensor system uses a plug and play approach to combine the sensors,
the sensor interface, the microcontroller, the wireless communication core and
the power management into a miniaturized system. In this common front-end
architecture, we should only customize the sensors and the microcontroller
software to the selected application.
A generic sensor interface chip is an important part of this universal platform.
Such a generic sensor interface is designed as a modular system including sev-
eral configurable building blocks (Fig. 2.2). These configurable blocks can be
Sensor interface chip
Analog front-end
Address 0
Address 1
CB 1 CB 2 CB 3 Address 2
Address 3
Configuration Configuration Configuration
Address 4
Address 5
Address 6
Configuration Configuration Configuration
Address 7
Address 8
CB 4 CB 5 CB 6 Address 9
Address 10
Address 11
Address 12
Act Address 13
Configuration Configuration Address 14
LF clock Main clock Address 15
Sample Conversion Reset

timer timer Configuration interface
Configuration Configuration
POR Ack Dout Dav Write Dclk Din Activate
Figure 2.2. Functional description of the modular sensor interface chip.

programmable amplifiers, filters, data converters, clocks, etc. The configura-

tion settings and the combination of these blocks can be changed according to
the needs of the application. Furthermore, the modular system can be easily
expanded with new building blocks to provide extra features. The settings of
these blocks are stored in a configuration SRAM. The configuration interface
allows the microcontroller to change the settings in a power efficient way. It
contains modes for fast and complete reconfiguration. In the fast reconfigu-
ration, only the settings of one specified address are changed. This is very
useful during the calibration of one specific interface parameter. Furthermore,
it is a handy manner to enter or leave a low power mode during operation. In
the complete reconfiguration, all the settings of the sensor interface chip are
reprogrammed. The normal operation flow stops during both configuration
modes. So, the sensor interface chip is in a reset state. After completion of the
programming phase, the sensor interface chip leaves the reset state and starts
operating autonomously. It amplifies the sensor signals and converts them to a
digital code. These digitized sensor data are transferred asynchronously to the
microcontroller.
3.1 Programming flow

Four pins are used during the programming of the sensor interface: Write,
Activate, Dclk and Din. When Write is high, the microcontroller is busy with
programming and the operation flow is in a reset state. During this phase, the
sensor interface loads the serial configuration data Din at the falling edge of
Dclk. The Activate pin represents the configuration state of the sensor interface.
Activate is low, when the configuration is not finished. After the configuration is
complete, all the configuration flags of the SRAM memory are high, this results
in a high Activate signal. This event is detected by the microcontroller, which
lets the sensor interface enter the operation mode with the new configuration
settings (Write becomes low).
The serial input data Din are generated in the following 16-bit format: ERN0
ERN1 A0 A1 A2 A3 D0 . . . D9. The first two bits (ERN0 and ERN1) encode the
type of the input word. This word can be interpreted as a complete programming
instruction, a fast programming instruction or a configuration word.
1 If (ERN0 ERN1) equals (1 0), the sensor interface chip needs to be fully
reprogrammed. So, all the configuration flags become low. In this case, the
other 14 bits in the word are dont care bits.
2 If (ERN0 ERN1) equals (0 1), only address A0 A1 A2 A3 needs to be

reprogrammed. So, only the configuration flag of this address becomes low.
3 If (ERN0 ERN1) equals (0 0), the data D0 . . . D9 are loaded at the config-
uration address A0 A1 A2 A3. Hence, the configuration flag of this address
becomes high.
Figs. 2.3 and 2.4 show the flow charts for the complete and fast configura-
tion modes.
The proposed interface combines simplicity with low power consumption.
It uses only 4 pins and saves a lot of energy, since the clock Dclk is only
provided during the programming phase. Furthermore, the fast reconfiguration
is an energy and time efficient option to change only one parameter during the
operation. This is attractive in many applications. As an example, we consider
Write bit becomes high
The operation flow is reset

(reset = 1, act = 0 and Dav = 0)
The first input word Din is loaded
It has (ERN0,ERN1) = (1,0)
All configuration flags are low
Activate becomes low
The code word

(0,0,address0,data0) is loaded
The data0 are written in the
SRAM at address0
The configuration flag of
address0 becomes high
The code word
SRAM at address1
The code word

SRAM at address15
Activate becomes high
Write becomes low
The operation flow starts
Figure 2.3. Flow chart for the complete reconfiguration mode.

Write bit becomes high
The operation flow is reset

(reset = 1, act = 0 and Dav = 0)
The first input word Din is loaded
It has (ERN0,ERN1) = (0,1) and
addressx
addressx becomes low
Activate becomes low
The code word

(0,0,addressx,datax) is loaded
The datax are written in the
SRAM at addressx
addressx becomes high
Activate becomes high
Write becomes low
The operation flow starts
Figure 2.4. Flow chart for the fast reconfiguration mode.
the case of the smart energy management in event-triggered microsystems.

In order to save the supply energy, the system is in a sleep mode. If some
important event is happening, the fast reconfiguration makes the system enter
the operational mode quickly and an accurate monitoring can start.
3.2 Operational flow

The programmed sensor interface chip operates autonomously. Its duty cycle
operation is controlled with the sample and conversion timers. The LF clock
and sample timer are running all the time, while the analog front-end and main
clock are only operating in the active mode. During this active mode, the
signal is converted into a digital output code. After this conversion, the sensor
interface chip enters the standby mode (the analog front-end and the main
clock are switched off) and puts the data available signal, Dav, high. This
event wakes the microcontroller in order to load the digital data, Dout. The
acknowledgement signal, Ack, becomes high after a successful transfer. This
resets Dav and eventually Ack becomes low. When the sample timer reaches
Nsample counts, it is reset and the process starts again (Fig. 2.5).
The microcontroller and sensor interface chip work independent in this op-
eration flow. The communication is only set up for data transfer from the
sensor interface chip to the microcontroller. The exact time and duration of
The interface chip is programmed

(Activate=1, Write=0, Dav=0, Ack=0)
The sample timer starts counting
After 1 count, Act becomes high
The bias currents of the analog front-

end are powered
Active mode
After 2 counts, Reset becomes low
The conversion process starts
The conversion is finished, the data are

transferred to Dout
One main clock period later, Act gets
low, reset and Dav become high
The analog front-end is switched off
Standby mode
The microcontroller loads the digital

data, Ack becomes high
Dav becomes low
Ack becomes low
The sample timer reaches Nsample counts
The sample timer restarts counting

from zero
Figure 2.5. Operational flow of the modular microsystem.
this asynchronous transfer are unimportant. The system functions correctly, if

the data transfer is completed during the standby mode of the sensor interface.
4. Conclusion
A generic platform for autonomous sensors would be a significant step to-
wards low cost, flexible and easy to use sensor nodes for the smart environ-
ment. Such a general multisensor microsystem consists of a multisensor ar-
ray, a sensor interface chip, a microcontroller, a wireless link and a power
management. An open module architecture with plug and play approach al-
lows to create an adequate solution for each application. Hence, most of the
applications benefit from a generic sensor interface architecture, where only
the sensors and the microcontroller software are customized to the selected
application.
In order to create such a generic platform, we have first derived the proper-
ties and design options for the different parts of the multisensor microsystem.
Secondly, a new modular architecture was presented, which allows to create
a complete sensor interface chip out of configurable blocks. The combination
and the settings of these blocks can be changed according to the varying needs
of the application. Furthermore, the sensor system can be expanded with addi-
tional building blocks during the development phase. The fast reconfiguration
offers a power and time efficient option to change only one interface parameter
during operation. The programmed sensor interface functions autonomously
and performs an asynchronous data transfer to the microcontroller.
Chapter 3
GENERIC SENSOR INTERFACE CHIP
1. Introduction
A modular design approach for autonomous sensors was presented in the
previous chapter. These concepts are used to create an ULP Generic Sensor
Interface Chip (GSIC). The GSIC performs an interface to a broad range of
capacitive sensor applications with medium accuracy (8-10 bits) and low speed
requirements (bandwidth <100 Hz).
This chapter presents the specifications, design and results of the GSIC. In the
first section, the different types of micromachined capacitive sensors are studied.
This results in a classification for capacitive sensors, which eventually leads to
the specifications for the GSIC. Thereafter, several front-end architectures for
capacitive sensors are studied. An optimal architecture is presented, which
achieves the required specifications with a minimal total power consumption.
The complete sensor interface chip contains Capacitance-to-Voltage converters,
a Switched Capacitor (SC) amplifier, a modulator, an LF clock, a main
oscillator, timing circuits, a bandgap reference and bias circuits. The design of
these blocks is described in sections 3.2 to 3.7. All these blocks are highly
configurable. The many configuration settings allow to optimize the interface
for a broad range of applications (section 4). The noise calculations of the
interface chip are presented in section 5. Finally, the GSIC is tested in state-
of-the-art pressure and accelerometer applications. The implemented pressure
monitoring system achieves a power consumption of 7.3 W for a 10 Hz sample
frequency and 8-bit accuracy in the 100 to 130 kPa range. In the acceleration
monitoring system, we measured a 10.3 W power consumption for a 10 Hz
sample frequency and 9-bit accuracy in the 1 g range. Furthermore, the
performance of these systems is compared with other generic sensor interfaces
and dedicated (U)LP pressure and accelerometer systems.
2. Capacitive sensors
Capacitive sensors can measure different types of physical signals like
humidity, acceleration, pressure and position. Capacitive sensors are suitable
for autonomous sensor applications since they dissipate no power and offer a
high sensitivity [Pue93]. The main disadvantage is the presence of high para-
sitic elements. Fig. 3.1 shows a simple electrical model of a single capacitive
sensor, including the effects of a shunting conductance Gp and two parasitic
capacitances Cp1 and Cp2 .
Mechanical capacitive sensors have a higher sensitivity, lower power con-
sumption, better temperature performance and are less sensitive to drift than
piezoresistive sensors. However, piezoresistive sensors have a simpler struc-
ture, fabrication process and readout circuit, since the resistive bridge provides
a low impedance output voltage. Hence, capacitive sensors are most often used
in low power and high performance applications. Mechanical capacitive sen-
sors can be developed with bulk or surface micromachining [Fre98, Yaz98].
In bulk micromachining, the wafer is etched from the backside to form the
desired structures in the silicon substrate. On the contrary, surface microma-
chined devices are fabricated from thin films deposited on the substrate. The
surface micromachining technique is compatible with CMOS technology and
allows to integrate the sensor and the interface circuit on the same die. This
reduces the device size and the parasitic capacitances significantly. However,
the smaller dimensions result in smaller capacitance values. Moreover, the sur-
face micromachined (accelerometers and pressure) sensors have a much lower
sensitivity and a larger mechanical noise due to their smaller mass. This re-
sults in harder noise requirements for the readout circuit, which gives a higher
power consumption for the input amplifiers. Hence, we will mainly focus on
bulk micromachined capacitive sensors in our ULP generic capacitive sensor
readout.
In order to characterize the capacitive sensors from different kinds of appli-
cations, we define the mean capacitance C0 and the relative full-scale deviation
Figure 3.1. An electrical model of a single capacitive sensor with sense capacitance Cx , para-
sitic shunt conductance Gp and parasitic capacitances Cp1 and Cp2 .
Generic Sensor Interface Chip 19
as:
Cx,min + Cx,max
C0 = (3.1)
2
C
= (3.2)
C0
where Cx,max and Cx,min are the maximal and minimal capacitance in the given
sensor application. In order to develop a generic capacitive sensor interface with
ULP consumption the following two difficulties need to be solved:
The various capacitive transducer applications that have been reported show
a wide range of mean capacitances and relative full scale deviations. Fig. 3.2
gives a graphical view of and C0 on different types of capacitor sensor
applications found in literature [Mas98, Pue97, Pue00, Lap96, Yaz03, Sel97,
Sei90, Tay00, Pue90, DB02, Cha02, Cha00, Egg00, Kan00, Lac03]. The
and C0 values depend on the type of excitation (acceleration, pressure,
humidity, etc.), the physical input range of the intended application, the
sensor structure and the technology.
The reduction of the effect of the parasitic elements with the lowest power
consumption requires new interface architectures.
The proposed readout circuit provides an interface to single and differential

capacitive sensor applications with 1 pF < C0 < 15 pF, 0.05 < , 200 fF
< C(= C0 ) < 10 pF, Cp1 < 50 pF and Cp2 < 50 pF.
0
10

1
10
Accelerometer
Pressure sensor
Humidity sensor
0 5 C (pF) 10 15
0
Figure 3.2. The relative full-scale deviation, , as a function of the mean capacitance, C0 , for
several capacitive microsystems.
3. Generic Sensor Interface Chip for capacitive sensors

A modular Generic Sensor Interface Chip (GSIC) for capacitive sensors
will be developed. The chip is equipped with many configuration settings
to offer an interface for a broad range of capacitive sensors. Furthermore, it
contains a smart energy management, which adapts the averaged power con-
sumption according to the speed and accuracy requirements of the selected
application.
The GSIC contains a microcontroller interface, a configuration memory and
the following configurable blocks (Fig. 3.3): LF clock, sample timer, reference
and bias circuits, main oscillator and clock generation circuits, Capacitance-
to-Voltage (C-V) converters, Switched Capacitor (SC) amplifier, voltage-to-
current (VI) converter, modulator, decimation counter and conversion timer.
The SC interface converts a capacitance variation C in a proportional voltage.
It consists of two C-V converters and an SC amplifier. In the C-V converters,
the sense capacitance, Cx , is converted to a proportional voltage. The SC am-
plifier amplifies the difference between the outputs of the C-V converters and
produces a quasi continuous input voltage for the modulator (VI converter
and modulator). The main oscillator and clock generation circuits provide the
clock signals to the capacitive sensor interface and the decimation counter. The
reference and bias circuits generate the bias currents for the sensor interface and
the main oscillator. The capacitive sensor interface, the main oscillator, the ref-
erence and the bias circuits are only powered in active mode (Act is high). The
LF clock and sample timer are used for the timing during low power standby
operation. They are the only parts of the system that are operating continu-
ously. Therefore, they are implemented with a very low current consumption
of approximately 500nA.
Reference and bias circuits Configuration Configuration

Vdd
Vss+Vbg
-
LF clock Sample timer
resetLF
+
Act Configuration Configuration

Iptat
F1 ...5 Fmod
Ground Oscillator and Conversion
clock generation timer
ActCV ActSC ActVI Actmod
Rbe
reset resetmod resetcounter Activate
Interface Din
Configuration Microcontroller Dclk
Write
Rref SRAM
Vss
SC interface
F1F1dF2F2d F3F3dF4F4dF5 S-D modulator
Configuration Dav
C-V converters SC VI Decimation Dout
Sensor Ack
and buffered amplifier converter Modulator counter
bitstream
reference
Configuration Configuration
Fmod resetmod
Figure 3.3. Functional description of the Generic Sensor Interface Chip for capacitive sensors.
3.1 Front-end architecture

Over the last decade, several readout circuits for capacitive sensors have
been reported. These circuits can be divided into three groups [Yaz04]: square
wave driven AC-bridge with voltage amplifier, harmonic driven AC-bridge with
voltage amplifier and SC circuit.
A. Square wave driven AC-bridge with voltage amplifier

The square wave driven AC-bridge with voltage amplifier is shown in Fig. 3.4.
The circuit consists of a half bridge, with the capacitors Cx and Cx , driven by
two opposite AC-signals Vref + and Vref . The amplitude of the bridge output
is proportional to the capacitance variation C. This voltage is amplified and
demodulated, which results in an output voltage:
C
Vout = Av Vref (3.3)
2C0 + Cp
This architecture does not eliminate the effect of the parasitic capacitance, Cp .
This degrades the performance significantly.
B. Harmonic driven AC-bridge with voltage amplifier

Fig. 3.5 shows the harmonic driven AC-bridge with voltage amplifier. The
bridge output is held at virtual ground by an op-amp with resistive feedback,
which reduces the effect of Cp . The drive signal, Vm , needs to be sinusoidal
(frequency fdrive ) to avoid errors induced by distortion. If fdrive is smaller than
the bandwidth of the amplifier, the output voltage after demodulation equals:
Vout = 2fdrive Vm Rf C (3.4)
The need for a sinusoidal driving voltage complicates the design of the front-end
significantly.
C. SC circuit
The SC circuit charges the sense capacitors with an opposite polarity and inte-
grates these charges on a capacitor, Cint (Fig. 3.6). Hence we obtain an output
Sync. Vout
Cx Demod. LPF
Vref+
Amp
Vref- Cp
Cx
Figure 3.4. Square driven AC-bridge with voltage amplifier.

Vm+ Cx Rf
Sync. Vout
- Demod. LPF
Vm- Amp
Cp
Cx
+
Figure 3.5. Harmonic driven AC-bridge with voltage amplifier.
voltage, which equals:

C
Vout = Vref (3.5)
Cint
This circuit also eliminates the effect of the parasitic capacitances and does not
need complex driving voltages.
D. Open loop ULP architecture

Many sensor systems reported in literature use off-the-shelf analog-to-digital
converters. This approach complicates the design of the analog front-end, be-
cause a buffered analog voltage has to be transferred to the ADC chip. As shown
in Fig. 3.7, the system can be considerably simplified by merging the analog
part of a first order ADC with the front-end circuit and by implementing the
digital filtering and processing in the microcontroller or DSP [Mei02, Rie93].
Many capacitive sensor interfaces use the sensor directly in a modulator
structure [Lem99, Wan98, Kul06, Kaj02]. This leads to high power consump-
tion, because the capacitors must be charged and discharged on the rhythm of
the high oversampling clock of the modulator. Most of these sensor interfaces
are designed for closed loop accelerometers. In these circuits, the electrostatic
feedback force is used to keep the sensor mass in its balanced position, which
results in a high linearity. Hence, the mechanical transfer characteristic acts
Freset
F1
Vref+ Cx Cint
-
Vout
Amp
Vref- Cp LPF
F2 Cx
+
Figure 3.6. Switched Capacitor circuit.

Smart signal processor

Feedback
1 bit DAC
__
Sensor Gate Microcontroller
Sensor interface Filter Comparator clock Bitstream or DSP
+
Clock
Non el. Analog Digital
Figure 3.7. Smart sensor system architecture with first order modulator.
as a second order filter in the loop. As a consequence, the stability and

the performance of the readout circuit strongly depend on the specific sensor
[Pet06]. Moreover, these systems need an important start-up time to bring the
sensor mass close to equilibrium [Kul03]. So, it is not possible to operate them
in a power efficient duty cycle.
To combine the advantages of a modulator structure and still maintain
low power consumption, we introduce a new open loop architecture, which
uses a low clock frequency, 8 kHz, for the SC interface (C-V converters and a
SC amplifier) and a higher clock frequency, 128 kHz, for the modulator
(Fig. 3.8). Reducing the clock frequency of the SC interface increases the
influence of the parasitic shunt conductance. Since the shunt conductance is
highly dependent on the pollution and condensation, it can cause a serious
reliability problem. So, the decrease of this effect is an important issue in the
design of an ULP interface for capacitive sensors. The effect can be reduced
by performing a dedicated series of eight measurements as explained in [Li00].
However, the method is not power efficient, since it requires a long measurement
time.
Cref1
Gp Fchop
Capacitance
to voltage
converter
Cx Cp2 Fchop
Ch
F1 F5
SC Sigma
Amplifier Delta
Vref Cref2 Modulator
F2 Cp1 Gp F5
Ch
Capacitance
to voltage
converter
Cx
Cp2
Figure 3.8. Capacitive sensor readout architecture.

This work presents a C-V converter, which uses class AB circuit techniques
and Correlated Double Sampling (CDS) operation to reduce the influence of
the parasitic shunting conductance while maintaining a low clock frequency
and low power consumption. The C-V converter performs charge leakage sup-
pression, which is several times higher than more conventional designs with the
same power consumption. The front-end also contains an enhanced chopping
scheme, which eliminates the effect of mismatches between both C-V convert-
ers. The capacitive sensor interface has two modes of operation. The first mode
is for single sensor operation with on chip reference capacitor, where the ref-
erence capacitor Cref 2 needs to be programmed to approximate C0 . The other
mode is for differential sensor operation, where the on chip reference capaci-
tor Cref 1 (or Cref 2 ) is programmed to compensate for the offset between Cx
and Cx . In both modes the amplification factor ASC of the SC amplifier and
the feedback capacitor Cf need to be programmed for optimal accuracy of the
interface.
3.2 Capacitance-to-Voltage convertor

Fig. 3.9 shows a conventional state-of-the-art capacitance-to-voltage con-
verter. It is known that this circuit is very effective in reducing the effects of
the parasitic capacitors Cp1 and Cp2 .
During the sampling phase 1 , the sense capacitor Cx is charged. During
the signal phase 2 , Vp becomes a virtual ground and the charge is transferred
to the feedback capacitor Cf . At the end of the signal phase, assuming an
ideal charge transfer, the voltage at the output of the C-V converter equals
Vref Cx /Cf . In reality the charge transfer will be imperfect. During 2 a part
of the signal charge is leaking away through the parasitic shunt conductance Gp .
This leakage charge has three contributions. The first contribution is due to the
finite transient response of the Operational Transconductance Amplifier (OTA),
which causes the potential Vp to settle in a certain time to the virtual ground
potential, and during this transient time a charge will leak away. The second
F1d
Gp
F1d Vp F2 Cf
Vref
-
Cx
F2d F1 + Cl
Cp1 Cp2
Figure 3.9. Conventional state-of-the art Capacitance-to-Voltage converter.

contribution is due to the offset voltage of the OTA, which creates a leakage
charge proportional to Vof f set Gp Tphase2 . The third contribution originates
from the finite DC gain Av of the OTA, which in turn leads to a leakage charge
proportional to (Vout /Av )Gp Tphase2 .
In [Li02], Li et al. use the conventional state-of-the-art C-V converter in
combination with a chopping technique and the three-signal autocalibration
method. This interface eliminates the offset leakage and compensates for the
drift in the readout electronics. Unfortunately, the technique is not energy
efficient, since it requires four measurement cycles and three extra voltage
sources to determine Cx .
Fig. 3.10 shows the proposed ULP C-V converter. This interface needs only
one voltage source Vref and one measurement cycle to determine Cx . It uses
Correlated Double Sampling (CDS) to eliminate the effect of the offset voltage
on the sensor interface [Lam83]. During 1 , the capacitor Cs samples the offset
voltage. During 2 , the sampled offset is subtracted from the instantaneous one.
The C-V converter uses a class AB OTA with cascode output stage
(Fig. 3.11, 3.12). The class AB operation is a power efficient solution to re-
duce the transient leakage. After the transition from phase 1 to phase 2 , the
tail current of the OTA is boosted, which speeds up the charge transfer from
the sense capacitor to the feedback capacitor. During settling, the voltage Vp
approaches the virtual ground and the tail current falls back to a low quiescent
level.
In order to understand the operation of the OTA circuit [Har99], it is important
to note that the common source voltage of M1 and M2 is forced by the internal
negative feedback to follow the larger of the two voltages Va and Vb . When
Vin is smaller than Vin+ , the output voltage Voutop2 of the op-amp op2 is
pulled to the positive supply and the common source voltage Vs equals Vb . So,
Vs is only determined by Vin+ , the bias current Ibias and the dimensions of the
transistor Mb . Since the input transistors are biased in weak inversion mode,
Cf F2
Gp F1d F1d
F1d Cs
Vref Vp
-
Cx
F2d F1 + Cl
Cp1 Cp2
Figure 3.10. Capacitance-to-Voltage converter with correlated double sampling and class AB
operation.
M9 M10
- M3 M4 -
Vb Va
Op1 Ibias Op2
+ +
M12
Vs
Vpb
M1 Ma Mb M2
Va Vb
Vout
Vin- Vin+
M11
Vnb
Cl
M7 M5 M6 M8
Figure 3.11. Class AB OTA.
M3a M4a M3
Voutop
Cc
M1a M2a
Vs
Vina+ Vina-
Ibiasop
Figure 3.12. Internal feedback op-amp op.
current boosting for small Vin = Vin+ Vin is proportional to:

IM 1 Vin+ Vin
= exp (3.6)
IM 2 nUT
where n is the weak inversion slope factor and UT = (kT /q) is the thermal
voltage. For larger Vin the input transistors leave the weak inversion region
and the current boosting is less steep than predicted by equation 3.6. Input tran-
sistors with larger W/L ratio improve the current boosting, since they operate
in a deeper weak inversion mode.
The internal feedback op-amps, op1 and op2, need to be fast Miller OTAs to
realize a short tracking time and a high current efficiency. The stability of this
feedback mechanism is guaranteed by the compensation capacitor Cc . The op-
amps, op1 and op2, are loaded by the gate source capacitances of M1 and M2 .
Hence, larger dimensions (W/L) of M1 and M2 result in a larger Cc and a higher

current consumption of op1 and op2 to maintain a fast tracking response. So,
an optimal AB OTA needs to consider both the current boosting capabilities and
the internal feedback behavior. In our designed AB OTA, the internal feedback
mechanism has a phase margin of 81 deg (small signal approximation) and a
tracking time smaller than 5 s. The op-amps, op1 and op2, consume a current
of 125 nA. The current boosting, IM 1 /IM 2 , equals 85 for an input voltage,
Vin , of 0.4 V (OTA in unity-gain feedback configuration). The class AB OTA
uses a cascode output stage to achieve a high gain.
The clocks 1d and 2d are slightly delayed with respect to 1 and 2 .
This makes the charge injections appear as an offset on the outputs of the C-
V converters [Joh97]. The SC amplifier amplifies the difference between the
outputs of both C-V converters. So, the effects of charge injection in the C-V
converters are eliminated.
In order to minimize power consumption, one should choose the interface
clock frequency as low as possible. Our ability to reduce the clock frequency
is limited by the accuracy considerations. First of all, a low clock frequency
will make CDS less effective in reducing 1/f noise. Secondly, lowering the
clock frequency results in undersampling the OTA noise bandwidth by a too
high factor and degrades the systems noise performance. Lastly, a low clock
frequency will increase the effects of electrostatic forces on accelerometers
[Bao00, Pue96]. Depending on these arguments, calculations show that a clock
frequency of 8 kHz is an optimal choice for our interface.
Figure 3.13. Charge leakage during signal phase 2 .

In order to compare a conventional C-V converter to the proposed one, both

structures are biased (Fig. 3.9 vs. Fig. 3.10) with the same average current
consumption (3.5 A). Fig. 3.13 shows the charge leakage during phase 2
for both C-V converters with the following properties: Vref = 1 V, Cx = 10 pF,
Cp2 = Cp1 = 5 pF and Gp = 0.01 S, feedback capacitor Cf = 27 pF, Cs = 5 pF
and Cl = 10 pF. It can be seen that the new C-V converter performs a much faster
charge transfer. The offset and gain leakage are almost perfectly eliminated by
the high DC gain OTA and the CDS operation. Our C-V converter has a charge
leakage, which is approximately three times lower than the traditional one. The
simulated leakage charge equals 7 fC, which is approximately 0.07 % of the
signal charge. So, the error is smaller than 10 bits for this application.
3.3 Chopping scheme

The mismatch between the switches and feedback capacitors of both C-V
converters causes errors in the transfer characteristic of the sensor front-end.
It also makes the interface more susceptible to interference. This can cause
an extra loss in accuracy. With mismatches, we obtain the following outputs,
VCV,+ and VCV, , for the C-V converters:

Qof f set 1
VCV,+ = Q+ + Qof f set + + QEM I C
(3.7)
2 Cf + 2 f

Qof f set 1
VCV, = Q + Qof f set + QEM I C
(3.8)
2 Cf 2 f
Q Q
where Q+ and Q , Qof f set + of f set
2 and Qof f set of f set
2 , QEM I and
Cf Cf
Cf + 2 and Cf 2 are the sampled sense charges on Cx and Cx ,
the charge injections, the common mode electromagnetic interference and the
feedback capacitors.
These equations can be simplified to:
Q+ Q Q+ + Q Cf Qof f set QEM I Cf
VCV,+ VCV, = + + +
Cf Cf 2Cf Cf Cf Cf
(3.9)
This corresponds to a mismatch induced error:

Q+ + Q Cf Qof f set QEM I Cf
Errormismatch = + + (3.10)
Q+ Q 2Cf Q+ Q Q+ Q Cf
The proposed chopping scheme provides a solution for this problem. In this
scheme, a pseudo differential structure is built, where the capacitive sensor ele-
ments (Cx and Cx ) are connected to each C-V converter for an equal number of
Conversion time (= 2 ms)
+ - - ++ - - ++ - - ++ - - + Enhanced chopping
+ - + - + - + - + - + - + - + - Conventional chopping
Ts (= 125 ms)
Figure 3.14. Conventional and enhanced chopping scheme during one modulator conver-
sion period.
1
Enhanced chopping
Conventional chopping
Interference reduction
0.8 Without chopping
0.6
0.4
0.2
0
0 0.1 0.2 f/f 0.3 0.4 0.5
s
Figure 3.15. Frequency dependence of the interference reduction without chopping, with con-
ventional chopping and with enhanced chopping.
interface periods. This modulates the effects of the mismatch with the chopping
sequence. These components are filtered by the low pass operation of the
modulator.
Fig. 3.14 compares the interference reduction of the enhanced chopping
with a conventional chopping scheme for a conversion time of 2 ms.
The enhanced chopping scheme is more efficient in eliminating low frequency
interference (f < 0.1fs , where fs is the SC interface frequency of 8 kHz)
(Fig. 3.15).
3.4 SC amplifier
Fig. 3.16 shows the fully differential SC amplifier, which amplifies the dif-
ference between the outputs of the C-V converters [Mar87]. It uses a correlated
F4 C3
F3 C2 F3d
Ch
Vin+ F3d C1 F4d F5 Vout+
+ -
F4
- +
Vin- F C1 C2 F4d F5 Vout-
3d Ch
F3 F3d
C3
F4
F3
F3d
F4
F4d
F5
Figure 3.16. Fully differential SC amplifier with slew enhanced Correlated Double Sampling
scheme.
double sampling scheme, which does not require any resetting of the output in
each clock period. This topology is power efficient, since it allows more relaxed
op-amp specifications for low frequency inputs. To reduce the influence of the
charge injection, the clocks 3d and 4d are slightly delayed with respect to
3 and 4 . The outputs of the SC amplifier are sampled on the hold capacitors
Ch and Ch during phase 5 , so a quasi continuous output voltage is provided
to the input of the analog-to-digital converter.
The behavior of the SC amplifier can be characterized in a state model.
Fig. 3.17 shows the half circuit for the differential mode characteristics.
F4 C3
F3 Q3 F3d
C2
Vin F3d C1 Q2 Ch
- F4d F5
Va Vout
F4 Q1
+
Figure 3.17. Half circuit of the SC amplifier (study of differential characteristics).
During the sampling phase 3 , we obtain the following set of equations:

q1,n 1 = Vin,n 1 Va,n 1 C1 (3.11)
2 2 2
q2,n 1 = Va,n 1 C2 (3.12)

2 2

q3,n 1 = q2,n 1 q2,n1 q1,n 1 q1,n1 + q3,n1 (3.13)
2 2 2
q3,n 1
Vout,n 1 = Va,n 1 2
(3.14)
2 2 C3

Vout,n 1 = Adif f Va,n 1 Vof f set (3.15)
2 2
where Adif f and Vof f set are the differential gain and the offset of the OTA.
During the signal phase 4 , we obtain the following set of equations:
q1,n = Va,n C1 (3.16)

q2,n = q1,n q1,n 1 + q2,n 1 (3.17)
2 2
q3,n = Vout,n C3 (3.18)

q2,n
Vout,n = Va,n (3.19)
C2
Vout,n = Adif f (Va,n Vof f set ) (3.20)
With these sets of equations, we can derive a discrete state model of the form:
Xn = AXn1 + BVn (3.21)
Yn = CXn (3.22)
In this model, we take the state vector Xn = (x1,n = q1,n 1 , x2,n = q2,n 1 ,
2 2
x3,n = q3,n 1 , x4,n = q1,n , x5,n = q2,n , x6,n = q3,n ), the input vector
2
Vn = (v1,n = Vin,n 1 , v2,n = Vof f set ) and the output vector Yn = (y1,n =
2
Vout,n 1 , y2,n = Va,n 1 , y3,n = Vout,n , y4,n = Va,n ).
2 2
The step response of the amplifier converges to a voltage

C1 C1 + C2 C1 + C2 Vof f set
lim Vout,n = Vin 1 2 + (3.23)
n C2 C2 Adif f C2 Adif f
Hence, the gain error is reduced with a factor 1/(Adif f 2 ). This would allow
the use of a low gain OTA. However, the transient response is also affected by
the differential gain of the OTA. Adif f needs to be larger than 10000 to achieve
an error smaller than 0.004 after one clock period (Fig. 3.18).
The SC amplifier uses a fully differential folded cascode OTA with SC com-
mon mode feedback to achieve a high gain (Fig. 3.19). The input transistors of
the OTA are biased in weak inversion to reduce the power consumption. The
phase margins of the OTA and the common mode feedback equal 78 deg and
82 deg. The phases 3 , 4 and 5 take 3/8, 5/8 and 2/8 of the clock period.
The equivalent load capacitor, Cload , equals:
8 8
Cload = (Cl + Cc,cmf b + Cs,cmf b ) = Ch (3.24)
3 2
0
10
A =1000
diff
1 A =100
(Vout,nVout,ideal)/Vout,ideal
10 diff
A =10000
diff
2
10
3
10
4
10
5
10
6
10
1 2 3 4 5 6 7 8 9 10
n
Figure 3.18. Transient error as a function of the clock cycle, n.

M10 M11
Vcmfb Vcmfb
Voutmin Voutplus
M9
M8
M12 M13 M14 M15
M1 M2
Reset_ Reset Reset_ Reset
Vinplus Vinmin
Vpb Vpb
Cl Cl
M6 M7
Vnb Vnb
M29 M30
Reset Reset
M4 M3 M5
Vna Vna Vna
F3d_ F4d_ F4d_ F3d_

M19 M18
M16
M26 M25
M28
Voutplus Voutmin
Ground Ground
F4d F4d
Cs,cmfb Cs,cmfb
Cc,cmfb Cc,cmfb
F4d_ F4d_
M21 M20
M24 M23
Vpa Vcmfb Vpa

M17
M27
M22
Reset_
F3d_ F4d F4d F3d_
Vpa
Figure 3.19. Differential folded cascode OTA with SC common mode feedback.
where the contribution of the feedback capacitor CC12+C

C1
2
in the load capacitor
C2 C1
can be neglected, because C1 +C2 << Ch , Cl .

gm
The equivalent GBW = 2C load
of the OTA equals 160 kHz, which is 20
times the clock frequency. Since the SC amplifier has a maximal programmable
gain of 16, the settling error will be smaller than exp(2 20 16 ) 0.05%.
So, the SC amplifier performs an adequate settling behavior for each gain
setting.
3.5 modulator
The modulator structure allows one to adapt the system and its energy
consumption to the selected sensor application. The accuracy of the ADC is
determined by the conversion time, i.e., the number of oversampling clock
periods that are used to acquire the digital bit code. Since many sensor signals
have a very small bandwidth (typical order of a few tens of Hz) and a medium
resolution is sufficient (8-10 bits), the system can operate in a duty cycle, where
the analog readout circuitry is only for a short period of time in the ON-state.
Other important advantages of the architecture are the immunity against
digital interference and locking effects and the relaxed specifications for the
analog components [Bos88].
The modulator and the digital decimation filter become more complex with
increasing order of the sigma delta ADC. For higher order modulators the risk
for instability is higher, so only 1st and 2nd order modulators can be realized
with a reduced complexity [Jes00]. The choice between a 1st and 2nd order
modulator is an important dilemma. A 2nd order modulator has the advantage
that it gives 6 dB per octave oversampling ratio more resolution. Hence, the
measurement time can be decreased to achieve the same resolution as in a
1st order modulator. However, a 2nd order structure needs a higher order
decimation filter and a more complex modulator. A 1st order modulator can
use a simple digital counter as decimation filter. This counter can easily be
implemented on the sensor interface chip, which reduces power consumption.
Considering the overall medium resolution and low speed requirements, we
have opted for a 1st order modulator. For more accurate and faster appli-
cations a 2nd order modulator would probably be a better choice.
Traditionally, most of the modulators are SC realizations. However the
first order modulator presented in this work is a Continuous Time (CT)
implementation. For this type of modulator the bandwidth and slew rate re-
quirements are less stringent, which reduces the power consumption and makes
the structure less sensitive to noise and digital interference. On the feed-
back side, the reference voltage of a SC modulator requires buffering to attain
the oversampling speed (128 kHz). The DAC of the CT integrator can be
implemented with current sources, which do not load the voltage reference dy-
namically. The main drawback of this structure is the accuracy limitation by
the non-linearity of the voltage-to-current converter. For medium resolution
applications this does not pose any problem, because the total harmonic distor-
tion of the voltage-to-current converter can easily be better than -60 dB. Most
of the CT modulators presented in literature are higher order modulators, which
are designed for medium and high-speed applications. However, our first order
sigma delta modulator is an ULP low frequency design for autonomous sensor
systems. So, a review of the dominant error sources is necessary to achieve an
optimal design for our application. The most important issues, that have been
addressed in the design of the CT modulator, are given below.
The charge injections, induced by a falling and a rising transition in the
feedback DAC, are not perfectly equal in magnitude. So, every falling and
rising transition pair injects an extra charge in the integrator capacitor, Cint ,
which results in an accumulating error charge on Cint . This charge injec-
tion imbalance creates a non-linearity, which is larger in the middle of the
modulator input range, since at zero input the highest number of transitions
occurs. The charge injection imbalance has several causes: the difference
between the rising and the falling edge delay time of the quantizer output
[Ada86] (this phenomenon is more dominant for high speed modulators),

the mismatch between the DAC switches and the residual voltage between
the positive and negative input of the integrator op-amp.
The current leaking away from the integrator can also cause non-linearity
problems [Fee91]. This effect can be reduced by using an integrator op-amp
with a high DC gain and by using a current DAC and a voltage-to-current
converter with high output impedances.
The clock jitter creates a random variation in the timing. This introduces a
random change in the amount of charge delivered to the CT loop between
successive iterations [Che99].
The non-idealities of the quantizer.
The noise current injected in Cint limits the dynamic range of the modulator.
The total noise contains contributions of the voltage-to-current converter and
the current DAC.
The CT modulator can be a Non-Return-to-Zero (NRZ) or a Return-to-
Zero (RZ) implementation. For NRZ modulators, the charge injection depends
on the previous DAC symbols. In this case, the feedback DAC codes (101)
and (110) have a different effect on the charge transfer to Cint . This creates
a distortion in the conversion result. These effects are eliminated in a RZ
modulator. During each clock cycle, this modulator switches according the
quantizer decision for part of the time and resets to zero for the rest. In this
way, independent of the quantizer decision, both a rising and a falling edge
appear in the DAC pulse. Hence, the intersymbol interference is reduced. On
the other hand, the clock jitter is lower in NRZ than in RZ modulators, since
NRZ modulators have a smaller number of transitions. Because the charge
injection imbalance dominates the clock jitter in our design, we have chosen
for a RZ implementation.
A. RZ CT modulator
Fig. 3.20 shows the proposed RZ CT modulator. After the reset signal
becomes low, switches 1 and 3 are closed and switches 2 and 4 are open. The
output current of the VI converter is integrated on the capacitor Cint . The
regenerative comparator executes his decision during the modulation phase
mod . At the end of this phase, the output of the comparator is settled and
saved in the following latch. After the rising edge of the feedback clock f b ,
the switches in the current DAC are stimulated. If the state changes, the break
operation is executed before the make operation. The clocks zero1 and zero2
are created out of mod by a non-overlapping clock generator circuit. The delay
is chosen such that the falling edge of zero1 comes after the rising edge of f b .
Fmod
Fzero2
M1 M3 Delay
switch1 switch3 Ffb
Reset
Iref Iref
Fmod
Delay
Fzero1
Reset switch1
switch2 switch4 Reset Vout1
Ffb clock switch2
Vout2
Vin+ M2 M4 S
VI Cint Q R
converter -
Fzero2
Vin- Fzero1
+
switch3
Fmod S Vout1
Q_ R switch4
Vout2
Ffb clock
Reset
Figure 3.20. Simplified schematic of the RZ CT modulator with VI converter, a current

DAC, an integrator, a comparator, a break-before-make feedback scheme and clocks.
Hence, the influences of charge injections due to transitions in the switches 1

to 4 are eliminated.
B. Voltage-to-current converter
Fig. 3.21 shows the VI converter, which transforms a differential voltage into a
single ended current [Kwa91]. It contains a fully differential core to eliminate
the even order distortion components. It uses p-type input transistors biased in
subthreshold region with substrate to source connection, so the non-linearity and
noise of the input transistors are limited. It has a high-impedance cascode output
stage to reduce the integrator leakage. The total harmonic distortion decreases
M15 M3 M4 M16
Vpa Vpa
M13 M1 M2 M14
Vpb Vpb
Vin+ Vin-
Iout
M11 M12
Vnb C1 C2 Vnb
M9 M5 M7 M8 M6 M10
Vna Vna
Figure 3.21. The voltage-to-current converter.

with increasing bias current. The voltage-to-current converter uses 1.8 A.

This allows a total harmonic distortion of -73 dB for a 0.5 Vpp differential sine
input with frequency 100 Hz.
The VI converter has an internal feedback mechanism, which causes unsta-
bility when the compensation capacitors, C1 and C2 , are omitted. The small
signal scheme is used to study the frequency behaviour of the VI converter
(Fig. 3.22).
The analysis of this network gives the following closed loop transfer function:
Vout T
= (3.25)
Vin N
Ct Cc s2 + Cn1 gm1 s + gm1 gm7

Cc (Ct + Cn1 + Cn2 )s2 + Cc (gm7 + 2G)s + gm1 gm7
The equivalent open loop transfer function can be calculated as:
TOL T
= (3.26)
NOL N T
Ct Cc s2 + Cn1 gm1 s + gm1 gm7

Cc (Cn1 + Cn2 )s2 + Cc (2G + gm7 )s + 2G(go1 + go5 ) + gm7 go1
The internal loop gain of the designed VI converter (C1 = C2 =10 pF) has a
GBW of 33 kHz and a phase margin of 85 deg (Fig. 3.23).
The simulated noise output current during one modulator period ( 8s) equals
190 pARM S . This is much smaller than the required 1.25 nA, so the noise of
the VI converter is negligible.
go3
Ct Vout 2G
Vin
Cn2
gm1(Vin-Vout) go1
Cc go7
V1
Cn1 go5 gm7 V1
Figure 3.22. Half circuit of the input stage of the VI converter (small signal approximation).
Figure 3.23. Bodeplot of the equivalent open loop transferfunction of the VI converter (C1 =
C2 =10 pF).
C. Current DAC
Fig. 3.24 shows the current DAC, which uses a cascode output stage and a break-
before-make timing scheme. This break-before-make timing scheme consists
M15 M14 M1
Vpa
M16 M13 M2
Ground Vpb
Switch1 Switch2
Actmod
M3 M4
R1 R2 Ground Iout
M5 M6
M9 M11 M7
Switch3 Switch4
Vnb
M10 M12 M8
Vna
Figure 3.24. The current DAC.

M5 M7
Vout1 Vout2
M1 M3 M9 M11
Reset Clock Clock Reset
M2 M4 M6 M8 M10 M12
Vss Vin1 Vin2 Vdd
Figure 3.25. Ratioed SR latch with reset and clock switches.
of two SR latches (Fig. 3.25). When the outputs of the quantizer (Q and Q )
change, the latches first open the inactive switches and thereafter they close
the active switches. This ensures that at all times the switched current source
has a high output impedance. The simulated noise output current during one
modulator period ( 8s) equals 100 pARM S (=(Iref +,noise + Iref ,noise )/2).
This is much smaller than the required 1.25 nA, so the noise of the current DAC
is negligible.
D. Integrator
The integrator uses a two-stage op-amp to achieve a high DC gain (Fig. 3.26).
This limits the integrator leakage. After every transition, a residual voltage
M7 M11 M8 M9
Vpa
M1 M2
Vin- Vin+
C
Vout
M5 M6
Vnb Vnb
M3 M4 M10
Vna Vna Vna
Figure 3.26. The integrator op-amp.

swing Vr occurs at the inverting input of the integrator (Fig. 3.27).

2Iref
Vr (3.27)
2GBW Cint
This results in a charge injection imbalance,
2Iref
Qimbalance (CpDAC+ CpDAC ) (3.28)
2GBW Cint
where CpDAC+ and CpDAC are the parasitic capacitors of the DAC in the
positive respectively negative state. The integrator has a GBW = 250 kHz,
Cint = 10 pF and Iref = 125 nA to reduce this charge injection imbalance. The
phase margin of the integrator op-amp equals 72 deg.
E. Comparator
Fig. 3.28 shows a regenerative comparator, which is used as a quantizer [Yin92].
This discrete time comparator combines a fast response with low power con-
sumption. The comparator consists of a differential input pair (M2 /M3 ), a top
and bottom regeneration loop (M11 /M12 and M4 /M5 ) with transfer transistors
(M6 /M7 ) and pre-charge transistors (M9 /M10 ) and a switch for resetting (M8 ).
The comparator operates in three phases: the reset phase, the bottom and top
regeneration phases. latch is high and reg is low in the reset phase. This
disconnects and resets the bottom and top regeneration latches. The differential
pair injects a differential current, proportional to the comparator input voltage
difference, into the bottom regeneration loop and generates a voltage difference
Figure 3.27. The residual voltage during a DAC transition (NRZ mode).
over M8 . This voltage will act as the initial imbalance for the regeneration.
When latch goes down, the imbalance is regenerated by the bottom regenera-
tion loop (bottom regeneration phase). After reg rises, the bottom and the top
regeneration loops are connected and they both start to regenerate the imbalance
(top regeneration phase).
The offset of this comparator is determined by the mismatch between the
input pair and the bottom regeneration loop. The mismatches in the top regen-
eration loop and the transfer switches can be neglected, since the imbalance
is already significantly larger at the start of the top regeneration phase. The
implemented comparator has a 3 offset of approximately 12 mV. This effect
is negligible in our modulator.
A non-overlapping clock generator creates the clocks for the comparator
(Fig. 3.28). The clock generation circuit uses two different delay cells: delay
cell 1 implements a large delay between 200 ns and 400 ns, dependent on the
supply voltage and process tolerances, whereas delay cell 2 gives a delay of a
few ns. Both delay cells are placed in the feedback path. Hence, the delays
Freg
M1 M9 M11 M12 M10

Vpa
M2 M3
Vinplus Vinmin
Voutmin Voutplus
M6 M7
Flatch
M8
M4 M5
Fmod_ Delay cell

Flatch
Delay 1
Delay 2
Freg
Fmod
Figure 3.28. The regenerative comparator with clock generation circuit.

only effect the start of the reset and top regeneration phases (latch and reg
become high), but they have no impact on the exact timing of the quantizer
decision phase (reg goes down). Consequently, the operation of the quantizer
is immune to power supply variations and process tolerances.
F. Performance
A decimation counter is used at the output of the bitstream to obtain a full digital
code. The conversion time of the modulator is programmable between 256, 512
and 1024 clock cycles. So, the resolution can be varied between 8, 9 and 10 bits.
The ADC performance is measured on the final GSIC circuit (section 6).
Figs. 3.29, 3.30, 3.31 and 3.32 compare the measured Integral and Differential
Non-Linearity (DNL and INL) of the proposed RZ modulator with a NRZ
version for a modulator clock of 128 kHz and a conversion time of 8 ms (=1024
clock cycles). Both the RZ and NRZ modulator achieve the specified resolution
of 10 bits (DNL and INL are smaller than 0.5 LSB). The NRZ modulator has a
larger DNL near midrange. This is caused by charge injection imbalance.
3.6 Bandgap reference, bias system and buffered reference

voltage
The bias and reference system of the sensor interface consists of a bandgap
reference, a reference current, bias branches for the main oscillator, modulator,
VI converter, SC amplifier and C-V converters and a buffered reference voltage
for the C-V converters.
Differential nonlinearity error

0.5
0.4
0.3
DNL error (LSB)
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
500 400 300 200 100 0 100 200 300 400 500
Output code
Figure 3.29. The measured differential non-linearity of the RZ modulator (10-bit mode).
Differential nonlinearity error

0.5
0.4
0.3
DNL error (LSB)
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
500 400 300 200 100 0 100 200 300 400 500
Output code
Figure 3.30. The measured differential non-linearity of the NRZ modulator (10-bit mode).
Integral nonlinearity error

0.5
0.4
0.3
INL error (LSB)
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
500 400 300 200 100 0 100 200 300 400 500
Output code
Figure 3.31. The measured integral non-linearity of the RZ modulator (10-bit mode).
A. Bandgap reference and bias system

Fig. 3.33 shows the bandgap reference and bias system. The bandgap reference
circuit creates a voltage, which is independent of the power supply voltage
and the temperature. The bandgap reference circuit adds a Proportional To
Absolute Temperature (PTAT) voltage to the base emittor voltage, Vbe , of the
bipolar transistor Q3 . This results in a temperature stable bandgap voltage, Vbg .
Integral nonlinearity error

0.5
0.4
0.3
INL error (LSB)
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
500 400 300 200 100 0 100 200 300 400 500
Output code
Figure 3.32. The measured integral non-linearity of the NRZ modulator (10-bit mode).
Vertical bipolar transistors, implemented in CMOS technology, have well

known temperature characteristics and show almost no dependency on process
parameters [Wan00]. The base emittor voltage, Vbe , can be written as [Mei86,
Bak00]:
kT IC
Vbe = ln (3.29)
q Is
where Is is the saturation current and IC the collector current. The saturation
current Is strongly depends on the temperature, according to the equation:

qVg0
Is = CT exp (3.30)
kT
where Vg0 is the extrapolated bandgap voltage at 0 K and C and are constants.
Taking this into account, we can write equation 3.29 as
kT IC
Vbe = Vg0 + ln (3.31)
q CT
Because Ic /CT < 1, the value of the ln function is negative, which results in a
negative temperature coefficient for Vbe . The collector current IC is proportional
to the absolute temperature (tempco KIC ) in the chosen bandgap circuit. So,
we can write equation 3.31 as:
kT KIC T
Vbe = Vg0 + ln (3.32)
q CT
M13 M38
act act
M24 M25 M26 M27 M28 M29 M30 M31
M3 M4 M5 M6
Vbias
M23
M7 M8 M10 M9
Vbias M32 M33 M34 M35 M36 M37
gnd
Cstart M21 M22

Generic Sensor Interface Chip
ActCV ActSC ActVI Actmod
Figure 3.33.
Vbg
M11 M12 M14 M15
nact act
IbiasCV IbiasSC IbiasVI Ibiasmod IbiasMCLK
C
M43
Vbiaspa
R2
R1
M16
nact M17 M18 M19 M20
M42
M2 Rref
M1
Q1 Q2 Q3 Ibias Vbiaspb
R3 R4
M38 M41
Bandgap reference and bias system.

Vbiasnb
M39 M40
Vbiasna
45
With the base emitter voltage, Vbe (Tr ), at the reference temperature Tr , we can
rewrite this equation as

T T kT T
Vbe = Vg0 1 + Vbe (Tr ) + ( 1) ln (3.33)
Tr Tr q Tr
Equation 3.33 can be written as the sum of a constant term (Vg0 ), a linear term
(T ) and higher order terms (O(T 2 )):
Vbe (T ) = Vg0 T + O(T 2 ) (3.34)
where
kTr
Vg0 = Vg0 + ( 1) (3.35)
q
1
= Vg0 Vbe (Tr ) (3.36)
Tr

k Tr
O(T ) = ( 1)
2
T Tr + T ln (3.37)
q T
The cascode mirrors (M4 /M7 , M5 /M8 ) provide the same current through
the bipolar transistors Q1 and Q2 . Both transistors are implemented with a ratio
of 1:n (Q1 = 1 element, Q2 = n elements). Hence, the voltage difference over
the resistor R1 equals
kT
Vbe1 Vbe2 = ln n (3.38)
q
The bias current through R1 is proportional to the absolute temperature. This
current is mirrored by the transistors M6 /M9 , which results in a PTAT voltage
R2 kT
Vptat = ln n (3.39)
R1 q
This PTAT voltage compensates the negative tempco of the Q3 base emitter
voltage for n = 10 and R2 /R1 = 8. The resistor R1 has a nominal value of
125 k. This sets the bias current in the bandgap circuit to 500 nA (300 K).
Hence, the circuit performs an adequate start-up behavior (a few tens of s) and
reduces the low-level injection effects [Wan00] in the bipolar transistors. The
circuit provides a bandgap voltage Vbg of 1.135 V, with a variation of 0.3 % in
the specified temperature range from 40 C to 85 C.
A buffer op-amp, with load resistor Rref , converts the bandgap voltage
into a reference current. The buffer op-amp uses a symmetrical input core
to enhance the Power Supply Rejection Ratio (PSRR) and to eliminate the
systematic offset. A Miller compensation capacitor, C, is added to obtain an
overdamped frequency response with GBW of 15 kHz and a phase margin of
77 deg. The resistor Rref is placed in close proximity of the resistor R from the
voltage-to-current converter (common centroid layout). So, the operation of the

system will be in first order independent of the temperature. The cascodes in the
reference circuit make the bias currents immune to power supply variations. The
bias currents are distributed from the central reference circuit to the slave bias
circuits. Hence, the bias voltages for the n- and p-type (cascode) transistors are
generated locally for each building block. This reduces the effects of systematic
mismatches between distant regions on the chip. The C-V converters (and
buffered reference voltage Vref ), SC amplifier, VI converter and modulator can
be switched off separately, by opening a switch in their bias branches. A switch
in the start-up circuit of the bandgap reference offers a power down option for
the total analog part of the system (act = 0).
B. Buffered reference voltage

The buffered reference voltage needs to charge the sensor capacitances Cx , Cx
and Cp1 on the rhythm of the SC interface clock (8 kHz). It is realized by
converting the bias current with a feedback resistor Rf to a voltage Vref above
analog ground (Fig. 3.34). The two stage op-amp contains a symmetrical cas-
code input pair followed by a class AB output buffer with adaptive load [You98]
(Fig. 3.35). The symmetrical input core reduces the systematic offset and en-
hances the power supply rejection. The class AB buffer is a power efficient
solution to provide high peak currents at the output load. After 1 turns on,
the transistor M19 leaves the saturation region (Vds < Vgs VT ). Hence, node
A becomes high impedant and an increased current flows from transistor M20
to the sensor capacitances. When the output voltage approaches its final value,
transistor M19 returns to saturation region and the adaptive load becomes low
impedant. Hence, the current consumption decreases to a low quiescent level.
The implemented class AB buffer consumes 650 nA in the input stage (bias
transistor M3 ), 250 nA in M17 /M13 and 75 nA in M19 /M15 . With these
settings, the output (M20 /M16 ) has a quiescent current of 2.5 A. The com-
Rf
-
M2 Vref
Vbiasnb
+
ground
M1
Vbiasna
Vss
Figure 3.34. Buffered reference voltage.

M21
nresetCV
M11 M10 M3 M17 M19 M20
Vbiaspa Vbiaspa
M18
M12 M1 M2 Vbiaspb
Vbiaspb Vinplus Vinmin
C1
A
B Vref
M14 M8 M9
Vbiasnb Vbiasnb
C2
Vbiasnb
M15 M13 M4 M5 M6 M7 M22 M16

Vbiasna resetCV
Figure 3.35. Buffer op-amp with class AB output stage.
pensation capacitors, C1 and C2 equal 7 pF. The buffered reference voltage

has a stable transient response (2 to 1 ) with an overshoot of 2% (without
oscillation) and a 0.1 % settling time of 50 s for the maximum load capacitor
of 100 pF (worst case). This design is compared with conventional class A
buffers and it is proven to be five times more efficient in power consumption.
3.7 Main clock, clock generation circuits and LF clock

The clocks for the C-V converters, SC amplifier and modulator are derived
from the main oscillator. The main clock and the analog read-out electronics
are switched off during OFF-state (act = 0). So, only the 8 kHz LF clock is
running (very low power consumption). Timers are used to set the duration of
the ON and OFF state. Hence, we have a system with programmable duty cycle
operation and a very small OFF-state current consumption of approximately
0.5 A. The clock frequencies are also programmable to cope with technology
variations.
A. Main clock
The main clock of the GSIC is a 512 kHz square wave relaxation oscilla-
tor [Wak89]. The oscillator operates as follows (Fig. 3.36). If we assume
that the SR latch is in the state Vout1 = 0 and Vout2 = 1, the capacitor
C1 is shorted to analog ground and the current Icharge is loading the ca-
pacitor C2 . When the potential at node A becomes higher than Vturn , the
state of the SR latch is changed and the capacitor C1 is loaded. Hence,
we obtain a power supply independent oscillation frequency, which equals
Icharge /(2CVturn ).
Vdd
M1 M3 M5 M7 M9
Vbiaspa
M2 M4 M6 M8 M10
Vbiaspb
Icharge
M11 M16
B A
M12 M13 M14 M15

reset reset
C2
C1
Vturn Vturn
AGND AGND
+
+
-
Vdd
M17
Vbiaspa
M18
Vbiaspb
reset
Vturn
S
R
Vout2
Vout1
AGND
Figure 3.36. Square wave relaxation oscillator.
The capacitor C, the turning potential Vturn and the charge current Icharge
are chosen to reduce the clock jitter Tj, with respect to the modulator period
(Tmod 8s), so that
4 Tj, 1
< (3.40)
3 Tmod 1024
Hence, we attain enough accuracy for the modulator.

The clock jitter stems from the noise current in the oscillator capacitor and
the input noise voltages of the comparators. The noise current is integrated on
the capacitor C during one clock period, so its effect is negligible compared to
the broadband input noise of the comparator. Hence, we can approximate the
M4 M6 M7 M13 M23 M19

act
M5
M14 M18
Vout
M2 M3 M8 M9 M10 M11 M15 M17

Vin+ Vin-
M1 M21 M12 M22 M16 M20

Vbiasna nact nact
Figure 3.37. Fast continuous time comparator with low quiescent current.
clock jitter by [Abi83]

6 Vn
Tj, = Tosc (3.41)
2 V
where Tosc is the oscillator period ( 2s), is a constant, Vn is the equiv-
alent rms noise input voltage of the comparator and V is the triangle wave
peak-to-peak voltage. The clock jitter equals 2.9 ns in the implemented oscil-
lator, which satisfies equation 3.40.
The oscillator uses fast continuous time comparators with low quiescent
current (Fig. 3.37) [Bak98]. The first stage is a low gain, high bandwidth
preamplifier that drives a latch. The preamplifier uses small size transistors that
are biased in moderate inversion region. Hence, a high bandwidth is realized
with a minimum power consumption. The output mirrors of the preamplifier
are realized with a ratio M4 /M6 of 1/4 to further reduce the current of the input
stage. The latch performs the decision. Its transistors (M8 , M9 , M10 and M11 )
have equal size to reduce the comparator hysteresis. The latch outputs are used
to drive a self-biased differential amplifier. During the transition, the current
in the self-biased differential amplifier is boosted to perform a fast behavior.
After the decision is made, it falls back to a very low quiescent current. The
output of the self-biased differential amplifier drives a push-pull output driver.
The implemented comparator causes an oscillator propagation delay of 40 ns
for a current consumption of 2.5 A.
B. LF clock
The 8 kHz LF clock is used for the timing between subsequent samples. It
is the only part of the system, which operates continuously. The LF clock
is biased when the power supply VDD is put on [Kha03]. After the GSIC is
R2
M11
Vp
M6 M7
M10
Vdd
Vp
M1 M2 M3 M4 M5
Vp Vn
M8 M9
Vn Cstart
M12 M17
M13 M14 M15 M16

resetLF resetLF
C C
Vturn Vturn
AGND AGND
+
+
-
Vdd
M18
Vp
resetLF
Vturn
S
R
Vout2
Vout1
R1
AGND
Figure 3.38. LF oscillator with bias circuit.
programmed, resetLF becomes low and the LF clock and sample timer start
operating (operational flow, section 3.2). The LF clock is implemented as a
relaxation oscillator with a bias circuit (Fig. 3.38). The oscillator uses simple
two-stage open-loop comparators.
4. Configuration settings
The GSIC is equipped with many programming settings to offer an inter-
face to a broad range of capacitive sensor applications. The capacitive sensor
interface has two modes of operation. The first mode is for single sensor op-
eration with on chip reference capacitor, where the reference capacitor, Cref 2 ,
needs to be programmed to approximate the base capacitance of the sensor, C0 .
The other mode is for differential sensor operation, where the on chip refer-
ence capacitor, Cref 1 (or Cref 2 ), is programmed to compensate for the offset
between Cx and Cx . The amplification factor, ASC , of the SC amplifier and
the feedback capacitor, Cf , of the C-V converters need to be programmed for
optimal accuracy of the interface. The sample period is 6-bit programmable
between 8 ms and 512 ms and the ADC accuracy is selectable between 8, 9
or 10 bits for a conversion time of 2, 4 or 8 ms. Hence, the averaged power

consumption is strongly related to the accuracy and speed requirements of the
selected application. Both the LF clock and main oscillator are programmable
to cope with technology variations.
Table 3.1 shows the configuration addresses, their databits and a descrip-
tion of their functions. A databit with symbol X is not used and has to be
interpreted as a dont care bit. A databit with symbol T is reserved for test
purposes. The addresses 9 to 15 are unused. So, the modular architecture can
be expanded with extra configurable blocks, such as a temperature sensor or a
(bio)potential instrumentation amplifier. Hence, one can make a system that
can interface with several types of sensors in different time intervals. This will
result in applications such as a personal health assistant, which monitors the
blood pressure, body temperature and heart rate.
Table 3.1. Description of the configuration memory of the Generic Sensor Interface Chip.
D0 D1 D2 D3 D4 D5 D6 D7 D8 D9
a
0 X X X X X X T T M od1 M od0
1b X X X X X X LF3 LF2 LF1 LF0
2c X X X X X X CL3 CL2 CL1 CL0
3d X X S5 S4 S3 S2 S1 S0 AD1 AD0
4e X X Cr1 7 Cr1 6 Cr1 5 Cr1 4 Cr1 3 Cr1 2 Cr1 1 Cr1 0
5f X X Cr2 7 Cr2 6 Cr2 5 Cr2 4 Cr2 3 Cr2 2 Cr2 1 Cr2 0
6g X X X X X X Cf 3 Cf 2 Cf 1 Cf 0
7h X X X X ASC5 ASC4 ASC3 ASC2 ASC1 ASC0
8i X X ActCV ActSC ActV I Actmod T T T T
9 X X X X X X X X X X
a The M od1 and M od0 bits select the interface mode (single or differential).
b The LF3 . . . LF0 bits select the LF clock frequency.
c The CL . . . CL bits select the main oscillator frequency.
3 0
d The sample period is selected by S . . . S . The LF clock, sample timer and analog front-end are off, if
5 0
these bits are low (resetLF = 1, Act = 0, Reset = 1). AD1 and AD0 set the conversion resolution (8,
9 or 10 bit). The analog front-end is always off, if both bits are low (Act = 0, Reset = 1).
eC
r1 7 . . . Cr1 0 set the reference capacitor Cref 1 .
fC
r2 7 . . . Cr2 0 set the reference capacitor Cref 2 .
gC
f 3 . . . Cf 0 set the feedback capacitor Cf .
hA
SC5 . . . ASC0 set the SC amplification factor ASC .
i Act
CV , ActSC , ActV I , Actmod offer a power down option for the C-V converters, SC amplifier, VI
converter and modulator.
Furthermore, a computer algorithm is developed, which estimates the opti-

mal Cref , Cf and ASC for each application. This algorithm is expressed as an
optimization problem, which minimizes the total error. The objective function
considers the full-scale loss, leakage error, settling error, noise and ADC accu-
racy. The algorithm is presented in chapter 4. The obtained values for Cref , Cf
and ASC are the starting points of a calibration cycle. During this procedure,
the settings can be adjusted in order to cope with the technology variations of
the sensors and the interface chip. Furthermore, it is possible to change the
configuration settings after the system is operational. Hence, we can adapt the
system to changes in the environment (e.g. available supply energy) or drift
phenomena.
5. Noise
The mechanical noise of the sensor, the electrical noise of the interface and
the quantization noise of the ADC contribute to the total noise in the system.
The electrical noise of the SC interface is studied in this section. In 5.1, a
methodology is presented to estimate the noise in SC circuits in a simple and
accurate manner. Subsection 5.2 presents the noise calculations in the C-V
converter. The global noise performance of the SC interface is studied in 5.3.
5.1 Bennet model

The noise of the SC interface (C-V converters and SC amplifier) is calcu-
lated with the Bennet model [Ben48]. This model states that the noise can be
described as an infinite sum of discrete sinusoidal components. These com-
ponents have a different frequency, an equal amplitude and a random phase,
which is uniformly distributed in the interval [, ]. The sum of all the Ben-
net components equals the total noise power, which is given by the product of
the Power Spectral Density (PSD) and the bandwidth.
With the Bennet model, we can calculate the noise as follows:
Each noise source is presented by an equivalent sinusoidal source with am-

n , random phase and pulsation . The corresponding phasor
plitude u
n exp(j).
equals u
Thereafter, we compute the output phasor, Uout , at time t2 (i.e. the end of
the signal phase).
The power spectrum of the noise at t2 , Sout,t2 (), is given by:

1
|Uout |2 d
Sout,t2 ()d = d Sun (f ) (3.42)
2 2n
u
where Sun is the power spectrum of the noise source. With Sout,t2 as a
function of the frequency, equation 3.42 can be written as:

1
Sout,t2 (f ) = 2Sun u2 d (3.43)
2 out,g
where uout,g is the normalized output amplitude at t2
|Uout |
uout,g = (3.44)
u
n
The output signal is sampled by the following SC stage. This process can
be modelled by an ideal Sample and Hold operation [Fis82]. Hence, the
output spectrum after sample and hold, Sout,h (f ), is given by:

d j2f 1 2 f
Sout,h (f ) = Sout,t2 exp sinc (3.45)
fs f2 fs
d
where the discrete power spectrum Sout,t2 is given by:

d j2f
Sout,t2 exp = fs2 Sout,t2 (2f + 2kfs ) (3.46)
fs
k=
This equation expresses the noise aliasing caused by the sampling operation.
The noise has a finite bandwidth, BWnoise . If we take this in consideration,
we can approximate the infinite sum by

N
Sout,t2 (2f + 2kfs ) Sout,t2 (2f + 2kfs )
k= k=N

BWnoise
N = 10round (3.47)
fs
The RMS-value of the sampled noise at the output, noise , is then given by:

fs /2 fs /2
noise = Sout,h (f )df = 2 Sout,h (f )df (3.48)
fs /2 0
5.2 Noise calculations

The noise in the C-V converters and the SC amplifier contains contributions
of the switches and the OTAs (thermal and 1/f noise). All these components
C2 F2
sw6
F1 sw3 F1
F1 C3 sw4
Vref
sw1 - F2
C1 Una
sw5 F1 sw2 Cls
+ Cl
F2 Cp1 Cp2
Figure 3.39. C-V converter with switches, sw1 to sw6 , and OTA noise source, Suna .
are uncorrelated. So, the total noise can be found as a squared superposition of
their RMS values. The effects of the individual noise sources are estimated with
the Bennet model. Hereafter, only the noise calculations of the C-V converter
are presented. Similar computations have been made for the SC amplifier.
The noise in the C-V converter is caused by the input referred noise, Una , of
the OTA and the thermal noise of the switches (Fig. 3.39). The input referred
noise spectral density of the OTA, Suna , equals:
8kT KF
Suna (f ) = Nex + (3.49)
3gm f
where Nex is the noise excess factor [San94] and KF is the 1/f noise factor,
which depends on the technology, the type and the size of the transistors. The
switches can be replaced by their ON-resistance, Ron , in series with a white
noise source
Sunsw (f ) = 4kT Ron (3.50)
A. Noise of the OTA

During the sampling phase 1 , the noise of the OTA, Una , is sampled on
capacitor C3 (Fig. 3.40). At the end of the sampling phase, t1 , this results in a
C3
-
Vs Una
+ Cl
Figure 3.40. OTA noise during the sampling phase, 1 .

sampled voltage, Vs , given by:

1
1 1
Vs = exp(jt1 )Una j 1
1+ j 1
(3.51)
2GBW1 + A 2GBW1 + A
where A is the DC gain of the OTA and GBW1 is the gain bandwidth of the
OTA during the sampling phase.
Cl
GBW1 = GBW (3.52)
Cl + C3
The sampled and the instantaneous noise are subtracted during the signal
phase 2 (Fig. 3.41). Hence, the equivalent noise source at t2 equals
Veq = Una exp(jt2 ) Vs (3.53)
So, the phasor V equals
C2 Veq
V = (3.54)
j 1
(C1 + C2 + Cp2 ) 2GBW 2
+ A + C2
where GBW2 is the gain bandwidth of the OTA during the signal phase.
Cl
GBW2 = GBW C2 (C1 +Cp2 )
(3.55)
Cl + Cls + C2 +C1 +Cp2
The phasor of the output voltage, Uout , is then given by:

C1 + Cp2
Uout = V V (3.56)
C2
The obtained Uout is used to calculate uout,g (3.44). Thereafter, the output
rms noise, noise , is computed according to the procedure from subsection 5.1
(3.43-3.48).
C2
V Veq
-
C1 Vs Una
Cl Cls Uout
Cp2 +
Figure 3.41. OTA noise during the signal phase, 2 .

The noise at the output contains contributions of the thermal and 1/f noise
components of the OTA. The thermal component causes a noise voltage, noise ,
which
is independent of the transconductance gm . The bandwidth of the C-V
converter is proportional to the transconductance, gm . So, the noise aliasing
is also proportional to gm . On the other hand, the OTA noise input spectrum
decreases with gm (3.49). Hence, the noise voltage is not effected by gm .
decreases with a higher load capacitor, Cl .

is proportional to Cp2 /C1 .
The 1/f noise is adequately suppressed by the CDS operation.
B. Noise of switch 1
At the end of 2 , the thermal noise of switch 1 creates an indirect and a direct
noise component at the output. The indirect component stems from the sampled
noise (at the end of 1 ) on the series parallel combination of C1 , Cp2 , C2 and
C3 . During the signal phase, these noise charges are transferred to the feedback
capacitor Cf . The direct component is caused by the thermal noise during 2 .
This component is filtered by the low pass operation of the OTA, whereas the
indirect component is only limited by the 1/(2RON C) cut-off behavior. Since
the latter frequency is much higher, we can neglect the direct component.
The sampled (indirect) noise component is evaluated in Fig. 3.42. After
solving this network, we obtain the output phasor (at t2 ):
Q1 () Qp () Q2 () Q3 ()
Uout = + (3.57)
C2 C3
The found Uout is used to calculate the rms noise value, noise . This noise value
is independent of RON .

is inverse proportional to C1 (with a constant Cp2 /C1 ).
decreases with increasing Cp2 /C1 .
Q2 RON
4kTRON RON Q1 Q3
RON
Qp RON
Figure 3.42. Equivalent scheme for the noise of switch 1, during 1 .

C. Noise of switch 2
The noise of switch 2 is also estimated by its indirect component. The solution
of the network (Fig. 3.43) gives the following output phasor:
Q1 () + Qp () + Q2 () Q3 ()
Uout = + (3.58)
C2 C3
The rms noise value

increases with Cp2 /C1 .
D. Noise of switch 3
The noise of switch 3 is estimated by its indirect component. The solution of
the network (Fig. 3.44) gives the following output phasor:
Q1 () + Qp () Q2 () Q3 ()
Uout = (3.59)
C2 C3
The rms noise value

decreases with increasing Cp2 /C1 .
Q2 RON
RON Q1 Q3
RON
Qp RON
4kTRON

Q2 RON 4kTRON
RON Q1 Q3
RON
Qp RON
E. Noise of switch 4

The noise contribution of switch 4 can be approximated by the sampled kT /C3
noise on C3 .
F. Noise of switches 5 and 6

The noise of switches 5 and 6 has only a direct component, so it is negligible.
5.3 Effective number of bits

The noise of the sensor system depends on the application (characteristic sen-
sor curve, parasitic capacitors, etc.) and the configuration settings of the front-
end circuit. A measure for the accuracy of a sensor system has to consider both
the sensor and the interface imperfections. To characterize the noise behavior
of a generic system, the quantity effective number of bits is defined as follows:
Assume we have a sensor with characteristic S(x), where x is a physical
quantity (acceleration, pressure, humidity, . . . ), which varies between xmin and
xmax . If one applies a sinusoidal signal between xmin and xmax at the input
of the sensor, the interface will produce a fundamental harmonic Vout,f und,rms
and a disturbing component Vout,noise+dist,rms caused by noise and distortion.
The effective number of bits equals:

Vout,f und,rms
Ebits = log2 0.2925 (3.60)
Vout,noise+dist,rms
The noise contributions of the buffered reference voltage Vref , the op-amps
and the switches in the C-V converters and the SC amplifier were evaluated.
The sampled noise, bsample , presented at the input of the ADC during one
SC interface clock period (0.125 ms) is shown in Fig. 3.45. This figure gives
the sampled noise as a function of and C0 , assuming Cp1 = Cp2 = 10pF,
Gp = 0.01S and a single linear sensor characteristic:

x 0.5(xmax + xmin )
Cx (x) = C0 1 + (3.61)
xmax xmin
So, only the noise contributions of the SC interface are taken into account.
Figure 3.45. Sampled noise as a function of and C0 .
During one measurement period, several SC interface cycles are performed.

This means that the effective number of bits, bmeas , during one measurement
equals
bmeas = bsample + 0.5log2 (NSC ) (3.62)
where NSC is the number of SC interface cycles during one measurement.
Furthermore, the accuracy limitations by the modulator need to be considered.
These quantization errors depend on the measurement time. With a conversion
time between 256 and 1024 modulator periods (= 16 - 64 SC interface periods),
the simulated front-end accuracy is at least 8 bits for each sensor in the specified
range.
6. Experimental results
The GSIC is designed in a 0.5m CMOS technology (AMIS) for a supply
voltage between 2.7 and 3.3 V. The chip measures 3.2 mm by 2.9 mm (including
the IO ring) (Fig. 3.46).
Special care has been taken to reduce the influence of the digital and higher
frequency electronics on the sensitive analog read-out. For this purpose, the C-
V converters are placed on the opposite side of the ADC, the main oscillator and
the clock generation circuits. Furthermore, five different supplies with separate
IO pads are used: a supply for the external circuits in the IO ring (V SSE,
V DDE), a supply for the internal low noise transistors in the IO ring (V SSI,
V DDI), a supply for the digital core cells (V SSCO, V DDCO), a supply for
the guard rings (V SSS, V DDS) and an analog dual supply (AV SS, AGN D,
Figure 3.46. Die photograph of the Generic Sensor Interface Chip (size = 3.2 mm x 2.9 mm,
including IO ring).
AV DD). The guard rings are placed in close proximity of the digital circuits
and higher speed analog circuits. Hence, they provide a low impedance return
path for the injected current pulses [Ing97]. An n-well on the ground potential
AGN D shields the analog high ohmic poly resistors and capacitors from the
substrate.
The system has a total current consumption of approximately 40A dur-
ing ON-state. Table 3.2 shows the current consumption of the analog blocks.
The averaged power consumption can be tailored towards the speed and ac-
curacy requirements of the application. Hence, the averaged current con-
sumption is smaller than 20A for sensor applications with small bandwidth
(<50 Hz) and medium resolution requirements (Fig. 3.47). As an exam-
ple, we consider the cases of a pressure sensor and an inclinometer
application.
Table 3.2. Current consumption of the analog building blocks.
Analog building blocks Current (A)

Bandgap reference, bias system and buffered reference voltage Vref 13.5
Main clock and clock generation circuits 8
C-V converters 2 x 3.5
SC amplifier 4
VI converter 1.8
Modulator 4.1
Total ON-state current 38.4
LF clock (runs continuously) 0.5
8bit mode
9bit mode
Current consumption (A)
10bit mode
1
10
0
10 1 2
10 10
Sample frequency (Hz)
Figure 3.47. Averaged current consumption of the GSIC as a function of the sample frequency
and the resolution mode.
6.1 Pressure monitoring system

In the diagnosis of urinary incontinence, one wants to observe the contraction
of the bladder muscle as the bladder fills and empties. For this purpose, we
use a single capacitive absolute pressure sensor (B012FB, VTI Technologies)
to monitor the bladder pressure in the range of 100-130 kPa [Coo05]. This
application needs a sampling frequency of 10 Hz and 8-bit accuracy. A duty
cycle of 2 % can be used, which results in a measured power consumption
of approximately 7.3W (3 V supply). The test PCB contains the GSIC, the
pressure sensor and a connector to a pump (Fig. 3.48). The sensor is placed in
the cavity of the connector. In this way, we can apply a reliable pressure on the
sensor membrane. Fig. 3.49 shows the digital output code as a function of the
Figure 3.48. Test PCB with pressure sensor, connector and GSIC (covered with globtop coat-
ing).
100
Digital output code
50
50
100
100 105 110 115 120 125 130

Pressure (kPa)
Figure 3.49. Measured digital output code as a function of the applied pressure.
applied pressure. The behavior is slightly non-linear over pressure, inherent to

the performance of the sensor. The linearization is accomplished by numerical
correction using a 3 points calibration on the physical sensor model (provided
by the manufacturer). The remaining pressure error after this linearization is
smaller than 0.1 kPa (Fig. 3.50). The repeatability of the output is, for each
applied pressure, within 1 count. Hence, the dominant noise source is the
quantization.
Furthermore the system is tested (with different settings) in the pressure range
from 60 kPa to 120 kPa. For this test, we have selected a sample frequency of
with linearization
0.2 without linearization
Pressure error (kPa) 0
0.2
0.4
0.6
0.8
1.2
1.4
100 105 110 115 120 125 130
Pressure (kPa)
Figure 3.50. Pressure error after 3 points calibration with the sensor model.
10 Hz and 10-bit resolution. The measured power consumption equals 16.4W

(3 V supply). Fig. 3.51 shows the digital output code as a function of the applied
pressure. The remaining pressure error after this linearization is smaller than
0.1 kPa (Fig. 3.52). The repeatability of the output is, for each applied pressure,
within 1 count. Hence, the dominant noise source is the quantization.
500
400
300
Digital output code
200
100
0
100
200
300
400
500
60 70 80 90 100 110 120
Pressure (kPa)
Figure 3.51. Measured digital output code as a function of the applied pressure.
1
with linearization
without linearization
0
Pressure error (kPa)
1
5
60 70 80 90 100 110 120
Pressure (kPa)
Figure 3.52. Pressure error after 3 points calibration with the sensor model.
6.2 Inclination monitoring system

The SCG10Z-G001CC low g accelerometer (VTI Technologies) is used in an
inclination monitoring system. This sensor is tested with several configuration
settings of the GSIC. In the first test, the functionality of the sensor system is
evaluated (9-bit mode). In the second test, we have verified that the operation
of the GSIC is immune to power supply variations from 2.7 V to 3.3 V. The
last test shows us that the transition from 8-bit to 10-bit mode has a negligible
effect on the acceleration measurement characteristic.
A. Functionality test
In this test set-up, we select a 10 Hz sample frequency and 9-bit resolution to
measure the inclination from -90 to 90 degrees (equivalent with -1 g to 1 g ac-
celeration). With these settings, the GSIC consumes 10.3W (3 V supply). The
test PCB contains the accelerometer and the GSIC. It is connected to an incli-
nometer calibration fixture (Model 5560, Robert A. Denton) (Fig. 3.53). In this
static set-up, we can vary the orientation of the test accelerometer (gravitational
field) in steps of 10 degrees.
Fig. 3.54 presents the measured digital output code as a function of the in-
clination. Fig. 3.55 shows the equivalent output characteristic as a function
of the acceleration. In Fig. 3.56, we see that the non-linearity acceleration
error is smaller than 0.03 g. The repeatability of the output is, for a con-
stant acceleration, within 1 counts. Hence, the dominant noise source is the
quantization.
Figure 3.53. Inclination measurement set-up with test PCB (GSIC and accelerometer) con-
nected to the calibration fixture.
250
200
150
Digital output code
100
50
0
50
100
150
200
250
80 60 40 20 0 20 40 60 80
Inclination (degrees)
Figure 3.54. The measured digital output code as a function of the inclination.
250
200
Digital output code 150
100
50
0
50
100
150
200
250
1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1
Acceleration (g)
Figure 3.55. The measured digital output code as a function of the acceleration.
0.05
0.04
Acceleration error (g)
0.03
0.02
0.01
0
0.01
0.02
0.03
0.04
0.05
1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1
Acceleration (g)
Figure 3.56. The non-linearity acceleration error as a function of the acceleration.
B. Test with different power supplies

The battery supply voltage of autonomous sensors decreases during the lifetime
of the system. Furthermore, systems powered by energy scavengers (which
take their energy from the environment, e.g. body warmth or vibrations) deal
with a varying power source. Hence, it is important that the operation of
250
2.7 V
200 3.3 V
100
50
0
50
100
150
200
250
80 60 40 20 0 20 40 60 80
Figure 3.57. The measured digital output code as a function of the inclination for supply
voltages of 2.7 and 3.3 V.
the sensor system is immune to a broad range of supply voltages. For this
purpose the GSIC is tested in 9-bit mode with a 10 Hz sample frequency
for a supply voltage between 2.7 V and 3.3 V. The GSIC consumed 8.2W
in the 2.7 V test and 12.4W in the 3.3 V test. In Fig. 3.57, we see that
the measured acceleration characteristic is nearly independent from the power
supply.
C. Influence of the resolution mode

The configuration settings of the GSIC can be changed, while the system is op-
erational. This allows to adapt the system to environmental changes. So, we can
compensate for drift phenomena and we can decrease the average power con-
sumption (sample frequency and resolution mode) when the available energy
is getting low. A lower resolution mode corresponds with a shorter conver-
sion time (10-bit => 8 ms, 8-bit => 2 ms). As a consequence, a change
from 10-bit to 8-bit mode also creates a different electrostatic force on the
mechanical sensor. In order to evaluate this phenomenon, we performed tests
with the G001CC accelerometer in 10-bit and 8-bit mode for a sample fre-
quency of 10 Hz. The GSIC consumed 16.8W in 10-bit mode and 6.9W
in 8-bit mode with a power supply of 3 V. In Fig. 3.58, we see that the tran-
sition from 8-bit to 10-bit mode has a negligible effect on the accelerometer
characteristic.
500
10bit
400 8bit
200
100
0
100
200
300
400
500
80 60 40 20 0 20 40 60 80
Figure 3.58. The measured digital output code as a function of the inclination for 10-bit and
8-bit resolution mode.
7. Performance comparison
The energy per accuracy level, En, gives a metric for the efficiency of the
sensor interface. It is defined as follows:
Pav
En = (3.63)
fsample 2b
Where Pav , fsample and b are the averaged power consumption, the sample
frequency and the accuracy (in a bit number). The type of output also effects
total efficiency. Sensor interfaces with digital outputs only need a serial or
parallel interface to transfer the data to the microcontroller. This transfer can
be extremely fast and power efficient, which allows the microcontroller to stay
for a long period in sleep mode. On the other hand, sensor interfaces with
analog output still need an external ADC, which consumes additional power.
While sensor interfaces with quasi-digital output (frequency, period or duty
cycle output) require a counting process, performed by the microcontroller, to
yield the digital code. Hence, the microcontroller cannot enter the sleep mode
during the conversion process. Tables 3.3 and 3.4 compare the energy per
accuracy level of the GSIC with other generic interfaces and dedicated (U)LP
capacitive sensor interfaces. We see that the GSIC outperforms the existing
generic sensor interfaces. It even achieves an efficiency, which is better than
most of the dedicated state-of-the-art ULP interfaces.
Table 3.3. Performance comparison between the GSIC, other generic sensor interfaces (C0 =
10.3pF, C = 1 pF) and dedicated (U)LP pressure sensor systems.
ref output range accuracy fsample Pav En

[Coo05] quasidigital 100-130kPa 0.04kPa 10Hz 28W 3.7nJ
[DG96] digital - 11 bit 2Hz 12W 3nJ
[VTI] digital 30-120kPa 18Pa 6.5Hz 84W 2.6nJ
[Kol04] digital 50-650kPa 1kPa 2Hz 1.5W 1.3nJ
[VDG96] quasidigital 100-130kPa 300aF 10Hz 5mW 150nJ
[Yaz00] analog 100-130kPa 1fF 20Hz =2xBW 2.2mW 110nJ
this work digital 100-130kPa 120Pa 10Hz 7W 2.7nJ
Table 3.4. Performance comparison between the GSIC, other generic sensor interfaces (C0 =
2.5pF, C = 0.4pF) and dedicated (U)LP accelerometer systems.
ref output range accuracy fsample Pav En

[Ana] analog +/-2g 185mg (=6.6 x rms) 40Hz 32W 37nJ
[ST] digital +/-2g 1mg 80Hz =2xBW 1.9mW 6nJ
[Col] analog +/-2g 1.4mg (=6.6 x rms) 42Hz =2xBW 600W 5nJ
[Yaz00] analog +/-1g 1fF 20Hz =2xBW 2.2mW 275nJ
this work digital +/-1g 4mg 10Hz 10W 2nJ
8. Conclusion
In this chapter, the specifications, design and results of the Generic Sen-
sor Interface Chip for capacitive sensors have been presented. The GSIC is
designed as a complete system, which is optimized for Ultra Low Power con-
sumption. It contains the following blocks: Capacitance-to-Voltage converters,
a SC amplifier, a modulator, a bandgap reference and bias circuits, a main
oscillator and clock generation circuits, an LF clock, a configuration SRAM
and an interface to the microcontroller.
The SC interface (two C-V converters and a SC amplifier) converts a capac-

itance variation, C, in a proportional voltage. It operates on a lower clock
frequency, 8 kHz, than the modulator to reduce the power consumption.
Reducing the clock frequency of the SC interface increases the influence of
the parasitic shunt conductance. A C-V converter, which uses class AB tech-
niques and Correlated Double Sampling operation is developed. This C-V
converter is approximately three times more efficient in reducing the effects
of the shunt conductance than conventional C-V converters (with the same
current consumption). The outputs of both C-V converters are chopped and
filtered by the modulator to eliminate the effects of mismatches and
to perform an adequate reduction of the common mode interference. The
fully differential SC amplifier amplifies the difference between the outputs
of both C-V converters and provides a quasi continuous input voltage for
the modulator.
The implemented CT modulator consists of a VI converter, a modulator
and a decimation counter. The CT modulator has the advantage that it
does not need a buffered reference voltage to attain the oversampling speed
(128 kHz). This results in a significant decrease of the power consumption.
The error sources in ultra low power CT modulators were examined.
The dominant error source is caused by charge injections. This effect is
eliminated by implementing a RZ switching scheme.
The bandgap reference and bias circuits provide the bias currents to the sen-
sor interface and the main oscillator. The system is immune to temperature
variations in the range from 40 C to 85 C. Furthermore, its operation
is quasi independent of the power supply in a range from 2.7 to 3.3 V. A
switch in the start-up circuit of the bandgap reference offers a power down
option for the total analog part of the system.
The main oscillator and clock generation circuits provide the clock signals
to the SC interface and the modulator. The main clock and the analog
read-out electronics are switched off during OFF-state. A conversion timer
is used to set the duration of the ON-state.
The 8 kHz LF clock is used for the timing during low power standby oper-
ation. It is the only part of the system, which operates continuously. It has
a very small current consumption of approximately 0.5A.
The GSIC has many configuration settings to provide an interface to a broad
range of capacitive sensor applications. The capacitive sensor interface has
two modes of operation. The first mode is for single sensor operation with
on chip reference capacitor, where the reference capacitor, Cref 2 , needs
to be programmed to approximate C0 . The other mode is for differential
sensor operation, where the on chip reference capacitor, Cref 1 (or Cref 2 ),
is programmed to compensate for the offset between Cx and Cx . In both
modes the amplification factor, ASC , of the SC amplifier and the feedback
capacitor, Cf , of the C-V converters need to be programmed for optimal
accuracy of the interface (chapter 4).
The sample frequency is 6-bit programmable between 2 Hz and 125 Hz and
the ADC accuracy is selectable between 8, 9 or 10 bits for a conversion time
of 2, 4 or 8 ms. Hence, the averaged power consumption can be adapted to

the accuracy and speed requirements of the selected application.
Both the LF clock and main oscillator are programmable to cope with tech-
nology variations.
The GSIC is designed in a 0.5m CMOS technology and measures 3.2 mm

by 2.9 mm (including the IO ring). The total system consumes merely 40A
in operational mode with a 3 V supply. The duty cycle management adapts
the energy consumption according to the accuracy and speed requirements of
the application. This results in a measured power consumption of 7.3W in a
pressure monitoring system (10 Hz sample frequency and 8-bit accuracy in the
100 to 130 kPa range). For the acceleration measurement system, it achieves
a power consumption of 10.3W for a sample frequency of 10 Hz and 9-bit
accuracy in the 1 g range.
Chapter 4
ALGORITHM FOR OPTIMAL CONFIGURATION

SETTINGS
1. Introduction
Generic sensor interface design reduces the costs and offers a handy solution
for multisensor applications. However, previous generic readout circuits often
experienced a loss of performance and an increased power consumption. The
ability to program the front-end would have an important impact on the accuracy
of the sensor readout. The new design methods reported in this chapter allow
us to create a generic sensor interface with a minimum loss. Furthermore, an
algorithm is provided, which calculates the optimal interface settings for each
application. These settings enable a state-of-the-art performance from our ULP
generic sensor interface.
Section 2 studies the non-ideal effects (full-scale loss, leakage error, settling
error and noise) of configurable capacitive sensor interfaces. General design
methods are derived, which are illustrated on the generic interface architecture.
Section 3 presents an algorithm that estimates the optimal configuration of the
interface for each application. This algorithm is expressed as an optimization
problem, which minimizes the total error. The minimized objective function
considers the full-scale loss, leakage error, settling error, noise and ADC accu-
racy. In section 4, we apply the algorithm in a prototype biomedical pressure
monitoring system.
2. Programmability
The front-end architecture consists of a Switched Capacitor (SC) interface
followed by a modulator. The SC interface works on a lower clock frequency,
8 kHz, than the modulator, 128 kHz, to reduce power consumption. In the
Capacitance-to-Voltage (C-V) converter, the sense capacitance, Cx , is converted
to a proportional voltage. The SC amplifier amplifies the difference between the
outputs of both C-V converters and produces a quasi continuous input voltage for
the modulator. The capacitive sensor interface has two modes of operation.
The first mode is for single sensor operation, where the reference capacitor Cref
needs to be programmed to approximate C0 . The other mode is for differential
sensor operation, where the on chip reference capacitor is used to compensate
for the offset between Cx and Cx . In both modes, the amplification factor ASC
of the SC amplifier and the feedback capacitor Cf of the C-V converters need
to be programmed for optimal accuracy of the interface.
The programmability (step size and range) of Cref , Cf and ASC strongly
influences the use of the dynamics and the settling performance of the interface.
This section introduces methods to estimate these influences mathematically.
Furthermore, the appropriate step sizes and ranges of Cref , Cf and ASC are
calculated for our generic architecture.
2.1 Full-scale loss

We consider a general sensor characteristic C(x), where x varies between
xmin and xmax (Fig. 4.1). If Cref , Cf and ASC have unlimited programmabil-
ity, the input of the modulator sees a voltage characteristic between 0.5V f spp
and 0.5V f spp . Due to the limitation of the programmability, one cannot use
the full input range of the ADC (Fig. 4.2). The full-scale loss V f srel is given
by:
Vmax + Vmin
V f srel = (4.1)
V f spp
where Vmax and Vmin are the deviations between the ideal (unlimited pro-
grammable interface) and realistic (limited programmable interface) ADC input
characteristics at xmax and xmin .
Cxmax
C0
Cxmin
Xmin Xmax
Figure 4.1. General sensor characteristic C(x).

Algorithm for Optimal Configuration Settings 75
0.5 Vfspp
Ideal curve DVmax
Realistic curve
DVmin
-0.5 Vfspp Xmin Xmax
Figure 4.2. The input voltage of the modulator for an unlimited (ideal curve) and limited
(realistic curve) programmable SC interface.
The full-scale loss has two causes: firstly, the components are programmable
between certain boundaries, Pmin and Pmax ; secondly, they are programmable
in finite steps, Pstep .
The optimal settings are calculated with an algorithm, which maximizes the
accuracy (see section 3). This objective function considers not only the full-
scale loss, but also the settling error, leakage error, noise and ADC accuracy.
Due to the boundaries of the programming components, the optimal settings
may result in a full-scale loss that is significant. For our capacitive sensor
interface, this deterministic full-scale loss is given by:
dVref C0 ASC
V f srel = 1 (4.2)
Cf V f spp
where d = 1 for single sensors and d = 2 for differential sensors.

The finite step size makes the programmed value P differ from the ideal
value Pi by an amount P = P Pi . This deviation P can be considered as
a stochastic variable with a uniform probability distribution between 0.5Pstep
and 0.5Pstep . This results in a stochastic full-scale loss.
2.2 Programmability of Cref

The deviation, Cref , of Cref to C0 (d = 1) or Cref to Cof f set (d = 2)
results in an offset voltage, Vof f set , towards the ideal ADC input characteristic
(Fig. 4.3):
ASC Cref Vref
Vof f set = (4.3)
Cf
0.5 Vfspp
Ideal curve
DVfs
Realistic curve
0
DVoffset
-0.5 Vfspp
Xmin Xmax
Figure 4.3. The full-scale loss V f s for an offset deviation Vof f set .
If we assume that Cf and ASC have unlimited programmability, we find:

2Cref
V f srel = (4.4)
dC
We consider Cref as a stochastic variable with a uniform probability func-
tion between 0.5Cref,step and 0.5Cref,step . Hence, the full-scale loss is also
uniformly distributed with mean value zero and standard deviation :
2Cref,step
(V f srel ) = (4.5)
dC 12
This standard deviation gives a measure for the loss in dynamics, which is
caused by the finite step size, Cref,step , of Cref . In order to allow a good
exploitation of the system dynamics, we postulate that (V f srel ) needs to be
smaller than 25% for our interface (200fF< C <10pF and 1pF< C0 <15pF).
In Fig. 4.4 we see that this condition is met for a Cref,step value of 75 fF.
Hence, we obtain a Cref capacitance, which is 8-bit programmable between
0 and 19.125 pF (= Cref,max > C0,max ) in steps of 75 fF. The matching
properties of a standard analog CMOS technology do not pose any problem for
the development of such an 8-bit capacitor array (INL<< 0.5LSB).
2.3 Programmability of Cf
Fig. 4.5 shows the ULP C-V converter (section 3.2). During the sampling
phase 1 , the sense capacitor, Cx , is charged. During the signal phase 2 , Vp
becomes a virtual ground and the charge is transferred to the feedback capacitor,
Cf . At the end of the signal phase, assuming an ideal charge transfer, the voltage
at the output of the C-V converter equals Vref (Cx /Cf ). In reality, the charge
1
(Vfs ) (%) 10
rel
0
10
1 2 3 4 5 6 7 8 9 10
C (pF)
Figure 4.4. Standard deviation of the full-scale loss as a function of C with Cref,step = 75
fF (d = 1).
transfer will be imperfect due to the finite transient response of the OTA. This
transfer error has two contributions: the leakage error and the settling error.
The potential Vp will settle in a certain time to the virtual ground, during this
time some charge will leak away through the parasitic shunt conductance, Gp ,
resulting in a leakage error. At the end of 2 , the potential Vp will be slightly
different from the ground level, so a small charge remains on Cp2 +Cx , resulting
in the settling error.
The C-V converter uses a class AB OTA with cascode output stage to reduce
the leakage and settling errors (Figs. 4.6 and 4.7). After the transition from
phase 1 to phase 2 , the tail current of the OTA is boosted, which speeds up
the charge transfer from the sense capacitor to the feedback capacitor. During
Cf F2
Gp F1d F1d
F1d Cs
Vref Vp
-
Cx
F2d F1 + Cl
Cp1 Cp2
Figure 4.5. Capacitance-to-Voltage converter with correlated double sampling and class AB
operation.
M9 M10
- M3 M4 -
Vb Va
Op1 Ibias Op2
+ +
M12
Vs
Vbiaspb
M1 Ma Mb M2
Va Vb
Vout
Vin- Vin+
M11
Vbiasnb
Cl
M7 M5 M6 M8
B : 1 1 : B
Figure 4.6. Class AB OTA.
settling, the voltage Vp approaches the virtual ground and the tail current falls
back to a low quiescent level.
During 2 , the charge transfer passes the following three phases:
During the first phase, the internal feedback op-amps op set the potential Vs
to follow the larger of the two voltages Va and Vb .
During the second phase, the current is strongly boosted and the output
current of the OTA, Iout , becomes:

Vin+ Vin
Iout = BIbias exp 1
nUT

Vin+ Vin
BIbias exp (4.6)
nUT
M3a M4a M3
Voutop
Cc
M1a M2a
Vs
Vina+ Vina-
Ibiasop
Figure 4.7. Internal feedback op-amp.

During the third phase, the current boosting is nearly exhausted and the OTA
output current shows a small signal settling behavior

Vin+ Vin
Iout = BIbias exp 1
nUT

Vin+ Vin
BIbias (4.7)
nUT
The total leakage error leakerr has contributions of the three phases:
t1 t3
Gp BIbias f t
leakerr = Vin (0)dt + nUT exp dt
Cx Vref 0 0 nUT Cout
t2
Vin (0) BIbias f t
nUT ln exp + dt (4.8)
0 nUT nUT Cout
with
Vref Cx
Vin (0) =
Cx + Cp2 + Cf Cl /(Cf + Cl )
Cf (Cx + Cp2 )
Cout = Cl +
Cf + Cx + Cp2
Cf
f = (4.9)
Cf + Cx + Cp2
The time intervals t1 , t2 and t3 indicate the duration of each phase and
are given by the following formulas:
Vgs3 Cc
t1
Ibiasop

Vin (0) Cout nUT
t2 exp(1) exp
nUT f BIbias
t3 T2 (4.10)
where T2 is the duration of the signal phase 2 .

The contributions of the first two phases are negligible in the settling error,
settlingerr , then we obtain:

(Cp2 + Cx )nUT exp BI bias f t3
Cout nUT
settlingerr (4.11)
Cx Vref
The leakage and settling errors decrease with increasing Cf . Simulations show
that our interface has a good settling behavior if the feedback factor f is greater
than 0.25. This implies that Cf,max needs to be higher than 1/3(Cx,max +
Cp2,max ) to provide a sufficient settling for the whole application range. On
the other hand, a larger Cf needs a larger chip area and a larger amplification
factor, ASC , for the SC amplifier to use the full input range of the ADC. This
results in a higher GBW specification and hence a higher power consumption
for the SC amplifier. If we make Cf 4-bit programmable between 0 and 27 pF
(= Cf,max ) in steps of 1.8 pF, we can always find a suitable Cf for the specified
application range.
In order to evaluate the validity of the leakage charge model, we compare
the mathematical model (4.8-4.10) with Cadence simulations. Fig. 4.8 shows
the leakage charge as a function of Cx and Cp2 with Gp = 0.01S and optimal
configuration settings (algorithm section 3, with d = 1, = 0.5 and b =
10bits). In Fig. 4.9, we see that the model has an accuracy of 5% for most of
the capacitor values. Only for small values of Cx (< 3pF) the error rises to
approximately 20%.
2.4 Programmability of ASC

The limited programmability of ASC results in a full-scale loss (Fig. 4.10):
ASC
V f srel = (4.12)
ASC
The full-scale loss contains contributions of the deterministic full-scale loss

(limited range ASC,max ) and the stochastic full-scale loss (limited step size
ASC,step ).
8
7
leakage charge (fC)
6
5
4
3
2
1
20
15 9 10
7 8
10 6
4 5
5 3
1 2
Cp2 (pF) Cx (pF)
Figure 4.8. The leakage charge as a function of Cx and Cp2 (mathematical model=meshed
curve, Cadence simulation=marker points.
25
(leakmodleaksim)/leaksim (%)
20
15
10
5
0
5
10
20
15 9 10
7 8
10 6
4 5
5 3
1 2
C (pF) Cx (pF)
p2
Figure 4.9. The relative deviation between the mathematical leakage charge model and the
Cadence simulation points as a function of Cx and Cp2 .
We assume that the full output range of the C-V converters is used for small
. Hence, we find that the interface can always provide an optimal usage of the
ADC input range (deterministic full-scale loss V f s(det) = 0) if
V f spp
ASC,max > (4.13)
dmin V f sCV
where V f sCV is the maximum allowable output voltage of the C-V converters.
Consequently, the output of the C-V converter will not saturate.
In our interface ( min = 0.05, V f sCV = 1V and V f spp = 0.5V for
single sensors d = 1) we choose ASC,max = 15.75 accoring to condition
(4.13).
0.5 Vfspp
Ideal curve 0.5 DVfs
Realistic curve
0.5 DVfs
-0.5 Vfspp Xmin Xmax
Figure 4.10. The full-scale loss V f s for a slope deviation.

For the stochastic full-scale loss, we consider ASC as a stochastic variable

with uniform probability function between 0.5ASC,step and 0.5ASC,step . We
obtain:
ASC,step
(V f srel ) = (4.14)
12ASC
If we assume that Cf,op is programmed for optimal settling and the deterministic
full-scale loss is zero, we find:
dC0 Vref ASC,step
(V f srel ) = (4.15)
Cf,op V f spp 12
where
C
if V f spp Cf,max
0 Vref d
< ASC,max
Cf,op = Cf,max (4.16)
else
C0 Vref d
Cf,op = ASC,max (4.17)
V f spp
In order to offer a standard deviation of the full-scale loss smaller than 25%
for capacitive sensor applications with > 0.05 and 200fF< C < 10pF, we
program ASC between 0 and 15.75 in steps of 0.25 (Fig. 4.11).
2.5 Noise
The noise of the system depends on the configuration settings of the front-end
circuit. The noise presented at the input of the modulator Vrms,noise contains
1
10
(Vfs ) (%)
rel
0
10
1 2 3 4 5 6 7 8 9 10
C (pF)
Figure 4.11. Standard deviation of the full-scale loss as a function of C, with ASC,step =
0.25 and d = 2 (differential sensors).
contributions of the C-V converter and the SC amplifier. This noise is calculated
with the techniques from section 5.
3. Optimal settings
For every application, optimal settings for Cref , Cf and ASC can be obtained.
They are the solution of an optimization problem, which minimizes the total
error (4.18).
Given: a capacitive sensor application with the following properties: d, C0 ,
, Cp1 , Cp2 , Gp and an ADC accuracy of b bits. For a single sensor application
(d = 1), we set Cref C0 . In a differential sensor application (d = 2), we
program Cref to compensate for the offset between Cx and Cx .
The algorithm minimizes the objective function:

(leakerr (Cx = C0 ))2 + (settlingerr (Cx = C0 ))2
2 b 2 (4.18)
2b 2 noise
+ 1V f srel + 1V f srel
where

(V f srel (det))2 + ( (V f srel (ASC )))2
V f srel = (4.19)
+ ( (V f srel (Cref )))2
This formula has contributions of the leakage error (4.8), settling error (4.11),
the noise bnoise , the ADC accuracy b, the deterministic full-scale loss (4.2) and
the stochastic full-scale losses due to ASC (4.14) and Cref (4.5).
The feasible solutions of the problem are subject to the following four con-
straints:
The outputs of the C-V converter are not allowed to saturate.

V f sCV Cf Vref 1 + C0 (4.20)
2
The ADC is not allowed to saturate.

Vref dC0 ASC
V f spp (4.21)
Cf
The lower and upper bound of Cf .
0 Cf 27pF (4.22)
The lower and upper bound of ASC .

0 ASC 15.75 (4.23)
4. Results
The computer algorithm is implemented in Matlab as a constraint based gra-
dient search optimization function and performs a smooth convergence. As
an example, we consider the case of bladder pressure measurements for uro-
dynamic tests. In this application we use a single capacitive absolute pressure
sensor (B012FA, VTI Technologies) to monitor the pressure in the range of 100-
130 kPa. The application properties are: = 0.1, C0 = 10.3pF, Gp = 0.001S
and Cp1 = Cp2 = 30pF. The ADC accuracy is set to 10 bits. Fig. 4.12 shows
the error function (4.18) for the feasible ASC and Cf values (4.20-4.23). The
algorithm gives Cref = 10.275pF, Cf = 16.2pF and ASC = 7.25 as optimal
solution. For these values, we obtain a leakage error of 0.01%, a negligible
settling error, a full-scale loss of 4% and noise bnoise of 11.5 bits. These set-
tings are the starting points of a calibration cycle. During this procedure, the
settings can be adjusted in order to cope with the technology variations of the
sensors and the interface chip. The measured full-scale loss is approximately
7%. The repeatability of the output is, for each applied pressure, within 1
Figure 4.12. Error function (4.18) for the feasible ASC and Cf values, the optimal settings
ASC = 7.25 and Cf = 16.2pF result in an error of 0.11%. The white region violates the
constraints (non-feasible region).
count. Hence, the dominant noise source is the quantization, as was predicted
by the calculated model (4.18).
5. Conclusion
The limited programmability of generic sensor interfaces results in a loss of
dynamic range (full-scale loss). Hence, the influence of noise increases and the
ADC accuracy is not exploited optimally. Furthermore, it is important that one
maximizes the interface accuracy for each application. This can be expressed
as an optimization problem. The minimized objective function considers the
leakage error, settling error, noise, ADC accuracy, deterministic and stochastic
full-scale losses. The obtained values for Cref , Cf and ASC estimate the
optimal configuration. These settings are the starting points of a calibration
cycle. During this procedure, the settings can be adjusted in order to cope with
the technology variations of the sensors and the interface chip.
Chapter 5
PHYSICAL ACTIVITY MONITORING SYSTEM
1. Introduction
The developed Generic Sensor Interface Chip enables low cost autonomous
sensor nodes for several state-of-the-art applications. This is illustrated in a
demonstrator for physical activity monitoring. The system can be used to
monitor the wellness of both humans and animals. The miniaturized sensor
node contains an accelerometer, the GSIC, a microcontroller and a wireless
transceiver. A computer interface displays the sensor data in real-time and
allows to change the interface settings during operation.
Section 2 presents the background and motivation for physical activity mon-
itoring. The implementation of the system is given in section 3.
2. Background and motivation

A physical activity monitoring system is useful for the surveillance and eval-
uation of patients [Ste03]. Such systems are used to motivate clinical groups
to exercise, including people with diabetes, obesity and congestive heart fail-
ure. Furthermore, the motion sensor technology helps to improve the quality
of rehabilitation programs and program changes.
The activity monitoring system contains an accelerometer, which monitors
the low frequent body movements (bandwidth < 20Hz). The accuracy of the
system is significantly increased by filtering the high frequent vibrations. For
this purpose, a sensor with low pass characteristic is used. Extra filtering is
performed by the read-out electronics (e.g. modulator and digital filtering
techniques). In most cases, the observation of the physical activity takes several
days or even weeks. Hence, Ultra Low Power sensor nodes are necessary to
enable long term monitoring.
Activity monitoring systems are also used for animal welfare observations. In
this application, the activity, heart rate and body temperature gives an indication
for the health of livestock [Wou95]. Hence, the productivity in farms can be
significantly increased.
3. Implementation
The physical activity gives a good indication for the health and wellness. For
this purpose, we have developed a wireless autonomous sensor system, which
monitors the acceleration in the range from -7.5 g to 7.5 g. This physical activity
monitoring has been implemented as a 3D stack, which contains a sensor, a
microcontroller and a wireless layer. The sensing layer is developed in this
work, whereas the microcontroller and wireless layers were already presented
in [Tor04]. Each layer has been made on a 14 mm x 14 mm printed circuit
board. The boards are connected vertically into a miniaturized cube (Fig. 5.1).
The sensor layer performs the sensing, amplification and analog-to-digital
conversion of the acceleration signal. It contains a differential capacitive ac-
celerometer (G012BA of VTI Technologies) and the GSIC (naked die covered
with globtop coating).
The microcontroller layer implements the data processing and control of the
entire module. An MSP430 microcontroller (Texas Instruments) was selected
for its low active power (0.6 nJ/ instruction), low standby power (2W) and fast
wakeup from standby to active mode (6s). To meet the size requirements, a
bare die component is used, which is encapsulated with globtop coating.
Figure 5.1. 3D stacked autonomous sensor cube, containing a sensor, a microcontroller and a
wireless layer (size 14mm x 14mm x 12mm).
Physical Activity Monitoring System 89
100
Digital output code
50
50
100
0 50 100 150 200 250

Sample
Figure 5.2. Output data for a sinusoidal acceleration (f=15Hz, peak-to-peak amplitude=10g,
offset=-1g).
The wireless layer consists of a single-chip short-range 2.4 GHz transceiver

(nRF2401, NORDIC, 18 nJ/bit) with a custom designed coplanar integrated
folded dipole antenna.
80
Power spectral density (dB/bin)
60
40
20
20
40
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
f/fs
Figure 5.3. Power Spectral Density of the output data for a sinusoidal acceleration (f=15Hz,
peak-to-peak amplitude=10g, offset=-1g).
The sensor node communicates with a remote basestation, which is imple-

mented on the PC (USB stick). A Labview computer interface displays the data
in real-time and allows to change the interface settings remotely.
We have selected a 120 Hz sample frequency and 8-bit accuracy for this
application. Hence, the GSIC operates in a duty cycle of 24%, which results
in a measured average current consumption of 16A. The total system (GSIC,
microcontroller and wireless core) consumes 700A. The dynamic performance
of the system is tested with shaker tests. In these tests, we applied a 15 Hz,
10 g (peak-to-peak) sinusoidal acceleration with 1 g (gravitational field) offset
on the sensor node (Fig. 5.2). Fig. 5.3 shows the measured output spectrum.
The harmonics in the spectrum correspond to a non-linearity of 4.1%. The rms
noise to signal ratio equals 0.7%.
4. Conclusion
A modular autonomous sensor node for physical activity monitoring has
been developed. The 3D stacked cube is approximately 1 cm3 and contains a
sensor board (accelerometer and GSIC), a microcontroller board and a wireless
board. In this application, the GSIC consumes approximately 48W for a
sample frequency of 120 Hz and 8-bit accuracy. The digital sensor data are
temporarily stored in the memory of the microcontroller. No data compression
is performed, although application specific algorithms could be implemented in
the microcontroller. The wireless transceiver sends the sensor data in packages
of four words to the remote USB transceiver. The computer interface displays
the data in real time and allows to change the configuration settings of the GSIC
via the bidirectional wireless communication link.
Chapter 6
CONCLUSION
1. Realized developments
An increasing number of medical diagnostics, comfort, entertainment and
sports applications are making use of sensor systems in and around the body.
Power autonomy is still a major challenge in these applications and a significant
part of the power consumption stems from the sensor interface circuitry. Most
often, the sensor interface is tailored towards one specific application. This
leads to a high recurrent design cost. An Ultra Low Power (ULP) generic
multi-sensor interface would offer a solution to both problems. Several research
groups have already developed generic sensor interface architectures. Because
the power dissipation of these systems is in the mW range, the power autonomy
problem is not solved. ULP sensor interface circuits have also been reported,
but these are dedicated to a specific sensor and therefore do not address the
generic application requirement.
This work presents a new generic architecture for autonomous sensor nodes.
The modular design methodology provides a flexible way to build a complete
sensor interface out of configurable blocks. The combination and the settings of
these blocks can be changed according to the varying needs of the application.
Furthermore, the sensor system can be expanded with additional building blocks
during the development phase.
This architecture is illustrated in a Generic Sensor Interface Chip (GSIC)
for capacitive sensors. The GSIC contains a microcontroller interface, a con-
figuration memory and the following configurable blocks: LF clock, sample
timer, reference and bias circuits, main oscillator and clock generation circuits,
Capacitance-to-Voltage (C-V) converters, Switched Capacitor (SC) amplifier,
voltage-to-current (VI) converter, modulator, decimation counter and conver-
sion counter.
The sensor interface converts a capacitance variation, C, into a proportional

digital output code. It consists of a SC interface followed by a modulator.
The SC interface (two C-V converters and a SC amplifier) operates on a lower
clock frequency, 8 kHz, than the modulator, 128 kHz, to achieve very low power
consumption. A new C-V converter with class AB and Correlated Double
Sampling (CDS) operation is developed. This C-V converter is approximately
three times more efficient in reducing the effects of the shunt conductance
than conventional C-V converters (with the same current consumption). The
mismatch between both C-V converters makes the system more susceptible to
interference. This can cause an extra loss in accuracy. The proposed chopping
scheme provides a solution for this problem. Hence, a pseudo differential
structure is built, where the capacitive sensor elements are connected to each
C-V converter for an equal number of interface periods. This modulates the
effects of the mismatch with the chopping sequence. These components are
filtered by the low pass operation of the modulator. The fully differential
SC amplifier amplifies the difference between the outputs of the C-V converters
and provides a quasi continuous input voltage for the modulator. The CT
modulator consists of a VI converter, a modulator and a decimation counter. It
produces an 8-bit, 9-bit or 10-bit code for a conversion time of 2, 4 or 8 ms
(INL and DNL are in all modes smaller than 0.5 LSB).
The main oscillator and clock generation circuits provide the clock signals
to the capacitive sensor interface and the decimation counter. The LF clock
is used for the timing during low-power standby operation. The reference and
bias circuits generate the bias currents for the sensor and the main oscillator.
The GSIC performs an interface to single and differential capacitive sen-
sors with 1pF< C0 <15pF, 200fF< C <10pF and 0.05< (= C/C0 ).
For single sensor operation, the on chip reference capacitor, Cref , needs to be
programmed to approximate C0 . For differential sensor operation, the on chip
reference capacitor is programmed to compensate for the offset between Cx and
Cx . In both modes the amplification factor, ASC , of the SC amplifier and the
feedback capacitor, Cf , of the C-V converters need to be programmed for op-
timal accuracy of the interface. The sample frequency is 6-bit programmable
between 2 Hz and 125 Hz. Both the LF clock and main oscillator are pro-
grammable to cope with technology variations.
The GSIC is designed in a 0.5m CMOS technology for a supply voltage
between 2.7 and 3.3V. The chip has a size of 3.2mm by 2.9mm. The GSIC
is tested with several pressure sensors and accelerometers. In the pressure
monitoring system, we measured an averaged power consumption of 7.3W
for a 10 Hz sample frequency and 8-bit accuracy in the 100 to 130 kPa range.
For the acceleration measurement system, it achieves a power consumption of
10.3W for a sample frequency of 10 Hz and 9-bit accuracy in the 1 g range.
The energy per accuracy was defined as a performance metric. It is shown that
Conclusion 93
the GSIC outperforms the existing generic sensor interfaces. It is even more
efficient than most of the dedicated ULP sensor interfaces.
A physical activity monitoring system is implemented in a 1 cm3 stack.
This demonstrator consists of a sensor layer (the GSIC and an accelerometer),
a microcontroller layer and a wireless layer. The bidirectional wireless link
(from the sensor node to the computer) makes it possible to display the data in
real time and to change the interface settings remotely. Hence, we have made a
smart autonomous sensor node, which can adapt at any time to changes in the
environment.
2. Suggestions for future work

Based on the developments presented in this work, some suggestions for
future work in the field of autonomous sensor interfaces are summarized:
The implemented GSIC can only interface with capacitive sensors. But
the modular architecture can be expanded with extra configurable blocks,
such as a temperature sensor or a (bio)potential interface. For this purpose,
programmable filters, instrumentation amplifiers, etc. need to be developed.
Hence, one could make a microsystem that can interface with several types
of sensors in different time intervals. This will result in applications such
as a personal health assistant, which monitors the blood pressure, body
temperature and heart rate.
Many biomedical applications need an extremely miniaturized sensor node.
One could reduce the size of the 3D stack with special packaging techniques
(e.g. 3D solder ball interconnect technology). Furthermore, the activity
monitoring system could be implemented on a flexible substrate, which is
very suitable for human body applications (smart bracelet). However, this
packaging technique effects the mechanical response (linearity, damping,
etc.) of the accelerometer system. New measurement set-ups should be
developed to estimate the real life performance and reliability of such a
system.
One could integrate a microcontroller core or a digital signal processing unit
with the interface electronics. Potential benefits are a reduced power con-
sumption (less buffering, processing unit is more dedicated to smart sensor
applications), a smaller size and a lower interconnection cost. The main
drawback is the increased effect of the substrate noise on the interface elec-
tronics [Ast04]. For this purpose, new guarding techniques, which separate
the sensitive ULP read-out electronics from the noise generating digital core,
should be explored.
The battery supply of the autonomous sensor node can be replaced by en-
ergy scavengers with conversion electronics. These electronics convert the
energy from the scavengers into a reliable supply voltage. They can be in-
tegrated with the GSIC (low cost) or implemented in a special technology
(higher conversion efficiency).
The narrowband communication front-end (NORDIC chip and folded dipole
antenna) can be replaced by an Ultra Wide Band solution. This would reduce
the power consumption significantly.
The drift in autonomous sensor nodes has to be studied. New algorithms
should separate the sensor signal from the long term drift. These algorithms
are able to adapt the interface settings of the GSIC (optimal settings) and to
digitally compensate for the sensor drift.
References
[Abi83] A. Abidi and R. G. Meyer, Noise in relaxation oscillators, IEEE Journal of

Solid-State Circuits, SC-18(6):794802, 1983.
[Ada86] R. Adams, Design and implementation of an audio 18 bit analog-to-digital con-
verter using oversampling techniques, Journal of Audio Engineering Society,
pages 153166, 1986.
[AMI] AMIS, AMIS-52100, Datasheet, http://www.amis.com.
[Ami00] R. Amirtharaja, S. Meninger, J. O. Mur-Miranda, A. Chandrakasan, and J. Lang,

A micropower programmable dsp powered using a mems based vibration to elec-
tric energy converter, In Proceedings IEEE International Solid-State Circuits
Conference, pages 362363, 2000.
[Ana] AnalogDevices, ADXL202E, Datasheet and application note: Minimizing power
consumption of iMEMS accelerometer, http://www.analog.com.
[Arm00] C. Arm, J.-M. Masgonty, and C. Piguet, Double latch clocking scheme for low
power ip cores, In Integrated Circuit Design Power and Timing Modeling, Opti-
mization and Simulation (PATMOS), pages 217224, 2000.
[Asa98] G. Asada, M. Dong, T. S. Lin, F. Newberg, G. Pottie, and W. J. Kaiser, Wireless
integrated network sensors: low power systems on a chip, In Proceedings of
European solid-state circuit conference (ESSCIRC), pages 916, 1998.
[Ast04] A. Astaras, S. P. Beaumont, T. B. Tang, A. F. Murray, L. Wang, and D. S. R. Cum-

ming, Noise analysis on integrated multisensor microsystems, In Proceedings
IEEE Sensors Conference, pages 995998, 2004.
[Bak98] R. Baker, H. Li, and D. Boyce, CMOS circuit design, layout and simulation,
chapter 26, pages 686691, IEEE Press, 1998.
[Bak00] A. Bakker and J. Huijsing, High-accuracy CMOS smart temperature sensors,

Kluwer Academic Publishers, 2000.
[Bao00] M. Bao, H. Yang, H. Yin, and S. Shen, Effects of electrostatic forces generated
by the driving signal on capacitive sensing devices, Sensors and Actuators A,
84:213219, 2000.
96 References
[Ben48] W. Bennet, Spectra of quantized signals, Bell Syst. Tech. J., BSTJ-27:446472,
1948.
[Bos88] B. Boser and B. A. Wooley, The design of sigma-delta modulation analog-to-

digital converters, IEEE Journal of Solid-State Circuits, 23(6):12981308, 1988.
[Cat04] M. Catrysse, Wireless power and data transmission for implantable and wearable
monitoring systems, PhD thesis, Katholieke Universiteit Leuven, 2004.
[Cha00] S. Chatzandroulis, D. Tsoulakas, and P. A. Neukomm, A miniature pressure

system with a capacitive sensor and a passive telemetry link for use in implantable
applications, IEEE Journal of Microelectromechanical systems, 9(1):1823, 2000.
[Cha02] A. V. Chavan and K. D. Wise, A monolithic fully integrated vacuum sealed cmos
pressure sensor, IEEE Transactions on Electron Devices, 49(1):164169, 2002.
[Che99] J. A. Cherry and W. M. Snelgrove, Clock jitter and quantizer metastability in

continuous time sigma-delta modulators, IEEE Transactions on Circuits and
Systems-II: Analog and Digital Signal Processing, 46(6):661676, 1999.
[Cla03] W. Claes, M. De Cooman, W. Sansen, and R. Puers, A 136 w/channel autonomous

strain-gauge datalogger, IEEE Journal of Solid-State Circuits, 38(12):22802287,
2003.
[Col] Colibrys, MS7002, Datasheet, http://www.colibrys.com.
[Coo05] J. Coosemans and R. Puers, An autonomous bladder pressure monitoring system,

Sensors and Actuators A, 123124:155161, 2005.
[Cra90] S. B. Crary, W. G. Baer, J. C. Cowles, and K. D. Wise, Digital compensa-

tion of high-performance silicon pressure transducers, Sensors and Actuators A,
2123:7072, 1990.
[DB02] D. De Bruyker, Intelligent silicon pressure sensors, PhD thesis, Katholieke Uni-
versiteit Leuven, 2002.
[Des05] G. Despesse, T. Jager, J. Chaillout, J. Leger, and S. Basrour, Design and fabri-
cation of a new system for vibration energy harvesting, In Proceedings of Ph.D.
research in Microelectronics and Electronics conference (PRIME), volume 1, pages
199202, 2005.
[DG96] B. De Geeter, O. Nys, and J.-P. Bardyn, A wide temperature range micropower
sensor interface circuit, In Proceedings of European solid-state circuit conference
(ESSCIRC), pages 136139, 1996.
[Egg00] T. Eggers, C. Marschner, B. Clasbrummel, R. Laur, J. Binder, K. Takahata, and

Y. Gianchandani, Advanced hybrid integrated low-power telemetric pressure mon-
itoring system for biomedical applications, In mems, pages 329334, 2000.
[Enz96] C. Enz and G. Temes, Circuit techniques for reducing the effects of op-amp
imperfections: autozeroing, correlated double sampling, and chopper stabilization,
Proceedings of the IEEE, 84(11):15841614, 1996.
[Fee91] O. Feely and L. O. Chua, The effect of integrator leak in sigma delta modulation,
IEEE Transactions on Circuits and Systems, 38:12931305, 1991.
References 97
[Fis82] J. H. Fischer, Noise sources and calculation techniques for switched capacitor
filters, IEEE Journal of Solid-State Circuits, SC-17(8):742752, 1982.
[Fre98] P. J. French and P. M. Sarro, Surface versus bulk micromachining: the contest for
suitable applications, Journal of Micromechanics and microengineering, 8:4553,
1998.
[Har99] R. Harjani, R. Heineke, and F. Wang, An integrated low-voltage class ab cmos

ota, IEEE Journal of Solid-State Circuits, 34(2):134142, 1999.
[Hos97] B. Hosticka, Integrated sensor systems in CMOS technology, pages 219241,

Analog Circuit Design, W. Sansen, J. Huijsing, R. J. van de Plassche (eds.) Kluwer
Academic Publishers, 1997.
[Hua03] Q. Huang, M. Qin, Z. Zhang, M. Zhou, L. Gu, H. Zhu, D. Hu, Z. Hu, G. Xu, and
Z. Liu, Weather station on a chip, In Proceedings IEEE Sensors Conference,
pages 11061113, 2003.
[Ing97] M. Ingels and M. Steyaert, Design strategies and decoupling techniques for re-
ducing the effects of electrical interference in mixed-mode ics, IEEE Journal of
Solid-State Circuits, 32(7):11361141, 1997.
[Jac03] P. Jacobs, J. Janssens, T. Geurts, and J. Crols, A 0.35m cmos fractional-n trans-
mitter for 315/433/868/915 mhz ism applications, In Proceedings of European
solid-state circuit conference (ESSCIRC), pages 425428, 2003.
[Jes00] P. G. A. Jespers, D to A and A to D architectures, analysis and simulation, chapter 7,

pages 423427, Oxford University Press, 2000.
[Joh97] D. A. Johns and K. Martin, Analog integrated circuit design, pages 423427,
John Wiley & Sons, 1997.
[Kaj02] T. Kajita, U. K. Moon, and G. Temes, A two chip interface for a mems accelerom-
eter, IEEE Transactions on Instrumentation and Measurement, 51(4):853858,
2002.
[Kan00] U. Kang and K. D. Wise, A high-speed capacitive humidity sensor with on-chip
thermal reset, IEEE Transactions on Electron Devices, 47(4):702710, 2000.
[Kha03] Q. A. Khan, S. K. Wadhwa, and K. Misri, Low power startup circuits for voltage
and current reference with zero steady state current, In ISLPED, pages 184188,
2003.
[Kol04] C. Kolle, W. Scherr, D. Hammerschmidt, G. Pichler, M. Motz, B. Schaffer,

B. Forster, and U. Ausserlechner, Ultra low power monolithically integrated
capacitive pressure sensor for tire pressure monitoring, In Proceedings IEEE
Sensors Conference, pages 244247, 2004.
[Kul03] H. Kulah, Closed-loop electromechanical sigma-delta microgravity accelerome-

ters, PhD thesis, University of Michigan, 2003.
[Kul06] H. Kulah, J. Chae, N. Yazdi, and K. Najafi, Noise analysis and characterization of a
sigma-delta capacitive microaccelerometer, IEEE Journal of Solid-State Circuits,
41(2):352361, 2006.
98 References
[Kwa91] T. Kwan and K. Martin, An adaptive analog continuous time cmos biquadratic
filter, IEEE Journal of Solid-State Circuits, 26:859867, 1991.
[Lac03] J. Laconte, V. Wilmart, J.-P. Raskin, and D. Flandre, A high-sensitivity capacitive

humidity sensor using 3-layer patterned polyimide sensing film, In Proceedings
IEEE Sensors Conference, pages 106107, 2003.
[Lam83] K. K. Lam and M. A. Copeland, Noise cancelling switched-capacitor filtering

technique, IEE Electronics Letters, 19:810811, 1983.
[Lap96] D. Lapadatu, M. De Cooman, and R. Puers, A double-sided capacitive miniaturized

accelerometer based on photovoltaic etch-stop technique, Sensors and Actuators
A, 53:261266, 1996.
[Lem99] M. Lemkin and B. Boser, A three axis micromachined accelerometer with a

cmos position sense interface and digital offset trim electronics, IEEE Journal of
[Leo05] V. Leonov, P. Fiorini, S. Sedky, T. Torfs, and C. Van Hoof, Thermoelectric mems
generators as a power supply for a body area network, In Dig. Int. Conf. on
Solid-State Sensors and Actuators (Transducers), pages 291294, 2005.
[Li00] X. Li and G. C. M. Meijer, Elimination of shunting conductance effects in a

low cost capacitive sensor interface, IEEE Transactions on Instrumentation and
Measurement, 49:531534, 2000.
[Li02] X. Li and G. C. M. Meijer, An accurate interface for capacitive sensors, IEEE

Transactions on Instrumentation and Measurement, 51(5):935939, 2002.
[Mar87] K. Martin, L. Oczolak, Y. S. Lee, and G. Temes, A differential switched-capacitor

amplifier, IEEE Journal of Solid-State Circuits, 22(1):104106, 1987.
[Mas98] A. Mason, N. Yazdi, A. V. Chavan, K. Najafi, and K. D. Wise, A generic multiele-

ment microsystem for portable wireless applications, Proceedings of the IEEE,
86(8):17331746, 1998.
[Mei86] G. C. M. Meijer, Thermal sensors based on transistors, Sensors and Actuators

A, 10:103125, 1986.
[Mei02] G. C. M. Meijer and X. Li, Smart sensor interface electronics, In Proceedings

of the 23rd international conference on microelectronics (MIEL), pages 106107,
2002.
[Mid89] S. Middelhoek and S. Steigerwald-Audet, Silicon sensors, Academic Press,

London, 1989.
[Nor] Nordic, nRF24E1, Datasheet, http://www.nordicsemi.com.
[Per94] J.-F. Perotto, C. Lamothe, C. Arm, C. Piguet, E. Dijkstra, S. Fink, E. Sanchez, and
J.-P. Wattenhofer, An 8-bit multitask micropower risc core, IEEE Journal of
[Pet06] V. P. Petkov and B. Boser, High-order electromechanical sigma delta modulation

in micromachined inertial sensors, IEEE Transactions on Circuits and Systems-I:
Fundamental Theory and Applications, 53(5):10161022, 2006.
References 99
[Pue90] R. Puers, E. Peeters, A. Van Den Bossche, and W. Sansen, A capacitive pressure
sensor with low impedance output and active suppression of parasitic effects,
Sensors and Actuators A, 2123:108114, 1990.
[Pue93] R. Puers, Capacitive sensors: when and how to use them, Sensors and Actuators
A, 3738:93105, 1993.
[Pue96] R. Puers and D. Lapadatu, Electrostatic forces and their effects on capacitive
mechanical sensors, Sensors and Actuators A, 56:203210, 1996.
[Pue97] R. Puers and P. Wouters, Adaptable interface circuits for flexible monitoring of
temperature and movement, Journal of Analog Integrated Circuits and Signal
Processing, 14:193206, 1997.
[Pue99] R. Puers, Sensor, sensor interfacing and front-end data management for stand-
alone microsystems, Journal of Micromechanics and microengineering, 9:R1R7,
1999.
[Pue00] R. Puers, M. Catrysse, G. Vandevoorde, R. J. Collier, E. Louridas, F. Burny,

M. Donkerwolcke, and F. Moulart, A telemetry system for the detection of hip
prosthesis loosening by vibration analysis, Sensors and Actuators A, 85:4247,
2000.
[Rab02] J. Rabaey, J. Ammer, T. Karalar, S. Li, B. Otis, M. Sheets, and T. Tuan, Picoradios
for wireless sensor networks: the next challenge in ultra low power design, In
Proceedings IEEE International Solid-State Circuits Conference, pages 200361,
2002.
[Rie93] F. R. Riedijk, Integrated smart sensors with digital bus interface, PhD thesis,
Technische Universiteit Delft, 1993.
[Ryc05] J. Ryckaert, C. Desset, A. Fort, M. Badaroglu, V. De Heyn, P. Wambacq, G. Van der

Plas, S. Donnay, B. Van Poucke, and B. Gyselinckx, Ultra-wideband transmit-
ter for low power wireless body area networks: design and evaluation, IEEE
Transactions on Circuits and Systems-I: Fundamental Theory and Applications,
52(12):25152525, 2005.
[San82] W. Sansen, On the integration of an internal human conditioning system, IEEE

Journal of Solid-State Circuits, 17:513521, 1982.
[San94] W. Sansen and K. R. Laker, Design of analog integrated circuits and systems,
McGraw-Hill, 1994.
[Sei90] H. Seidel, H. Riedel, R. Kolbeck, G. Muck, W. Kupke, and M. Koniger, Capacitive

silicon accelerometer with highly symmetrical design, Sensors and Actuators A,
21:312315, 1990.
[Sel97] A. Selvakumar and K. Najafi, A high-sensitivity z-axis capacitive silicon microac-

celerometer with a torsional suspension, IEEE Journal of Microelectromechanical
systems, 7(2):192200, 1997.
[ST] ST, LIS3LV02DQ, Datasheet, http://www.st.com.

100 References
[Ste03] B. G. Steele, B. Belza, K. Cain, C. Warms, J. Coppersmith, and J. Howard, Bodies

in motion: monitoring daily activity and exercise with motion sensors in people with
chronic pulmonary disease, Journal of Rehabilitation Research & Development,
40(5):4558, 2003.
[Ste05] T. Sterken, P. Fiorini, and R. Puers, Parametrical study of miniature generators for
large motion applications, In Proceedings of Eurosensors XIX, pages MC1MC2,
2005.
[Tan02] T. B. Tang, E. A. Johannessen, L. Wang, A. Astaras, M. Ahmadian, A. F. Murray,

J. M. Cooper, S. P. Beaumont, B. W. Flynn, and D. S. R. Cumming, Toward a
miniature wireless integrated multisensor microsystem for industrial and biomed-
ical applications, IEEE Sensors Journal, 2(6):628635, 2002.
[Tay00] F. E. H. Tay, J. Xu, and V. J. Logeeswaran, A differential low-g microaccelerometer

with mg resolution, Sensors and Actuators A, 86:4651, 2000.
[Tor04] T. Torfs, S. Sanders, C. Winters, S. Brebels, and C. Van Hoof, Wireless network
of autonomous environmental sensors, In Proceedings IEEE Sensors Conference,
volume 2, pages 923926, 2004.
[VDG96] F. M. L. Van Der Goes, Low-cost smart sensor interfacing, PhD thesis, Technische
Universiteit Delft, 1996.
[VTI] VTI, SCP1000-D01, Datasheet, http://www.vti.fi.
[Wak89] M. Wakayama and A. Abidi, A 30 mhz low-jitter high-linearity cmos voltage-

controlled oscillator, IEEE Journal of Solid-State Circuits, 24(2):325330, 1989.
[Wan98] B. Wang, T. Kajita, T. Sun, and G. Temes, High accuracy circuits for on-chip
capacitive ratio testing and sensor readout, IEEE Transactions on Instrumentation
and Measurement, 47(1):1620, 1998.
[Wan00] G. Wang and G. C. M. Meijer, The temperature characteristics of bipolar transistors

fabricated in cmos technology, Sensors and Actuators A, 87:8189, 2000.
[War01] B. A. Warneke, M. Last, B. Leibowitz, and K. S. J. Pister, Smart dust: com-

municating with a cubic-millimeter computer, IEEE Computer Magazine, pages
4451, 2001.
[War03] B. A. Warneke, Ultra-low energy architectures and circuits for cubic millime-
ter distributed wireless sensor networks, PhD thesis, University of California at
Berkeley, 2003.
[Won04] L. S. Y. Wong, S. Hossain, A. Ta, J. Edvinsson, D. H. Rivas, and H. Naas, A very

low power cmos mixed signal ic for implantable pacemaker applications, IEEE
Journal of Solid-State Circuits, 39(12):24462456, 2004.
[Wou95] P. Wouters, Intelligent and low power sensor interface systems for biotelemetry
applications, PhD thesis, Katholieke Universiteit Leuven, 1995.
[Yaz98] N. Yazdi, F. Ayazi, and K. Najafi, Micromachined inertial sensors, Proceedings

of the IEEE, 86(8):16401659, 1998.
References 101
[Yaz00] N. Yazdi, A. Mason, K. Najafi, and K. D. Wise, A generic interface chip for capac-
itive sensors in low-power multi-parameter microsystems, Sensors and Actuators
A, 84:351361, 2000.
[Yaz03] N. Yazdi and K. Najafi, A high-sensitivity silicon accelerometer with a folded-

electrode structure, IEEE Journal of Microelectromechanical systems, 12(4):
479486, 2003.
[Yaz04] N. Yazdi, H. Kulah, and K. Najafi, Precision readout circuits for capacitive mi-
croaccelerometers, In Proceedings IEEE Sensors Conference, pages 2831, 2004.
[Yin92] G. M. Yin, F. Op t Eynde, and W. Sansen, A high-speed cmos comparator with

8-b resolution, IEEE Journal of Solid-State Circuits, 27:208211, 1992.
[Yos97] Y. Yoshii, A. Nakajo, H. Abe, K. Ninomiya, H. Miyashita, N. Sakuri, M. Kosuge,

and S. Hao, 1 chip integrated software calibrated cmos pressure sensor with mcu,
a/d convertor, d/a convertor, digital communication port, signal conditioning circuit
and temperature sensor, In Dig. Int. Conf. on Solid-State Sensors and Actuators
(Transducers), pages 14851488, 1997.
[You98] F. You, S. H. K. Embabi, and E. Sanchez-Sinencio, Low-voltage class ab buffers

with quiescent current control, IEEE Journal of Solid-State Circuits, 33(6):
915920, 1998.
Index
Accelerometer, 65, 88 Decimation filter, 34

Adaptive load, 47 Delay cell, 41
ADC accuracy, 43, 52 Differential sensor, 24, 51
Amplification factor, 51 Digital signal processing unit, 8, 93
Asynchronous data transfer, 14 Duty cycle operation, 1314, 20, 33, 48
Autonomous sensor cube, 8890
Effective number of bits, 5960
Bandgap reference voltage, 4346 Electromagnetic interference, 2729
Bennet model, 53 Electrostatic force, 22, 27
Bias circuits, 47 Energy conversion electronics, 94
Bipolar transistor Energy efficiency, 69
low-level injection effects, 46 Energy per accuracy level, 69
Bladder pressure observation, 6264 Energy scavengers, 9, 93
Break-before-make, 36, 38
Buffered reference voltage, 4748
Feasible solutions, 83
Bulk micromachining, 18
Feedback capacitor, 51
Flexible substrate, 93
Calibration, 6, 53, 63, 84
Folded cascode OTA, 3233
look-up table algorithm, 8
Folded dipole antenna, 89
polynomial evaluation, 8
Full-scale loss, 7475
Capacitance-to-Voltage converter, 2429
deterministic, 75
Charge injection imbalance, 35, 40, 43
stochastic, 75
Charge injections, 27, 36
ASC , 82
Chopping, 7, 2729
Cref , 76
enhanced, 29
Class AB
buffer, 4748 Guard rings, 61
OTA, 2527
Clock jitter, 35, 49 Harmonic driven AC-bridge, 21
Closed loop accelerometers, 22
Common mode feedback, 32 Inclination monitoring, 6568
Computer algorithm, 53, 84 Inclinometer calibration fixture, 65
Configuration settings, 5153 Integrator, 3940
Constraint, 83 Integrator leakage, 35, 36, 39
Constraint based optimization, 83 Intersymbol interference, 35
Continuous time comparator, 50
Correlated Double Sampling, 7, 25 Labview computer interface, 90
slew enhanced, 2932 Leakage error, 25, 27, 7780
Current boosting, 2527 LF clock, 5051
Current DAC, 3839 frequency, 52
104 Index
Main oscillator, 4850 Reference current generator, 4647

frequency, 52 Regenerative comparator, 4042
Mean capacitance, 19 Relative full-scale deviation, 19
Microcontroller, 8, 69, 88, 93 Relaxation oscillator, 4850
Mismatch, 27
Modular sensor interface chip, 1014, 20, 52 Sample period, 52
SC amplifier, 2933
Noise, 35, 5360, 82 SC interface, 21
electrical, 53 Self-biased differential amplifier, 50
mechanical, 53 Sensor non-ideality, 67
quantization, 53 cross-sensitivity, 6
Noise excess factor, 55 drift, 7, 94
Non-linearity, 43 non-linearity, 6
differential (DNL), 43 production spread, 6
integral (INL), 43 Settling error, 7780
Non-overlapping delayed clocks, 27, 30, 41 Shaker tests, 90
Sigma Delta modulator, 22, 29, 3343
Objective function, 53, 83 CT, 3435
Operational flow, 1314 NRZ, 35
Optimal settings, 83 RZ, 3536
Output, 69 SC, 34
analog, 69 Single sensor, 24, 51
digital, 69 Smart energy management, 14, 20, 33, 52, 61
quasi-digital, 69 Solder ball interconnect technology, 93
Square wave driven AC-bridge, 21
Parasitic capacitance, 18, 24 SR latch, 39
Parasitic shunt conductance, 18, 23, 25, 77 State model, 3032
Personal health assistant, 93 Step size, 74
Physical activity monitoring, 8788 Substrate noise, 93
Piezoresistive sensors, 18 Supply strategy, 61
Power management, 910 Surface micromachining, 18
active devices, 9
passive devices, 10
Three-signal autocalibration, 25
Pressure sensor, 62
Total harmonic distortion, 34, 37
Programmability, 7383
ASC , 8082
Cf , 7680 USB transceiver stick, 90
Cref , 7576
Programming flow, 1113 Vertical bipolar transistors, 44
Proportional to absolute temperature, 44 Voltage-to-current converter, 34, 3637
Quantizer, 35, 4042 Weak inversion, 26, 32

Wireless transceiver, 9
Range, 74 high-frequency, 9
Reconfiguration, 1113 narrowband, 9, 89, 94
complete, 1113 ultra wide band, 9, 94
fast, 1113 low-frequency, 9
Reference capacitor, 24, 51 inductive coupling, 9

Ultra Low Power Capacitive Sensor Interfaces

Uploaded by

Copyright:

Available Formats

Ultra Low Power Capacitive Sensor Interfaces

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ultra Low Power Capacitive Sensor Interfaces

Uploaded by

Copyright:

Available Formats

ULTRA LOW POWER CAPACITIVE

CHRIS VAN HOOF

ISBN 978-1-4020-6231-5 (HB)

Printed on acid-free paper

All Rights Reserved

3.6 Bandgap reference, bias system and buffered

The increasing performance of smart microsystems merging sensors, signal

The drive towards an intelligent environment has lead to an increased need

(ISFET), a standard PN-junction silicon temperature sensor and a dual elec-

In chapter 2 an overview of the most important design aspects for au-

Sensor interface chip

Sensor 4 Operational settings

Figure 2.1. General multisensor microsystem for autonomous sensor applications.

The most difficult non-ideality to compensate is drift. Knowledge of the

2.2 Sensor interface chip

reconfiguration of the sensor front-end with a reduced complexity (limited num-

Table 2.1. Energy consumption of various microcontrollers.

Microcontroller Energy (pJ/instruction/bit)

2.4 Wireless link

2.5 Power management

3. Modular design methodology

Sensor interface chip

Sample Conversion Reset

POR Ack Dout Dav Write Dclk Din Activate

Figure 2.2. Functional description of the modular sensor interface chip.

programmable amplifiers, filters, data converters, clocks, etc. The configura-

3.1 Programming flow

2 If (ERN0 ERN1) equals (0 1), only address A0 A1 A2 A3 needs to be

Write bit becomes high

The operation flow is reset

All configuration flags are low

Activate becomes low

The code word

The code word

Write becomes low

The operation flow starts

Figure 2.3. Flow chart for the complete reconfiguration mode.

Write bit becomes high

The operation flow is reset

The code word

Write becomes low

The operation flow starts

Figure 2.4. Flow chart for the fast reconfiguration mode.

the case of the smart energy management in event-triggered microsystems.

3.2 Operational flow

The interface chip is programmed

After 1 count, Act becomes high

The bias currents of the analog front-

After 2 counts, Reset becomes low

The conversion process starts

The conversion is finished, the data are

The microcontroller loads the digital

Ack becomes low

The sample timer reaches Nsample counts

The sample timer restarts counting

Figure 2.5. Operational flow of the modular microsystem.

this asynchronous transfer are unimportant. The system functions correctly, if

GENERIC SENSOR INTERFACE CHIP

The proposed readout circuit provides an interface to single and differential