Nano Electronics

This page intentionally left blank
Introduction to Nanoelectronics
Increasing miniaturization of devices, components, and integrated systems requires
developments in the capacity to measure, organize, and manipulate matter at the
nanoscale. This textbook is a comprehensive, interdisciplinary account of the technology
and science that underpin nanoelectronics, covering the underlying physics, nanostruc-
tures, nanomaterials, and nanodevices.
Without assuming prior knowledge of quantum physics, this book provides a unifying
framework for the basic ideas needed to understand the recent developments in the eld.
Following an introductory description of recent trends in semiconductor and device
nanotechnologies, as well as novel device concepts, materials for nanoelectronics are
treated, covering methods of growth, fabrication and characterization. Treatment then
moves to an analysis of nanostructures, including recently discovered nanoobjects, and
concludes with a discussion of devices that use a simple scaling-down approach to
copy well-known microelectronic devices, and nanodevices based on newprinciples that
cannot be realized at the macroscale.
Numerous illustrations, homework problems and interactive Java applets help the
student to appreciate the basic principles of nanotechnology, and to apply them to real
problems. Written in a clear yet rigorous and interdisciplinary manner, this textbook is
suitable for advanced undergraduate and graduate students in electrical and electronic
engineering, nanoscience, materials, bioengineering, and chemical engineering.
Further resources for this title, including instructor solutions and Java applets, are
available online at www.cambridge.org/9780521881722.
Vladimir V. Mitin is a Professor and Chair of the Department of Electrical Engineering
at the University of Buffalo, State University of New York. He has co-authored eight
books, and over 400 professional publications, including ten patents.
Viatcheslav A. Kochelap is Professor and Head of the Theoretical Physics Department
at the Institute of Semiconductor Physics National Academy of Sciences, Ukraine. He
has published over 200 journal articles.
Michael A. Stroscio is a Professor in the departments of Electrical and Computer Engi-
neering, and Bioengineering, at the University of Illinois at Chicago. He is a Fellow of
the IEEE, the American Physical Society, and the AAAS.
Introduction to Nanoelectronics
Science, Nanotechnology, Engineering,
and Applications
VLADI MI R V. MI TI N
University of Buffalo, State University of New York
VI ATCHESLAV A. KOCHELAP
Institute of Semiconductor Physics National Academy of Sciences, Ukraine
MI CHAEL A. STROSCI O
University of Illinois at Chicago
CAMBRIDGE UNIVERSITY PRESS
Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, So Paulo
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
First published in print format
ISBN-13 978-0-521-88172-2
ISBN-13 978-0-511-36811-0
Cambridge University Press 2008
2007
Information on this title: www.cambridge.org/9780521881722
This publication is in copyright. Subject to statutory exception and to the provision of
relevant collective licensing agreements, no reproduction of any part may take place
without the written permission of Cambridge University Press.
ISBN-10 0-511-36811-9
ISBN-10 0-521-88172-2
Cambridge University Press has no responsibility for the persistence or accuracy of urls
for external or third-party internet websites referred to in this publication, and does not
guarantee that any content on such websites is, or will remain, accurate or appropriate.
Published in the United States of America by Cambridge University Press, New York
www.cambridge.org
hardback
eBook (Adobe Reader)
eBook (Adobe Reader)
hardback
Contents
Preface page vii
List of notation xii
1 Toward the nanoscale 1
2 Particles and waves 11
2.1 Introduction 11
2.2 Classical particles 11
2.3 Classical waves 13
2.4 Waveparticle duality 23
2.5 Closing remarks 28
2.6 Problems 29
3 Wave mechanics 33
3.1 Introduction 33
3.2 The Schr odinger wave equation 33
3.3 Wave mechanics of particles: selected examples 38
3.4 Atoms and atomic orbitals 55
3.6 Problems 63
4 Materials for nanoelectronics 65
4.1 Introduction 65
4.2 Semiconductors 66
4.3 Crystal lattices: bonding in crystals 68
4.4 Electron energy bands 73
4.5 Semiconductor heterostructures 84
4.6 Lattice-matched and pseudomorphic heterostructures 87
4.7 Organic semiconductors 95
4.8 Carbon nanomaterials: nanotubes and fullerenes 98
4.10 Problems 107
vi Contents
5 Growth, fabrication, and measurement techniques for nanostructures 109
5.1 Introduction 109
5.2 Bulk crystal and heterostructure growth 110
5.3 Nanolithography, etching, and other means for fabrication of
nanostructures and nanodevices 115
5.4 Techniques for characterization of nanostructures 120
5.5 Spontaneous formation and ordering of nanostructures 127
5.6 Clusters and nanocrystals 134
5.7 Methods of nanotube growth 136
5.8 Chemical and biological methods for nanoscale fabrication 141
5.9 Fabrication of nanoelectromechanical systems 157
5.11 Problems 163
6 Electron transport in semiconductors and nanostructures 165
6.2 Time and length scales of the electrons in solids 165
6.3 Statistics of the electrons in solids and nanostructures 172
6.4 The density of states of electrons in nanostructures 180
6.5 Electron transport in nanostructures 183
6.7 Problems 215
7 Electrons in traditional low-dimensional structures 218
7.2 Electrons in quantum wells 218
7.3 Electrons in quantum wires 228
7.4 Electrons in quantum dots 231
7.6 Problems 238
8 Nanostructure devices 242
8.2 Resonant-tunneling diodes 242
8.3 Field-effect transistors 255
8.4 Single-electron-transfer devices 269
8.5 Potential-effect transistors 276
8.6 Light-emitting diodes and lasers 284
8.7 Nanoelectromechanical system devices 306
8.8 Quantum-dot cellular automata 317
Appendix: tables of units 323
Index 325
Preface
Welcome to the amazing nanoworld! In this book you will nd fundamental princi-
ples in nanoscience and basic techniques of measurement, as well as fabrication and
manipulation of matter at the nanoscale. The book discusses how these principles, tech-
niques, and technologies are applied to the newest generation of electronics, known as
nanoelectronics.
The science of atoms and simple molecules, and the science of matter frommicrostruc-
tures to larger scales, are both well established. A remaining, extremely important, size-
related challenge is at the nanoscale roughly the dimensional scales between 10 and 100
molecular diameters where the fundamental properties of materials are determined and
can be engineered. This eld of science nanoscience is a broad and interdisciplinary
eld of emerging research and development.
Nanotechnology is concerned with materials, structures, and systems whose compo-
nents exhibit novel and signicantly modied physical, chemical, and biological prop-
erties due to their nanoscale sizes. A principal goal of nanotechnology is to control and
exploit these properties in structures and devices at atomic, molecular, and supramolecu-
lar levels. To realize this goal, it is essential to learn howto fabricate and use these devices
efciently. Nanotechnology has enjoyed explosive growth in the past fewyears. In partic-
ular, nanofabrication techniques have advanced tremendously in recent years. Obviously,
revolutionary changes in the ability to measure, organize, and manipulate matter on the
nanoscale are highly benecial for electronics with its persistent trend of downscaling
devices, components, and integrated systems. In turn, the miniaturization required by
electronics is one of the major driving forces for nanoscience and nanotechnology.
Practical implementations of nanoscience and nanotechnology have great importance,
and they depend critically on training people in these elds. Thus, modern education
needs to address the rapidly evolving facets of nanoscience and nanotechnology. A new
generation of researchers, technologists, and engineers has to be trained in the emerg-
ing nanodisciplines. With the purpose of contributing to education in the nanoelds,
we present this textbook providing a unifying framework for the basic ideas needed to
understand recent developments underlying nanoscience and nanotechnology, as applied
to nanoelectronics. The book grew out of the authors research and teaching experience
in these subjects. We have found that many of the ideas and achievements in elds under-
lying nanoscience and nanotechnology can be explained in a relatively simple setting,
if the necessary foundational underpinnings are presented properly. We have designed
this textbook mainly for undergraduate students, who will be trained in diverse elds
viii Preface
including nanoscience, physics of material devices, electrical and optical engineering,
materials science and engineering, and mechanical engineering. It can be helpful also
for training students in bioengineering and chemical engineering. To reach such a broad
audience, materials are presented in such a way that an instructor can choose the level
of presentation depending on the backgrounds of the students. For example, we have
included Chapters 2 and 3 in part for students who have not taken a quantum mechanics
course. An analogy with wave elds elastic waves and optical waves is exploited
widely to introduce wave mechanics of particles and quantumprinciples, which play key
roles in the interpretation of the properties of nanomaterials.
One of us (V.V.M.) has taught the course for students in the second semester of their
sophomore year. For students at this level, Chapters 2 and 3 were covered in detail and,
consequently, there was not enough time to cover all of the devices that are discussed in
Chapter 8. If students using the book have previously taken courses on quantummechan-
ics and electromagnetics, the instructor may start from Chapter 4. This book may be also
used as an introductory graduate or senior undergraduate course. Another of us (M.A.S.)
has used Chapters 2 and 3 as the introduction to a graduate course on nanoelectronics for
a class with students drawn from electrical engineering, materials engineering, chem-
ical engineering, mechanical engineering, and physics. By covering Chapters 2 and 3
at the beginning of the course, the students can then proceed from this common basis
in quantum mechanics and other underlying areas of physics to cover more advanced
topics, either in the current text or in other texts such as Quantum Heterostructures by
V. Mitin, V. Kochelap, and M. Stroscio. The latter approach has been used by M.A.S. in
teaching nanoelectronics to graduate students with diverse backgrounds in many disci-
plines within engineering and the physical sciences. For this purpose, we include details
of derivations and mathematical justication of concepts in some sections. These details
can be omitted from an undergraduate curriculum.
The book contains homework problems on various subjects. These problems illustrate
the basic material and help students to understand and learn the basic principles of the
nanoscience and the nanotechnology.
* * * * *
Essentially, the chapters are organized into three main groups.
Chapters 13 are of an introductory character. In Chapter 1, we present in concise form
the main subject of the book. The recent and diverse trends in semiconductor and device
nanotechnologies, as well as novel concepts of nanodevices, are reviewed. These trends
make it clear why understanding the fundamentals of nanoscience is of great importance.
Chapters 2 and 3 are written for students who have not taken a quantum mechanics
course. An analogy with wave elds (elastic waves and optical waves) is exploited widely
to introduce wave mechanics of particles and the quantum principles, which play key
roles in the interpretation of the properties of nanomaterials.
In Chapter 2, we explain that the fundamental laws of physics governing particles
and material elds in the nanoworld are different from those that apply to familiar
macroscopic phenomena. Starting with an analysis of an arbitrary wave eld (elas-
tic vibrations in solids, electromagnetic elds, etc.), we nd particle-like behavior of
this wave eld for small wave amplitudes and (or) for spatial scales larger than the
Preface ix
wavelength. By analyzing particle motion, we show that at small spatial scales a particle
can not be characterized by exact coordinates and momentumand that it behaves rather as
an extended wave-like object. This analysis establishes the essence of the waveparticle
duality which is an underlying principle of nanophysics.
In Chapter 3, we discuss the basic physical concepts and equations related to the
behavior of particles in the nanoworld. We introduce the Schr odinger wave equation for
particles and determine the ways in which to calculate observable physical quantities.
Keeping in mind the diverse variants of nanostructures, by using wave mechanics we
analyze a number of particular examples, which highlight important quantum proper-
ties of particles. Many of the examples analyzed can serve as the simplest models of
nanostructures and are exploited in later chapters.
Chapters 4 and 5 are devoted to materials used in nanoelectronics, methods of their
growth, and fabrication and characterization techniques.
In Chapter 4, we present an overview of the basic materials that are exploited in
nanoelectronics. We start with semiconductor materials as the principal candidates for
use in nanoelectronics, because they offer great exibility in the control of the elec-
tronic and optical properties, and functions, of nanoelectronic devices. We show how,
through proper regimes of growth, doping by impurities, and sequent modications and
processing, one can fabricate nanostructures and nanodevices starting from bulk-like
materials. Then, we introduce other materials that have properties of great potential in
nanoelectronics. Organic semiconductors and carbon nanotubes are discussed.
In Chapter 5, the principal methods of materials growth and nanodevice fabrication are
presented. We start with an analysis of fabrication of nanodevices on the basis of perfect
materials and continue by considering processing techniques. All stages of fabrication
and methods of processing are considered in detail. Then, we discuss special regimes of
material growth, when nanostructures (mainly quantum dots) are formed spontaneously
due to the growth kinetics. These approaches to the production of nanostructures and
nanoelectronic devices actually represent evolutionary improvements in the growth
and processing methods applied previously in microelectronics. Nanoscale objects like
carbon nanotubes and biomolecules require, in general, other techniques for production.
These innovative techniques are also highlighted in this chapter.
We pay special attention to the most important characterization techniques, such
as atomic-force microscopy, scanning tunneling microscopy, and transmission electron
microscopy, among others.
Also in Chapter 5, we review advances in nanotechnology that came from syn-
thetic chemistry and biology. These include chemical and biological methods of surface
nanopatterning for preparing nanostructured materials with predened and syntheti-
cally programmable properties. The basic ideas related to these chemical and biological
approaches are discussed. Finally, we study the methods of fabrication of a new class of
devices commonly known as nanoelectromechanical systems (NEMS).
Chapters 68 include analyses of electron properties of nanostructures, traditional
low-dimensional systems, and recently discovered nano-objects.
In Chapter 6, transport of charge carriers is analyzed. Important aspects of transport
regimes are elucidated by comparing the time and length scales of the carriers with
x Preface
device dimensions and device temporal phenomena related to operating frequencies.
Then, we consider the behavior of the electrons in high electric elds, including hot-
electron effects. For short devices, we describe dissipative transport and the velocity
overshoot effect as well as semiclassical ballistic motion of the electrons. We present
ideas on quantum transport in nanoscale devices.
To distinguish the nanostructures already having applications fromthe newly emerging
systems, we refer to the former as traditional low-dimensional structures (quantumwells,
quantum wires, and quantum dots). These structures are considered in Chapter 7.
In Chapter 8, we consider newly emerging electronic, optical, and electromechan-
ical devices based on nanostructures. First, we discuss the devices which resemble
well-known microelectronic devices using a simple scaling-down approach. Examples
include such heterostructure devices as the eld-effect and bipolar transistors as well as
injection bipolar lasers. Then, we study nanodevices based on new physical principles,
which can not be realized in microscale devices. Among these are resonant-tunneling
devices, hot-electron transistors, single-electron-transfer devices, monopolar injection
cascade lasers, nanoelectromechanical devices, and quantum-dot cellular automata. We
understand that there are more devices to be reviewed. For example, not enough attention
has been paid to progress in silicon device technology. The ideas and results presented
provide an understanding of near-future developments in nanoelectronics and optoelec-
tronics that are occurring as a result of advances in nanotechnology. This will encourage
students to learn more about nanoelectronics.
The authors have many professional colleagues and friends from numerous countries
who must be acknowledged. Without their contributions and sacrices this work would
not have been completed. Special thanks go to the Division of Engineering Education
and Centers (EEC) in the Directorate for Engineering and to the Division of Materials
Research (DMR) in the Directorate for Mathematical and Physical Sciences (MPS), and
especially to the program manager Mary Poats at the National Science Foundation for
the partial support of this work through the Nanoscale Science and Engineering (NSE)
Nanotechnology in Undergraduate Education (NUE) Program. The help of Dr. Nizami
Vagidov in preparing gures and editing the text is especially appreciated.
V.V.M. acknowledges the support and active encouragement of the faculty of the
Department of Electrical Engineering and the School of Engineering and Applied
Sciences as well as the members of the Center on Hybrid Nanodevices and Systems,
especially Dr. Andrei Sergeev, at the University at Buffalo State University of NewYork.
He also extends his thanks to the undergraduate students who took the course EE 240
Nanotechnology, Engineering, and Science as well as to Matthew Bell, who was the
teaching assistant for the course. Their feedback helped him in the choice and presenta-
tion of the material that is presented in this book. Undergraduate students, Garun Vagidov
and Jonathan Bell, helped with some editorial work. V.V.M. is also grateful to his family
and friends for their strong support and encouragement, as well as for their understanding
and forgiveness of the fact that he did not devote enough time to them while working on
the book, and especially to his mother, grandson Anthony, and granddaughter Christina
whom he missed the most.
Preface xi
V.A.K. thanks his colleagues Professor S. V. Svechnikov, Professor A. E. Belyaev, Pro-
fessor F. T. Vasko, Dr. V. I. Pipa and Dr. B. A. Glavin fromthe Institute of Semiconductor
Physics at Kiev for numerous fruitful discussions of general problems and perspectives of
nanoscience and nanoelectronics. The wide spectrum of research directions and creative
atmosphere prevailing in the Theoretical Physics Department stimulated his involvement
in various nanoscience activities. He acknowledges permanent contacts with graduate
and PhD students, which gave him valuable feedback during the work on the textbook.
V.A.K. is deeply grateful to all his family members for their understanding and permanent
support.
M.A.S. extends his sincere thanks to Deans Prith Banerjee and Larry Kennedy, at the
College of Engineering, University of Illinois at Chicago (UIC) for their active encour-
agement and their longstanding efforts to promote excellence in research at the UIC.
M.A.S. gratefully acknowledges the generous support and enlightened encouragement
of Richard and Loan Hill. Special thanks go to Dr. Dwight Woolard of the US Army
Research Ofce, Drs. Daniel Johnstone, Todd Steiner, and Kitt Reinhardt of the AFOSR,
Dr. Rajinder Khosla, Dr. James W. Mink and Usha Varshney of the National Science
Foundation, Dr. Daniel Herr of the Semiconductor Research Corporation, and Dr. John
Carrano of the Defense Advanced Research Projects Agency for their encouragement
and interest. M.A.S. acknowledges the essential roles that several professional colleagues
and friends played in the events leading to his contributions to this book; these people
include Professor Richard L. Magin, Head of the Bioengineering Department at the
University of Illinois at Chicago (UIC), Professors Robert Trew, Gerald J. Iafrate, M. A.
Littlejohn, K. W. Kim, and R. and M. Kolbas, and Dr. Sergiy Komirenko of the North
Carolina State University, Professors G. Belenky and S. Luryi and Dr. M. Kisin of the
State University of New York at Stony Brook, Professors George I. Haddad, Pallab K.
Bhattacharya, and Jasprit Singh, and Dr. J.-P. Sun of the University of Michigan, Pro-
fessors Karl Hess and J.-P. Leburton at the University of Illinois at Urbana-Champaign,
Professor L. F. Register of the University of Texas at Austin, and Professors H. Craig
Casey and Steven Teitsworth of Duke University. M.A.S. also thanks family members
for their support while this book was being prepared; these include Anthony and Norma
Stroscio, Mitra Dutta, and Elizabeth, Gautam, and Marshall Stroscio.
Notation
Symbols
A amplitude of wave
A) average value of A
a lattice constant
a
0
length of carboncarbon bond in carbon nanotubes
a
1
. a
2
. a
3
basis vectors
a
i
basis vectors of lattice
B magnetic eld
C capacitance
d translation vector
D diffusion coefcient
d
sp
spacer thickness
E energy of a particle
E
F
Fermi level
E
g
bandgap
e elementary charge
F electric eld
F
0
amplitude of electric eld
f frequency
f
SET
frequency of Bloch oscillations
f vector of force
F distribution function
F
F
Fermi distribution function
G conductance
G
0
quantum of conductance
H Hamiltonian operator
H magnetic eld
H direction of the nearest-neighbor hexagon rows
H total energy, Hamiltonian function
h Plancks constant
h wave energy density
h
1D
wave energy density for a one-dimensional medium
h
--
Plancks constant divided by 2
Notation xiii
J current density
I current
I
T
tunnel current
I wave intensity
i quantum-mechanical ux of the particles
k wavevector
k
B
Boltzmanns constant
L inductance
L
T
thermal diffusion length
l
e
mean free path between two elastic collisions
l orbital quantum number
l angular momentum
l
coherence length
L
E
inelastic scattering length
L
x
, L
y
, L
z
dimensions of a sample
M mass of resonator
M magnetic dipole moment
m magnetic quantum number
m
effective mass of electron

m
0
mass of an electron in vacuum
m
HH
heavy-hole mass
m
LH
light-hole mass
m
SH
split-off hole mass
N
s
sheet concentration of donors
N
depl
sheet concentration of ionized acceptors
n principal quantum number
n
s
sheet concentration of electrons
P() Hermite polynomial
P
b
Probability of nding electron under the barrier
q wavevector
Q quality factor
Q amount of deposited material
R radial function
r magnitude of radius vector
r coordinate vector
r
0
Bohrs radius
R radius of quantum dot
R tube radius of carbon nanotube
R reection coefcient
S spin intrinsic angular momentum
S cross-section
S
z
projection of the spin of electron
s distance between tip and surface
s phase velocity of traveling wave
xiv Notation
s spin of electron
t time
t
tr
transit time
T time period
T vector corresponding to tube axis of carbon nanotube
T ambient temperature
T
e
electron temperature
T
d
translation operator
u
k
(r) Bloch periodic function
u displacement of atoms from their equilibrium positions
V potential energy
V volume
V
b
barrier height
V
0
volume of primitive cell
:
d
average(drift) velocity
: velocity
:
h
velocity of hole
W crystalline potential
U
M
potential energy
z
0
characteristic length
Z atomic number
dimensionality factor
spring constant
(x) Diracs delta-function
c relative mismatch of lattice constants of the substrate and epilayer
c dielectric constant of the medium
c
0
permittivity of free space
energy
+ potential
+
b
built-in Schottky voltage, Schottky barrier
+
0
applied voltage
phase
polar angle
gyromagnetic ratio
A
1D
elastic modulus of string
wavelength
j electron mobility
j
ph
partial electron mobility, determined by phonon scattering
j
im
partial electron mobility, determined by impurity scattering
set of quantum numbers
vector of polarization
dimensionless coordinate
O angular frequency of a particle
O ohm
Notation xv
frequency
q
frequency of harmonic oscillator
+(r, t) non-stationary wavefunction
+
(r, t) complex conjugate of wavefunction +(r, t)

(r) stationary wavefunction
three-dimensional density
1D
linear density of string
density of states
conductivity
O theta-function
polar angle
E
mean free time between two inelastic collisions
d
decay time of exural vibrations
e
mean free time between two elastic collisions
(z) wavefunction
electron afnity
Abbreviations
BT bipolar transistor
CMOS complementary MOS, i.e., NMOS and PMOS on the same chip
DPN dip-pen nanolithography
FET eld-effect transistor
JBT homojunction BT
JFET junction FET
HBT heterojuction BT
HEMT high-electron-mobility transistor
HFET heterojunction FET
HOMO highest occupied molecular orbit
LUMO lowest unoccupied molecular orbit
MES metalsemiconductor
MESFET metalsemiconductor FET
MODFET modulation-doped FET
MOS metaloxidesemiconductor
MOSFET metaloxidesemiconductor FET
QUIT quantum interference transistor
RTD resonant-tunneling diode
SIMOX separation by implantation of oxygen
SMS semiconductormetalsemiconductor
VMT velocity-modulation transistor
1 Toward the nanoscale
This book provides the foundations and the main ideas emerging from research that
underlies the applied eld called nanoelectronics. Nanoelectronics promises to improve,
amplify, and partially substitute for the well-known eld of microelectronics. The prex
micro denotes one millionth and, as applied to electronics, it is used to indicate that
the characteristic sizes of the smallest features of a conventional electronic device have
length scales of approximately a micrometer. The prex nano denotes one billionth.
Thus, in nanoelectronics the dimensions of the devices should be as many as a thousand
times smaller than those of microelectronics.
Such a revolutionary advance toward miniaturization of electronics is based on the
recently developed ability to measure, manipulate, and organize matter on the nanoscale
1 to 100 nanometers, i.e., 1 to 100 billionths of a meter. At the nanoscale, physics,
chemistry, biology, materials science, and engineering converge toward the same princi-
ples and tools, and form new and broad branches of science and technology that can be
called nanoscience and nanotechnology.
Advancing to the nanoscale is not just a step toward miniaturization, but requires the
introduction and consideration of many additional phenomena. At the nanoscale, most
phenomena and processes are dominated by quantum physics and they exhibit unique
behavior. Fundamental scientic advances are expected to be achieved as knowledge in
nanoscience increases. In turn, this will lead to dramatic changes in the ways materials,
devices, and systems are understood and created. Innovative nanoscale properties and
functions will be achieved through the control of matter at the level of its building blocks:
atom-by-atom, molecule-by-molecule, and nanostructure-by-nanostructure. The molec-
ular building blocks of life proteins, nucleic acids, carbohydrates are examples of
materials that possess impressive properties determined by their size, geometrical fold-
ing, and patterns at the nanoscale. Nanotechnology includes the integration of manmade
nanostructures into larger material components and systems. Importantly, within these
larger-scale systems, the active elements of the system will remain at the nanoscale.
The driving forces underlying developments at the nanoscale have at least two major
complementary components scientic opportunities and technological motivations.
Scientic opportunities
The progress in physics, chemistry, and biology at the nanoscale represents a natural step
in advancing knowledge and understanding Nature. Scientic perspectives on this route
are conditioned rst of all by new quantum phenomena in atomic- and molecular-scale
structures and by the interaction of large numbers of these small objects. Indeed, the
fundamental laws of physics in the nanoworld differ from those that apply to familiar
macroscopic phenomena. Instead of classical physics, that works so well for macroscopic
phenomena, the motion of particles and systems in the nanoworld is determined by the
so-called wave mechanics or quantummechanics. Abasic principle of nanophysics is the
fundamental concept that all matter, including electrons, nuclei, atoms, electromagnetic
elds, etc., behaves as both waves and particles. This waveparticle duality of all matter
is strikingly apparent at the nanoscale. For dealing with a large number of particles or
systems, the statistical laws are important. Statistical physics on the nanoscale is also
fundamentally different fromthat on the macroscale. In general, phenomena that involve
very large numbers of small interacting particles or systems follow different rules from
those involving only a few of them. Cooperative behavior of many-object systems is
revealed clearly at the nanoscale. Besides the phenomena just discussed, there are other
classes of phenomena that are important for science at the nanoscale.
It is appropriate here to refer to the famous 1959 lecture of the Nobel Prize laureate
Professor Richard Feynman with the title There is plenty of room at the bottom, where
he discussed the problem of manipulation and controlling things on a small scale.
Feynman did not just indicate that there is room at the bottom, in terms of decreasing
the size of things, but also emphasized that there is plenty of room. In his lecture,
Feynman justied the inevitable development of concepts and technologies underly-
ing the nanoworld and presented his vision of exciting new discoveries and scientic
perspectives at the nanoscale.
Technological motivations
Achievements in nanoscience and nanotechnology will have tremendous multidisci-
plinary impact. The benets brought by novel nanotechnologies are expected for many
important practical elds of endeavor. These include materials and manufacturing, elec-
tronics, computers, telecommunication and information technologies, medicine and
health, the environment and energy storage, chemical and biological technologies, and
agriculture. Having stated the purpose of this text, we consider now more detailed moti-
vations for the development of electronics at the nanoscale.
In general, progress in electronics is stimulated, in part, by the enormous demands for
information and communication technologies as well as by the development of numer-
ous special applications. The continuous demands for steady growth in memory and
computational capabilities and for increasing processing and transmission speeds of
signals appear to be insatiable. These determine the dominant trends of contemporary
microelectronics and optoelectronics. One of the main trends of the progress in electron-
ics was formulated by Intel co-founder Dr. Gordon Moore as the following empirical
observation: the complexity of integrated circuits, with respect to minimum component
cost, doubles every 24 months. This statement formulated forty years ago is known as
Moores law and provides an estimate of the rate of progress in the electronics industry.
Specically, Moores law predicts that the number of the basic devices transistors on
Toward the nanoscale 3
a microchip doubles every one to two years. This is possible only if progressive scaling
down of all electronic components is realized.
Electronics exploits the electrical properties of solid-state materials. A simple and
intuitive classication of solids makes a distinction between dielectrics and metals, i.e.,
dielectrics are non-conducting materials whereas metals are good conducting materials.
Semiconductors occupy the place in between these two classes: semiconductor materials
are conducting and optically active materials with electrical and optical properties vary-
ing over a wide range. Semiconductors are the basic materials for microelectronics and
remain the principal candidates for use in nanoelectronic structures because they exhibit
great exibility in terms of allowing the control of the electronic and optical properties
and functions of nanoelectronic devices. Accordingly, to a large extent, we will analyze
the trends of electronics in the context of semiconductor technology.
It is instructive to illustrate these trends and achievements through the example of
Si-based electronics. Indeed, contemporary microelectronics is based almost entirely
on silicon technology, because of the unique properties of silicon. This semiconductor
material has high mechanical stability as well as good electrical isolation and thermal
conductivity. Furthermore, the thin and stable high-resistance oxide, SiO
2
, is capable
of withstanding high voltages and can be patterned and processed by numerous meth-
ods. Silicon technology also enjoys the advantage of a mature growth technology that
makes it possible to grow Si substrates (wafers) of larger areas than for other semi-
conductor materials. The high level of device integration realizable with Si-based elec-
tronics technology may be illustrated by the important integrated circuit element of any
computer, controller, etc. the dynamic random access memory (DRAM). The main
elements of DRAM based on complementary metaloxidesemiconductor technology
(Si-CMOS) are metaloxidesemiconductor eld-effect transistors (MOSFETs). For Si
MOSFETs, channels for ow of electric current are created in the Si substrate between
the source and drain contacts, and the currents are controlled by electrodes metal gates
which are isolated electrically by very thin SiO
2
layers, which have become thinner than
10 nm.
Figure 1.1 illustrates the evolution of the DRAM size and transistor gate size as
functions of time. Besides transistors and capacitors, the chip contains metallic line
connections: local, intermediate, and global wiring. Figure 1.1 illustrates the steady
scaling down of all characteristic sizes and increasing levels of integration. For example,
the 64-Mbit DRAM chip contains approximately 10
8
transistors per cm
2
, each with
feature sizes of the order of 0.3 m. The transistors in this DRAM as well as those of
the more highly integrated 256-Mbit chip operate as conventional devices and obey the
laws of classical physics. The next generation of devices is entering the nanoscale regime
where quantummechanics is important; indeed, as we will discuss in this book, quantum
mechanics becomes dominant on the scale of approximately one to ten nanometers for
devices that operate at room temperature. According to Fig. 1.1, todays technology has
already reached the nanoscale and newer device concepts should be implemented before
2010.
One of the factors driving the huge production and wide use of microelectronic sys-
tems is the relatively low cost of their fabrication. Moreover, despite their increasing
Table 1.1 A roadmap for Si-based microelectronics (predictions of the Semiconductor
Industry Association)
1995 1998 2001 2004 2007 2010
Memories, DRAM
Bits per chip 64 M 256 M 1 G 4 G 16 G 64 G
Cost per bit (milli-cent) 0.017 0.007 0.003 0.001 0.0005 0.0002
Cost per chip (US$) 11 18 30 40 80 130
Logic, microprocessors
Transistors per cm
2
4 M 7 M 13 M 25 M 50 M 90 M
Cost per transistor (milli-cent) 1 0.5 0.2 0.1 0.05 0.02
Power supply (V) 3.3 2.5 1.8 1.5 1.2 0.9
Parameters
Minimum feature size (m) 0.35 0.25 0.18 0.13 0.10 0.07
Wafer size (in.) 8 8 12 12 16 16
Electrical defect density per m
2
240 160 140 120 100 25
The data are from U. K onig, Physica Scripta, T68, 90, 1996.
Figure 1.1 Technology nodes and minimum feature sizes from application ITRS Roadmap:
MPU, Micro Processing Unit; ASIC, Application-Specic Integrated Circuit. Used with
permission, from W. Klingenstein (2002). Technology Roadmap for Semiconductors.
http://broadband02.ici.ro/program/klingenstein 3d.pdf, page 15.
C _
InneonTechnologies AG,
2002.
complexity, microelectronic systems continue to be produced at lower costs. In Table 1.1
the costs per bit and costs per chip as well as the associated performance levels are given
as functions of the integration level. One can see that every three years the number of
bits per chip has increased by a factor of four (even faster than predicted by Moores law)
and the cost per bit has decreased by a factor of two or more.
In the same table, the integration levels of logic circuits and microprocessors are
forecast. We see that, for this case, device integration is also large but will increase
slightly slower than for DRAMs. The cost of the principal elements of logic circuits
transistors is signicantly greater, but it also tends to decrease. The forecast for the
necessary power supply presented in the table portrays a slow, but persistent, decrease.
Thus, one can expect favorable trends for the power consumption of microelectronic
systems.
The bottomof the table presents the necessary technological parameters for ultra-high
integration: minimum feature sizes, diameters of wafers, and electrical defect densities.
The large wafer size allows a greater number of devices to be fabricated on a chip. The
density of electrical defects (i.e., crystal imperfections which affect electrical properties)
is characteristic of the quality of the wafers. Table 1.1 forecasts that wafer diameters will
be continuously increased, while the defect density decreases by a factor of six per
decade; currently, they must be limited to several tens per m
2
.
After this overview of the dominant driving forces in nanoscale development, we will
mention briey other general issues important for this eld. These include improving
materials, fabrication and measurement techniques on the nanoscale, and novelties in
the operation principles of nanodevices.
Improving materials on the nanoscale
In the processes of achieving minimum device sizes and ultra-high levels of integration
it is necessary to identify the limiting and critical parameters for improved performance.
In reality, these parameters depend on the integrated elements of each individual material
system. For example, for transistors two parameters of the host material are of special
importance: the ultimate electron velocity and the limiting electric eld which does not
induce electric breakdown. Further improvements in the parameters can be achieved
through materials engineering.
Silicon plays the central role in electronics. However, semiconductors other than sil-
icon can be used. In particular, compound semiconductors constitute a general class
of semiconductors that has been used increasingly in recent decades. As examples of
forming compound semiconductors, every particular element in column III of the peri-
odic table of elements may be combined with every element in column V to form a
so-called IIIV compound, which is semiconducting. Then, two or more discrete com-
pounds may be used to form alloys. A common example is aluminumgallium arsenide,
Al
x
Ga
1x
As, where x is the fraction of column III sites in the crystal occupied by Al
atoms, and the fraction 1 x is occupied by Ga atoms. As a result, it becomes possible
not only to make discrete compounds, but also to realize a continuous range of materials
for tailoring necessary electronic properties. As for Si technology, the growth of silicon
germanium (Si
x
Ge
1x
) alloys facilitates the control of the properties of materials over a
considerable range of the electrical parameters. These techniques are exploited widely
in microelectronics.
Further revolutionary modication and engineering of materials can be accomplished
by using heterostructures with nanoscale features. Heterostructures are structures with
two or more abrupt interfaces at the boundaries between the different semiconductor
materials. With modern material-growth techniques, it is possible to grow structures
with transition regions between adjacent materials that have thicknesses of only one
or two atomic monolayers. This allows one to fabricate multilayered semiconductor
structures with nanoscale thicknesses.
The simplest multilayered structure has a single heterojunction, i.e., a single-
heterojunction structure is made of two different materials. At the interface of such
a heterojunction, the electronic properties are changed to improve selected physical
characteristics. In particular, electrons can be conned in a thin layer near the interface.
In fact, the layers with conned electrons can be made so thin that wave-like behavior
that is, quantum-mechanical behavior of the electrons becomes apparent. The same
phenomena occur for diverse multilayered nanoscale structures that can be grown with
high quality.
By using nanostructures, it is possible to modify the electronic properties of a great
variety of a nanoscale devices. Indeed, we live in a three-dimensional world, where a
particle can, in principle, move in all three directions. Quantum effects on the nanoscale
determine the properties of electrons in nanostructures: the nanostructures can be made
in such a way that the electron motion becomes two-dimensional, one-dimensional, or
even zero-dimensional. These nanostructures are known as low-dimensional quantum
heterostructures and are called quantumwells, quantumwires, and quantumdots, for the
cases where the electrons are conned in one, two, and three dimensions, respectively.
Such an impressive example of manipulation of the properties of the current-carrying
particles clearly illustrates fundamentally new possibilities for electronics that become
viable at the nanoscale.
Fabrication techniques on the nanoscale
The progress in heterostructure technology has been made possible largely as a result of
new advances in fabrication techniques. In Table 1.2, we provide a very brief summary
of some important steps now used in the growth, characterization, and processing of
heterostructures. In the period of the 1970s and 1980s, molecular-beam epitaxy was
invented, developed, and employed to fabricate high-quality and ultra-thin layers and
superlattices. Qualitative electron-beamand X-ray microscope technologies were used to
characterize the perfectness of structures, includinginterface disorder. Duringthis period,
lithographic and etching methods suitable for microscale and nanoscale devices were
proposed and realized. In the 1980s and later, new epitaxial techniques were developed;
these included metalorganic vapor-phase epitaxy and metalorganic molecular-beam
epitaxy, among others. These innovations made possible the fabrication of layers with
atomic-level accuracy. Lithography and etching methods were improved to the point
that they can be used for nanoscale structuring. Desirable spatial-modulation doping by
impurities has become possible, including -doping that is, doping of one or a few
atomic monolayers.
These approaches to the production of nanostructures and nanoelectronic devices
represent evolutionary improvements in the growth and processing methods applied
previously in microelectronics. Advances in nanotechnology allow, in principle, the uti-
lization of methods and concepts from other areas of science and engineering. Synthetic
Table 1.2 Advances in growth, characterization, and processing of quantum
heterostructures
1970s1980s Growth and fabrication methods
Molecular-beam epitaxy
Ultra-thin-layer fabrication
Superlattice fabrication
Characterization methods
Lithographic microstructuring
Qualitative electron-beam and X-ray microscopies
1990s2000s Growth and fabrication methods
Metalorganic vapor-phase epitaxy
Metalorganic molecular-beam epitaxy
Fabrication to atomic-layer accuracy
-Doping
Controlled strained layers
Fabrication methods based on chemistry and biology
Assembling inorganic nanoblocks with biomolecules
Characterization methods
Lithography and etching for nanostructuring
Dip-pen nanolithography
Quantitative electron-beam and X-ray microscopies
Scanning tunneling microscopy (STM)
Atomic force microscopy (AFM)
Picosecond and femtosecond spectroscopy
Terahertz time-domain spectroscopy
chemistry and even biology have much to offer for emerging nanotechnologies. Some
fundamental concepts coming from these elds can successfully be exploited for the
synthesis of nanomaterials and nanodevices. These include chemical and biological
methods of growth of nanoscale objects such as carbon nanotubes and biomolecules
surface nanopatterning, and preparing nanostructured materials with predened, syn-
thetically programmable properties from common inorganic building blocks with the
help of DNA interconnect molecules, etc.
Improvement in characterization methods for the nanoscale
Progress in the renement of fabrication techniques for making nanostructures depends
on the great improvements made in characterization methods. Some of these methods are
indicated in Table 1.2. In particular, composition and dopant distribution, lattice strain,
and other parameters within nanostructures must be known with atomic-scale precision.
Currently, the manipulation of a single atom (ion) in a solid is possible. New tools
scanning tunneling microscopy and atomic-force microscopy which portend numerous
applications in high-precision fabrication have emerged. Picosecond and femtosecond
spectroscopy have progressed substantially and they have been applied to characterize
the electronic and lattice properties of heterostructures. Finally, terahertz time-domain
spectroscopy was developed, which makes it possible to measure electric signals with
time resolution at the level of 10
12
seconds.
New principles of device operation at the nanoscale
Fundamental questions arise when conventional principles of device operation fail as
a result of entering the nanoscale domain. One of the effects is almost collisionless
motion frequently called ballistic motion of electrons ying through short devices.
As mentioned previously, the nanoscale domain is the realm of quantum physics.
Indeed, scaling down of devices and their integration above the level corresponding to
250 Mbits on a single chip makes it necessary to take into account new regimes and even
to modify the principles underlying device operation. Further device downscaling and
higher integration densities for information capacities exceeding 1 Gbit per chip imply
the need to investigate using quantum regimes of operation in future years. Quantum-
mechanical effects are not only important for operation of future integration circuits, but
also are already used for generation of ultra-high-frequency electromagnetic emission.
A relevant example is that of resonant-tunneling phenomena in nanoscale multilayered
structures, which creates a foundation for microwave emission up to 1000 GHz.
A number of such device-related quantum effects has been discovered. New physics
and new quantum effects always pass ahead of new devices exploiting these effects that
have made a substantial impact on device technology. We mention here just a few quan-
tum effects: 1970, the proposal of multilayered structures; 1974, the resonant-tunneling
effect; 1978, the modulation doping effect; 1980, conduction of polymers; 1985, the
discovery of the buckyball, C
60
; 1993, the discovery of single-walled carbon nanotubes;
and 1996, nanoelectromechanical systems (NEMSs). Some of these effects will be ana-
lyzed in this book. Here is a short list of some of the novel quantum devices: 1979, the
injection quantum-well laser; 1983; the Microwave DBRTD Oscillator; 1984, the hot-
electron transistor; 1998, the quantum-wire carbon nanotube eld-effect transistor; 1998,
polymer-based transistors and light-emitting devices; 2001, sensors based on NEMSs;
2001, sub-terahertz IIIV compound nanoscale eld-effect transistor; and 2006, sub-
terahertz InP and SiGe bipolar transistors. There is a delay time between the discovery
of the effect and the device concept, but the delay time is decreasing steadily. The fol-
lowing examples show this tendency. The effect of resonant tunneling was discovered
in 1974; the device the microwave double-barrier resonant-tunneling diode (DBRTD)
oscillator was realized in 1983. The rst quantum wires with one-dimensional electron
motion were studied in 1986; their rst application in lasers occurred in 1995. In both
cases the delay time was 9 years. The discovery of single-walled carbon nanotubes
was made in 1993 and in only 5 years the carbon nanotube transistor was fabricated. The
same is valid for the development of nanoelectromechanical systems and their applica-
tions for a number of sensors, etc. Thus, for contemporary electronics there is an evident
acceleration of the implementation of fundamental physical effects.
Besides quantumeffects, reducing device dimensions results in a decrease in the num-
ber of electrons participating in the transfer of an electric signal. As a result, nanoscale
devices may operate on the basis of single-electron transfer. Various novel single-electron
devices have been proposed and demonstrated. By reducing the sizes of quantum dots
to 100

A or less, it is possible to operate with single electrons at temperatures near or
close to room temperature.
The great technological advances brought about in mainstream microelectronics and
nanoelectronics can be used for the fabrication of other classes of nanodevices. One
such approach is based on quantum dots arranged in locally interconnected cellular-
automata-like arrays. The fundamental idea of operation of cellular automata is to encode
information using the charge conguration of a set of quantum dots. Importantly, in the
quantum-dot cellular-automata approach, the information is contained in the arrange-
ment of charges of the dots, rather than in the ow of the charges, i.e., electric current.
It can be said that the devices interact by direct Coulomb coupling rather than via the
current through the wires.
Another approach employs both electrical and mechanical properties of nanostruc-
tures. The new generation of devices and systems based on this approach is commonly
referred to as nanoelectromechanical systems (NEMSs). Indeed, on the nanoscale a
strong enhancement of coupling between electronic and mechanical degrees of freedom
occurs. This electromechanical concept may be used for the development of a new class
of devices that includes nanomachines, novel sensors, and a variety of other new devices
functioning on the nanoscale. Thus, NEMSs may supplement the traditional electronics
that works solely with electrical signals.
Nanotechnology for optoelectronics
Thus far, we have reviewed nanoscience and nanotechnology as applied to electronic
devices, i.e., devices in which electrical properties are exploited and which operate with
electrical input and output signals. Another class of devices is comprised of optoelec-
tronic devices, which are based on both electrical and optical properties of materials
and operate with both optical and electrical signals. An important and growing trend is
that optoelectronics complements microelectronics in many applications and systems.
First of all, optoelectronics provides means to make electronic systems compatible with
lightwave communication technologies. Furthermore, optoelectronics can be used to
accomplish the tasks of acquisition, storage, and processing of information. Advances
in optoelectronics make signicant contributions to the transmission of information via
optical bers (including communication between processing machines as well as within
them), to the high-capacity mass storage of information on laser disks, and to a number
of other specic applications. Clearly, optoelectronic devices have a huge number of
diverse applications.
The principal components of optoelectronic systems are light sources, sensitive optical
detectors, and properly designed light waveguides, for example, optical bers. These
devices and passive optical elements are fabricated with optically active semiconductor
materials. Semiconductor nanostructures and, in particular, quantum heterostructures
provide new means to enhance many optical and electro-optical effects. For example,
both of the most widely used light sources light-emitting diodes and laser diodes
may be improved greatly when nanostructures such as quantum wells, quantum wires,
and quantum dots are exploited as active optical elements.
As for the previously studied case of microelectronics, the trends in optoelectronics
involve scaling down the sizes of these devices as well as achieving high levels of
integration in systems such as arrays of light diodes, laser arrays, and integrated systems
with other electronic elements on the same chip. Optoelectronics benets substantially
through the use of nanotechnology and becomes competitive with its microelectronic
counterpart.
In conclusion, the current and projected trends in electronics lead to the use of nanos-
tructures and to the reliance on novel quantum effects as an avenue for realizing further
progress. These recent and diverse trends in semiconductor and device technologies as
well as in novel device concepts are driving the establishment of a new subdiscipline
of electronics based on nanostructures, i.e., nanoelectronics. This subdiscipline and its
foundations are studied in this book.
More general information on nanoscience, nanotechnology, and nanostructures, and
their potential, may be found in the following reviews:
R. Feynman, Theres plenty of room at the bottom, American Physical Society
Meeting, Pasadena, CA, 29 December 1959; originally published in Caltechs
Engineering and Science Magazine, February 1960; reprinted as R. P. Feynman,
Innitesimal machinery, Microelectromechanical Systems, 2, 1 (1993); (see, for
example, www.zyvex.com/nanotech/feynman.html).
National Nanotechnology Initiative: The Initiative and Its Implementation Plan,
National Science and Technology Council, Committee on Technology, Washington
DC, 2000 (see, for example, www.nano.gov).
H. Kroemer, Quasielectric elds and band offsets: teaching electrons new tricks,
Rev. Mod. Phys., 73, 783 (2001).
The International Technology Roadmap for Semiconductors (Semiconductor Industry
Association, San Jose, CA, 2002 update).
2 Particles and waves
2.1 Introduction
The evolution of microelectronics toward reduced device sizes has proceeded to a degree
that renders conventional models, approaches, and theories inapplicable. Indeed, for
objects with sizes of 100 nanometers or less it is frequently the case that the length scales
associated with fundamental physical processes are comparable to the geometrical size
of the device; also, fundamental time scales associated with physical processes are of
the order of the time parameters for nanodevice operation. Therefore, on the nanoscale
the theories and models underlying modern nanoelectronics become more complicated,
and rely more and more on basic science.
Generally, in the nanoworld the fundamental laws of physics that govern particles and
material elds differ from those that apply to familiar macroscopic phenomena such as
the motion of a baseball or a train. Instead of classical mechanics, that works so well
for macroscopic phenomena, the motion of particles in the nanoworld is determined
by the so-called wave mechanics or quantum mechanics. An underlying principle of
central importance for nanophysics is the fundamental concept that all matter, including
electrons, nuclei, and electromagnetic elds, behaves as both waves and particles, that
is, waveparticle duality is a basic characteristic of all matter.
At rst glance, wave properties and particle properties for the same physical object
are hardly compatible. To understand waveparticle duality, we will briey review, in
the following two subsections, the basic properties of particles and waves known from
classical physics.
2.2 Classical particles
A particle can be characterized by the momentum vector p and the kinetic energy E that
depends on the momentum. Here and throughout this book we will use the bold fonts for
vectors, i.e., p is the vector and p [p[ is its absolute value. The change of momentum
with time is dened by Newtons second law:
dp
dt
= f. (2.1)
where t is time and f is the vector of an external force. From Eq. (2.1) it follows that, if
the force is absent, then dp,dt = 0, i.e., p = constant. This is the so-called momentum
conservation law valid for a mechanical system in the absence of external forces. For
simplicity in classical mechanics, we assume that any particle we consider has a very
small size in comparison with the space where the particle is located. We refer to such a
particle as a point particle.
The coordinate vector, r, of a point particle and the particle velocity, v, are related by
the well-understood relation
dr
dt
= v. (2.2)
To obtain the relationship among the velocity, v, the momentum, p, and the energy of a
particle, E, one has to calculate the power associated with the force f (the work of the
force f on the particle per unit time). So, we have to multiply the left- and right-hand
sides of Eq. (2.1) by v, resulting in v dp,dt = fv. (Remember that the scalar product of
two vectors a and b is dened as ab = a
x
b
x
a
y
b
y
a
z
b
z
.) The right-hand side of the
transformed version of Eq. (2.1), fv, is equal to the rate of the energy change dE,dt
(dE,dt = fv), and we obtain the relation
dE
dt
= v
dp
dt
. (2.3)
Using the chain rule of function differentiation (d f (x),dt ) = (d f ,dx)(dx,dt ), we deter-
mine how the velocity of the particle is related to the momentum and energy:
v =
dE
dp
. (2.4)
Here, the derivative with respect to the vector p also gives the vector v with components
:
x
=
dE
dp
x
. :
y
=
dE
dp
y
. and :
z
=
dE
dp
z
.
Let us consider an important case of a particle moving in a potential eld. The force
is dened as the derivative of a potential V(r) with respect to the particle coordinate:
f = dV,dr. Note that, for the vector operator d,dr {d,dx. d,dy. d,dz], one often
uses another notation: d,dr , so that dV,dr is the so-called gradient of function
V(r), V(r). On multiplying the left- and right-hand sides of Eq. (2.1) by v, using the
denition of Eq. (2.2) and the chain rule, we nd
v
dp
dt

dV
dr
dr
dt
=
d
dt
(E V(r)) = 0.
The value of the kinetic energy plus the potential energy,
H E V(r). (2.5)
represents the total energy of the particle, H. The above calculations tell us that the
total energy of a particle in a potential eld does not change during its motion. So, we
have demonstrated the law of energy conservation, dH,dt = 0. When H is considered
as a function of two variables p and r, it is called the Hamiltonian function, or simply
the Hamiltonian. Remarkably, the partial derivatives of H give us both fundamental
equations (2.1) and (2.4): dp,dt = f = H,r and dr,dt = v = H,p. As we shall
see, the Hamiltonian of classical physics also plays an important role in the formulation
of quantum mechanics.
A point particle moving in free space may be characterized by a mass m and by the
kinetic energy:
E =
p
2
2m
. (2.6)
The latter dependence is frequently referred to as the energy dispersion. Here E is an
isotropic function of p. From the denition of kinetic energy given by Eq. (2.6) we nd
that
v =
p
m
. (2.7)
So, the velocity and the momentum are collinear vectors. Then, Newtons second law,
Eq. (2.1), can be rewritten in its usual form:
m
d
2
r
dt
2
= f. (2.8)
Now, Eq. (2.5) takes the form that we will use often in this book:
H
p
2
2m
V(r). (2.9)
One of the important results following from classical mechanics is that, if we know the
particle position r
0
and its momentum p
0
(or velocity v
0
) at an initial moment t
0
, from
Eqs. (2.1)(2.4), we can nd the position and the momentum (velocity) of the particle
at any given moment of time t for any given f or V(r).
Equations (2.1)(2.8) are the equations of classical mechanics. All of the variables,
such as r, p, E, and v, are continuous variables. Importantly, [p[ can have any value,
including zero, i.e., p = 0 and E = 0 are allowed.
For a particle, say an electron, moving inside of a crystal (a metal, a dielectric, a
semiconductor, etc.), the interaction of this particle with the crystal generally makes the
relationship between E and p the dispersion relation more complicated. In particular,
E may be an anisotropic function, and the velocity and momentum may be noncollinear
vectors. Examples of such energy dependences will be given in the problems for this
chapter.
2.3 Classical waves
We are all familiar with a lot of examples of waves and wave processes. These include
sound waves in air, sea waves, and elastic waves in solids, electromagnetic waves, and
gravitational waves. Generally, in classical physics wave motion arises in extended con-
tinuous media with an interaction between the nearest elements of the medium. Such
an interaction gives rise to the transfer of a distortion (an excitation) from one ele-
ment to another and to a propagation of this distortion through the medium. Despite the
z
a
z
n
= na z
n1
= (n 1)a z
n+1
= (n + 1)a
n+2
= (n + 2)a z
n2
= (n 2)a
u
n
n n + 1 n + 2
n 2 n 1
M M M
u
n+1
u
n1
M M
z
Figure 2.1 A linear chain of identical atoms of mass M: u
n
are displacements of atoms from their
equilibrium positions and the restoring force acting on the nth atom is f
n
= (u
n
u
n1
)
(u
n
u
n1
). Note: displacements u
n
are not shown to scale.
differences in the particular nature of waves, wave motion has much in common for
different media. We introduce wave properties by analyzing the following simple model.
We will construct a model of a one-dimensional medium, elements of which are
represented by atoms connected by massless springs. Vibrations in such a linear atomic
chain are governed by the laws of classical mechanics. The chain is supposed to be
innitely long. Let the equilibrium distance between atoms be a. Thus, the equilibrium
position of the nth atom is z
n
= na, and the displacement of this atom from its position
is denoted by u
n
. Figure 2.1 depicts such a linear chain of identical atoms of mass M.
The springs represent interatomic forces, i.e., interaction between nearest elements of
the medium. If the displacements of atoms from their equilibrium positions are not too
large, the restoring forces in the chain obey Hookes law,
f = u. (2.10)
where u is a change of the spring length, is the spring constant, and f is the force
exerted by the spring. Now, we can apply Eq. (2.10) for the total force, f
n
, acting on the
nth atom coupled with its two nearest neighbors by two springs as
f
n
= (u
n
u
n1
) (u
n
u
n1
). (2.11)
Hence, the Newton equation of motion, Eq. (2.8), for the nth atom is
M
d
2
u
n
dt
2
= (2u
n
u
n1
u
n1
). (2.12)
This set of linear differential equations, Eq. (2.12), in principle, describes wave-like
processes. However, we will make a further simplication and modify this discrete set
of equations to obtain one equation describing a continuous medium. Such a continuous
medium with elastic forces between its elements is, obviously, a string. To make this
transformation to the continuous case, we shall consider the discrete coordinate to be
continuous, z
n
z, and replace the nite difference in Eq. (2.12) with a derivative:
u
n
u
n1
a

u
z
and
(2u
n
u
n1
u
n1
)
a
2

2
u
z
2
. (2.13)
Then, we obtain the equation describing the displacement, u, of a string:
1D
2
u
t
2
A
1D
2
u
z
2
= 0. (2.14)
where
1D
= M,a is the linear density of the string, and A
1D
= a is the so-called elastic
modulus of the string. The innite set of ordinary differential equations, Eq. (2.12), is
replaced by a single partial differential equation. Since we started our derivation for a
mechanical system, for which the energy (per atom) can be dened, it is instructive to
nd a similar characteristic for the continuous mediumdescribed by Eq. (2.14). Assume,
for a moment, that the spring under consideration is of nite large length, L. Then, let
us multiply Eq. (2.14) by u,t and integrate it over the length L:
_
L
0
dz
_
1D
2
u
t
2
u
t
A
1D
2
u
z
2
u
t
_
= 0.
By integrating the second term by parts by using the standard relationship
_
nd: =
n:
_
: dn, we nd the following identity for a unit length of the string:
t
1
L
_
L
0
dz
_
1D
2
_
u
t
_
2
A
1D
2
_
u
z
_
2
_
A
1D
L
__
u
z
u
t
_
z=L
_
u
z
u
t
_
z=0
_
= 0.
To draw further conclusions, we dene an average of a quantity, A, over a piece of the
string, Z, as A = (1,Z )
_
(Z)
A dz. For a long string, the average should not depend on
the length, Z, of the piece of string. Now, as L , the latter identity will be satised
if the value
h
1D
=

1D
2
_
u
t
_
2
A
1D
2
_
u
z
_
2
(2.15)
is independent of time, i.e., h
1D
,t = 0, and the value h
1D
is conserved. In fact, the
rst term on the right-hand side of Eq. (2.15) is, obviously, the density of kinetic energy
over the length of the string, while the second term is the elastic (potential) energy. Thus,
h
1D
has the meaning of the energy density for our one-dimensional continuous medium.
As expected, the energy density is conserved, if external forces are absent.
Our one-dimensional analysis, which assumes that the atoms can move only along
a single direction z, can be generalized to a three-dimensional elastic medium; see
Problem 4 of this chapter. Now, the displacement becomes a three-dimensional vector,
u, and, instead of Eq. (2.14), we obtain
2
u
t
2
A
_
2
u
x
2

2
u
y
2

2
u
z
2
_
= 0 (2.16)
with r = {x. y. z] being the vector coordinate and being the three-dimensional den-
sity, i.e., the mass of a unit volume of the medium; A is the elastic modulus of the
medium. Since it was introduced as a single elastic modulus, independent of direction,
Eq. (2.16) is valid for an isotropic medium. The energy density of a three-dimensional
elastic medium is
h =

2
_
u
t
_
2
A
2
_
u
r
_
2
. (2.17)
Here, the bar denotes an average over a small volume LV(A = (1,LV)
_
(LV)
A dV),
and
u
r

u
x
x

u
y
y

u
z
z
= div u.
Finally, we can rewrite Eq. (2.16) in the standard form of the wave equation:
2
u
t
2
s
2
_
2
u
x
2

2
u
y
2

2
u
z
2
_
= 0. (2.18)
where we introduce a new parameter s, the meaning of which will be claried below. For
our model
s =
_
A,. (2.19)
Although we derived Eq. (2.18) for a particular model of the elastic medium, the equation
can be applied to describe a wide class of physical vector elds as exemplied by
u(x. y. z) the displacement eld associated with a wave in an elastic medium. If the
characteristic of a wave eld is a scalar value, say n, in Eq. (2.18), we should simply
substitute u n to describe the case of a scalar physical eld.
Now we will analyze solutions to Eq. (2.18) for some cases in which the solutions
are particularly simple. In many cases, such solutions are associated with wave-like
processes. We may look for solutions of the form
u(t. r) = Acos(qr t ) Bsin(qr t ). (2.20)
where A and B are arbitrary vectors, and q are unknown parameters; is known as
the angular frequency of the wave and q is called the wavevector. By substituting such a
form for u(t. r) into Eq. (2.18), we easily nd that Eq. (2.20) is a solution of Eq. (2.18),
if
2
= sq
2
. The relationship between and q = [q[ is called the dispersion relation:
= s[q[. (2.21)
Importantly, there is no limitation to the wavevector q: a solution can be found for any
q. This is valid only for innitely extended media, for which the wavevector can be a
continuous vector.
Because the two terms in Eq. (2.20) behave similarly, we can discuss basic properties
of these solutions based on the example of sinusoidal waves:
u(t. r) = Bsin(qr t ). (2.22)
The argument of the sine function is the phase of the wave, = qr t , and B is the
amplitude of the wave. Let the coordinate r be given, then we obtain a function that
oscillates in time with an angular velocity . The frequency denes the rate of change
of the phase with time t (radians per unit time). The period of time associated with a
single oscillation is
T = 2,.
Accordingly, T is known as the period. If the time t is xed, Eq. (2.22) represents a
function that oscillates as the coordinate changes. These oscillations are characterized
by the wavevector, q (or wavenumber, q). The wavevector denes the rate at which the

u
z
B
B
0
(a) t = 0
u
z
(b) t = T/4
0
4
l 4
l
l l
l= 2p /q
B
B
4
l
4
l
l l
Figure 2.2 A propagating wave u = B sin(qz t ). At time t = 0 the wave u = B sin(qz t )
at the point z = ,4 has displacement u = B sin(,2) = B. At time t = T,4 we have
t = (2,T)(T,4) = ,2; the same displacement will occur at the point z = 0: u =
B sin(qz t ) = B sin(,2) = B.
phase changes with the coordinate (radians per unit length). One can introduce the spatial
period to represent the distance for which u undergoes an oscillation of one cycle. It is
called the wavelength,
= 2,q. (2.23)
The relationship between the time period and the spatial period is dened by Eqs. (2.21)
and (2.23):
=
2
s = Ts. (2.24)
The waveforms of Eqs. (2.20) and (2.22) are often referred to as traveling plane
waves. Indeed, their phase changes in a single direction along the wavevector q and
surfaces of constant phase are planes perpendicular to q. Thus, in the case of a plane
wave in a three-dimensional medium, the wave parameters do not depend on the two
coordinates perpendicular to q. A traveling wave is illustrated in Fig. 2.2, where this
wave is shown at two different moments of time. Now, we can clarify the meaning of
the parameter s. Let the wavevector be directed along the z direction. Then, the wave
phase is = qz t = q[z (,q)t ]. So we can conclude that a given magnitude of
the phase moves with the velocity ,q = s, i.e., the parameter s in the wave equation
(2.18) is the phase velocity of the traveling wave. In the waveform of Eq. (2.22), the
constant B is the wave amplitude. The average of u
2
(t. r) can be expressed in terms of
B as u
2
= B
2
,2, where we use the formulae [sin(qr t )]
2
= [cos(qr t )]
2
=
1
2
.
According to Eq. (2.15), the amplitude relates to the wave energy density. Indeed, we
note that
2
_
u
t
_
2
A
2
_
u
r
_
2
=
B
2
2
_
2
[cos(qr t )]
2
q
2
A[cos(qr t )]
2
_
=
B
2
2
_
2
2

q
2
A
2
_
=
B
2
2
_
2
2

2
2
_
=
B
2
2

2
.
In obtaining this result, we have taken into account that q
2
=
2
,s
2
and s =

A,.
Thus, we obtain h =
2
B
2
,2. Importantly, the energy density for traveling plane waves
is independent of the coordinate and it is proportional to
2
. If = 0, there is no wave
and there is no energy associated with it. Note that, for the waveforms of Eqs. (2.20) and
(2.22), averaging over a small volume of the medium is equivalent to averaging over the
period of oscillations.
For traveling waves, one often introduces also the wave intensity, which is the density
of the energy ux. This energy ux represents the energy transferred by the wave per unit
time through a unit cross-section perpendicular to q. The intensity is a vector directed
along q with the absolute value
I = s h =
s
2
B
2
2
. (2.25)
Besides the waveform of Eq. (2.20), a traveling wave can be presented in a complex
form,
u = Ae
i(qrt)
. (2.26)
where the amplitude Ais generally a complex quantity. Indeed, Eq. (2.26) is a solution to
the wave equation (2.18). For some cases, it is convenient to operate with such a complex
waveform. However, one should remember that true physical quantities always have real
values.
Now we consider the important wave phenomenon known as wave interference. Sup-
pose that two sinusoidal waves of the same frequency propagate from different sources
through the medium. The sources of the waves are generally at different locations, so
the waves reach a point of observation r, in general, with different phase shifts
1
(r) and
2
(r):
u
1
(t. r) = B
1
sin(t
1
(r)). u
2
(t. r) = B
2
sin(t
2
(r)). (2.27)
The resulting wave eld is u(t. r) = u
1
(t. r) u
2
(t. r). In experiments, instead of
the wave amplitude, it is the intensity of the wave that is measured in many cases.
From Eq. (2.25) the intensity is proportional to u
2
. Straightforward calculation of u
2
gives us
u
2
=
1
2
_
B
2
1
B
2
2
2B
1
B
2
cos[
1
(r)
2
(r)]
_
. (2.28)
In deriving Eq. (2.28), we have used the identity
sin x sin y =
1
2
cos(x y)
1
2
cos(x y).
i.e.,
sin(t
1
(r))sin(t
2
(r)) =
1
2
cos(
1
(r)
2
(r))
1
2
cos(2t
1
(r)
2
(r)).
and the fact that the average of cos(2t
1
(r)
2
(r)) is zero. Thus, the intensity
of the resulting wave consists of three contributions: the term related to the wave
S
0
S
2
S
1
Light
Screen
I
n
t
e
r
f
e
r
e
n
c
e

p
i
c
t
u
r
e
Figure 2.3 A double-slit experiment: S
1
and S
2
form the double slit.
coming from the rst source, the term related to the wave from the second source,
and an additional term. This third contribution describes the effect of wave inter-
ference. It depends critically on the wave phase difference. The interference contri-
bution can be positive (constructive interference at ,2 - (
1
2
) - ,2 when
cos (
1
2
) > 0), or negative (destructive interference at ,2 - (
1
2
) - 3,2
when cos(
1
2
) - 0). Importantly, the interference effect can be observed only
for waves with the same frequency (otherwise the averaging leads to zero interfer-
ence contribution). The waves of Eq. (2.27) with time-independent phase shifts are
also known as coherent waves. If waves are characterized by a phase shift (
1
2
),
which jitters randomly in time, the waves are incoherent and no interference effect
occurs.
The simplest example illustrating the interference effect is the double-slit optical
experiment. Let two slits be illuminated by a light wave from a single source. These two
slits become two sources of coherent waves. The superposition of these waves generates
an interference pattern of fringes, as shown in Fig. 2.3.
The wave analyzed above travels along the vector q. Using the wavevector q in
Eq. (2.22) at the same frequency, we obtain another wave traveling in the opposite
direction. According to Eq. (2.20), a combination of these waves is also a solution to
Eq. (2.18):
u(t. r) = B
sin(qr t ) B
sin(qr t ).
(a) n = 1 (b) n = 2 (c) n = 3
L
Figure 2.4 Quantization of oscillations in the form of standing waves. The gray area is a
standing-wave pattern. Solid lines show amplitudes of oscillations at an instantaneous moment
of time. Only waves with integer half-wavelengths exist: L = n,2. Note: for demonstration
purposes, the amplitudes are shown not to scale.
where again B
and B
are arbitrary constant vectors. The two waves can also be

interpreted as incident and reected waves.
An important case is that of a wave propagating between two reecting walls
placed at z = 0 and z = L. The waves depend on a single coordinate z: u(t. z) =
B
sin(qz t ) B
sin(qz t ). If the walls are rigid there are no displacements

at the walls, so we should use the boundary conditions for the waves at z = 0 and
z = L: u(z = 0) = u(z = L) = 0. One of the many physical embodiments of this sit-
uation are represented by waves in a continuum elastic medium bounded by rigid
boundaries at z = 0 and z = L; another is given by a string attached to two xed
boundaries at z = 0 and z = L. From the boundary condition at z = 0, we nd
B
= 0, thus u(t. z) = B
[sin(qz t ) sin(qz t )] = 2B
sin(qz) cos(t ).
(Note: sin x sin y = 2 sin[
1
2
(x y)]cos[
1
2
(x y)]). The boundary condition at z = L
can be satised if, and only if, sin(qL) = 0. The latter requires the so-called quantiza-
tion of the wavevector q, qL = n, or
q = q
n
=

L
n. n = 1. 2. 3. . . .. (2.29)
where q
n
is called the wavenumber. That is, there exist only waves of a special form,
u = B sin(q
n
z) cos(
n
t ). (2.30)
at discrete values of the wavevector q
n
and frequency
n
= sq
n
. This important class of
waves is known as standing waves. From Eqs. (2.23) and (2.30), we have the relation
L = (,2)n. For standing waves, a strictly integer number of half-waves may be put
between the reecting walls. Figure 2.4 illustrates standing waves of various wavelengths
for an arbitrary instant of time. The longest possible wave (n = 1) corresponds to = 2L
when a half-wave ts between the reecting walls; see Fig. 2.4(a). To calculate the
energy density of the standing waves, we can use Eq. (2.15). Then, the energy density is
h
n
= B
2
2
n
,4. It is independent of coordinate, as found for the traveling plane wave.
Obviously, the energy ux is now zero. It is important to stress again the difference
between traveling waves (with arbitrary and continuous values of q including q 0,
hence , in innite media) and standing waves with quantized (or discrete) values
of q
n
with the minimum wavenumber q
1
= ,L.
The waveform given by Eq. (2.20) is actually a simple specic example of a more
general class of wave elds. Indeed, due to the linear character of the wave equation,
a sum of an arbitrary number of partial solutions is also a solution of the equation.
This property of waves satisfying linear wave equations is known as the superposition
principle. According to this principle, we may write the solution to Eq. (2.18) in the
general (complex) form as
u(t. r) =
q
A
q
e
i(qr
q
t)
. (2.31)
Here, the summation is taken over the wavevectors q. In addition,
q
relates to q through
the dispersion relation (2.21). The amplitudes of waves contributing to u depend, in
general, on the wavevectors. The superposition principle is the basis for many important
phenomena, including interference, formation of standing waves, and diffraction.
Electromagnetic waves in free space
One of the most important examples of waves is that of electromagnetic waves, i.e.,
oscillating electromagnetic elds. These elds are responsible for the most basic prop-
erties of matter from the nanoworld to the scale of the universe, and they are exploited
for a number of technologies critically important for modern society. The fundamentals
and applications of electromagnetic elds constitute a separate and extremely important
eld of science. In this book, we will deal with electromagnetic elds only briey in
order to illustrate the general character of the wave equation (2.18).
Electromagnetic waves are joint electric and magnetic elds that oscillate in both space
and time. In the simplest homogeneous case, both the electric eld F and the magnetic
eld H are governed by the wave equation in the form (2.18), where one should replace
u F or H and s c, where c is the velocity of light in free space and = qc. The
wave equation, for example, for the electric eld reads as
2
F
t
2
c
2
_
2
F
x
2

2
F
y
2

2
F
z
2
_
= 0. (2.32)
Now, one can write the electric eld in the form of a plane wave similar to Eq. (2.22):
F(r. t ) = bF
0
sin(qr t ).
where F
0
= bF
0
, F
0
is the amplitude of the electric eld, and b is the vector of the
polarization of the wave which denotes the direction of F(r. t ). The parameters q and
have the same meaning as above: the wavevector and the angular frequency of the wave.
Alternatively, it is possible to use a complex form of the plane wave,
F(r. t ) = ibF
0
e
i(qrt)
. (2.33)
but only the real part of this formula has physical meaning. The same kind of equation
may be written for the magnetic eld H.
Direction of
propagation
F
H
x
y
z
Figure 2.5 A plot of electric, F, and magnetic, H, elds as functions of z at time t = 0.
In free space, the vector F is always perpendicular to q, so that, if q is xed, the
electric eld has, in general, two possible polarizations of the electromagnetic wave that
correspond to the two orthogonal directions perpendicular to q. The magnetic eld vector
H is, in turn, perpendicular to both q and F. Propagation of the electric and magnetic
elds is illustrated by Fig. 2.5.
Equations of the form of Eq. (2.32) are valid for a homogeneous or nearly homo-
geneous dielectric medium, with only the replacement c c,
c, where c is the dielec-

tric constant of the medium:
2
F
t
2

c
2
c
_
2
F
x
2

2
F
y
2

2
F
z
2
_
= 0. (2.34)
Since c > 1, the velocity of light, c,
c, in a dielectric medium is less than that in free

space. The frequency, , and wavevector, q, are related through the dispersion relation
=
c
c
[q[.
The energy of the wave can be characterized by the density of the electromagnetic
energy, which is dened as
W = c
0
cF
2
(t ) =
1
2
c
0
cF
2
0
. (2.35)
where F
2
(t ) represents the time average of F
2
(t ) and c
0
is the permittivity of free space:
c
0
= 8.851 10
12
Fm
1
(1 F = 1 CV
1
) . We can dene the intensity of the wave as
the energy ux through unit area perpendicular to the wavevector q:
I =
c
2
c
0
cF
2
0
. (2.36)
The above description is associated with the classical description of the electromagnetic
elds.
In closing this section on classical waves, we summarize that wave motion occurs
in extended media and waves themselves are always delocalized physical entities that
generally extend over large distances.
2.4 Waveparticle duality
In this section we will study how any physical object may behave simultaneously as
a particle and as a wave. To understand this so-called waveparticle duality, we use a
two-way road from waves to particles and from particles to waves.
From waves to particles
We start with a well-known optical example, which will help us to overcome the gap
between classical waves extending over space and discrete particles having certain coor-
dinates. Visible light behaves as an electromagnetic wave exhibiting interference, diffrac-
tion, etc. On the other hand, we often see light as a light ray, which can propagate, be
reected, and be deected. The trajectory of such a light ray is very similar to a par-
ticle trajectory. The class of optical effects for which light can be described in terms
of rays constitutes the discipline of geometrical optics. The transition from wave optics
to geometrical optics is justied if the wavelength of light is much smaller than the
characteristic scales of the problem being considered: _ L, where L can be a distance
of light propagation, the size of an illuminated object, a scale of inhomogeneity of the
medium, etc. Consider the last case with a smooth inhomogeneity, which appears as a
smooth coordinate dependence of the dielectric constant c(r) in Eq. (2.34). We will use
a waveform similar to Eq. (2.26): F = Ae
i
. For monochromatic light (light of a single
frequency) we can write =
0
t . Here,
0
depends only on r. We can introduce the
wavevector again as q = ,r =
0
,r. In an inhomogeneous medium the wavevec-
tor and the wave amplitude depend on the coordinate vector r: q = q(r); A = A(r). On
substituting the chosen waveform, F = Ae
i
, into Eq. (2.34) we obtain
2
Ae
i
c
2
c(r)
_
2
A
x
2
e
i
2i
A
x
0
x
e
i
A
_
0
x
_
2
e
i

_
= 0. (2.37)
where, for simplicity, we have written only the derivatives with respect to the x-
coordinate. Within the geometrical-optics approximation, the phase (t. r) is large:
[[ 1. Moreover, both contributions to the phase are large too: t. [
0
[ 1. Since the
phase is large, the leading terms in our calculations, namely the very rst term (propor-
tional to (,t )
2
=
2
) and the third term in the brackets (proportional to (
0
,x)
2
),
are quadratic with respect to the phase derivatives. Keeping these leading terms and
cancelling out common multipliers, we obtain an equation for
0
:
_
0
r
_
2
=
2
c(r)
c
2
. (2.38)
If we nd a solution of Eq. (2.38), we can dene the so-called wave surfaces on which
0
is constant:
0
(r) =constant. For a given point r, the direction of the wave is determined
by the wavevector q = d
0
,dr; the wave direction is perpendicular to the wave surface.
After we have found the wave surfaces and calculated q(r), we may construct the ray
trajectories, as presented in Fig. 2.6. In the simplest case of a homogeneous medium,
a
b
c
1
2
3
ray
ray
ray
Figure 2.6 A sketch of the construction of ray trajectories by using the wavefronts. Rays a, b, and
c are perpendicular to the wavefronts 1, 2, and 3.
where c does not depend on the coordinate, we obtain a trivial result: q = constant, i.e.,
the ray propagates along a straight path.
Thus, if the wavelength is small in comparison with the characteristic scales of the
system being considered, wave motion can be described in terms of ray trajectories. Up
to this point, our discussion has been based totally on classical physics.
Thus far, we have not discussed the wave amplitude (or the wave intensity). The wave
amplitudes enter the equations only as parameters that may be dened if the parameters
characterizing the light sources are known. However, the wave amplitude is directly
related to the kind of interpretation we are using: a classical interpretation or a quantum
interpretation. Indeed, it has been established that the classical-physics approach is valid
only for large intensities of waves. If the intensity of a wave is small, other laws of
quantum physics dene the wave eld.
Let us consider briey basic quantum concepts related to wave elds. According to
Max Planck and Albert Einstein, electromagnetic waves interacting with matter can be
emitted and absorbed in discrete energy portions (quanta) photons. The energy of a
photon, E, is proportional to the frequency of the wave:
E = h
--
. (2.39)
where h
--
= 1.06 10
34
J s (Joule-second) is the so-called reduced Planck constant;
h = 2h
--
is calledsimplyPlancks constant. Eachphoton, like a particle, has a momentum
p = h
--
q. (2.40)
where q is the wavevector. (Equation (2.40) is formulated for photons in free space.)
Apparently, the direction of the momentum vector p coincides with the direction of
propagation of the wave. For each of the two possible polarizations of the radiation, one
introduces appropriate characteristics of photons: each polarization of light, b, is related
to a certain photon. Thus, a photon can be labeled by two indices, {q. b].
Table 2.1 A comparison between classical and quantum quantities
Classical quantity Corresponding quantum quantity
Density of optical energy, W Number of photons, N
q.b
= WV,(h
--
)
Optical intensity, I(r) Photon-ux density, I(r),(h
--
)
Total optical power, P Photon ux, P,(h
--
) = N
q.b
c
Equation (2.39), which represents the energy of a quantum of light, may be readily
generalized for electromagnetic waves in a dielectric medium with dielectric constant c:
h
--
=
h
--
c
c
q. (2.41)
It is very important that different quanta of an electromagnetic eld do not interact with
each other, as is reected by the linear character of the eld in free space. An interaction
between these modes is possible only in special media. Such media are called nonlinear
optical media.
By introducing quanta, i.e., particles of electromagnetic eld, we may understand
Eq. (2.38) and Fig. 2.6 in a newway. Let us multiply this equation by the factor h
--2
,(2m),
where m is an inessential parameter. Then, we may write
p
2
2m

h
--2
2
c(r)
2mc
2
= 0
upon taking into account the relations q = d
0
,dr and p = h
--
q. On dening a potential
energy as V(r) = h
--2
2
c(r),(2mc
2
), we come to the equation H p
2
,(2m) V(r) =
0, which is a direct analogue of the energy conservation law for a particle (compare this
with Eq. (2.9)). That is, Eq. (2.38) describes trajectories of quanta of the wave eld.
In quantum theory, instead of the wave intensity one introduces the number of quanta
(or photons), N
q.b
. If each of the quanta transfers the energy h
--
, then the intensity,
dened as the energy ux density, is I = h
--
cN
q.b
,V, where V is volume and N
q.b
,V is
the densityof quanta withwavevector q. Because the quantumpicture has tocoincide with
the classical picture for a large number of photons, one can match the latter relationship
and Eq. (2.36) for the classical intensity. From this comparison, it is possible to nd the
relationship between the classical amplitude of the electric eld, F
0
, and the number of
quanta:
F
0
=
_
2h
--
N
q.b
c
0
cV
. (2.42)
Our discussion of the relationship between electromagnetic waves and photons pro-
vides an example of the waveparticle duality that is ubiquitous in quantum physics. A
comparison of some characteristics of electromagnetic elds in the classical and quan-
tum interpretations is given in Table 2.1. For a wave of small intensity (small number
of photons, N
q.b
) the quantum description is more suitable, whereas for a substantial
intensity (large N
q.b
) a classical wave interpretation may be used.
From particles to waves
Consider a particle, say an electron. On the basis of classical mechanics one can attempt
to characterize it by a mass, m, and a vector representing the momentum, p. In classical
physics we can knowwith certainty that a particle is at a certain position in space, r. From
a quantum-mechanical point of viewa particle is characterizedbythe wavefunction, (r).
Frequently, quantum mechanics is referred to as wave mechanics. The wavefunction
is, in general, distributed in space. The main property of the wavefunction is that the
value [(r)[
2
dr gives the probability of nding a particle inside of a small volume
dr around point r. Thus, the wavefunction, , may be interpreted as the probability
amplitude corresponding to the probability density for nding a particle at a particular
point of space, r. That is, there is a direct analogue between [[
2
and the square of
the electromagnetic wave magnitude [F
0
[
2
, which determines the photon density, as
discussed previously.
Probabilistic behavior is one of the key features of quantum mechanics; thus, a word
of explanation is necessary to dene what is meant by probability in this context.
To understand probability in a quantum-mechanical context, it is convenient to have in
mind the following situation. Imagine an ensemble of similarly prepared systems. By
similarly prepared we mean identical systems as far as any physical measurement is
concerned. Now, if a measurement is made on one of the systems to determine whether
a particle is in a particular volume element, the result will be denite: either a particle is
there, or it is not. When the same measurements are made on a large number of similarly
prepared systems, the number of times a particle is found in the xed volume is taken
as a measure of the probability of nding a particle in the elementary volume.
For the simplest case of a particle in free space, the wavefunction has the form of a plane
wave, as introduced previously:
+(r. t ) = Ae
i
= Ae
i(krOt)
. (2.43)
where = kr Ot is the phase, k is the wavevector of the particle, A is the amplitude
of the wave, and O is the angular frequency associated with the energy of the particle.
The wavevector, or more precisely its magnitude k = [k[, is related to the wavelength of
a particle, :
=
2
k
. (2.44)
According to the de Broglie relationship, the momentum of a particle is related to a
wavelength associated with the particle through the equation
=
2h
--
p
. (2.45)
From Eqs. (2.44) and (2.45) we obtain a relation between the wavenumber and the
momentum of a particle: p = h
--
k, or, in the vector form,
p = h
--
k. (2.46)
This relationship coincides exactly with Eq. (2.40) for quantization of electromagnetic
waves. This simplest case demonstrates the way in which one may attribute to a particle
both particle-like and wave-like properties.
Within the plane-wave description of a free particle, it follows from Eq. (2.43) that
there is an equal probability of nding a particle in any point of space:
[+(r. t )[
2
= +(r. t )+
(r. t ) = [ A[
2
= constant.
where +
(r. t ) is the complex conjugate of +(r. t ).

This result appears to be in contradiction with the classical description of a particle.
Just as for the previously discussed case of electromagnetic elds, this contradiction
can be resolved by introducing the uncertainty principle. This principle has the form of
conditions restricting the range of coordinates and the range of momenta that can be
measured simultaneously for a particle. Uncertainties in the quantities Lp and Lr have
to satisfy the following inequalities:
Lp
x
Lx h. Lp
y
Ly h. Lp
z
Lz h. (2.47)
Thus, if a particle is localized in a region of space of width Lx, the uncertainty in
the x-component of its momentum will be greater than or equal to h,Lx. The particle
described by the wavefunction of Eq. (2.43) has a certain momentum p = h
--
k, but is
completely delocalized in space.
Note that the phase of the wave in Eq. (2.43) depends on time. The angular frequency
of the oscillations of this phase is related to the energy of the particle E through O E,h
--
.
The latter again coincides exactly with the energy of the wave quanta introduced by
Eq. (2.39). This kind of temporal phase dependence e
iEt ,h
--
and remains valid for
any complex system under stationary conditions, including the condition of a constant
external eld.
Another important fact is that the superposition principle discussed previously for clas-
sical waves is valid for particle waves. Thus, typical wave phenomena such as interference
and diffraction should be observed for particles. One of the requirements necessary to
observe these effects is coherence of the waves participating in the superposition. In par-
ticle language this means, rst of all, that particles should be monoenergetic. Indeed,
direct experiments with monoenergetic electrons have proved the occurrence of inter-
ference and diffraction of the electron waves. Since the famous 1927 DavissonGermer
experiment on diffraction of electrons by metal crystals, numerous experiments conrm-
ing the wave nature of particles have been done. Recent (1989) experiments by Akira
Tonomura with diffraction of electrons involved repeating a double-slit experiment that
had been performed with light; see Fig. 2.7. In these experiments, direct conrmation of
wave-like properties of electrons was obtained. The experimental setup of Tonomuras
experiment is presented in Fig. 2.7(a). It consists of (i) an electron gun that emits, one
by one, electrons with high velocity; (ii) an electron biprism (electron splitter); (iii) a
detector of diffracted electrons; and (iv) a CCD (charge-coupled device) camera that
records and displays the positions of the registered electrons. Ten electrons per second
were emitted by the source. For the rst several minutes the picture on the CCD screen
reected a chaotic distribution of electrons. Gradually the build-up of registered electrons
(a)
(1) (2)
(4) (3)
(b)
Figure 2.7 Tonomuras experiment. (a) A schematic diagram of the double-slit experiment.
(b) The build-up of interference fringes at various times: pictures on the monitor after (1)
10 electrons, (2) 200 electrons, (3) 6000 electrons, and (4) 140 000 electrons had been detected.
Electrons were emitted at a rate of 10 per second. Used with permission, from A. Tonomura.
(2006). Double-slit Experiment. (http://hqrd.hitachi.co.jp/globaldoubleslit.cfm), gures 1 and 2.
C _
Hitachi, Ltd. 1994, 2006. All rights reserved.
produced an interference pattern characteristic of experiments involving diffraction of
light in a two-slit experiment.
To estimate the particle wavelength and to understand the consequences of the uncer-
tainty principle, let us assume that a free electron moves with a velocity of about
10
7
cm s
1
. The mass of a free electron is m
0
= 9.11 10
28
g; thus, p
0
= m
0
: =
9 10
21
g cm s
1
, k
0
= p
0
,h
--
= 8.7 10
6
cm
1
, and the de Broglie wavelength of a
free electron is
0
= 2,k
0
= 7.2 10
7
cm = 72

A. If we need to measure both the
position and the momentum of the electron, and we impose the limit of 10% accuracy
on the value of its momentum, i.e., Lp
0
= 9 10
22
g cm s
1
and Lk
0
= 8.7 10
5
cm
1
, we cannot predict the position of this electron with an accuracy greater than
Lx = h,Lp
0
= 2,Lk
0
= 7.2 10
6
cm = 720

A. This value is about ten times
greater than the wavelength of the electron!
According to our estimates, we see that electron wavelengths have very small values.
For a material particle with a larger mass, the wavelength is even smaller. That is why in
most cases of ordinary life we do not observe wave-like behavior of particles. As we will
learn later, the wavelength of the electron in solids is ten times larger than that in vacuum,
i.e., up to 1000

A = 100 nm (or more). This is why, for an electron in the situations
encountered in nanoelectronics, wave-like physical properties are its key attributes and
they determine the basic properties of this nanoworld.
2.5 Closing remarks
This chapter emphasizes a fundamental property of all existing matter, which may be
called waveparticle duality. Starting with an analysis of an arbitrary wave eld (elastic
vibrations in solids, electromagnetic elds, etc.), we found particle-like behavior of this
2.6 Problems 29
wave eld at small wave amplitudes and (or) spatial scales larger than the wavelength.
By analyzing particle motion, we have seen that at small spatial scales a particle can not
be characterized by exact coordinates and momentum (velocity) and behaves rather as
an extended wave-like object. Thus, wave properties and particle properties for the same
physical object are compatible. This is the essence of waveparticle duality; indeed, this
is an underlying principle of central importance for nanophysics.
Contemporary microelectronics and optoelectronics exploit electric and light phenom-
ena, which are determined entirely by the properties of electrons in solids. Estimates of
the de Broglie wavelength of electrons in solids give us a value of about 10 nmand larger,
i.e., emerging nanoelectronics inevitably should be based on the wave mechanics.
For those who want to look deeper into waveparticle duality we recommend the
following textbooks:
R. P. Feynman, R. B. Leighton, and M. Sands, Lectures on Physics, vol. 3 (New York,
Addison-Wesley, 1964).
W. E. Lamb Jr. The Interpretation of Quantum Mechanics, edited and annotated by J.
Mehra (Princeton, MA, Rinton Press, 2001).
The problems presented below have two aims: to help the reader to attain a better
understanding of the denitions and principles stated in this chapter, and to illustrate
some unusual behavior of the electrons in solids.
In the following chapters, we will introduce basic denitions and equations of quan-
tum (wave) mechanics and analyze simple instructive examples that illustrate the main
quantitative and qualitative features of nanophysics.
2.6 Problems
1. For manysemiconductor materials usedincontemporaryelectronics, the relationship
between the momentum and energy of an electron is given by the implicit formula
p
2
2m
= E
_
1
E
E
g
_
.
where m
is the so-called effective mass of the electron and 1,E

g
is the so-called non-
parabolicity parameter. The formula has two solutions for unknown E: for electrons
(E > 0) and another for the so-called holes (E - 0).
(a) Find both solutions for E.
(b) Derive the expression for v for electrons only.
(c) Determine the electron velocity in free space and compare it with the expression
derived in (b).
As an example, consider GaAs, for which m
= 0.067m
0
and E
g
= 1.42 eV.
Note: the expression for the kinetic energy of an electron in free space is
E =
p
2
2m
0
.
2. Some metals and semiconductor materials instead of having an isotropic parabolic
energy dispersion, E = p
2
,(2m
), have anisotropic parabolic energy dispersion,

E =
_
1
m
_
i j
p
i
p
j
. (2.48)
i.e., for such cases the electron mass is no longer a scalar, but is instead a tensor. Let
the reciprocal effective-mass tensor (1,m
)
i j
be
_
1
m
_
i j
=
_
m
1
t
0
0 m
1
l
_
. (2.49)
where i. j = x. y. Here, the parameters m
t
and m
l
are the so-called transverse and
longitudinal effective masses of the conduction electron. For this case, the dispersion
relation (2.6) is simplied to
E =
p
x
2
2m
t
p
y
2
2m
l
. (2.50)
For the given momentum vector p = {[p[sin . [p[cos ]
(a) calculate the velocity v = dE,dp, and
(b) plot momentum vectors and velocity vectors corresponding to momentum vectors
with three values of , = 30
. 45
. and 60
, and take [p[ = 1. Consider n-Ge, for

which m
t
= 0.019m
0
and m
l
= 0.95m
0
.
Notice that the directions of the velocity vectors and momentum vectors do not coincide
(i.e., they are not collinear).
3. Consider a particle of mass m, which moves along a single coordinate z. Assume
that the potential force is f = z. Find the corresponding potential energy. Show that
the general solution to Eq. (2.8) is A sin(
,m t ), with A and being arbitrary

parameters. That is, the particle oscillates around the point z = 0 with the frequency
=
,m. Particular values of parameters A and determine the magnitude and the
phase of these oscillations.
4. To illustrate the method of generalization of the one-dimensional model, one can
consider a square (two-dimensional) lattice with identical atoms placed at the corners
of the squares. Each atom interacts with its nearest neighbors, i.e., with four atoms, as
shown in Fig. 2.8. The equilibrium position of each atom is determined now by two
integer numbers, say n and m. The coordinates of the atoms can be written as r
n.m
=
{x
n
. y
m
] = {na. ma]. The atom with a given n and m has neighbors with coordinates
r
n1.m
. r
n1.m
. r
n.m1
, and r
n.m1
. If u
n.m
are displacements of the atoms from their
equilibrium positions, their new vector-coordinates are r
/
n.m
= r
n.m
u
n.m
. Now the
displacements u
n.m
and the forces are vectors. According to Hookes law, the force
acting on the {n. m]th atom from its nearest neighbors is
f
n.m
=
_
(u
n.m
u
n1.m
) (u
n.m
u
n1.m
) (u
n.m
u
n.m1
) (u
n.m
u
n.m1
)
_
.
2.6 Problems 31
x
u
n,m
M
M M
M M M
M M M
y
{n, m}
{n + 1, m}
{n 1, m}
{n, m + 1} {n + 1, m + 1} {n 1, m + 1}
{n 1, m 1}{n, m 1} {n + 1, m 1}
(u
n,m
)
y
(u
n,m
)
x
Figure 2.8 A square two-dimensional lattice.
Calculate f
n.m
, keeping only terms that are linear with respect to displacements u
n.m
.
Show that the force f
n.m
has a form similar to Eq. (2.11), i.e.,
( f
n.m
)
x
=
_
(u
n.m
)
x
(u
n1.m
)
x
(u
n.m
)
x
(u
n1.m
)
x
_
.
etc. Write down the Newton equations for the displacements u
n.m
. Using the previously
discussed replacement of the discrete system by the continuous medium, obtain a two-
dimensional wave equation. Find expressions for the density and the elastic modulus of
the two-dimensional elastic medium considered.
5. The simplest example of the interference effect is given by the superposition of two
sinusoidal waves (u
1
= Bsin
1
(r. t ) and u
2
= Bsin
2
(r. t )). The resulting wave is a
sinusoidal wave of the form u
int
= B
int
sin
3
(r. t ).
(a) Find
3
(r. t ).
(b) Find B
int
.
(c) For the specic case of
1
(r. t ) =
1
(r) t and
2
(r. t ) =
2
(r) t , nd
3
(r. t )
and B
int
, and discuss the differences between your answers for this specic case and
answers (a) and (b) obtained before.
6. In addition to the solutions analyzed in Section 2.2, the wave equation (2.14)
has an innite number of solutions of other forms. Let u(z) be an arbitrary func-
tion that satises Eq. (2.14), for which rst and second derivatives du,dz, and
d
2
u,dz
2
can be calculated. Show that functions u(z st ) are the solutions to the wave
equation.
7. Using the energy density equation
h =

2
_
u
t
_
2
A
2
_
u
r
_
2
and the equation for a standing wave
u = Bsin(q
n
z)cos(
n
t ).
where [B[ =1 cm, s =340 ms
1
, =1.29 kg m
3
, V = L
x
L
y
L
z
, and L
x
= L
y
= L
z
=
2 m.
(a) Calculate the energy density of the wave for n = 1, 2, and 3.
(b) Calculate the total energy of the wave for n = 1, 2, and 3.
8. Consider a wave eld inside of a box of dimensions L
x
, L
y
, and L
z
. The walls of
the box are supposed to reect waves (a mirror box). Find three-dimensional solutions
of Eq. (2.18). Calculate wavenumbers and frequencies of the standing waves. Calculate
the total energy of the standing waves.
9. The superposition principle of Eq. (2.31) allows one to present any wave eld as
a combination of plane waves. The function u(t. z) = Be
[(zst ),Lz]
2
is a wave pulse
propagating through the medium with the velocity s. By using the Fourier transform of
u, present this wave eld as a sum of plane waves.
10. Consider visible (yellow) light with the wavelength = 600 nm (600 10
9
m).
Assume a light wave propagating in free space with the intensity density I =1 mWcm
2
(10
3
J s
1
cm
2
). Calculate the electric eld amplitude F
0
(in units of Vcm
1
). Find
the wavevector, photon momentum, and energy. Estimate the density of quanta N
q.b
.
3 Wave mechanics
3.1 Introduction
Inthis chapter we discuss the basic physical concepts andequations relatedtothe behavior
of particles in the nanoworld. We introduce the Schr odinger wave equation for particles
and determine the ways to calculate observable physical quantities. We nd that, in wave
mechanics, the motion of a particle conned to a nite volume is always character-
ized by discrete values of the energy and standing-wave-like wavefunctions, i.e., such a
motion is quantized. While motion in an innite space (i.e., free motion) is not quantized
and is described by propagating waves, the energy of the particle is characterized by a
continuous range of values.
Keeping in mind the diverse variants of nanostructures, by using wave mechanics we
analyze some particular examples, which highlight important quantumproperties of par-
ticles. Many of the examples analyzed can serve as the simplest models of nanostructures
and will be exploited in following chapters to understand the fundamentals of processes
in nanoelectronics.
3.2 The Schr odinger wave equation
From the previous chapter, we conclude that nanosize physical systems are quantum-
mechanical systems, inasmuch as their sizes are comparable to typical de Broglie wave-
lengths of the particles composing these systems. In dealing with quantum-mechanical
systems, one aims at determining the wavefunction of a single particle or of the whole
system. As we will demonstrate in the subsequent discussion, knowledge of the wave-
function in quantum mechanics is sufcient to describe completely a particle or even
a system of particles. This means that, if we know the wavefunction of a system, we
can, in principle, calculate all macroscopic parameters that dene the properties of the
system.
The wavefunction of a particle + satises the principal equation of quantum mechan-
ics, the time-dependent Schr odinger wave equation,
i h
--
+
t

H+ = 0. (3.1)
34 Wave mechanics
where, h
--
, the reduced Planck constant, h
--
= h,(2), was introduced earlier, and the
operator

H is the Hamiltonian of the system:
H =
h
--2
2
2m
V(r). (3.2)
In quantummechanics, the Hamiltonian is an operator, in contrast to the case in classical
mechanics, where it is a function. Let us consider the consequences of the basic equation
of quantum mechanics, Eq. (3.1). The Hamiltonian operator in quantum mechanics is
constructed by using the classical formof the Hamiltonian of Eq. (2.9), where the particle
momentum p is replaced by the momentum operator p = i h
--
,r. Thus, the rst term
in Eq. (3.2) is the operator of the kinetic energy, with
2
=

2
x
2

2
y
2

2
z
2
(3.3)
being the Laplacian operator. On comparing Eqs. (3.1) and (3.2) for particles with
Eq. (2.18) for wave elds, we may notice that both equations have second derivatives with
respect to the spatial variable, r, and, correspondingly, rst and second derivatives with
respect to time, t . Despite the latter difference, the solutions of Eq. (3.1) are expected to
be in wave-like form.
If the potential V(r) is time-independent, one can separate the dependences on the
time and spatial coordinates:
+(r. t ) = e
iEt ,h
--
(r). (3.4)
where (r) is a complex function of the spatial coordinates only. The time-dependent
wavefunction, +(r. t ), is often called the non-stationary wavefunction, while (r) is
referred to as a stationary wavefunction. By substituting Eq. (3.4) into Eq. (3.1), one
gets the time-independent Schr odinger equation:
_
h
--2
2
2m
V(r)
_
(r) = E(r). (3.5)
In Eqs. (3.4) and (3.5), E is the total energy of a particle. One of the important properties
of solutions to Eq. (3.5) is the orthonormality of solutions
i
(r) and
j
(r) corresponding
to different values of the energy, E
i
and E
j
:
_

i
(r)
j
(r)dr
i j
. (3.6)
where
i j
=
_
1. if i = j.
0. if i ,= j.
The major task of quantum mechanics is to solve the Schr odinger wave equation,
Eq. (3.1).
As we have mentioned already, the wavefunction of a particle in free space
(V(r) = 0) has a plane-wave form+(r. t ) = Ae
i(krOt )
(Eq. (2.43)). On substituting this
Unbound state E
3
Potential well V (z)
E
3
E
2
E
1
Wavefunction y
1
(z)
Bound state E
1
E
z
V
y
2
(z)
y
3
(z)
0
Unbound state E
2
1
2
3
Figure 3.1 Three types of solutions of the Schr odinger equation for a one-dimensional well of
arbitrary form.
wavefunction into the Schr odinger equation (3.1), one gets the relationship between the
electron wavevector and its energy,
E = h
--
O =
h
--2
2m
_
k
2
x
k
2
y
k
2
z
_
=
p
2
2m
. (3.7)
which coincides with the classical relationship between the particles momentum, p, and
its energy, E.
In general, the exact value of the energy E characterizes the system only for the so-
called stationary-state case, when the potential energy and, therefore, the Hamiltonian
do not depend on time. In addition, for a stationary situation, a determination of energy
requires an innite time of observation (measurement). If this time, Lt , is nite, the
inaccuracy of a measurement of the energy, LE, should satisfy the following inequality:
LE Lt h. (3.8)
This is the uncertainty relation between the energy and the time. Inequalities (2.47) and
(3.8) constitute fundamental uncertainty relations in quantum physics.
Equation (3.5) has the form known as an eigenvalue equation. The energy, E, is its
eigenvalue and the wavefunction, (r), is its eigenfunction. The eigenvalue E may run
over discrete or continuum values, depending on the shape of the potential function,
V(r), and the boundary conditions.
In order to illustrate both possible types of solutions of the Schr odinger equation
and energy states, as well as to clarify the tasks which arise, let us consider the one-
dimensional problem for a system with potential energy V(r) = V(z) shown in Fig. 3.1.
Here the vertical axis depicts the energy E, and z represents only one space coordinate.
The potential has a negative minimum at z = 0, tends to zero as z , and saturates
to a nite value V
at z . This potential is the most general formof a potential well.

At this point, a short qualitative discussion serves to emphasize that the boundary con-
ditions dene the type of solution. These solutions will be obtained and discussed in more
36 Wave mechanics
detail in due course. Among the possible solutions of the Schr odinger equation (3.5) with
the chosen potential, there can exist solutions with negative energy E = E
i
- 0. The
wavefunction corresponding to energy E = E
1
- 0 is represented in Fig. 3.1 by curve 1.
One of the peculiarities of solutions with negative energy is that the spatial region with
classically allowed motion, where the kinetic energy p
2
,(2m) = E V(z) > 0, is cer-
tainly restricted. In the classically forbidden regions where V(z) > E, the -function
decays as [z[ . The states of particles, like those just described, are the so-called
bound states and they are characterized by a discrete energy spectrum. Consider next the
possible solutions for the energy range 0 E V
as shown by the line labeled E

2
in
Fig. 3.1. These solutions exist for any values of E; they are nite as z and pene-
trate slightly into the barrier region V(z) > E as shown by curve 2 in Fig. 3.1. In the limit
z , these solutions may be represented as a sum of two waves traveling in oppo-
site directions: one wave is incident on, and the other is reected from, the barrier. For
each energy in the range 0 E V
there is only one solution satisfying the physical

requirements. For any energy E > V
there exist two independent solutions; see curve 3

corresponding to the energy E
3
. One solution may be chosen in the form of the wave
propagating from left to right. At z it has only one component, namely the wave
overcoming the barrier, and as z there is a superposition of incident and reected
waves. It must be emphasized that the wave reected from the barrier, when its energy
exceeds the height of the barrier, arises only in quantum physics. The other wavefunc-
tion can be chosen in the form of waves propagating from right to left. The latter is an
example of continuum energy spectra with energies E > V
when wavefunctions are

nite far away from the potential relief. These considerations demonstrate the impor-
tance of boundary conditions for Eq. (3.5): if the decay of wavefunctions (i.e., 0 at
z ) away from the potential well is required, the discrete energies and the bound
states can be found and they are of principal interest; if the boundary conditions cor-
respond to the incident wave, continuous energy spectra will be obtained. To conclude
the discussion of boundary conditions, let us mention that we often use potential ener-
gies with discontinuities. In this case, at the point of discontinuity of the potential, we
should require both continuity of the wavefunction and continuity of the derivative of the
wavefunction with respect to coordinates.
Since the Schr odinger equation is linear, it is clear that, if a function + is a solution
of the equation, then any function of the form of constant + is also a solution of the
same equation. To eliminate this ambiguity, we have to take into account the probabilistic
character of the wavefunction. Indeed, if a physical systemis enclosed in a nite volume,
the actual probability of nding a particle in this volume must equal 1, i.e.,
_
[+(r. t )[
2
dr = 1. (3.9)
Equation (3.9) is called the normalization condition. It provides the condition needed to
determine the constant multiplicative factor of the wavefunction for the case of a system
of nite size.
If an innite volume is under consideration and the integral of Eq. (3.9) does not exist,
there are other normalization conditions instead of Eq. (3.9). Consider, for example, the
so-called scattering problem, in which electrons come from innity and are scattered
by a local potential. This case corresponds to energies E
2
and E
3
for particle motion in
the potential sketched in Fig. 3.1. For this problem one can assume the incident wave
to be a plane wave with a given amplitude A: +(z. t ) = Ae
ikz
e
iEt ,h
--
. Then, the amount
of scattered waves will be proportional to the amplitude A, due to the linearity of the
Schr odinger equation. Very often the last condition is referred to as an initial condition
instead of as a boundary condition because we are dealing with the state before the
scattering and states after the scattering.
In an overview of quantum mechanics it is very useful to introduce the density of
particle ow, i. In classical physics, the density of ow is a vector, which species the
direction of particle ow and has a modulus equal to the number of particles crossing a
unit area perpendicular to the area per unit time. In quantum mechanics, this quantity is
given by the following formula
i =
ih
--
2m
(
). (3.10)
For example, the density of ow of particles described by the plane wave of Eq. (2.43)
is straightforwardly found to be
i =
h
--
k
m
[ A[
2
=
p
m
[ A[
2
= v[ A[
2
.
Average values of physical quantities
In light of the probabilistic character of the description of quantum-mechanical systems,
we have to clarify how to determine the average values of quantities that characterize
such systems. The simplest case is the calculation of the average coordinate of a particle.
Indeed, the absolute square of the normalized wavefunction gives the actual probabil-
ity per unit volume of nding a particle at a particular point in space, as studied in
Section 2.4. Hence, the average value of a particular coordinate, say z, is given by
z) =
_

z dr =
_
z[[
2
dr. (3.11)
Thus, the integral over space gives the mean, or expectation value, of coordinate z. It
must be stressed again that the meaning of the expectation value is the average of a
number of measurements of the coordinate z carried out over an ensemble of identical
particles.
Equation (3.11) can be generalized to a more general form for the calculation of the
expectation value of any observable a:
a) =

A) =
_

A dr. (3.12)
where the operator

A is associated with the observable a. From the denition of the
expectation value, Eq. (3.12), one can see that, if
a
is an eigenfunction of the operator
A and corresponds to a certain eigenvalue a,
A
a
= a
a
. (3.13)
38 Wave mechanics
then the expectation value of the physical observable a coincides with the eigenvalue:
a) = a. For example, if the wavefunction is the solution of the Schr odinger equation
(3.5), one may calculate the mean energy, E):
E) =

H) =
_

E

H
E
dr = E. (3.14)
However, if a particle were in a state with no well-dened energy, say a state that is
characterized by the superposition of the solutions
E
i
,
=
i
C
i
E
i
. (3.15)
one would obtain the average energy in the form
E) =
i
[C
i
[
2
E
i
with
i
[C
i
[
2
= 1. (3.16)
Here, we take into account the orthogonalization and normalization conditions of
Eqs. (3.6) and (3.9).
It is necessary to explain the differences between the two cases given by Eqs. (3.14)
and (3.16). The rst case is related to a system characterized by a wavefunction, which is
an eigenfunction of the operator

A; in this particular case

A =

H. The second case cor-
responds to the situation described by a superposition of the eigenfunctions of the same
operator. Measurements of the value of the energy for the rst case would reproducibly
give the same result, E. In the second case, the measurements would give us different
probabilistic results: energies E
i
will be measured with their probabilities [C
i
[
2
, and only
their average, E), remains the same.
Importantly, by using the Schr odinger equation, straightforward calculations of deriva-
tives of the average vector-coordinate r) with respect to time show that
m
d
2
dt
2
r) =
_
dV(r)
dr
_
.
i.e., the classical Newton equation (2.8) is recovered in terms of expectation values
obtained using wave mechanics.
Thus, the Schr odinger equation describes the evolution of the wavefunction of the
quantum-mechanical systemof particles. Its solution with proper boundary or/and initial
conditions gives us all the information necessary to calculate macroscopic parameters
of the physical system and device operation being analyzed.
3.3 Wave mechanics of particles: selected examples
The main principles of quantum mechanics have been discussed in Section 3.2. To
understand new peculiarities of particles arising due to the waveparticle duality, these
basic principles of wave mechanics are applied here in several instructive examples.
The simplest of these are related to the so-called one-dimensional case in which only
one-dimensional solutions are considered.
3.3 Wave mechanics of particles 39

V = V =
W
a
v
e
f
u
n
c
t
i
o
n
s
,

y
n
(
z
)
Q
u
a
n
t
i
z
e
d

e
n
e
r
g
i
e
s
,

e
n
z
0
z
0
L
2
L
2
L
2
L
2
n = 4
n = 3
n = 2
n = 1
Figure 3.2 Solutions of the Schr odinger equation for a quantum well with innite barriers.
Eigenenergies,
n
, are dened as
n
= n
2
1
, where
1
= h
--2
2
,(2mL
2
).
Of course, we live ina three-dimensional world. Thus, wavefunctions of real particles
are dependent on the three-dimensional coordinate vector r. However, often, the potential
energy is dependent on a single coordinate, or two coordinates. For such cases, there
exists a signicant simplication of the wave-mechanical description. Indeed, let V
be a function of the z-coordinate only. Then, we can introduce the two-dimensional
wavevector k
[[
= {k
x
. k
y
] and write the wavefunction as
+(t. r) = e
i(k
x
xk
y
yOt )
(z). (3.17)
i.e., the wavefunction is a plane wave propagating in the {x. y]-plane with the amplitude
modulated in the z direction. By substituting Eq. (3.17) into Eq. (3.1), we nd the one-
dimensional equation
_
h
--2
2m
d
2
dz
2
V(z)
_
(z) = 0. (3.18)
where
= E
h
--2
k
2
[[
2m
. (3.19)
Here E. h
--2
k
2
[[
,(2m), and can be interpreted as the total energy, the kinetic energy of
free motion along the x- and y-coordinates, and the energy associated with motion
in the potential V(z), respectively. Equations (3.17) and (3.18) demonstrate for the case
being considered that the motion along the z-coordinate is independent of motion in the
two other directions and can be analyzed in terms of one-dimensional motion.
A particle between two impenetrable walls
Initially, let us consider the one-dimensional case, when a particle is placed between two
impenetrable (rigid) walls at z = L,2, as depicted in Fig. 3.2. Thus, L is the width of
40 Wave mechanics
a one-dimensional well with innitely high potential barriers, namely V(z) = 0 inside
of the well and V(z) = outside of the well:
V(z) =
_
0. for [z[ - L,2.
. for [z[ > L,2.
(3.20)
The spatially dependent part of the wavefunction (z) satises the time-independent
Schr odinger equation (3.18):
_
h
--2
2m
d
2
dz
2

_
(z) = 0. (3.21)
In the general case, the solution to Eq. (3.21) is
(z) = Ae
iKz
Be
iKz
. (3.22)
where we have introduced
K =
_
2m
h
--2
. (3.23)
Impenetrable walls are viewed as boundaries that exclude the particle. Specically, for
the case of impenetrable walls, the probability of nding the particle at z L,2 or
z L,2 is zero, i.e.,
(L,2) = (L,2) = 0. (3.24)
These are, thus, the boundary conditions for the wavefunction of a particle in a quantum
well with impenetrable walls. By applying these boundary conditions to the wavefunction
(3.22), we obtain
Ae
iKL,2
Be
iKL,2
= 0 for z = L,2.
Ae
iKL,2
Be
iKL,2
= 0 for z = L,2. (3.25)
A nontrivial solution of this algebraic system of equations exists if, and only if the
following determinant is equal to zero:
e
iKL,2
e
iKL,2
e
iKL,2
e
iKL,2
= 0.
which results in
sin(KL) = 0 or KL = n. (3.26)
The latter equation determines the eigenvalues, K
n
:
K
n
=

L
n. n = 1. 2. 3. . . .; (3.27)
n = 0 is excluded because for K
0
= 0 Eqs. (3.25) and (3.22) give (z) 0. From
Eq. (3.23) and (3.27), we nd possible energies of this particle:
n
=
h
--2
K
2
n
2m
=
h
--2
2
2mL
2
n
2
. (3.28)
By substituting Eq. (3.27) into Eq. (3.25), it is easy to check that at a given n the
relationship between coefcients A and B is B = e
in
A = (1)
n1
A. Thus, if n is an
odd integer number, we obtain symmetric wavefunctions from Eq. (3.22):
n
(z) =
_
2
L
cos
_
nz
L
_
. n = 1. 3. . . . (3.29)
If n is an even integer, we obtain anti-symmetric wavefunctions:
n
(z) =
_
2
L
sin
_
nz
L
_
. n = 2. 4. . . . (3.30)
In these functions, the factor
2,L arises as a result of the normalization condition:

_

[
n
(z)[
2
dz = 1. (3.31)
The integer n is called the quantum number. Actually, we see that the energies
n
do
not depend on the sign of the quantum number n. The same is valid for the physically
important quantity [
n
(z)[
2
. Thus, we can use only, say, positive quantum numbers,
n > 0. Let us write down the four lowest states explicitly:
n = 1.
1
=
h
--2
2
2mL
2
.
1
=
_
2
L
cos
_
z
L
_
;
n = 2.
2
= 4
1
.
2
=
_
2
L
sin
_
2z
L
_
;
n = 3.
3
= 9
1
.
3
=
_
2
L
cos
_
3z
L
_
;
n = 4.
4
= 16
1
.
4
=
_
2
L
sin
_
4z
L
_
. (3.32)
These solutions allow one to draw some important conclusions.
First, the energy spectrum of the particle conned by a potential well is discrete. That
is, instead of a continuous change of energy as is characteristic in classical physics
a quantum particle placed in a well may have only certain discrete energy values. In
other words, the energy spectrum becomes quantized, as shown in Fig. 3.2. Often, one
refers to this type of the spectrum as a set of discrete energy levels. Interestingly, for the
case being considered, the interlevel distances,
n1
n
, increase with the number of
levels n.
Second, the lowest energy level (usually called the ground state) is not zero; it is nite.
That is, the particle can not have zero energy! Actually, this is a direct consequence of
the uncertainty principle. Indeed, a particle placed in a well is localized in a space region
of size L, i.e., the uncertainty Lz is less than or equal to L. According to Eq. (2.47),
such a localization leads to an uncertainty in the momentum Lp Lp
z
h,L. Then,
the momentum of the particle can be estimated by p Lp, i.e., the momentum is not
zero and there is a non-zero total energy.
Third, the wavefunctions are, in fact, standing waves, and strictly integer numbers of
half-waves may exist between these impenetrable walls. This result is mathematically
42 Wave mechanics
e
2
e
3
e
1
0
V= V
b
z
V = V
b
L
2
L
2
0
Q
u
a
n
t
i
z
e
d

e
n
e
r
g
i
e
s
,

e
n
Figure 3.3 Eigenenergies of a particle in a quantum well with nite potential barriers.
similar to the case of the quantization of standing elastic waves between rigid walls
depicted in Fig. 2.4.
A particle in a quantum well with nite potential barriers
Now we consider the case of particle connement in a potential well with potential
barriers of nite height. We accept the following idealized form of the potential as
illustrated in Fig. 3.3:
V(z) =
_
0. for [z[ L,2.
V
b
. for [z[ L,2.
(3.33)
where V
b
and L are the height and the width of the well, respectively. In this case, since
the energy spectrum is quantized such a potential well is frequently referred to as a
quantum well with nite potential barriers.
Using our classical-physics intuition, we may expect that a particle will be conned
in the well if its energy - V
b
. Let us focus just on this case. First, we have to solve
Eq. (3.18). Inside the well, V(z) = 0, and the solution is a simple combination of sine
and cosine functions, as in the previous case of Eqs. (3.29) and (3.30). So, it is convenient
to rewrite it in a form that combines them,
(z) = C cos(k
w
z) D sin(k
w
z). for [z[ L,2. (3.34)
where
k
w
=
_
2m,h
--2
. (3.35)
and C and D are arbitrary constants. Outside the well the solution has the form
(z) =
_
Ae
b
(zL,2)
. for z L,2.
Be
b
(zL,2)
. for z L,2.
(3.36)
where
b
=
_
2m( V
b
),h
--
2
. As a result of the symmetry of the problem, we can
choose either even or odd combinations in Eqs. (3.36) and (3.34). Consequently, conti-
nuity of the wavefunction implies that A = B for the even solutions and A = B for the
odd solutions. Now we have two constants for odd solutions and two for even solutions.
In particular, for even solutions we nd
(z) =
_
C cos(k
w
z). for [z[ L,2.
Ae
b
(zL,2)
. for [z[ L,2.
(3.37)
where the signs and correspond to positive and negative values of z, respectively.
The next step in nding the solution is to match the wavefunctions and their derivatives
with respect to z at the points z = L,2. For example, for even solutions we obtain from
Eq. (3.37) the system of algebraic equations
C cos(k
w
L,2) A = 0.
(3.38)
Ck
w
sin(k
w
L,2) A
b
= 0.
This system of algebraic equations has solutions if the corresponding determinant is
equal to zero. So, for even and odd solutions we obtain
tan
_
k
w
L
2
_
=

b
k
w
. (3.39)
cot
_
k
w
L
2
_
=
b
k
w
. (3.40)
These trigonometric equations can be solved numerically, but it is more instructive to
analyze them graphically. Let us transform them into the following equations:
cos(k
w
L,2) = k
w
,k
0
. for tan(k
w
L,2) > 0. (3.41)
sin(k
w
L,2) = k
w
,k
0
. for cot(k
w
L,2) - 0. (3.42)
where k
0
=
_
2mV
b
,h
--2
. The signs and in Eq. (3.41) are to be chosen when
values of cos(k
w
L,2) are positive and when they are negative, respectively. The same is
valid for and regarding sin(k
w
L,2) in Eq. (3.42) for the odd solutions.
The left-, L, and right-, R, hand sides of Eqs. (3.41) and (3.42) can be displayed on
the same plot as the functions of k
w
; see Fig. 3.4. The right-hand sides of Eqs. (3.41) and
(3.42) are linear functions with slope equal to k
1
0
. The left-hand side is a cosine or sine
function. Intersections of these two curves give us values k
w.n
for which our problem has
solutions satisfying all necessary conditions.
To analyze the results, we note that the problem is characterized by two independent
parameters: the height of the well, V
b
, and the width, L. We can x one of these parameters
and vary the other. Let us vary the height of the well V
b
, i.e., the parameter k
0
. In this case
the left-hand side of Eq. (3.41) and the corresponding curves Lin Fig. 3.4 do not change,
but the slope of the line, k
w
,k
0
, is controlled by k
0
. We can see that at small k
0
(small
V
b
), when the slope is large, as shown by the dasheddotted line in Fig. 3.4, there is only
one solution k
w.1
corresponding to small k
0
. This rst solution exists at any k
0
and gives
the rst energy level
1
= h
--2
k
2
w.1
,(2m). As k
0
increases a new energy level emerges at
44 Wave mechanics
slope k
0
1
0
LR
n=1
1
k
w,1
n=2
n=3
n=4
p
L
2p
L
3p
L
4p
L
k
w
Figure 3.4 A graphical solution for Eqs. (3.41) and (3.42). Solutions correspond to the
intersections of the curves. The dashdotted line corresponds to small
0
, resulting in only one
solution of Eq. (3.41). The dotted line corresponds to a critical value of
0
at which a second
level appears in the well as a result of the solution of Eq. (3.42). The solid line corresponds to an
intermediate value of
0
leading to four solutions.
L (A)
100 200 300 0
5
10
15
N
u
m
b
e
r

o
f

b
o
u
n
d

s
t
a
t
e
s
(a)
(b)
Figure 3.5 The number of bound states of a square well plotted as a function of the well
thickness: (a) V
b
= 224 meV, m = 0.067m
0
; (b) V
b
= 150 meV, m = 0.4m
0
.
k
w
= k
0
= ,L, as is shown by the dotted line in Fig. 3.4, with energy slightly below
V
b
, (
2
V
b
). When k
0
increases further, the rst and second levels become deeper and
a third level occurs in the well, and so on. Indeed, new levels occur when the parameter
_
(2mV
b
L
2
),(
2
h
--2
) becomes an integer.
Figure 3.5 depicts the number of bound states as a function of well thickness for two
specic values of V
b
andfor twoparticle masses. The parameters chosenare characteristic
for articial quantum wells fabricated on the basis of AlGaAs/GaAs materials (see the
next chapter). One can see that for thin quantum wells (L - 100

A) only a few energy
levels can occur.
The rst example we discussed is related to an innitely deep well, when k
0
. In
this case, the slope of the linear function in Fig. 3.4 tends to zero, solutions correspond
to k
w
= n,L, and the energy levels are given by Eq. (3.28) as discussed earlier.
Thus, as for the rst example, for a quantum well with walls of nite height, we have
also found discrete energy levels. The lowest energy level (the ground state) is also
non-zero. However, the number of discrete levels is nite. Another important phe-
nomenon is the penetration of a particle under a barrier. Equation (3.36) describes
the wavefunction under a barrier. To explain this effect, consider the energy (2.9) for a
classical particle: = p
2
,(2m) V(z). Since p
2
is always positive, particle motion is
possible in the spatial region with V(z) 0. For - V
b
, this classically allowed
region coincides with the region within the quantum well. The motion of such a clas-
sical particle is always restricted to a nite interval of the coordinates and we never nd
the classical particle outside of the region where its total energy is less than the poten-
tial energy: i.e., a classical particle treats any barrier as an impenetrable barrier. The
quantum-mechanical analysis shows that the wavefunctions are nite at any coordinate
and we can nd a particle even in classically forbidden regions, or, as one says, under
the barrier.
The effect of penetration of a particle under a barrier is the so-called tunneling effect. It
is principally a quantum-mechanical phenomenon. The probability of nding the particle
rapidly decreases as we move away from the classically allowed interval.
A conning potential with quadratic coordinate dependence
Nowwe will briey analyze the very important case of particle connement in a potential
with quadratic coordinate dependence:
V(z) =
1
2
d
2
V
dz
2
z
2
1
2
m
2
z
2
. (3.43)
where we introduce intentionally specic notation for the second derivative of the poten-
tial V. Indeed, in classical physics, the force corresponding to this potential is a linear
function: f (z) = m
2
z and the solutions to Newtons equation (2.8) are functions
oscillating with frequency ; see Problem 3 of Chapter 2. This type of particle motion is
that of a harmonic oscillator. Moreover, the rst non-zero term in the Taylor expansion
of any potential near its minimum is quadratic with respect to displacement from its
minimum as in Eq. (3.43). This is why the example of the harmonic oscillator has wide
applicability.
For the harmonic oscillator potential, the time-independent Schr odinger equation
(3.18) can be rewritten as
d
2
dz
2

2m
h
--2
_

m
2
z
2
2
_
= 0.
It is convenient to introduce the characteristic length, z
0
=
_
h
--
,(m). In dimensionless
coordinates, = z,z
0
, the wave equation takes the simple form
d
2
d
2

_
2
h
--
2
_
= 0. (3.44)
Consider rst at very large coordinates:
2
2,(h
--
), when the second term in
brackets dominates. Then, Eq. (3.44) reduces to
d
2
d
2

2
= 0. (3.45)
46 Wave mechanics

n = 0
n = 1
n = 2
e
2
=
2
5hw
e
1
=
2
3hw
e
0
=
2
hw
classical
turning point
z
0
E
n
e
r
g
y
Figure 3.6 Eigenenergies and eigenfunctions of a particle in a parabolic quantum well (harmonic
oscillator).
The asymmetric solution of the latter equation is proportional to e
2
,2
. Taking into
account that the solutions should be nite at [[ , we can look for a wavefunction
that will be a solution of Eq. (3.44) for all values of in the form
() = e
2
,2
P().
It can be shown that, to satisfy the normalization condition (3.31) for wavefunctions,
the unknown function P() should be a polynomial of a special form, which is called a
Hermite polynomial. This is possible only at certain values of the energy:
n
= h
--
_
n
1
2
_
. n = 0. 1. 2. . . . (3.46)
For each value of n, the function P() can be calculated by using derivatives:
P
n
() = (1)
n
e
2 d
n
d
n
e
2
.
Returning to the coordinate z, we present the three lowest states as
n = 0.
0
=
1
2
h
--
.
0
(z) =
_
m
h
--
_
1,4
e
z
2
,(2z
2
0
)
;
n = 1.
1
=
3
2
h
--
.
1
(z) =
_
m
h
--
_
1,4
z
z
0
e
z
2
,(2z
2
0
)
; (3.47)
n = 2.
2
=
5
2
h
--
.
2
(z) =
_
m
4h
--
_
1,4
_
1 2
z
2
z
2
0
_
e
z
2
,(2z
2
0
)
.
The quadratic potential, energy positions, and wavefunctions are shown in Fig. 3.6.
Importantly, in contrast to the previous examples, the interlevel distance is strictly con-
stant and equals h
--
. This case can be classied as the case of equidistant energy levels.
As in the previous example, the particle penetrates, i.e., tunnels under, a barrier of nite
value. For comparison, we indicate that the classically allowed region of motion for the
particle is restricted by the two crossings of the curve V(z) and the line E = constant,
as illustrated by Fig. 3.6.
Concluding these examples of particles placed into conning potentials, we emphasize
the fundamental signicance of Plancks constant, h
--
: all quantized energies and interlevel
distances are proportional to h
--
. If we formally suppose that h
--
0, the effects of energy
quantization, tunneling, etc., will disappear. That is, this limit corresponds to classical
mechanics.
Quantized electromagnetic waves as harmonic oscillators
The harmonic oscillator model has a number of applications. The most important is
quantization of electromagnetic waves. This is one more illustration of the fundamental
principle of waveparticle duality.
In Section 2.3, we described briey the quantization of electromagnetic waves. The
crossover from a classical wave characterized by the wavevector q and polarization to
its quantized analogue is based on introducing the energy quantum given by Eq. (2.39)
and the photon number N
q.
. The energy of the wave must increase linearly with N
q.
. In
other words, energy states of the eld are equidistant, just as for the harmonic oscillator
studied above. This link between the quantized eld and the harmonic oscillators is
developed as follows. According to quantum physics, the electromagnetic eld consists
of an innite number of modes (waves), each of which is characterized by a wavevector,
q, and a specic polarization, . Each mode {q. ] may be described in terms of a
harmonic oscillator of frequency = cq,
c. Here, c and c are the speed of light and

the dielectric constant, respectively. Correspondingly, the energy separation between
levels of this quantum-mechanical oscillator is h
--
= (h
--
c,
c)q (see Eq. (2.41)). This

oscillator may be in the non-excited state, which manifests the ground-state or zero-
point vibrations of the electromagnetic eld. The existence of this zero-point energy is a
purely quantum-mechanical phenomenon, as manifested by the nite lowest energy for
particles in a conned potential. The oscillator may be excited to some higher energy
level. Let the integer N
q.
be a quantum number of this level; then, the energy of the
electromagnetic eld associated with the oscillator in mode {q. ] is
W
q.
V =
_
N
q.

1
2
_
h
--
. (3.48)
where W
q.
is the energy density of the mode and V is the volume of the system.
Equation (3.48) is the quantum analogue of the energy density given by the classi-
cal equation (2.35). The excited-level number, N
q.
, is interpreted as the number of
quanta, or the number of photons in the mode under consideration. At large pho-
ton numbers, N
q.

1
2
, we immediately obtain the classical relationships given in
Table 2.1.
The use of such an interpretation of wave-eld quantization allows one to explain
all known effects of the interaction of electromagnetic elds with matter. For example,
48 Wave mechanics
0 k
x
k
y
e
1
e
2
e
3
E
Figure 3.7 An example of two-dimensional subbands. The subbands are represented by the
paraboloids in {E. k
x
. k
y
]-space;
1
,
2
, and
3
indicate the bottoms of subbands.
the existence of zero-point vibrations completely explains spontaneous emission from
atoms, an effect that does not exist in classical physics.
Low-dimensional subbands and low-dimensional systems
Nowwe return to the analysis of quantization effects related to particles. By generalizing
the examples of conned potentials previously considered, we can conclude that the
energy spectrum for a potential conning a particle in one dimension is
E = E
n.k
[[
=
n
h
--2
2m
_
k
2
x
k
2
y
_
. (3.49)
The energy spectrum depends on three parameters: two continuous parameters, k
x
and
k
y
, and the discrete quantum number, n. Such a spectrum is interpreted as a set of energy
subbands: a set of paraboloids with minima at quantized energies
n
, as shown in Fig. 3.7.
According to Eq. (3.49), a particle characterized by a given quantumnumber n can move
freely only in the {x. y]-plane. We can say that the particle possesses only two degrees of
freedom. In such a case, the particle may be viewed as a particle that is free to move in two
dimensions. The two-dimensional character of particles becomes apparent when energy
intervals between subbands are large. This, in turn, is possible when the potential well has
a width in the nanometer regime. Such a quantum well with low-dimensional particles
(electrons) can be fabricated by using thin (tens of nanometers) semiconductor layers.
Simple examples are just thin suspended lms or free-standing structures (depending on
the fabrication method), as illustrated by Fig. 3.8. Another example is a thin layer placed
between two other materials. To explain this example, we will dene in Section 4.5 the
energy required to remove an electron from a semiconductor to a vacuum outside of the
semiconductor the so-called electron afnity. If a thin lm possesses a large electron
afnity and surrounding materials are of smaller afnities, the difference between the
electron afnities explains the formation of a potential well for the electrons. Figure 3.9
(a) Suspended thin film (b) Free-standing thin film
Figure 3.8 Examples of suspended and free-standing thin lms.
z
L B
A
A
Figure 3.9 A three-layer structure where layers A create potential barriers for electrons from
layer B.
illustrates a heterostructure system forming quantum wells, which are important for
nanoelectronics.
Quite similarly, connement of a particle in two directions leads to additional energy
quantization and leaves only one degree of freedom for particle-wave propagation. The
energy for this case is
E = E
n
1
.n
2
.k
x
=
n
1
.n
2

h
--2
2m
k
2
x
. (3.50)
where the quantized energy
n
1
.n
2
depends on two integer quantum numbers, n
1
and n
2
,
and motion along the propagation direction (assumed to be the x direction) is charac-
terized by the one-dimensional wavevector k
x
. Thus, the spectrum consists of a set of
one-dimensional subbands as shown in Fig. 3.10 (compare the imaginary slices of the
parabaloids of Fig. 3.7 at xed values of k
y
with the parabolas of Fig. 3.10). Such a quan-
tized particle is free in one dimension and the corresponding articial structure is called
a quantum wire. Currently, there exist several technological methods for the fabrication
50 Wave mechanics
0 k
x
e
11
e
12
e
13
E
Figure 3.10 An example of one-dimensional subbands. The subbands are presented by the
parabolas in {E. k
x
]-space;
11
,
12
, and
13
indicate the bottoms of subbands.
Figure 3.11 A scanning electron microscope image of free-standing InP quantum wires. Reprinted
with permission from Thomas M artensson, Patrick Carlberg, et al., Nanowire arrays dened by
nanoimprint lithograph, Nano Lett., 4 (4), 699702 (2004).
C _
American Chemical Society.
of quantum wires. For illustration, in Fig. 3.11, an array of free-standing quantum wires
made from InP is presented.
In the following sections we will show that transformation of spectra from three-
dimensional to low-dimensional energy subbands drastically changes the behavior of
the particles.
Quantum-box, dot, and zero-dimensional systems
So far, we have considered examples where a particle is conned in one or two directions.
This leads to quantization of the electron spectrum, resulting in two-dimensional or one-
dimensional energy subbands. But there still is at least one direction for free propagation
of the particle along the barriers of conning potentials. The advances in semiconductor
y
x
z
L
y
L
x
L
z
Figure 3.12 A sketch of a quantum box, L
x
L
y
L
z
, embedded in a matrix.
technology allow one to go further and fabricate heterostructure-based structures, in
which all existing degrees of freedom of electron propagation are quantized. These so-
called quantum dots, or quantum boxes, or zero-dimensional systems, are much like
articial atoms, and they exhibit extremely interesting behavior.
To consider the energy spectrum of a zero-dimensional system, we have to study
the Schr odinger equation (3.5) with a conning potential that is a function of all three
coordinates and that connes the electron in all three directions. The simplest case is the
quantumbox in the formof a parallelepiped with impenetrable walls. The corresponding
potential, V(x. y. z), is
V(x. y. z) =
_
0. inside of the box,
. outside of the box,
(3.51)
where the box is restricted by the conditions 0 x L
x
. 0 y L
y
, and 0 z L
z
;
see Fig. 3.12. Using the results of the one-dimensional analysis discussed previously,
one can write down the solutions of the Schr odinger equation for a box:
E
n
1
.n
2
.n
3
=
h
--2
2
2m
0
_
n
2
1
L
2
x
n
2
2
L
2
y
n
2
3
L
2
z
_
. n
1
. n
2
. n
3
= 1. 2. 3. . . ..
n
1
.n
2
.n
3
(x. y. z) =
_
8
L
x
L
y
L
z
sin
_
xn
1
L
x
_
sin
_
yn
2
L
y
_
sin
_
zn
3
L
z
_
. (3.52)
Of fundamental importance is the fact that E
n
1
.n
2
.n
3
is the total electron energy, in con-
trast to the previous cases, where the solution for the bound states in a quantum well and
quantum wire gave us only the energy spectrum associated with the transverse conne-
ment. Another unique feature is the presence of three discrete quantum numbers, n
1
, n
2
,
and n
3
, resulting straightforwardly fromthe existence of three directions of quantization.
Thus, we obtain three-fold discrete energy levels and wavefunctions localized in all three
dimensions of the quantum box. It is important to stress here that the elimination of a
degree of freedom brings about a quantum number that substitutes for a continuous vari-
able, i.e., a component of momentum (or wavevector) corresponding to that particular
degree of freedom. In a quantum dot of a parallelepipedal shape we have three quantum
52 Wave mechanics
numbers, n
1
, n
2
, and n
3
, that substitute for the three components of the wavevector, k:
k
x
, k
y
, and k
z
.
The discrete spectrum in a quantum box and the lack of free propagation of a particle
in any direction are the main features distinguishing quantum boxes from quantum wells
and quantum wires. As is well known, these features are typical for atomic systems as
well.
The similarity with atoms is seen from another example of a potential the spherical
dot. In this case, the potential can be modelled as
V(r) =
_
0. for r R.
. for r R.
(3.53)
It is instructive to compare this result with Eq. (3.20). We assume that the center of the
dot coincides with the center of coordinates; r is the magnitude of the radius-vector, r,
and R is the radius of the spherical dot.
As we have found above, a quantum number (continuous or discrete) can be attributed
to each of the degrees of freedomof the particle. For a problemwith spherical symmetry,
like that given by Eq. (3.53), it is convenient to introduce the following three degrees
of freedom: radial motion (along the vector coordinate) and two rotations, which can
be described by two polar angles, and . As in the case of a parallelepiped-shaped
quantumdot, the three degrees of freedomwill bring about three quantumnumbers. One,
associated with the motion along the radius, r, substitutes for the wavevector component
k
r
, and the two other quantum numbers, associated with rotations, l and m, substitute
for the wavevector components k
and k
.
The electron wavefunction in a spherical quantum dot can be found in the form
n.l.m
(r. . ) = R
n
(r)Y
lm
(. ). (3.54)
where R
n
(r) is the radial function and Y
lm
(. ) are the so-called spherical functions,
which are angle-dependent. We will consider the simplest case of a wavefunction with
spherical symmetry, which corresponds to Y
00
= 1. For spherical symmetry both quan-
tum numbers associated with rotations are equal to zero: l = m = 0; this corresponds to
k
= k
= 0, i.e., the wavevector, k, is parallel to the radius, r. On rewriting the radial

function as R
n
(r) =
n
(r),r we nd the Schr odinger equation for
n
(r) to be similar to
that in Cartesian coordinates:
_
h
--2
2m
2
r
2
E
_
n
(r) = 0.
Thus, as a result of the spherical symmetry the problem reduces formally to a one-
dimensional equation. For the case in which there is an innitely high barrier surrounding
the quantum dot, the wavefunction vanishes for r > R and it follows that R
n
(r) = 0 for
r > R. One can obtain the solutions of this equation as
R
n
(r) =
A sin(k
w
r)
r
. (3.55)
where k
w
=
2m
0
E,h
--
at r - R, R
n
(r) = 0 at r > R, and Ais an arbitrary constant. The
condition that R
n
(r) = 0 for r > R leads to the result that sin(k
w
R) = 0. This equation
2 2
L
z
V
b
i
in
i
r
i
t
L 0
e
Figure 3.13 Transmission of electrons through the barrier region.
is the same as Eq. (3.26) but with KL replaced by k
w
R. So, k
w
is quantized: k
w
= n,R
(compare this with Eq. (3.27) and its analogy with Eq. (3.28)) One can obtain
E
n.0.0
=
h
--2
2
n
2
2mR
2
. n = 1. 2. . . .. (3.56)
where zero indexes show that the values of l and m are equal to zero. This series of levels
is the same as for the one-dimensional quantum well with innitely high barriers.
Quantum reection, transmission and tunneling effects
Now we consider quantum effects arising for continuous energy spectra of particles.
Consider the simplest potential in the form of a rectangular barrier, as shown in
Fig. 3.13:
V(z) =
_
V
b
. for [z[ L,2.
0. for [z[ > L,2.
(3.57)
where V
b
, the barrier height, is greater than zero, and L is the barrier width. In such a
potential, the particle is not conned. The problem should be formulated as a scattering
problem: an incident particle moves, for example, from the left (from z ), then it
can be reected back from the barrier or transmitted through the barrier into the region
z > L,2. The main goal of the analysis is to nd the probabilities of reection and
transmission. It is convenient to dene these quantities as the ratios of reected, i
r
, and
transmitted, i
t
, particle uxes to the incident ux i
in
:
R() =
i
r
i
in
and T() =
i
t
i
in
. (3.58)
The uxes should be calculated according to the denition of Eq. (3.10).
There exist two distinct energy intervals, such that the behavior of the particle is
expected to exhibit different characteristics depending on which energy interval the
particles energy is in: 0 - - V
b
and > V
b
. In classical physics, for the rst energy
interval, obviously the reection coefcient is R = 1, while the transmission is T = 0;
for the second energy interval R = 0 and T = 1.
54 Wave mechanics
In wave mechanics, we have to solve Eq. (3.18) with the potential (3.57). The solutions
are
(z) =
_
_
_
e
ikz
re
ikz
. z L,2.
ae
z
be
z
. L,2 z L,2.
t e
ikz
. z L,2.
(3.59)
To the left of the barrier we introduce the incident wave with a unit magnitude, e
ikz
and
a reected wave, re
ikz
. To the right of the barrier there is only a transmitted wave, t e
ikz
.
Then, the exponential factors are
k =
2m
0
h
--
and =
_
2m
0
(V
b
)
h
--
.
For - V
b
, is a real number, whereas for > V
b
, is an imaginary number. The
parameters r. a. b, and t are still arbitrary functions of , which we nd by matching
these solutions and their derivatives at the walls of the barrier, z = L,2. The uxes at
z can be calculated in terms of r and t : i
in
= 1. i
r
= [r[
2
, and i
t
= [t [
2
. Omitting
the procedure for calculation of r and t , we write down the obvious relationship
R T = 1.
and the two different results for the two energy intervals are as follows:
T =
1
1
_
k
2
2
2k
_
2
sinh
2
(2L)
. - V
b
. (3.60)
T =
1
1
_
k
2
K
2
2kK
_
2
sin
2
(2KL)
. > V
b
. (3.61)
In the above formulas we used K
2
=
2
.
Let us consider rst the case of an energy less than the barrier height, i.e., Eq. (3.60).
Classical results can be obtained formally at the limit of zero Plancks constant, h
--
. Indeed,
when h
--
0, we obtain , sinh(2L) and T 0. That is, no transmission
through the barrier occurs, as expected. In reality, this classical result is realized for
high and wide barriers (V
b
. L ). However, at nite values of these parameters, we
always obtain a nite probability of particle transmission through the barrier, which is,
as we discussed previously, the tunneling effect. For L 1, sinh(2L) =
1
2
e
2L
, the
second term in the denominator of Eq. (3.60) predominates and we can approximate the
formula as
T
16k
2
2
(k
2
2
)
2
e
4L
. (3.62)
Thus, the tunneling effect is determined primarily by the exponential factor. According
to Eq. (3.59) for the wavefunction, the probability of nding the particle under the barrier
is also exponentially small.
For the second case, corresponding to an energy greater than the barrier height, when
classical physics gives strictly T = 1, from the quantum-mechanical equation (3.61), we
0.0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Barrier height
10
0
.
3

e
V
(a)
e (eV)
T
(b)
Barrier height
20
0
.
3

e
V
T
e (eV)
0.0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Figure 3.14 Dependences of the transmission coefcient, T, on electron incident energy, , for
various thicknesses of the barrier, L: (a) L = 10

A and (b) L = 20

A. The height of the barrier
V
b
= 0.3 eV.
nd that T reaches 1 only at sin(2KL) = 0, i.e., at certain resonance energies, when
2KL = n, withn beinganinteger. Otherwise, T - 1withminima at 2KL = (n
1
2
).
All of this means that there exists a reection of the particle even at large energy.
Figures 3.14(a) and 3.14(b) illustrate the transmission coefcient for the barrier shown in
the inset. Both energy intervals, 0 - - V
b
and > V
b
, are presented. The transmission
coefcient for - V
b
is nite due to tunneling, but it is small; whereas for > V
b
the
coefcient is close to 1. Oscillations in T with energy are explained by another quantum
effect overbarrier reection of the particles.
Concluding this analysis, we emphasize the great signicance of quantum effects
in continuous energy spectra, and that these effects determine the basic electrical and
optical properties of nanostructures.
3.4 Atoms and atomic orbitals
It is instructive to compare the simple model of a spherically symmetric potential of
Eq. (3.53) and the simplest hydrogen atom. This atom consists of the positive proton
(nucleus) and the negatively charged electron, which interact according to the Coulomb
law:
V(r) =
1
4c
0
e
2
r
. (3.63)
where c
0
is the permittivity of free space, e is the elementary electrical charge, and r is
the distance between proton and electron. The negative sign in Eq. (3.63) indicates that
the electron and the proton are attracted to each other.
The potential of Eq. (3.63), shown in Fig. 3.15, has some similarities to that of a
quantumdot with impenetrable walls. This similarity allows us to suppose that the energy
of the electron in the hydrogen atomwould be quantized. There are two major differences
56 Wave mechanics

0
5 10 15 20 0 5 10 15 20
n = 2
E
(
e
V
)
r (A)
n = 1
n = 3
5
15
10
Figure 3.15 Eigenenergies of an electron in a hydrogen-atom potential well.

between these potentials: (i) the proles of the potentials (compare Eqs. (3.53) and (3.63))
and (ii) the reference points for zero energy. Namely, the zero energy was located in the
quantum dot and the electron should have innite energy to escape into the potential
walls, whereas in the hydrogen atom the energy of a free electron is taken to be zero at
r = and the energy at r = 0 is equal to . For further analysis of the hydrogen-
atom model, the free-electron mass, m
0
, is used in this section because the motion of
nuclei can be neglected due to the fact that the mass of the nucleus is almost 2000 times
greater than the free-electron mass.
The ground state of the hydrogen atom may be described by the so-called radially
symmetric function, (r):
(r) =
1
r
0
e
r,r
0
. (3.64)
where r
0
is the characteristic radius of the ground state. It can be estimated as follows.
Let us evaluate the kinetic energy of an electron conned in the sphere of radius r
0
by
using the uncertainty conditions of Eq. (2.47): p h
--
,r
0
; then, the kinetic energy can be
evaluated as
p
2
2m
0
h
--2
2m
0
r
2
0
. (3.65)
Thus, the total electron energy, H, at r = r
0
is
H(r
0
) = V(r
0
)
p
2
2m
0
=
1
4c
0
e
2
r
0
h
--2
2m
0
r
2
0
. (3.66)
To nd the r
0
that corresponds to the minimum of the total energy, H, we nd the rst
derivative of H(r
0
) with respect to r
0
and equate the result to zero:
dH(r
0
)
dr
0
=
1
4c
0
e
2
r
0
2

2h
--2
2m
0
r
3
0
= 0. (3.67)
The solution of Eq. (3.67) gives the formula for r
0
:
r
0
=
4c
0
h
--2
m
0
e
2
. (3.68)
which is called the Bohr radius. Now we can nd the energy of the ground state E
1
:
E
1
=
1
4c
0
e
2
2r
0
=
m
0
e
4
(4c
0
)
2
2h
--2
. (3.69)
Numerical estimates give the following values for the energy and the radius of the
ground state of the hydrogen atom: E
1
= 13.5 eVand r
0
= 0.529

A(1

A = 10
10
m =
0.1 nm).
An arbitrary atomic state is characterized by the following three quantum numbers,
n, l, and m:
principal number n = 1. 2. 3. . . .;
orbital number l = 0. 1. 2. . . .. n 1; or l = s. p. d. . . .;
magnetic number m = 0. 1. 2. . . .. l .
The series of energy levels for the hydrogen atom is the following: 1s, 2s, 2p, 3s, 3p,
3d. . . . The ground-state wavefunction of Eq. (3.64) corresponds to the case of n = 1
and l = m = 0. This wavefunction has the eigenenergy E
1
of Eq. (3.69). The general
expression for the nth energy level of the hydrogen atom is given by
E
n
=
E
1
n
2
. (3.70)
Importantly, the energy spectrum of the hydrogen atom depends on only one discrete
number n and does not depend on the orbital and magnetic numbers l and m. This means
that some states of the atomcan have the same energy. This situation is called degeneracy
of the states.
Results from a number of detailed optical experiments have supported this classi-
cation of energy spectra of hydrogen atoms and conrmed the energy values with high
accuracy. However, it turns out that nature is more complicated. Additional non-optical
experiments revealed a new property of quantum systems, which has no analogue in
classical physics.
Indeed, in classical physics a particle with angular momentum l can have a magnetic
dipole moment M= l with the coefcient called the gyromagnetic ratio. The latter
parameter is associated with the fundamental parameters of the electrons:
=
e
2m
0
c
. (3.71)
where e, m
0
, and c are the elementary charge, free-electron mass, and speed of light,
respectively. If an external magnetic eld, B, is applied, an additional potential energy
58 Wave mechanics
of the particle in the magnetic eld arises:
U
M
= BM. (3.72)
Let the magnetic eld be nonuniform, say it depends on the z-coordinate. Then the force
acting on the particle will be
f
z
=
U
M
z
= M
z
B
z
. (3.73)
As a result, a beam of atoms with magnetic moments will split into components cor-
responding to different values of M
z
. In special experiments, known as SternGerlach
experiments, with hydrogen atoms in the ground s state, when l = m = 0 and thus
M
z
= 0, a splitting of the beam into two symmetrically deected components has been
detected. It provided evidence of the existence of a dipole magnetic moment, which we
have not taken into account yet. It has been assumed that the electron has an intrinsic
angular momentum which is responsible for the results of the experiment. This intrin-
sic angular momentum S is called spin. The projection of the spin, S
z
, of an electron
in the magnetic eld, B, takes on the half-integer values
1
2
and
1
2
. The spin effect
explains also the results of a number of other experiments with electrons. Moreover, it
has been found similarly that some spin characteristic can be attributed to any particle.
For example, photons light quanta have a spin equal to 1.
Thus, the complete description of an energy state of the hydrogen atomshould include
additionally the spin number S
z
. Now the series of states including the spin degener-
acy can be written down as 1s(2). 2s(2). 2p(6). 3s(2). 3p(6). 3d(10). . . . The numbers in
brackets indicate the degeneracies of the states. For example, the nomenclature 2p(6)
describes six states of 2p type (l = 1) with the same energy, but with different spins (
1
2
)
and different magnetic numbers (m = 1. 0. 1).
Spin is a very important characteristic. It is responsible for a number of quantum
effects. Moreover, it determines the character of the population of different energy levels
by particles: in a system consisting of identical electrons only two (with different spins)
or fewer electrons can be found in the same state at the same time. Since spin is included
among the quantum numbers, two electrons with opposite spins and all other quantum
numbers being the same should be considered as electrons in different quantum states.
The Pauli exclusion principle can be reformulated as follows: no two electrons can be
in the same state. This is the so-called Pauli exclusion principle. This is one more quan-
tum principle. The Pauli exclusion principle is especially important for many-electron
systems.
The classication of energy levels used for the hydrogen atoms may be applied to
many-electron atoms. In the case of many-electron atoms, we should take the nuclear
charge as Ze, where Z is the atomic number in the Periodic Table of elements (Fig. 3.16),
and take into account the Pauli exclusion principle. The electronic structure of complex
atoms can be understood in terms of the lling up of energy levels. One can introduce
internal energy shells and external energy shells (or valence energy states). The latter
are lled up by the outermost valence electrons and play the major role in determining
the chemical behavior of complex atoms.
Transition Metals
Lanthanides
Actinides
1
2
3 4 5 6 7 8 9 10 11 12
13 14 15 16
17 18
1
3
11
19
37
55
87 88 103 104 105 106 107 108 109 110 111 112
132.9
(223.0) (226.0) (262.1) (261.1) (262.1) (263.1) (264.1) (265.1) (266.1) (269.1) (272.1) (277.1)
137.3 175.0 178.5 180.9 183.8 186.2 190.2 192.2 195.1 197.0 200.6 204.4 207.2 209.0 (209.0) (210.0) (222.0)
56 71 72
38 39 40 41
73 74 75 76 77 78 79 80 81 82 83 84 85 86
42 43 44 45 46 47 48 49 50
39.10 40.08 44.96 47.87 50.94 52.00 54.94 55.85 58.93 58.69 63.55 65.39 69.72
20 21 22 23 24 25 26 27 28 29 30
5 6 7 8 9
1 2
10
10.81 12.01 14.01 16.00 19.00 20.18
1.008 4.003
31
12
22.99 24.30
4
H
Li Be
Na
K Ca Sc Ti V Cr Mn Fe Co Ni Cu Ga Zn
B C N
P S Cl Ar
O F Ne
He H
14 15 16 17 18 13
28.09 26.98 30.97 32.06 35.45 39.95
Si Al
32 33
51 52
72.61 74.92
121.8 127.6
Ge As
34
78.96
Se
35
79.90
Br
36
83.80
Kr
Sb Te
53
126.9
I
54
131.3
Xe Sr Rb
Cs
Fr Ra Lr
57
89 90 91 92 93 94 95 96 97 98 99 100 101 102
138.9
(227.0) 232.0 231.0 238.0 (237.0) (244.1) (243.1) (247.1) (247.1) (251.1) (252.1) (257.1) (258.1) (259.1)
La
Ac Th Pa U Np Pu Am Cm Bk Cf Es Fm Md No
58
140.1
La
59 60 61 62 63 64 65 66 67 68 69 70
140.9 144.2 (144.9) 150.4 152.0 157.2 158.9 162.5 164.0 167.3 168.9 173.0
Pr Nd Pm Sm Eu Gd Tb Dy Ho Er Tm Yb
Rf Db Sg Bh Hs Mt
Ba Lu Hf Ta W Os Ir Pt Au Hg Tl Pb Pi Po At Rn Re
Y Zr Nb Mo Tc Ru Rh Pd Ag Cd In Sn
Mg
1.008
6.941 9.012
85.47 87.62 88.91 91.22 92.91 95.94 (97.91) 101.1 102.9 106.4 107.9 112.4 114.8 118.7
Figure 3.16 The Periodic Table of elements.
While the energy levels of the hydrogen atom exhibit some degree of degeneracy,
i.e., they are independent of the angular momentum described by the orbital number l
and its projection described by the magnetic number m, in many-electron atoms, the
energies depend not only on the principal quantum number n, but also on the orbital
number l. Therefore, the electrons can be subdivided into subshells corresponding to
different orbital numbers l. The spectroscopic notations for these subshells are the same
as above: s. p. d. f. . . . For a given l there are 2l 1 values of m and two values of
the spin S
z
=
1
2
. Thus, the total number of electrons that can be placed in subshell l is
2(2l 1). Periods of the Periodic Table of elements are constructed according to lling
of the shells. Hydrogen, H, is the rst element of the table. It has one electron occupying
the 1s state, i.e., l = 0. The next atom is helium, He. It has two electrons, which occupy
the lowest s states in the same shell, but have opposite spins; this is referred to as the 1s
2
electron conguration. So, the 1s shell is completely lled by two electrons.
The next period of the table begins with lithium, Li. It has three electrons, with the
third electron sitting on the 2s-type level. Thus, we may immediately represent the
electron conguration of Li as 1s
2
2s
1
. After lithium follows beryllium, Be, with the
conguration 1s
2
2s
2
. In the next element boron, B, the fth electron begins to ll up
the next subshell: 1s
2
2s
2
2p
1
. For the p shell, we have l = 1 and m = 1. 0. 1. Thus,
the valence electron in Be can take three values of the magnetic quantum number, m,
and two values of the spin, S
z
. In general, we may put six electrons in the p shell, and
so on. For nanoelectronics the elements of group IV and groups III and V that con-
stitute semiconductor crystals are important: silicon (Si, whose atomic number, and
number of electrons, is 14), germanium (Ge, atomic number 32), gallium (Ga, atomic
60 Wave mechanics
nucleus nucleus
AB
AB
A B
A B
Antibonding orbital
AB
Bonding orbital
AB
-
+
+ +
(b)
(c)
(a)
Lower level
Higher level
Atomic orbitals
A
and
B
with equal energies of atoms
A and B
y
y y
y y
y
y
y
Figure 3.17 Formation of bonding and antibonding atomic orbitals (two atoms).
number 31), arsenic (As, atomic number 33), etc. Their electron congurations are as
follows:
group IV: Si : 1s
2
2s
2
2p
6
3s
2
3p
2
. Ge: 1s
2
2s
2
2p
6
3s
2
3p
6
3d
10
4s
2
4p
2
;
group III: Ga: 1s
2
2s
2
2p
6
3s
2
3p
6
3d
10
4s
2
4p
1
; (3.74)
group V: As: 1s
2
2s
2
2p
6
3s
2
3p
6
3d
10
4s
2
4p
3
.
One can see that, both for Si and for Ge, the outer valence electron congurations are
very similar (the same number of electrons in the same p state). Thus, we can expect that
their chemical and physical properties are also similar.
Atoms with completed subshells are stable and chemically inert. Examples are helium,
He, with the conguration 1s
2
2s
2
; neon, Ne, with the conguration 1s
2
2s
2
2p
6
; and argon,
Ar, with the conguration 1s
2
2s
2
2p
6
3d
10
4s
2
4p
6
.
In general, atoms with not completely lled-up subshells, that is, those which have
valence electrons, are expected to be chemically active, i.e., they can form chemical
bonds. Coupling between atoms in molecules and solids is determined by the type of
the electron wavefunctions. Often, these wavefunctions are called atomic orbitals. For
example, the wavefunction of a two-atom molecule should be a combination (a hybrid)
of two orbitals. Such a hybrid can be either a symmetric function, or an antisymmetric
one. Let this hybrid be a symmetric function (constructive interference of the orbitals), as
shown in Fig. 3.17(c). Then, there is a nite probability of nding the electron between
the nuclei. The presence of electrons between two nuclei, Aand B, attracts both nuclei and
keeps them together. The corresponding state is called a bonding state of two atoms and
the corresponding wavefunction,
AB
, is called the bonding orbital. The antisymmetric
hybrid (destructive interference of the orbitals) does not bond atoms and is called an
antibonding state, see Fig. 3.17(b). The simplest example of the bonding effect is given
by the two-atom hydrogen molecule: the bonding state constructed from s states of
the hydrogen atoms corresponds to an energy lower than the energy of two uncoupled
(c)
(d)
x
x
z
z
y
y
p
z
orbital
p
x
orbital
(a) (b)
x
z
y
s orbital p
y
orbital
+
-
x
z
y
+

y
+
y
+
-
+
+
-
-
Figure 3.18 Atomic orbitals: (a) s, (b) p
y
, (c) p
z
, and (d) p
x
.
hydrogen atoms. This results in a stable H
2
molecule. The antibonding state corresponds
to an energy greater than the energy of two free hydrogen atoms. In the antibonding state
the atoms repel each other.
Actually, the formation of bonding and antibonding states of two hydrogen atoms is
a particular example of a more general and very important result: when two identical
atoms approach each other closely, their energies change and, instead of a single atomic
(degenerate) energy level, two levels arise, as illustrated in Figs. 3.17(b) and 3.17(c). This
is the so-called energy splitting. Thus, bonding and antibonding orbitals correspond to
these two levels. Three atoms drawn together are characterized by three close levels, etc.
Now, consider in detail atoms of group IV. As presented above, for these atoms the
outer shell has four electrons: two in s states and two in p states, i.e., their conguration
can be thought as a core plus the s
2
p
2
shell. Thus, atoms of group IVcan formfour bonds
with other atoms in molecules or solids (one bond for one electron). The wavefunctions
of these outer-shell states have very different spatial congurations. Figure 3.18 illus-
trates the wavefunctions (orbitals) for an s state and for p states with different angular
momentumprojections m. Thus, the s orbital is spherically symmetric without any angu-
lar dependence. The 3p orbitals are anisotropic and can be considered as perpendicular
to each other. The hybrid orbitals forming bonds with other atoms have two electrons
with opposite spins. In fact, because of interaction between electrons the s and p orbitals
overlap signicantly and realistic bonds are formed as linear combinations of s and p
orbitals, which are called sp
3
hybridized orbitals. Importantly, when the three directed p
orbitals are hybridized with the s orbital, the negative parts of the p orbitals are almost
cancelled out, so that mainly positive parts remain, as shown in Fig. 3.19(a). Thus, the
hybridized orbitals present directed bonds in space, as shown in Fig. 3.19(b). These
directed bonds of atoms are responsible for the particular crystal structures of elements
62 Wave mechanics
x
z
y
s orbital
x
z
y
p
y
orbital
+
+
-
+
sp
3
orbital
x
z
-
(a)
(b)
+
y
+
+
+
+
Figure 3.19 Formation of hybridized sp
3
orbitals (a) and four sp
3
hybridized orbitals in Si (b).
in group IV. In the next chapter, we will analyze these crystals using the denitions
discussed above.
3.5 Closing remarks
In this chapter we have presented basic denitions and equations of quantum (wave)
mechanics. We found that two cases of particle motion should be distinguished: (i)
quantized motion always occurs with in a nite volume (for example, in a potential
well); (ii) free motion occurs in an innite space (practically, in a large volume). We
discussed the differences in particle behavior between these two cases.
We analyzed simple instructive examples that illustrate the main qualitative fea-
tures of quantum physics. Then, we briey discussed the electronic properties of
atoms, which we should use for the analysis of nanoelectronic systems. Many of the
basic quantum-mechanical concepts introduced above will be applied in the follow-
ing chapters to study semiconductor nanostructures and the principles of operation of
nanodevices.
For those who want to look deeper into the basic principles of quantum physics we
recommend the following textbooks:
L. Schiff, Quantum Mechanics (New York, McGraw-Hill, 1968).
D. Saxon, Elementary Quantum Mechanics (San Francisco, CA, Holden-Day, Inc.,
1968).
R. P. Feynman, R. B. Leighton, and M. Sands, Lectures on Physics, vol. 3 (New York,
Addison-Wesley, 1964).
The problems presented below illustrate the main denitions given in this chapter and
allow one to practice with the simplest quantum-mechanical examples.
3.6 Problems 63
3.6 Problems
1. Suppose that a particle is placed into a region of space of volume V. Using the
normalization condition of Eq. (3.9), nd the amplitude A of the plane wave (2.43).
2. Assume that the wavefunction is a superposition of two plane waves of the same
energy: + =
1
2
with
1.2
= A
1.2
e
i(k
1.2
rOt )
. Calculate the probability density of
nding a particle at the point r, i.e., calculate [+(r)[
2
. Discuss the interference effect and
its dependence on the wavevectors k
1
and k
2
.
3. Assume that an electron with mass m
0
= 9.8 10
31
kg is placed in a quantumwell
with two impenetrable walls and that the distance between the walls is L = 10
6
cm.
Calculate the three lowest subband bottom energies:
1
.
2
, and
3
. For these stationary
states, nd the probability density of nding the electron at the middle of the well at
z = 0.
4. Consider an electron, which is placed into a quantum well with width L = 10
6
cm
and barrier height V
b
= 300 meV. Calculate the lowest energy level and compare your
result with results for the previous problem. Since the probability density for nding the
particle at a point z is [(z)[
2
, the probability of nding the particle in the classically
forbidden regions is
P = 2
_

L,2
[(z)[
2
dz.
Use the solution for the lowest energy level to calculate P. Find the number of discrete
energy levels in such a quantum well.
5. The hydrogen molecule, H
2
, has the frequency of oscillations = 8.2 10
14
s
1
.
The reduced mass, which determines the relative displacement of hydrogen atoms, is
m = 0.84 10
27
kg, which is half of the mass of a hydrogen atom. Using the model
of a harmonic oscillator, estimate the characteristic displacement, z
0
, during vibrations
of this molecule.
6. In classical physics, for a system under equilibrium conditions, the equipartition
principle is valid. According to this principle the average energy of any oscillator is
k
B
T, where k
B
= 1.38 10
23
J K
1
is Boltzmanns constant and T is the ambient
temperature. Calculate the energy of zero-point vibrations of an electromagnetic wave
of wavelength 10 m. Calculate the average number of photons, N
q.
, that will be excited
with this wavelength if the ambient temperature is equal to 500 K.
7. Consider two quantumobjects, a cubic box and a spherical dot surrounded by impen-
etrable walls. Suppose that the volumes of the classically allowed regions are equal.
Compare the ground-state levels for the box and the dot.
64 Wave mechanics
8. By using approximate formulae for the transmission coefcient, calculate the prob-
ability of tunneling of the conduction electron through an AlGaAs layer of thickness
20

Aembedded within a GaAs matrix, for a barrier height equal to 0.3 eV, with the
electron mass in GaAs m = 0.067m
0
, where m
0
= 9.8 10
31
kg is the mass of a free
electron.
9. One of the experimental facts which constitutes one of the fundamentals of wave
mechanics is the observation of discrete optical spectra (spectral lines) of emission of
hydrogen atoms. Using the formulae for the energy levels of a hydrogen atom, E
n
,
calculate a general expression for the emission frequencies. Find the wavelengths for
several lines of the so-called Lyman series, for which one of the states participating in
emission is the ground state 1: 1 2. 1 3. 1 4. . . .
4 Materials for nanoelectronics
4.1 Introduction
After the previous introduction to the general properties of particles and waves on the
nanoscale, we shall now overview the basic materials which are exploited in nanoelec-
tronics. As discussed in Chapter 1, electronics and optoelectronics primarily exploit the
electrical and optical properties of solid-state materials. The simplest and most intuitive
classication of solids distinguishes between dielectrics, i.e., non-conducting materials,
and metals, i.e., good conducting materials. Semiconductors occupy a place in between
these two classes: semiconductor materials are conducting and optically active materials
with electrical and optical properties varying over a wide range. Semiconductors are the
principal candidates for use in nanoelectronic structures because they allow great exi-
bility in the control of the electrical and optical properties and functions of nanoelectronic
devices.
The semiconductors exploitedinmicroelectronics are, ingeneral, crystalline materials.
Through proper regimes of growth, subsequent modications and processing, doping by
impurities, etc., one can fabricate nanostructures and nanodevices starting from these
bulk-like materials.
Other physical objects that demonstrate promising properties for nanoelectronics were
discovered recently, for example carbon nanotubes. These wire-like and extended objects
are of a few nanometers in cross-section. They can be produced with good control of
their basic properties; in particular, they can be fabricated as either semiconductors or
metals. Various types of processing techniques have been shown to be viable for the
fabrication of electronic nanodevices from carbon nanotubes.
In this chapter, we consider various materials that have applications for nanoelectron-
ics. We start with the classication of dielectrics, semiconductors, and metals. Then,
we dene electron energy spectra, which determine the basic properties of the electrons
in crystals. For nanoelectronics, a critical issue is the engineering of electron spectra,
which can be realized in heterostructures. Thus, we analyze basic types of semiconductor
heterostructures. Finally, we briey describe organic semiconductors and carbon-based
nanomaterials, among them carbon nanotubes and such nano-objects as fullerenes.
E
1
E
n
e
r
g
y

b
a
n
d
Large number of atoms
(a)
E
1
E
1
E
1
E
1
(b)
Energy level E
3
Overlap
Energy level E
2
Energy level E
1
Energy bands E
1
, E
2
, and E
3
3
2
1
Figure 4.1 (a) Formation of an energy band. The ground energy level E
1
of a single atom evolves
into an energy band when many identical atoms are interacting with each other. (b) Energy
bands 1, 2, and 3 that correspond to single-atom energy levels E
1
, E
2
, and E
3
, respectively.
4.2 Semiconductors
In every solid, electrons can be characterized in terms of their energy levels. In crystals,
allowed electron energies typically have an energy band structure that may be understood
as follows. In Section 3.4, we pointed out that, if two atoms, each with the same energy,
come close to each other, the composite two-atom system is characterized by two close
energy levels. Similarly, for a system of N atoms, every energy level of the isolated atom
is split into N closely spaced levels. This assembly of close levels may be considered as
an energy band. Figure 4.1(a) illustrates schematically the formation of such an energy
band from a single atomic level. Since a single atom has a series of energy levels, the
electron energies in a crystal constitute a series of energy bands that may be separated
by energy gaps, or may overlap as illustrated by Fig. 4.1(b). As soon as the energy bands
are formed, the electrons should be thought of as collectivized: they can no longer be
attributed to specic atoms, since the energy bands characterize the whole system of N
atoms.
Acrucial point is the lling of the bands by electrons. We shall use the Pauli exclusion
principle, which we considered in the previous chapter. That is, no two electrons can be
in the same state. It is possible, as an example, for two electrons to be in the same energy
state, but these electrons must be in different spin states; thus the electrons are in fact in
different overall states. Under equilibrium conditions and at low ambient temperature,
the lowest energy levels should be populated. As we will see later, the most important
electrons are those inthe upper populatedbands. Then, principally, we obtaintwopossible
cases.
First, all bands are completely lled and the lled bands are separated from the upper
(empty) bands by an energy gap. This is the case illustrated by Figs. 4.2(a) and 4.2(b) for
dielectrics (insulators) with bandgaps E
g
>5 eVand for semiconductors with bandgaps
E
g
- 5 eV, respectively. Actually, there is no difference between lling up energy bands
for a dielectric (insulator) and for a semiconductor. The difference is in the energy gap
4.2 Semiconductors 67
(a)
Empty
Filled
E
g
Empty
E
g
Filled
Partially Filled
Overlap
Empty
Filled
(b)
(d)
(c)
Filled
Figure 4.2 Energy bands in the cases of a dielectric, a semiconductor, and a metal. (a) The case
of a dielectric: lled valence band and empty conduction band; E
g
> 5 eV. (b) The case of a
semiconductor: lled valence and empty conduction band (at low temperature); E
g
- 5 eV.
(c) Electrons in the partially lled band may gain energy from an electric eld and, as a result,
transfer to empty (free) levels of higher energy, thus exhibiting electric conductivity.
Accordingly, a structure with a partially lled band corresponds to a metal. (d) For overlapping
bands the available electrons ll states in both bands.
between the upper lled band and the next empty band: for semiconductors this energy
gap is much smaller than it is for dielectrics, as is illustrated in Figs. 4.2(a) and 4.2(b).
This upper band is empty, at least at low temperature. Later, we will study the electronic
band structure of semiconductor materials in more detail.
Second, the upper bands that contain electrons are not completely lled as in
Figs. 4.2(c) and 4.2(d); this case corresponds to a metal. Indeed, to exhibit conduc-
tivity under an applied electric eld, an electron should experience acceleration and a
gain in energy. That is, an electron should be able to gain a small amount of energy
and be transferred to close, but higher, energy levels. If all energy levels (a whole band)
are already lled up, the electron can not participate in conduction processes. This is
the case of a dielectric (an insulator). In contrast, if the energy band is not completely
lled and there are empty energy levels available for the electrons, in an electric eld
the electron can move to upper levels and gain energy in a process that corresponds to
electrical conductivity. This is the case of a metal.
Semiconductors include elements fromthe central portion of the Periodic Table of ele-
ments columns II to VI, as shown in Table 4.1. At the center of Table 4.1 is silicon, Si,
the backbone material of modern electronics. Silicon plays the central role in electronics
Table 4.1 The central portion of the Periodic Table of elements
Group II Group III Group IV Group V Group VI
Be B C N O
Mg Al Si P S
Zn Ga Ge As Se
Cd In Sn Sb Te
Hg Tl Pb Bi Po
just as steel plays a dominant role in metallurgy. Below Si is germanium, Ge. Nowadays,
Ge is rarely used by itself; however, GeSi alloys play an increasingly important role in
todays electronics technology. Besides the elemental materials, contemporary electron-
ics also uses combinations of elements from group III and group V, and combinations
of elements of group II and group VI, as well as some more complicated combinations.
These combinations are called compound semiconductors. By combining each element
fromgroup III with N, P, As, Sb, and Bi fromgroup V, 25 different IIIVcompounds can
be formed. The most widely used compound semiconductor is GaAs (gallium arsenide)
and all IIIV semiconductors are used to fabricate so-called heterostructures. A het-
erostructure is made of two different materials with a heterojunction boundary between
them. The specic choice of heterostructure depends on the application.
Two or more compounds may be used to formalloys. Acommon example is aluminum
gallium arsenide, Al
x
Ga
1x
As, where x is the fraction of group III sites in the crystal
occupied by Al atoms, and 1 x is the fraction of group III sites occupied by Ga atoms.
Hence, nowwe have, not just 25 discrete compounds, but a continuous range of materials.
As with the IIIV compounds, every element shown in column II may be used together
with every element in column VI to create IIVI compounds, and again, by combining
more than two of these elements, it is possible to create a continuous range of materials.
As a result, it is possible to make compositionally different IVIV, IIIV, and IIVI
compounds.
4.3 Crystal lattices: bonding in crystals
We start with the denition of crystals. A crystal is a solid in which the constituent
atoms are arranged in a certain periodic fashion. That is, one can introduce a basic
arrangement of atoms that is repeated throughout the entire solid. In other words, a crystal
is characterized by a strictly periodic internal structure. Not all solids are crystals. In
Fig. 4.3, for comparison, we present a crystalline solid (a), a solid without any periodicity
(a so-called amorphous solid) (b), and a solid in which only small regions are of a
single-crystal material (a so-called polycrystalline solid) (c). As might be expected,
crystalline materials can be the most perfect and controllable materials. Before studying
periodic arrangements of atoms in crystals, we shall discuss different types of bonding in
crystals.
4.3 Crystal lattices 69
(a) Crystalline
(b) Amorphous
(c) Polycrystalline
Figure 4.3 Three types of solids: (a) ordered crystalline and (b) amorphous materials are
illustrated by microscopic views of the atoms, whereas (c) a polycrystalline structure is
illustrated by a more macroscopic view of adjacent single-crystalline regions, each of which has
a crystalline structure as in (a).
Ionic crystals
Ionic crystals are made up of positive and negative ions. The ionic bond results primarily
fromattractive mutual electrostatic interactionof neighboringions withopposite charges.
There is also a repulsive interaction with other neighbors of the same charge. Attraction
and repulsion together result in a balancing of forces that leads to the atoms being in
stable equilibrium positions in such an ionic crystal. As for the electronic conguration
in a crystal, it corresponds to a closed (completely lled) outer electronic shell. A good
example of an ionic crystal is NaCl (salt). Neutral sodium, Na, and chlorine, Cl, atoms
have the congurations Na
11
(1s
2
2s
2
2p
6
3s
1
) and Cl
17
(1s
2
2s
2
2p
6
3s
2
3p
5
), respectively.
That is, the Na atom has only one valence electron, while one electron is necessary to
complete the shell in the Cl atom. It turns out that the stable electronic conguration
develops when the Na atom gives one valence electron to the Cl atom. Both of them
become ions, with opposite charges, and the pair has the closed outer-shell conguration
(like inert gases such as helium, He, and neon, Ne). The inner shells are, of course,
completely lled both before and after binding of the two atoms. In general, for all
elements with almost closed shells, there is a tendency to form ionic bonds and ionic
crystals. These crystals are usually dielectrics (insulators).
Covalent crystals
Covalent bonding is typical for atoms with a low level of lling up of the outer shell. An
excellent example is provided by a Si crystal. As we discussed in the previous chapter, the
electron conguration of Si can be represented as core 3s
2
3p
2
. To complete the outer
3s
2
3p
2
shell, a silicon atom in a crystal forms four bonds with four neighboring silicon
atoms. The symmetry of the hybrid sp
3
orbitals dictates that these neighboring atoms
should be situated in the corners of a tetrahedron as shown in Fig. 4.4. Then, the central
Si
Si
Si
Si
Si
Figure 4.4 Four sp
3
-hybrid bonding orbitals in a crystal of Si.
Si atom and each of its nearest-neighbor Si atoms share two electrons. This provides so-
called covalent (chemical) bonds (symmetric combinations of the sp
3
orbitals) in the Si
crystal. The four bonding sp
3
orbitals forman energy band that is completely lled by the
valence electrons. This band is called the valence band. The antisymmetric combination
of the sp
3
orbitals with destructive interference, as in Fig. 3.17, leads to antibonding
states and to the rst empty energy band. The latter band is called the conduction band
of the Si crystal. Not surprisingly, a covalent bond of this type plays a major role for Ge,
which is also from group IV of the Periodic Table of elements.
In fact, both of the types of atomic bonding discussed previously may exist simultane-
ously in a crystal, for example, in IIIV compounds. Indeed, the electron congurations
in Ga and As are core 4s
2
4p
1
(Ga) and core 4s
2
4p
3
(As). When a GaAs crystal is
formed, the As atom gives one valence electron to the Ga atom, which makes them both
ions. The Coulomb interaction of these ions contributes to the ionic bonding in IIIV
compounds. But now each ion has only two 2s electrons and two 2p electrons, which are
not enough to ll the shell completely. Therefore, the rest of the bonding goes through the
formation of the sp
3
-hybridized orbitals. This is the covalent contribution to the crystal
bonding. We can conclude that the IIIVcompounds are materials with mixed bonding
partly ionic and partly covalent.
Metals
Metals, such as Na, K, and Ca, consist of ions regularly situated in space. Each atom
contributes an electron to an electron sea, in which the ions are embedded. The system
as a whole is neutral and stable. The electrons contribute signicantly to the binding
energy of metals.
4.3 Crystal lattices 71
Table 4.2 Binding energies for different types of crystals
Type of crystal coupling Crystal Energy per atom (eV)
Ionic NaCl 7.9
LiF 10.4
Covalent Diamond, C 7.4
Si 3.7
Ge 3.7
Metallic Na 1.1
Fe 4.1
Al 2.4
Molecular and CH
4
0.1
inert-gas crystals Ar 0.8
There exist other bonding mechanisms in solids, in molecular crystals and inert-gas
crystals. However, these crystals have small binding energies, that is, they are not stable
enough to be used widely in electronics. In Table 4.2, for comparison, we present binding
energies per atom for different types of crystals. We can see that ionic and covalent
crystals typically have binding energies in the range 110 eV, whereas molecular and
inert-gas crystals are weakly coupled systems. Metals have intermediate coupling. The
binding energy of a crystal is an important parameter, since it determines the stability of
the crystal, its aging time, the applicability of various treatment processes, etc.
Crystal lattices
Now, we returntothe discussionof crystal periodicity. The periodic arrangement of atoms
(ions) in a crystal forms the lattice. The positions of atoms in the lattice are dened as
the sites. In principle, atoms always perform small-amplitude oscillations around the
sites. However, in many cases we can neglect these small-amplitude oscillations and
think of a crystal as a system of regularly distributed atoms (ions). In such a perfect and
periodic crystal lattice, we can identify a region called a unit cell. Such a unit cell is a
representative of the entire lattice, since the crystal can be built by regular repeats in
space of this element. The smallest unit cell is called the primitive cell of the lattice. The
importance of the unit cell lies in the fact that by studying this representative element one
can analyze a number of properties of the entire crystal. The primitive cell determines
the fundamental characteristics of the crystal, including the basic electronic properties.
One of the most important properties of a perfect crystalline lattice is its translational
symmetry. Translational symmetry is the property of the crystals being carried into
itself under parallel translation in certain directions and for certain distances. For any
three-dimensional lattice it is possible to dene three fundamental noncoplanar primitive
translation vectors (basis vectors) a
1
. a
2
, and a
3
, such that the position of any lattice site
can be dened by the vector R = n
1
a
1
n
2
a
2
n
3
a
3
, where n
1
. n
2
, and n
3
are arbitrary
integers. If we construct the parallelepiped using the basis vectors, a
1
. a
2
, and a
3
, we
obtain just the primitive cell.
a
1
a
2
a
1
a
1
a
2
B
C
A
a
2
Figure 4.5 A two-dimensional lattice. Three unit cells are illustrated by A, B, and C. Two basis
vectors are illustrated by a
1
and a
2
.
a
2
a
1
a
3
(a)
a
2
a
1
a
3
(b)
a
2
a
1
a
3
(c)
Figure 4.6 Three types of cubic lattice: (a) simple, (b) body-centered, and (c) face-centered.
Basis vectors of the three different primitive cells are a
1
, a
2
, and a
3
.
Translational symmetry is illustrated in Fig. 4.5, where for simplicity we present a
two-dimensional lattice of which several unit cells (A, B, and C) are shown. The cell A
obviously has the smallest magnitudes of vectors a
1
and a
2
. Thus, the basis vectors of
the primitive cell are the vectors a
1
and a
2
from the unit cell A. An arbitrary lattice site
is at the point R = n
1
a
1
n
2
a
2
with integers n
1
and n
2
. To visualize the translational
symmetry of this lattice one can start from any point of the lattice and nd all other
equivalent positions in space by just applying translations that are integer multiples of
the basis vectors.
Since many of the crystals used in electronics are of the so-called cubic symmetry,
here we consider briey such cubic lattices. For them, the unit cell may be selected in the
form of a cube. There are three different types of cubic lattices. The simple cubic lattice
has atoms located at each corner of the cube, as shown in Fig. 4.6(a). The body-centered
cubic lattice has an additional atom at the center of the cube, as shown in Fig. 4.6(b).
The third type is the face-centered cubic lattice, which has atoms at the corners and at
the centers of the six faces as depicted in Fig. 4.6(c).
Specically, the basic lattice structure for diamond, C, silicon, Si, and germanium,
Ge, is the so-called diamond lattice. The diamond lattice consists of two face-centered
cubic structures with the second structure being shifted by a quarter of a diagonal of the
a
Figure 4.7 The diamond lattice is of the face-centered type of cubic lattices. The tetrahedral
bonding arrangement of neighboring atoms is clear.
rst cube, or by a distance a
1
,4 a
2
,4 a
3
,4 from the rst face-centered structure;
here vectors a
1
, a
2
, and a
3
are vectors of the rst cube as dened in Fig. 4.6(a). In
Fig. 4.7 the atoms of the rst face-centered structure are shown in black and some atoms
of the second one are gray. Thus, a diamond lattice contains twice as many atoms per
unit volume as does a face-centered cubic lattice. The four nearest-neighbor atoms to
each atom are shown in complementary shading for easier visualization. The parameter
a which characterizes the cubic lattice is the so-called lattice constant. The lattice of
volume V = a
3
consists of eight atoms.
Besides the translations, the crystal symmetry contains other symmetry elements, for
example, specic rotations around high-symmetry axes. In cubic crystals, axes directed
along the basis vectors are equivalent and they are the symmetry axes. It is convenient
to use a system of coordinates built on the basis vectors. Three symmetry axes may be
denoted as [100], [010], and [001]. (Here we have used notations that are common in
crystallography: [100] is a unit vector in the x direction, [010] in the y direction, and [001]
in the z direction. All important symmetry directions of a cube are shown in Fig. 4.8(a).
The directions of the type [110] and [111] are also important crystal directions. It is
evident that these directions are equivalent to the opposite ones, [
10] and [
1], as well
as to analogous ones. In Fig. 4.8(a), we show the symmetry directions of a cubic crystal.
If the crystal is carried into itself on rotation through an angle 2,n about some axis that
passes through the crystal, then this axis is said to be an n-fold axis C
n
. For example, in
a cubic crystal there are three four-fold axes C
4
and four three-fold axes C
3
, as shown in
Fig. 4.8(b). The symmetry elements of the lattice make analysis of crystals properties
much simpler.
4.4 Electron energy bands
Summarizing the above analysis, we conclude that a crystal consists of nuclei and elec-
trons. The valence electrons are collectivized by all nuclei and we can expect them to
0
0
1
010
0
x
z
y
1
0
0
1
0
0
1
1
0
010
1
1
1
110
1
1
0
1
0
1
0
0
1
(a)
x
y
z
C
3
C
3
C
3
C
3
C
4
C
4
C
4
(b)
Figure 4.8 (a) Symmetry directions for cubic crystals; (b) rotational symmetry directions in
cubic crystals.
be relatively weakly coupled to atoms. This allows one to think about a crystal as a
system with two relatively independent subsystems: the atom (ion) subsystem and the
electron subsystem. In this section, we will consider the electron subsystem of crystals.
Actually, in a crystal an electron moves in the electrostatic potential created by posi-
tively charged ions and all other electrons. This potential is frequently referred to as the
crystalline potential, W(r). We will consider one-particle states for electrons in an ideal
crystal and we will present a classication of these states and nd a general form for the
wavefunctions and energies.
An electron in a crystalline potential
For an ideal crystal the crystalline potential is periodic with the period of the crystalline
lattice. Let a
j
, with j = 1. 2. and 3. be the three basis vectors of the lattice that dene the
three primitive translations. The periodicity of the crystalline potential, W(r), implies
that
W
_
r
3
j =1
n
j
a
j
_
= W(r). (4.1)
where r is an arbitrary point of the crystal and n
j
are some integers. The one-particle
wavefunction should satisfy the time-independent Schr odinger equation
H(r) =
_
h
--2
2m
0
2
W(r)
_
(r) = E(r). (4.2)
where m
0
is the free-electron mass and

His the crystalline Hamiltonian. Equation (4.2)
neglects interactions between electrons and that is why the wavefunction, (r), is often
called the one-particle wavefunction.
Because of the potential periodicity of Eq. (4.1), the wavefunctions (r) may be
classied and presented in a special form. To nd this form we introduce the translation
operator,

T
d
, that acts on the coordinate vector r as
T
d
r = r d. d =
3
j =1
n
j
a
j
. (4.3)
On applying this operator to the wavefunction we nd that the function

T
d
(r)
(r d) is also a solution of Eq. (4.2) for the same energy E. Let us assume that
the electron state with energy E is not degenerate. Then, we conclude that the two
wavefunctions (r) and (r d) can differ only by some multiplier, C
d
:
(r d) = C
d
(r). (4.4)
From the normalization condition
_
[(r d)[
2
dr = 1 (4.5)
we obtain [C
d
[
2
= 1. Two different translations d
1
and d
2
should lead to the same result
as the single translation d = d
1
d
2
, i.e., C
d
1
C
d
2
= C
d
1
d
2
. From this result it follows
that C
d
may be represented in an exponential form:
C
d
= e
ikd
= exp
_
ik
3
j =1
n
j
a
j
_
. (4.6)
where k is a constant vector. Thus, fromEq. (4.4) we get the wavefunction in Bloch form,
(r) = e
ikd
(r d) = e
ikr
u
k
(r). (4.7)
where
u
k
(r) = e
ik(rd)
(r d). (4.8)
One can check that the so-called Bloch function u
k
(r) is a periodic function:
u
k
(r d
/
) = u
k
(r). d
/
=
3
j =1
n
j
a
j
.
Therefore, the stationary one-particle wavefunction in a crystalline potential has the form
of a plane wave modulated by the Bloch function with the lattice periodicity. The vector
k is called the wavevector of the electron in the crystal. This wavevector is one of the
quantum numbers of electron states in crystals.
By applying the so-called cyclic boundary conditions to the crystal with a number of
periods N
j
along the direction a
j
,
(r N
j
a
j
) = (r). N
j
. (4.9)
we nd for k
ka
j
N
j
= 2n
j
. n
j
= 1. 2. 3. . . .. N
j
. (4.10)
These allowed quasi-continuum values of k form the so-called rst Brillouin zone of
the crystal. They are just those energy bands which we discussed above by using simple
qualitative considerations. It is important that the symmetry of the Brillouin zone in
k-space is determined by the crystal symmetry.
Let the one-particle energy corresponding to the wavevector k be E = E(k). If the
wavevector changes within the Brillouin zone, one gets a continuum energy band; i.e.,
an electron energy band. At xed k, the Schr odinger equation (4.2) has a number of
solutions in the Bloch form:
.k
(r) =
1
V
e
ikr
u
.k
. (4.11)
where enumerates these solutions and, thus, the energy bands. Owing to the crys-
tals periodicity, the Bloch function can be calculated within a single primitive cell.
In Eq. (4.11) we normalize the wavefunction
.k
for the crystal volume V = NV
0
;
N = N
1
N
2
N
3
and V
0
are the number and volume of the primitive crystal cell,
respectively. From the normalization of the wavefunction
.k
(r) one obtains
1
V
0
_
V
0
[u
.k
[
2
dr = 1. (4.12)
where the integral is calculatedover the primitive cell. This formula allows one toestimate
the order of the value of u
.k
: [u
.k
[ 1.
Thus, through this analysis we have established an extremely important property of
the electrons in crystalline solids: despite the interaction of an electron with atoms and
other electrons, in a perfect lattice the electron behaves much like a free electron. The
electron can be characterized by a wavevector k and, thus, it possesses the momentumh
--
k.
By considering phenomena that have spatial scales much greater than distances between
atoms (ions) in the primitive cell, we may omit the Bloch function u
.k
and describe the
electrons by a wavefunction in the formof the plane wave
k
(r) = A exp(ikr), just as for
a free particle. However, the wavevector changes inside the Brillouin zone in a manner
that is specic for a given crystal and, in general, the energy dispersions E = E
(k) can
differ from that of the free electron considerably.
The holes
According to the discussion given in Section 4.2, some of the energy bands are completely
lled, while the others are almost empty. For our purposes, two of the bands are of great
importance: the upper lled band and the lowest empty band. They are called the valence,
E
v
(k), and the conduction, E
c
(k), band, respectively.
One of the ways to get an electron into the conduction band is to transfer an electron
from the valence band to the conduction band. Thus, for analysis of the valence band, it
is useful to adopt the concept of a hole as a new quasiparticle; i.e., by a hole we refer
to an electron missing from the valence band. These quasiparticles can be introduced
and described on the basis of simple considerations. If the valence band is full, the total
wavevector of all electrons in the valence band is zero:
k
v
=
i
k
i
= 0. (4.13)
where the sum accounts for all occupied valence states. Let us assume that one of the
electrons with wavevector k
e
is removed from the valence band. The total wavevector of
the valence electrons becomes
k
v
=
i
k
i
= k
e
. (4.14)
On the other hand, removing this electron is identical to the creation of a hole in
the valence band. One can attribute the wavevector of Eq. (4.14) to this hole: k
h
= k
e
.
Then, the energy of the valence electrons decreases by the term E
v
(k
e
), and, thus, one can
also attribute the energy E
h
(k
h
) = E
v
(k
e
) to this hole. If the energy band is symmetric;
i.e., E
v
(k) = E
v
(k), we can write for the hole energy
E
h
(k
h
) = E
v
(k
e
) = E
v
(k
e
) = E
v
(k
h
). (4.15)
Thus, we can characterize the hole by a wavevector, k
h
, and an energy, E
h
(k
h
), and
consider the hole as a new quasiparticle created when the electron is removed from
the valence band. In the conduction band, the electron energy, E
c
(k), increases as the
wavevector, k, increases. Conversely, in a valence band, near the maximum energy of
the band the electron energy, E
v
(k), decreases as k increases. However, according to
Eq. (4.15), the hole energy increases with the hole wavevector, k
h
. That is, the hole
behaves as a usual particle. Thus, one can introduce the velocity of the hole, :
h
=
E
h
(k
h
),k
h
, andthenemployNewtons laws, etc. The absence of a negative charge inthe
valence band brought about when an electron is removed makes it possible to characterize
a hole by a positive elementary charge; that is, the holes carry positive electrical charge.
It is worth emphasizing that the similarity between the electrons and holes is not
complete: the holes exist as quasiparticles only in a crystal, whereas the electrons exist
also in other physical media, as well as in the vacuum.
Symmetry of crystals and properties of electron spectra
Usually, the energy dispersion relations, E
(k), are very complex and can be obtained

only numerically in the context of approximate methods.
Fortunately, the Brillouin zone possesses a symmetry which directly reects the sym-
metry of the unit cell of the crystal in coordinate space. If a crystal is mapped into itself
due to transformations in the form of certain rotations around the crystalline axes and
of mirror reections, one can speak about the point symmetry of directions in the crys-
tal. In the Brillouin zone, this symmetry generates several points with high symmetry
with respect to the transformations of the zone in k-space. The extrema of the energy
dispersion E
(k) always coincide with these high-symmetry points. In particular, this

fact allows one to simplify and solve the problem of obtaining electron spectra. Near
extrema, the energy spectra can be approximated by expansion of E
(k) in series with

respect to deviations from the symmetry points. Such an expansion can be characterized
by several constants which dene the reciprocal effective mass tensor:
_
1
m
_
i j
=
_
_
_
m
1
xx
m
1
xy
m
1
xz
m
1
yx
m
1
yy
m
1
yz
m
1
zx
m
1
zy
m
1
zz
_
_
_
. (4.16)
where i and j denote x-, y-, and z-coordinates.
Table 4.3 Symmetry points in group IV semiconductors and III--V compounds
Symmetry Position of extremum in
point k-space Degeneracy
I 0 1
L (,a)[111]. (,a)[
111]. (,a)[1
11]. (,a)[11
1] 4
L (2,a)[100]. (2,a)[010]. (2,a)[001]. [ [ - 1 6
X (2,a)[100]. (2,a)[010]. (2,a)[001] 3
W
p
z
p
y
p
x
L
K
X
Figure 4.9 The rst Brillouin zone of cubic crystals. Points of high symmetry, I, L, and X points,
are shown.
Thus, due to the crystal symmetry, the problem of nding an electron energy spec-
trum E
(k) is reduced to the following steps: (a) determination of the high-symmetry

k points of the Brillouin zone; (b) calculation of energy positions of extrema; and
(c) analysis of effective masses or other parameters of an expansion of E
(k) within
extrema.
The structure and symmetry of the Brillouin zone for cubic crystals of group IV
semiconductors and of IIIVcompounds are very similar. Figure 4.9 shows the Brillouin
zone of these semiconductor materials. The symmetry points are shown in Fig. 4.9 and
presented in Table 4.3. Evidently, as a result of crystal symmetry, several points have
the same symmetry; indeed, they are mapped into themselves under proper symmetry
transformations. Such a degeneracy of the symmetry points is indicated in Table 4.3. In
particular, the I, L, X, and L points are of central importance. They give the positions
of the extrema of the electron energy in IIIV compounds, Ge, and Si.
The bandstructures for GaAs and Si are presented in Fig. 4.10. The energy dispersions
along two symmetric directions of the wavevectors [111] (from I to L) and [100] (from
I to X) are shown. In each case, the energy, E, is taken to be zero at the top of the
highest valence band, which is located at the I point for both of these materials. Since

E
g

GaAs
1
1
2
3
E
(
e
V
)
E
(
e
V
)
1
1
2
3
Si
E
g
L X X L
Figure 4.10 Bandstructures of Si and GaAs: the bandgap of Si is E
g
= 1.12 eV; and the bandgap
of GaAs is E
g
= 1.42 eV.
electrons tend to be near the energy minima, one can think of electrons as being located
inside marked regions of k-space. Frequently, these regions in k-space are referred to as
energy valleys, or simply valleys. Materials with several valleys are called many-valley
semiconductors. For IIIV compounds there is only one valley around the point k = 0
(the I valley); however, in the case of Si there exist six L valleys in accordance with
the degeneracy of the L points. It is worth emphasizing that also other symmetry points
can play a considerable role in processes occurring far from equilibrium. The highest
valence band and the lowest conduction band are separated by energy bandgaps, E
g
. Let
us postpone the analysis of valence bands for a while and focus on conduction bands.
Conduction bands have different structures for the groups of materials considered in
this section. The main difference is that for Si and Ge the lowest minima are located
at L and L points, respectively, whereas for most IIIV compounds there is only one
lowest minimum, which is at the I point. The difference is not simply quantitative;
indeed, it is qualitative and leads to a series of important consequences in the behavior of
electrons.
As discussed previously, within the I, L, X, and L points the electron dispersion
curves, E
(k), can be expanded in series with respect to deviations from the minima,
k
:
E
(k) = E(k
)
1
2
_
h
--2
m
_
i j
(k
i
k
i.
)(k
j
k
j.
). (4.17)
For the I point, we have the simplest case of the so-called isotropic effective mass, m
:
_
1
m
I
_
i j
=
1
m
i j
. (4.18)
For L, X, and L points, i.e., for L, X, and L valleys, the reciprocal effective-mass
tensor has only two independent components, corresponding to the longitudinal, m
l
, and
transverse, m
t
, masses. For example, in the case of Si, for a coordinate system with the
Table 4.4 Energy-band parameters for Si and Ge
Group IV Si Ge
Type of bandgap Indirect Indirect
Lowest minima L points L points
Degeneracy 6 4
E
g
(eV) 1.12 0.664
Electrons
m
l
,m
0
0.98 1.64
m
t
,m
0
0.19 0.082
Holes
m
hh
,m
0
0.50 0.44
m
lh
,m
0
0.16 0.28
L
so
(eV) 0.044 0.29
z-axis along the axis where the Lvalley under consideration is located and the two other
axes perpendicular to the rst, one can obtain
_
1
m
_
i j
=
_
_
m
1
t
0 0
0 m
1
t
0
0 0 m
1
l
_
_
. (4.19)
where m
l
and m
t
are called the longitudinal and transverse effective masses, respectively.
For groupIVsemiconductors, Si andGe, the energyparameters are presentedinTable 4.4.
The degeneracy, given in Table 4.4, indicates the existence of six and four equivalent
energy valleys for Si and Ge, respectively. According to Fig. 4.9, the L points are located
at the edges of the Brillouin zone, i.e., only half of each energy valley lies inside the rst
Brillouin zone. This reduces the effective number of the valleys to four. In Table 4.4 the
bandgaps, E
g
, and the electron longitudinal, m
l
, and transverse, m
t
, effective masses are
also presented.
For IIIV compounds, different situations occur for different materials: some of these
compounds are direct-bandgap crystals and others are indirect materials. Thus, for the
conduction bands of these materials the conduction band edge can be found at the I point,
at the L point, or at the X point. In Table 4.5, energy-band parameters for three sets of
IIIV compounds are shown; the direct or indirect nature of the crystals is indicated.
Now we shall consider valence bands. Their structure is more complicated. For group
IV semiconductors and for IIIV compounds, the top of the valence bands at k = 0
has high degeneracy, i.e., several valence bands have the same energy at this point. The
degeneracy occurs because these bands originate from the bonding of three p orbitals of
the atoms composing the crystals. Thus, if one neglects the interaction between the spin
of the electrons and their motion (so-called spinorbit interaction), one obtains three
degenerate valence bands, each of which is also doubly degenerate as a result of electron
spin. In fact, the spinorbit interaction causes a splitting of these six-fold degenerate
states. At k = 0 they are split into (i) a quadruplet of states with degeneracy equal to 4
and (ii) a doublet of states with degeneracy equal to 2. This splitting of valence bands,
Table 4.5 Energy-band parameters for III--V compounds in the As family
InAs GaAs AlAs
Type of gap Direct Direct Indirect
Lowest minima I point I point X points
E
g
(eV) 0.354 1.42 2.16
Electrons
m
,m
0
0.025 0.067 0.124
Holes
m
hh
,m
0
0.41 0.50 0.50
m
lh
,m
0
0.26 0.07 0.26
L
so
(eV) 0.38 0.34 0.28
L
so
, at k = 0 is shown in Fig. 4.10. One often refers to the lower valence band as the
split-off valence band. At nite k, the spinorbit interaction leads to further splitting of
the upper valence band into two branches: the heavy- and light-hole bands. Parameters
characterizing these bands are the heavy-hole, m
hh
, and light-hole, m
lh
, masses. The
effective masses of the light hole, m
lh
, and heavy hole, m
hh
, as well as the distance to the
split-off band, L
so
, for Si, Ge, and some IIIV compounds are presented in Tables 4.4
and 4.5.
It is worth mentioning that, despite the relative complexity of the picture presented
for energy bands, the description of electron properties in terms of the electron and hole
quasiparticles by using several E
(k) dependences is incommensurably simpler than

operating with the enormous number of valence electrons in the crystal (10
23
cm
3
).
Direct-bandgap and indirect-bandgap semiconductors
One of the important conclusions, which can be drawn from the energy-band picture
described above is related to the optical properties of the crystals.
In Section 2.4, following M. Planck and A. Einstein, we found that electromagnetic
radiation (light) can be thought of as a ux of photons with certain energy and momentum
values given by Eqs. (2.39) and (2.40). The absorption and emission of light can be inter-
preted as the absorption and emission of discrete portions of the light or photons with
a specic energy and momentum. Let us apply these ndings to consider lightcrystal
interactions. In a crystal absorbing or emitting a photon of sufcient energy, an electron
may be transferred between the valence and conduction bands. For a given frequency of
light, , such a transition is possible if the energy and momentum conversation laws are
satised
E
c
(k
1
) E
v
(k
2
) = h
--
.
k
1
k
2
= q.
where k
1
andk
2
are the wavevectors of the electrons participatinginphototransition; here,
q is the photon wavevector. The sign, (), in the second equation stands for photon
emission (absorption). Throughout the whole optical spectral region (from infrared to
ultraviolet light), wavelengths are much greater than electron de Broglie wavelengths,
as estimated in Chapters 2 and 3. The photon wavevectors q = 2,, in turn, are much
smaller than the electron wavevectors ([k
1
[. [k
2
[ [q[). This property reduces the above
equations to
k
1
k
2
= k and E
c
(k) E
v
(k) = h
--
.
In other words, under absorption and emission of the light electrons transferred between
the valence and conduction bands practically preserve their wavevectors. In other words,
the electron wavevector changes very little. In an energy scheme like that presented
in Fig. 4.10, the processes of lightcrystal interaction can be interpreted as vertical
electron interband transitions. Another conclusion following from this analysis is that
the interband phototransitions are possible only for light with energy quanta exceeding
the bandgap, h
--
E
g
. This nding implies that a pure semiconductor crystal is optically
transparent for light with h
--
- E
g
( > 2ch
--
,E
g
. with c being the velocity of light in
vacuum). For some optoelectronic applications, the spectral range near the onset of
absorption/emission of light is critically important.
On combining this analysis with the previously described bandstructures of different
materials we can see that phototransitions induced by light with energy near the bandgap
are possible in semiconductors for which the conduction and valence bands have a
minimum and a maximum, respectively, at the same I point. For example, in GaAs one
can transfer an electron from the valence band to the conduction band directly without a
change in its momentum. Crystals of this type are called direct-bandgap semiconductors.
In contrast, in order to move an electron from the valence band to the conduction band
in Si and Ge, one needs not only to add an amount of energy greater than the minimum
energy difference between the conduction and valence bands to excite an electron, but
also to change its momentum by a large amount (comparable to the scale of the Brillouin
zone). Such a semiconductor is called an indirect-bandgap semiconductor.
Summarizing, the bandstructure of a semiconductor material determines both elec-
trical and optical properties. Manipulation of electrons using light, i.e., optoelectronic
functions, is easier for a direct-bandgap semiconductor, such as GaAs. In contrast, silicon
and other group IV materials are optically relatively inactive. The situation is changed
in nanoscale Si and SiGe structures, where the momentum conservation law is no longer
rigorously obeyed.
Bandstructures of semiconductor alloys
As emphasized in previous discussions, the energy bandstructure of a particular semi-
conductor determines its electrical and optical properties. For naturally existing semi-
conductor crystals such as monatomic Ge and Si, and binary GaAs, their xed, and
unalterable, energy bandstructures restrict their applications. One of the powerful tools
for varying the bandstructure is based on alloying two or more semiconductor materi-
als. Some alloys exhibit well-ordered crystal structures. Though an alloy always has a
disorder of the constitutive atoms, contemporary technology facilitates partial control of
Table 4.6 Bandgaps for III--V alloys
Alloy E
g
(eV)
Al
x
Ga
1x
As 1.42 + 1.247x
Al
x
In
1x
As 0.360 + 2.012x + 0.698x
2
Ga
x
In
1x
As 0.360 + 1.064x
Ga
x
In
1x
Sb 0.172 + 0.139x + 0.415x
2
Al
x
Ga
1x
Sb 0.726 + 1.129x + 0.368x
2
Al
x
In
1x
Sb 0.172 + 1.621x + 0.430x
2
Table 4.7 Effective masses for the alloy Al
x
Ga
1x
As
Type of minimum Effective mass, m
,m
0
I point 0.067 + 0.083x
X minima 0.32 0.06x
L minima 0.11 + 0.03x
Heavy hole 0.62 + 0.14x
Light hole 0.087 + 0.063x
this disorder and produces high-quality crystals. The properties of such materials can be
interpreted in terms of nearly ideal periodic crystals.
Consider an alloy consisting of two components: A, with a fraction x, and B, with
a fraction 1 x. If A and B have similar crystalline lattices, one can expect that the
alloy A
x
B
1x
has the same crystalline structure, with the lattice constant a
c
given by
a combination of lattice constants of materials A, a
A
, and B, a
B
. The simplest linear
combination leads to the equation (Vegards law)
a
c
= a
A
x a
B
(1 x). (4.20)
Then, the symmetry analysis can be extended to these types of alloys. For SiGe alloys
and IIIV compounds, this leads us to the previously discussed symmetry properties
of the energy bands. Since the bandstructures are similar, one can characterize certain
parameters of the alloy as functions of the fraction x. This approximation is often called
the virtual-crystal approximation. For example, the bandgap of an alloy can be repre-
sented as E
alloy
g
= E
g
(x). Such approximate dependences are given in Table 4.6 for IIIV
alloys. They correspond to the bandgaps, E
g
, at I points.
As the composition of an alloy varies, the internal structure of energy bands changes
signicantly. For example, in the case of Al
x
Ga
1x
As alloys, the lowest energy minimum
of the I conduction band of GaAs is replaced by the six X minima of AlAs as the value
of x is increased. Indeed, near the composition x 0.4, the alloy transforms from a
direct- to an indirect-bandgap material. The x-dependences of the effective masses for
various electron energy minima as well as for heavy and light holes for Al
x
Ga
1x
As are
presented in Table 4.7.
Clearly, the established capability of fabricating a variety of high-quality materials
provides an excellent tool for modifying the fundamental properties of materials.
4.5 Semiconductor heterostructures
Further modication and engineering of material properties is possible with the use of
heterostructures. Heterostructures are structures with two or more abrupt interfaces at
the boundaries between the regions of different materials. With modern techniques for
growth of materials, it is possible to grow structures with transition regions between
adjacent materials that have thicknesses of only one or two atomic monolayers.
Band offsets at heterojunctions
Let us consider a junction between two different semiconductor materials, which gener-
ates an abrupt change in the energy gap as well as an abrupt change in the conduction-
and valence-band energies. These abrupt changes result in band-offset steps.
To understand the principal novel features brought about by an abrupt energy change
in the energy bandstructure, we need to deviate from the approach of the previous
sections where, while considering the energy bands of semiconductors, we analyzed
energy structures in terms of relative positions of the bands in each of the semiconductors.
In this approach, absolute values of the energies were not important and only relative
positions of the bands were taken into account. However, if two different materials are
brought together, the absolute values of energies become critically important. There is a
simple way to compare energy bands of different materials. Let us introduce the vacuum
level of the electron energy, which coincides with the energy of an electron outside
of a material. It is obvious that the vacuum level may be taken to have the same value
for any material. One can characterize the absolute energy position of the bottom of the
conduction band with respect to this level, as shown in Fig. 4.11. The energy distance
between the bottomof the conduction band and the vacuumlevel, , is called the electron
afnity. In other words, the electron afnity is the energy required to remove an electron
from the bottom of the conduction band to outside of a material, i.e., to the so-called
vacuum level. Thus, if we know the electron afnities for different materials, we know
the values of the conduction-band bottoms with respect to each other.
With this denition of electron afnity, one can calculate the discontinuity in the
conduction band at an abrupt heterojunction of two materials, A and B:
LE
c
= E
c.B
E
c.A
=
A
B
. (4.21)
where
A.B
are the electron afnities of materials A and B. Similarly, one can calculate
the discontinuity of the valence band for the same heterojunction:
LE
v
= E
v.B
E
v.A
=
B
A
LE
g
. (4.22)
where LE
g
= E
g.A
E
g.B
is the bandgap discontinuity for the heterojunction, with
E
g.A
and E
g.B
being the bandgaps of materials Aand B, respectively. Thus, if this simple
4.5 Semiconductor heterostructures 85
Vacuum level
Bottom of the conduction band
Top of the valence band
E
l
e
c
t
r
o
n

a
f
f
i
n
i
t
y
F
o
r
b
i
d
d
e
n

e
n
e
r
g
y

g
a
p
E
g
Figure 4.11 This diagram illustrates the electron afnity and vacuum level in a crystal.
approach the electron-afnity rule is applicable to a pair of semiconductor materials,
one can calculate band offsets for an ideal heterojunction. Furthermore, if three materials,
say A. B. and C, obey this rule, the following transitivity property is valid:
LE
v
(A,B) LE
v
(B,C) LE
v
(C,A) = 0.
where LE
v
(A,B) is the valence-band discontinuity at the A,B interface. Hence, it is
possible to calculate the band offset for one of three junctions if parameters for two of
them are known.
Unfortunately, this rule fails for many semiconductor pairs. One reason for this failure
is the dissimilar character of chemical bonds in adjacent materials. The formation of new
chemical bonds at such a heterojunction results in charge transfer across this junction
and the consequent reconstruction of energy bands, which leads to the breakdown of the
electron-afnity rule. In real heterojunctions, band offsets can depend on the quality of
the interface, conditions of growth, etc.
On combining different values of the electron afnity and the energy bandgap, we
can expect different band line-ups at the interface between two semiconductor materials.
In Fig. 4.12, sketches of the three possible types of band discontinuity are presented.
The most common line-up is of the straddling type presented in Fig. 4.12(a), with
conduction- and valence-band offsets of opposite signs, and the lowest conduction-band
states occur in the same part of the structure as the highest valence-band states. This case
is referred to as a type I heterostructure. The most widely studied heterojunction system,
GaAs/Al
x
Ga
1x
As, is of this kind for x - 0.4. The next sketch, Fig. 4.12(b), depicts a
heterostructure where the lowest conduction-band minimum occurs on one of the sides
(b) (c)
A
B B
A
E
v
E
c
E
v
E
c
E
c,B
(a)
A
B
E
v
E
c
E
c,A
E
v,A
E
g,A
E
g,B
E
v,B
Figure 4.12 Three types of interfaces: type I (a); type II (b); and broken-gap lineup (c). The
energy bandgaps of the materials A and B are indicated in (b).
eF
eF
eF
+eF
(a) (b)
+eF
h
e
A B A
x
B
1x
(c)
+eF
h
e
A
B
A
x
B
1x
Figure 4.13 (a) A external electric eld, F, simply tilts the conduction and valence bands of a
semiconductor. The forces eF and eF acting on the electron and the hole are equal in
magnitude and opposite in direction. (b) The same direction of forces eF
e
and eF
h
on the
electron and the hole caused by quasielectric eld in the conduction band, F
e
, and in the valence
band, F
h
. (c) Forces eF
e
and eF
h
of opposite directions for electrons and holes.
and the highest valence-band maximum on the other, with an energy separation between
the two less than the lower of the two bulk bandgaps. This case represents a type II
heterostructure. The combination AlAs/Al
x
Ga
1x
As for x > 0.4 and some Si/Si
x
Ge
1x
structures are of this kind. Figure 4.12(c) illustrates a broken-gap line-up, in which the
bottom of the conduction band on one side drops below the top of the valence band on
the other. The example of this band line-up is given by InAs/GaSb, with a break in the
forbidden gap at the interface of the order of 150 meV.
Graded semiconductors
Very often, graded semiconductors are used instead of abrupt heterointerfaces. To illus-
trate the idea, consider rst a homogeneous piece of a semiconductor, say, a piece of
uniformly doped silicon, but with an electric eld applied. Then, the band diagramlooks
like that illustrated in Fig. 4.13(a) and is represented simply as two parallel tilted lines
corresponding to the conduction- and valence-band edges. The separation between the
two lines is the energy bandgap of the semiconductor; the slope of the two band edges
is the elementary charge e multiplied by the electric eld F. When an electron or a hole
Table 4.8 Lattice constants for cubic
semiconductor materials (T = 300 K)
Semiconductor Lattice constant (

A)
SiC 3.0806
C 3.5668
Si 5.4309
GaP 5.4495
GaAs 5.6419
Ge 5.6461
AlAs 5.6611
InP 5.8687
InAs 6.0584
is placed into this structure, a force eF acts on the electron and eF acts on the hole;
the two forces are equal in magnitude and opposite in direction.
Slopes of conduction- and valence-band edges arise in the case of a graded transition
from one material to another. Graded transitions from a narrow-bandgap to a wide-
bandgap semiconductor that correspond to the abrupt heterojunctions of Figs. 4.12(a)
and 4.12(b) are shown in Figs. 4.13(b) and 4.13(c). As is obvious from Figs. 4.13(b)
and 4.13(c), in the case of graded heterostructures there is a built-in electric eld that
acts on electrons and holes. This eld is called quasielectric. The quasielectric eld
does not exist in homogeneous crystals; that is why graded heterostructures can be used
for new devices where the existence of a built-in electric eld is required. Examples of
materials used in graded nanostructure devices are Si
x
Ge
1x
and Al
x
Ga
1x
As, where
x changes in the direction of growth. Graded structures and the accompanying quasi-
electric forces introduce a newdegree of freedomfor the device designer and allowhimto
obtain effects that are basically impossible to obtain using only external (or real) electric
elds.
4.6 Lattice-matched and pseudomorphic heterostructures
Now we shall consider some of the principal problems that arise in the fabrication of
heterostructures. In general, one can grow any layer on almost any other material. In
practice, however, the interfacial quality of such articially grown structures can vary
enormously. Evenwhenone fabricates a structure fromtwomaterials of the same groupor
fromcompounds of the same family, the articiallygrownmaterials of the heterostructure
may be very different from the corresponding bulk materials. First of all, the quality of
the materials near heterointerfaces depends strongly on the ratio of lattice constants for
the two materials.
In Table 4.8, lattice constants for several group IV semiconductors and IIIV com-
pound semiconductors are presented; all of the cases presented represent cubic crystals.
The lattice constants for some other materials can be found fromFig. 4.14. Depending on
the structural similarity and lattice constants of the constituent materials, there exist two
5.4 5.6 5.8 6.0 6.2 6.4 6.6
0.0
0.5
1.0
1.5
2.0
2.5
3.0

GaP
AlP
ZnSe
CdS
Ge
Si
GaAs
AlAs
InP
AlSb
CdTe
InSb
HgTe
InAs
a
0
()
E
g
(
e
V
)
GaSb
Figure 4.14 Room-temperature bandgaps, E
g
, as functions of the lattice constant, a
0
, for selected
IIIV and IIVI compounds and selected group IV materials and their alloys.
essentially different classes of heterointerfaces: lattice-matched and lattice-mismatched
materials. Prior to an analysis of both classes, we highlight other factors affecting the
quality and usefulness of heterointerfaces.
Valence matching
Since there are still no rigorous rules for how one can realize a given level of quality for
heterojunctions, we consider a few examples, which illustrate the problems.
If lattice matching were the only obstacle, the Ge/GaAs system would be the ideal
heterosystem, because, according to Table 4.8, it would allow one to realize the ideal
combination of group IV semiconductors and IIIV compounds. Indeed, on the basis
solelyof lattice-constant matching, the Ge/GaAs systemappears tobe the most promising
candidate. However, it turns out that there is the problem of chemical compatibility for
this heterostructure. Covalent bonds between Ge on the one hand and Ga or As on the
other are readily formed, but they are what could be called valence-mismatched bonds,
meaning that the number of electrons provided by the atoms is not equal to the canonical
number of exactly two electrons per covalent bond. Hence, the bonds themselves are not
electrically neutral. Consider a hypothetical idealized (001)-oriented interface between
Ge and GaAs, with Ge to the left of a mathematical plane and GaAs to the right, as
shown in Fig. 4.15. In GaAs, an As atom brings along ve electrons (resulting in 5,4
electrons per bond) and is surrounded by four Ga atoms, each of which brings along
three electrons (3,4 electron per bond), adding up to the correct number of two (8,4)
electrons per GaAs covalent bond. However, when, at a (001) interface, an As atom
has two Ge atoms as bonding partners, each Ge atom brings along one electron per
bond, which is half an electron more than is required for bonding. Loosely speaking,
the As atom does not know whether it is a constituent of GaAs or a donor in Ge. As
a result, each GeAs bond acts as a donor with a fractional charge, and each GeGa
Ge Ga As
[001]
[
1
1
0
]
-
-
Figure 4.15 Departure from electrical neutrality at a mathematically planar (001)-oriented
Ge/GaAs interface. The different atomic species, Ga or As atoms, and Ge atoms, do not bring
along the correct number of electrons to form electrically neutral GaGe or AsGe covalent
bonds of two electrons per bond. Reprinted with permission from W. A. Harrison, E. Kraut
et al., Polar heterojunction interfaces Phys Rev. B, 18(8), 4402 (1978).
C _
1978 by the
American Physical Society.
bond acts as an acceptor with the opposite fractional charge. To be electrically neutral,
a Ge/GaAs interface would have to have equal numbers of both charges, averaged not
only over large distances, but also locally. Given chemical bonding preferences, such an
arrangement will not occur naturally during epitaxial growth. If only one kind of bond
were present, as in Fig. 4.15, the interface charge would support a large electric eld
of 4 10
7
Vcm
1
. Such a huge eld would force atomic rearrangements during the
growth, trying to equalize the numbers of GeAs and GeGa bonds. However, these
rearrangements will never go to completion, but will leave behind locally uctuating
residual charges, with deleterious consequences for the electrical properties of materials
and any device applications.
Interfaces with perfect bond-charge cancellation may readily be drawn on paper, but
in practice there are always going to remain some local deviations from perfect charge
compensation, leading to performance-degrading random potential uctuations along
the interface. This argument applies to other interfaces combining semiconductors from
different columns of the Periodic Table of elements.
The above discussion pertains to the most widely used (001)-oriented interface. The
interface charge at a valence-mismatched interface actually depends on the crystallo-
graphic orientation. It has been shown that an ideal (112) interface between group IV
and IIIV compounds exhibits no interface charge. An important example is the GaP-
on-Si interface that has a sufciently low defect density; as a result, it is used in various
devices grown on Si. After these comments, we return to the discussion of the role and
importance of lattice matching.
Figure 4.16 Lattice-matched materials (a) and lattice-mismatched materials (b); the resulting
structure is strained (pseudomorphic) if the upper layer B adopts the lattice of substrate A (c).
Lattice-matched materials
For lattice-matched structures, the lattice constants of the constituent materials are nearly
matched; i.e., the lattice constants are within a small fraction of a percent of each other.
There is no problem, in general, in growing high-quality heterostructures with such
lattice-matched pairs of materials. By high-quality we mean that the interface struc-
ture is free of lattice imperfections such as interface defects, etc. Such imperfections
result in poor electrical and optical properties and may lead to fast and widespread
degradation of the structure. Figure 4.16(a) illustrates a lattice-matched layer B on a
substrate A. One can expect that the layer can be grown on the substrate if both mate-
rials are from the same group and the binding energies and crystal structures are very
similar.
According to the data of Fig. 4.14 and Table 4.8, the AlGaAs/GaAs system is an
example of a lattice-matched material. The system has a very small mismatch in the
lattice constants of only about 0.1% over the entire range of possible Al-to-Ga ratios
in the AlGaAs. As a result, such heterostructures can be grown free of mechanical
strain and signicant imperfections. Hence, these structures provide a practical way
of tailoring bandstructures. In addition to these tailored electronic parameters, elastic
and other lattice properties can be different in layers composing such a lattice-matched
heterostructure.
Lattice-mismatched materials
The case of lattice-mismatched structures is characterized by a nite lattice mismatch.
Figure 4.16(b) depicts this case. If one tries to match these lattices, strain in the plane
of growth and a distortion along the growth axis arise. Thus, one obtains a strained
layer with a lattice deformation. The lattice-mismatched structure can be characterized
also by the relative mismatch of the lattice constants of the substrate, a
A
, and the
epilayer, a
B
:
c =
a
A
a
B
a
A
. (4.23)
Consider an elastic deformation of a lattice. It can be characterized by the vector of
relative displacement u. The displacement denes how any lattice point r moves to a
new position, r
/
= r u, as a result of the deformation. Different regions of the crystal
can be deformed differently. Thus, the displacement depends on coordinates: u = u(r).
In fact, only relative displacements are important. They are determined by the strain
tensor:
u
i j
=
1
2
_
u
i
u
j
i
_
. (4.24)
where the tensor u
i j
has the following components:
u
i j
=
_
_
u
xx
u
xy
u
xz
u
yx
u
yy
u
yz
u
zx
u
zy
u
zz
_
_
. (4.25)
Here i and j denote x, y, and z components;
i
= x. y, and z for i = 1. 2, and 3,
respectively. In this discussion, we consider only diagonal components of u
i j
. They
determine a change in the crystal volume, from V to V
/
, produced by the strain:
=
V
/
V
V
= u
xx
u
yy
u
zz
. (4.26)
The elastic energy density (elastic energy per unit volume) of a crystal may also be
expressed in terms of the strain tensor. For cubic crystals, this energy is given by
U =
1
2
c
11
_
u
2
xx
u
2
yy
u
2
zz
_
c
44
_
u
2
xy
u
2
xz
u
2
yz
_
c
12
_
u
xx
u
yy
u
xx
u
zz
u
yy
u
zz
_
. (4.27)
where c
11
. c
12
, and c
44
are elastic constants or elastic moduli of the crystal. Equa-
tion (4.27) is a generalization of the expression for the potential energy of an elastic
isotropic medium derived in Section 2.3. The elastic constants c
12
and c
44
are responsi-
ble for anisotropy of crystals, this is why they equal zero for an elastic isotropic medium.
As a result, Eq. (2.17) is recovered in this case with A = c
11
.
The stress tensor is dened in terms of derivatives of the elastic energy density with
respect to strain-tensor components:
i j
=
U
u
i j
. (4.28)
Boundary conditions at a surface or at an interface may be formulated in terms of the
stress tensor:
i j
N
j
= f
i
. (4.29)
where N is a vector perpendicular to the surface and f is an external force applied to the
surface.
These equations are sufcient for calculations of the strain of a layer A grown on
a mismatched substrate B. Let the lattice constants of these two materials be a
A
and
a
B
, respectively. In this discussion, both materials are assumed to be cubic crystals and
the direction of growth is along the [001] direction. If the layer A adopts the lattice
periodicity of the substrate B, the in-plane strain of the layer is
u
xx
= u
yy
= u
[[
= 1
a
B
a
A
. (4.30)
There should be no stress in the direction of growth. Thus, from Eq. (4.29) it follows that
zz
= 0. On calculating
zz
from Eqs. (4.27)) and (4.28), and from the obtained result,
zz
= c
11
u
zz
c
12
(u
xx
u
yy
), we nd the strain in the direction perpendicular to the
layer:
u
zz
=
2c
12
c
11
u
[[
. (4.31)
Thus, the strain can be found through the mismatch of the lattice constants.
The strain results in two types of effects: (1) the strain can generate various imperfec-
tions and defects; and (2) the strain in the layer leads to a change in the symmetry of the
crystal lattice, for example, from cubic to tetragonal or to rhombohedral, etc. Of course,
the latter effect can modify the energy bandstructure of the layer.
Strained pseudomorphic heterostructures
Here, we consider imperfections generated by strain from a lattice mismatch. In order
to understand the nature of the formation of imperfections in a layered structure, let
us consider the characteristic energies of the structure. First of all, a layer grown on a
substrate with a mismatched lattice should possess extra elastic energy, E
el
, caused by
the strain. This energy is a function of the thickness of the layer, d, and increases with
increasing d. In the simplest case of uniform strain, the elastic energy can be calculated
through its density U: E
el
= U d S, where S is the area of the layer. On the other
hand, the generation of mist defects requires some energy. Let us denote this energy by
E
im
. If the extra elastic energy exceeds the energy associated with the imperfection, i.e., if
E
el
(d) > E
im
, the system will relax to a new state with lower energy and imperfections
will be generated. That is, the extra strain energy is the main physical reason for the
instability and degradation of heterostructures fabricated from materials with a large
mismatch of lattice constants.
Since the value of E
im
remains nite even in thin layers, for certain thicknesses we
may get E
el
(d) - E
im
. Thus, there is not sufcient strain energy and imperfections will
not be generated. Such strained heterostructures can be of high quality. Hence, in some
approximations, for each pair of materials there exists a critical thickness of the lay-
ers, d
cr
; if d - d
cr
the lattice mismatch is accommodated by the layer strain without
the generation of defects. The corresponding layered systems are called pseudomorphic
heterostructures. In general, a pseudomorphic layer of material possesses some charac-
teristics similar to those of the substrate and may possibly have the same lattice structure
as the substrate material. In our case, a crystalline semiconductor layer grown on another
semiconductor takes on the in-plane lattice periodicity of the substrate semiconductor.
Figure 4.16(c) illustrates the case when the deposited layer adopts the lattice periodicity
of a substrate material. Examples of such systems are the Ga
1x
Al
x
As/Ga
1x
In
x
As and
GaAs/Ga
1x
In
x
As structures. In fact, these heterostructures are used to improve the char-
acteristics of the so-calledheterojunction-eld-effect transistors whichwill be considered
in Chapter 8. In spite of signicant mismatches of lattice constants, these structures are
virtually free of interface defects, because of the small, nanometer-scale, thicknesses
of the pseudomorphic layers used in the fabrication of functioning heterojunction-eld-
effect transistors.
It is sometimes possible to grow defect-free systems with layer thicknesses exceeding
the critical thickness. However, such systems are metastable and this may lead to device
degradation as a result of the generation of mist defects driven by temperature effects or
other external perturbations. Central to the stability of pseudomorphic structures is the
question of whether or not the strain energy leads to damage of the materials when the
structures are subjectedtovarious forms of external stress andprocessing. The experience
accumulated in this eld shows that in the case of small strain energy the heterostructures
are stable. For example, in the case of the GaP/GaAsP layered system, the strain energy
is about 10
3
eV per atom. Since this quantity is rather small in comparison with the
energy required to remove the atom from its lattice site, this system can be stable for
sufciently thin layers.
The above-discussed strain states are shown in Fig. 4.17 as a function of x for Ge
x
Si
1x
layers grown on Si substrates. The phase diagram the critical thickness of the layer
versus the Ge fraction consists of three regions: strained layers with defects at large
thicknesses, nonequilibrium (metastable) strained layers without defects at intermedi-
ate thicknesses, and equilibrium and stable layers without defects at small thicknesses.
According to these results, a stable Ge layer on Si (the largest mist) can not be grown
with a thickness greater than 10

A or so.
Let us consider the Si/Ge system in more detail. This system is very interesting
and important because it has opened new horizons for silicon nanotechnology and Si-
based applications. The data of Table 4.8 show that heterostructures based on Si and Ge
materials should be designed so that they are always pseudomorphic.
First of all, the stability and quality of these Si/Ge pseudomorphic heterostructures
depend strongly on the thicknesses of the strained layers, as discussed previously. In
fabricating Si/Ge structures, one grows specic numbers of Si andGe atomic monolayers.
Thus, layer thicknesses can be characterized by the numbers of these monolayers. Let
n and m be the numbers of Si and Ge monolayers, respectively. This system is known
as the Si
n
/Ge
m
superlattice. The second important factor that determines the quality of
these structures is the material of the substrate on which the superlattices are grown.
We have discussed the case of Ge
x
Si
1x
layers grown upon a Si substrate; see Fig. 4.17.
For the fabrication of Si/Ge superlattices, the substrates of choice are frequently either
Ge
x
Si
1x
alloys or GaAs.
Let us consider Ge
x
Si
1x
as a substrate. The elastic energyof a strainedsystemdepends
on the alloy composition of the substrate. Figure 4.18 illustrates this dependence for
various numbers of monolayers for the symmetric case, n = m. In accordance with the

Ge
x
Si
1x
/Si
Strain + Defects
(Equilibrium)
S
t
r
a
i
n

(
M
e
t
a
s
t
a
b
l
e
)
S
t
r
a
i
n

(
E
q
u
i
l
i
b
r
i
u
m
)
Misfit (%)
0 1 2 3 4
0
Germanium Fraction, x
0.2 0.4 0.6 0.8 1.0
10
100
1000
1 m
C
r
i
t
i
c
a
l

L
a
y
e
r

T
h
i
c
k
n
e
s
s
,

d
c
r
Figure 4.17 The stabilitystrain diagram for a Ge
x
Si
1x
layer on Si substrate. After R. People and
J. C. Bean, Erratum: Calculation of critical layer thickness versus lattice mismatch for
Ge
x
Si
1x
/Si strained-layer heterostructures Appl. Phys. Lett. 47, 322 (1985), Appl. Phys. Lett.
49, 229 (1986). Reprinted with permission from R. People and J. C. Bean, Appl. Phys. Lett., 49
229 (1986).
C _
1986 American Institute of Physics.
n = 1
2
3
4
5
6
1 0.5
Ge fraction
E
l
a
s
t
i
c

e
n
e
r
g
y

(
a
.
u
.
)
a
Figure 4.18 The elastic energy of strained Ge
n
/Si
n
superlattices on Ge
x
Si
1x
substrate with
various numbers of monolayers n as a function of the Ge fraction in the Ge
x
Si
1x
substrate. The
numbers on the curves indicate n.
previous discussion, the elastic energy increases with increasing thicknesses of strained
layers for a given substrate material. Because of this, one employs superlattices with
a few monolayers: 2 (n. m) 5. Figure 4.18 also shows a nontrivial dependence of
the strain energy on the alloy composition of the substrate; the minimal strain energy is
expected for x in the range from 0.4 to 0.6.
Another important characteristic of pseudomorphic Si/Ge structures is the distribution
of the elastic energy over the monolayers of the superlattice. It was shown that the most
homogeneous distribution over layers occurs for the Si/Ge alloy with x 0.5. From
this point of view, Si
0.5
Ge
0.5
substrates are preferable. However, these results depend
strongly on the orientation of the substrate material. Often, the direction of growth on
the substrate is chosen to be the [001] direction.
Although the technology of the growth and fabrication of Si/Ge structures is still in a
developing stage, from our short analysis one can see that there is some qualitative and
even quantitative knowledge concerning the behavior responsible for the stability and
perfection of these structures.
Lattice-matched heterostructures
Let us return to Fig. 4.14 and discuss lattice-matched heterostructures in more detail.
From this gure, we can determine the lattice constants of various compounds. First of
all, one can see that the GaAs/AlAs system is really unique because the lattice constants
have almost identical values. In order to achieve lattice matching for other cases, it
is possible either to combine a binary compound and a ternary compound, or to use
ternaryternary compounds having appropriate ratios of atomic species within each
layer. For example, in the case of GaInAs/InP structures, lattice matching is achieved
exactly only for Ga
0.47
In
0.53
As, in which the ratio of Ga to In is 47 to 53 in the GaInAs
layer; for the other ratios, the GaInAs layer is not lattice-matched with the InP. Moreover,
the wide-bandgap Ga
0.51
In
0.49
P material is compatible with the narrow-bandgap GaAs
material.
Inconclusion, the broadrange of possibilities for controllingbandgaps andbandoffsets
for both electrons and holes, as well as electron and hole effective masses, provides
the basis for energy-band engineering. Through such energy-band engineering, it is
possible to design and fabricate high-quality heterostructures with designated optical
and electrical properties. If one can not achieve the desired properties using lattice-
matched compositions, it is possible to employ strained pseudomorphic structures.
4.7 Organic semiconductors
In recent years, organic molecules have been shown to have properties that make them
suited for novel electronic and optoelectronic devices. Such novel devices include organic
light-emitting diodes, electronic circuits in mechanically exible layers, crystal displays,
novel molecular electronic devices, carbon-nanotube-based devices for data processing,
and bioelectronic devices.
H
C
-bond
-bond
H H
H
C
Figure 4.19 A schematic diagram of C
2
H
4
(ethene) illustrating -bonds and -bonds between
carbon atoms.
Some organic molecules including a variety of hydrocarbons have been shown
to have properties, such as relatively high conductivity and states with energy differ-
ences, LE = h
--
, corresponding to near-infrared, visible, and ultraviolet wavelengths,
= 2c,, that make themsuited for novel electronic and optoelectronic devices. These
organic molecules may be deposited and bound on surfaces in specied patterns, using
techniques described in Chapter 5, so that they form patterned layers with nanoscale
dimensions. Thus, active layers in devices may be made very thin so that carrier transit
time through a device is reduced, making possible very fast devices. Many organic poly-
mers are exible and it is thus possible to make electronic and optoelectronic devices on
exible thin lms. Flexible displays such as a computer display that may be rolled up
into a pen represent one of the novel applications of organic molecules in electronics
and optoelectronics.
The importance of organic molecules which formconducting polymers is illustrated by
the fact that Alan Heeger, Alan MacDiarmid, and Hideki Shirakawa won the Nobel Prize
for Chemistry in 2000 for their pioneering research on conducting polymers. Working
together, they learned how to make very pure samples of polyacetylene and found ways
of doping them with bromine by exposing them to bromine vapor. They discovered
that it was possible to increase the conductivity of polyacetylene by seven orders of
magnitude! How does this work? To understand this, consider the - and -bonds in a
simple hydrocarbon such as ethene, C
2
H
4
, as depicted in Fig. 4.19. The localized -bond
leads to the bonding of the two carbon atoms in a simple hydrocarbon such as ethene.
The extended -bonds are found in many hydrocarbons; these molecules are said to be
-conjugated. The extended -bonds are delocalized in space along the length of the
molecule and are thus similar to the extended energy bands that are formed when atoms
are combined to form a crystal. Thus, these extended -bonds open the way for charge
transport along the polymer chain.
Polyacetylene, shown in Fig. 4.20, is a hydrocarbon that has alternating single and
double bonds leading to slightly different CCinteratomic bond distances and therefore
to energy levels separated in energy by different amounts, LE, as discussed in Chapter 3
for bonding and antibonding orbitals; see Fig. 3.17. The differences between the energy
-bonds
Alternating
single and
double bonds
-bonds
Figure 4.20 A polymer with carbon backbone having alternating single and double bonds
between carbon atoms; such -conjugated polymers have extended wavefunctions facilitating
charge transport along the chain. Polyacetylene is a case in point.
Region of -bonds
and charge transport
Alternating
single and
double bonds
Region of -bonds
and charge transport
-
+
Figure 4.21 A -conjugated polymer (without -bonds shown) with an acceptor impurity. The
acceptor atom such as bromine or iodine attracts and captures one of the electrons in the
extended -states and results in a region where there is a decit of electron charge a hole. The
captured electron is relatively immobile compared with the hole, which moves along the
polymer.
levels differ from one hydrocarbon polymer to another, making it possible to produce
conducting polymers with different electrical and optical properties; -conjugated poly-
mers like polyacetylene have overlapping -orbitals that conduct charge. The doping
of a -conjugated polymer with an acceptor impurity is illustrated in Fig. 4.21. The
acceptor removes an electron from the polymer and leaves the polymer with a decit
of negative charge, i.e., with a positive charge a hole (more detailed discussion about
doping is given in Section 5.3). This positive charge attracts neighboring electrons and
produces a local distortion of charge near the positive charge. This distorted charge
density is known as a polaron. A polaron propagates along the molecule in a manner
that bears some similarity to the propagation of a hole in a semiconductor. Indeed, the
polaron behaves as a positive (screened) particle with an effective mass that is different
from the mass of the electron. It is observed for polymers like polyacetylene that the
mobility of the hole is high and resulting conductivities are close to those of metals.
Thus, doping leads to conductivities in -conjugated polymers that make them useful in
electronic devices.
4.8 Carbon nanomaterials: nanotubes and fullerenes
In this section we will study newnanoscale objects carbon-based nanomaterials. These
include the so-called carbon nanotubes and fullerenes. These nanomaterials exhibit many
unusual properties. In particular, the carbon nanotubes exhibit metallic, semiconductor,
and dielectric properties and have great promise for applications in nanoelectronics.
We start with a brief review on carbon and ordinary carbon materials. Carbon, C, is
a group IV element, as is silicon, Si. Carbon is known to be associated with rich and
diverse chemistry and carbon atoms participate in the formation of a great number of
molecules. Before the discovery of carbon-based nanomaterials, there were only two
known forms of solid carbon: diamond and graphite. Diamond is a dielectric with a
very large electron energy bandgap (6 eV). Diamond can be p-doped, thus it should be
considered as a wide-bandgap semiconductor. It is an indirect-bandgap semiconductor.
In contrast, graphite is a semimetal. The structural and electronic properties of both
of these carbon materials are not very promising for common electronic applications.
However, the wide-bandgap properties of semiconducting diamond can be exploited in
high-temperature electronics.
Carbon nanotubes
Actually, carbon nanotubes have the form of seamlessly rolled single sheets of carbon
atoms. The cylindrical sheets may have diameters of only a few nanometers. They are
very small objects that exhibit many different structures and properties. These nan-
otubes have no macroscopic analogues, such as the bulk semiconductors that served as
analogues for traditional semiconductor nanostructures. Thus, to study the structure of
the carbon nanotubes, one should use the most advanced microscopy techniques: atomic
force microscopy and scanning tunneling microscopy, which we will consider in the next
chapter.
Acarbonnanotube is composedof carbonatoms. Adefect-free single-wallednanotube
consists of a single cylinder and is characterized by the tube diameter and a quantity
knownas the helicity. Tounderstandthe structure of carbonnanotubes, we will start witha
two-dimensional graphite sheet, showninFig. 4.22(a). The single sheet of graphite, called
graphene, has the formof a honeycomb-like lattice. Let a
1
and a
2
be the graphene lattice
vectors and n and m be integers. The diameter and helicity of the nanotube are uniquely
characterized by the vector C = na
1
ma
2
(n. m) that connects crystallographically
equivalent sites on a two-dimensional graphene sheet. Here a
1.2
are in units of a
0
3 with
a
0
being the carboncarbon distance. By using the vector C a carbon nanotube can be
constructed by wrapping a graphene sheet in such a way that two equivalent sites of the
hexagonal lattice coincide. The wrapping vector C denes the relative location of these
two sites. In Fig. 4.22(a), the wrapping vector C connects the origin (0, 0) and the point
with coordinates (11, 7). Thus, a nanotube denoted by indices (11, 7) is formed. Atube is
called an armchair tube if n equals m, and a zigzag tube in the case of m = 0. Wrapping
vectors along the dotted lines leads to tubes that are of zigzag or armchair form. All other
wrapping angles lead to chiral tubes whose wrapping angle is specied relative either to
4.8 Carbon nanomaterials 99
armchair zigzag
a
r
m
c
h
a
ir
c
h
ir
a
l
zigzag
t
u
b
e

a
x
i
s
(11,0)
(0,7)
a
2
a
1
(11,7)
C
T H
(0,0)
(b) (a)
Figure 4.22 (a) The relation between the hexagonal carbon lattice and the chirality of carbon
nanotubes. A carbon nanotube can be constructed from a single graphite sheet, called graphene,
by rolling up the sheet along the wrapping vector C. (b) Fragments of armchair and zigzag
carbon nanotubes.
the zigzag direction or to the armchair direction = 30
. Both and the wrapping

angle (chiral angle) are in the range (0. 30
) as a result of the hexagonal character

of the carbon two-dimensional lattice of the graphene. Dashed lines are perpendicular
to C and run in the direction of the tube axis indicated by the vector T. The solid
vector His perpendicular to the armchair direction and species the direction of nearest-
neighbor hexagon rows. The angle between T and H is the chiral angle . The unit
cell of a nanotube can be constructed by nding the smallest lattice vector T which
connects equivalent points of the lattice. It can be shown that this vector is given by
T = [(n 2m)a
1
(2n m)a
2
],q. (4.32)
where a
1
and a
2
are the basis vectors of the graphene lattice; q is dened as follows:
q =
_
l. if n m is not a multiple of 3l.
3l. if n m is a multiple of 3l.
where l is the greatest common divisor of n and m. The number of hexagons, N, in the
unit cell is
N =
2(n
2
m
2
nm)
q
. (4.33)
The tube radius, R, is given by
R =
[C[
2
=
3a
0
2
_
n
2
m
2
nm . (4.34)
The length of the carboncarbon bond can be estimated as a
0
0.14 nm. It is easy to see
that, for a general chiral nanotube, the number of atoms contained in the unit cell is very
large. For example, an (8, 7) tube has a radius of about 0.57 nm and contains 676 atoms
in the unit cell. In determining the number of atoms, we have taken into account that
each hexagon contains two carbon atoms: n = 8, m = 7, n m = 1, q = 1, N = 338,
and the number of atoms is 2N = 676. In Fig. 4.22(b), the structures of two particular
examples of the armchair and zigzag types of nanotubes are sketched.
Figure 4.23 Single-walled carbon nanotubes of two different diameters.
No. 10
No. 11
No. 7
No. 8
No. 1
H
1 nm
T
j
Figure 4.24 Atomically resolved scanning tunneling microscope images of individual
single-walled carbon nanotubes. Reprinted with permission, from P. Moriarty, Nanostructured
materials, Rep. Prog. Phys., 64, 369 (2001).
C _
IOP Publishing Limited.
In Figs. 4.23 and 4.24, atomically resolved images of individual single-walled carbon
nanotubes are shown. The images were obtained by using scanning tunneling microscopy.
On the surface of the cylinders (tube images), it is seen clearly that the lattice structure
has a spacing between carbon atoms equal to a
0
0.14 nm. Such a lattice image makes
it possible to identify the chirality of the tube. Dashed arrows in Fig. 4.24 represent the
tube axis T and solid arrows indicate the direction of nearest-neighbor hexagon rows
H; compare this with Fig. 4.22(a). Tubes denoted by 10, 11, and 1 are chiral, whereas
tubes 7 and 8 have a zigzag and an armchair structure, respectively. Tube 10 has a chiral
angle = 7
and a diameter of 1.3 nm, which corresponds to the (11, 7) type shown in
Fig. 4.22(a). A hexagonal lattice is plotted on top of image 8 to prove that the non-chiral
armchair structure applies for this case.
In carbon nanotubes, an electron can propagate freely only in the direction along
the tube axis. Thus, electron or hole transport in these structures has the properties of
one-dimensional transport and the carriers can be characterized by a one-dimensional
wavevector k. As for any periodic structure, we can think of the electron energies as a
system of energy bands dependent on the wavevector k. The energy bands of a single-
walled nanotube can be constructed from the energy bands of the graphene presented
in Fig. 4.22(a). For the case of a multi-walled carbon nanotube, the energy bands may
be constructed from the energy bands of a layer of graphite having a thickness equal
to the thickness of the wall of the nanotube. Indeed, suppose that we nd the energy
bands of the graphene to be E
N
(K) with K = {K
x
. K
y
] being the two-dimensional
wavevector. Instead of K
x
and K
y
, we can use projections of K onto the tube axis, T,
and onto the wrapping vector, C, k and K
C
, respectively; i.e., the energy is E
N
(K
C
. k).
However, along the C direction there is a periodicity. Thus, we can use the so-called
cyclic boundary conditions, as we did in Section 4.4 to obtain the Brillouin zones. This
results in quantization of the K projection onto the vector C:
K
C
=
2N
1
a
0
3
(4.35)
with N
1
being an integer. For the electron energy, E(k), we obtain
E(k) = E
N
(K
C
. k) = E
N
_
2N
1
a
0
3
. k
_
E
N.N
1
(k). (4.36)
That is, each of the initial bands of the graphene N is split into a number of one-
dimensional subbands N
1
. In addition, the character of the splitting depends on the
wrapping type, i.e., on the values of (n. m) for the nanotube. Then, classication as
a metal, semiconductor, or dielectric can be made for these one-dimensional crystals.
For different (n. m) values, distinct properties of the nanotube can be expected. A more
detailed theory predicts that when (n m) is divisible by three the single-walled carbon
nanotubes are of metallic nature; otherwise, they are semiconductors with bandgaps that
depend on the nanotube diameter:
E
g
2E
CC
a
0
d
. (4.37)
where E
CC
is the binding energy of two carbon atoms, a
0
is the carboncarbon distance,
and d is the diameter of the nanotube. Direct measurements support these considerations.
Specically, for semiconducting tubes, an inverse linear dependence of the bandgap on
the nanotube diameter is conrmed, and the coefcient of proportionality is in good
agreement with theoretical expectations. It turns out that both types of carriers can exist
in those nanotubes, i.e., both the electron and the hole can be responsible for electric
conductivity of the tubes.
Similarly to the electronic properties, the mechanical properties of nanotubes are
related closely to those of a single graphite sheet. Since the graphite sheet is very stiff
in the in-plane direction, one can expect nanotubes to have a similar stiffness constant
along the tube axis.
Indeed, it was found that, in contrast to the electronic properties, the elastic moduli of
nanotubes are almost insensitive to the tube diameter and the chiral angle. Their values
are comparable to those of diamond or of a graphite sheet. In general, nanotubes are very
stiff. For comparison, their elastic moduli are as much as ve times larger than those of
steel!
The response of the nanotubes to a large deformation is also remarkable. Indeed, most
hard materials fail with a strain (see Eq. (4.26)) of about 1% or even less, because of
generation of defects. The carbon nanotubes can sustain up to 15% tensile strain before
fracture. Together with high stiffness, this provides a critical tensile strength of a single
nanotube as much as 300 to 400 times larger than that for steel.
In conclusion, carbon nanotubes represent a new class of nanostructures, which dif-
fer from traditional solid-state device structures. The nanotubes can be fabricated with
good control of their basic properties, including the electron bandstructure. They can
be fabricated as semiconductors either with electron conductivity, or with hole con-
ductivity. They can be contacted to metals and various types of processing techniques
have been shown to be viable for the fabrication of electronic devices from carbon
nanotubes.
Following the discovery of carbon nanotubes, it was recognized that this kind of
perfectly organized nanostructure should not be limited to carbon. It was found that for-
mation of nanotubes is a generic property of materials with anisotropic two-dimensional
layered structures. The structures of this type are called inorganic nanotube structures;
examples include WS
2
, MoS
2
, V
2
O
5
, and BN nanotubes. The study of these novel struc-
tures has led to the observation of a few interesting properties and promises potential
applications, particularly in nanoelectronics.
Buckyball fullerenes
Now we consider the so-called buckyball fullerenes which represent the fourth major
formof pure carbon. The previously discussed diamond and graphite and the well-studied
carbon nanotubes are the other three.
The buckyball fullerene is a molecule with the chemical formula C
60
and is one of
the best-known nanoscale objects in nanoscience. Figure 4.25 depicts a buckyball. It
consists of 60 carbon atoms occupying equivalent sites. Each atom is bonded to three
other atoms. Single and double CC bonds occur. The lengths of the two types of bond
are 0.146 nm and 0.140 nm, respectively. That is, these bonds are practically identical.
In Fig. 4.25 they are represented by light and dark-shaded lines. To form almost an ideal
buckyball, nearest neighbors of carbon atoms are organized in pentagons and hexagons.
Every pentagon in the case structure is surrounded by ve hexagons. The truncated
icosahedron structure shown in Fig. 4.25 has 12 pentagonal faces and 20 hexagonal
faces at 60 vertices with a C atom at each vertex.
Figure 4.25 A schematic drawing of the C
60
buckyball (fullerene). Single and double bonds are
shown by light and dark shaded lines.
Because of the near-spherical shape of the truncated icosahedron, we can describe the
electronic energy states of the buckyball fullerene by using the classication of quantum
states developed for a spherically symmetric potential (Section 3.3) and the simplest
atoms (Section 3.4). As was found there, the electron energy states can be classied by
angular momentum quantum numbers (orbital numbers l). In the C
60
fullerene, the
total number of valence electrons is 240. Of them, 180 electrons are involved in
the relatively deep-lying -bonding energy level, so that the electronic properties of the
fullerene are determined primarily by the remaining 60 electrons involved in -bonds.
A total of 50 electrons may be accommodated in quantum states with orbital num-
bers from l = 0 to l = 4. With two-fold spin degeneracy, this gives the following level
populations: 2 6 10 14 18, as illustrated by Fig. 4.26. The remaining 10
electrons ll the energy level with the orbital number l = 5. Actually, the total number
of these states is 2 (2 l 1) = 22. Now, we recall that strictly speaking a trun-
cated icosahedron differs from a sphere and has a lower symmetry. On analyzing this
lower symmetry, one nds splitting of l-states into three groups, as shown in Fig. 4.26
(h
1u
. t
1u
. and t
1g
denote different types of symmetry). The lowest ve-fold (excluding
spin degeneracy) levels accommodate the remaining 10 electrons and can be designated
as the highest occupied molecular orbitals. Two other groups of l = 5 states are lifted
considerably; together with the l = 6 states they create the lowest unoccupied molecular
orbitals.
Conceptually, the electronic properties of an individual fullerene molecule can be
modied via replacement of a C atom with an atom having a greater or smaller number
of valence electrons. Such a process can be thought of as a kind of doping. An example
is azofullerene with the chemical formula C
59
N, where one C atom is replaced by a N
atom.
l
5
5
6
6
5
4
3
3
2
1
0
t
1g
t
1u
h
1u
(HOMO)
(LUMO)
(LUMO + 1)
Figure 4.26 A diagram of the energy states of 60 valence electrons in the fullerene molecule.
After P. Moriarty, Nanostructured materials, Rep. Prog. Phys., 64, 355 (2001).
C _
IOP
Publishing Limited.
The existence of unoccupied molecular orbitals for C
60
explains effects of intermolec-
ular interactionof the buckyballs. For example, twoC
60
molecules, beingneutral systems,
attract each other through polarization of electron clouds. Such an attraction is caused by
van der Waals forces. Owing to these forces, two fullerenes compose the dimer C
60
C
60
with the coupling estimated to be about 0.27 eV. Similar van der Waals forces pro-
vide for interaction and adsorption of the fullerenes onto surfaces of various materials.
Figure 4.27 depicts the fullerene molecules adsorbed onto a Si surface. Since the van der
Waals forces are relatively weak, the fullerene molecules readily diffuse on the surface
and, in large concentrations, they can form hexagonally close-packed islands, as shown
in Fig. 4.28.
Then, the buckyballs themselves form face-centered cubic crystals (at room tem-
perature) with large cohesive energy in this crystal (1.6 eV per molecule). Inter-
estingly, the lattice constant of the C
60
crystal has the unusually large value of
1.42 nm.
Summarizing, the fullerenes are natural and very stable nano-objects manifesting a
number of interesting physical and chemical properties, which can be controlled and
used in nanoscience applications.
4.9 Closing remarks
We began this chapter with the denition of crystalline materials. We introduced two
very important components of crystals, i.e., the electron subsystem and the crystalline
lattice. The electronic applications of a material are determined primarily by its electronic
Figure 4.27 An image of fullerene molecules adsorbed on a silicon surface (at low
concentration). Reprinted with permission, from P. Moriarty, Nanostructured materials,
Rep. Prog. Phys., 64, 306 (2001).
C _
Figure 4.28 At large concentrations, the fullerenes form a hexagonal structure on a surface.
Reprinted with permission, from P. Moriarty, Nanostructured materials, Rep. Prog. Phys., 64,
358 (2001).
C _
properties. As a result of the periodicity of crystals, one can interpret the electron energy
in terms of the energy bands. Depending on the relative positions of the energy bands
and their lling by electrons, a material can be described as a dielectric, semiconductor,
or metal. We found that, despite the interaction of the electrons with the great number of
atoms and ions composing a crystal, one can attribute a quasimomentum (a wavevector)
to an electron and that, frequently, within the actual range of the wavevectors, the electron
energy can be approximated by a simple dependence on wavevector. Several newnotions
that are necessary to describe electrons in these cases were introduced: (1) the positions
of the energy minima, (2) the bandgap, and (3) the effective masses.
We analyzed semiconductor alloys and learned that for many cases of practical interest
the alloy energy spectra can be described in a manner similar to that used to describe
pure crystals. These alloys may be engineered to produce considerable variations in
the electron parameters.
Next, we explained the formation of discontinuities in both valence and conduction
bands when two materials are brought together to form a heterojunction. This effect
provides the practical basis for modifying electron energy spectra, and it is particularly
useful for spatial modulation of the potential proles experienced by electrons and holes
as well as for the creation of various articial heterostructure-based nanostructures with
energy barriers; these nanostructures include quantum wells, quantum wires, quantum
boxes, and superlattices.
By analyzing the state of the art in the fabrication of heterostructures, it was estab-
lished that high-quality heterostructures can be produced using materials with similar
crystal properties, for example, materials from the same group. The ratio of lattice con-
stants of the two materials is a critical parameter for such heterostructures. If these lattice
constants almost coincide, as in the case of lattice-matched materials, one can produce a
heterostructure without strain due to the absence of lattice mismatch or mist imperfec-
tions. An example of such a lattice-matched systemis the AlGaAs/GaAs heterostructure.
For lattice-mismatched materials we have found that only thin layers can accommodate
the lattice mismatch and retain near-perfect crystalline structure. The resultant structures
are pseudomorphic strained layers. The strain in pseudomorphic heterostructures leads
to a set of new phenomena. In particular, it affects the energy spectra in the strained
layers. An example is the Si/Ge heterostructure.
Finally, we considered the structural and electronic properties of novel nanosized
objects such as carbon nanotubes and carbon buckyballs.
More information on crystal symmetry, electron energy spectra, and the general theory
of strain effects can be found in the following books:
C. Kittel, Quantum Theory of Solids (New York, John Wiley & Sons, Inc., 1963).
I. Ipatova and V. Mitin, Introduction to Solid-State Electronics (New York, Addison-
Wesley Publishing Company, 1996).
Abrief history of semiconductor heterostructures and clear motivation for their imple-
mentation is given by Professor Herbert Kroemer in his Nobel Lecture:
H. Kroemer, Quasielectric elds and band offsets: teaching electrons new tricks,
Rev. Mod. Phys., 73, 783 (2001).
Detailed treatments of semiconductor alloys and heterojunctions as well as reviews of
methods of their fabrication and doping are presented in the following references:
J. Singh, Physics of Semiconductors and Their Heterostructures (NewYork, McGraw-
Hill, 1993).
4.10 Problems 107
V. V. Mitin, V. A. Kochelap, and M. A. Stroscio, Quantum Heterostructures (New
York, Cambridge University Press, 1999).
The following publications are devoted to calculations and discussions of electron
energy spectra and wavefunctions in various quantum semiconductor structures:
T. Ando, A. B. Fowler, and F. Stern, Electronic properties of two-dimensional sys-
tems, Rev. Mod. Phys., 54, 437 (1982).
G. Bastard, Wave Mechanics Applied to Semiconductor Heterostructures (New York,
Halsted Press, 1988).
C. Weisbuch and B. Vinter, Quantum Semiconductor Structures (San Diego, CA,
Academic Press, 1991).
A detailed description of carbon nanotubes and carbon buckyballs is presented in the
following recent publications:
M. S. Dresselhaus, G. Dresselhaus, and P. C. Eklund, Science of Fullerenes and
Carbon Nanotubes (San Diego, CA, Academic Press, 1996).
P. Moriarty, Nanostructural materials, Rep. Prog. Phys., 64, 297 (2001).
4.10 Problems
1. Using the lattice constant of silicon, a = 5.43 10
8
cm, and the fact that the
number of Si atoms per unit volume, a
3
, is eight, calculate the number of atoms per
1 cm
3
and the density of the crystalline silicon (silicons atomic weight is 28.1 g mole
1
).
2. Estimate the volume of the rst Brillouin zone in k-space for a simple cubic lattice
with the lattice constant a = 5 10
8
cm. Assume that the average energy of the elec-
trons is 3k
B
T,2, where k
B
and T are Boltzmanns constant (k
B
= 1.3810
23
J K
1
)
and the ambient temperature, respectively. Estimate the volume occupied by electrons
in k-space at T = 300 K with the effective mass m = 0.1m
0
, with m
0
being the free-
electron mass. Compare these two volumes and discuss whether Eq. (4.17) is valid for
electrons with the parameters specied previously.
3. Three electrons of the same energy are placed into three different energy valleys
of silicon. The valleys are located at L points of the [100]-, [010]-, and [001]-axes.
Assuming that all three electrons move along the same direction, say [100], and using
Eqs. (4.17) and (4.19), nd the ratio of the velocities for these electrons.
4. Consider a valence band consisting of light- and heavy-hole branches. Assume that a
heavy hole with energy E is transferred to a light-hole state with the same energy E. Find
the ratios of the quasimomenta and the velocities of the hole in the initial and nal states.
5. For Al
x
Ga
1x
As alloy, nd the composition having an energy bandgap equal to
2 eV using Table 4.6. For this alloy, determine the effective masses in the I and X
valleys using Table 4.7.
6. Assume that for some applications it is necessary to use a lm of InGaAs of high
quality, which can be grown on an InP substrate. By using Eq. (4.20) and data presented
in Table 4.8, nd the lattice-matched composition of this alloy, its lattice constant, the
energy bandgap, and the wavelength of the light corresponding to this bandgap.
7. Assume that the conduction-band offset for an Al
x
Ga
1x
As/GaAs heterojunction is
60% of the difference of the bandgaps of these materials. Find the composition of the
AlGaAs layer necessary for the resulting heterojunction to have an energy barrier for the
electrons equal to 0.3 eV. Calculate the energy barrier for the holes.
8. On the basis of the values of lattice constants given in Table 4.8, explain why it is fea-
sible to grow stable Al
x
Ga
1x
As/GaAs and In
x
Al
1x
As/In
y
Ga
1y
As heterostructures;
explain why it is difcult to grow stable heterostructures of GaP/SiC and InP/SiC.
9. Find the numbers of hexagons per unit cell in zigzag and armchair nanotubes. Cal-
culate the radii of (7, 8) and (11, 10) nanotubes.
5 Growth, fabrication, and
measurement techniques for
nanostructures
5.1 Introduction
Having reviewed the basic properties of materials exploited in nanoelectronics, we shall
nowstudy the principal methods of high-quality material growth and nanodevice fabrica-
tion. Methods for the growthof perfect materials withcontrollable properties are critically
important for nanostructure fabrication. Indeed, stringent requirements must be met for
the growth of crystals for nanosize devices. These requirements include many factors and,
rst of all, ultra-high quality and purity, both controlled within extremely close limits.
The following examples illustrate the term ultra-high quality. For Si crystals used in
nanodevices, concentrations of controlled impurities currently reach concentrations of
less than one part in ten billion (1 in 10
10
). For the case of Ge, this number is in the
range of 1 in 10
13
10
14
. The quality of a silicon crystal being used for nanoelectronics
can be characterized in terms of the density of defects: they must be limited to several
tens per 1 m
2
(!) of the Si wafer according to the Semiconductor Road Map, that was
discussed in Chapter 1. The basic methods of growth of perfect crystalline materials and
multilayered heterostructures we will discuss in Section 5.2.
To fabricate a nanostructure and a nanodevice two approaches can be undertaken.
The rst is based on a previously grown perfect material with further processing. This
includes a number of fabrication stages and methods (nanolithography, etching, implan-
tation, selective doping, etc.). In Section 5.3 we review these methods. The second
approach exploits special regimes of material growth, when nanostructures are formed
spontaneously due to the growth kinetics. Such a growth regime can control size,
shape, and other properties of the nanostructures. Details of this approach are given in
Section 5.4.
Progress in the renement of fabrication techniques for making nanostructures
depends on the great improvements made in characterization methods. In particular,
composition and dopant distribution, lattice strain, and other parameters within nanos-
tructures must be known with atomic-scale precision. Currently, the manipulation of a
single atom (ion) in a solid is possible. In Section 5.5, we shall present the most impor-
tant characterization techniques, such as atomic-force microscopy, scanning tunneling
microscopy, transmission electron microscopy, and others.
The previously mentioned approaches for the production of nanostructures and nano-
electronic devices actually represent evolutionary improvements in the growth and
110 Growth, fabrication, and measurement techniques
processing methods applied previously in microelectronics. Nanoscale objects like
carbon nanotubes and biomolecules require, in general, other techniques for produc-
tion. These innovative techniques are highlighted in Sections 5.6 and 5.7.
Advances in nanotechnology require the utilization of methods and concepts from
almost every area of science and engineering. Synthetic chemistry and even biology
have much to offer for this emerging eld. Some fundamental concepts coming from
these elds can successfully be exploited. These include chemical and biological meth-
ods of surface nanopatterning, and preparing nanostructured materials with predened,
synthetically programmable properties from common inorganic building blocks with
the help of DNA interconnect molecules. The basic ideas related to these chemical and
biological approaches are discussed in Section 5.8.
The great technological advances brought about in mainstream microelectronics and
nanoelectronics can be used for the fabrication of another class of nanodevices that
employs both electrical and mechanical properties of nanostructures. This new genera-
tion of devices is commonly called nanoelectromechanical systems (NEMSs). Indeed,
a strong enhancement of coupling between electronic and mechanical degrees of free-
dom appears on the nanometer length scale. This results in a new class of devices
that includes nanomachines, novel sensors, and a variety of new devices functioning
on the nanoscale. In Section 5.9 we study the methods of fabrication of this class of
nanodevices.
5.2 Bulk crystal and heterostructure growth
Though technological methods and especially regimes of growing various types of crys-
tals are generally different, they have a lot in common. Here we consider the common
steps in growing pure materials using Si technology as an example.
Single-crystal growth
The following three steps are necessary to produce high-quality silicon crystals:
(i) production of metallurgical-grade silicon (impurity level 5 10
16
cm
3
); (ii)
improvement of the latter material up to electronic-grade silicon (the level of impu-
rities is reduced to 5 10
13
cm
3
or less); and (iii) conversion to single-crystal Si
ingots.
Metallurgical-grade silicon is typically produced via reaction of silicon dioxide (SiO
2
)
with C in the form of coke:
SiO
2
2C Si 2CO. (5.1)
which requires very high temperature (1800
C). Coke is a coal fromwhich most of the

gases have been removed. The silicon obtained at this step is not single-crystalline and
is not pure enough for electronic applications, though it is good for some metallurgical
applications such as the production of stainless steel.
Molten Si
Growing
crystal
Seed
Crucible
Pull
Wafers
Figure 5.1 The Czochralski method for the growth of bulk semiconductors.
Further reductions in impurities can be achieved by carrying out the following reaction
of the silicon with dry HCl:
Si 3HCl HSiCl
3
H
2
. (5.2)
Now we obtain the trichlorosilane HSiCl
3
, which is typically in a liquid state with
a boiling point of 32
C. Simultaneously with HSiCl

3
, other chlorides of impurities,
such as FeCl
3
, are formed. Since their boiling points are different, the simple fractional
distillation technique can be applied: the mixture of HSiCl
3
and other impurity chlorides
is heated and then condensed in a series of distillation towers at appropriate temperatures.
By this technique HSiCl
3
is separated from impurities. The following reaction with H
2
then converts the trichlorosilane into highly pure electronic-grade silicon:
2HSiCl
3
2H
2
2Si 6HCl. (5.3)
The pure Si obtained by this process is still polycrystalline.
The nal process, which converts polycrystalline silicon into single-crystal Si ingots,
is based on the Czochralski method. In this method, a seed Si crystal provides a template
for growth. First, this seed crystal is lowered into the molten Si material. (The melting
point of Si is 1412
C.) Then it is raised very slowly so that the molten material touching
the seed crystallizes as the seed is withdrawn from the molten material. Rotation of the
seed crystal, stabilization of the temperature eld, and other tricks are used to growhighly
homogeneous ingots. The Czochralski method is illustrated in Fig. 5.1. Importantly, this
technologyfacilitates dopinginthe course of crystal growth. Indeed, one canintentionally
add precise quantities of impurities (dopants) into semiconductor melts to provide for
regions of crystallization having the desired doping concentrations. This technique is
used widely in growing silicon, germanium, and, with some modications, compound
semiconductors.
As the single-crystal ingot is grown, it is mechanically processed to obtain wafers of
thicknesses of hundreds of micrometers, as shown schematically in Fig. 5.1. The wafers
are used subsequently for producing individual devices, or integrated circuits, or for the
GaAs
Substrate
Sources
Shutters
Molecular beams
Ga
As
Si
Al
Be
Figure 5.2 The MBE method for the growth of GaAs/AlGaAs heterostructures.
fabrication of much more complicated structures. In the following discussion, we focus
on the technique of growing single-crystal layers above a substrate wafer, which is called
epitaxial growth, or just epitaxy.
Epitaxial growth
Fabrication of a crystal layer upon a wafer of a compatible crystal makes it possible to
obtain very well-controlled growth regimes and to produce high-quality crystals with the
desired crystalline orientation at temperatures typically well below the melting point of
the substrate. During the epitaxial growth, several methods of delivering the necessary
atoms to the growing layer can be used. The most developed methods are molecular-beam
epitaxy (MBE), chemical-vapor deposition (CVD), and liquid-phase epitaxy (LPE). Here
we shall discuss MBE and CVD in more detail.
Molecular-beam epitaxy
The MBE method can be realized in a high vacuum, where molecular or atomic beams
deliver onto a substrate the necessary components for growing the desired crystalline
layer. For example, suppose that we want to grow an AlGaAs layer on GaAs. Then,
the substrate will be GaAs and the atomic beams are uxes of the elements Al, Ga,
and As, as well as beams of dopants (typically, Si is used for n-doping and Be for
p-doping). Sources of the elements are contained in separately heated chambers. The
evaporated elements formbeams, which are separately and closely controlled, collimated,
and directed onto the substrate surface, as illustrated by Figs. 5.2 and 5.3. Typical
ux densities in the beams are of 10
14
10
16
atoms cm
2
s
1
. The substrate is held
at relatively low temperature (600

C for GaAs), while densities of the components
in the beams are large. This provides effective growth of the layer. A slow growth rate
(1 monolayer per second), which is often referred to as layer-by-layer growth, results in
Figure 5.3 A typical MBE chamber. Used with permission from
www.ece.cmu.edu,dwg,research,mbechamber.html.
C _
2007, ECE.
the growth of a high-quality layer. By controlling shutters for each beam, one can produce
abrupt changes in crystal compositions and doping concentrations on the scale of one
monolayer.
Chemical-vapor deposition
This epitaxial method allows one to realize a low-temperature growth regime and to
use high-purity chemicals for delivering the necessary atoms for growth of a crystalline
layer. The layers can be grown onto a seed crystal or substrate frommixtures of chemical
vapors containing both semiconductor elements and dopants.
Wafers
Vent
Pedestal
Susceptor
Quartz chamber
Gas baffle
Gas inlet
Radio-frequency
source
Figure 5.4 The CVD method for the growth of heterostructures.
The chemical-vapor deposition is conducted in a reaction chamber called a reactor,
as illustrated by Fig. 5.4. In the reactor, a typical pressure of chemicals is 10
4
Pa and
heating is achieved by power from a microwave radio-frequency source.
In the case of growth of Si layers, several different gases containing Si atoms can
be used. They include silicon tetrachloride (SiCl
4
), silane (SiH
4
), and dichlorosilane
(SiH
2
Cl
2
). Inthe use of silicontetrachloride, the followingreactionwithhydrogenoccurs:
SiCl
4
2H
2
Si 4HCl.
The reaction can be conducted at temperatures in the range of 11501250

C. In the
case of using silane and dichlorosilane, the reaction can be conducted at even lower
temperatures (10001100

C). These temperatures are well below the melting point of
Si. Thus, these reactions release atoms of Si, and the relatively low-temperature regimes
provide efcient crystal growth onto the seed.
For growth of A
III
B
V
compounds, the following reactions are used:
A
III
(CH
3
)
3
B
V
H
3
A
III
B
V
3CH
4
.
These reactions take place at temperatures of 600700

C. Dopants can be delivered,
for example, by silane (n-doping) or diethylzinc (p-doping).
It is important that epitaxial methods can be applied to produce new materials that are
difcult to grow by other methods. Examples are wide-bandgap nitrides of the group III
elements. These include InGaN and AlGaN compounds.
In conclusion, epitaxial growth on the basis of CVD and MBE methods has become
the dominant technique for the fabrication of perfect multilayered crystals of nanoscale
thicknesses. Extremelyhighuniformityof layers has beendemonstratedtobe achievedby
these epitaxial methods. Both group IV elements and IIIV compounds are successfully
5.3 Nanolithography, etching, and other means 115
grown with thickness control of the order of one monolayer. Different types of doping
uniform doping, modulation doping, and delta-doping are realized with high accuracy.
Since in the chemical reactor the partial pressures of chemicals are much higher than the
pressure in the molecular beams of the MBE method, the rate of crystal growth realized
in the CVD method is higher than that of MBE. The former may be used in industrial
production, while the latter is rather well suited for research laboratories.
5.3 Nanolithography, etching, and other means for fabrication
of nanostructures and nanodevices
In the previous section, we studied methods of growing high-quality single-crystal wafers
and crystalline multilayered structures. We found that thicknesses of the layers may be on
the nanometer scale. However, to produce an individual device or electric circuit scaled
down to nanosize in two, or three, dimensions, one needs to exploit additional methods.
First of all, on a wafer it is necessary to produce patterns corresponding to features of the
nanodevice or circuitry. This may be done by using one of the so-called nanolithography
methods.
Let us start with photolithography. This method involves the generation of a reticle,
which is essentially a transparent quartz plate with the necessary pattern. Opaque regions
on the reticle are made up of an iron-oxide layer, which absorbs ultraviolet (UV) light.
A pattern on the reticle is usually made by a computer-controlled electron beam, which
moves as prescribed by pattern-generation software. An electron-beam-sensitive material
(the electron-beam resist) is placed onto the iron-oxide-covered quartz. The resist is
exposed selectively by the electron beam and then the exposed material (positive resist)
is removed. Finally, the iron-oxide layer can be selectively removed by etching to generate
the desired pattern on the quartz plate.
During the next step, a thin uniform layer of photoresist is deposited onto the wafer
surface. There are two types of photoresists: positive and negative. For positive resists,
the resist is exposed with UV light wherever the underlying material is to be removed.
In these resists, exposure to the UV light changes the chemical structure of the resist
so that it becomes more soluble. The exposed resist is then washed away by developer
solution, leaving windows of the bare underlying material. The mask, therefore, contains
an exact copy of the pattern which is to remain on the wafer. Negative resists behave in
just the opposite manner. Exposure to the UV light causes the negative resist to become
polymerized, and thus more difcult to dissolve. Therefore, the negative resist remains
on the surface wherever it is exposed, and the developer solution removes only the
unexposed portions. Masks used for negative photoresists, therefore, contain the inverse
(or photographic negative) of the pattern to be transferred. Figure 5.5 shows the steps
involved in photolithography, as well as the pattern differences generated by the use of
positive and negative resists. Positive resists are now the dominant type of resists used
in fabrication processes.
To transfer the patterns onto the wafer surface, the mask should be aligned with the
wafer. Once the mask has been aligned accurately with the pattern on the wafer surface,
Illuminating radiation
Mask
Photoresist
Film to be patterned
Substrate
Develop
Etch pattern
Imaging system
Illuminated areas
Ion-oxide layer
Quartz plate
Negative resist:
exposed rendered
insoluble
Positive resist:
exposed is
removed
Figure 5.5 Basic lithographic processes.
the photoresist is exposed through the pattern on the mask to high-intensity UV light.
One of the exposure methods, the contact method, is shown in Fig. 5.5. In this method,
the photoresist is exposed to UV light while the wafer is in the contact position with the
mask. As a result of the direct contact between resist and mask, very high resolution is
possible in contact printing.
One of the last steps in the photolithography process is development. The results
obtained after exposure and development are shown in Fig. 5.5 both for negative and for
positive resists.
Actually, photolithography processes have been known for decades. To use them for
nanostructures, we have to understand the limitations imposed by the wave nature of
light. Indeed, as we discussed in Chapter 2, simple geometrical ray optics is applied for
dimensions greater than the wavelength of light. The wave effect diffraction of light
restricts the application of photolithography to minimum dimensional scales of about
the wavelength of light. Thus, shorter wavelengths are preferable. Examples of some of
the shorter wavelengths in use are 0.365 m for UV mercury lamps and 0.193 m for
ArF excimer lasers. With these UV sources, it is possible to achieve linewidths of about
0.25 m and 0.15 m, respectively. Further penetration into the deep-UV region appears
to be extremely difcult.
The previously mentioned diffraction limit is much smaller for X-rays, electron beams,
and ion beams. Thus, advances in nanolithography are occurring as a result of the use
of these short-wavelength beams. For example, electrons with an energy of 10 keV have
a wavelength of about 0.1

A; i.e., less than the lattice constants of any crystal. Now, the
Plasma radicals and ions
Etch gas Pump
Wafer
Cathode
Anode (Chamber)
RF
power supply
Figure 5.6 Reactive-ion etching.
ultimate linewidth is determined by interaction with the photoresist layer. In principle,
it is possible to reach linewidths below 0.1 m for direct writing with an electron beam
onto the photoresist.
Etching
As soon as a resist pattern is formed, an etching method is generally used to pro-
duce the desired structures as shown in Fig. 5.5. There are many etching methods.
A widely used method is wet chemical etching. For example, dilute HF is used to
etch a SiO
2
layer covering silicon. The HF reacts with SiO
2
and does not affect the
photoresist or silicon. That is, this wet chemical etch is highly selective. However,
the rate of etching is the same for any direction, lateral or vertical, so the etching is
isotropic. Using an isotropic etching technique is acceptable only for relatively large
structures. For nanosize structures, anisotropic etching with faster vertical etching is
preferable.
Anisotropic etching generally exploits a physical process, or some combination of
both physical and chemical methods. The best-known method of anisotropic etching is
reactive-ion etching. Reactive-ion etching is based on the use of plasma reactions. This
method works as follows. An appropriate etching gas, for example a chlorouorocarbon,
lls the chamber with the wafers. The pressure is typically reduced, so that a radio-
frequency (RF) voltage can produce a plasma. The wafer we want to etch is a cathode
of this RF discharge, while the walls of the chamber are grounded and act as an anode.
Figure 5.6 illustrates a principal scheme for the ion-etching method. The electric voltage
heats the light electrons and they ionize gaseous molecules, creating positive ions and
molecular fragments (so-called chemical radicals). Being accelerated in the electric eld,
the ions bombard the wafer normal to the surface. This normal incidence of bombarding
ions contributes to the etching and makes the etching highly anisotropic. This process,
unfortunately, is not selective. However, the chemical radicals present in the chamber
give rise to chemical etching, which, as we discussed, is selective. Now we see that the
method combines both isotropic and anisotropic components and can give good results
for etching on the nanoscale.
Doping
As we discussed in Section 4.2, a perfect semiconductor is a dielectric at low tempera-
tures: the valence band is completely lled with valence electrons, while the conduction
band is totally empty. When the crystal temperature rises, some electrons can be excited
into the conduction band, which results in the creation of a pair: an electron in the con-
duction band and a hole in the valence band. Thus, for the electron concentration, n, and
the hole concentration, p, we obtain
n = p = n
i
(T).
Here, n
i
(T) is the temperature-dependent intrinsic density of electrons and holes. Such
a perfect semiconductor is itself called an intrinsic material. In practice, the intrinsic
concentrations are small and can not be controlled effectively.
Thus, it is important to nd a method to create the electron or/and hole concentrations
needed for each particular device application. The most common method of varying
electron and/or hole concentrations in a crystal is the process of adding impurities to the
material; i.e., the doping process. The process is based on the following physical picture.
Whenanimpurityis addedtoa semiconductor, additional energylevels are contributedby
the impurity to the energy structure of the crystal. Many of these additional energy levels
fall near the bandgap energies. Here, we consider two especially important cases: (1) the
impurity levels are close to the conduction band and (2) they are close to the valence band.
In the former case the impurities can be donors of electrons to the conduction band, while
in the latter case they can be acceptors; i.e., they will accept electrons from the valence
band and thereby generate holes in the valence band. For example, impurities fromgroup
V of the periodic table of elements (P, As, Sb) are donors and impurities from group III
(B, Al, Ga, In) are acceptors for group IV semiconductors. The energy distances of
donor levels from the conduction band and of the acceptor level from the valence band
are much smaller than the energy bandgap. This promotes thermally induced ionization
of these impurities even at low temperatures and the creation of conducting electrons
or holes. In Fig. 5.7, the population of energy states in semiconductors with donors and
acceptors is illustrated for zero and nite temperatures.
We can perform the following simple estimates of energy levels of donors and accep-
tors. Consider rst, for example, an atom from group V embedded in a semiconductor
material of group IV. Such an atom has ve valence electrons, of which four can par-
ticipate in formation of covalent bonds with neighboring group IV atoms. But the fth
electron is an extra electron. We can consider this electron to be moving around a positive
ion, i.e., the impurity can be thought of as a hydrogen atom embedded in a dielectric
medium. If the radius of the electron state for this atom is large, we can suppose that the
electron has the effective mass m
, which is characteristic for the conduction band. By

Figure 5.7 Doping of semiconductors: (a) donor doping and (b) acceptor doping.
applying the results of Section 3.4 that were obtained for the hydrogen atom, we nd the
ground-state energy of such a hydrogen-like state of the impurity:
E
1
=
m
e
4
(4c
0
c)
2
h
--2
. (5.4)
with c
0
and c being the permittivity of free space and the dielectric constant of the mate-
rial, respectively. For an acceptor atom, we may apply the same approach, considering a
hole moving around a negatively charged ion. Equation (5.4) will be valid if the effective
mass of the electron in the conduction band, m
, is replaced by the effective mass of a

hole, m
h
, in the valence band: m
m
v
. Typical energies of donor and acceptor levels
in silicon, for example, are about 3060 meV from the conduction and valence bands,
and the ionization occurs at temperatures above several tens of degrees Kelvin.
Similarly to the previously considered case of materials of group IV, in IIIV com-
pounds the atoms from group VI (S, Se, Te) are donors when they occupy sites of group
V atoms. Atoms from group II (Be, Zn, Cd) act as acceptors when substituting group
III atoms. A more complicated case occurs when doping is accomplished with atoms
of group IV; for example, with Si or Ge. These impurities are called amphoteric, which
means that they can act either as donors or as acceptors, depending on the sites they
occupy. If such an atom occupies a group III site, it brings an additional electron and
acts as a donor. When occupying a group V site, the atom accepts an electron and acts
as an acceptor. In GaAs, typically, Si impurities occupy Ga sites; thus they are typically
donors. However, if during the growth process there are As vacancies, Si atoms can ll
these vacancies and serve as acceptors.
In conclusion, the doping of semiconductor materials provides a powerful tool
for realizing the desired type of conductivity electron conductivity or hole
conductivity and fabricating structures with the desired values of electric resistance.
Doping methods provide a means for the control of the electronic properties over a
wide range of values. These methods are well suited for application in the fabrication of
semiconductor nanostructures.
5.4 Techniques for characterization of nanostructures
Progress in the fabrication, study, and use of nanostructures would not be possible with-
out adequate techniques for the characterization of these structures. These techniques
should allow one to determine the shape and geometrical parameters of nanostructures,
the distribution of chemical composition, the strain elds, etc. Knowing all of these,
one can predict the electronic and optical properties which will ultimately be relevant in
applications. The questions of what the geometrical parameters of nanostructures are,
and what their shapes are, can be addressed using atomic-force microscopy (AFM) or
scanning tunneling microscopy (STM). Cross-sectional variants of STM and transmis-
sion electron microscopy (TEM) are sufciently powerful tools for the investigation of
buried structures. X-ray diffraction and some photoluminescence methods may also be
used to determine size and shape, but in general AFM, STM, and TEM have better res-
olution. On the other hand, X-ray diffraction is a very powerful tool for measuring strain
elds, defects, and imperfections. For the determination of statistically relevant proper-
ties, averaging over many individual nanostructures is necessary. STM and TEM often
do not facilitate the study of large enough areas of a sample to yield quantitative data.
Photoluminescence and diffraction techniques make possible the study of intrinsically
large ensembles of nanostructures, so that a good statistical average is obtained automat-
ically, whereas the quantication of uctuations is sometimes difcult. The X-ray and
optical methods are routinely applied for material characterization, and discussion about
them can be found elsewhere.
In this section, we shall present a brief overview of new techniques based on AFM,
STM, and TEM, which can be applied for a detailed characterization of nanostructures.
Scanning tunneling microscopy
This novel technique yields surface topographies and work-function proles on an atomic
scale directly in real space. In Section 4.5, we explained that the removal of an electron
from the conduction band of a solid requires a certain amount of energy called the
electron afnity. For a metal or a doped semiconductor, when the conduction band is
partially lled, the energy needed to remove an electron is lower and it is called the
work function. Let us consider two conducting solids separated by a space. In terms of
classical physics, a transfer process of an electron from one solid into another can be
thought of as an electron transfer over a vacuum barrier. The process requires additional
energy and because of this it has a small probability. In Section 3.3 we found that,
according to quantum mechanics, a particle can penetrate a classically forbidden spatial
5.4 Techniques for characterization 121
s
B
C
A
I
T
V
T
s
P
y
P
x
P
z
V
z
Control
Figure 5.8 The principle of operation of the scanning tunneling microscope.
region under a potential barrier. This phenomenon was called tunneling. Thus, electron
transfer between two solids can occur as a tunneling process through (under) the vacuum
barrier. Tunneling experiments have been performed, for example, by using two metal
lms separated by vacuum or a solid-state insulator (a sandwich structure). Each of the
metal lms can be considered as an electrode and when a voltage bias is applied to
these electrodes a so-called tunneling electric current is produced. This current can give
information on electronic properties, but obviously the information will be averaged
over the area of the metal-lm surface. By appropriate shaping of one of the electrodes
spatial resolution of far smaller scales than that of sandwich structures can be achieved.
Since vacuum is conceptually a simple tunnel barrier, such experiments pertain directly
to the properties of the electrodes and their bare surfaces. Clearly, vacuum tunneling
offers fascinating and challenging possibilities to study surface physics and many other
related topics.
The principle of STM, which is based on electron tunneling, is straightforward. It
consists essentially of a scanning metal tip (one electrode of the tunnel junction) over
the surface to be investigated (the second electrode), as depicted in Fig. 5.8. The metal
tip is xed to a rectangular piezodrive marked as P
x
. P
y
and P
z
. The piezodrive is made
of a piezoceramic material. The tunnel current, I
T
, is a sensitive function of the gap
between the tip and the surface, s; i.e., I
T
V
T
exp(A
1,2
s), where is the average
barrier height; the numerical value of A is equal to unity if is measured in eV and
s in

A. Obviously the barrier height,
0
, is equal to the work function for a metal
or doped semiconductor. With a typical value of of several eV, I
T
changes by an
order of magnitude for every 1

A change of the gap, s. The control unit, CU, applies
a DC voltage, V
z
, to the piezodrive, P
z
, such that I
T
remains constant when scanning
the tip with P
x
and P
y
over the surface. At constant function , V
z
(x. y) yields the
topography of the surface, that is z(x. y), directly, as illustrated at a surface step in
Fig. 5.8. The curvature of the tip causes the abrupt step to appear to be smeared over
a distance . For a constant tunneling current, changes in the work function, , are
compensated by corresponding changes in the distance, s. Thus, a lower work function
at a contamination spot C would be observed as an additional surface structure denoted
by B in the Fig. 5.8. These work-function-induced structures and true structures can,
however, be separated by the following method. Let the gap, s, be modulated by Ls while
scanning at a frequency higher than the cut-off frequency of the control unit. Thus, the
modulation is no longer compensated by the feedback loop of the control unit. As a result,
the current will be modulated by LI
T
. Then, the ratio j
d
L(ln I
T
),Ls
1,2
directly
gives the work function revealing the spot C, in the simple situation shown in Fig. 5.8.
Since V
z
. I
T
, and j
d
can be measured, the topography and the work function can be
reconstructed.
This principal scheme of tunneling microscopy provides (i) stability of a vacuum gap
in the sub-

A range and (ii) a lateral resolution in the

A range. This requires excellent
vibration damping and very sharp tunnel tips. The rst requirement is met by using
highly developed and clever mechanical means. In fact, it is possible to use bungee
cords if they are properly placed and have the desired elastic properties! Tunnel tips used
nowadays are typically made of tungsten or molybdenum wires with tips of overall radii
-1 m. However, the rough macroscopic grinding process creates many rather sharp
minitips. The tunnel current is extremely sensitive to the vacuum gap, s; this is why
the minitip closest to the sample denes the whole current through the tip. Actually,
the lateral resolution is given by the width of the tunnel channel, which is extremely
narrow. Additionally, focussing of the tunneling current (in addition to the geometrical
one) occurs due to a local lowering in height of the tunnel barrier at the apex of the tip.
At present, the resolution of scanning tunneling microscopy reaches 0.05

A vertically
and well below 2

A laterally.
Scanning tunneling microscopy is subject to some restrictions in application: only
conductive samples can be investigated, and measurements usually have to be performed
in ultra-high vacuum.
On the other hand, the tunnel current is sensitive to material composition and strain.
Atomic resolution in both lateral and vertical directions makes STM an ideal tool for
the investigation of growing surfaces and facets at this scale, which can give insight into
growth mechanisms. STM systems attached to a growth chamber allow measurements
to be made without breaking the vacuum after growth.
The tunnel current in STM is sensitive only to a thin layer at a sample surface, and
therefore it might seem that buried structures are beyond the scope of STM studies.
However, buried structures can be studied by STM. Indeed, after cleaving samples STM
can be performed at the cleavage edge. Such a cross-sectional STM can reveal details
on the inner structure of buried nano-objects. In Fig. 5.9(a), the measured and simulated
cross-sectional STMproles for a stack of InAs islands on GaAs substrate are shown. It is
seen that the buried InAs islands have the shape of truncated pyramids. A compositional
intermixing in the islands was found, with the GaAs composition decreasing linearly
from 0.4 at the base to 0 at the top of the islands. The corresponding lattice-parameter
distribution in the growth direction is shown in Fig. 5.9(c). This indicates directly an
increase of compressive strain in the GaAs matrix above and below the islands.
2.5 nm
Measurement
height ()
0
6
2
z

|
|

[
0
0
1
]

(
n
m
)
y || [110] (nm) y || [110] (nm)
2
10
5
0.55
0.60
0.65
0
position along [001] (nm)
l
a
t
t
i
c
e

c
o
n
s
t
a
n
t

a
l
o
n
g

[
0
0
1
]

(
n
m
)
5 10
0
6
2
2
10 10 0 10
1
(b)
(d)
(a)
(c)
2
[001]
[110]
Simulation
3 4
Figure 5.9 Cross-sectional scanning tunneling microscopy (STM): (a) an image of a stack of
InAs islands in GaAs; (b) comparison between measured and simulated height proles for a
similar sample; (c) the lattice parameter in the growth direction in an InAs island (the
experimental data were obtained from cross-sectional STM; the solid line is from a simulation
assuming an In content increasing from island base to island apex); and (d) the electronic
wavefunction measured at two different tip biases, compared with simulations for the ground
and the rst excited states. Reprinted with permission from J. Stangl, V. Hol y et al., Structural
properties of self-organized semiconductor nanostructures, Figs. 25 and 26, Rev. Mod. Phys.
76, 725783 (2004).
It is remarkable that, apart from providing structural information, low-temperature
scanning tunneling spectroscopy has been used for wavefunction mapping of single-
electron states in nanostructures. In Section 3.4, while studying the electron states in
spherically symmetric potentials we introduced the states of different symmetries: s, p,
etc. When they are applied to the InAs dots (islands) the STM methods directly reveal
s-, p-, d-, and even f-type states as made visible by an asymmetry of the electronic
structure, which can be attributed to a shape asymmetry of the islands. Simulation of
the electron ground state and rst excited state of an InAs island corresponds well
to the STM image, showing that the wavefunctions in such islands are indeed atom-
like; see Fig. 5.9(d). In the panel on the left, these are electrons in the ground state;
in the panel on the right, electrons from both the ground state and the rst excited
state contribute to the measured electron distribution. These two measurements were
performed at different voltages at the STM tip: at a low bias of 0.69 V, only s electrons
contribute, whereas at a larger bias of 0.82 V, both s and p electrons contribute to the STM
image.
Figure 5.10 Atomic-force microscopy in the contact mode. The tip has dimensions of about
3050 nm at the end.
Atomic-force microscopy
An atomic-force microscope measures the force between the sample surface and a very
ne tip. The force is measured either by recording the bending of a cantilever on which
the tip is mounted the contact mode or by measuring the change in resonance fre-
quency due to the force the tapping mode. For example, in Fig. 5.10, the contact-mode
AFM technique is sketched. With a typical resolution of several nanometers laterally
and several

A vertically, AFM is ideally suited to characterize the shapes of nanostruc-
tures. For large scan sizes up to 100 m100 m, the lateral arrangement can also be
obtained. With AFM, any surface can be investigated; hardly any sample preparation is
required. A drawback of AFM is that only structures on a surface can be investigated.
Furthermore, most semiconductor materials oxidize under ambient conditions, so that,
strictly speaking, the AFM images usually show the surface of this oxide. When obtain-
ing quantitative data such as lateral sizes and heights of structures, this has to be kept in
mind, as well as the fact that the image is actually a convolution of the samples surface
morphology with the shape of the microscope tip. Figure 5.10 is a schematic view of a
contact-mode AFM. Essentially, a micrometer-size cantilever has an extremely sharp tip
attached to it, which is sharpened to about 3050 nmat the end. Alow-power probe laser
beam is reected off the top of the cantilever and into a four-quadrant photodetector,
which records the position of the reected beam. Note that the probe beam need not be
perfectly aligned (as long as some part of the beam is reected into the detector, and the
surface does not reect too heavily into the detector), and need not even be smaller than
the detector (since the difference between the quadrant signals allows the determination
of the beam position). The photodetector measures the position of the reected beam,
which in turn gives information about the position of the cantilever and hence the tip. If
the whole apparatus is raster-scanned across the surface (or the sample is scanned under
the microscope), then an image of the surface relief can be generated.
(b)
Figure 5.11 PbSe islands with [001]-type facets: (a) an AFM image of the top surface of a
PbSe/PbEuTe island multilayer; (b) an AFM image of an area of 3 m 3 m of the top
surface of a PbSe/PbEuTe island multilayer. Islands are arranged in a regular array up to the
sixth-nearest neighbor. Reprinted with permission, from J. Stangl, V. Hol y et al., Structural
properties of self-organized semiconductor nanostructures, Figs. 25 and 26, Rev. Mod. Phys.
76, 725783 (2004).
Examples of the quantitative analysis of AFM images are shown in Fig. 5.11. There,
the top surface of PbSe/PbEuTe multilayers is shown. Both materials are semiconductors.
From Fig. 5.11(a), one can see that PbSe forms triangular pyramids with [001]-type side
facets. The lateral ordering can also be analyzed by AFM. In Fig. 5.11(b), a hexagonal
in-plane arrangement of the pyramids is evident.
Transmission electron microscopy and scanning electron microscopy
Among the methods which allowone to see things at the nanometer scale, two types of
electron microscopy play an important role. Transmission electron microscopy (TEM)
makes possible the visualization of thin slices of material with nanometer resolution.
This technique has subnanometer resolution, and, in principle, can resolve the electron
densities of individual atoms. A TEM operates much like an optical microscope, but
uses electrons instead of visible light, since the wavelength of electrons is much smaller
than that of visible light. As we have already discussed, the resolution limitation of any
microscopy is based on the wavelength of the probe radiation. As studied in Chapter 2,
the electron wavelength is much smaller than that of visible light. Since electrons are used
instead of light, glass lenses are no longer suitable. Instead, a TEM uses magnetic lenses
to deect electrons. Beyond this, a TEM is very similar to a conventional microscope,
complete with condenser lenses, objective lenses, and projector lenses.
In a TEM, the electrons are collimated fromthe source and passed through the sample,
and the resulting pattern of electron transmission and absorption is magnied onto a
viewing screen. The image is typically recorded with a charge-coupled-device (CCD)
camera, whose working element is a Si chip with the surface divided into a large array of
pixels sensitive to charge carried by electrons. In scanning electron microscopy (SEM),
Scanning Electron
Microscope
Transmission Electron
Microscope
Source
Condenser
Objective
System
Scan Coils
Sample
Sample
Detector
Screen
Projector Lens
Figure 5.12 Schematic diagrams of a scanning electron microscope (SEM) and a transmission
electron microscope (TEM).
the electron beam is not projected through the whole sample area. Instead, it is raster-
scanned across the surface, and the secondary electrons, or X-rays, emitted from the
surface are recorded. This generates a lower-resolution image, but allows the direct
mapping of surface features, and can even be used for elemental analysis. Both types of
electron microscope are depicted schematically in Fig. 5.12.
The electron microscopy techniques are widely used due to their very high spatial
resolution and sensitivity to composition. TEM can be performed either on thin slices
parallel to the sample surface (plane-view TEM) or on cross-sectional slices. Hence,
buried nanostructures can be examined well by TEM, with some restrictions due to
specimen preparation: in many cases, the lateral island diameter is comparable to the
slice thickness. The image analysis is often not straightforward but requires elaborate
image-analysis techniques and/or model calculations. Compared with other techniques,
usually very small areas are investigated, so that no statistically averaged values can be
obtained.
We present here only one example of atomically resolved transmission electron
microscopy, from which the positions of unit cells, strain, and composition informa-
tion were derived. The results obtained, after digital analysis of the lattice image, are
given in Fig. 5.13: the strain distribution (i.e., distortion of atoms constituting the island
from their regular positions) for an InGaAs island on a GaAs substrate is depicted.
Remarkably, these techniques make it possible to visualize a detailed map of the strain
for an object of size a few tens of nanometers. From Fig. 5.13, it is seen clearly how
5.5 Spontaneous formation and ordering 127
Figure 5.13 Strain distributions obtained from the TEM images of InGaAs islands in GaAs by
using the method of digital analysis of lattice images. Reprinted with permission, from J. Stangl,
V. Hol y et al., Structural properties of self-organized semiconductor nanostructures, Fig. 27(a),
Rev. Mod. Phys. 76, 725783 (2004).
the strain increases at the apex of the InGaAs island, while around the island the strain
changes its sign.
In conclusion, powerful characterization techniques have been developed to study
nanosize objects. The techniques give three-dimensional images in real space and on
an atomic scale in all three dimensions. The methods are nondestructive. They provide
means to perform structural and chemical analyses of the materials used in nanostruc-
tures. Moreover, these techniques make it possible to observe and measure directly the
electron distributions inside the nanostructures; that is, it is possible to observe the
electron probability densities!
5.5 Spontaneous formation and ordering of nanostructures
From the previous section, it is evident that crystal-growth and device-fabrication tech-
niques are highly developed and are already having an impact on nanoscale semiconduc-
tor structures and devices. Every step toward realizing and perfecting articial nanos-
tructures involves a number of new physical and chemical processes and requires very
serious efforts and technical innovations.
Importantly, Mother Nature points to another way to produce nanostructures. Indeed,
the phenomenon of spontaneous formation of periodic domain structures in solids with
a macroscopic periodicity has been known for several decades. Progress in TEM, STM,
and AFM is facilitating the reliable and accurate observation, investigation, and control
of surfaces and periodic structures with the characteristic periodicity of 1100 nm. This
opens a new way to use self-organizing growth processes as well as the formation of
periodically ordered structures on semiconductor surfaces for the direct fabrication of
quantum nanostructures and devices.
d d d
(a) (b) (c)
Figure 5.14 Three types of periodic structures: (a) periodically faceted surfaces; (b) planar
domains; and (c) three-dimensional strained islands.
[001]
[110]
Unreconstructed Reconstructed
Figure 5.15 Reconstruction of Si surface.
There are three distinctive classes of spontaneously formed nanostructures on a sur-
face, as illustrated in Fig. 5.14. These nanostructures are (1) periodically faceted surfaces
(Fig. 5.14(a)), (2) periodic structures of planar domains (Fig. 5.14(b)), and (3) ordered
arrays of three-dimensional coherently strained islands in lattice-mismatched heteroepi-
taxial systems (Fig. 5.14(c)). Despite the fact that the geometries of the three classes
are different, there exist common features for all of these nanostructures. The main one
is that the driving force of the periodic ordering is a long-range elastic interaction. The
elastic-strain eld is created due to the discontinuity of the intrinsic elastic properties on
domain boundaries and/or by the lattice mismatch between two materials composing the
heteroepitaxial system. The three classes depicted in Figs. 5.14(a)(c) are equilibrium
structures. In particular, they can be observed upon annealing of the crystal, or upon
interruption of the crystal growth. Here we restrict our discussion to the formation of
three-dimensional coherently strained islands.
We begin with a brief discussion of the strain arising at surfaces of solids. Since atoms
in the surface layer of any material are in a different environment from that experienced
by those in the bulk, the surface layer energetically favors a lattice parameter different
from the bulk value in the directions parallel to the surface. Being adjusted to the bulk
lattice, the surface layer is intrinsically stretched or compressed. Therefore, the surface is
characterized by intrinsic surface stress. At the surface, even the symmetry of the crystal
can be changed; this is known as surface reconstruction. Figure 5.15 illustrates such a
reconstruction of the Si surface.
FM VW SK
Figure 5.16 Three growth modes of heteroepitaxial systems: Frankvan der Merwe (FM),
VolmerWeber (VW), and StranskiKrastanow (SK).
To some extent the intrinsic surface stress of a solid is analogous to the surface tension
of a liquid. However, there is a fundamental difference between the properties of a liquid
surface and those of a solid surface. Indeed, liquids are typically isotropic. The processes
of formation and deformation of a liquid surface are identical and can be described by
a single parameter that characterizes the energy of cutting of bonds on the surface.
This is the so-called surface energy. However, in crystals the process of the formation
of a surface is quite different: the distance between atoms changes and symmetry can
change. This is quite different from the simple cutting of bonds. In addition, crystals are
anisotropic and the energy needed to create a free surface of a given orientation depends
on the orientation.
These features of the surfaces of solids give rise to different epitaxial growth regimes.
Three growth modes are distinguished traditionally. They are the (1) Frankvan der
Merwe (FM), (2) VolmerWeber (VW), and (3) StranskiKrastanow(SK) growth modes.
Respectively, they can be described as (1) layer-by-layer growth, (2) island growth (three-
dimensional), and (3) layer-by-layer plus island growth. These three modes are illustrated
in Fig. 5.16. The particular growth mode for a given system depends on the surface
energies and on the lattice mismatch between the material of the substrate and that of
the grown layer. The growth regimes discussed in Section 5.2 are of the FM type.
In the following discussion we dene the surface energy as the excess energy of a very
thin distorted layer (one or two monolayers) at the free surface of a solid. The interface
energy is the excess energy of a thin layer at the interface between two solids. In lattice-
matched systems, the growth mode is governed only by interface and surface energies.
Let
1
and
2
be the surface energies of the substrate and epitaxial layer (epilayer),
respectively, and let
12
be the interface energy. Then, if the sum of the epilayer surface
energy and the interface energy is lower than the energy of the substrate surface, i.e.,
1
>
2
12
, the deposited material wets the substrate and the FMgrowth mode occurs.
Achange in
2

12
can drive a transition fromthe FMto the VWgrowth mode. These
two modes of coexistence of crystalline materials are quite analogous to those of a liquid.
For a strained epilayer with small interface energy,
12
, the initial growth may occur
layer-by-layer; however, as the layer becomes thicker it may lower its increasing strain
UF
DI
CI
Figure 5.17 Elastic-strain relaxation during StranskiKrastanow growth (schematic). Light gray
areas denote the substrate, dark gray areas denote the lattice-mismatched epilayer. The lines
symbolize lattice planes. Top, uniformly strained lm (UF); middle, dislocated relaxed islands
(DI); and bottom, coherently strained islands (CI).
energy by forming isolated islands in which strain is relaxed. This results in the SK
growth mode. Thus, the SK mode depends critically on the lattice mismatch. Three
scenarios for strain relaxation are sketched in Fig. 5.17. Figure 5.17(a) corresponds to
uniform pseudomorphic growth without any stress relaxation, which can exist for a very
thin strained layer, as discussed in Section 4.7. Figure 5.17(b) corresponds to a dislocated
relaxed island. Finally, the island in Fig. 5.17(c) is coherently strained. From the point
of view of the use of self-forming nanosize islands, coherently strained structures are of
most importance, because of their high quality.
The scenarios presented in Fig. 5.17 occur as a result of the interplay of several
parameters: (1) the ratio of the surface energy and the energy of the dislocated interface,
(2) the amount of deposited material Q, and (3) the lattice mismatch, c (see Eq. (4.23)
and Fig. 4.16). As an example, a diagram of possible growth regimes is depicted in
Fig. 5.18. In Fig. 5.18, the various growth regimes are identied in terms of the amount
of deposited material, Q, versus lattice mismatch, c.
For a dense system of islands, the elastic interaction between islands via deformation
of the substrate is essential. The system of interacting islands is then a system of elastic
domains where the energy minimum corresponds to a structure with periodic domains.
Thus, there is a possibility of growing an ordered system of nano-islands.
Following this general analysis of formation of three-dimensional islands under pseu-
domorphic growth of crystalline materials, we consider particular examples of such
self-growing nanosize heterostructures.
The rst experimental evidence of the formation of the coherently strained islands
was obtained by transmission electron microscopy of the InAs/GaAs system (GaAs is
the substrate, while InAs is the epilayer). As follows from the data of Table 4.8, this is
the lattice-mismatched system with mismatch parameter c 7%. Coherently strained
and stable islands have been found for many systems: Ge/Si, GeSi/Si (Si substrate of
[100] orientation), AlInAs/GaInAs, InAs/InP, CdSe/ZnSe, and others.
0.0 0.1 0.2
0
1
2
3
4
5
Q
R
1
R
2
R
3
R
1
R
2
R
3
FM SK VW
FM
VW
SK
1
SK
2
1

2 3

Figure 5.18 An equilibrium phase diagram of a lattice-mismatched heteroepitaxial system as a

function of the total amount of deposited material Q and the lattice mismatch c. The small
panels at the top and bottom illustrate the morphology of the surface in the six growth modes
described in the text. The small empty triangles indicate the presence of stable islands, while the
large shaded ones refer to ripened islands. Reprinted with permission from I. Daruka and A.
Barabasi, Dislocation-free island formation in heteroepitaxial growth: a study at equilibrium,
Phys. Rev. Lett., 79, 3708 (1997).
C _
1997 by the American Physical Society
To study the formation of the islands, the growth-interruption method typically is
applied. This method allows one to control the amount of the deposited material even
belowone deposited monolayer at the same growth temperature. For example, at a growth
temperature of 480

C the following features were revealed. When the average thick-
ness of InAs deposition reaches a critical value of 1.61.7 monolayers, a morphological
transition from an InAs layer to three-dimensional InAs islands occurs. After two mono-
layers of InAs have been deposited, an array of well-developed islands is observed. The
average lateral size, the height, and the density of the islands are 100140

A, 5080

A,
and 10
10
10
11
cm
2
, respectively. Figure 5.19 depicts a single InAs island on GaAs
substrate studied by scanning tunneling microscopy.
In general, results of growth and post-growth annealing studies show that the islands
grown in the SKmode are usually stable against ripening. Under given growth conditions
they have well-dened sizes and shapes. For example, in SiGe/Si, essentially only four
forms of islands are observed: (1) shallow pre-pyramids, (2) square pyramids, (3) hut
clusters elongated pyramids, and (4) large domes with facets in several directions.
Figure 5.20 depicts a diagram that links possible shapes of islands with their volume
and the mismatch parameters for Ge or GeSi on Si substrate. In the rst stage of growth,
shallow pre-pyramids appear that later convert to pyramids and then to domes. Pyramids
and domes are clearly observed during growth at higher temperatures, whereas the much
smaller hut clusters form at lower growth temperatures. The typical lateral sizes of the
[110]
[110]
50
Figure 5.19 A single InAs island on GaAs substrate. Reprinted with permission, from J. Stangl, V.
Hol y et al., Structural properties of self-organized semiconductor nanostructures, Fig. 12(b),
Rev. Mod. Phys. 76, 725783 (2004).
Figure 5.20 Shape transition of Ge or SiGe islands grown on Si [001] during growth (solid arrow),
postgrowth annealing (dotted arrow), and Si capping (dashed arrow). The solid curves represent
the critical volumes for pyramids and domes. Corresponding STM images are embedded in the
diagram. Reprinted with permission, from J. Stangl, V. Hol y et al., Structural properties of
self-organized semiconductor nanostructures, Fig. 14, Rev. Mod. Phys. 76, 725783 (2004).
islands are 8001,000

A and their heights are 60100

A. All of these features of island
growth can be explained by analyzing surface and interface energies, as well as the
elastic energy of strained materials. Crystalline anisotropy is also an important factor in
the formation of islands.
As for lateral correlation of the island, the mutual inuence of islands on the same
surface is rather weak and can be observed experimentally only for growth at very
low rates; i.e., close to thermodynamic equilibrium. Typically, the lateral self-ordering
Figure 5.21 AFM images of the surface of a Ge layer grown on lithographically prepatterned Si
[001] substrates. In the sample shown on the top left, the islands are arranged in a regular array
along two orthogonal [110]-directions. In the sample shown on the bottom right, the unit vectors
of the two-dimensional array of pits are oriented along the [100]- and [110]-directions, leading
to a 45
island alignment. X-ray diffraction (top right and bottom left) demonstrates the high
degree of ordering of islands. Reprinted with permission from J. Stangl, V. Hol y et al.,
Structural properties of self-organized semiconductor nanostructures, Fig. 21, Rev. Mod.
Phys. 76, 725783 (2004).
of islands can yield short-range ordered patterns with astonishing regularity; however,
this process does not result in true long-range ordering. Ordering can be realized by
using island nucleation on lithographically prepatterned substrates. Similarly to the case
for nonpatterned substrates, for prepatterned substrates the nucleation of the islands
is affected by the surface curvature and the surface stress. Thus, the substrate can be
prepared to provide initial nucleation at selective locations on the substrate. Various
methods can be used; e.g. shallow etching of trenches in Si and subsequent overgrowth
with a GeSi multilayer, followed by Ge nucleation; and direct growth of Ge in two-
dimensional periodically etched pits. The latter method results in long-range-ordered
island structures, as shown in Fig. 5.21. Nucleation occurs at the intersection of the side
facets within the pits. Thus, a combination of self-organizing processes in crystal growth
with prepatterning methods may be used to achieve precise placement of individual
islands, as well as high ordering throughout nanosize heterostructures.
In conclusion, for device applications, uniformislands with controlled positioning are
required. This goal is difcult to achieve using self-organization alone, but a combination
with other techniques may be successful. The following three routes are most promising:
(1) a combination with conventional lithography will allowcontrolled positioning of self-
organizing nanostructures without losing their inherent advantages; (2) seeded growth
with catalytic nanoparticles facilitates the fabrication of structures signicantly smaller
than the SK islands; and (3) a combination of self-assembly with the techniques of
organic chemistry and biochemistry. Nowadays, the controlled fabrication of extremely
small structures of a fewnanometers in all three dimensions and their selective addressing
seems quite feasible.
5.6 Clusters and nanocrystals
The spontaneously formed and ordered nanostructures studied in the previous section
are merely one type of a number of known examples of nanoclusters and nanocrystals,
which can be grown by using various technological methods.
A nanocluster can be thought of as a size-dependent collection of atoms (from several
atoms to several thousands of atoms). In an ideal case, the cluster is isolated. Isola-
tion means the absence of foreign chemical species within the cluster volume, or on
its surface. Obviously, the simplest way to realize isolation is to synthesize a cluster
under vacuum conditions and then keep it in an inert-gas environment. Typically, an
ideal cluster has a high density of unsatised dangling bonds on its surface. Such ideal
unsupported clusters are not very useful for functional nanostructured materials and
devices. Indeed, one needs to manipulate the clusters, to place them onto a surface in
a certain order, to provide interaction with them, etc. However, if the total number of
atoms in a cluster N is large, the fraction of atoms at the cluster surface which can have
unsatised bonds is of the order of N
2
3
. The ratio of atoms on the surface to N decreases
as 1,N
1
3
and a big cluster is almost an ideal cluster. Importantly, the dangling bonds
have a high reactivity and thus a semiconductor cluster prepared under high-vacuum
conditions will readily be oxidized on exposure to the atmosphere. Practically, cluster
dangling bonds can be terminated articially by using an organic, or inorganic, addi-
tive. Such a passivation of a cluster surface leads to an effective functionalization of the
cluster.
The following two techniques of cluster and nanocrystal fabrication are different from
the growth methods considered in the previous section: gas-phase and colloidal cluster
syntheses.
Under gas-phase cluster synthesis, clusters are grown in a gas, prior to their passivation
and deposition onto a surface. Such clusters are formed when the vapor pressure of
atoms composing the clusters is much larger than it should be under equilibrium at a
given temperature. Nonequilibrium atom vapors can be created by various methods, for
example by laser evaporation of a solid, laser- or thermally induced decomposition of
species containing necessary atoms, etc.
For example, vacuum laser-induced decomposition of silane leads to formation of Si
clusters as ultra-small nanocrystals. Typically, the nanocrystal sizes range from 3 nm to
10 nm. The clusters can be deposited onto the surface of a metal, graphite, or silicon. In
Fig. 5.22, an STM image of Si clusters is presented. One can see the atomic structure
of the Si-(111) surface and several Si clusters of various sizes. On the left of the image,
the detailed structure of a larger Si cluster is observable. Interestingly, the surface of the
Si crystalline substrate oriented normally to the (111) direction has a large reactivity to
5.6 Clusters and nanocrystals 135
Figure 5.22 Silicon nanoclusters absorbed on a (111) surface of Si. Several clusters, of differing
sizes, are seen on this STM image. Reprinted with permission from D. W. McComb, B. Collings
et al. An atom-resolved view of silicon nanoclusters, Chem. Phys. Lett. 251, 812 (1996).
Permission from Elsevier.
Si clusters. This explains the large sticking coefcient for Si clusters. As a result, on
being deposited onto the surface, these clusters do not diffuse onto the substrate at room
temperature.
Compound semiconductors can also be synthesized in the gas phase. For example,
laser evaporation of materials containing Ga and As atoms in an inert gas produces
GaAs clusters with dimensions of tens of nanometers. If these clusters are exposed to
air, they are capped by an As-rich oxide shell. Then, the clusters can be deposited
onto a surface (Si-(100), for instance) and manipulated by using an atomic-force
microscope.
The colloidal synthesis of almost monodisperse nanoparticles is based on controlled
nucleation and growth of clusters in a precursor-containing solution. Both metallic and
semiconducting clusters can be grown. The most studied case is synthesis of IIIV
and IIVI compounds. The method involves injection of reagents containing the cluster
constituents (for example, Cd and Se) into a hot solvent, where nucleation of CdSe
occurs. Molecules of the solvent cap and thereby passivate the nucleated cluster. For
the above example, tri-n-octylphosphide (TOPO) is used as a solvent. The reagents are
Me
2
Cd and TOPSe.
A hot solution of TOPO is favorable for nucleation of CdSe. TOPO capping the
clusters signicantly slows the cluster growth and makes the properties of the product
more controllable. Careful control of the solution temperature enables production of
nanoparticles with a small size dispersion. The size dispersion of nanocrystals can be
within a few percent of the average diameter. The latter is a few to tens of nanome-
ters. The shape of nanocrystals is close to spherical. On being deposited onto a surface,
these nanocrystals form an ordered lateral structure. In Fig. 5.23, a tunneling elec-
tron microscope image is presented, which provides evidence of the formation of a
50 nm
Figure 5.23 A transmission electron microscopy image of CdSe nanocrystals of size 4.8 nm
grown by the colloidal method and then deposited on a surface. X-ray diffraction demonstrates
the high ordering of these nanocrystals. Reprinted with permission, from P. Moriarty,
Nanostructured materials, Fig. 19(b), Rep. Prog. Phys. 64, 324 (2001).
C _
IOP Publishing
Limited.
self-organized 10-nm CdSe nanocrystal superlattice. Interestingly, articial passivation
of these clusters modies intercluster interaction. The latter is responsible for intercluster
spacing. Thus, the ordering of CdSe and other clusters on a surface can be controlled by
the regime of their fabrication in the colloid. In general, colloidal chemistry is conve-
nient for engineering of basic properties of nanocrystals, including their shape, size,
and state of surface. For example, more sophisticated nanocrystals prepared in a colloid
can include a multilayered coating: the CdSe clusters can be covered by a layer of HgS
followed by a layer of ZnSe and so on.
Clusters and nanocrystals possess a number of properties different from those of
bulk materials. Indeed, they occupy an intermediate place between individual natural
atoms/molecules and bulk crystals. For semiconducting materials, the clusters have
electron energy bands practically the same as those of corresponding bulk materials.
However, their small sizes restrict electron motion, giving rise to electron connement
and quantization in all three directions. As a result, the fundamental electrical, optical,
and mechanical properties are modied. In the next chapter we will study some of these
modications.
5.7 Methods of nanotube growth
From the discussion of carbon nanotubes in Chapter 4, it is clear that the properties
of these nano-objects are considerably different from the properties of other nanos-
tructures and nanodevices fabricated on the basis of bulk-like materials. The same
is valid for methods of growth of nanotubes. In this section, we consider several
growth methods for carbon nanotubes. Basically, these methods are arc-discharge, laser-
ablation, and chemical-vapor deposition. Arc-discharge and laser-ablation methods were
Figure 5.24 Catalytic growth of a (10, 10) armchair carbon nanotube using a metal atom (large
black sphere) as a catalyst. Carbon atoms are depicted by white balls. Reprinted with
permission, from J. Charlier and S. Iijima, Growth mechanisms, in M. S. Dresselhaus,
G. Dresselhaus, and Ph. Avouris (Eds.), Carbon Nanotubes, Fig. 11, Topics in Applied Physics
vol. 80 (Berlin, Springer-Verlag, 2001), 5579.
historically the rst approaches used to fabricate nanotubes. Chemical-vapor-deposition
growth methods are new and highly promising for scale-up of defect-free nanotube
materials.
Carbon nanotubes are constituted solely of carbon atoms. Thus, any growth method
for these nanostructures should rst provide carbon atoms, and then condense the car-
bon vapor under certain nonequilibrium conditions at temperatures below the graphene
melting point (4100 K).
Carbon-vapor condensation can result in several different forms of carbon-containing
products: fullerenes, graphitic layers, nanotubes, etc. To obtain a considerable yield of the
nanotubes, corresponding growth regimes should be used. For example, relatively high
temperatures (10001300 K) are necessary to form single-walled nanotubes. Indeed,
the rolling-up of a graphene sheet into a tube requires additional energy, especially for
small diameters of single-walled tubes. Then, it was found that the key role in nanotube
growth can be played by catalytic processes involving transition metals (iron, nickel,
cobalt, etc.). Particular catalytic mechanisms are complicated.
Figure 5.24 illustrates one such mechanism for growing a (10, 10) armchair nanotube
(white balls) with a Ni (or Co) atom(large black sphere) chemisorbed onto the open tube
edge. Though Co and Ni atoms are strongly bound, they are still very mobile at the edge
of the growing tube. Consequently, the metal catalyst keeps the tube open as a result
of its mobility around the open edge, ensuring that any pentagons or other high-energy
local structures are rearranged to hexagons. The latter occurs through the exchange
mechanism: the metal catalyst assists two incoming carbon atoms (or a C
2
molecule) in
the formation of carbon hexagons, thus increasing the tubes length.
I
He
Arc-discharge
CVD
Laser ablation
C
n
H
m
Catalyst
(c)
(b)
(a)
Laser
pulse
Figure 5.25 Schematic diagrams of arc-discharge (a); laser ablation (b); and chemical-vapor-
deposition (c) methods of growth of carbon nanotubes.
Arc-discharge and laser ablation
In an arc-discharge, carbon atoms are evaporated by a plasma of helium gas ignited
by high currents passed through opposing carbon anode and cathode as illustrated by
Fig. 5.25(a). Evaporationof carbonatoms inanarc-discharge is a stronglynonequilibrium
process, thus these atoms should undergo sequential condensation at temperatures below
the melting point of graphite (3000 K). Carbon nanotubes arise as one of the products
of this process. The synthesized multi-walled nanotubes have lengths of the order of
10 m and diameters in the range 530 nm. The nanotubes are typically bound together
by molecular van der Waals interactions and form tight bundles.
For the growth of single-walled tubes, a metal catalyst is needed in the arc-discharge
system. For example, producing substantial amounts of single-walled nanotubes is pos-
sible by arc-discharge with the use of a carbon anode containing a small percentage
of cobalt catalyst in the discharge camera. As a result, abundant single-walled nan-
otubes are generated in the soot material. Optimization of the growth of single-walled
carbon nanotubes in an arc-discharge is achieved by using a carbon anode contain-
ing a large atomic percentage of a transition metal (for example, up to 4% nickel
catalyst).
Figure 5.26 A bundle of carbon nanotubes grown by laser ablation. Reprinted with permission,
from Hongjie Dai, Nanotube growth and characterization, in M. S. Dresselhaus,
G. Dresselhaus, and Ph. Avouris (Eds.), Carbon Nanotubes, Fig. 4, Topics in Applied Physics,
vol. 80, pp. 2953 (Berlin, Springer-Verlag, 2001).
The process of growth of high-quality single-walled nanotubes takes place also during
the so-called laser-ablation (laser-oven) method illustrated by Fig. 5.25(b). The method
utilizes intense laser pulses to ablate a carbon target containing 0.5 atomic percent of
catalytic nickel and cobalt. The target is placed in a tube-furnace heated to 1200
C.
During laser ablation, a ow of inert gas is passed through the growth chamber to
carry the grown nanotubes downstream to be collected on a cold nger, where carbon
condensation occurs. The resulting tubes are primarily in the form of ropes consisting
of tens of individual nanotubes close-packed into hexagonal crystals via the van der
Waals interactions. Packaging of individual nanotubes into a bundle is clearly seen in
Fig. 5.26.
Typically, arc-discharge and laser-ablation methods lead to a number of by-products:
fullerenes (graphitic polyhedrons, some with enclosed metal particles), amorphous car-
bon, etc. Further purication is necessary to obtain the nanotubes. This process involves
reuxing the nanotubes in a nitric acid solution for an extended period of time, which oxi-
dizes away amorphous carbon particles and removes metal catalyst species. Production
of single-walled nanotubes of high quality by laser ablation and arc-discharge is rela-
tively simple and has already led to the wide availability of samples useful for studying
their fundamental properties.
Chemical-vapor deposition
A schematic experimental setup for chemical-vapor-deposition growth is depicted in
Fig. 5.25(c). The growth process involves heating a catalyst material to high temper-
atures in a tube furnace and owing a hydrocarbon gas through the tube reactor for a
period of time. Materials grown over the catalyst are collected upon cooling the systemto
room temperature. Parameters controlling nanotube growth are the hydrocarbons, cata-
lysts, and growth temperature. The active catalytic species are typically transition-metal
nanoparticles formed on a support material such as alumina. The growth mechanism
involves the dissociation of hydrocarbon molecules catalyzed by the transition metal,
and dissolution and saturation of carbon atoms in the metal nanoparticle. The precip-
itation of carbon from the saturated metal particle leads to the formation of tubular
carbon solids. Tubule formation is favored over formation of other forms of carbon
such as graphitic sheets with open edges. This is because a tube contains no dangling
bonds.
Relatively high temperatures (10001300 K) are necessary to form single-walled
nanotubes with small diameters and allow the production of nearly defect-free nanotube
structures. Among all hydrocarbon molecules, methane is the most stable at high tem-
peratures against self-decomposition. The methane chemical-vapor-deposition approach
is promising for enabling scale-up of defect-free nanotube materials to the kilogram or
even ton level.
Directed growth of single-walled nanotubes
Ordered, single-walled nanotube structures can be grown directly by methane chemical-
vapor deposition on catalytically patterned substrates. Consider, for example, a method
developed to grow suspended nanotube networks on substrates containing lithograph-
ically patterned silicon pillars. The growth starts with developing a liquid-phase cat-
alyst precursor material that has the advantage over solid-state catalysts of allowing
the formation of uniform catalyst layers for large-scale catalytic patterning on surfaces.
The precursor material consists of a triblock copolymer, aluminum, iron and molybde-
num chlorides in mixed ethanol and butanol solvents. The aluminum chloride provides
an oxide framework when oxidized by hydrolysis and calcination in air. The triblock
copolymer directs the structure of the oxide framework and leads to a porous cata-
lyst structure upon calcination. The iron chloride also can lead to catalytic particles
needed for the growth of nanotubes. The catalyst precursor material is rst spun into
a thin lm on a polydimethyl siloxane stamp, followed by contact printing to transfer
the catalyst precursor selectively onto the tops of pillars pre-fabricated on a silicon sub-
strate. The stamped substrate is calcined and then used in chemical-vapor-deposition
growth.
Remarkably, the nanotubes grown fromthe pillar tops tend to be directed frompillar to
pillar. The directed growth of suspended single-walled nanotubes is presented in Fig. 5.27
for three different congurations of the pillars: (a) a nanotube power-line-like structure,
(b) a square of nanotubes, and (c) an extensive network of suspended nanotubes. Such
a directed growth can be understood as follows. Nanotubes are nucleated only on the
tower-tops since the catalytic stamping method does not place any catalyst materials on
the substrate below. As the nanotubes lengthen, the methane ow keeps the nanotubes
oating and waving in the wind since the ow velocity near the bottom surface is
substantially lower than that at the level of the tower-tops. This prevents the nanotubes
from being caught by the bottom surface. The nearby towers on the other hand provide
5.8 Chemical and biological methods 141
Figure 5.27 Pillar tops are connected by suspended single-walled nanotubes that form (a) a
nanotube power-line-like structure; (b) a square of nanotubes; and (c) an extensive network of
suspended nanotubes. Reprinted with permission, from Hongjie Dai, Nanotube growth and
characterization, in M. S. Dresselhaus, G. Dresselhaus, and Ph. Avouris (Eds.), Carbon
Nanotubes, Fig. 9, Topics in Applied Physics, vol. 80 (Berlin, Springer-Verlag, 2001), pp. 2953.
xation points for the growing tubes. If a waving tube contacts an adjacent tower, the
tubetower van der Waals interactions will catch the nanotube and hold it aloft. By such a
growth mode the suspended nanotubes can be made of macroscopic length. For example,
tubes longer than 150 m can be grown.
Growth of isolated nanotubes on a specic surface site
It is of importance that the chemical-vapor-deposition method allows one to grow
individual nanotubes at specic sites on at SiO
2
substrates. The approach involves
methane chemical-vapor deposition onto substrates containing catalyst islands patterned
by electron-beam lithography. Nanotube chips with isolated nanotubes grown from the
islands have already been obtained. This growth approach readily leads to tubes orig-
inating from well-controlled surface sites, and makes possible the development of a
controlled method to integrate nanotubes into addressable structures.
5.8 Chemical and biological methods for nanoscale fabrication
The technologies studied in previous sections are based primarily on physical processes
and methods. The use of chemical and biological methods opens new ways for nanofab-
rication. It is obvious that chemical synthesis can produce a wide range of nanoparticles,
including inorganic, organic, and biological nanosystems. In addition, chemistry and
biology can be complementary to existing physical nanofabrication methods.
In particular, selected techniques of chemistry and biology portend alternatives to
conventional photolithography. One approach is based on self-assembly phenomena. It
is known that there are molecules that, due to physical, biological, or chemical processes,
tend to congregate into long-range periodic structures. They can facilitate the realization
of regular periodic masks on the surface of a metal or a semiconductor with a character-
istic period of 1050 nmand with internal pores of diameter 525 nm. Such resolution is
superior to that of conventional photolithography, which is now of the order of 100 nm.
In addition, the self-assembly methodology has potential for being much less expensive
and less time-consuming than the electron-beam lithography techniques currently used
for writing such small structures.
The chemistry methods can be used to tailor the chemical composition and structure
of a surface. This can be achieved through a new direct-write tool dip-pen nanolithog-
raphy, which generates surface-patterned chemical functionality with a length scale of
1100 nm.
In turn, achievements in biological science generate principally new methodologies
for preparing nanostructured materials with predened and synthetically programmable
properties from given inorganic building blocks.
Chemical and biological innovations for nanotechnologies are briey considered in
this section.
Chemical self-assembly of nanoscale structures
Here we consider chemical systems based on block copolymers. They exhibit self-
ordering, which can be used for nanopatterning.
A block copolymer is a macromolecule that consists of several polymer blocks, which
typically can be grown in a few steps. In Fig. 5.28 an example of a triblock polymer
is presented. The synthesis of this macromolecule starts with the polymerization of the
styrene monomer. The amount of monomer supplied for the rst step is only sufcient
to create a miniature styrene chain (the rst block) with an average degree of polymer-
ization of 9. (Note: the degree of polymerization represents the number of monomeric
units in a macromolecule or a block.) A second block with a similar average degree of
polymerization grows when another chemical isoprene is added. In the third step,
carbon dioxide (CO
2
) installs carboxyl end groups at the top of the previous miniature
diblock copolymers. One terminus of the triblock (the lower) is hydrophobic while the
opposite terminus is hydrophilic. These triblock molecules have a rodcoil architecture
with a stiff rodlike segment (the second block) covalently connected to more torsionally
exible segments. Chemical processes, typically, lead to a polydispersity of resulting
triblock molecules; however, their dispersity is small: the ratio of the mass-averaged to
number-averaged molecular masses ranges from 1.06 to 1.1.
Several triblock polymers can compose different clusters, as illustrated in Fig. 5.28.
On being deposited onto a surface, triblock polymers form a macromolecular thin lm
with a high degree of self-ordering, as can be seen from the scheme in Fig. 5.29. The
triblock polymer lm has a typical standing-up structure. The shaded circle indicates
a chemisorbing headgroup and the open circle an endgroup, which can be chosen from
Figure 5.28 Molecular graphics of the triblock self-assembling molecule (left) and a cluster of
triblock macromolecules (center and right), shade-coded for energy in the various sectors of the
cluster. Reprinted with permission from S., Stupp, V. LeBonheur et al., Supramolecular
materials: self-organized nanostructures, Science, 276, 384 (1997).
C _
1996 AAAS.
a variety of chemical functionalities. A top view of a real lm obtained by transmission
electron microscopy is depicted in Fig. 5.30. The evident ordering of macromolecules
in the lm is driven by the stiff rodlike segments of the triblocks. Specically, a balance
of attractive and repulsive forces mediates the formation of macromolecular cells, as
for a crystal. In general, block copolymers easily cover the surfaces of various metals
and semiconductor crystals and form two-dimensional periodic structures of various
symmetries ranging from low-symmetry to square and hexagonal structures. Articially
built polymer lms have thicknesses ranging from tens to hundreds of nanometers with
pores of various sizes and various shapes.
End group
Head group
Substrate
Backbone
Figure 5.29 A schematic diagram of a triblock polymer lm; the polymer is shown standing-up.
After F. Schreiber, Self-assembled monolayers: from simple model systems to
biofunctionalized interfaces, J. Phys.: Condens. Matter, 16, R881 (2004).
C _
IOP Publishing
Limited.
Figure 5.30 A transmission electron micrograph of a lm formed by the triblock molecules,
revealing regularly sized and shaped aggregates that self-organize into superlattice domains.
Reprinted with permission from S. Stupp, V. LeBonheur et al., Supramolecular materials:
self-organized nanostructures, Science, 276, 384 (1997).
C _
1997 AAAS.
There are two ways to create nanostructures using self-assembly of block copolymers:
(i) precipitation of metal or other inorganic crystals onto a polymer mask attached to a
substrate; and (ii) using the polymer mask for the subsequent processing of the substrate,
or as a support for additional auxiliary masks. After the mask has been formed on the
surface of a semiconductor, the device-fabrication technology remains the same as with
the conventional nanofabrication and employs dry ion or selective etching, oxidation,
metallization, diffusion, selective growth, etc.
Biological methods
As examples of the use of biological methods as elements of nanoscale technologies, we
consider nanopatterning and nanoassembly techniques that employ proteins and DNA,
Amino Group
Side R Group
Carboxylic Acid Group
H
H H OH
C C
R
N
O
Figure 5.31 A generic amino acid.
as well as the use of biochemical molecular-recognition principles for the assembly of
nanoscale inorganic building blocks into macroscopic functional materials.
First, let us consider some of the basic properties of proteins. Proteins are formed by
binding together sequences of amino acids to form chain-like molecules whose links
are amino acids. Proteins consisting of only a few amino acids are known as peptides.
Important subgroups of proteins are enzymes and antibodies. Proteins constitute one
of the most important classes of biomolecules and they are sometimes referred to as
the machines of the body in view of the large number of diverse functions that they
perform. As mentioned previously, amino acids are the building blocks of all proteins.
Amino acids are among the most important molecules found in biological systems. In
fact, by binding amino acids together to form chains of amino acids, it is possible to
synthesize all known peptides (short chains of amino acids), all known proteins (long
chains of amino acids), and all antibodies (selected proteins). Antibodies are proteins
that have high afnities for binding to specic molecules known as antigens. Thus, an
antibody in a cellular environment has a high probability of binding to its antigen if
the antibody and antigen are in close proximity. Peptides, proteins, and antibodies are
enormously important in biology. As just one example, proteins function as gates that
control the ow of ions into and out of neurons.
Figure 5.31 illustrates a generic amino acid. All amino acids are derived from the
generic structure shown in Fig. 5.31 by varying the side group, R. When two amino
acids are in close proximity to each other, the carboxyl group on the end of one amino
acid has an afnity to bind to the amino group on the end of the other amino acid. When
this bonding occurs, the amino group gives up an H atom, and the carboxyl group gives
up an OH moiety. The liberated OH and H form H
2
O and an amide bond, CONH,
remains as the bond linking two amino acids. In Fig. 5.32, three of the twenty common
amino acids are depicted. Glycine (G) is the amino acid with the simple side group
of a single hydrogen (H) atom. Arginine (R) and aspartic acid (D) are also shown in
Fig. 5.32. As discussed previously, when amino acids like glycine (G), arginine (R), and
aspartic acid (D) are in water, as is generally the case in biological systems, the amino
(NH
2
) group on the terminus of one amino acid has an afnity (tendency) for binding
to the carboxyl (COOH) group on another amino acid; after such a binding event, the
two amino acids are bound together by an amide bond (CONH bond), and a water
molecule, H
2
O, is produced that becomes part of the surrounding water environment.
Glycine
(Gly or G)
CH
H
COOH
NH
2
Arginine
(Arg or R)
NH
2
CH
2
CH
2
CH
2
COOH CH
NH
NH
2
NH
C
Aspartic Acid
(Asp or D)
CH
2
OH
O
C
CH COOH
NH
2
Figure 5.32 Examples of some amino acids.
Such a process provides a very simple example of how chemistry may be used to self-
assemble molecules and structures. By binding a few amino acids together a peptide is
formed. From this discussion it is clear that amino acids are indeed the building blocks
of many biomolecules, including peptides, proteins, and antibodies. As explained and
illustrated previously, these biomolecules peptides, proteins, and antibodies may be
formed through binding of amino and carboxylic groups at the two ends of the amino
acids. The full set of twenty common amino acids has the following members: alanine
(A), arginine (R), asparagine (N), aspartic acid (D), cysteine (C), glutamine (Q), glycine
(G), glutamic acid (E), histidine (H), isoleucine (I), leucine (L), lysine (K), methionine
(M), phenylalanine (F), proline (P), serine (S), threonine (T), tryptophan (W), tyrosine
(Y), and valine (V).
Just as the binding of NH
2
and COOH groups plays a role in linking many
biomolecules, they may be used to bind a COOH-functionalized (COOH-coated) quan-
tum dot to the amino terminus of a peptide, protein, or antibody. In this way, we mimic
a self-assembly technique found in nature as a key step in our arsenal of nanofabrica-
tion techniques. It is also possible to use another chemical bond found in nature the
thiol bond to provide a frequently used technique for assembling nanodevice struc-
tures. Specically, the amino acid cysteine, depicted in Fig. 5.33, and several other
amino acids (leveine, valine, methionine, and serine), have special uses in chemical self-
assembly since each of these amino acids contains a sulfur (S) atom in its side group, R.
Specically, the sulfur atom has an afnity for binding to a (111) surface of gold and it
binds to quantumdots like CdS through an SS or thiol bond. These thiol bonds are also
instrumental in the phenomenon of protein folding, where they cause cysteine molecules
at various points along a given protein to be attracted to each other, thus causing protein
folding. Indeed, these SS bonds are the strongest bonds found of all possible bonds
among amino acids in protein systems.
Leveine
(Lev or L)
CH COOH NH
2
CH
CH
2
CH
3
CH
3
Valine
(Val or V)
CH COOH NH
2
CH
2
CH
3
CH
3
Methionine
(Met or M)
CH COOH NH
2
CH
2
CH
3
CH
2
S
Cysteine
(Cys or C)
CH COOH NH
2
CH
2
SH
Serine
(Ser or S)
CH COOH NH
2
CH
2
OH
Figure 5.33 Additional examples of amino acids.
The afnity of sulfur to bind gold is sufcient to result in the wide use of SAu
bonds to bind molecules to Au surfaces. In the eld of molecular electronics, there are
efforts to use molecules as wires, and it is necessary to bind these wires to metal contacts.
By chemically binding a sulfur atom to the end of such a molecular wire, it is possible
to bind the S-functionalized ends of the wire to Au. The use of these SAu bonds in
molecular electronics is so prevalent that they are referred to as the alligator clips of
molecular electronics.
Nowthat we have considered basic properties of proteins, we turn to their applications
in nanofabrication. It is known that proteins form the external surface of many bacteria.
Analogously to the case of block copolymer lms analyzed previously, surface-layer
proteins can be used as nanoscale biological masks. As shown schematically in Fig. 5.34,
surface-layer proteins can be isolated frombacterial cells, and subsequently reassembled
on the surfaces of a solid. The surface-layer proteins can be deposited upon various
metals and semiconductor crystals. On their surfaces, surface-layer proteins form two-
dimensional periodic structures of various symmetries. The sizes of the elementary cells
of these surface crystals range from 3 nm to 30 nm. The thickness of an articially built
surface layer is about 28 nm, and it has pores of various sizes (28 nm) and shapes. In the
upper part of Fig. 5.35, one such periodic structure is shown. The gure illustrates typical
processing steps for nanopatterning: (a) deposition of surface-layer protein crystals onto
the substrate; (b) metallization by deposition of metal; (c) dry etching, which allows the
transfer of the pattern to the substrate; and (d) a plane viewof the resulting nanostructure
with highly ordered holes.
liquid
solid
reassembly
isolation of

S-layer proteins
S
-
l
a
y
e
r

p
r
o
t
e
i
n
bacterial cell
envelope
Figure 5.34 A schematic drawing of the isolation of surface-layer proteins from bacterial cells
and their reassembly into crystalline arrays in suspension at a solid support.
Substrate
Protein
S-layer
Pore
(a)
Mask deposition
Hole sites
Protein
Dry etching
(c)
Metal (oxide)
Metal overlayer
Metal shadowing
(b)
Protein
(d)
Hole sites
Metal (oxide)
on top of
protein
Figure 5.35 Processing steps for nanopattern transfer: (a) deposition of S-layer protein crystals
onto the substrate; (b) shadow metallization by electron-beam deposition; (c) dry etching to
transfer the pattern to the substrate; and (d) a plane view of the idealized nanostructure. After
T. A. Winningham, S. G. Whipple, et. al., Pattern transfer from a biological nanomask to a
substrate via an intermediate transfer layer, J. Vac. Sci. Technol., 19, 17961802 (2001).
Reprinted with permission from Thomas A. Winningham, Steven G. Whipple, and Kenneth
Douglas, Journal of Vacuum Science & Technology B, 19, 1796 (2001).
C _
2001 AVS The
Science & Technology Society.
Figure 5.36 A scanning electron microscopy image of GaAs semiconductor wires obtained by
means of the surface-layer protein technology. The wires have a length of 80 nm and a diameter
of about 8 nm. Reprinted with permission, from M. Haupt, S. Miller et. al., Nanoporous gold
lms created using templates formed from self-assembled structures of inorganic-block
copolymer micelles, Advanced Materials, 15, 829831 (2003), Fig. 3(a), p. 831.
Then, such a surface superlattice of holes can be used, for example, to fabricate
semiconductor nanoscale free-standing wires. Figure 5.36 depicts a scanning electron
microscopy image of semiconducting GaAs wires fabricated using surface-layer protein
technology. The wires have a length of 80 nm and a diameter of about 8 nm at the top
of each wire, while the base is thicker. High ordering of free-standing nanoscale wires
is obtained using this biology-based method of nanopatterning.
Biological methods facilitate the realization of the assembly of nanoscale inorganic
building blocks into functional materials. As an example, the biological macromolecule
DNA may be used as a synthetically programmable assembler of nanoscale structures
in order to fabricate a wide variety of different ensembles of selected nanocomponents.
Such ensembles may be assembled in a variety of architectures. This method is based on
the molecular-recognition properties associated with DNA. DNA may be prepared and
functionalized with virtually any chromophore, acceptor, donor, or an active group in
an automated fashion. It is known that the molecular-recognition properties arise due to
base pairing in DNA strands. Specically, the commonly encountered double-stranded
DNA is composed of two single strands of DNA. These single strands are composed
of linear sequences of the four DNA bases adenine (A), cytosine (C), guanine (G), and
thymine (T). The two single strands bind together to form a single strand as a result of
the high afnities for G and T to bind together and C and A to bind together. Thus, as
a simple example, GTCAC and TGACA bind together to form a double-stranded DNA
molecule. The two single strands, GTCAC and CAGTG, are complementary to each
other. These molecular-recognition processes can guide the assembly of nanoparticles
into extended structures. In principle, the method selects nanoparticles with certain
chemical compositions and sizes, and it controls also the distance and coupling between
the particles in the resulting nanostructured materials.
A
B
A
A A
A
A
A
B
B
B
B B
B
A B
Modification with
3
' ' ' '
'
'
thiol TACCGTTG 5
Modification with
5 AGTCGTTT 3 thiol
Au nanoparticles
Addition of linking DNA duplex
5 ATGGCAAC
TCAGCAAA 5
Further oligomerization
and settling
A
B
A
A
A
A
A
A
B B
B
B
B
B
B
B
A
A
B
B
A
B
A
A
B
B
A
B
A
B
Figure 5.37 Formation of a biomoleculeinorganic cluster hybrid. Reprinted with permission,
from C. A. Mirkin, Programming the assembly of two- and three-dimensional architectures
with DNA and nanoscale inorganic building blocks, Inorg. Chem., 39, 22582272 (2000).
C _
As nanoparticle building blocks, nanocrystals of metals, semiconductors, mag-
netic particles, fullerenes, etc. are currently available. The sizes of these nanocrys-
tals/nanoparticles can be controlled precisely, in many cases from 1 nm to 20 nm in
diameter.
The formation of a biomoleculeinorganic cluster hybrid is one of the important bio-
logical methods for creatingnanoscale architectures. This process canbe illustratedbythe
following example. Let two different sequences of non-complementary eight-base-pair
DNA be synthesized with alkanethiol endgroups. Then, two batches of Au nanoparti-
cles are functionalized with these DNA strands. In the scheme presented in Fig. 5.37 this
functionalizing process is shown in the upper part of the gure. Two particular sequences
of DNA molecules are indicated as well. If the two batches of Au particles were simply
mixed, there would be no DNA recognition and hence no particle aggregation. However,
the addition of linker DNA strands changes the situation. Such a linker molecule is com-
posed of a DNA molecule that has three regions: a central region of double-stranded
DNA and two end regions that are composed of unpaired single strands of DNA. These
unpaired single-stranded ends are known as sticky ends. If the linker DNA molecules
contain eight-base-pair sticky ends that are complementary to the base pairs attached
Figure 5.38 A transmission electron microscopy image of a two-dimensional structure of Au
nanoparticles linked by DNA molecules. Reprinted with permission, from C. A. Mirkin,
Programming the assembly of two- and three-dimensional architectures with DNA and
nanoscale inorganic building blocks, Inorg. Chem., 39, 22582272 (2000).
C _
American
Chemical Society.
to the Au particles, the particles start to aggregate, as shown in the lower part of the
scheme in Fig. 5.37. The process of linking these nanoparticles with DNA molecules is
known as oligomerization since DNA molecules are also known as oligonucleotides.
In this example, the so-called sticky ends, 5
/
ATGGCAAC and TCAGCAAA5
/
are sin-
gle strands of DNA bound to opposite ends of a segment of double-stranded DNA.
The double-stranded portion of this molecule is represented by the ladder symbol in
Fig. 5.37. As is illustrated in Fig. 5.37, the 5
/
ATGGCAAC sticky end binds to the
3
/
thiolTACCGTTG complementary molecule which is bound to an Au quantum dot by
this bond. Likewise, the TCAGCAAA5
/
sticky end binds to the 5
/
AGTCGTTT3
/
thiol
complementary molecule which is bound to an Au quantum dot by the thiol bond.
Thus, an object can be built from nanosized building blocks linked by DNA molecules.
By such a method, both two-dimensional and three-dimensional ordered nanosys-
tems have been fabricated. Figure 5.38 depicts a transmission electron microscopy
image of a two-dimensional structure of Au nanoparticles linked by DNA molecules.
Similarly, semiconductor nanoclusterDNA hybrids have been used to construct
extended macroscopic structures of CdSe/ZnS quantum dots interconnected with DNA
molecules. The wavelength of light emitted by a quantum dot is determined by its size as
well as by the material. Table 5.1 summarizes the bandgaps of some direct-bandgap mate-
rials that are used to fabricate quantum dots. The materials represented in Table 5.1 have
wavelengths spanning the ultraviolet, visible, and infrared regions of the electromagnetic
spectrum.
When we integrate DNA with inorganic building blocks, we can exploit some of the
properties of the latter. For example, semiconductor nanocrystals hybridized with DNA
can be used as uorescent biological labels, because of specic optical properties of
Table 5.1 The bandgaps of the direct-bandgap
bulk materials used for fabrication of quantum dots
Compound semiconductor Bandgap (eV)
AlN 6.15
CdS hexagonal 2.4
CdS cubic 2.55
CdSe hexagonal 1.75
CdSe cubic 2.17
CdTe 1.49
PbS 0.41
PbSe 0.27
ZnS 3.68
GaN 3.36
these nanosclusters such as their emission at well-dened wavelengths. These optical
properties will be discussed in the next chapters.
Although DNA arguably is the most tailorable and versatile molecule for organizing
nanoscale materials into extended structures, its use has some limitations. Most notably,
it is not a high-temperature material, and therefore the structures initially generated from
DNA interconnects will not be stable at elevated temperatures.
As illustrated previously, NH
2
and COOH groups play a role in linking many
biomolecules, and may be exploited further in interconnecting ensembles of quantum
dots. Consider the case in which ZnS-coated CdSe quantum dots (QDs) coated with
carboxyl groups are functionalized with GGGC peptides. These CdSeZnS QDs func-
tionalized with GGGC peptides will be denoted simply as CdSeZnSGGGC. This
commonly used notation does not specify the number of peptides bound to each quan-
tum dot. In practical applications, the number of peptides may vary from one to twenty
or more depending upon the application. In the present case, where the outermost amino
acid is a cysteine molecule with its sulfur-containing side group, the C amino acids
may be used for chemical self-assembly. Indeed, if, as the rst step in the fabrication
process, an Au wire is immersed in a beaker containing a typical density of 10
16
CdSe
ZnSGGGCcomplexes per cm
3
, many of the CdSeZnSGGGCcomplexes will bind to
(111) surfaces of the Au wire since the S atoms in the cysteine side groups have afnities
for binding to (111) surfaces of Au. In the next fabrication step, the CdSeZnSGGGC-
coated Au wire is immersed in a beaker containing 10
16
CdS nanocrystals per cm
3
, and
there is subsequent binding of CdS nanocrystals to the C amino acids of the GGGC
biomolecules that are attached to the CdSeZnS nanocrystals that are bound to the Au
wire. Upon repeated alternating immersions of the quantum-dot functionalized wire in
the two beakers of CdSeZnSGGGC and CdS, the nanocrystals are assembled as illus-
trated in Fig. 5.39. The nanocrystals assembled in this manner have densities in excess
of 10
17
per cm
3
! These densities are orders of magnitude higher than those achievable by
current commercially used semiconductor-device-fabrication techniques. This example
illustrates the use of chemical self-assembly to integrate ensembles of semiconductor
nanostructures. To create electrically functional integrated semiconductor nanocrystal
Figure 5.39 A quantum-dot network assembled using chemically directed assembly techniques.
networks using chemically directed assembly with molecular links, it is necessary to use
molecular links that conduct electricity.
Since proteins (and short proteins peptides) constitute one of the major classes of
biomolecules, an understanding of the electronic properties of these general classes of
molecules is of potential importance in bioelectronics. Charge transport in peptides has
recently been studied by inserting amino acids containing a side group containing a
natural chromophore, which is a molecule that produces charge carriers when exposed
to light. In this way, charge can be introduced into the peptide by illuminating the chro-
mophore in the side group. In addition, phenylalanine, tryptophan, and tyrosine have
been considered theoretically as possible conducting elements of peptides since their
side groups are rich in conductive -bonds. As we discussed for conducting polymers,
such -bonds provide charge conduction. Results from the chromophore-based stud-
ies of charge transport indicate that charges move along these peptide-based wires at
speeds of about 1.5 10
5
cm s
1
, a relatively slow speed from the standpoint of Si-
and GaAs-based electronic devices. This relatively slowtransport implies that electronic
systems of integrated nanocrystals with peptide-based interconnects must be designed
with architectures that overcome the limitations imposed by carrier-transport speeds in
peptides; perhaps, the design of massively parallel architectures offers a possible reso-
lution of the difculties imposed by the limited propagation speed in the peptide-based
interconnects.
Our previous discussion of the use of DNA as a self-assembling agent makes it clear
that DNAmay be used to link metallic and semiconducting nanostructures. There are also
current research efforts aimed at using DNA as molecular wires that conduct electric
current. Of course, these efforts stem from the growing interest in the use of DNA
as a charge-carrying element in bioelectronic devices. In these efforts, a key parameter
determining the change of transport properties of DNAis the highest occupied molecular
orbital (HOMO). As shown in Fig. 5.40, the commonly encountered type of DNA is a
double-stranded helix-shaped molecule with each of the two strands being composed
of sequences of the following four bases: adenine (A), cytosine (C), guanine (G), and
thymine (T). As discussed previously, these sequences may contain the A, C, G, and T
bases in any order, and the adjacent bases on the two strands are pairs of either GC or
e
h+
h+ h+
h+
Light
Quantum Dot
DNA
E
v
h+
E
HOMO
Figure 5.40 Transition of a hole from a quantum dot to DNA.
AT. For a simple double strand of DNA that contains only GC pairs, the energy levels
of the DNA exhibit a gap of the approximate magnitude of 2.0 eV between the HOMO
and the lowest unoccupied molecular orbital (LUMO).
In molecules, energy states may be lled, partially lled, or empty just as in semicon-
ductors. As discussed earlier in this book, the valence band of an undoped semiconductor
is nearly full of electrons at low temperatures, and the conduction band is nearly empty
under these conditions. For molecules, it is traditional to refer to the HOMO simply as
the HOMO level or the HOMO band. The highest energy in the HOMO band, there-
fore, plays a role analogous to the high-energy edge, or top, of the valence band, E
v
, of
a semiconductor. Similarly, there is an energy gap between the top of the HOMO band
and the low-energy edge, or bottom, of the next (empty) molecular orbital, known as the
LUMO, which is referred to frequently as the LUMO band. The bottom of the LUMO
band plays a role analogous to the low-energy edge of the conduction band, E
c
.
One of the current approaches to studying charge transport in DNA is to bind DNA
to a semiconductor quantum dot and to inject charge into the quantum dot as shown in
Fig. 5.40. As shown in the lower portion of Fig. 5.40, it is desirable in such experiments
to pick a semiconducting material for the quantum dot that has some of its energy states
aligned with some of the orbitals of the DNA. In the case illustrated in Fig. 5.40, the
valence-band energy is selected so that it falls just below the HOMO of the DNA. For
a DNA molecule that is composed of a pure strand of G bases bound to a pure strand
of C bases, the top of the HOMO band has an energy of 7.34 eV as measured relative
to the vacuum level. Two readily available colloidal semiconductor quantum dots with
valence bands having energies close to the HOMO band energy are TiO
2
and ZnO.
TiO
2
has conduction- and valence-band energies of 4.21 eV and 7.41 eV relative
to the vacuum level, respectively. ZnO has conduction- and valence-band energies of
4.19 eV and 7.39 eV relative to the vacuum level, respectively. For either of these
semiconductors, the energy alignment of the valence-band edge with the HOMO band
is as shown in Fig. 5.40. With such a band alignment, holes in the semiconductor may
be injected into the DNA.
In practical experiments on charge transport using these QDDNA structures, light
with energy greater than the bandgap of ZnO (3.2 eV) or TiO
2
(also 3.2 eV) is used to
create electronhole pairs in the quantum dot. Since ZnO and TiO
2
are both indirect-
bandgap materials, these electronhole pairs do not recombine as rapidly as they do in
direct-bandgapmaterials. (This is one of the reasons whyTiO
2
is usedinsolar cells, where
there is a need to maximize the production of electric currents from photo-produced
carriers.) As shown in Fig. 5.40, the photo-produced holes may escape the quantum dot
and move into the DNAwire. Alternatively, we may viewthis process as an electron near
the top of the HOMOband making a transition into the quantumdot, where it recombines
with a hole in the valence band. Experiments based upon these techniques are currently
being conducted by several research groups to assess the current-carrying capability
of DNA. One of the interesting ndings emerging from such studies is that charges
(holes) tend to become trapped near guanine-rich regions of DNA molecules. This is
not surprising, since the ionization potentials of GC and AT base pairs are 7.34 eV
and 7.99 eV, respectively. Since these ionization potentials correspond to the energies
required for an electron to transition from the top of the HOMO band to the (unbound)
continuum, it follows that the HOMO band-edge energy has a local maximum near
guanine-rich sites along the DNA. Accordingly, a hole that is propagating along the
DNA wire and dissipating its energy as it propagates by producing vibrational modes
in the DNA, as an example may become trapped at the guanine-rich sites. The charge-
transport properties of DNA are still not understood fully. However, if conducting DNA
wires may be engineered by selecting appropriate base sequences, it may be possible to
use DNA not just as a self-assembling agent of complex ensembles of nanocomponents,
but also as an electrically active linking elements in these ensembles of nanostructures. It
may even be possible to design novel types of ultra-high-density information-processing
systems that greatly exceed the maximuminformation-processing capabilities realizable
by the downscaling of todays integrated circuits! At present, it is reasonable to expect
that years of research will be required to determine whether such nanostructure-based
chemically self-assembled systems are useful for advanced information processing.
Dip-pen nanolithography
Since the invention of the scanning tunneling microscope and similar techniques, there
have been attempts to develop new nanolithography methods. In particular, STM- and
AFM-based methods were applied to oxidize, scrape, or etch nanostructures on surfaces.
Some of these methods were mentionedinprevious sections. However, proposedmethods
are generally limited to the growth of thin oxides on selected metal and semiconductor
surfaces, or to multistep etch procedures that cannot be generalized to parallel and
productive patterning of nanostructures.
Finally, dip-pen nanolithography (DPN) was introduced as a direct-write scanning-
probe-based lithography in which an AFMtip is used to deliver chemical reagents directly
to nanoscopic regions of a target substrate, as shown in Fig. 5.41.
The DPN technique is a type of soft lithography, where the word soft refers to the
chemical composition of the nanostructures which can be fabricated. They are made
Solid Substrate
Water Meniscus
Tip with Nanoscale
Dimensions
Molecule to be
Assembled
Writing Direction
Molecular
Transport
Figure 5.41 The principal writing element of a dip-pen lithographic system is a nanoscale
structure having a tip with nanoscale dimensions. Fluids owing over the surface of the tip are
deposited onto the surface of a substrate in the region where a meniscus is formed between the
tip and the surface. Reprinted with permission, from C. A. Mirkin, Programming the assembly
of two- and three-dimensional architectures with DNA and nanoscale inorganic building
blocks, Inorg. Chem., 39, 22582272 (2000).
C _
of organic ligands rather than solid-state materials. Though it is unlikely that dip-pen
technologies can displace conventional solid-state fabrication methods, they are highly
complementary. Many interesting scientic and important practical issues pertaining to
miniaturization can be addressed as a result of having these soft lithography methods,
whereby molecules can be patterned in a controlled fashion on the sub-100-nm length
scale. For example, such a method can generate nanoscale molecule-based conduct-
ing structures and provide their contacting with macroscopically addressable electrodes
prepared via conventional microfabrication methods.
In inventing DPN, the idea was to miniaturize a 4000-year-old technology: the technol-
ogy of the quill or dip-pen. The largest difference is that we wanted to do on a nanoscopic
scale what a quill pen can do on a macroscopic scale.
Considering the fundamentals of DPN, rst we should mention that the simple idea
of transporting any ink through a nanoscale AFM tip to a surface via the well-known
capillary action does not work properly. Instead, the basic idea is to design a system with
an ink that would chemically react with a substrate of interest. The scheme in Fig. 5.41
represents the main DPN elements: the moving AFM tip is coated with ink molecules;
a water meniscus forms between the tip and the solid substrate; ink molecules react
with the substrate. Such an effect of reaction of a chemical with a surface is called
chemisorption. Thus chemisorption acts as a driving force for moving the molecules
from the tip to the substrate and then results in formation of stable one-molecule-thick
nanostructures. The key to making this approach successful is to select organic molecules
with low water solubilities so that the chemisorption driving force facilitates the control
of tip-to-substrate transport properties. This would prevent the uncontrolled, nonspecic
adsorption and accumulation of multilayers of molecules on the surface.
Figure 5.42 illustrates the results of chemisorption. The panel on the left shows the
coating painting of a 1-m
2
area of the gold surface with octadecanethiol via the
use of DPN. The formation of a one-monolayer structure is proved by the more detailed
(a) (b)
Figure 5.42 (a) The painting via DPN by octadecanethiol on gold. (b) A lattice-resolved image of
an octadecanethiol nanostructure deposited on a single crystal of gold. Reprinted with
permission, from C. A. Mirkin, Programming the assembly of two- and three-dimensional
architectures with DNA and nanoscale inorganic building blocks, Inorg. Chem., 39, 22582272
(2000).
C _
lattice-resolved image of the transported octadecanethiol, as depicted in the panel on the
right of Fig. 5.42. The observed lattice is hexagonal with an intermolecular spacing of
5

A, the known lattice constant for an octadecanethiol monolayer on gold formed by a
number of other methods.
The application of the method is shown in Fig. 5.43. The panel on the left shows a
molecular grid of octadecanethiol on gold. The grid is a single molecule thick, with 100-
nm-wide lines. The panel on the right demonstrates an array of uniformly deposited 450-
nm-diameter dots. Note that attaining this type of shape regularity and reproducibility
would not be possible with a conventional pen, even over macroscopic dimensions.
Currently, writing with DPN generates a nanostructure with approximately 6070 nm
linewidths. The smallest fabricated structures are 15-nm-diameter dots spaced
5 nm apart. DPN has been developed to pattern a variety of inksubstrate combinations.
The method is compatible with many inks, from small organic molecules to organic
and biological polymers and from colloidal particles to metal ions and sols; patterned
surfaces range from metals to insulators and semiconductors. In Table 5.2, examples of
combinations of chemical inks and surfaces are listed.
In conclusion, chemistry and biological methods can be applied for fabrication of
nanosystems. Importantly, these new methods can be complementary with the conven-
tional semiconductor nanotechnologies.
5.9 Fabrication of nanoelectromechanical systems
In previous sections, we studied state-of-the-art semiconductor technologies that may be
used to produce nanoscale structures and devices for electronics. High-quality structures
Table 5.2 Examples of ink--substrate combinations used in DPN
Ink Substrate Notes
Alkylthiols Au 15 nm resolution on single-crystalline surfaces,
50 nm on polycrystalline surfaces
Ferrocenylthiols Au Redox-active nanostructures
Silazanes SiO
x
Patterning on oxides
GaAs
Proteins Au, SiO
x
Both direct write and indirect assembly
Conjugated polymers SiO
x
Polymer deposition veried spectroscopically
and electrochemically
DNA Au, SiO
x
Sensitive to humidity and tip-silanization conditions
Fluorescent dyes SiO
x
Luminescence patterns
Metal salts Si, Ge Electrochemical and electrolytic deposition
Colloidal particles SiO
x
Viscous solution patterned from tip
(a) (b)
Figure 5.43 The nanoscale grid (left) and the dot array (right) generated via DPN. Reprinted with
permission, from C. A. Mirkin, Programming the assembly of two- and three-dimensional
architectures with DNA and nanoscale inorganic building blocks, Inorg. Chem., 39, 22582272
(2000).
C _
were necessary in order to obtain superior electronic properties. The great achievements
and the advances brought about in mainstream electronics by these technologies can be
used for the exploration of other nanodevices commonly called nanoelectromechanical
systems (NEMSs). This class of devices includes nanomachines, novel sensors, and a
variety of new devices that function on the nanoscale.
Nanomechanical devices promise to revolutionize measurements of extremely small
displacements and extremely weak forces, particularly at the molecular scale. Indeed,
with surface and bulk nanomachining techniques, NEMSs can now be built with masses
approaching a few attograms (1 attogram = 10
18
g) and with cross-sections of about
Mechanical
system
transducer
Control Electrical
control signal
Output
transducer
Electrical
output signal
Input
transducer
Electrical
input signal
Mechanical
response
Mechanical
stimulus
Mechanical
perturbation
Figure 5.44 A schematic representation of a three-terminal electromechanical device.
10 nm. The small mass and size of NEMSs give them a number of unique attributes that
offer immense potential for newapplications and fundamental measurements. In general,
the potential applications of NEMSs are likely to be enormous and could benet a diverse
range of elds, ranging from nanoelectronics to medicine and biotechnology. In this
section, we study the basic concepts of NEMSs, the technology for NEMS fabrication,
and the challenges arising in this eld.
An electromechanical device can be thought of as a two-, three-, or, generally, multi-
terminal transducer that provides input stimuli (i.e., signal forces), and reads out a
mechanical response (i.e., output displacement). At additional control terminals, elec-
tric signals can be applied and subsequently converted by the control transducers into
varying forces to perturb the properties of the mechanical element in a controlled
and useful manner. The generic picture of a NEMS is shown in Fig. 5.44, where
the input, output, and control transducers/terminals are presented schematically. The
basic mechanical element of a NEMS is a nanosize suspended lm, a membrane, or a
beam; in the following discussion, we will use the term beam. Easy exural deforma-
tions of such mechanical elements provide high mechanical responsivity for a NEMS.
Electronic devices to which the beam is coupled are assumed to be of comparable
dimensions.
Let us consider the fabrication of the basic mechanical component of a NEMS. The
crystal and heterostructure growth, and processing techniques, which have been stud-
ied previously are used widely to produce suspended semiconductor structures. These
techniques can be applied to bulk silicon, epitaxial silicon, and systems based on IIIV
compounds and to other materials.
The procedure for fabricating a suspended structure is illustrated in Fig. 5.45. In
its simplest form, the procedure starts with a heterostructure that contains structural
(b) (a)
(d) (c)
Figure 5.45 A general scheme for the fabrication of NEMS suspended structures using structural
(gray) and sacricial (dark) layers on a substrate (dark gray). (a) Three-layer base
heterostructure, (b) etching-mask deposition, (c) anisotropic etching, and (d) selective wet
etching of the sacricial layer.
(gray) and sacricial (dark) layers on a substrate (dark gray), as in Fig. 5.45(a). Masks
on the top of this heterostructure can be patterned by a combination of optical and
electron-beam lithography, followed by thin-lm-deposition processes. The resulting
mask protects the material from beneath during the next stage; see Fig. 5.45(b). Unpro-
tected material around the mask is then etched away using a plasma etching process
as in Fig. 5.45(c). Finally, a local chemically selective etching step removes the sac-
ricial layer from the specic regions to create a freely suspended nanostructure that
is both thermally and mechanically isolated; see Fig. 5.45(d). This procedure can be
repeated several times and combined with various deposition processes to produce
the necessary mechanical nanostructure for a particular device. The exibility of the
process allows one to apply this general scheme to diverse materials and to fabri-
cate fully suspended structures with lateral dimensions of approximately a few tens of
nanometers.
Consider, for example, the important case of silicon nanomachining by the use of
the so-called SIMOX (Separation by IMplantation of OXygen) process. The procedure
starts from a Si wafer, which is processed by a large dose of oxygen-ion implantation.
The implanted wafer is annealed at high temperature to form a SiO
2
layer of 0.051 m.
Above the SiO
2
layer formed, a single Si crystal layer is then grown. This top layer is 0.1
0.2 m thick. As the result, one obtains a Si/SiO
2
/Si heterostructure that corresponds
Figure 5.46 Four different NEMSs fabricated by SIMOX method. From A. N. Cleland,
Foundations of Nanomechanics, Fig. 11.4 (Berlin, Springer-Verlag, 2003).
to the gray/dark/dark gray structure of Fig. 5.45. All stages are already illustrated in
Fig. 5.45: the top Si layer is patterned and an etching mask is created as in Fig. 5.45(b);
to the top Si layer anisotropic etching is applied as shown in Fig. 5.45(c); then, the oxide
is subjected to selective wet etching to create nally a suspended structure as depicted
in Fig. 5.45(d). The SIMOX process can be carried out with wafers of large area (4 to
6 inches in diameter) and facilitates the integration of a number of NEMSs and other
electronic devices on a chip. Figure 5.46 illustrates some suspended structures fabricated
by this method.
In conclusion, due to advanced technologies, fabrication of new nanoelectromechani-
cal systems becomes possible. These systems have dimensions sosmall that their mechan-
ical motion (vibrations) are coupled to the electrons much more strongly than in the case
of bulk-like, massive samples. A number of nanodevices can be built on the basis of
NEMSs. Chapter 8 provides additional discussion of these devices.
5.10 Closing remarks
In this chapter, we focussed on methods used for high-quality material growth and for
nanodevice fabrication. We studied the growth of perfect crystals and multilayered het-
erostructures. We found that there has been a considerable and persistent improvement
of traditional semiconductor technologies for material processing. We also found that
these advances facilitate the fabrication of nanostructures and nanodevices with excel-
lent precision and high reproducibility, and with the necessary electrical, optical, and
mechanical properties.
Moreover, we analyzed novel approaches to producing nanostructures based on special
regimes of material growth(the StranskiKrastanowregime), inwhichnanostructures are
formed spontaneously due to the growth kinetics. The processes of self-organization and
self-ordering of nanostructures (quantumdots, quantumwires, etc.) give rise to a newand
unique way to produce high-density ensembles of nanodevices with desired properties.
Apart from the traditional methods of fabrication, those adapted for microelectronics
and nanoelectronics, we studied fabrication techniques for such original nano-objects
as nanotubes.
We described both improved and novel characterization and growth methods that
facilitate the control of nanostructure geometries with atomic-scale precision as well as
the manipulation of a single atom or ion in a crystal.
The latest achievements in the chemical and biological sciences have generated new
approaches to nanofabrication that are complementary to conventional semiconduc-
tor technologies. We presented several examples of the new chemical and biological
methods.
Finally, we studied in detail the fabrication of a new class of nanodevices, nanoelec-
tromechanical systems. Such nanosized systems are so small that the motion of a few
electrons may strongly affect the mechanical vibrations of nanostructures.
More information on crystal growth and device fabrication can be found in the fol-
lowing books and papers:
R. H. Hendel, S. S. Pei et al., Molecular-beam epitaxy and the technology of
selectively-doped heterostructure transistors, in GalliumArsenide Technology, ed.
D. K. Ferry (Indianapolis, IN, Howard W. Sams, 1985).
K. Ploog, Delta-doping in MBE grown GaAs: concept and device application, J.
Cryst. Growth, 81, 304 (1987).
D. Bimberg, M. Grundman, and N. N. Ledentsov, Quantum Dot Heterostructures
(Chichester, John Wiley & Sons, 1999).
The following publications are devoted to self-organized growth of nanostructures:
R. N otzel, Self-organizing growth of quantum dot structures, Semicond. Sci. Tech-
nol. 11, 1365 (1996).
V. A. Shchukin and D. Bimberg, Spontaneous ordering of nanostructures on crystal
surfaces, Rev. Mod. Phys., 71, 1125 (1999).
J. Stangl, V. Hol, and G. Bauer, Structural properties of self-organized semiconductor
nanostructures, Rev. Mod. Phys., 76, 725 (2004).
Scanning tunneling and atomic force microscopy techniques are described in the book
C. J. Chen, Introduction to Scanning Tunneling Microscopy (New York, Oxford Uni-
versity Press, 1993).
Detailed analysis of fabrication methods of carbon and other nanotubes is presented in
M. S. Dressellhaus, G. Dressellhaus, and P. C. Eklund, Science of Fullerenes and
Carbon Nanotubes (San Diego, CA, Academic Press, 1996).
In the following two publications the chemical and biological nanofabrication methods
are reviewed:
C. A. Mirkin, Programmingthe assemblyof two- andthree-dimensional architectures
with DNAand nanoscale inorganic building blocks, Inorg. Chem., 39, 22582272
(2000).
5.11 Problems 163
D. S. Ginger, H. Zhang, and C. A. Mirkin, The evolution of dip-pen nanolithography,
Angewandte Chem., 43, 3045 (2004).
A description of fabrication of various types of NEMSs is presented in
A. N. Cleland, Foundations of Nanomechanics (Berlin, Springer-Verlag, 2003).
5.11 Problems
1. Describe the main differences between the Czochralski method of crystal growth
and epitaxial growth. Which of these approaches can be applied to grow multilayered
crystalline structures?
2. In the molecular-beam epitaxy method, the rate of crystal growth is characterized
by the ux density, J, of atoms, constituting the growing lm. For binary crystals AB,
like GaAs, SiGe, etc., the uxes of deposited components A and B are to be equal:
J
A
= J
B
= J,2. Let the crystal density
AB
be given. Using given J.
AB
, and masses of
atoms AandB, calculate the time necessarytogrowa lmof thickness d. Estimate numer-
ically the growth time for GaAs lm of thickness 100 nm at J = 10
15
atoms cm
2
s
1
;
(
GaAs
= 5.316 g cm
3
).
3. When a photolithographic method is applied, the effect of diffraction of light restricts
the minimal scale of an illuminated pattern necessary for further processing to fabricate
a nanostructure. Explain the advantages in the use of short-wavelength light sources in
nanolithography.
Assume that the minimal thickness of a light line d
min
is related to the wavelength
of illumination as d
min
,2. Calculate and compare the minimal scales of devices
fabricated by exploitation of three laser sources: a red HeNe laser ( =0.63 m), a UV
KrF laser ( = 0.243 m), and a UV ArF laser ( = 0.19 m).
4. Apply the hydrogen model of Eq. (5.4) for energy levels of donors in GaAs and
InAs. Use the effective masses according to Table 4.5 and set the dielectric constants, c,
equal to 12.8 and 15.5, respectively. Calculate the ionization energies and estimate the
radius of ground donor states for these materials. Find the average number of primitive
cells covered by a single donor electron.
5. Explain the role of lattice mismatch for regimes of self-organizing growth of nanos-
tructures. For which of the following heterostructures is the formation of nano-islands
possible: GaAs/AlAs (the lattice constants are a
0
= 5.64 and 5.66

A, respectively),
GaAs/InAs (a
0
= 5.64 and 6.06

A), and Si/Ge (a
0
= 5.43 and 5.65

A).
6. Discuss the differences between the working principles of the nanoscale techniques
of scanning tunneling microscopy and atomic-force microscopy.
7. Scanning tunneling microscopy (STM) is a new and important technique
used to probe and characterize nanostructures. Explain why atomic-scale precision
measurements are possible when a macroscopic tip of dimension about 1 m is used.
What are the limitations for applications of STM?
8. Some chemical and biological methods of formation of surface nanopatterns can
be complementary to standard semiconductor technologies. Which of these meth-
ods can be used for creation of periodic patterns and which are for arbitrary surface
patterning?
6 Electron transport in semiconductors
and nanostructures
6.1 Introduction
In previous chapters we studied advances in materials growth and nanostructure fabri-
cation. In the case of electrons, we paid primary attention to the quantization of their
energy in nanostructures. In fact, electronics relies upon electric signals, i.e., it deals with
measurements of the electric current and voltage. Controlling and processing electric
signals are the major functions of electronic devices. Correspondingly, our next task will
be the study of transport of charge carriers, which are responsible for electric currents
through nanostructures.
The possible transport regimes of the electrons are dependent on many parameters and
factors. Some important aspects of these regimes can be elucidated by comparing the
time and length scales of the carriers with device dimensions and device temporal phe-
nomena related to operating frequencies. Such an analysis is carried out in Section 6.2.
In Sections 6.3 and 6.4 we discuss the role of electron statistics in transport effects.
Then, we consider the behavior of the electrons in high electric eld, including so-called
hot-electron effects. Analyzing very short devices, we describe dissipative transport and
the velocity-overshoot effect. Finally, we consider semiclassical ballistic motion of the
electrons and present ideas on quantum transport in nanoscale devices in Section 6.5.
6.2 Time and length scales of the electrons in solids
We start with an analysis of possible transport regimes of the electrons in nanostructures.
Since there is a large number of transport regimes, we introduce their classication in
terms of characteristic times and lengths fundamentally inherent to electron motion.
Electron fundamental lengths in solids
As we already noted in Section 4.3, the characteristic length in a crystalline solid is
the lattice constant a
0
. However, the scales which are relevant to the charge carriers are
typically much larger than a
0
. This fact, as stressed in Chapter 4, allows one to neglect
ne crystalline structure and consider an electron as an almost free particle, in particular
by assigning to the electron an effective mass that may differ fromthe mass of an electron
in vacuum.
166 Electron transport
1
2
3
4
Dimensionless electron effective mass (m
*
/m
0
)
E
l
e
c
t
r
o
n

w
a
v
e
l
e
n
g
t
h
,

l

(
)

T = 300 K
Figure 6.1 Electron wavelength versus the electron effective mass for room temperature
(T = 300 K). Points 1 through 4 correspond to InSb, GaAs, GaN, and SiC, respectively.
The rst fundamentally important length is the de Broglie wavelength of an electron
in a solid. For a free particle this length was introduced in Chapter 2; see Eq. (2.45). For
an electron in a semiconductor nanostructure with the effective mass m
the de Broglie
wavelength is typically greater than that of a free electron
0
:
=
2h
--
p
=
2h
--
2m
E
=
0
_
m
0
m
.
0
=
2h
--
2m
0
E
. (6.1)
where E is the electron energy and m
0
is the mass of the electron in vacuum. In Fig. 6.1
the value is shown as a function of m
,m
0
. Points 14 on the curve indicate wavelengths
for electrons in InSb, GaAs, GaN, and SiC, respectively. We have used effective masses
m
,m
0
equal to 0.014, 0.067, 0.172, and 0.41, respectively, for these materials and we
have assumed that the electron energy is E = k
B
T. Here T = 300 K is the ambient
temperature and k
B
is Boltzmanns constant. We see that the de Broglie wavelength of
an electron in typical semiconductors with m
in the range (0.011)m

0
is of the order
of 73073

A; i.e., it is really much larger than the lattice constants for the materials
presented in Table 4.8. As the temperature decreases to 3 K, the de Broglie wavelength
increases by one order of magnitude. Thus the wavelength becomes comparable to the
sizes of semiconductor structures and devices fabricated by modern nanofabrication
technology.
Size of a device and electron spectrum quantization
Let us introduce a geometrical size of a semiconductor sample L
x
L
y
L
z
, as shown
schematically in Fig. 6.2. Without loss of generality we assume that L
z
- L
y
- L
x
. If
6.2 Time and length scales 167
L
z
L
y
L
x
Contacts
(a) (b)
Figure 6.2 (a) Geometrical sizes of a semiconductor sample (L
z
- L
y
- L
x
) and (b) a sample
with contacts; the electron transport occurs along the x-direction.
the system is free of randomness and other scattering mechanisms are sufciently weak,
the electron motion is quasiballistic and the only length with which the geometrical sizes
need be compared is the electron de Broglie wavelength . Since only an integer number
of half-waves of the electrons can t into any nite system, instead of a continuous energy
spectrumand a continuous number of the electron states, one obtains a set of discrete elec-
tron states and energy levels, each of which is characterized by the corresponding number
of half-wavelengths. This is frequently referred to as quantization of electron motion.
Depending on the dimensions of the system, one can distinguish the following cases.
(a) The three-dimensional or bulk-like case, when the electron spectrum quantization is
not important at all,
_ L
x
. L
y
. L
z
. (6.2)
and an electron behaves like a free particle characterized by the effective mass m
.
(b) The two-dimensional or quantum-well case, when the quantization of the electron
motion occurs in one direction while in the other two directions electron motion is
free:
. L
z
_ L
y
. L
x
. (6.3)
Such a case was discussed in Chapter 3 for the example of a potential energy depen-
dent on a single coordinate. The electron energy for this case is given in the form of
two-dimensional subbands, as described by Eq. (3.49).
(c) The one-dimensional or quantum-wire case, when the quantization occurs in two
directions, so that the electron moves freely only in one direction along the wire:
L
z
. L
y
. _ L
x
. (6.4)
Such a case was discussed in Chapter 3 for the example of a potential energy depen-
dent on two coordinates. The electron energy for this case is given in the form of
one-dimensional subbands, as in Eq. (3.50).
(d) The zero-dimensional or quantum-box (quantum-dot) case, when the quantization
occurs in all three directions and the electron can not move freely in any direction:
L
x
. L
y
. L
z
. . (6.5)
The simplied models for this case also were analyzed in Chapter 3. The energy
spectrum is discrete.
The last three cases also illustrate the quantum size effects in one, two, and three
dimensions, respectively. If at least one geometrical size of a device is comparable to the
electron wavelength, a quantum-mechanical treatment of the problemis strictly required.
Let us analyze the conditions when and the reasons why carriers lose their wave-
like behavior so that they can be considered as classical particles. There are two major
reasons. The rst is non-ideality of the system, which leads to electron scattering. The
second is related to nite temperature and electron statistics.
Electrons in solid-state devices are subjected to scattering by crystal imperfections,
impurities, lattice vibrations, interface roughness, etc. These scattering processes are
divided into two groups: elastic and inelastic. In classical physics, an elastic collision
leads to a change only in the particle momentum (wavevector), whereas in an inelastic
collision both the momentum and the energy change. An essential property of an elastic
collision is that it does not destroy the phase of the electron. Indeed, after an elastic
scattering the energy remains unchanged and the electron wavefunction +(r. t ) consists
of different components, which are of type e
iOt
e
ik
j
r
. All components have the same time-
dependent phase e
iOt
. Thus, the spatial distribution of the electron density [+(r. t )[
2
=
[(r)[
2
remains independent of time. In other words, elastic scattering does not destroy
the coherence of electron motion. The same is true for the case of two or more impurities:
the spatial wave pattern is generally complex, but it remains coherent.
Using the semiclassical language, if
e
refers to the mean time between two elastic
scattering events, we can dene the mean free path of the electrons between elastic
scattering events as l
e
= :
e
, where : is the average electron velocity. Therefore, even
for distances exceeding l
e
the wave-like properties of electrons are coherent.
Inelastic scattering leads to a new result. This scattering produces electron waves
with different energies and the resulting wavefunction has a complex dependence on
both position and time; the beating of different wave components in time washes out
the coherence effects. Let
E
be the mean time between two inelastic collisions. The
distance the electron propagates between these collisions is frequently called the inelastic
scattering length, L
E
. The electron preserves its quantum coherence for distances less
than L
E
and it loses coherence for larger distances. Generally, L
E
> unless extremely
nonequilibriumconditions exist. Often L
E
far exceeds the mean free path, l
e
. In this case,
the electron undergoes many elastic collisions before losing its energy. This process is
known as diffusion and its displacement during
E
is known to be
L
E
=
_
D
E
(
E

e
). (6.6)
where the diffusion coefcient D is given by D = :
2
e
,, with = 3 for a three-
dimensional electron gas, = 2 for a two-dimensional electron gas, and = 1 for a
one-dimensional electron gas. Usually,
E
and L
E
decrease as the temperature of the
system increases.
The second reason for averaging out the quantumbehavior is the temperature effect on
the electron statistics. Indeed, at nite temperatures, there are electrons with signicantly
different energies and this leads to a large spreading of the wavefunction phases and the
coherence in the electron system is destroyed. One can estimate the characteristic length
L
T
associated with such a temperature-related smearing of the coherence. Electron
energy broadening of the order of k
B
T leads to a spreading of the phases with time,
t , that goes as L = t (k
B
T,h
--
). Thus, the time of spreading, t , can be estimated as
the instant
T
, for which L 1, i.e.,
T
= h
--
,(k
B
T). If the only scattering is elastic
scattering, an electron diffuses in space over a distance of about
Dt during the time t

exceeding the mean-free-ight time
e
. Therefore, during the time interval
T
one obtains
the thermal diffusion length L
T
=
D
T
=
_
Dh
--
,(k
B
T). At distances exceeding L
T
,
the coherence of the electron will be lost.
In fact, the dephasing effects caused by inelastic collisions and the temperature spread-
ing of phases exist simultaneously. The spatial scale associated with the loss of quantum-
mechanical coherence should be determined by the smaller of these two lengths:
l
. min{L
E
. L
T
]. (6.7)
Electron transport is determined by the wavefunction, i.e., by the superposition of the
scattered electron waves. From the above considerations we can conclude that the coher-
ence length, l
, denes the limit belowwhich electron transport has a quantumcharacter.

Devices with geometrical sizes of the order of the coherence length are no longer char-
acterized by macroscopic material parameters such as conductivity, average velocity,
etc. Such systems are called mesoscopic systems. The proper theory for the description
of mesoscopic devices is therefore quantum theory; the properties of such mesoscopic
systems are determined by wave-like phenomena and are thus strongly dependent on
the geometry of the sample, contacts, positions of scatterers, etc. For cases in which the
transport distance, L
z
, is long compared with l
, the device can be described within the

framework of classical physics.
Quantum and classical regimes of electron transport
Let us compare the previous discussion of fundamental lengths with characteristic device
sizes to illustrate and explain possible electron-transport regimes. For simplicity, we
suppose that transport occurs along one dimension, say the x direction. The total current
in each of the other two directions is zero, but these transverse sizes of the device can be
important too.
Quantum and mesoscopic regimes of transport
We can dene two non-classical regimes. If the de Broglie wavelength exceeds the length
of a device L
x
,
L
x
. (6.8)
and l
e
, electron-transport is described in terms of the quantum ballistic transport
regime. If the coherence length, l
(also referred to frequently as the dephasing length),

exceeds L
x
and ,
l
> L
x
. . (6.9)
electron transport is described in terms of the mesoscopic transport regime.
The classical transport regime
In the case when the size L
x
exceeds the dephasing length,
L
x
> l
. (6.10)
electron transport is described in terms of the classical regime. If the dimension L
x
is
less than the mean free path,
l
e
> L
x
. (6.11)
electron transport is described in terms of the classical ballistic regime, which means
that electrons can move through the device along classical trajectories without collisions.
If the dimension L
x
is greater than the mean free path,
L
x
l
e
. (6.12)
electron transport is of a diffusive nature. If L
x
l
E
l
e
. electrons do not lose their
energy in moving across the device. Such transport is called quasiballistic transport. In
the absence of an electric eld the electrons preserve their energy under the quasiballistic
regime. On combining the above-discussed inequalities with inequalities (6.2)(6.4), one
can see that there are three classical transport regimes for one, two, and three-dimensional
electrons.
If the transverse dimensions, L
z
and L
y
, are bothgreater thanthe de Broglie wavelength
but they are comparable to one of the characteristic classical lengths, the transport regime
is characterized by transverse classical size effects. In this case, collisions with the device
boundaries affect electron transport through the device. For example, if one or both of
the transverse dimensions are of the order of the mean free path,
L
z
. L
y
l
e
.
the resistance of the device depends strongly on the properties of the side boundaries.
Roughness of the boundaries increases the resistance and entirely controls it if L
z
. L
y
_
l
e
.
If the transverse dimensions become comparable to one of the diffusion lengths, we are
dealing with another kind of classical size effect, namely, diffusive classical size effects.
For example, if L
z
or L
y
is of the order of the energy relaxation length L
E
the device
boundaries provide an additional energy-relaxation channel. This diffusive size effect
controls the mean energy of nonequilibriumelectrons. For convenience the classication
of the possible transport regimes is presented in Table 6.1.
Table 6.1 Classification of transport regimes
Intercontact distance, L
x
, is comparable to
Quantum regime the electron wavelength, L
x

Mesoscopic regime Intercontact distance is less than the dephasing length,
L
x
l
Classical regime (one-, two-, and Intercontact distance exceeds the dephasing length,
three-dimensional electron L
x
> l
:
transport) classical ballistic regime, l
e
L
x
quasiballistic regime (energy-conserving):
L
E
L
x
l
e
. l
transverse size effects:

effect related to the mean free path, L
z
. L
y
l
e
diffusion effects, L
z
. L
y
L
E
Time scales and temporal (frequency) regimes
The time scales which characterize transport phenomena determine the temporal and
frequency properties of materials and devices.
There are two fundamental times dening the character of electron-transport behavior:
the time between two successive scattering events, or the free-ight (scattering) time,
e
;
and the time which characterizes the duration of a scattering event,
s
. Under ordinary
conditions
e

s
. In fact, it is usually assumed that the scattering event is instanta-
neous, i.e.,
s
0. In this case either classical or quantum theory can be applied for
the description of electron behavior, depending on the length scales. If, however,
e
is
comparable to or smaller than
s
, which may happen under extremely strong scattering
of nonequilibrium electrons, the quantum description of electron behavior is required
regardless of the size of the system.
In classical transport regimes, the characteristic times and their relationships to the
device sizes are of critical importance. They determine temporal and frequency regimes
of device operation. For example, the transit time t
tr
= L
x
,: determines the duration of
signal propagation through a device; here : is the electron velocity. Therefore, t
tr
denes
the ultimate speed limit of the device: the device cannot effectively operate in the time
range less than t
tr
or at frequencies greater than t
1
tr
. This explains one of the trends of
modern electronics: scaling down the device sizes.
The times related to transverse dimensions, t
b
= L
z.y
,: (near the ballistic regimes),
or t
D
= (L
z.y
)
2
,D (for diffusive size effects), determine features of electron transport at
frequencies of the order of t
1
b
, or t
1
D
.
In quantum mechanics, as described in Chapter 3, if external potentials are time-
independent, the electrons are in stationary states. In this case, despite a possible complex
dependence of the wavefunction on position, the temporal evolution of a stationary state
is always determined by an exponential factor exp[i(E,h
--
)t ]. If the alternating external
eld of an angular frequency is applied to the stationary electron system the response
of the electron system may be referred to one of the following three different regimes
depending on the frequency of the external eld.
Ultra-high (quantum) frequencies
If h
--
is comparable to the characteristic stationary electron energy state, E, the nature
of the electrons response will be essentially quantum-mechanical. Only transitions
between states with the energy difference LE = h
--
are allowed. If E is quantized,
the interaction is possible only at resonance frequencies. By varying the device size,
one can vary the energy spectrum and, as a result, change the frequency properties over
a wide range. The kinetic times
e
.
E
. etc. lead to a broadening of these resonances.
If this broadening exceeds the energy separations between quantized levels, the dis-
crete quantum behavior changes to a continuum-like behavior reminiscent of classical
mechanics.
If h
--
_ E the electrons response to an alternating eld is classical (quantization of
transitions can be neglected). In the classical picture, the external alternating eld will
cause periodic electron acceleration and deceleration. The scattering interrupts these
accelerations and decelerations. Depending on how many scattering events occur during
the one period, we distinguish two different regimes of electron behavior.
High (classical) frequencies
If
e
1 the electron motion during one period is not interrupted by scattering. In
accordance with classical mechanics, the electron momentum oscillates with a phase
opposite to that of the eld.
Low frequencies
If
e
_1 the electron undergoes many scattering events during one period of the
external eld. Multiple scattering during the period brings the electron into a quasi-
stationary state, which follows the oscillations of the external eld. In other words, the
electron momentum oscillates in phase with the eld.
In closing this section we conclude that depending on the device dimensions, the
temperature, and other conditions there is a variety of transport regimes. Each of these
regimes demonstrates peculiar properties and requires a physical description suited to
the relevant conditions.
6.3 Statistics of the electrons in solids and nanostructures
For further analysis of nanostructures, we need to review briey the basic properties of
many-electron systems. Indeed, any semiconductor material consists of a vast number
of electrons, i.e., it is essentially a many-electron system. For such a system, the key
question is how particles are distributed over the different energy states which charac-
terize these particles. For example, if the particles move freely, we may be interested
in information about their distribution over the velocities; if the motion of the particles
is quantized, knowledge of their distribution over the energy levels is necessary, etc.
Knowing such distributions, we may nd all average characteristics of the many-particle
systems.
6.3 Statistics 173
The rules and principles according to which particles occupy the energy states in many-
particle systems constitute the so-called physical statistics. In the physical statistics, to
describe the occupation of states by particles, one uses a distribution function of par-
ticles. Under equilibrium, the distribution over the energy levels determines entirely
the properties of many-particle systems. The distribution function has the meaning
of the probability of nding particles with a given energy, E. Let E
l
be the energy
level of particles in a many-particle system, the index l numbers the energy levels.
Then, the distribution function can be thought of as a function of the energy, F(E
l
).
Obviously,
l
F(E
l
) = N. (6.13)
where N is the total number of the particles. It turns out that the statistical principles
in classical and quantum physics are different. The quantum physics brings statistical
features that are absent in the classical description. These features are associated with the
fact that elementary particles, including electrons, are identical and it is impossible, in
principle, to specify their coordinates and trace a given electron. In addition, an internal
characteristic of a particle, the spin, plays a very important role in many-particle physics.
Although a denition of spin is absent in classical physics, we start with a brief review
of classical statistics.
Classical statistics
In classical physics, under equilibriumthe distribution function has an exponential form.
This is the so-called Boltzmann distribution
F
B
(E) = Ce
E,(k
B
T)
. (6.14)
with C and T being the normalization constant and the ambient temperature, respec-
tively. The normalization constant can be found from Eq. (6.13). In general, statis-
tics in classical physics does not restrict the number of particles occupying an energy
level E.
As an example of the application of the Boltzmann distribution, consider the distribu-
tion of free particles with mass m over the velocities. First, we suppose that the system
under consideration is uniform, i.e., external forces are absent. Then, the energy of these
particles coincides with the kinetic energy E = E
kin
= mv
2
,2, where v is the particle
velocity. For three-dimensional particles, we have v = {:
x
. :
y
. :
z
] and each of the veloc-
ity components can vary in the interval [. ]. Thus, by rewriting Eq. (6.14) we
can introduce the distribution function as
F
:
(v) = C
:
e
mv
2
,(2k
B
T)
. (6.15)
Since the velocity is a continuous value, the meaning of the distribution F
:
is the follow-
ing. If we dene an innitesimally small volume d
3
: d:
x
d:
y
d:
z
around a given
velocity v in the velocity space, then the number of particles whose velocity is inside
this volume is F
:
(v)d
3
:. That is, actually, F
:
is the density of the distribution over the
velocity. To nd the normalization constant, we can integrate F
:
over all possible v, and
then we obtain the total number of the particles N. As a result, the distribution function,
F
:
(v), can be written in the form
F
:
(v) = N
_
m
2k
B
T
_
3,2
e
mv
2
,(2k
B
T)
= N
_
m
2k
B
T
_
3,2
e
m(:
2
x
:
2
y
:
2
z
),(2k
B
T)
.
(6.16)
This is the so-called Maxwellian distribution of the particles. By using the Maxwellian
distribution, we can calculate average characteristics of the so-called ideal gas of parti-
cles. For example, the average energy per particle is
E =
1
N
_
d
3
:
mv
2
2
F
:
(v) =
3
2
k
B
T. (6.17)
the average velocity at equilibrium is v = 0, etc.
Fermi statistics for electrons
Now we return to the more general case of quantum statistics. As we stressed at the
beginning of this section, the spin of the particle plays the crucial role for quantum
statistics.
The denition of the spin of a particle was introduced in Section 3.4 as an additional
internal degree of freedom. Though one can compare spin with classical rotation, in
fact, spin is strictly a quantum-mechanical quantity and differs substantially from its
classical analogue. The principal quantitative characteristic of spin is a dimensionless
quantity called the spin number s. It is well established experimentally that an electron
has a spin number equal to
1
2
. If one xes an axis in space, the projection of the electron
spin on this axis can be either
1
2
or
1
2
. A complete description of an electron state
requires a set of quantum numbers: three of them correspond to motion of a particle
in space, say l = {l
1
. l
2
. l
3
], and one more corresponds to a spin, s. According to the
classication discussed in Section 3.4, this case corresponds to a two-fold degeneracy
for each energy level.
For many actual cases the electron spin is not important in altering the energy spectra,
or the spatial dependence of wavefunctions, etc. There is one crucially important con-
sequence of the fact that the electron spin number is a half integer. Indeed, the particles
with half-integer spin numbers obey the Pauli exclusion principle, which we introduced
in Section 3.4. It reads as follows: any quantum state {l. s] can be occupied by a sin-
gle particle only. In other words, two electrons in a system can not be simultaneously
in the same quantum state. Stated in another way, two electrons may be in the same
energy state if their spin quantum numbers are different (the so-called degenerate state);
if one spin quantum number is
1
2
, the other must be
1
2
. The degeneracy is removed
if there is interaction between electron spin and electron translational (orbital) motion,
which is known as spinorbital interaction. In this case, the electron spin affects the
electrons spatial properties, and electrons with spin
1
2
and
1
2
have different energies.
In Fig. 6.3, we illustrate possible populations of the energy levels by the electrons for
two cases, degenerate and nondegenerate energy levels.
6.3 Statistics 175
(a) (b)
Spin splitting Spin degeneracy
Figure 6.3 Occupation of energy levels by the electrons: (a) spin-degenerate levels and
(b) nondegenerate (spin-split) levels.
F
F
(E)
0
E
F E
T
1
= 0
T
4
T
3
T
2
T
1
< T2 < T3 < T4
1
1/2
Figure 6.4 How the Fermi distribution function varies with crystal temperature.
It is clear that the Pauli exclusion principle leads to a new, non-classical statistics of
electrons. Such statistics is called the Fermi statistics. Under equilibrium the occupation
of the energy levels is described by the Fermi distribution function:
F
F
(E
l.s
) =
1
1 e
(E
l.s
E
F
),(k
B
T)
. (6.18)
where T is the temperature of the system, E
l.s
is the energy of the quantum state charac-
terized by the set of quantum numbers l and s, and E
F
is the so-called Fermi energy or
Fermi level. The evolution of the Fermi function F
F
(E
l.s
) with the temperature is shown
in Fig. 6.4, where we use the temperature parameters T
4
> T
3
> T
2
> T
1
with T
1
= 0.
Importantly, the Fermi energy can be related to the total number of electrons through
the normalization condition of Eq. (6.13). Taking into account explicitly the summation
over the spin, we obtain
l.s
F
F
(E
l.s
) = N. (6.19)
which gives E
F
= E
F
(N. T).
FromEq. (6.18), one can see that in accordance with the Pauli principle the occupation
of any energy state, {l. s], dened by Eq. (6.18), is always less than or equal to 1.
At high temperatures the second term in the denominator of Eq. (6.18) is substantially
larger than unity and the Fermi distribution is close to the Boltzmann distribution:
F
F
(E) e
(E
F
E),(k
B
T)
. (6.20)
Equations (6.20) and (6.14) are the same when the normalization constant C is equal to
C = exp
_
E
F
k
B
T
_
. (6.21)
The corresponding curve is shown schematically in Fig. 6.4 at T = T
4
.
In the limit of low temperatures, T 0, the function F
F
transforms into a step
function:
F
F
(E) =
_
1. E
F
> E.
0. E
F
- E.
(6.22)
i.e., F
F
(E) = 1 for the energy levels below the Fermi energy E
F
since all levels with
E - E
F
are occupied and F
F
(E) = 0 for energies above E
F
since these levels are empty.
In this limit, the electron system is frequently referred to as a highly degenerate electron
gas.
Now we can apply Fermi statistics to the electrons in a conduction band. Let n be
the concentration of electrons in the conduction band with energy dispersion E(k). We
accept that the energy spectrum is independent of the spin. Thus, the set of the quantum
numbers, l, is identical to the set of the electron wavevectors, k. According to the Fermi
distribution, the probability of nding an electron with the wavevector k is
F
F
(E(k)) = 2
1
1 e
(E(k)E
F
),(k
B
T)
. (6.23)
where the factor 2 comes from the spin degeneracy. The Fermi energy, E
F
, and the
electron concentration, n, are related through the following equation:
n =
N
V
=
2
V
k
1
1 e
(E(k)E
F
),(k
B
T)
. (6.24)
with V being the volume of the crystal. The summation in the latter formula can be
converted to an integration. Indeed, according to the analysis given in Section 4.4 and
Eq. (4.10), the electron wavevector takes the following values: k
x
L
x
= 2l
1
. k
y
L
y
=
2l
2
, and k
z
L
z
= 2l
3
, with l
1
, l
2
, and l
3
being integers. Here we introduced the crystal
dimensions L
x
, L
y
, and L
z
. (The crystal dimensions are related to the basis vectors of the
lattice a
i
and the number of primitive cells N
i
as L
x
= a
x
N
x
, etc.) Thus, summation over
k is equivalent to summation over l
i
. The latter can be approximately calculated via the
integral:

l
1
.l
2
.l
3
(. . .)
_ _ _
dl
1
dl
2
dl
3
(. . .). Since the distribution function depends
on E(k), it is convenient to express the latter integral in terms of integration over k. We
can use the relationships
Ll
1
=
L
x
2
Lk
x
. Ll
2
=
L
y
2
Lk
y
. Ll
3
=
L
z
2
Lk
z
.
6.3 Statistics 177
Since L
x
L
y
L
z
= V, we can nally write
k
(. . .) =
V
(2)
3
_ _ _
dk
x
dk
y
dk
z
(. . .). (6.25)
This procedure of replacement of the summation over discrete k by integration over
continuous k is useful for the calculation of average quantities.
As an example, let us calculate the Fermi energy E
F
of the electron system at low
temperatures (T 0). The energy spectrumof the electrons is supposed to be isotropic,
i.e., it depends only on the modulus [k[ = k: E(k) = E(k) = h
--2
k
2
,(2m
), where m
is
the effective mass of the electron. Let the electron concentration be n. According to the
Fermi statistics, electrons will occupy all energy states below the Fermi energy. Since
E(k) is an increasing function of k, it follows, for T 0, that all states with k k
F
are
occupied, where k
F
is the so-called Fermi wavevector dened through E(k
F
) = E
F
. When
calculating the concentration according to Eq. (6.24), we have to performthe summation
(integration) over all occupied states, i.e., k k
F
. For these k we have F
F
= 1, and it
follows that
n =
2
V
kk
F
1 =
2
V

V
(2)
3
_ _ _
[k[k
F
d
3
k.
Evaluationof this integral gives the volume of a sphere of radius k
F
, i.e., 4k
3
F
,3. Then, we
obtain a relationship between the Fermi wavevector k
F
and the electron concentration n:
k
F
= (3
2
n)
1,3
. Finally, the Fermi energy of the degenerate electrons in a bulk crystal is
E
F
= (3
2
)
2,3
h
--2
n
2,3
2m
. at T 0. (6.26)
In this case, E
F
increases as the 2,3 power of the electron concentration n. Because the
Fermi function contains the exponential factor, the low-temperature limit corresponds to
the condition E
F
k
B
T. In metals and heavily doped semiconductors, the electron gas
remains degenerate up to room temperature. For example, in the case of a GaAs crystal
withthe effective electronmass M
= 0.067m
0
, where m
0
is the mass of the free electron,
at the concentration n = 10
17
cm
3
, we nd k
F
= 1.43 10
6
cm
1
and E
F
=11.6 meV.
This energy corresponds to a temperature of 135 K. Thus, at T -135 K the electron gas
concentration n = 10
17
cm
3
in GaAs can be considered as degenerate and it is possible
to use the above estimates for the Fermi energy, E
F
, and the Fermi wavevector, k
F
.
The degenerate electron gas is an interesting and very important physical system. This
limiting case facilitates understanding a number of complex phenomena in a simple way.
Indeed, as emphasized previously, in a degenerate electron gas all states below E
F
are
occupied. Let us imagine that a small external perturbation is applied to such a many-
electron system. The perturbation rst of all will cause a redistribution of electrons
between the energy states. However, belowthe Fermi level all states are completely lled
and no redistribution is possible. Instead, only those electrons that are at the Fermi level,
i.e., that have energy just equal to E
F
, can be affected by the perturbation. This results in
the fact that only a small portion of the electrons can participate in the crystals response
to the perturbation. One can say that these active electrons are on the so-called Fermi
surface in k space and the size of the Fermi surface determines the basic properties
+k
F
k
y
k
y
k
x
k
F
k
F
}}
k
x
k
z
k
x
k
F
k
F
(a)
(b)
(c)
0
0
0 k
F
Figure 6.5 The Fermi surfaces for (a) three-, (b) two-, and (c) one-dimensional electron gases.
of the degenerate electron gas. For bulk-like crystals with the simple energy spectrum
considered above, the Fermi surface is just a sphere of radius k
F
. The Fermi surface for
a three-dimensional electron gas is presented in Fig. 6.5(a). Using Eq. (2.4) it is easy to
show that an electron on the Fermi surface has the velocity :
F
= h
--
k
F
,m
.
We can consider low-dimensional electron systems quite similarly. As discussed in
Section 4.5, by using heterostructures it is possible to fabricate articial potential wells,
which conne the electrons fromthe conduction band in such narrowlayers that electron
motion across the layers becomes quantized. This results in electron energies in the form
of low-dimensional subbands given by Eq. (3.49) for the so-called quantum wells:
E
l
3
(k
[[
) =
l
3

h
--2
k
2
[[
2m
. (6.27)
where l
3
(l
3
=1, 2, . . . ) and k
[[
, the two-dimensional wavevector, determine the electron
motion in the plane of the layer. The Fermi distribution function in the formof Eq. (6.18)
can be used to calculate the two-dimensional electron concentration (sheet concentration
of electrons), n
2D
:
n
2D
= n d =
2
S
l
3
.k
[[
1
1 e
(E
l
3
(k
[[
)E
F
),(k
B
T)
. (6.28)
where we have introduced the sheet concentration of electrons, n
2D
(the number of elec-
trons per unit area), the thickness d, and the area S of the conned layer; obviously
V = d S. Actually, exactly as found for bulk crystal, Eq. (6.28) establishes the rela-
tionship between the sheet concentration, n
2D
, and the Fermi energy, E
F
.
6.3 Statistics 179
The sheet concentration of the electrons can be explicitly calculated in the limit of
a degenerate electron gas. Let us suppose that the temperature is low and that only the
lowest subband is populated by the electrons. Then, in the sum over l
3
in Eq. (6.28) we
should keep only one term with l
3
= 1:
n
2D
=
2
S
k
[[
1
1 e
(E
1
(k
[[
)E
F
),(k
B
T)
. (6.29)
Calculation of the right-hand side of this equation can be simplied by replacing the
summation over k
[[
by integration, similarly to that of Eq. (6.25):
k
[[
(. . .) =
S
(2)
2
_ _
dk
x
dk
y
(. . .). (6.30)
where k
x
and k
y
are the components of the two-dimensional vector k
[[
. In the limit of
T 0, the sheet concentration n
2D
can be calculated by dening a two-dimensional
Fermi wavevector k
[[.F
via the relationship
E
F
=
h
--2
k
2
[[.F
2m
. (6.31)
The Fermi surface for two-dimensional carriers is a disk of radius k
[[.F
in k
[[
space,
as shown in Fig. 6.5(b). On performing the integration of Eq. (6.30) over the disk, we
nd
k
[[.F
= (2n
2D
)
1,2
and E
F
=
h
--2
m
n
2D
. at T 0. (6.32)
That is, the Fermi energy, E
F
, increases as the rst power of the electron sheet concen-
tration, n
2D
.
For a nanostructure in which the electron motion is restricted in two directions, i.e.,
for the quantum wire, the energy spectrum is given by Eq. (3.50):
E
l
2
.l
3
(k
x
) =
l
2
.l
3

h
--2
k
2
x
2m
. (6.33)
where it is assumed that the x direction is the only direction of free motion. By applying
a procedure similar to that used above, we may see that for low temperatures the Fermi
surface shrinks to two points in the one-dimensional k space: k
x
= k
F
with k
F
=
1
2
n
1D
. Here, n
1D
is the linear electron concentration, which can be dened by the
formula n
1D
= N,L, with N being the total number of electrons in the wire and L
being the wire length. Subsequent evolution of the Fermi surface of the electrons as
the dimensionality is lowered is illustrated in Figs. 6.5(a)(c). The Fermi energy of
one-dimensional electrons is given by
E
F
=

2
h
--2
8m
n
2
1D
. for T 0. (6.34)
On comparing the Fermi energies obtained for different dimensionalities of the electron
gas, we can conclude that lowering the dimensionality of the gas gives rise to a more
rapid increase of E
F
with increasing concentration for low-dimensional systems.
The concept of Fermi statistics is one of the fundamental ideas of modern solid-state
physics and extremely important for nanoelectronics. The Fermi statistics is applied
widely in the description of nanoelectronic devices.
6.4 The density of states of electrons in nanostructures
To complete the analysis of the occupation of the energy levels by electrons in nanos-
tructures, we shall study another physical quantity characterizing the occupation, known
as the density of states. The reason for introducing this quantity is the following. From
the results presented in the previous chapters, we see that the electron energy spectra
in nanostructures with conduction electrons are complex and consist of a series of sub-
bands. The distances between subbands are determined by the prole of the conning
potential, while inside each subband the spectrum is continuous and these continuous
spectra overlap. To characterize these complex spectra, it is convenient to introduce a
special function known as the density of states, (E), which gives the number of quantum
states d N(E) in a small interval dE around energy E:
d N = (E)dE. (6.35)
If the set of quantum numbers corresponding to a certain quantum state is designated as
, the general expression for the density of states is dened by
(E) =
(E E
). (6.36)
where E
is the energy associated with the quantum state . At this point, it is useful to
introduce Diracs famous -function:
(x) =
_
0. for x ,= 0.
. for x 0.
(6.37)
though
_

(x)dx = 1. (6.38)
Diracs -function is used when performing an integration. The main rule of such an
integration is the following:
_
b
a
dx (x x
0
)+(x) = +(x
0
). if a - x
0
- b. (6.39)
where +(x) is an arbitrary well-behaved function.
As a simple example of a calculation of the density of states by using the denition
of Eq. (6.36), let us calculate this quantity for electrons in a bulk crystal. As discussed
previously, for such a case the set of quantum numbers is {k. s]. Assuming the energy
to be independent of the spin, E(k) = h
--2
k
2
,(2m
). Replacing the summation by an

integration as in Eq. (6.25), we obtain
3D
(E) =
2 V
(2)
3
_ _ _
dk(E E(k)). (6.40)
6.4 The density of states 181
(E) (E)
(E) (E)
(a) (b)
(c) (d)
0
0
0 0
E E
E
E
e
1
e
2
e
3
e
11
e
111
e
a
e
b
e
a
e
b
e
c
Figure 6.6 The density of states of electrons, (E), in systems of different dimensionalities: (a) a
bulk crystal, (b) a quantum well, (c) a quantum wire, and (d) a quantum dot. Here
1
,
11
, and
111
are the ground states in a quantum well, quantum wire, and quantum dot, respectively;
2
and
3
are higher states in a quantum well and
a
,
b
, and
c
are higher states in a quantum wire
and in a quantum dot.
Because E(k) actually depends on the modulus k, spherical coordinates may be used.
Integration over two angles gives 4 and the triple integral is reduced to the single
integral
3D
(E) =
2V
(2)
3
4
_

dk k
2
(E E(k)). (6.41)
Now, we replace the integration over k by an integration over E(k) = E. Then, taking
into account that k
2
= (2m
,h
--2
)E and dk =
_
m
,(2h
--2
E) dE,
3D
(E) =
V
2
2
_
2m
h
--2
_
3,2
_

0
dE
E(E E). (6.42)

Finally, by using the rule of Eq. (6.39), we obtain
3D
(E) =
_
m
h
--2
_
3,2
V
2E. (6.43)
The obtained density of states for three-dimensional electrons is depicted in Fig. 6.6(a).
For electrons in a quantum well with the energy spectrum given by Eq. (6.27), the
set of quantum numbers includes a spin quantum number, s, a quantum number, l
3
,
characterizing the transverse quantization of the electron states, and a continuous two-
dimensional vector, k
[[
. Hence, {s. l
3
. k
[[
]. There is a two-fold spin degeneracy of
each state, (s =
1
2
), so that
2D
(E) = 2
l
3
.k
x
.k
y
_
E
l
3

h
--2
_
k
2
x
k
2
y
_
2m
_
. (6.44)
In order to calculate the sum over k
x
and k
y
, we can apply the replacement, Eq. (6.30),
with S being the area of the surface of the quantum well, S = L
x
L
y
, where L
x
and
L
y
are the sizes of the quantum well in the x and y directions, respectively. Calculations
of the integrals give us
2D
(E) =
m
S
h
--2
l
3
_

0
dE (E
l
3
E) =
m
S
h
--2
l
3
O(E
l
3
).
(6.45)
where O(x) is the Heaviside step-function:
O(x) =
_
1. for x > 0.
0. for x - 0.
(6.46)
Very often the density of states per unit area,
2D
,S, is used to eliminate the size of
a sample. Each term in the sum of Eq. (6.45) corresponds to the contribution from one
subband. The contributions of all subbands are equal and independent of energy. As
a result, the density of states of two-dimensional electrons exhibits a staircase-shaped
energy dependence with each step being associated with one of the energy states,
l
3
.
The height of each step is universal and depends only on the effective electron mass.
Figure 6.6(b) depicts the two-dimensional density of states. By comparing the densities
of states for the electrons in bulk crystals per unit volume and those of quantum wells
per unit surface we may see that the differences between the two- and three-dimensional
cases are most pronounced in the energy regions of the lowest subbands. For large l
3
the staircase function lies very close to the bulk curve
3D
(E) and coincides with it
asymptotically.
Similarly, we can nd the density of states of a one-dimensional electron gas with the
energy spectrum given by Eq. (6.33). The result of calculations is
1D
(E) =
l
2
.l
3
l
2
.l
3
(E).
where
l
2
.l
3
(E) =
L
_
2m
h
--2
1
_
E
l
2
.l
3
O(E
l
2
.l
3
). (6.47)
Here, L is the length of the wire. Schematically,
1D
(E) for one-dimensional electrons is
shown in Fig. 6.6(c). The characteristic feature of the one-dimensional density of states
is its divergence near the bottom of each of the one-dimensional subbands. The density
of states then decreases as the kinetic energy increases. This behavior is very remarkable
because it leads to a whole class of newelectrical and optical effects peculiar to quantum
wires.
Now, we can consider the ultimate case of the density of states for zero-dimensional
electrons, i.e., for electrons in quantum dots. According to the denition of Eq. (6.36),
in the case of quantum dots or boxes the spectra are discrete. Thus, the density of states
is simply a set of -shaped peaks, as depicted in Fig. 6.6(d). For an idealized system, the
peaks are very narrowand innitely high, as illustrated in Fig. 6.6(d). In fact, interactions
between electrons and impurities as well as collisions with lattice vibrations bring about
a broadening of the discrete levels and, as a result, the peaks for physically realizable
systems have nite amplitudes and widths. Nevertheless, the major trend of sharpening
of the spectral density dependences as a result of lowering the dimensionality of the
system is a dominant effect for nearly perfect structures at low temperatures.
The dramatic changes in the electron density of states caused by dimensionally con-
ning crystals manifest themselves in a variety of major modications in conductivity,
optical properties, etc. Indeed, as we will see, these modications in the density of states
also lead to new physical phenomena.
6.5 Electron transport in nanostructures
As indicated in Section 6.2, there are distinct regimes of electron motion in nanostruc-
tures. In this section, we will study several instructive examples of electron transport
induced by an electric bias. We start with the simplest low-eld dissipative classi-
cal electron transport, which is realized at large intercontact distances, according to
Eqs. (6.10) and (6.12). This transport regime can occur in bulk samples, quantum wells,
and quantum-wire structures.
Classical dissipative transport
In a solid the electrons are in constant motion. However, this motion is chaotic as a result
of random scattering by imperfections, lattice vibrations, interface roughness, etc. As a
result there is no preferred direction for electron motion. Thus, the net electron ux and
electric current are equal to zero. If an electric eld F is applied to the solid, an electric
force, eF, acts on each of the electrons (here we suppose that the electron charge is
equal to e). Though the chaotic character of electron motion can remain, a directional
net drift of electrons is induced by the electric force. Figure 6.7 demonstrates chaotic
electron motion without and with the electric eld applied. To describe this motion, we
can use the Newton equation (2.8) for the average velocity of the electrons, v. To take
into account the scattering processes, which lead to the loss in the directed component
of the velocity, we introduce an additional term that contains a friction force:
m
dv
dt
=
m
e
v eF. (6.48)
where
e
can be interpreted as the momentum relaxation, or free-ight time, which was
discussed in Section 6.2. Obviously, the electron motion described by the above equation
is dissipative in nature.
For the stationary state, dv,dt = 0, we obtain
v =
e
e
m
F = jF. (6.49)
where we have introduced the parameter j, called the electron mobility. The mobility
is one of the basic characteristics of electron transport in the case of low electric elds.
The denition of j,
j =
e
e
m
. (6.50)
Figure 6.7 Chaotic diffusion motion of an electron at equilibrium (a) and when an electric eld F
is applied (b).
shows that the mobility is larger for materials with small effective mass m
and with
suppressed scattering, i.e., for large
e
. The negative sign in Eq. (6.49) reects the fact
that the electrons move in the direction opposite to the electric eld, because of their
negative charge.
Having the electron average (drift) velocity, v, and the electron concentration, n, we
can calculate the electric current density as
J = evn = ejnF = F. (6.51)
Here,
= ejn (6.52)
is the so-called specic conductivity, and, as we can see, it depends on both the electron
concentration, n, and the electron mobility, j. The result given by Eq. (6.51) is known
as Ohms law. If the specic conductivity and the geometrical dimensions of a sample
are known, we can readily calculate the total electric current:
I = J S = SF. [I[ =
+
0
R
. (6.53)
where we introduce the cross-section of the sample S, the voltage drop across the sample
+
0
= L
x
F, and the sample length L
x
. The resistance R is
R =
L
x
S
. (6.54)
Equations (6.49)(6.51) are valid for systems of any dimensionality, where dissipa-
tive transport occurs. These include three-dimensional, two-dimensional, and one-
dimensional systems. For all of them, the electron mobility is a representative char-
acteristic of the classical transport regime.
To dene measurement units of the mobility, we must rst discuss briey the units of
measurement of several entities introduced in the preceding discussion. (1) The charge
is measured in units of coulombs, C. The elementary charge of the electron, e, is equal
to 1.6 10
19
C. (2) The electron concentration is measured in m
3
, m
2
, and m
1
for
three-dimensional, two-dimensional, and one-dimensional systems, respectively. (3) The
electric eld is measured in units of volts per meter, Vm
1
. (4) Accordingly, the units
of mobility are m
2
V
1
s
1
. (5) The current is measured in amperes, A, i.e., coulombs
per second, Cs
1
. (6) For the current density and conductivity of Eq. (6.51), we obtain
different units for electron systems with different dimensionalities. For example, in a
bulk crystal the current density is measured in amperes per unit cross-section, Am
2
,
and the units of conductivity are 1 O
1
m
1
with Obeing the ohm the unit of electrical
resistance.
Returning to the discussion of mobility, we shall stress that the magnitude of the
mobility depends on the particular material and scattering mechanisms in this material.
Typically, the scattering mechanisms in semiconductors include scattering by impuri-
ties, imperfections, and lattice vibrations. If the rst two scattering mechanisms can
be avoided in pure and high-quality samples, lattice or, as one frequently says, phonon
scattering can not be avoided in principle. The total scattering rate is a sum of rates of
scattering due to particular mechanisms. Each of the rates is proportional to a scattering
probability and inversely proportional to the mean free time. This leads to the conclu-
sion that also the mobilities due to two or more scattering mechanisms should be added
inversely:
1
j
=
1
j
ph
1
j
im
. (6.55)
where j
ph
. j
im
. . . . are partial mobilities determined by phonon scattering, impurity
scattering, etc. Phonon scattering depends in an obvious way on temperature: at low
temperatures, when thermal lattice vibrations are suppressed, the rate of this type of
scattering is small, but it increases with the temperature. This results in a nite value of
mobility even in pure crystals. For example, at room temperature the electron mobilities
of Si and GaAs bulk crystals are limited to 1350 cm
2
V
1
s
1
and 8500 cm
2
V
1
s
1
,
respectively. The same order of magnitude for jis realized in low-dimensional systems at
roomtemperature. When the temperature decreases, mobility increases and the scattering
by imperfections and impurities becomes the limiting factor, as shown in Fig. 6.8. These
scattering mechanisms can be avoided in low-dimensional systems, where mobility can
reach values above 10
5
10
6
cm
2
V
1
s
1
.
The electron current in the form of Eq. (6.51) is valid for uniform conductors. If the
electron concentration, n, is spatially dependent, the electrons, naturally, diffuse from a
region of high concentration to one of a low concentration. This produces an electron
ux opposite to the gradient of the electron concentration: dn,dr. The diffusive
contribution to the current can be written as
J
D
= eD
dn
dr
eD
r
n. (6.56)
Figure 6.8 The temperature dependence of the electron mobility in Si for a system with two
mechanisms of scattering, lattice scattering and impurity scattering. The concentration of
impurities, N
D
, is indicated near each curve. After W. F. Beadle, J. C. C. Tsai et al. (Eds.), Quick
Reference Manual for Semiconductor Engineers (New York, Wiley, 1985).
where D is the diffusion coefcient, that was discussed in Section 6.2. Thus, in a nonuni-
form conductor, the total current is composed of both drift and diffusive contributions:
J = ejFn eD
r
n. (6.57)
It is easy to establish the fundamental relationship between two kinetic coefcients,
the mobility, j, and the diffusivity, D. Indeed, let us apply Eq. (6.57) to equilibrium
conditions. Under equilibrium, the electric current is absent, i.e., J = 0. The electron
concentration, n(r), can be expressed via the electrostatic potential, +(r), according
to Boltzmanns distribution in the form of Eq. (6.14): n(r) = n
0
exp[e+(r),(k
B
T)]. On
putting this dependence and the electric eld, F(r) = d+,dr, into Eq. (6.57) and equat-
ing the current to zero, we obtain the so-called Einstein relationship:
D
j
=
k
B
T
e
. (6.58)
Thus, having the electron mobility j, one can easily calculate the diffusion coefcient,
D.
The results just discussed are relevant to steady-state electron transport. However,
the Newton equation (6.48) can describe the behavior of electrons in an arbitrary time-
dependent electric eld, F(t ). Since any dependence, F(t ), can be represented by using
a Fourier transform, without loss of generality we can analyze the case of a harmonic
dependence of the eld:
F(t ) = F
cos(t ). (6.59)
where F
is the eld magnitude and is the frequency of eld oscillations. For further
analysis, we use the well-known formula cos(t ) = Re[e
it
]. Here Re[. . .] means
calculation of the real part of an expression in the brackets. Thus, the physical eld is
F(t ) = Re[F
e
it
].
It is worth noting that calculations with exponential functions like e
it
are always
simpler than those with sine and cosine functions. For this reason, the following approach
is commonly used. Instead of the eld in the form of Eq. (6.59) in the Newton equation,
one uses the complex representation of the eld:
F(t ) = F
e
it
.
When a solution to Eq. (6.48) with the complex eld is found, the real part of this solution,
that has a physical meaning, can be calculated easily.
To apply this approach, we look for the solution of Eq. (6.48) with a complex eld of
the exponential form v
e
it
. On substituting this form into Eq. (6.48), we immediately
obtain
v
=
e
m
e
1 i
e
F
. (6.60)
In this case, the electron velocity should be calculated as v(t ) = Re[v
e
it
]; i.e.,
v(t ) =
e
e
m
_
cos (t )
1
2
2
e

e
sin (t )
1
2
2
e
_
. (6.61)
Thus, the cosine electric eld of Eq. (6.59) induces electron motion, with both cosine
and sine contributions. It is instructive to rewrite Eq. (6.61) as
v(t ) =
e
e
m
cos(t )
_
1
2
2
e
. (6.62)
where we introduce the phase shift of the electron velocity with respect to the phase
of the eld (6.59). This phase shift can be found from the equation tan =
e
. Here,
> 0, which implies a delay of the electrons with respect to the eld changes. Obviously,
the delay exists only for non-zero frequencies and arises because of the friction force
in the Newton equation. The second conclusion, which follows from Eq. (6.62), is that
the magnitude of velocity oscillations decreases when the eld frequency increases. In
the limit of very high frequencies
e
1, the alternating electron velocity vanishes.
Having the electron velocity, v(t ), and concentration, n, we can calculate the time-
dependent electric current similarly to Eqs. (6.51) and (6.52). However, more common
is the use of the complex representation of the current density:
J(t ) = ()F
e
it
. (6.63)
where we introduce the complex conductivity:
() =
e
2
e
n
m
_
1
1
2
2
e
i

e
1
2
2
e
_
. (6.64)
This conductivity reduces to its steady-state value in the limit
e
0, and, in this
limit, Im[] 0. At nite frequencies, both contributions, Re[] and Im[], are
0 1 2 3 4 5 6 7 8 9 10
0.0
0.2
0.4
0.6
0.8
1.0
1
2
wt
e
s(wt
e
)
s(0)
Figure 6.9 Dependences of the real (curve 1) and imaginary (curve 2) parts of the complex
conductivity of Eq. (6.64) on frequency ; (0) = ( = 0).
important for an alternating current; in particular, at = 1,
e
the two contributions
are equal numerically. The latter property is used typically to determine
e
by varying
the frequency, . Importantly, the electron scattering time,
e
, determines the high-
frequency properties of a material. In particular, at
e
1 the conductivity vanishes
and the electron subsystem of the material does not react to a high-frequency eld.
The relationship given by Eq. (6.64) is known as the Drude formula. In Fig. 6.9 the
dependences of Re[()] and Im[()] versus are shown.
The conductivity () is a specic characteristic of a material. If the material is
homogeneous and its geometrical parameters are known, we can use to relate the
total alternating current

I and the alternating voltage

+ = +
e
it
, both in complex
representation:
I(t ) = I
e
it
. I
=
+
Z()
. Z() =
L
x
()S
. (6.65)
Here, in the second relation, we omit the vector designation; L
x
and S are the intercontact
distance and the cross-section of the sample, respectively. By comparing these results
with those of the steady-state case given by Eqs. (6.53) and (6.54), one may see that,
instead of the resistance, R, a frequency-dependent parameter, Z(), is introduced. It
is called the impedance. The impedance is a complex function that characterizes the
electrical properties of the whole sample. The impedance can be introduced for any
nonhomogeneous sample and device.
Dissipative transport in short structures
The mobility is a specic characteristic of a uniform conducting material. In a sample
of extended dimension L
x
, the electric eld is almost uniform. It can be estimated as
F = +
0
,L
x
, with +
0
being the applied voltage, and the mobility determines basically

n
+
i n
+
(b)
x
V(x)
E
F
(c)
V(x)
(a)
virtual cathode
x
Figure 6.10 (a) The structure schematic of an n
in
diode. Potential energy proles, V(x),

in a diode with charge-limited dissipative transport: (b) for an unbiased diode and (c) for a
biased diode.
the electric resistance and the current, according to Eq. (6.53). In short samples, another
electrical effect can contribute essentially toward the electron transport and the current.
The effect arises due to nonuniform redistribution of the electrons and, thus, of the
electric charge across the short sample. The charge induced by the current affects the
potential distribution along the sample, making it strongly nonuniform and dependent
on the current. As a result, the dependence of the total current on the applied voltage
is no longer linear. The electron transport in this regime is called space-charge-limited
transport. Since in this text we are interested particularly in very short structures, it is
instructive to review briey the space-charge-limited transport under the condition of
Eq. (6.2), when we can introduce mobility.
Let us consider a short sample with two contacts as depicted in Fig. 6.2(b). The two-
contact device is called the diode and the body of the sample between the contacts is
called the base of the diode. Assume for simplicity that the base is not doped, i.e., there
are no conducting electrons in the base. The contacts can be fabricated by heavy n-type
doping of the contact regions; they are called n
regions. In such a case, this device is an

n
i (insulator)n
diode as shown in Fig. 6.10(a). Let + designate the electrostatic

potential. Then, the potential energy of the electrons is V = e+. In Fig. 6.10(b), a
sketch of the potential energy of an unbiased diode is presented: the electrons in the
contact regions on the left and right are separated by a high potential barrier in the base.
If an electric voltage +
0
is applied, the potential relief changes, as shown in Fig. 6.10(c).
The potential energy, V(x) = e+(x), decreases and some of the electrons can overcome
the potential barrier, penetrate from the emitter electrode (the cathode) into the diode
base, and contribute to the current. This bias-induced effect is referred to as electron
injection from the electrode to the base. Obviously, with increasing bias, the maximum
of the potential shifts toward the cathode and becomes lower, and the injection current
increases. This physical picture can be described by the following simple model, which
is valid at large electric biases. The current density, J, can be written via the mobility
characteristic for the undoped base, j, the injected electron concentration, n(x), and the
electrostatic eld, F(x) = d+,dx, as in Eq. (6.51):
J = ejn(x)F(x). (6.66)
Importantly, for the chosen conguration, see Fig. 6.10(a), the cathode is on the left and
the anode is on the right. The potential, +, increases with x, and the electric eld, F,
and current density, J, are both negative. As a result of the continuity condition, in the
stationary state the current density through the diode is conserved: J = J
0
=constant,
where J
0
is the absolute value of the current density. From the latter relationship, we can
determine the concentration of the injected electrons, n(x):
n(x) = J
0
,(ejF(x)). (6.67)
For the electrostatic eld, we can use Poissons equation,
d
2
+
dx
2
=
dF
dx
=
en(x)
c
0
c
. (6.68)
where c is the dielectric constant of the base material and c
0
is the permittivity of free
space. After substitution of n(x) from Eq. (6.67), Eq. (6.68) can be rewritten in terms of
the eld, F:
F
dF
dx
=
J
0
c
0
cj
or F dF =
J
0
c
0
cj
dx. (6.69)
This equation should be supplemented by a boundary condition for the eld. We can use
the fact that at the maximum of the potential barrier:
d+
dx
= F = 0. (6.70)
At a large electric bias, this maximum is shifted close to the cathode and we can set
F(x = 0) 0. This approach neglects the processes which occur in a very narrow
region between the real cathode and the potential maximum. It is called the virtual-
cathode approximation and it is used widely for such a simplied analysis. Integration
of Eq. (6.69) leads to the solution
1
2
F
2
(x) =
J
0
c
0
cj
x or F(x) =
_
2J
0
c
0
cj
_
1,2
x
1,2
. (6.71)
where the negative value of the square root is chosen for the solution in accordance
with the above-discussed sign of the electric eld. Thus, the distribution of the potential,
+(x), is given by
+(x) =
_
x
0
dx F(x) =
_
8J
0
9c
0
cj
_
1,2
x
3,2
. (6.72)
From the total voltage drop +(L
x
) = +
0
, we can nd the relationship between the
current density, J
0
, and the bias, +
0
, i.e., the currentvoltage characteristic:
J
0
J
0.d
=
9c
0
cj
8
+
2
0
L
3
x
. (6.73)
This is the so-called MottGurney law for the diode with dissipative electron trans-
port. We have found that the currentvoltage characteristic becomes strongly nonlinear
because of the space-charge effect. The space-charge effect can be characterized, partic-
ularly, by the average concentration of the injected electrons, n :
n =
1
L
x
_
L
x
0
n(x)dx =
3
2
c
0
c+
0
eL
2
x
. (6.74)
which increases with the applied voltage.
Equation (6.49) and F(x) facilitate the calculation of the average time of electron
transit through the diode, t
tr.d
:
t
tr.d
=
_
L
x
0
dx
1
:(x)
=
_
L
x
0
dx
1
jF(x)
=
4
3
t
0.d
. (6.75)
where
t
0.d
=
L
2
x
j+
0
. (6.76)
Here, t
0.d
is the transit time of the electrons drifting in the average electric eld, +
0
,L
x
,
i.e., when the space-charge effects are neglected. As follows fromEq. (6.75), these effects
increase the transit time by the factor 4,3.
In general, non-stationary electrical properties of a device are characterized by the
current induced in response to a time-dependent external voltage bias. If the bias is of a
frequency , the current response is given by the impedance according to Eq. (6.65). For
a MottGurney diode biased with a steady-state voltage, +
0
, the complex impedance
can be calculated exactly:
Z() =
6R
d
O
3
_
(Osin O) i
_
O
2
2
1 cos O
__
. (6.77)
where
R
d
=
d+
0
dJ
0
=
4L
3
x
9c
0
cjS+
0
is the differential resistance of the diode at steady state calculated from Eq. (6.73),
I
0
= J
0
S. S = L
y
L
z
is the cross-section of the sample, and O = t
tr.d
; i.e., it is
determined via the average time, t
tr.d
, of electron transit through the diode; see Eq. (6.75).
The dependences of Re[Z(O)] and Im[Z(O)] are presented in Fig. 6.11. The impedance
of Eq. (6.77) differs considerably from the impedance of a material with the Drude-
like high-frequency behavior. Importantly, as the voltage bias increases, the impedance
decreases and the current response increases +
0
; moreover, the spectral width of the
response becomes larger (1,t
tr.d
+
0
). Both conclusions provide evidence that a short
0 1 2 3 4 5 6 7 8 9 10
0.0
0.2
0.4
0.6
0.8
1.0
wt
tr,d
Z(wt
tr,d
)
Z(0)
1
2
Figure 6.11 Dependences of the real (curve 1) and imaginary (curve 2) parts of the impedance of
Eq. (6.77) on frequency ; Z(0) = Z( = 0).
diode with space-charge-limited dissipative transport under a high bias can operate at
higher frequencies than can a doped diode with Drude conductivity.
To conclude this analysis of dissipative electron transport in short samples, we want
to point out that the main results obtained are valid at large biases, when the diffusive
contribution to the current can be neglected:
j[F(x)[ D
dn(x)
dx
. (6.78)
(A related analysis is proposed in Problem 7.) For such conditions, the transport is
determined to a large extent by the space-charge effects, the electric eld is highly
nonuniform, and the injection current increases quadratically with the bias, as Eq. (6.73)
shows. Under high bias, the transit time through the diode decreases and the device
remains electrically active in a frequency range enlarged proportionally with the bias.
Hot electrons
Now, we return to large uniform crystals to study the effect of high electric elds on
the drift velocity and the currentvoltage characteristics. Equations (6.49)(6.51) were
obtained under the assumption that the relaxation time,
e
, is constant; i.e., it does not
depend on the magnitude of the electric eld. When the eld increases, the electron gas
is far from equilibrium. In particular, the average electron energy increases. This may
be understood from the following qualitative considerations. The electron energy in an
electric eld may be described through the relationship
dE
dt
= e(vF)
E E
eq
E
. (6.79)
The rst term on the right-hand side corresponds to the power gained by the electron
from the electric eld; the second term represents the rate of electron-energy losses.
The losses are proportional to the deviation of electron energy, E, from its equilibrium
value, E
eq
, and
E
is the energy relaxation time introduced in Section 6.2. The electron
energy increases in the electric eld until the overall energy balance is maintained. In
the stationary case, when dE,dt = 0, Eq. (6.79) yields an electron energy, E, given by
E = E
eq
e(vF)
E
. (6.80)
According to Eq. (6.49), the velocity, v, is linearly proportional to the electric eld, F.
Thus, the mean electron energy increases as the square of the electric eld and
exceeds the equilibriumvalue, E
eq
. It is convenient and conventional in semiconductor
electronics to consider the electron effective temperature, T
e
, instead of the mean
electron energy. The relationship between the temperature and the mean energy can be
found at thermal equilibrium and is E = k
B
T
e
,2, where the factor is the dimension-
ality of the structure. Obviously, under thermal equilibrium the electron temperature, T
e
,
coincides with the lattice temperature, T. Under nonequilibrium conditions the two tem-
peratures, T, and, T
e
, may differ. The effective electron temperature expressed through
the mean electron energy serves as a gauge of the nonequilibrium state. If T
e
exceeds T
only slightly, and the electron transport still obeys Ohms law, we have warm electrons.
The case with T
e
T corresponds to the situation in which the electrons are far from
equilibrium. For such a situation, the electrons frequently are called hot electrons. The
electron temperature can reach magnitudes of about several thousand degrees Kelvin,
while the lattice can remain cold. For simple estimates it is accepted that transition from
the warm-electron regime to the hot-electron regime occurs at the electric eld F = F
he
,
when e(vF)
E
is equal to E
eq
= k
B
T. It is easy to estimate the heating electric eld,
F
he
. Indeed, in Eq. (6.80) the drift velocity v can be estimated from v = jF; thus, the
critical heating eld is
F
he
=
_
k
B
T
ej
E
. (6.81)
For hot electrons, the scattering processes themselves become dependent on the eld.
The linear relationship of Ohms law of Eq. (6.51) is no longer valid and the current
voltage characteristics, J = J(F), and the drift-velocityeld dependence, : = :(F),
can exhibit strongly nonlinear behavior, which depends on the electron bandstructure
and specic scattering mechanisms. In particular, these dependences are of different
shapes in the two most important materials. In group IV semiconductor materials, the
J(F) and :(F) dependences lead to the saturation effect at large elds, as shown in
Fig. 6.12(a), whereas in IIIV compounds, after a nonlinear portion with a rise, the
current and the velocity undergo a decrease in some interval of the elds, as shown
in Fig. 6.12(b). As follows from the analysis of different transport regimes given in
Section 6.2, for high-speed devices the important material parameter is the maximum
of the drift velocity, which can be achieved in high electric elds. A comparison of the
saturated and maximum velocities for relevent materials is presented in Table 6.2.
One can see that some of the IIIV compounds (including GaAs, InP, and InSb) have
velocities several times larger than those of Si, GaP, and AlAs. For Si the drift velocity is
restricted to a magnitude of about 10
7
cms
1
that can be reached for electric elds above
Table 6.2 Values of the saturated (for
Si, SiC, SiO
2
, AlAs, and GaP) and
maximum (for GaAs, InP, InAs, and
InSb) drift velocities for semiconductor
materials at room temperature
Velocity
Material (10
7
cms
1
)
Si 1
SiC 2
SiO
2
1.9
AlAs 0.65
GaP 1.1
GaAs 2
InP 2.5
InAs 4.4
InSb 6.5
0
0
(a)

Si
Low-field extrapolation
F
1/2
lines
F
0
lines
10
10
2
10
3
10 10
2
10
3
10
4
10
5
10
6
10
7
(b)
1
10
2
3
20 5 15 25 30 35
GaAs
InP
T = 300 K
77 K
193 K
298 K
Figure 6.12 Currentvoltage characteristics at large electric elds: (a) Si and (b) GaAs and InP.
Reprinted with permission, (a), from E. J. Ryder, Mobility of holes and electrons in high
electric elds, Phys. Rev., 90, 766 (1953).
C _
1953 by the American Physical Society.
several kVcm
1
. For GaAs, the velocity maximum of 2 10
7
cms
1
can be achieved
at elds 3.5 kVcm
1
.
Transient overshoot effects
Limitations in the drift-velocity characteristic for a particular material can be overcome
by utilizing another hot-electron effect known as velocity overshoot. In order to explain
this phenomenon, let us mention that our previous conclusions have been made for
the steady-state case, in which the electron distribution is stationary. In other words,
the previous analysis has referred to the electron properties averaged over times much
greater than the characteristic times of the system such as the mean-free-ight time,
0 1 2 3 4
0
1
2
3
4
5
6

v

(
1
0
7
c
m
s
1
)
v

(
1
0
7
c
m
s
1
)
Time (10
13
s) Time (10
12
s)
(a) (b)
20 kVcm
1
5 kVcm
1
10 kVcm
1
5 kVcm
1
1 kVcm
1
GaAs Si
0 1 2 3 4 5 6 7 8 9 10
0.0
0.5
1.0
1.5
10 kVcm
1
Figure 6.13 The overshoot effect transient response of the electron drift velocity to step-like
pulses of the electric eld at room temperature. The eld magnitudes are indicated for each plot:
(a) Si and (b) GaAs. Reprinted with permission, from V. Mitin, V. Kochelap, and M. Stroscio,
Quantum Heterostructures, Fig. 7.13 (Cambridge, Cambridge University Press, 1999).
momentum and energy relaxation times, etc. Now, we will consider processes that occur
in the electron system immediately following its deviation from the equilibrium state. In
this particular case, we will focus on the response to the electric eld in the form of a
step-function.
In general, the momentum relaxation time,
e
, entering Eq. (6.49) is shorter than
the energy relaxation time,
E
, which determines the electron energy; see Eq. (6.79).
Therefore, the velocity response to the electric eld step is faster than the energy
response for the case described by Eq. (6.79). Typically, if the scattering rate increases
with increasing electron energy, the electron velocity may exceed the stationary veloc-
ity during a time interval of the order of
E
. In other words, the transient velocity
is not just a function of the electric eld but also a function of the electron energy.
Indeed, the velocity adjusts itself quickly to the quite slowly changing energy, and
it follows that energy until the energy reaches the steady state. Initially, when the
electron energy has not reached the stationary value, the electron velocity correspond-
ing to the transient energy is higher than the velocity corresponding to the stationary
energy.
The overshoot effect is shown in Fig. 6.13 for Si and GaAs. The drift velocity is
presented as a function of time for several electric elds. The electric eld is assumed to
be switched on at time t = 0. The overshoot effect is pronounced in high electric elds.
The maximum of the transient velocity can exceed the stationary saturated velocity by
as much as two to four times.
From a physical explanation of the overshoot effect, one can understand how it is
possible to utilize this effect. Let us imagine that cold electrons enter an active region
of a semiconductor device through a contact. If there is a high electric eld in the active
region, the electrons will be accelerated. At some distance from the injecting contact the
x (m)
0.5
1.0

0.0
0.1
0.2
0.3
0.4
0.5

0.0 0.5 1.0
2
4
6

0.0 0.5 1.0
0
5
10
15
20

F
n
N
D
F

(
k
V
c
m
1
)

n
,

N
D

(
1
0
1
6

c
m
3
)
(a)
v

(
1
0
7

c
m
s
1
)

E

(
e
V
)

v
E
(b)
x (m)
0 0.0
Figure 6.14 Space-charge-limited transport with the velocity overshoot effect in an n
in
GaAs diode at L
x
= 0.37 m, T = 77 K, and +
0
= 0.5 V. (a) The doped regions, the electric
elds, F, and the injected electron concentration, n. (b) The drift velocity, :, and the average
energy of the injected electrons, E. From A. Ghis, E. Constant, and B. Boittiaux, Ballistic and
overshoot electron transport in bulk semiconductors and in submicronic devices, J. Appl. Phys.,
54, 214221 (1982). Reprinted wth permission from A Ghis, E Constant, and B. Boittiaux,
Journal of Applied Physics, 54, 214 (1393).
C _
electrons will reach the maximum overshoot velocity and after that their velocity will
gradually decrease to the stationary value. If the active region of the device is short and
comparable to the distance over which the overshoot effect takes place, electron transit
through this active region will occur at a velocity higher than the stationary velocity and
the overall transit time will be shorter. Consequently, the device will be able to operate
at higher speed and frequency.
As follows from Fig. 6.13, the characteristic time of the overshoot effect is about
tr
= 0.5 10
12
s for GaAs. Estimating the average velocity under the overshoot as
:
m
= (24) 10
7
cms
1
, we obtain an estimate for the device length necessary to realize
the discussed effect: L
x
:
m
tr
= 0.10.2 m. That is, the overshoot effect and ultra-
high-speed electrons are achievable in short (submicron) samples.
At high biases and currents, the space-charge effects should be taken into account
together with the velocity overshoot. In Fig. 6.14, we present the calculations of space-
charge-limited electron transport in a short n
in
GaAs diode, where n
are heavily
doped contacts and i is the undoped base of the diode. The step-like doping of the con-
tact regions is marked by N
D
. The eld, the concentration of the injected electrons and
their average drift velocity, and the average energy are shown. The electric eld and the
injected electron concentration are extremely nonuniform, which is similar to the pre-
viously analyzed case of dissipative electron transport. However, for a short diode, the
velocity overshoot in the diode base is clearly observed. The average energy increases
in the base, and drops only in the receiving contact because of strong scattering in
the heavily doped region. Interestingly, the drift velocity reaches a value of above 4
10
7
cms
1
and the electron energy exceeds 3500 K at a crystal temperature
of 77 K.
For the low-eld, hot-electron, and overshoot transport regimes studied, the processes
of dissipation play a dominant role in controlling the electron drift velocity. By scaling
down the sample dimensions, one can achieve an ultimately fast collisionless electron
transport.
Classical ballistic transport
According to Eq. (6.11), the classical ballistic (collisionless) regime occurs in very short
samples, where during their ight the electrons do not undergo any scattering. The nite
electric current and electric resistance in such a case occur exclusively due to the space-
charge effects. We can account for these effects in a ballistic diode with the geometry
depicted in Fig. 6.2(b). Similarly to the case of the above-presented short diodes with
dissipative and overshoot transport, the distribution of the electrostatic potential, +(x),
the concentration, n(x), and the velocity, :(x), have to be determined under collisionless
electron ight. We will exploit Poissons equation (6.68) and use the current density in
the form
J = en(x)v(x). (6.82)
The velocity, v, can be found from the Newton equation (2.8); however, we will use the
energy conservation law, which reads as
m
:
2
(x)
2
e+(x) =
m:
2
c
2
e+
c
= constant. (6.83)
Here, :
c
and +
c
are parameters corresponding to the injecting electrode (cathode). In
the case of a large electric bias, we can simplify the latter equation by considering that
the electrons are injected over the barrier with small velocity :
c
0. Then, we exploit
the virtual-cathode approximation discussed previously. It is convenient to set +
c
= 0
and obtain
:(x) =
_
2e+(x)
m
(6.84)
and
n(x) =
J
0
e:(x)
. (6.85)
where J
0
is the absolute value of the current density through the diode. As we noted in the
section devoted to dissipative transport, J is negative and it is convenient to introduce an
absolute value of current density, J
0
, which is constant throughout the entire sample. On
combining the latter formula with Poissons equation (Eq. (6.68)) we obtain the equation
for +(x):
d
2
+(x)
dx
2
=
J
0
c
0
c
_
m
2e+(x)
. (6.86)
On multiplying both sides of the last equation by d+(x),dx, we obtain an equation that
can be integrated. The result of integration is
1
2
_
d+
dx
_
2
=
2J
0
c
0
c
_
m
+
2e
C. (6.87)
The integration constant, C, can be set to zero, since within our approach as x 0 we
have d+,dx 0 and + 0. Now, we can rewrite the result as
1
+
1,4
d+
dx
= 2
_
J
0
c
0
c
_
m
2e
. (6.88)
Further integration gives us the nal result for the distribution of the electrostatic
potential:
+(x) =
3
2
_
J
0
c
0
c
_
2,3
_
m
2e
_
1,3
x
4,3
. (6.89)
Taking into account that +(L
x
) = +
0
, we can nd the currentvoltage characteristic of
the ballistic diode:
J
0.b
=
_
2
3
_
3,2
c
0
c
L
2
x
_
2e
m
+
3,2
0
. (6.90)
This equation is known as Childs law. It was found and veried rst for vacuum diodes,
where collisions of electrons were totally absent. Childs law differs considerably from
the currentvoltage characteristic of the dissipative diode given by Eq. (6.73).
Using Eq. (6.89), we can calculate other parameters of the ballistic diode. For example,
the transit time for the electrons is
t
tr.b
=
3
2
t
0.b
. t
0.b
= L
x
_
2m
e+
0
. (6.91)
with t
0.b
being the transit time of the ballistic electrons in the absence of space-charge
effects. The space-charge effects increase the transit time, t
tr.b
, by the factor 3,2.
It is instructive to compare the results obtained for space-charge-limited transport in
the dissipative and ballistic diodes. To make this comparison we assume that the lengths
of the diodes, L
x
, are equal and that the two diodes are biased equally. Then, from
Eqs. (6.75) and (6.91), for the ratio of the transit times we obtain
t
tr.b
t
tr.d
=
9
+
0
j
4
2eL
x
=
9
e
e+
0
4
2m
L
x
=
9l
e
8L
x
_1. (6.92)
where we used the relationship j = e
e
,m
, and introduced the maximum mean-free

path, l
e
=
e
:
m
, and the maximum electron velocity in the diode, :
m
=
2e+
0
,m
. The
ratio obtained is small, in accord with the condition of dissipative transport of Eq. (6.12).
That is, the ballistic diodes can provide much faster regimes of operation. Similarly, it
can be shown that, for a given electric bias, the currents in the ballistic diode are much
greater than those of the dissipative diode: J
0.b
,J
0.d
=
8
9
(
2
3
)
3,2
L
x
,l
e
1. Hence, the
ultimate collisionless transport has a number of advantages over the dissipative transport
due to the higher velocity of electrons.
In fact, in a real situation some collisions always occur. Then, this intermediate
case can be treated numerically by using a much more complex model. In Figs. 6.15(a)
and 6.15(b) the results of such numerical modeling for a GaAs diode of length L
x
=
0.4 m at temperature T = 300 K are presented. A typical value of mobility for high-
quality GaAs material is j =7500 cm
2
V
1
s
1
. For this mobility, the time of free ight
0 5 10 15
0.0
0.2
0.4
0.6
0.8
0.0 0.1 0.2 0.3 0.4 0.5
0.0
0.1
0.2

(b)
v (10
7
cms
1
)
F

(
v
,
x
)

x = 0.3 m
x = 0.05 m
x = 0 m
4
Lightly- doped slab Contact
2
x (m)
(a)
v
(
x
)

(
1
0
7
c
m
s
1
)

0
Contact
6
8
10
12
14
x = 0.1 m
v
T
Figure 6.15 The results of numerical modeling of a GaAs diode. (a) The distribution of electron
potential energy, e+(x), and average velocity, :(x), across the diode. (b) The distribution
function, F(:. x), at x = 0, 0.05, 0.1, and 0.3 m. The applied voltage, +
0
, is equal to 0.47 V;
the thermal velocity, :
T
, is equal to 2.6 10
7
cm s
1
. Reprinted with permission from H. U.
Baranger and J. W. Wilkins, Ballistic electrons in an inhomogeneous submicron structure:
thermal and contact effects, Phys. Rev. B, 30, 7349 (1984).
C _
1984 by the American Physical
Society.
is
e
= 2.9 10
13
s. At the given temperature, the thermal velocity of the electrons
is :
T
=
k
B
T,m
2.6 10
7
cms
1
. The average mean-free path of an electron is
l
e
= 0.075 m and L
x
,l
e
5. Thus, on average the electrons with velocity :
T
would
undergo as many as ve collisions during the ight across a uniformsample of the chosen
length, L
x
. For such a sample, the transport would be almost dissipative. For a strongly
biased diode, we reach quite different conclusions. In Fig. 6.15(a), the potential-energy
prole andthe average electronvelocityare presentedfor anappliedvoltage +
0
= 0.47V.
For comparison, the thermal velocity :
T
is also indicated. One can see that the maxi-
mum average velocity is above 7 10
7
cms
1
. The actual mean free path is estimated
to be larger, 0.07 m l
e
0.2 m. Thus, this example corresponds to an interme-
diate case between pure ballistic and dissipative transport. For such a case, different
electrons have different velocities and it is possible to characterize these electrons by
a distribution function, F, over the velocity, :. The distribution depends also on the
distance along the diode, x: F = F(:. x). In Fig. 6.15(b), the distribution function is
presented for various distances, x. The thermal distribution of the electrons in the cathode
is shown by the dotted line, and it is obviously the Maxwellian distribution of Eq. (6.16)
with the temperature T = 300 K. Actually, this distribution determines the velocities
of the injected electrons. The initial spreading of the electrons over velocities in the
cathode is one of the main differences from the simple model analyzed previously, in
which the electrons are injected with a near-zero velocity. Inside the diode, the distribu-
tion becomes strongly anisotropic: electrons with negative velocities are almost absent.
As the distance increases, the distribution becomes more and more anisotropic, with a
0 1 2 3 4 5 6 7 8 9 10
0.6
0.4
0.2
0.0
0.2
0.4
0.6
0.8
1.0
wt
tr,b
Z(wt
tr,b
)
Z(0)
1
2
Figure 6.16 Dependences of the real (curve 1) and imaginary (curve 2) parts of the impedance of
Eqs. (6.93) and (6.94) on frequency ; Z(0) = Z( = 0).
well-pronounced maximum. For larger distances, this maximum corresponds approxi-
mately to values given by the previously discussed model of pure ballistic transport. This
example highlights the real nature of an almost ballistic diode: some of the electrons par-
ticipate in transport and there occurs some spreading over the velocity. However, the aver-
age properties of the diode are very close to those obtained in the simple model based on
the classical Newton and Poisson equations.
The previously discussed simple model allows one to investigate high-frequency prop-
erties of the ballistic diode and calculate its impedance. The real and imaginary parts of
the impedance are
Re[Z(O)] =
12R
d
O
4
[2(1 cos O) Osin O]. (6.93)
Im[Z(O)] =
12R
d
O
4
[2 sin OO(1 cos O) O
3
,6]. (6.94)
Here, R
d
= d+
0
,dJ
0
is the differential resistance calculated with the use of the steady-
state currentvoltage characteristic of Eq. (6.90); O = t
tr.b
, i.e., the frequency measured
in units of the inverse transient time dened by Eq. (6.91). Both dependences, Re[Z(O)]
and Im[Z(O)], are presented in Fig. 6.16. According to Eqs. (6.93) and (6.94) and the
denition of R
d
, the magnitudes of Re[Z(O)] and Im[Z(O)] decrease as the voltage
bias, +
0
, increases:
R
d
=
d+
0
dJ
0
=
1
2
_
3m
e+
0
L
2
x
c
0
c
.
That is, the current response becomes larger at larger biases (R
d
1,
+
0
). Simultane-
ously, the frequency region within which the diode is still active increases with the bias
(
+
0
). These features are quite similar to those obtained for the dissipative Mott
Gurney diode; see Eq. (6.77). A principally new feature is the observable oscillations
of Re[Z()] and Im[Z()] with and the frequency windows arising with negative
E
F
x
(b)
(a)
Device Lead Lead
L R
Figure 6.17 (a) The structure schematic of a nanostructure mesoscopic device and (b) the
potential energy prole, e+(x), in the biased mesoscopic device.
values of Re[Z()]. The rst window with Re[Z()] - 0 occurs for between
6.3,t
tr.b
and 9,t
tr.b
. It is known that a negative Re[Z()] corresponds to an electrical
instability at frequency . Such an instability can be used for generation of high-
frequency electromagnetic oscillations, and we will discuss this phenomena in detail in
Section 8.2.
In conclusion, ballistic nanoscale devices are the fastest devices based on classical
electron transport. Both a nite electric current and a nite electric resistance in the
absence of electron scattering arise because of the electrostatic effects induced by redis-
tribution of charged electrons in the device. This is the reason why this case is associated
with charge-limited transport. The ballistic diodes are characterized also by frequency
windows within which the dynamic resistance of the device is negative, which signies
an electrical instability and the possibility of generating ultra-high-frequency electro-
magnetic radiation.
Quantum ballistic transport: the Landauer formula
The general theoretical description of the various quantum transport regimes is too
complex a problemto be presented in this book. We consider the simplest limiting case of
time-independent and low-temperature transport, when inelastic processes are negligible.
Thus, according to the classication given in Table 6.1 we consider now the mesoscopic
regime of electron transport. As has been pointed out previously, transport through
nanostructure mesoscopic devices depends on both the geometry of the nanostructure
and the leads (electrodes, contacts, wires, interconnects, etc.) which connect the device
to an external electric circuit. Hence, we should consider the whole system: the device
and the leads. Figure 6.17 illustrates the essential features of the simplied model. The
system consists of two leads, to which a bias is applied, two electron waveguides, L
and R (they can be thought of as quantum wires), and the device itself.
In order to avoid the need for a detailed description of the leads, we assume that the
leads are reservoirs of electrons, where energy and momentum relaxation processes are
so effective that the electrons remain in equilibrium even under a given applied voltage
bias. Hence, the boundary conditions at the interface between the leads and the device
are assumedtobe determinedbythe equilibriumFermi distributionfunctionof Eq. (6.18).
The electron concentration in the leads is so high that the electrostatic potential in each
lead is taken to be constant, as for the case of a metal. Let E
F
be the Fermi energy of the
electrons in the leads in the absence of a bias. On applying a voltage bias, +
0
, the Fermi
level in one of the leads becomes E
F
1
2
e+
0
, while that of the second lead becomes
E
F
1
2
e+
0
, as depicted in Fig. 6.17(b). Thus, the electron distribution functions are
F
F
_
E
1
2
e+
0
E
F
_
and F
F
_
E
1
2
e+
0
E
F
_
. (6.95)
at the left and right leads, respectively. Here E is the kinetic energy of the electrons.
In studying quantumwires in Section 3.3, we found that the electron spectrumconsists
of a series of one-dimensional subbands with the energies
E
k
x
.n.m
=
n.m
h
--2
k
2
x
2m
. (6.96)
Here, n and m are integers and k
x
is a one-dimensional wavevector directed along the
axis of the quantum wire. The electron wavefunction can be written as
(x. y. z) =
(y. z)
[[
(x). (6.97)
The component of the wavefunction
(y. z) describes the transverse distribution of

the electrons in the quantum wire, while
[[
(x) describes the translational motion of
electrons along the connecting quantum wires. It is generally composed of plane waves
e
ik
x
x
. Since electron transport occurs along the wires, we must analyze the wavefunc-
tion
[[
(x). We can write the wavefunction of the electrons,
[[
(x), at the interfaces
between the leads and the device. These interfaces are indicated in Fig. 6.17 by the L
and R cross-sections. The wavefunction,
[[.l
(x), of the electrons coming from the left
lead is
[[.l
(x) =
_
e
ik
l
(xx
l
)
r
l
e
ik
l
(xx
l
)
. x x
l
.
t
r
e
ik
r
(xx
r
)
. x x
r
.
(6.98)
Here, x
l
and x
r
denote the coordinates of the cross-sections L and R, respectively. That
is, at the cross-section L the wavefunction consists of incident and reected waves,
while at the cross-section R there is only the wave that has passed through the device.
The coefcients t
r
and r
l
are the amplitudes of these transmitted and reected waves,
respectively. These coefcients depend on the particular potential prole in the device,
its geometry, etc. In a similar manner, one can write the wavefunction,
[[.r
(x), of the
electrons coming from the right lead:
[[.r
(x) =
_
t
l
e
ik
l
(xx
l
)
. x x
l
.
e
ik
r
(xx
r
)
r
r
e
ik
r
(xx
r
)
. x x
r
.
(6.99)
In Eqs. (6.98) and (6.99), k
l
and k
r
are the wavevectors in the left and right cross-sections,
respectively.
The complex coefcients t
r
. r
l
. t
l
, and r
r
are connected by several fundamental rela-
tionships, which do not depend on the specic design of the device. The requirement of
current continuity for both wavefunctions,
[[.l
(x) and
[[.r
(x), gives
k
l
(1 [r
l
[
2
) = k
r
[t
r
[
2
and k
r
(1 [r
r
[
2
) = k
l
[t
l
[
2
. (6.100)
Other important relationships are
[r
l
[
2
= [r
r
[
2
and t
r
t
l
= t
r
t
l
. (6.101)
Then, using Eq. (6.100), we get
k
2
r
[t
r
[
2
= k
2
l
[t
l
[
2
. (6.102)
By substituting the wavefunctions of Eqs. (6.98) and (6.99) into Eq. (3.10), which
denes the particle ux, we calculate the incoming, i
in
, and outgoing, i
out
, electron ows:
i
in
= :
l
and i
out
= :
r
[t
r
[
2
. (6.103)
where :
l
= h
--
k
l
,m
and :
r
= h
--
k
r
,m
are the velocities at the L and R cross-sections,

respectively. The ratio of these quantities denes the transmission coefcient for
electrons moving through the device from the left to the right:
T
lr
(E) =
i
out
i
in
=
k
r
k
l
[t
r
[
2
. (6.104)
The transmission coefcient corresponding to the electrons moving through the device
from the right to the left equals
T
rl
(E) =
k
l
k
r
[t
l
[
2
. (6.105)
From Eqs. (6.102), (6.104), and (6.105), one can nd that
T
lr
(E) = T
rl
(E) = T(E
[[
). (6.106)
where
E
[[
=
h
--2
k
2
x
2m
(6.107)
is the kinetic energy corresponding to the longitudinal component of the electrons
momentum. Thus, the transmission coefcients are the same for both directions of
incoming electrons. The ratio of reected and incoming electron uxes denes the
reection coefcient:
R(E) =
i
r
i
in
= [r
l
[
2
= [r
r
[
2
. (6.108)
It is obvious that
T(E) R(E) = 1. (6.109)
Now, we can take into account contributions to the total current from all electrons
entering the device from both leads. Consider a state of the electrons, say in the left
lead, with quantum numbers k
x
. n, and m. The number of electrons in this state is given
by 2F
F
(E(k
x
. n. m)
1
2
e+
0
E
F
). where the factor 2 comes from spin degeneracy. If
the connection length is L
c
, the number of electrons per unit length of the connection is
2F
F
,L
c
. The total contribution to the electric current from the electrons entering from
the left is
I
l
=
2e
L
c
n.m
k
x
>0
:
[[
T(E
[[
)F
F
_
E(k
x
. n. m)
1
2
e+
0
E
F
_
. (6.110)
Similarly, for the electrons from the right lead one gets
I
r
=
2e
L
c
n.m
k
x
-0
:
[[
T(E
[[
)F
F
_
E(k
x
. n. m)
1
2
e+
0
E
F
_
. (6.111)
Therefore, the total current through the device is
I = I
l
I
r
=
2e
L
c
n.m
k
x
>0
:
[[
T(E
[[
)
_
F
F
_
E(k
x
. n. m)
1
2
e+
0
E
F
_
F
F
_
E(k
x
. n. m)
1
2
e+
0
E
F
__
. (6.112)
Since, for our model, the electron velocity, :
[[
, and the transmission coefcient, T, are
independent of the transverse quantum numbers, n and m, we can calculate the sum over
n and m. Taking into account the explicit form of the Fermi function, F
F
, it is convenient
to introduce the distribution function dependent only on the kinetic energy E
[[
:
F
F
(E
[[
) = 2
n.m
1
1 exp
_
E
[[
n.m
E
F
k
B
T
_. (6.113)
Next, as in Eq. (6.30) the summation over k
x
in Eq. (6.112) may be replaced by an
integration:
k
x
{. . .] L
c
_
dk
x
2
{. . .] = L
c
_
dE
[[
2h
--
:
[[
{. . .]. (6.114)
Finally, we obtain the following expression for the total current:
I = e
_
dE
[[
2h
--
T(E
[[
)
_
F
F
_
E
[[
1
2
e+
0
E
F
_
F
F
_
E
[[
1
2
e+
0
E
F
__
.
(6.115)
where the integration range runs over the kinetic energy of the longitudinal motion E
[[
.
Note that the electron velocity does not appear in the nal expression for the total current,
I . This general result may be applied to a variety of different cases. Let us consider two
of them.
A device macroscopically large in the transverse directions
In this case the transverse quantum numbers are the wavevectors n = k
y
and m = k
z
,
and
n.m
=
h
--2
2m
_
k
2
y
k
2
z
_
.
The function of Eq. (6.113) can be calculated explicitly,
F
F
(E
[[
) = S
m
k
B
T
h
--2
ln
_
1 exp
_
E
F
E
[[
k
B
T
__
. (6.116)
where S is the cross-sectional area. The term F
F
(E
[[
) has the meaning of the number of
electrons with energy E
[[
. Then the total current density, J = I ,S, is
J = e
m
k
B
T
h
--2
_
dE
[[
2h
--
T(E
[[
) ln
_
_
_
_
_
_
1 exp
_
E
F
E
[[
1
2
e+
0
k
B
T
_
1 exp
_
E
F
E
[[
1
2
e+
0
k
B
T
_
_
_
. (6.117)
This is indeed a useful result since Eq. (6.117) allows one to calculate the currentvoltage
characteristic of a nanostructure device and its dependence on the electron concentration,
temperature, etc. Interestingly, we found a nite current in this biased device, where no
scattering occurs. That is, the device possesses an electric resistance. The resistance is
explained by two factors: (1) quantum-mechanical reection of electron waves inside
the device and (2) the nite number of electrons which can be injected into the device.
Note that no electrical charge redistribution was considered, i.e., relatively small currents
were supposed.
Device conductance at low temperatures. The Landauer formula
Let us turn to the general result of Eq. (6.112). It can be simplied signicantly for near-
equilibrium transport at low temperatures. We know that in the limit of zero temperature
the Fermi distribution function, F
F
, becomes a step function:
lim
T0
F
F
(E E
F
) = O(E
F
E). (6.118)
If the applied voltage, +
0
, is small, the difference between the distribution functions in
the sum of Eq. (6.112) is equal to
F
F
_
E(k
x
. n. m)
1
2
e+
0
E
F
_
F
F
_
E(k
x
. n. m)
1
2
e+
0
E
F
_
= e+
0
(E
F
E).
(6.119)
where we have taken into account the fact that the derivative of the step-like Fermi
distribution function is a -function. Hence, the current is proportional to the voltage
bias, +
0
.
One can introduce the conductance of a nanostructure device as the ratio
G =
I
+
0
. (6.120)
From Eqs. (6.112), (6.114), (6.119), and (6.120) we obtain the conductance at low tem-
peratures in the form
G =
e
2
h
n.m.s
T(E
F
. n. m) = 2
e
2
h
n.m
T(E
F
. n. m). (6.121)
where the sum extends only over the electron states (n. m) with energy E - E
F
. The
coefcient in front of the sum of Eq. (6.121),
G
0
=
e
2
h
. (6.122)
is called the quantum of conductance; here h is Plancks constant; h = 2h
--
, where h
--
is the reduced Planck constant introduced previously in Chapter 2. The quantum of
conductance is equal to G
0
=39.6 S (1 siemens =1 S =1 AV
1
) and its inverse value
is 1,G
0
= 25.2 kO. Equation (6.121) is often called the Landauer formula.
Sometimes it is convenient to consider electron states corresponding to different quan-
tum numbers (n. m) in terms of separate electron conduction channels. In this latter
formulation, Eq. (6.121), may be rewritten in the form
G = 2G
0
n.m
T
n.m
with T
n.m
= T(E
F
. n. m). (6.123)
Here each channel (n. m) contributes G
0
T
n.m
to the conductance, G. If the channel cor-
responding to the (n. m) state is transparent to electrons, T
n.m
= 1, and the contribution
of this channel is equal to the quantum of conductance of Eq. (6.122).
This regime is also referred to as quantum ballistics. Thus, a nanoscale device with
quantum ballistic transport exhibits nite conductance (nite resistance). Generally, the
conductance depends both on the transmission coefcient of the device and on the Fermi
distribution functions in the leads. In particular, even if the device is entirely trans-
parent, T = 1, the conductance remains nite and is equal to the quantum of conduc-
tance, G
0
. An increase in the occupation of the upper low-dimensional subbands gives
rise to an unusual behavior of the conductance. Indeed, at low temperature, actually at
k
B
T _ E
F
, only those subbands for which the energy of the subband bottoms
n.m
- E
F
are occupied. If, on changing the Fermi energy, E
F
, a new subband starts to populate,
the conductance of the device increases in a step-like manner, independently of how
many electrons occupy this newsubband. In conclusion, the Landauer formula describes
quantumtransport in mesoscopic devices. It is valid at lowtemperature and small voltage
bias.
The simplest device with quantum ballistic transport is the so-called quantum point
contact. It can be fabricated in a number of fashions. The essence of the quantum point
contact is that two electron reservoirs are connected through a conducting region with
transverse dimensions, L
z
and L
y
, comparable to the de Broglie wavelength of the
electrons. In Fig. 6.18(a), a sketch of a point contact on the {x. y]-plane is depicted.
The electron transport and current occur along the x direction. For simplicity, we can
assume that the channel has a constant dimension, L
x
, in the x direction. If the variation
of the channel dimension in the y direction, L
y
(x), is adiabatically smooth, we can use
Eq. (3.50) for the one-dimensional energy subbands:
E
k
x
.n.m
=
h
--2
2
n
2
2m
L
2
z
h
--2
2
m
2
2m
[L
y
(x)]
2

h
--2
k
2
x
2m
. (6.124)
where we used approximate quantization of electron motion in the y direction. The
resulting potential proles are shown in Fig. 6.18(b), for a few subbands in the structure
d
x
E
x
E
F
m = 1, 2, 3
L
y
(x)
(a)
(b)
0
y = 0.5d + ax
2
y = (0.5d + ax
2
)
y
Figure 6.18 Geometrical representations of point contact (a) and quantized energy
1.m
in the
region of point contact (L
z
_ L
y.min
) (b). It is shown schematically that electrons can be in all
three subbands in the region with the densest hatching. Electrons can be only in the subband
1.1
in the region shown by the least-dense hatching.
with L
z
_ L
y.min
. For the indicated Fermi energy, we have only one channel open for
electron motion and the second channel has a small barrier. As soon as the Fermi energy
reaches the bottom of the next subband, a new channel opens for conducting electrons
and the conductance of this point contact undergoes a step-like increase. Typically,
such narrow constrictions clearly demonstrate the quantization of conductance. Since
the potential barriers in the constriction are quite wide, tunneling processes have small
probability. As a result, each of the channels is either almost open (the transmission
coefcient T = 1) or almost closed (T = 0); i.e., one will expect steps in the conductance
close to their universal value 2G
0
.
Consider as an example the narrow constriction dened by shallow etching on the
AlGaAs/GaAs structure with two-dimensional electrons. In order to performthe shallow
etching with well-controlled edges, a 13-nm-thick Al mask was formed by electron-
beam lithography and lift-off, using a eld-emission scanning electron microscope at
an acceleration voltage of 2.5 kV. The sample was then shallow etched, as described in
Chapter 4. After the etching mask was removed, an Al gate electrode was deposited,
covering the constriction. By applying a voltage to the gate, one can control the electron
density and, thus, the Fermi energy. A scanning electron micrograph of a shallow etched
constriction is shown in the inset of Fig. 6.19. The shape of the etched constriction
is characterized by two parameters: (1) the width of the constriction, d, and (2) the
curvature, a, of the parabolas describing the shallow etched walls of the constriction
0.6 0.7 0.8 0.9 1.0
16
14
12
0
2
4
6
8
10
31 K
20 K
11 K
1.3 K
G

(
e
2
/
h
)
0
(V)
Figure 6.19 The quantization of conductance, G, at temperatures T = 1.3, 11, 20, and 31 K.
Inset: a shallow etched constriction. From A. Kristensen, J. B. Jensen et al., Conductance
quantization above 30 K in GaAlAs shallow-etched quantum point contacts smoothly joined
to the background 2DEG, J. Appl. Phys., 83, 607 (1998). Reprinted with permission from
A. Kristensen, J. B. Jensen, M. Zaffalon, C. B. Srensen, S. M. Reimann, P. E. Lindelof,
M. Michel, and A. Forchel, Journal of Applied Physics, 83, 607 (1998).
C _
1998 American
Institute of Physics.
seen in the picture. If the {x. y]-coordinate system is placed in the center of symmetry
of the constriction, the etched walls are described by
y = (0.5d ax
2
). (6.125)
The quantum contact presented in Fig. 6.19 has the parameters d = 50 nm and
a = 0.001 25 nm
1
. Using the second term in Eq. (6.124), we can estimate the energy
distances between the lowest subbands in the narrow cross-section of the constriction;
they are 20 meV, 16 meV, 11 meV, etc. The sample exhibits well-dened conductance
quantization at lowtemperatures, as depicted in Fig. 6.19 for a series of measurements at
various temperatures. The rise of the conductance for each step is close to the universal
value 2G
0
.
Single-electron transport
In the previously discussed cases of electron transport, we assumed that the number of
electrons participating in this transport is so large that the discrete nature of the elec-
trons does not matter. For small devices operating with weak currents, this assumption
is no longer valid. Instead, a new type of charge-dependent electron transport, single-
electron transport, develops. Below we present the basic concepts of this transport
regime.
In general, in bulk-like materials and macroscopic devices the electron discreteness is
not manifested in average characteristics such as the local electron density, total electric
current, etc. However, it is well known that this discreteness manifests itself by con-
tributing to current noise (uctuations) even for macroscopic samples. For example, the
so-called shot noise is entirely due to the discreteness of the electron charge; it arises
because of the random process of electrons entering the device.
When the dimensions of the devices are scaled down, the role of the discreteness
of charge increases. In the case of ultra-small devices, the discreteness of the electron
charge gives rise to principally new effects in electron transport. This transport becomes
correlated; i.e., a transfer of one electron through the device depends upon the transfer
of others. The correlation appears because of the Coulomb interaction of individual
electrons. A new class of devices, referred to as single-electron devices, is based on such
processes.
Single-electron effects rely on a charging process that occurs when electrons enter
a tiny conducting sample. When the tiny conducting sample, often called the metal-
lic island, is extremely small, the electrostatic potential of the island signicantly
increases even when only one electron enters it. In general, the charging energy of a
sample is
E
C
=
Q
2
2C
. (6.126)
where Q is the charge and C is the capacitance of the sample. For a spherical island of
radius r, the capacitance, C, can be estimated as
C = 4c
0
cr. (6.127)
For example, an island with radius r = 10 nm has a capacitance of the order of 10 aF =
10
17
F, where F is the farad the unit of capacitance, C (1 aF = 1 attofarad = 10
18
F
and 1 F = 1 CV
1
). Then, with an increase in the voltage, L+ (L+ = LQ,C), which
is equal to e,C with the electron charge e = 1.602 10
19
C, the increase in the energy,
LE = e L+, reaches 16 meV. This is comparable to the thermal noise energy at room
temperature, k
B
T 26 meV. If one electron is transferred to the island, the Coulomb
repulsion prevents additional electrons fromentering the island unless the island potential
is intentionally lowered by an external bias. If the island potential is lowered gradually,
the other electrons can enter the island only one by one, with negligibly small power
dissipation.
We can dene a single-electron device as a nanostructure with so small a capacitance
that a single electron added to the device generates a measured voltage change on the
device. The basic properties of single-electron devices can be described by using the
example of a customary system: an insulated tunnel junction, I, between two conducting
electrodes, M (heavily doped semiconductor regions or metals), as shown in Fig. 6.20
for a metalinsulatormetal (MIM) system. Let this junction be characterized by the
~ 10
M M I
Tunnel
junction
Figure 6.20 The simplest MIM (metalinsulatormetal) system of single electronics: a tunnel
junction with a small capacitance, C. Single-electron transfer is shown schematically.
capacitance, C, and the conductance, G. We can suppose that the capacitance is roughly
proportional to the cross-section, S, of this junction. Thus, a small S implies a small
capacitance. The conductance of the junction, G, is small enough for one to consider
the system as a leaking capacitor. Let the system have a charge, Q. Then, when a single
electron leaks through the insulator, this event changes the initial electrostatic charge
just by the elementary charge, e: Q (Q e). Hence, according to Eq. (6.126), the
electrostatic energy of the junction changes by
LE
C
=
(Q e)
2
2C

Q
2
2C
=
e(2Q e)
2C
. (6.128)
The potential difference between the plates changes from
+
1
= Q,C (6.129)
to
+
2
= (Q e),C. (6.130)
Then, the single-electron transfer leads to a voltage uctuation across the junction equal
to
L+ = +
1
+
2
= e,C. (6.131)
and a corresponding uctuation of the leakage current,
LI = G L+. (6.132)
The current related to the single-electron transfer can be estimated from the following
qualitative considerations. The uncertainty relation of Eq. (3.8) between the energy and
time allows us to estimate the extreme limit of the tunneling time,
t
, for such a junction:
t

h
LE
C
. (6.133)
In turn, the electric current
I
t
=
LQ
Lt
(6.134)
0 e/2 e e/2 e
I
S
Q
Q
2
2C
E
C
=
Figure 6.21 The charging energy, E
C
, for a tunnel junction with the capacitance, C, and charge,
Q, of one of the plates. The solid arrow indicates energetically favorable electron-tunneling
events, the dashed arrow indicates an energetically unfavorable event, and the dotted double
arrow indicates tunneling events without a change in energy of the capacitor, E
C
. Inset: a circuit
with charging of a tunnel junction by a current source.
associated with the tunneling of a single electron (LQ = e, Lt =
t
, and LE
C
=
e
2
,(2C)) is
I
t

e
e
3
2hC
. (6.135)
If this current, I
t
, exceeds the current uctuation due to the voltage uctuation, L+,
namely
I
t
LI = G L+ = G
e
C
. (6.136)
the single-electron transfer processes will control the electric current through the junc-
tion. The latter inequality leads to the criterion
e
2
h
G. (6.137)
Note that the value on the left-hand side of this criterion just coincides with the quantum
of conductance, G
0
e
2
,h, introduced by the Landauer formula of Eq. (6.121).
As a result of this analysis, we can state that, if the Coulomb energy of charging of the
device is greater than the thermal energy, k
B
T, and the current, I , associated with the
single-electron transport is greater than uctuations of the leakage current, then electron
transport is correlated and single-electron effects are important. The criteria for such
single-electron transport can be formulated as follows:
C _
e
2
k
B
T
and G _
e
2
h
; (6.138)
i.e., both the capacitance and the leakage conductance of the device should be small.
Nowwe can discuss the physics of single-electron transfer by using the electric circuit
shown in Fig. 6.21. We assume a situation in which the tunnel junction in the circuit is
charged by a current source, I
S
. Starting with no net charge on the capacitor plates at
time t = 0, the current source slowly begins to charge the junction. For a given charging
rate and short time scales, the excess charge on the capacitor plates Q - e. Importantly,
I
0
0
e
C
e
C
+
Figure 6.22 The currentvoltage characteristic under Coulomb blockade. Within the voltage-bias
range from e,C to e,C the current is suppressed by Coulomb correlations.
macroscopic plates can be charged with a charge less than the charge of a single electron
just by shifting the conduction electrons in the plates with respect to the positive ions of
the lattice (the so-calledpolarizationcharge). Whenthe current source has chargedone of
the plates of the capacitor to a charge equal to e,2 and another plate to e,2, the plates
have the charge difference of 1e between them. The electrostatic energy of the junction
reaches the value marked by the dot in Fig. 6.21. Now 1e charge can tunnel through the
junction as shown by the dotted arrow in Fig. 6.21. Thus, before the tunneling the initial
charges on the plates were (e,2. e,2), whereas after the tunneling the charges are
(e,2. e,2). We see fromFig. 6.21 that the charge transfer does not change the energy
of the junction. The two charge congurations are, actually, identical. Because of this,
nothing can stop the electron from tunneling back to its initial state and so on. However,
the current source continues to charge the plates and the tunneling shown by the solid
arrow becomes energetically favorable (LE
C
in Eq. (6.128) becomes negative), and it
becomes energetically unfavorable for the electron to tunnel back to its initial capacitor
plate. Thus, in reality, we obtain tunneling of a single electron and the system starts
to accumulate the charge to provide for the tunneling of the next electron, which will
happen after some time has elapsed. The succession of electron-tunneling events means
that the electron ux is highly correlated. In Fig. 6.21 all discussed tunneling events
are shown schematically. From the considerations based on Eq. (6.128) one can see that
any electron transfer is prohibited for a small initial capacitor charge: e,2 - Q - e,2
(LE
C
for such Q is positive). This physical effect is called Coulomb blockade. If the
conditions of Eq. (6.138) are met, the charging energy plays the dominant role in the
system, the tunneling of an electron is energetically unfavorable, and at low temperature
tunneling is not possible at all (if it is blocked). This results in the specic current
voltage characteristic presented in Fig. 6.22. The main feature of such a characteristic
is the total suppression of the current in some nite interval of external voltage biases,
e,C - +
0
- e,C.
Importantly, the manifestation of the correlated transport strongly depends on the
external circuit to which the single-electron tunnel junction is attached. Let the exter-
nal circuit impedance be Z(). If the impedance is as low as [Z()[ _ G
1
0
for the
most important frequencies, 1,
t
e
2
,(hC), then charge uctuations in the circuit
are greater than the elementary charge, e, and all correlation effects are suppressed.
t
Q
0
e
2
e
2
+
T
SET
T
SET
=
1
f
SET
e
I
=
Figure 6.23 Bloch oscillations, showing the dependence of junction charging on time.
If the external circuit impedance is in the intermediate interval G
1
0
_[Z()[ - G
1
.
the junction exhibits Coulomb blockade in the bias range from e,C to e,C, as
shown in Fig. 6.22, but outside this range there is no correlation between tunneling
events.
Finally, an interesting current regime occurs if [Z()[ G
1
G
1
0
. when the exter-
nal circuit can be considered as a source of a xed direct current, I . This current causes
a recharging of the junction inside the range of Coulomb blockade without electron tun-
neling; it corresponds to a linear change in the charge with time: dQ,dt = I constant.
When the edge of the blockade range is reached, an electron tunnels through the junction.
The system nds itself again in the blockade range near the opposite edge and the
process repeats; see Fig. 6.23. Thus, one gets temporal oscillations of the charging with
a frequency, f
SET
, determined by the current:
f
SET
=
I
e
. (6.139)
This is the so-called frequency of single-electron tunneling (or Bloch) oscillations.
The previously discussed nonlinear currentvoltage characteristics, oscillations, and
other single-electron effects give rise to a principally new approach to low-energy elec-
tronics. This eld of single-electron devices is developing rapidly and has many potential
applications. Devices such as the single-electron transistor, the turnstile, and the single-
electron pump, have been proposed and realized on the basis of these effects. Though
these results were achieved at lowtemperatures, modern technology portends their exten-
sion to liquid-nitrogen temperature and even room temperature.
6.6 Closing remarks
In this chapter, we have shown that a number of different electron-transport regimes
can occur in semiconductors and their nanostructures. These regimes are characterized
by different values of the electron velocity and magnitude of the current, very different
currentvoltage dependences, etc. The electrons can behave as semiclassical particles,
or as quantum ones. If the device dimension along the current is much larger than the
electron de Broglie wavelength, electron motion, typically, is classical. It can occur in a
dissipative manner when electrons undergo multiple collisions caused by crystal defects
andlattice vibrations. The rate of these collisions determines the average electronvelocity
achievable at a given electric eld.
If the eld is small, the average velocity is a linear function of the eld. The coef-
cient in the linear relationship between the average velocity and the eld is the electron
(hole) mobility. In perfect materials and structures the mobility is high and limited
by scattering by lattice vibrations. For such systems, at low temperatures, at which
these vibrations are reduced, the mobility reaches a maximum magnitude that is lim-
ited by scattering by impurities. The mobility is one of the most important charac-
teristics for electronic applications. The electron concentration, mobility, and geome-
try of a sample determine its electrical conductance or electrical resistance. We have
analyzed modication of the transport in oscillating electric elds and found that the
current response is dened by a complex and frequency-dependent characteristic the
impedance.
Then, we considered the behavior of the electrons in high electric elds, explaining
hot-electron effects and transit-time effects. Among the latter, the velocity-overshoot
phenomenon, i.e., achievement of very high speed for a short period of time, is the most
important for ultra-high-speed devices.
Practically, the transient time effects, including velocity overshoot, can be realized in
short devices. In short devices, another type of physical effects should also be considered.
Specically, the formation of space charge affects the current. The electron space charge
increases the electric resistance; this case is known as space-charge-limited transport.
We have considered and compared several short diodes of different lengths, from the
dissipative diode to the ballistic device. We have found that their electrical properties are
markedly different from those of bulk-like samples; in particular, the currentvoltage
characteristics are essentially nonlinear. Calculations of the impedance, which deter-
mines the high-frequency properties of these diodes, showed that the range of frequen-
cies within which the device response is high is directly related to the time of electron
transit through the device. Thus, ultra-high frequencies can be achieved only for very
short (nanoscale) devices.
We have analyzed the simplest types of quantum transport and have come to a very
unusual conclusion the quantum device possesses a nite resistance (conductance)
even if scattering is entirely absent. The physical reason for such a nite resistance in the
absence of scattering is quantum-mechanical reection of electron waves from the inter-
faces between terminals and the quantum device. We have discussed the quantization of
the conductance in a quantum device with essentially one-dimensional electron motion.
The conductance quantization is related to electron energy quantization and manifests
itself as a multiple-step-like behavior of the electric current.
Finally, we have considered electric currents in devices so small that the electrons
can be transferred only one by one, because of Coulomb repulsion. The associated
single-electron-transport regime has very unusual properties, which can be exploited in
ultra-small electronic devices.
6.7 Problems 215
Additional information on the various regimes of electric current can be found in
K. Hess, Advanced Theory of Semiconductor Devices (NewYork, WileyIEEE Press,
1999).
D. K. Ferry, Semiconductors (New York, Macmillan Publishing Company, 1991).
I. I. Ipatova and V. Mitin, Introduction to Solid State Electronics (NewYork, Addison-
Wesley Publishing Company, 1996).
For much more information on quantumtransport and its applications to device physics
the reader can consult the following references:
R. Landauer, Electrical resistance of disordered one-dimensional lattices, Philos.
Mag., 21, 863 (1970).
H. Grubin, D. K. Ferry, and C. Jacoboni (Eds.), The Physics of Submicron Semicon-
ductor Devices (New York, Plenum Press, 1989).
C. Jacoboni, L. Reggiani, and D. K. Ferry (Eds.), Quantum Transport in Semiconduc-
tors (New York, Plenum Press, 1992).
Discussions of charge-limited transport and single-electron transport are presented,
for example, in the following references:
M. A. Lampert and P. Mark, Current Injection in Solids (New York, Academic Press,
1970).
K. K. Likharev, Correlated discrete transfer of single electrons in ultrasmall tunnel
junctions, IBM J. Res. Develop., 12, 144 (1988).
H. Koch and H. Lubbig (Eds.), Single-Electron Tunneling and Mesoscopic Devices
(Berlin, Springer-Verlag, 1992).
6.7 Problems
1. Consider the bulk crystal of GaAs with an electron effective mass m
= 0.067m
0
,
where m
0
is the free-electron mass, and mobility j = 10
5
cm
2
V
1
s
1
at T = 77 K.
Using the relationships introduced in Sections 6.2 and 6.5, and Eq. (6.50) (j =
e
e
,m
), calculate
the de Broglie wavelength, ;
the scattering time,
e
;
the thermal electron velocity, :
T
;
the mean free path, l
e
; and
the diffusion coefcient, D (D = :
2
T
e
,, = 3 for bulk material).
Determine the transport regimes for devices with feature sizes L
x
=0.05, 0.5, and 5 m.
2. Consider a bulk crystal, a quantum well of thickness 10 nm, and a quantum wire of
cross-section 10 nm10 nm. For such samples, the specic electron concentrations of
10
18
cm
3
, 10
12
cm
2
, and 10
6
cm
1
correspond to the same three-dimensional electron
density. For these three cases, by using the results of Section 6.3, calculate and compare
values of
the Fermi wavevector, k
F
, and corresponding de Broglie wavelength ( = 2,k
F
);
the Fermi energy, E
F
; and
the Fermi velocity, :
F
=
2E
F
,m
.
3. Assuming an ambient temperature T = 4 K, prove that the electron gas considered
in the previous problem is degenerate for all three cases. For a nondegenerate gas the
thermal energy is E
T
= (,2)k
B
T, where = 3, 2, and 1 for bulk, quantum well, and
quantumwire, respectively, and the thermal velocity is :
T
=
2E
T
,m
. Calculate these
parameters for further comparison. Using the Pauli exclusion principle, explain why the
Fermi energy and the Fermi velocity of the electrons are higher than the corresponding
characteristics E
T
and :
T
for a nondegenerate gas of particles with the same mass m
.
4. As studied in Section 4.4 for many semiconductor materials, the energy spectra of
the holes are degenerate and consist of two branches: the light holes, E
lh
= h
--2
k
2
,(2m
lh
),
and the heavy holes, E
hh
= h
--2
k
2
,(2m
hh
). Let both hole subbands be populated, with
concentrations n
lh
and n
hh
, respectively. At low temperature, the holes in both subbands
have the same Fermi energy. Find the relationship between the concentrations of the
heavy and light holes.
5. The density of states as a function of the energy has a staircase character for quantum
wells. Using Eq. (6.45), estimate the height of the stairs for the density per unit area in
the case of a GaAs quantum well.
6. Use the energy spectra of light and heavy holes presented in Problem 4 and compare
the densities of states for the light and heavy holes in bulk samples. Which of these two
hole subbands is more populated under equilibrium conditions?
7. Consider a short diode of length L
x
with space-charge-limited dissipative transport.
Let +
0
be the voltage applied to the diode. To make estimates for this device one should
specify the criteria of validity of the concept of the dissipative diode. For this, one should
combine Eq. (6.12) and the condition of the absence of electron heating in the diode,
i.e., F(x) - F
he
with F
he
dened by Eq. (6.81).
Using the estimate for the eld F +
0
,L
x
, prove that the diode concept works for
the voltage k
B
T,e - +
0
- L
x
_
k
B
T,(ej
E
). Show that these criteria can be met under
the condition L
x
>
j
E
k
B
T,e.
Assuming room temperature and the material parameters j = 10
3
cm
2
V
1
s
1
and
E
=10
10
s, calculate the critical length, L
x
. For L
x
=5 mand +
0
=0.1 V, determine
the electric current density, J, and the electron transit time,
tr
.
8. Consider a GaAs diode, in which the electrons have the effective mass m
=
0.067m
0
. Assume liquid-nitrogen temperature, T = 77 K. For such a temperature in
high-quality GaAs material, the typical mobility is j = 10
5
cm
2
V
1
s
1
.
6.7 Problems 217

U (eV)
x (nm)
0.2
0 50 10
Figure 6.24 A potential barrier U(x) = U
0
,cosh(x,d) with U
0
= 0.2 eV and d = 10 nm.
Estimate the mean free path, l
e
=
e
:
T
. Set the diode length L
x
= l
e
; i.e., assume
that the diode is a ballistic device.
Calculate the current density, J, and the transit time,
tr
, for the ballistic diode at +
0
=
0.1 V and compare your result with parameters obtained for Problem 7.
9. The Landauer formula of Eq. (6.121) includes the energy-dependent transmis-
sion coefcient, T(E). It is instructive to discuss this formula by using an exam-
ple for which the exact solution of the quantum-mechanical Schr odinger equation is
known. One such case is the electron motion in the potential of a special dependence:
U(x) = U
0
,cosh(x,d) (Fig. 6.24). One can see that this potential barrier has height equal
to U
0
and the characteristic spatial scale d. It was found that the transmission coefcient
has the form
T(E) =
sinh
2
(k
x
d)
sinh
2
(k
x
d) cosh
2
_
_
2
_
8m
U
0
d
2
h
--2
1
_
_
. for
8m
U
0
d
2
h
--2
> 1.
where k
x
=
2m
E,h
--
is the wavevector of the incident electron. Assume that only the
one-dimensional subband with n = m = 1 is populated by electrons. Set the following
parameters: U
0
= 0.2 eV, d = 10 nm, and m
= 0.067m
0
, as for GaAs.
Using the dependence T(E), calculate the conductance, G, as a function of the Fermi
energy.
7 Electrons in traditional
low-dimensional structures
7.1 Introduction
Now, we begin our analysis of novel developments in electronics that have resulted from
the use of nanostructures in modern electronic devices. Importantly, the attributes of
nanotechnology make it possible to pursue both devices with smaller dimensional scales
and novel types of device. Though the ongoing trend of miniaturization in electronics
is extremely important, the unique properties of electrons in nanostructures give rise
to novel electrical and optical effects, and open the way to new device concepts. The
electric current and voltage in a device are determined by two major factors: the con-
centration and the transport properties of the charge carriers. In nanostructures, these
factors can be controlled over wide ranges. In this and the next chapter we will study
nanostructures for whichthese basic factors that are important for the electronics are engi-
neered, whichare beingexploitedintensivelybothinresearchlaboratories andinpractical
nanoelectronics.
To distinguish the nanostructures already having applications fromthe newly emerging
systems, we refer to the former as traditional low-dimensional structures.
7.2 Electrons in quantum wells
In this section, we consider a few particular examples of nanostructures with two-
dimensional electrons.
As a basis for the further analysis, we will recall and develop several of the previously
introduced denitions and properties of an electron gas. In what follows, the effect of
the band offset arising at a junction of two semiconducting materials, which was dened
in Section 4.5 via electron afnities, is critically important. As discussed previously, the
electron afnity is the energy required to remove an electron from the bottom of the
conduction band of a specic material to vacuum. The difference in afnities of two
adjacent materials is equal to the conduction-band offset at their heterojunction. Another
useful parameter of a material is the work function, which is dened as an energy required
to remove an electron from the Fermi level of the material to vacuum. Finally, we will
use also the fundamental property of electron statistics which implies that in equilibrium
the Fermi level is constant throughout the whole system.
1
Charge transfer
E
F
E
c
V
b
Unstable state
(a)
z
Thermal equilibrium
z
E
c
E
F
+
+
+
+
(b)
V
b
(z)
e (z)
2DEG
Figure 7.1 The charge-transfer effect in a selectively doped single heterostructure. (a) The
step-like prole is unstable. (b) Charge transfer results in ionization of a layer of the doped
material, band bending, lining up the Fermi levels across the structure, and formation of an
electron channel at the interface.
Single modulation-doped heterojunctions
First, let us analyze qualitatively a single semiconductor heterojunction. We consider a
junction of two semiconductors with afnities such that they result in a discontinuity
of the conduction band as indicated by Fig. 7.1(a). Then, we assume that the barrier
semiconductor material is n-doped while the narrow-bandgap material on the right-hand
side of the structure is undoped. In real situations, the latter material is, usually, lightly
doped by acceptors; i.e., it is a p-type material.
The doped regions of the systemx the positions of the Fermi level, E
F
. At equilibrium,
the energy levels below E
F
are occupied by electrons. Accordingly, the scheme sketched
in Fig. 7.1(a) is unstable. The electrons will move toward the undoped crystal until an
electrostatic eld, brought about by the redistribution of electrical charge, bends the band
edges so that the Fermi level becomes constant across the materials; i.e., we obtain the
same Fermi level in the n-doped region and in the undoped region. Instead of having an
energy step as in Fig. 7.1(a), one obtains the situation shown in Fig. 7.1(b). That is, the
band edges are bent, there is ionization of impurities in some region of the doped part of
the system, and there are free electrons inside the potential well. This well is formed by
both (1) the bandgap discontinuity and (2) the electrostatic potential. The situation just
described is frequently referred to as formation of an electron gas at the interface.
Three important conclusions can be derived immediately from these considerations.
(i) Though both the initial materials were insulators (at least at low temperatures), now
at the interface, near the junction, one obtains an electron channel and electron
concentration that are nite even at T = 0 K.
(ii) Charge carriers in the potential well are separated spatially from their parent impu-
rities from the barrier side of the structure. Charged impurities, which usually lead
to large scattering rates and low mobilities, serve only as sources of carriers, and
scattering of electrons from the potential well by the impurities in the barrier is
suppressed as a result of the spatial separation of electrons from impurities.
(iii) The bending of the energy band creates a conning potential for carriers in one
direction, say along the z-axis. Hence there is a quantization of electrons in the z
direction and the establishment of a two-dimensional electron gas is quite possible.
220 Traditional low-dimensional structures
To complete the discussion of this simple model, let us note that in the real situation the
side of the heterostructure with the lower conduction-band edge is usually doped lightly
with acceptors. In this case, the right barrier of the well is higher than half of the bandgap
of the narrow-bandgap material. Of course, the residual acceptors reduce the electron
channel mobility but their effect is weak.
Here, we considered a heterostructure with the Fermi level dened only by the modula-
tion doping. These structures are called ungated heterostructures. For gated heterostruc-
tures the physical picture is slightly different. Such gated heterostructures will be studied
subsequently.
Basic equations describing the physics of the electrons at an interface
It is important to have an idea of the basic equations describing the electrons at an inter-
face. We began by considering a step-like discontinuity of the energy band and found that
the potential well is formed by charge transfer in space. Thus, the shape of the potential
is determined by the charges of all electrons on the interface and ionized impurities. On
the other hand, this many-body potential determines the motion of each electron and the
total number of ionized impurities. Thus, we face the so-called self-consistent problem:
the potential is dened by the concentration of electrons and ionized impurities, and it,
in turn, affects their redistribution. The simplest approach to this self-consistent problem
is to treat the electron quantization in the scheme wherein the potential is described by
a self-consistent electrostatic eld. The corresponding electrostatic potential, +(z), can
be thought of as a function dependent on only the coordinate z perpendicular to the
interface and is a solution of Poissons equation:
d
2
+(z)
dz
2
=
e
c
0
c
_
(r)[
2
F(E
) N
D
(z) N
A
(z)
_
. (7.1)
Here, e is the elementary electrical charge, c is the dielectric constant, and c
0
is the
permittivity of free space. The total charge consists of the charges of the electrons
and impurities. Let the wavefunctions of the electrons be
(r), with being the set

of electron quantum numbers. Then, [
(r) [
2
represents the probability of nding the
electron of the state labeled by at the point r. In order to calculate the contribution of
the electrons to the space charge we introduce the energy-dependent electron distribution
function, F(E
), which determines the probability of the electron occupying the energy

level E
. Thus, the electron charge density is e
(r)[
2
F(E
). The charge of the

nonuniformly distributed positive donors and negative acceptors is determined by their
densities, eN
D
(z), and, eN
A
(z), respectively.
We assumed that the potential, +(z), does not depend on the x- and y-coordinates.
Thus,
(r) can be factorized as discussed in Section 3.2:
(r) =
1
S
e
i(k
x
xk
y
y)
j
(z). { j. k
[[
]. (7.2)
where S is the area of the junction and k
[[
= {k
x
. k
y
]. The factorization (7.2) leads to
(r)
(r) = [
(r)[
2
=
1
S
[
j
(z)[
2
. (7.3)
which makes Eq. (7.1) dependent only on the z-coordinate. We also obtain the one-
dimensional Schr odinger equation for the z direction:
_
h
--2
2m
d
2
dz
2
V
b
(z) e+(z)
_
j
(z) =
j
j
(z). (7.4)
Here, the total potential energy consists of two contributions: the built-in potential of
the heterostructure, V
b
(z), and the self-consistent potential, e+(z). In our case, V
b
(z)
corresponds to the energy-band discontinuity at the junction see Fig. 7.1(a) which
depends only on the z-coordinate. We set
V
b
(z) = V
b
O(z).
where O(z) is the Heaviside step-function. In Eq. (7.4),
j
are the energies of two-
dimensional subbands. Thus, the total electron energy is given by Eq. (6.27):
E
j
(k
[[
) =
j

h
--2
k
2
[[
2m
.
Now, we can calculate the electron concentration:
[
2
F(E
) =
1
S
s. j.k
[[
[
j
(z)[
2
F
_
j

h
--2
k
2
[[
2m
_
=
j
[
j
(z)[
2
n
j
. (7.5)
Here, we also introduce the sheet density of electrons for the level j :
n
s. j
(E
F
)
1
S
s.k
x
k
y
F(s. k
[[
. j ) =
m
k
B
T
h
--2
ln
_
1 exp
_
E
F
j
k
B
T
__
. (7.6)
Actually, this result repeats Eq. (6.116). The quantity n
s. j
is a function of the temperature,
T, and takes the simplest form as T 0:
n
s. j
(T = 0) =
m
h
--2
(E
F
j
)O(E
F
j
). (7.7)
Equation (7.7) indicates that the level j is occupied if the Fermi level exceeds the
corresponding energy of quantization of transverse electron propagation,
j
.
It is necessary to formulate boundary conditions for the coupled differential equations
represented by Eqs. (7.4) and (7.1). For localized electron states, we may set
j
(z) 0
for z . For the electrostatic potential we suppose that d+,dz 0 for z
.
In conclusion, Eqs. (7.1), (7.4), and (7.5) together with the boundary conditions com-
plete the formulation of the self-consistent problem that describes the formation of a
two-dimensional electron gas at the interface as well as the electron quantization. By
integrating Eq. (7.1) over an innite range of z and using the boundary conditions, one
can nd a neutrality equation in the form
j
n
s. j

_

dz[N
A
(z) N
D
(z)] = 0. (7.8)
This result implies that, despite the charge transfer, the entire systemremains electrically
neutral.
E
g2
E
v
E
c
1
2
E
g1
E
g2
GaAs
AlGaAs
Vacuum level
(a)
(b)
1
E
g1
E
F
+
+
-
-
-
-
-
Figure 7.2 Schematic energy band diagrams of a selectively doped AlGaAs/GaAs
heterostructure before (a) and after (b) charge transfers have occurred. Relative positions of the
valence and conduction bands are given for both materials. The electron afnities are shown
conditionally (left). The Fermi level in the AlGaAs material is supposed to be pinned on the
donor level. The narrow-bandgap GaAs is slightly p-doped. After C. Weisbuch and B. Vinter,
Quantum Semiconductor Structures (San Diego, Academic Press, 1991).
Numerical analysis of a single heterojunction
Accurate analysis of the problemof electron-channel formation at a heterojunction can be
accomplished through numerical solution of the system of Eqs. (7.1)(7.5). Frequently,
to solve the Schr odinger equation one exploits the so-called variational method. Briey,
the method is based on the use of so-called trial wavefunctions containing a few well-
chosen parameters, which then are determined from the condition that the total energy
of the system should be a minimum.
If we assume that only one lowest subband with j = 1 is occupied by the electrons, a
simple trial wavefunction can be chosen in the form
1
(z) = 0 for z 0 and
1
(z) =
_
b
3
2
ze
bz,2
for z 0. (7.9)
which implies that the wavefunction is localized near the interface and is equal to zero
on the wide-bandgap side of the heterojunction.
Before discussing the numerical results, let us return briey to the picture of the
electron energy at the heterojunction in order to present both the conduction- and the
valence-band proles for specic heterojunctions. The left-hand side of Fig. 7.2 shows
the band edges of separated AlGaAs and GaAs; the distance to the vacuumlevel is shown
provisionally. It is assumed that AlGaAs is heavily n-doped and GaAs is doped lightly by
acceptors (p-doped). This doping brings about the pinning of the Fermi level at the donor
level on the AlGaAs side of the structure. On the right-hand side of the picture, one can
see the energy structure of the AlGaAs/GaAs junction with two depletion regions: one,
positively charged, on the AlGaAs side and another, negatively charged, on the GaAs
E
F
Al
0.3
Ga
0.7
As GaAs
1
d
sp
0
0.1
0.2
0.3
0 100 200 300 400 100 200 300
E
n
e
r
g
y

(
e
V
)
z ()
+
+
+
+
+
+
+
2DEG
Figure 7.3 Calculated self-consistent potentials, energy levels, and wavefunctions of an
Al
0.3
Ga
0.7
As/GaAs selectively doped heterostructure. The junction is situated at z = 0. The
spacer thickness is 50 and the donor binding energy of AlGaAs is chosen as E
D
= 50 meV.
After T. Ando, Self-consistent results for a GaAs,Al
x
Ga
1x
As heterojunction. I. Subband
structure and light-scattering spectra, J. Phys. Soc. Japan, 51, 3893 (1982).
side. For such a doping the quantumwell occurs only for electrons and does not occur for
holes, since the AlGaAs/GaAs heterojunction is of type I (according to the classication
of Chapter 4.)
In Fig. 7.3, the results of numerical solution of Eqs. (7.1)(7.5) are presented for
a selectively doped Al
0.3
Ga
0.7
As/GaAs heterojunction. Heavily n-doped Al
0.3
Ga
0.7
As
and lightly p-doped GaAs materials are separated by a thin undoped Al
0.3
Ga
0.7
As layer
with thickness d
sp
. Such an undoped layer is called the spacer layer or simply a spacer.
In calculations, the sheet electron concentration, n
s
, in the channel at the junction is
equal to the number of ionized donors per unit area, N
s
= 5 10
11
cm
2
. The sheet
concentration of ionized acceptors, N
depl
, is supposed to be much smaller: N
depl
= 5
10
10
cm
2
. The self-consistent potentials, energy levels, and wavefunctions are shown
for two different cases. For the rst case, shown by the dashed curve in Fig. 7.3, the
trial electron wavefunction is dened by Eqs. (7.9). The second case, shown by the solid
curve in Fig. 7.3, corresponds to a more sophisticated trial wavefunction, which can
penetrate under the barrier into the wide-bandgap part of the structure. In the region of
classically allowed motion, the wavefunctions for all of these approximations are similar
to each other. However, they are drastically different in the barriers. Generally, the trial
function of Eqs. (7.9) gives a reasonable value of the energy
1
, as well as the behavior of
the potential at large distances. However, the second trial function yields more accurate
results. In particular, it gives a slightly lower energy level. Results obtained for various
concentrations, n
s
, of electrons at the interface are summarized in Table 7.1.
In this table, the lowest energy level,
1
, the spatial scale of the electron connement,
z), and the probability of nding electrons in the barrier, P
b
, are presented for three
Table 7.1 Results of calculations of electron parameters for
two wavefunction approximations: I corresponds to the
function of Eqs. (7.9), and II to a more accurate trial function
n
s
, (10
11
cm
2
)
2 4 8
1
(meV) 38.6 56.6 86.1
I z) (

A) 116 99 82
P
b
(%) 0 0 0
1
(meV) 32.00 45.6 66.3
II z) (

A) 100 82.6 65.6
P
b
(%) 0.7 1.11 1.95
values of the electron concentration, n
s
. The value of z) was calculated as the quantum-
mechanical average of the z-coordinate of an electron, and P
b
=
_
0
dz[
1
(z)[
2
. One
can see that the relative height of the rst energy level,
1
, increases as the concentration
increases. The value of z) decreases with increasing n
s
as well as with the conning
electrostatic potential. Thus, we can see that the width of the electron channel is in the
range 60100

A. The probability of nding electrons in the wide-bandgap barrier is
small, but increases with increasing electron connement. Obviously, P
b
is equal to zero
for the wavefunction of Eqs. (7.9).
The examples given in our discussions illustrate the following major features of a
selectively doped heterojunction: (1) the formation of electron-conducting channels with
concentrations in the range n
s
= 10
11
10
12
cm
2
at any temperature, including T = 0 k;
(2) the spatial separation of the electrons from their parent donors with very low prob-
ability of electron penetration into the barrier less than or about 1% as well as
spatial isolation of the electrons from the p-doped narrow-bandgap material; (3) the
formation of a potential well for electrons with the potential prole self-consistently
dependent on the electron concentration; and (4) the quantization of electrons inside the
potential well with the resultant two-dimensional character of the electron spectrum and
with the electrons conned in the two-dimensional channel with a width of less than
100

A.
Control of charge transfer
We have considered a heterojunction formed by two semi-innite semiconductors with
xed concentrations of donors and acceptors as illustrated in Fig. 7.2. This results in
a conducting channel at the interface with a constant electron concentration that is
determined by the doping prole. The control of the conductivity, or more exactly, the
control of the resistance or its inverse, the conductance of the structure is necessary
in order to realize useful devices.
Let us consider the possibility of changing the conductance of a heterojunction
by controlling the electron concentration. For that purpose, we study the so-called
2DEG
n -AlGaAs
+
p-GaAs
Ungated heterostructure
(a)
Gated heterostructure
(b)
p-GaAs
Metal gate
n -AlGaAs
+
Figure 7.4 Schematic diagrams of (a) ungated and (b) gated heterostructures with
two-dimensional electron gas (2DEG).
1
e
b
l
d
d
sp
0
z
E
F
E
F
e
b
l
d
d
sp
0
z
E
F
E
F
1
(a) (b)
2DEG
+
+
+
+
+
+
+ +
+
+
+
+
Figure 7.5 Conduction-band diagrams for M/AlGaAs/GaAs heterostructures. The built-in

Schottky voltage controls the depletion region under the metallic gate. It results in (a) a
normally-off device for a narrow barrier and (b) a normally-on device for a wide barrier with
2DEG in the potential well (60 nm or wider).
gated heterojunction which is presented schematically in Fig. 7.4(b). For comparison,
Fig. 7.4(a) depicts the ungated heterojunction considered previously. The only difference
between them is the metal (M) contact placed on the top of the n
layer of the AlGaAs

barrier material in the gated structure. This metal semiconductor system (MES) is called
a MES structure. For GaAs-like materials, MES structures are of the most importance
for device applications, because these materials do not have a stable natural oxide, unlike
in the case of SiO
2
on silicon. Therefore, most of the electronic devices based on GaAs
use MES structures. One also refers to these structures as Schottky-gate structures.
Typically, under a metallic gate in GaAs-like materials there are extended depletion
regions, which occur because of a high built-in Schottky voltage, +
b
, of about 0.8 V.
Such a depletion region is known as a Schottky depletion region. The conduction-band
energy diagram for an M/AlGaAs/GaAs heterostructure is presented in Fig. 7.5 for two
thicknesses of the AlGaAs layer. The n-doped region is separated fromthe junction by an
undoped spacer. The structure is shown in Fig. 7.5(a) for a relatively thin AlGaAs layer,
and Fig. 7.5(b) depicts a structure with a thicker AlGaAs layer. For both cases, there
exists an extended depletion region, which affects the electron channel formed at the
AlGaAs/GaAs interface and provides two possibilities of controlling the structure. The
normally-off structure corresponds to Fig. 7.5(a). The depletion region extends through
both a thin AlGaAs layer and the junction. The bottom of the quantum well shifts up.
The Fermi level lies under the lowest energy subband. Thus, there are no electrons inside
the channel and the conductivity along the heterostructure is almost zero. The donors
in the AlGaAs doped region are ionized, and the electrons have left the semiconductor
part of the structure which is charged positively. In order to turn on the conductivity
of the device, it is necessary to apply a positive voltage to the metal gate. normally-off
structures can be fabricated by using a thin AlGaAs barrier.
The normally-on structure is illustrated in Fig. 7.5(b). In this case, the built-in voltage
drops across a thick AlGaAs layer so that the Fermi level lies above the lowest subband
and electrons populate the channel without an external voltage bias. This channel has
a nite conductivity under normal conditions. This case can be realized for sufciently
thick AlGaAs layers. Thus, in normally-on devices, one can control the conductance
of the channel by applying a negative voltage to the metal. A large voltage leads to a
depopulation of the channel and can switch the device off.
The calculated electron potential energy and quantized levels both for normally-off
and for normally-on devices are shown in Fig. 7.6 for various gate voltages, +
G
. The
Fermi level is taken to be at the zero energy. The values for the normally-off device were
calculated for an AlGaAs layer thickness of about 400

A; see Fig. 7.6(a). The quantum
well formed on the interface contains up to four quantized levels. A positive voltage
lowers the bottom of the conduction band of GaAs at the interface. The bottom touches
the Fermi level at +
G
= 0.3 V and the device is turned on at a threshold voltage of
about 0.8 Vwhen the rst quantized level touches the Fermi level. It is clearly seen that
in the AlGaAs barrier layer a potential minimum occurs and tends to be lowered with
increasing gate voltage. This phenomenon can result in a negative effect: the formation
of a second channel in this layer, which will collect electrons, screen the gate voltage,
and result in loss of control of the concentration of the two-dimensional electron gas at
the interface. The normally-on device is shown in Fig. 7.6(b). It has an AlGaAs layer
thickness greater than 600

A. The device can be switched off when a negative voltage of
about 0.5 V is applied to the gate.
The results just presented were obtained by numerical calculations. Let us consider
some experimental data related to the problem of modulation-doped heterostructures
where carrier concentrations and their mobilities have been measured simultaneously.
In Fig. 7.7 the sheet electron concentration controlled by the gate voltage is shown for
Al/AlGaAs/GaAs systems fabricated for high-electron-mobility transistors (HEMTs).
The curves correspond to various spacer thicknesses d
sp
. One can see that the electron
concentration can be varied over one order of magnitude. Saturation of the sheet concen-
tration at high positive voltage is caused by transitions of electrons to the potential well
which is formed in the middle depleted barrier region, as discussed previously. Figure 7.7
0 400 800 0
G
= 0.8 V
G
= 0.3 V
GaAs AlGaAs
G
= 0.5 V
G
= 1.5 V
GaAs AlGaAs
400 800
300
200
100
0
100
(a) (b)
Distance from gate ()
E
n
e
r
g
y

(
m
e
V
)
Figure 7.6 Calculated self-consistent potentials for conduction electrons in two M/AlGaAs/GaAs
heterostructures (a) corresponds to the normally-off device; (b) corresponds to the normally-on
device at room temperature. The Fermi level is at E = 0. Horizontal lines indicate the bottom
energy of the lowest four subbands. Dashed lines show donor levels. After B. Vinter, Subbands
and charge control in a two-dimensional electron gas eld-effect transistor, Appl. Phys. Lett.,
44, 307 (1984). Reused with permission from B. Vinter, Applied Physics Letters, 44, 307
(1984).
C _
shows that considerable changes in the properties of the structures occur with changes
in spacer thickness. Among the various structures presented in Fig. 7.7, the particular
structure with d
sp
= 0 is most likely to be useful for fabricating a normally-off device.
Indeed, by applying a positive voltage to the gate one can increase the electron concen-
tration in the channel. Structures with a thick spacer are well suited for normally-on
devices; in these structures, positive voltage does not change the concentration, whereas
a negative voltage reduces it sharply.
The spacer is an important element of modulation-doped structures because it partially
prevents the scattering of channel electrons by the heavily doped side of the heterostruc-
ture and increases the electron mobility. On the other hand, there is one negative effect
of a spacer. Increasing the spacer thickness leads to an increase of the potential drop
on the spacer and consequently to a lowering of the electrostatic potential that connes
electrons near the interface. Hence, a thick spacer causes a decrease in the electron con-
centration. This trade-off between the mobility and the carrier concentration requires an
optimization of the structural design for each particular device application.
n
depl
10
cm
2
Al
0.3
Ga
0.7
As/GaAs
T = 12 K
11
10
8
6
4
2
0
1 2 0 1
S
u
r
f
a
c
e

c
a
r
r
i
e
r

c
o
n
c
e
n
t
r
a
t
i
o
n
,

n
s
(
c
m
2
)
Gate voltage,
G
(V)
d
sp
=
0
45
75
180
Figure 7.7 Measured gate-voltage dependences of the channel density of two-dimensional
electrons in Al
0.3
Ga
0.7
As/GaAs structures at T = 12 K and various spacer thicknesses d
sp
.
All samples are Si-doped with N
D
= 4.6 10
17
cm
3
, except for the sample with 180- spacer,
which has N
D
= 9.2 10
17
cm
3
. The thickness of the doped Al
0.3
Ga
0.7
As layer is
7001200 . From K. Hirakawa, H. Sakaki, and J. Yoshino, Concentration of electrons in
selectively doped GaAlAs/GaAs heterojunction and its dependence on spacer-layer thickness
and gate electric eld, Appl. Phys. Lett., 45, 253 (1984). Reprinted with permission from K.
Hirakawa, H. Sakaki, and J. Yoshino, Applied Physics Letters, 45, 253 (1984).
C _
1984 American
Institute of Physics.
Here, we have considered systems with a single heterojunction. These systems can
be fabricated by a relatively simple technology and they have numerous applications,
but they suffer from greatly limited carrier concentration in the conduction channel. As
can be seen from Fig. 7.7, the typical surface concentrations are less than 10
12
cm
2
for
single-heterojunction devices. Higher concentrations of carriers in the conducting chan-
nel can be obtained in a double-junction system. This system is illustrated by Fig. 7.8,
where possible double-junction heterostructures are presented schematically. For het-
erostructures of type I, quantum wells are formed both for electrons (two quantized
levels are shown) and for holes (one quantized level is shown), whereas for a type-II
heterostructure a quantum well arises only for the electrons.
7.3 Electrons in quantum wires
A quantum wire is a conductive structure, wherein electron transport is constrained
primarily to be along a single direction. Let this direction be along x. For the two other
directions the y and z directions the quantum-mechanical connement of the electrons
is imposed by the heterointerface potentials, or by suitable externally applied electrostatic
potentials. Such a system is also called a one-dimensional electron system. Electron
7.3 Electrons in quantum wires 229
V
b,h
V
b,e
1,e
2,e
z
E
0 z
1,e
2,e
1,h
E
0 L
2
L
2
Valence
band
Conduction
band
L
2
L
2
(a) (b)
Figure 7.8 Two types of band diagrams of double heterostructures: (a) type I and (b) type II.
A
A
B
A
A
G G
(a) (b)
B
Figure 7.9 Quantum wire formed by etching (a) and split-gating (b) of the two-dimensional
electron gas.
motion along the free x direction is characterized by a one-dimensional wavevector k
x
.
The wavefunction has the form
+(x. y. z) = e
ik
x
x
n
1
.n
2
(y. z).
where
n
1
.n
2
(y. z) corresponds to conned transverse motion of the electron. The bound
states corresponding to transverse motion are enumerated by integers n
1
and n
2
and
usually are called one-dimensional subbands. In Section 3.3, we considered electron
connement in two directions and found subband energies of the form of Eq. (3.50).
If distances between the lowest subband and excited subbands are small in comparison
with both the thermal energy, k
B
T, and the Fermi energy of electrons, E
F
, scattering of
electrons from the lowest subband into higher subbands is relatively weak. The electrons
occupy primarily the lowest subband and behave almost as one-dimensional particles.
Currently, there are several methods for the fabrication of quantumwires, including the
direct growth of the wires. The most obvious method is to start with a two-dimensional
structure and to impose an additional connement to two-dimensional electron gas, as
shown schematically in Fig. 7.9. In the case of Fig. 7.9(a), the etching process is used to
realize geometrical restrictions for electron motion. In the case of Fig. 7.9(b), additional
connement is induced by an electrostatic potential that is applied to a metallic split gate
placed on the top of the heterostructure.
T = 4.2 K
wire length
g
(V)
0.8 0.9 1.0 1.1 1.2
G
(

2
e
2
/
h
)
0
2
4
6
8
10
12
1
(m)
0.4
0.2
2
5
7
10
Figure 7.10 The dependence of the conductance, G, of a quantum wire versus gate voltage, +
g
,
for seven different lengths of wires as indicated near each curve. After L. Worschech, F.
Beuscher, and A. Forchel, Quantized conductance in up to 20 m long shallow etched
GaAs/AlGaAs quantum wires, Appl. Phys. Lett., 75, 578 (1999). Reused with permission from
L. Worschech, F. Beuscher, and A. Forchel, Applied Physics Letters, 75, 578 (1999).
C _
1999
American Institute of Physics.
Electron transport in quantum wires
The principal distinct features of electron transport in quantum wires are related to
Landauer quantization of the conductance. This quantization can be observable for per-
fect wire structures at low temperatures. As an example, we discuss measurements of
the conductance for AlGaAs/GaAs quantum wires obtained by the etching method, as in
Fig. 7.9(a). The quantumwires were fabricated on modulation-doped AlGaAs/GaAs het-
erostructures grown by molecular-beam epitaxy. High-resolution electron-beam lithog-
raphy was used to dene the masks of the quantum wires. With the help of wet chemical
etching, the doped layer and the spacer both fabricated with AlGaAs as well as 70 nm
of GaAs were removed. From scanning electron microscopy, the geometrical width of
the quantumwires was determined to be 135 nm. The micrographs indicated a very small
sidewall roughness of the wires. Before etching, the electrons at the selectively doped
AlGaAs/GaAs heterointeface were characterized by a density of about 3 10
11
cm
2
and high mobility in the range (12)10
6
cm
2
V
1
s
1
. At low temperature (4 K), the
electron gas is degenerate and the mean free path of electrons having the Fermi energy is
estimated to be in the range 1020 m. Thus, for wires fabricated with lengths less than
10 m, we can expect Landauer quantization. To control one-dimensional subbands in
the wires, an aluminum top gate was evaporated onto the top of the structure. A voltage
applied to this gate changed the conning potential and subband energies. The results
are presented in Fig. 7.10 for various wire lengths. The conductance quantization is seen
clearly for the wires shorter than 5 m. Up to ten steps in the conductance are seen;
i.e., about ten one-dimensional subbands can be observed by changing the gate voltage.
Experiments with magnetic eld and temperature variation also reveal the quantization
effect and allow one to determine the concentration of one-dimensional electrons. The
typical concentration and subband spacing were found to be n
1D
6 10
6
cm
1
and
L
1.2
12.5 meV, respectively. The relatively large subband spacing equivalent to
140 K implies that the one-dimensional character of electron motion should be
unchanged even at temperatures of several tens of degrees Kelvin.
For higher temperatures, at whichelectronmotionbecomes semiclassical, the transport
properties of quantumwires based on semiconductor heterostructures are similar to those
of quantum wells and pure bulk-like materials. Thus, selectively doped quantum wires
should exhibit high electron mobilities under low electric elds and high drift velocities
characteristic of hot electrons under high elds.
The situation with electron transport in carbon nanotubes is quite different. In high
elds the electron drift velocities can reach magnitudes in the range (24) 10
7
cms
1
,
which are larger than those in perfect IIIV semiconductor compounds. However, the
electrical properties are affected signicantly by the surrounding environment, which
inuences the removal of heat fromcurrent-carrying nanotubes. For example, suspended
nanotubes display drastically different electron transport fromthat of those on substrates.
Figure 7.11 illustrates this difference. In the upper panel of this gure, scanning elec-
tron microscope images of two single-walled carbon nanotubes with Pt contacts are
shown. The left nanotube segment is not suspended and is in contact with a nitride-based
substrate. The right nanotube segment is suspended over a 0.5-m-deep trench. The
diameters of the nanotubes were measured by the atomic-force microscope to be in the
range 23 nm. In the lower panel, the results of measurements of the currentvoltage
characteristics for the 3-m length of suspended and non-suspended segments of the
nanotube are given. The measurements were conducted at room temperature in vacuum.
One can see that the non-suspended nanotube portion displays a monotonic increase
in the current, approaching 20 A under increasing voltage, V, while the current in
the suspended tube reaches a peak of 5 A followed by a pronounced current drop.
This strikingly different behavior is due to signicant self-heating effects of the wires
carrying ultra-high current densities. Indeed, in the case of the suspended tube the Joule
heating can not be removed effectively, because thermal uxes are possible only through
nanotube contacts. As a result, the temperature of the tube increases, especially in its
central region.
7.4 Electrons in quantum dots
A quantum dot can be dened as a material system in which electrons are conned in
all three directions. Some particular examples of such systems were described in previ-
ous chapters. They include the self-forming nanosized semiconductor islands, clusters,
and nanocrystals studied in Chapter 5, etc. Quantum dots can be fabricated by various
methods, including a direct way, starting from a structure with two-dimensional elec-
trons and using an etching process to impose geometrical restriction on the electrons
in two additional directions. A small metallic gate fabricated upon a two-dimensional
heterostructure also can provide the necessary electron connement under a negative
(a)
(b)
V (V)
I

(
A
)
Figure 7.11 (a) Carbon nanotubes on a substrate and suspended over a trench; (b) their
currentvoltage characteristics. Reprinted with permission from E. Pop, D. Mann et al.,
Negative differential conductance and hot phonons in suspended nanotube molecular wires,
Phys. Rev. Lett., 95, 155505 (2005).
C _
voltage applied to the gate. In Fig. 7.12, three types of quantum dots are shown: a
nanosized island self-forming under a special growth regime, colloidal nanocrystals, and
a quantum box fabricated in a controlled way from a two-dimensional heterostructure.
An example of a quantum dot formed by using the technique of fabricating a gate upon
a two-dimensional heterostructure will be discussed in the next chapter; see Section 8.4.
The most notable feature of quantum dots is that all existing degrees of freedom of
electron propagation are quantized, if a conning potential is deep and the dimensions
of the structure are comparable to the de Broglie wavelength of the electron, as dened
in Eq. (6.5). According to the analysis given in Section 6.2, such a system can be called
a zero-dimensional system. The latter terminology stresses the dramatic changes of the
electronic properties in quantum dots. Indeed, in a crystalline solid, as was discussed in
Chapter 4, electrons possess continuous energy bands with the occupation, width, and
8 nm
(b)
Molten Silicate
CdS
(c)
AlGaAs
GaAs
AlGaAs
(a)
GaAs
InAs
As
2
In
AlGaAs
Figure 7.12 Schematic representations of three different approaches for the fabrication of
quantum dots: (a) self-organized growth of nanosized islands, (b) nanocrystals in a colloid, and
(c) articial patterning and etching of a heterostructure with two-dimensional electron gas.
separation of these bands determining the fundamental electrical and optical properties
of the solid. At the other end of the length scale, for individual atoms the electronic state
density is discrete, resulting in, for example, the absence of simple electron transport
and intrinsically sharp spectral optical lines. In some respects, the electronic structure
of quantum dots might be said to fall somewhere between these two extremes. Thus,
quantum-dot structures are like large articial atoms, macroatoms.
For the analysis of the electronic properties of a quantum dot, a model of a conning
potential may be used. In Section 3.3, we considered two examples of these models:
the quantum-box model given by Eq. (3.51) and the spherical-dot model of Eq. (3.53).
The results obtained with these models for the energy spectra and wavefunctions can be
applied to semiconductor quantumdots, if the masses are taken to be the effective masses
of the material being analyzed. The height of the conning potential is determined by
the band offsets, see Fig. 4.12.
Consider semiconductor quantumdots fabricated froma two-dimensional heterostruc-
ture by using an etching process, or by colloidal synthesis, as illustrated in Figs. 7.12(b)
and 7.12(c). In these cases, depending on the heterostructure type, it is possible to realize
different types of carrier connement. The classication of heterostructure types was
presented in Section 4.5. If the initial heterostructure is of type I, both electrons and
holes can be conned in the same quantum-dot structure. If the initial heterostructure is
of type II, only one type of carriers, either electrons or holes, can be conned.
If electron motion is quantized in all three possible directions, we obtain a newphysical
object, a macroatom. Questions concerningthe usefulness of suchobjects for applications
naturally arise from the point of view of their electronic applications. A fundamental
question is the following: what is the current through a macroatom? A valid answer is
that there exists the possibility of passing an electric current through an articial atom
due to tunneling of electrons through quantum levels of the macroatom.
Importantly, that transport of charge occurs always in quanta of the elementary charge
of the electron. This discreteness has no practical consequences for current owing in
bulk materials, in two-dimensional structures, and even in quantum wires, because the
numbers of electrons transported are large. Charge transport in quantum-dot systems
is completely different. Indeed, it occurs as electron tunneling from a cathode through
the quantum dot to an anode. That is, the clearance space between the cathode and the
anode should be considered as a potential barrier in which a quantum dot is embedded.
In Section 3.3, while studying the tunneling effect, we found that the probability of
tunneling drastically depends on the thickness and height of the barrier: for large enough
thickness and barrier height, tunneling is essentially suppressed. However, tunneling
through a discrete state localized inside the barrier has much higher probability. The
whole tunneling process can be thought of as a sequential process: the capture of an
electron from the cathode to the localized state and then its emission to the anode. As a
result, the easiest way for the electrons to be transferred from the cathode to the anode
is for them to tunnel through the quantum dot. Then, an electron being captured to
the quantum dot blocks tunneling of other electrons. Electron transfer occurs in highly
correlated manner, one by one, at least at low temperatures. Examples of such single-
electron transport were discussed in Section 6.5. Thus, the electric current through a
quantum dot occurs in the regime of single-electron transport. Devices that use such
transport will be discussed in the next chapter.
In fact, electrical methods applied to quantum dots to realize useful devices are not
the only methods possible. The control of the electric current through the dots can also
be realized by means of light, sound waves, etc. Consider here optical control of the
dots and optoelectronic functions of zero-dimensional devices. The main peculiarities
of the optical properties of quantum dots arise due to electron and hole quantization.
In quantum dots fabricated using a type-I heterostructure, the carrier energies have the
form
E
QD
e
= E
QD
g

n
(n
1
. n
2
. n
3
). E
QD
h
=
p
(n
/
1
. n
/
2
. n
/
3
).
Here, E
QD
g
is the fundamental bandgap of the material of the quantum dots. For a type-I
heterostructure, E
QD
g
is less than the bandgap E
g
of the surrounding material into which
the dots are embedded;
n
and
p
depend on sets of three discrete quantum numbers,
{n
1
. n
2
. n
3
] and {n
/
1
. n
/
2
. n
/
3
], for the electrons and holes, respectively. Specic depen-
dences of E
QD
e
and E
QD
h
on the quantum numbers are determined by the materials used
for quantum dots and their environment, the geometry of dots, etc. The model depen-
dences of Eqs. (3.51) and (3.53) can be used for estimation of the energy levels. Owing
to these discrete energy spectra, quantumdots interact primarily with photons of discrete
energies:
h
--

2h
--
c
= E
QD
g

n
(n
1
. n
2
. n
3
)
p
(n
/
1
. n
/
2
. n
/
3
). (7.10)
20 nm
QD
W
D
GaAs
InGaAs
GaAs
X
R
100 nm
1.15 1.20 1.25 Energy (eV)
E
c1
QW
E
hh1
Figure 7.13 A sketch of quantum dot conning potentials for electrons and holes. Peaks in optical
spectra correspond to phototransitions between quantized states of the electrons and holes in the
quantum dot. Reprinted with permission from J. Tulkki and A. Heinamaki, Connement effect
in a quantum well dot induced by an InP stressor, Phys. Rev. B, 52 (11), 8239 (1995).
C _
1995
by the American Physical Society.
where c is the velocity of light and is the wavelength of the light. The different combi-
nations of quantum numbers {n
1
. n
2
. n
3
] and {n
/
1
. n
/
2
. n
/
3
] give a series of optical spectral
lines, for which interaction between the dots and light is efcient. Importantly, the fact
that E
QD
g
- E
g
implies that the light interacting with the dots is not absorbed by the
surrounding material. These considerations are illustrated in Fig. 7.13. The potential
proles of the conduction band, E
c
, and the upper (heavy-hole) valence band, E
hh
, are
depicted. The potential wells for electrons and holes represent the connement poten-
tials of the quantum dot. Quantized levels for electrons and holes are shown; arrows
indicate the possible phototransitions between different quantized states. For each such
transition, a characteristic line of the optical spectrum arises. For the example of an
InGaAs quantum dot grown inside a GaAs quantum well, the quantum-dot spectral lines
arise in the region of photon energies from 1.1 to 1.3 eV, as presented in Fig. 7.13. The
width of observed spectral lines depends primarily on two factors. First, the spectral
lines are broadened because of relaxation processes in the system. Second, it is typi-
cal that in experiments light interacts with numerous dots and there is some dispersion
of their sizes. This produces a dispersion in energy-level positions and additional line
broadening.
The optical control of the electric current owing through a quantum dot can be
explained with the help of a device that can be called a single-quantum-dot photodiode.
Depth
InGaAs
QD
i-GaAs
V
b
E
n
e
r
g
y
E
F
hv
n
+
-GaAs
E
F
Metal
Figure 7.14 A schematic energy band diagram of a single-quantum-dot photodiode for
photocurrent experiments, with an n
-contact on the left and a metal contact (Schottky contact)

on the right. After Figs. 1 and 13(b), H. Kenner, S. Suer et al., Recent advances in
exciton-based quantum information processing in quantum dot nanostructures, New J. Phys., 7,
184 (2005).
C _
The diode base is an i region (undoped, i.e., intrinsic region), into which a single quantum
dot is embedded. A schematic band diagram of such a photodiode under equilibrium,
when the Fermi level, E
F
, is the same throughout the whole structure, is shown in
Fig. 7.14. At the semiconductor/metal heterojunction a potential barrier V
b
(the Schottky
barrier) arises. An electric bias is to be applied to two terminals; in the case under
considerationtheyare the n
region(onthe left) anda metal contact (the Schottkycontact)

on the right. Under inverse biasing, an electric current is possible, in principle, due to the
interband tunneling mechanism. The probability of such tunneling is extremely small and
the current is negligible even if a quantumdot is embedded in the base. Illumination of the
diode by light resonant with phototransitions between discrete electron and hole levels in
the quantum dot leads to the excitation of an electronhole pair inside the dot. Now, the
electron and hole can tunnel more readily from the dot into the i region and contribute
to the electric current, the photocurrent. In Fig. 7.15, we present the experimental
results obtained fromexcitation of the ground state (n
1
= n
2
= n
3
= n
/
1
= n
/
2
= n
/
3
= 1)
of a single self-assembled In
0.5
Ga
0.5
As quantum dot embedded into a 360-nm-thick
intrinsic GaAs layer. Since InGaAs/GaAs heterostructures are of type I, the only optically
active part is the single In
0.5
Ga
0.5
As quantum dot. The experiments were carried out
at 4.2 K. In Fig. 7.15, the photocurrent is plotted as a function of the electric bias
for various wavelengths of illuminating light. The wavelengths are indicated on the
photocurrent curves. One can see that photocurrentvoltage dependences have, in fact, a
pronounced resonant character. This is explained by the fact that the quantized electron
and hole energies are shifted under an applied electric eld, as expected from the so-
called Stark effect observable for atoms and molecules. When these energies are such that
the wavelength given by Eq. (7.10) corresponds to the illuminating light, the light excites
T = 4.2 K
laser
= 927.35 nm
927.62 nm
928.03 nm
928.45 nm
927.1 nm
0.4 0.6 0.8 1.0 1.2 1.4
0
5
10
15
20
25
30
P
h
o
t
o
c
u
r
r
e
n
t

(
p
A
)
Bias voltage (V)
Figure 7.15 Photocurrent resonances for several different excitation wavelengths versus bias
voltage. Reprinted with permission, from H. Kenner, S. Suer et al., Recent advances in
exciton-based quantum information processing in quantum dot nanostructures, New J. Phys., 7,
184 (2005).
C _
electrons and holes inside the quantumwell, which produces the measured photocurrent.
As the applied bias increases, the energies are shifted to smaller values and the resonance
wavelength increases. In Fig. 7.15, spectra for excitation of the same ground state of the
dot for different biases are shown. The observed photocurrent spectra are very narrow
because a single dot is involved. Spectral broadening becomes visible at high biases when
the electron and hole energy levels decay as a result of the increased rate of tunneling
from the dot.
A photocurrent excited resonantly with an electrically tunable optical resonance can
nd numerous applications. Specically, such a quantum-dot photodiode facilitates opti-
cal manipulation of individual quantum states tunable with an electric bias. Such a
capability is necessary for devices required as components of systems used in quantum
information technology.
7.5 Closing remarks
In this chapter, we focussed on traditional quantum structures that are already being
exploited in nanoelectronics and optoelectronics. These include quantum wells, wires,
and dots. These structures provide electron connement in one, two, and three dimen-
sions, respectively. We found that connement of the electrons in any direction forbidding
free propagation in this direction leads to quantization of the electron energy.
We showed that the connement effect can be very strong. In particular, for one of the
most used and practical cases when electrons are conned at a heterointerface we
found that the electrons are localized within a thin spatial layer of thickness ranging from
5 nm to 10 nm depending on the electron concentration. This connement results in high
electron densities. For example, a surface concentration of conned two-dimensional
electrons of about 10
12
cm
2
corresponds to a bulk concentration of from 10
18
cm
3
to
2 10
18
cm
3
, which are typical values for heavily doped semiconductor bulk materials.
In quantum wells and wires, the carriers can be spatially separated from layers with
dopants, whichgenerate free carriers. Suchtechniques leadtoquantumstructures exhibit-
ing large electron mobility relative to the case of bulk structures. Together with high and
controllable electron density, this enhanced mobility results in high electric current den-
sities and high speeds of operation.
Quantumdots possess totally quantized spectra for electrons and holes. Electron trans-
port via the dots can occur in the tunneling regime. Accordingly, the optical properties of
the dots differ considerably fromthose of bulk samples, as well as fromthose of quantum
wells and wires. Quantum-dot spectra consist of a series of separate lines. The positions
of these spectral lines can be controlled by dot dimensions and geometry. Quantum dots
nd applications in optoelectronics.
For those who want to study traditional quantum structures in detail, we recommend
the following additional reading.
Results on electron quantization in SiO
2
/Si structures are presented in the review
T. Ando, A. B. Fowler, and F. Stern, Electronic properties of two-dimensional sys-
tems, Rev. Mod. Phys., 54, 437 (1982).
Discussions of particular examples of the quantum wells and wires based on IIIV
compounds can be found in the following references:
G. Bastard, Wave Mechanics Applied to Semiconductor Heterostructures (New York,
Halsted Press, 1988).
C. Weisbuch and B. Vinter, QuantumSemiconductor Structures (NewYork, Academic
Press, 1991).
Fabrication methods and properties of quantum dots are analyzed in the book
D. Bimberg, M. Grundmann, and N. N. Ledentsov, Quantum Dot Heterostructures
(Chichester, John Wiley & Sons, 1999).
7.6 Problems
1. Consider a quantum-well structure with the electron energy given by Eq. (6.27).
Assume that for the lowest two-dimensional subbands the intersubband distance L
21
=
1
is given. For T 0, nd the formula for the critical concentration n
c
of two-
dimensional electrons at which the second subband starts to be populated. Using
1
,
2
,
and L
21
for an innitely deep quantum well, estimate the critical concentration, n
c
, for
a quantum well of width L = 10 nm.
7.6 Problems 239
2. To obtain characteristics of the electrons conned at an interface, one can use the
electron wavefunction of the lowest quantized state in the form of Eq. (7.9), which
depends solely on the parameter b. Apply the denition of the expectation value of a
physical quantity presented in Eq. (3.12) and nd an expression for the average distance
of the electrons fromthe interface z). Calculate the mean square deviation of the location
of electrons from its average position; i.e., calculate (z z))
2
).
3. Near the semiconductor heterojunction the electrons can be conned in the direction
normal to the heterojunction. In Figs. 7.1 and 7.2, sketches of conning potentials are
presented. If the potential has high barriers, the lower energy levels may be studied with
some accuracy by applying the following triangular approximation:
V(x) =
_
eFx. for x > 0.
. for x 0.
Here, x = 0 corresponds to the heterojunction position, and F can be interpreted as
the conning electrostatic eld. For this triangular model of the quantum well, the
Schr odinger equation for the transverse component of the electronic wavefunction can
be solved exactly. The quantized electron energies are
n
=
_
e
2
h
--2
F
2
2m
_
1,3
p
n
.
where p
n
are parameters dened by some algebraic equation. The lowest parameters are
known to be p
1
2.35 and p
2
4.09. Using the effective mass of GaAs, m
=0.067m
0
,
evaluate the positions of the lowest energy levels for the AlGaAs/GaAs heterojunction
as a function of the eld F.
The electrostatic eld F can be related easily to the electron concentration n
s
inside
the triangular well. Indeed, ionized (positive) impurities in the wide-bandgap part of the
heterojunction (x - 0) determine the conning eld F. The number of these impurities
approximately equals the number of the electrons. Thus, in the region x > 0, Gauss
law gives the eld as F = 4en
s
,c, where c is the dielectric constant. Assuming a
concentration of n
s
= 10
12
cm
2
and taking c = 13, estimate the eld F and the energies
1
and
2
. Compare the separation between the rst two levels with the thermal energy
k
B
T at room temperature. Discuss the population of these levels by the electrons at T =
300 K.
4. In Section 4.4 on analyzing different types of crystals, we found that in silicon
there are six energy-equivalent minima (valleys). The electrons are mainly accumulated
near these minima. The minima are at the following crystalline directions: [100], [
100],
[010], [0
10], [001], and [00
1] (see Table 4.4). Within these minima, the electron energy

spectrum is given by Eq. (4.17) with effective mass tensors of the type of Eq. (4.19).
Let us consider a Si quantum-well layer oriented perpendicular to the [001] direction.
For free electrons collected in the pair of valleys along directions [001] and [00
1], the
electron Hamiltonian (kinetic energy) is
h
--2
2m
t
_

2
x
2

2
y
2
_
h
--2
2m
l
2
z
2
.
while for the other two pairs of valleys the Hamiltonians are
h
--2
2m
l
2
x
2

h
--2
2m
t
2
y
2

h
--2
2m
t
2
z
2
.
and
h
--2
2m
t
2
x
2

h
--2
2m
l
2
y
2

h
--2
2m
t
2
z
2
.
These Hamiltonians imply that the electron motion along the quantum-well layer and
that in the perpendicular direction are completely independent. The wavefunctions and
quantized energies should have the form of Eq. (7.2).
Assume that the potential barriers of the quantum well are innitely high. Employing
the results of Eqs. (3.27) and (3.28) for an innitely deep potential, nd energy subbands
for the different electronvalleys. Discuss the energysplittingof the valleys inthe quantum
well. Using the data for m
l
and m
t
presented in Table 4.4, determine which pair of valleys
will be of the lowest energy. In bulk Si crystals, the numbers of electrons in different
equivalent minima are equal. Can we expect an electron redistribution to be caused
by quantization in the well? Which valleys will be overpopulated and which will be
depleted?
5. Ananostructure consists of two semi-innite two-dimensional electron gases con-
nected through a quantumwire. The entire structure is fabricated froma single innite
quantum-well layer. Does a potential barrier arise for electron motion from one two-
dimensional gas to the another? Explain why the barrier arises.
Using the adiabatic approach described by Eq. (6.124), discuss the barrier heights for
the electrons from different two-dimensional subbands.
Assume that the thickness of the quantum-well layer fabricated fromGaAs is 5 nmand
that the quantum-wire width is 5 nm. Estimate the height of this barrier for the electrons
populating the lowest two-dimensional subband.
6. In IIIV compound semiconductors the energies of heavy and light holes coincide
at zero wavevector, as shown in Fig. 4.10. Energy levels exhibiting such a coincidence in
energy are known as degenerate energy levels. Quantization of the two types of holes lifts
this degeneracy. Using models with innite-potential walls for quantumwells, wires, and
dots, calculate the splitting between the lowest states of heavy and light holes in these
low-dimensional structures.
Obtain numerical estimates of this splitting for GaAs structures with the following
geometrical parameters: a quantum well of thickness 5 nm, a quantum wire of cross-
section 5 nm 5 nm, and a quantum dot of volume 5 nm 5 nm 5 nm. Parameters of
heavy and light holes are given in Table 4.5.
7.6 Problems 241
7. It is known that GaAs/AlGaAs heterostructures are of type I. Thus, both electrons
and holes are conned in low-dimensional structures fabricated from these materials.
Consider GaAs quantum dots embedded in an AlGaAs matrix. Use the parameters of
electrons and of heavy and light holes presented in Table 4.5. Explain what kind of holes
forms the hole ground state in a GaAs quantum dot.
To describe the quantum dots, apply the model of a potential box of Eq. (3.51). Using
Eq. (7.10), calculate the positions of spectral lines for quantum dots with the following
dimensions: (a) L
x
= L
y
= L
z
=5 nm and (b) L
x
=3 nm, L
y
=5 nm, and L
z
=7 nm.
The bandgap of GaAs is given in Table 4.5.
8 Nanostructure devices
8.1 Introduction
In previous chapters, fundamental physical processes on the nanoscale, analysis of nano-
materials, and nanofabrication methods were all discussed extensively. The knowledge
gained in these previous discussions makes it possible to consider and analyze a vari-
ety of different nanostructure devices. In this chapter, we consider electronic, optical,
and electromechanical devices. Some of these devices mimic well-known microelec-
tronic devices but with small dimensional scales. This approach facilitates applications
to devices with shorter response times and higher operational frequencies that operate
at lower working currents, dissipate less power, and exhibit other useful properties and
enhanced characteristics. Such examples include the eld-effect transistors and bipolar
transistors considered in Sections 8.3 and 8.5.
On the other hand, newgenerations of the devices are based on newphysical principles,
which can not be realized in microscale devices. Among these novel devices are the
resonant-tunneling devices analyzed in Section 8.2, the hot-electron (ballistic) transistors
of Section 8.5, single-electron-transfer devices (Section 8.4), nanoelectromechanical
devices (Section 8.7), and quantum-dot cellular automata (Section 8.8).
As a whole, the ideas presented in this chapter provide an understanding of the future
development of nanoelectronic and optoelectronic devices that may be realized through
the wide use of nanotechnology.
8.2 Resonant-tunneling diodes
Diodes or, in other words, two-terminal electrical devices, are the simplest active ele-
ments of electronic circuits. Some applications of diodes are based on their nonlinear
currentvoltage characteristics. Another important capability required of diodes is their
operational speed. Such high-speed operation implies that the feature sizes of diodes
should be as small as possible.
In previous sections, we studied two types of short n
in
diode, both with space-

charge-limited classical electron transport. These diodes used homostructures, i.e., it was
supposed that they are made from the same material with nonuniform doping. Among
the cases studied, the highest speed is achieved for the ballistic diodes. According to
the classication given in Section 6.2 for the extreme scaling down of the diode size,
V
b
L
w
V
b
n
+
-GaAs
n
+
-GaAs
GaAs
Al
x
Ga
1 x
As
x
As
x
As
x
As
n
+
-GaAs
Al
x
Ga
1
L
b
n
+
-GaAs
GaAs
Al
x
Ga
1
Al
x
Ga
1
L
b
L
w
GaAs
GaAs
(a) (b)
Figure 8.1 Layer designs for double-barrier resonant-tunneling structures. (a) Alternating layers:
n
-doped GaAs (substrate and one of the contacts), undoped AlGaAs (barrier), undoped GaAs
(quantum well), undoped AlGaAs (barrier), n
-doped GaAs (top contact). (b) The same as (a)

except for two additional undoped spacer layers of GaAs between the contacts and the barriers.
V
b
is the barrier height, L
b
the barrier thickness, and L
w
the quantum well width.
quantum transport must be used to describe the carriers. A very important example of
such nanoscale diodes is the so-called double-barrier resonant-tunneling diode. In this
section, we consider this type of nanoscale quantum diode.
The physics underlying the resonant-tunneling effect
We start by considering a double-barrier heterostructure as an example of resonant-
tunneling diodes. Figure 8.1 depicts a sequence of layers in such a structure. The top
and bottom parts of the structure are doped regions, while the barriers and well layers
are undoped. Figure 8.1(a) shows a specic structure in which a quantum-well layer of
GaAs is embedded between two Al
x
Ga
1x
As barrier layers. The top and bottom regions
of doped GaAs serve as contacts. Aslightly different design is shown in Fig. 8.1(b), where
two additional spacer layers separate the doped regions and the double-barrier part of
the structure. The purpose of these spacer layers is to prevent scattering of tunneling
electrons by impurities in the contact regions. The thicknesses of the well, barriers, and
spacers may be varied substantially. Inside the quantum well, several quantized levels
can exist. In fact, these levels are quasi-bound states, because there is a small but nite
probability of the electron tunneling out of the well. Quantum-mechanical tunneling is
responsible for the nite lifetimes of the electrons in those levels and leads to some
broadening of the quantum-well states.
Thus, a resonant-tunneling diode can be thought of as a system with two contacts,
with three-dimensional electron states, and a quantum well, with a two-dimensional
electron system. These three subsystems are weakly coupled through tunneling. Energy-
band diagrams of the structure are presented in Fig. 8.2 for three different voltage biases.
Figure 8.2(a) corresponds to the equilibriumcase, when no voltage is applied. In the well
under consideration, there exists at least one quasi-boundlevel; the case of a single level of
energy,
1
, is depicted in Fig. 8.2. Actually,
1
is the bottomof the lowest two-dimensional
E
c
col
E
F
E
c
em
1 E
F
e
0
(b)
1
E
c
em
E
c
col
E
F
e
0
E
F
(c)
(a)
E
c
E
F
E
c
1
E
F
d
z
Figure 8.2 Energy band diagrams of double-barrier resonant-tunneling structures:
(a) equilibrium conditions; (b) in-resonance; and (c) off-resonance. The electron potential
energy and populations of the energy band in contact n
-regions are sketched in the left and

right parts of the gure. E
em
c
and E
col
c
denote the positions of the bottoms of the conduction
bands under electric bias.
subband because there exists in-plane free electron motion. The parameters of the diode
are chosen to be such that, in the non-biased state of the diode, the quasi-bound level,
1
, lies above the Fermi energy, E
F
, in the contacts, as in Fig. 8.2(a). By applying a
voltage bias to the contacts, one can produce a downward shift of the level in the well.
For electrons with arbitrary energies, the probability of tunneling through the double-
barrier structure is very small. The structure is designed to prevent the thermal transfer
of electrons over the barriers. Therefore, the only situation favorable for transmission
of electrons through the structure is when the quasi-bound level lies below the Fermi
energy, E
F
, but above the conduction-band bottoms of the contacts. In this case, those
electrons from the emitter (the contact on the left) whose kinetic energy of the in-plane
(perpendicular) motion, E
= h
--2
k
2
z
,(2m
), coincides with
1
are transmitted through
the structure with nite probability. This is the so-called resonant-tunneling process,
which has the important attribute of exhibiting negative differential resistance.
Before analyzing this effect, we recall that the electrical properties of a simple con-
ductor with a linear currentvoltage characteristic are characterized by a resistance, R,
according to the formula
I = +
0
,R.
where +
0
is the applied voltage and I is the current. If a conductor has a nonlinear
currentvoltage characteristic, one can introduce the so-called differential resistance,
R
d
:
R
d
=
_
dI
d+
_
1
.
(b)
(a)
(c)
0
I
0
(d)
Figure 8.3 Resonant-tunneling diode currentvoltage characteristic: the portions labeled by
(a), (b), and (c) correspond to the physical situations illustrated in the previous gure. Portion
(d) shows what happens with the current when the second quasi-bound state enters into
resonance with the electrons from the emitter.
The term negative differential resistance is used to denote that R
d
- 0; this corre-
sponds to the unusual effect of a decrease of the current when the applied voltage
increases.
The following qualitative model of resonant tunneling explains the appearance of neg-
ative differential resistance. Before the level of the well reaches the resonance position,
the current through the diode is very small because it is controlled by non-resonant tun-
neling and by transport over barriers, both of which have low probabilities of occurring.
When the bias corresponds to the case of resonance energies, as depicted in Fig. 8.2(b),
the transmission coefcient and the electric current through the diode increase sharply.
Further increase in the current with increasing voltage bias continues until the resonance
level passes the bottom of the emitters conduction band. There are no electrons (the
conduction band is above the Fermi level, E
F
), as shown in Fig. 8.2(c), to tunnel reso-
nantly and the current decreases in spite of the increase in voltage bias. Consequently,
the currentvoltage characteristic of the structure contains a segment that exhibits neg-
ative differential resistance. The general form of the currentvoltage characteristic is
presented in Fig. 8.3. At large bias, further increase of the current can be realized either
by shifting other quasi-bound states so that they are resonant with electron energies in
the emitter, or by substantial evolution of the potential prole and non-resonant transport
through and over the barrier.
Quantitative characteristics of the resonant-tunneling effect
The resonant transmission of the electrons through the double-barrier diode can be pro-
videdbyphysicallydifferent processes. Conceptually, the simplest wayis direct quantum-
mechanical tunneling, which corresponds to the coherent tunneling process.
Coherent tunneling
In this case, the electron is characterized by a single wavefunction throughout the whole
quantum structure. Traditionally, the cathode and the anode of the diode are referred to
as the emitter and collector, respectively. An electron entering the structure from the
0.2 0.1 0.0 0.1 0.2 0.3 0.4
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0

2 E = 80 meV
T (E
) =
1 +
E
n
E
2
1
n
= 0.1 eV
E (eV)
T
(
E
)
E
Figure 8.4 The full width, 2 LE, of the half-maximum of the transmission peak.
emitter is described by the wavefunction
em
(r. z), which has the form of an incident
plane wave,
em
(r. z) = A
em
e
i(krk
em
z
z)
.
and an outgoing wave that has passed through the structure,
em
(r. z) = B
em
e
i(krk
col
z
z)
;
here, k and r are two-dimensional in-plane vectors, and z is directed perpendicular
to the layers. The quantities k
em
z
and k
col
z
are the z-components of the wavevector in the
emitter and collector, respectively. In an ideal layered structure, the vector k is conserved
because there are no forces acting in plane directions; thus, k is the same for incident
and outgoing waves. The relation of these wavevectors, the corresponding energies, E
em
and E
col
, and the applied voltage, +

0
, can be obtained from the energy conservation
law:
E
em
E
col

h
--2
_
k
em
z
_
2
2m

h
--2
_
k
col
z
_
2
2m
= e+
0
.
The coefcients A
em
and B
em
can be found by solving the Schr odinger equation with
the potential corresponding to the double-barrier structure under voltage bias. The ratio
of the coefcients B
em
and A
em
denes the transmission coefcient:
T(E
) =
[B
em
(E
)[
2
[ A
em
(E
)[
2
.
Instead of solving the Schr odinger equation, here we give a simplied formula for T(E
),
which allows us to introduce the basic parameters of the resonant-tunneling effect.
As follows from the qualitative analysis presented in the previous subsection, T(E
)
has to have sharp peaks in the vicinity of resonance energies,
n
. For a structure with
symmetric barriers, we can approximate these peaks with
T(E
) =
1
1
_
E
n
LE
_
2
. (8.1)
where 2 LE is the full width of the half maximum of the transmission peak. Equa-
tion (8.1) assumes that the probability of tunneling through the structure is unity when
the electron energy, E
, exactly coincides with a quasi-bound state,

n
. In Fig. 8.4, T(E
)
is plotted for several magnitudes of LE. One can see that the tunneling transmission
coefcient can be a very sharp function of the electron energy.
Besides the transmission coefcient, it is convenient to introduce the probability of
tunneling of the electrons out of the well per second, I; it is determined by LE:
I = LE,h.
According to the uncertainty relation we can estimate the lifetime, , of the electrons on
the quasi-bound level between the barriers as
= 1,(2I).
The tunneling probability per unit time can be expressed as the product of the attempt
rate, :
z
,(2L
w
), and the probability of tunneling through a single barrier with one
attempt, T:
I = T
:
z
2L
w
.
where L
w
is the thickness of the well and :
z
is the transverse velocity of the electron in
the well, which can be estimated from the relationship
m
:
2
z
,2 =
n
.
For the purpose of making illustrative numerical estimates, we choose the following set
of parameters: the energy of the rst level,
1
, is 50 meV, the height and thickness of
the barriers are V
b
= 300 meV and L
b
= 40

A, and the well width is L
w
= 100

A. For
this structure, we obtain :
z
= 5 10
7
cms
1
and I = 10
11
s
1
. Hence, the lifetime of
an electron in the well equals = 5 10
12
s. Both and I depend critically on the
height and thickness of the barriers.
The lifetime of the quasi-bound level depends on heterostructure parameters and
material combinations. In Fig. 8.5, the lifetimes of the quasi-bound states are depicted
for ve particular double-barrier structures. The structures are assumed to be made of
GaAs/AlGaAs, InGaAs/AlAs, InGaAs/GaAs, andInAs/AlSb. The widths of the quantum
wells are equal to 46

A, while the barriers have thicknesses ranging from 10

A to 60

A.
The heights of the barriers vary from 0.3 eV to 1.2 eV and are marked on the lines.
Figure 8.5 shows that the lifetime can vary over a very wide range.
0 1 2 3 4 5 6 7
0.01
0.1
1
10
F
i
r
s
t

q
u
a
s
i
-
b
o
u
n
d

s
t
a
t
e

l
i
f
e
t
i
m
e

(
p
s
)
Barrier thickness (nm)
a b e c d
well barriers
GaAs AlAs
In
0.53
Ga
0.47
As
In
0.53
Ga
0.47
As
GaAs Al
0.42
Ga
0.58
As
AlAs
InAs AlSb
GaAs
a
b
c
d
e
V
b
= 1 eV
V
b
= 1.2 eV
V
b
= 0.3 eV
Figure 8.5 The lifetime of the rst quasi-bound state in n-type double-barrier structures made
from the ve material systems presented in the inset. In each structure, the quantum well width
is xed at 46

A. The energy heights of the barriers are indicated next to the curves. After
T. C. L. G. Sollner et al., High-frequency applications of resonant-tunneling devices, in
Electronic Properties of Multilayers and Low-Dimensional Structures, J. M. Chamberlain et al.
(Eds). (New York, Plenum, 1990), pp. 283296.
For an asymmetric double-barrier structure, we dene two different transmission coef-
cients for the left-hand, T
l
, and right-hand, T
r
, barriers, respectively, so that the total
transmission coefcient, T, can be approximated by
T(E
) =
4T
l
T
r
(T
l
T
r
)
2
1
1
_
E
n
LE
_
2
. (8.2)
where LE h
--
I and I = (T
l
T
r
):
z
,(2L
w
). For asymmetric barriers, the maximum
transmission at E
=
n
is less than unity. It is important to recognize that these results
are obtained under the assumption that the electrons transit through the system without
phase-changing scattering, so that E
remains constant.
Using the parameters introduced for the double-barrier resonant-tunneling structures
and assuming low temperatures, it is possible to obtain the estimates for the current
voltage characteristics sketched in Fig. 8.3.
The formulas (8.1) and (8.2) describe the transmission of an electron with xed energy
E
. In order to calculate the electric current, we should take into account all of the
electrons which tunnel from the emitter and collector. We suppose that in the contact
regions heavily doped regions the thermal equilibrium for electrons is established in
a very short time. Therefore, to a reasonable degree of approximation we can assume
that the Fermi distribution gives the distribution of electrons entering the double-barrier
part of the structure. If a voltage bias, +
0
, is applied, the difference between the Fermi
energies of the emitter, E
em
F
, and collector, E
col
F
, is
E
em
F
E
col
F
= e+
0
.
In deriving the Landauer formula in Section 6.5, we obtained the equation for the net
current in a quantum device:
I = I
em
I
col
.
where, in terms of the present analysis, we introduced the current from the emitter to the
collector, I
em
, and the current from the collector to the emitter, I
col
. These currents may
be expressed in terms of the electron concentrations in the two electrodes, n
em
(E
) and
n
col
(E
):
I
em.col
=
e
2h
--
_
dE
T(E
)n
em.col
(E
). (8.3)
Here, the integration runs over the energies E
above the bottom of the conduction

band in the emitter and collector, respectively. The concentrations, n
em
and n
col
, are
given by Eq. (7.6). Let us consider low temperatures, at which the expressions for the
concentrations are simpler. Then, instead of Eq. (7.6) we obtain, for example,
n
em
(E
) =
m
h
--2
(E
F
E
). (8.4)
The electron transverse energy is limited by E
em
c
- E
- E
em
F
, otherwise the sheet
density of the tunneling electrons would be equal to zero. We get the following expression
for the current:
I =
em
2
2
h
--3
_
E
em
F
E
em
F
e+
0
dE
_
E
em
F
E
_
T(E
). e+
0
- E
em
F
. (8.5)
This formula takes into account the fact that, because all states in the collector with
energy E
- E
em
F
e+
0
are occupied, tunneling is possible only for those electrons with
energies satisfying the following conditions: E
em
F
> E
> E
em
F
e+
0
. For e+
0
> E
em
F
all emitter electrons can tunnel through the barriers and we have
I =
em
2
2
h
--3
_
E
em
F
E
em
c
dE
_
E
em
F
E
_
T(E
). e+
0
> E
em
F
. (8.6)
We suppose that the Lorentzians in Eqs. (8.1) and (8.2) are narrow relative to E
em
F
. Then,
the integration over E
in Eqs. (8.5) and (8.6) is straightforward and yields

I =
em
:
z
2h
--2
L
w
_
E
em
F

n
(+
0
)
_
T
l
T
r
T
l
T
r
. (8.7)
where we have used Eq. (8.2) for asymmetric barriers. This equation represents the
tunneling current through the bound state with energy
n
(+), which depends on the
voltage bias as illustrated previously by the example of Fig. 8.3(b). A peak value of
the current is reached when the bound-state energy is resonant with the bottom of the
conduction band in the emitter E
em
c
:
I
p
=
em
:
z
2h
--2
L
w
E
F
T
l
T
r
T
l
T
r
. (8.8)
Table 8.1 Resonance width and collision broadening of level
1
of Al
0.3
Ga
0.7
As/GaAs
double-barrier structures
I
r
/I
sc
L
w
(

A) L
b
(

A) I
r
(meV) at T = 300 K at T = 200 K at T = 70 K
50 70 1.3 10
2
6 10
3
1.9 10
2
2.6 10
1
50 50 1.5 10
1
7.5 10
2
8.3 10
1
3.1
50 30 1.76 0.88 1.3 3.6
20 50 6.03 3.02 4.56 124
When the bound-state energy,
n
, falls below the bottom of the conduction band in the
emitter, the current drops rapidly to the value determined by off-resonance tunneling
processes. This value can be estimated, if we assume that off-resonance transmission
coefcients are constant:
I
v
=
em
2
2
h
--3
E
2
F
T
l
T
r
. (8.9)
Since for the transmission coefcients the inequality T
l
. T
r
_1 is valid, the resonant
current of Eq. (8.8) is much greater than the off-resonance current of Eq. (8.9). The results
obtained for the coherent mechanism of tunneling through a double-barrier structure
support the qualitative discussion given at the very beginning of this section.
Sequential tunneling
Another process responsible for the resonant-tunneling effect is the so-called sequential
tunneling process. In the sequential tunneling scheme, electron transmission through the
structure is regarded somewhat articially as two successive transitions: rst from
the emitter to the quantum well and then from the well to the collector. It is important
to highlight the main difference between the previously studied coherent mechanism
and the sequential mechanism of resonant tunneling. The rst mechanism excludes any
electron collisions during the transition fromthe emitter to the collector. The second sce-
nario applies even when there is electron scattering inside the quantum well. Although
the coherent and sequential processes result in the same behavior of the double-barrier
resonant structures, it is possible and instructive to separate and compare these pro-
cesses. In the previous discussion, we dened a broadening, I, for the quasi-bound
state due to the tunneling process. If both the processes of tunneling and scattering
inside the well occur, the width of the quasi-bound state increases. Let us introduce
the width of the quasi-bound state, I
r
, as the full width of the half maximum of the
transmission peak. Let the collision broadening of this state be I
sc
. From the previ-
ous discussion, we conclude that coherent tunneling dominates if I
r
> I
sc
, whereas
the sequential processes dominate if I
r
- I
sc
. Table 8.1 illustrates different regimes
of tunneling for several specic Al
0.3
Ga
0.7
As/GaAs double-barrier structures at zero
bias. In this table, L
w
and L
b
are the thicknesses of the wells and barriers, respectively.
The ratio I
r
, I
sc
is calculated by using the estimate I
sc
= h
--
,, where is the scatter-
ing time. From the results of Chapter 6, this time may be either calculated from rst
80 K
300 K
Figure 8.6 Measured currentvoltage characteristics of a symmetric Al
0.48
In
0.52
As/Ga
0.47
In
0.53
As
double-barrier resonant-tunneling structure at 300 K (a) and 80 K (b). After Fig. 1, F. Capasso,
S. Sen et al., Quantum-effect devices, in S. M. Sze (Ed.), High-Speed Semiconductor Devices
(New York, Wiley, 1990). Reprinted with permission of John Wiley & Sons, Inc.
principles or deduced from experimental mobility measurements. For Table 8.1, the fol-
lowing values are assumed: when T =300 K, j =7000 cm
2
V
1
s
1
, and . 3 10
13
s; when T = 200 K, j = 2 10
4
cm
2
V
1
s
1
, and = 10
12
s; and when T = 70 K,
j 10
5
cm
2
V
1
s
1
, and 5 10
12
s. If I
r
, I
sc
> 1, elastic collisions can be
neglected and resonant tunneling is highly coherent. From Table 8.1 one can see that
this type of tunneling is typical for the case of low temperatures and thin barriers. In
the opposite limit, I
r
, I
sc
_1, the tunneling processes are more likely to be sequen-
tial. It is important to emphasize that the scattering affects also the magnitude of the
transmission coefcient. The maximum of the transmission probability decreases by the
factor I
r
,(I
r
I
sc
) when scattering takes place. This explains why resonant tunneling
is washed out at high temperatures and in structures with defects and impurities.
Negative differential resistance under resonant tunneling
As we have already seen fromqualitative discussions, resonant-tunneling structures man-
ifest strongly nonlinear currentvoltage characteristics. In particular, it is very impor-
tant that there is a portion of these characteristics for which the current decreases
while the voltage increases. Typical currentvoltage characteristics for a double-barrier
structure of AlInAs/GaInAs are presented in Fig. 8.6. The results are shown for two
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
J
(
1
0
5
A

c
m
2
)
GaAs/AlAs
J
p
/J
v
1.4
InAs/AlSb
J
p
/J
v
3.3
InGaAs/AlAs
J
p
/J
v
12
Voltage (V)
T = 300 K
Figure 8.7 The room-temperature current density versus voltage characteristics and peak-to-
valley ratios, J
p
,J
v
, for high-speed double-barrier RTDs made from three different material
systems: GaAs/AlAs, InAs/AlSb, and InGaAs/AlAs. After E. R. Brown, Resonant tunneling in
high-speed double barrier diodes, in Hot Carriers in Semiconductor Nanostructures: Physics
and Applications, J. Shah (Ed.) (Boston, MA, AT&T and Academic, 1992), pp. 469498.
temperatures, T = 80 K and T = 300 K. The structure is symmetric (similar barriers,
equal thicknesses of spacers, and doping of contacts); thus, there is an almost antisym-
metric currentvoltage characteristic. For the lower temperature, there is clearly seen a
portion with almost zero current at nite voltage biases, which corresponds to the posi-
tion of the resonant quasi-bound level above the Fermi level of the emitter and very small
non-resonance and overbarrier currents. When the level is shifted below the bottom of
the emitters conduction band (+
0
0.75 V), the current drops to almost zero values.
At room temperature, small overbarrier currents exist at any nite voltage bias and the
current drop is also considerably less than in the former case. An important parameter
for characterizing N-type negative differential resistance is the ratio of the maximum
and minimum values of the current the so-called peak-to-valley ratio. Achieving large
peak-to-valley ratios greatly enhances the possibility of applications utilizing negative
differential resistance. For the case presented in Fig. 8.6, the peak-to-valley ratio is 15
at 80 K. At room temperature, the ratio decreases to about 4, but still remains sufcient
for applications. The peak-to-valley ratio depends not only on the physical nature of
the negative differential resistance, but also on many technological and design factors.
Although the development of the double-barrier system is still in progress, for the most
optimized and perfect structures, a peak-to-valley ratio of about 20 or even higher may
be achieved at room temperature.
Comparisons of the currentvoltage characteristics for various double-barrier struc-
tures and the peak-to-valley ratios obtained for T =300 Kare presented in Fig. 8.7. Three
types of heterostructures currently used for resonant-tunneling diodes GaAs/AlAs,
InGaAs/AlAs, and InAs/AlSb are characterized by current densities exceeding
10
5
Acm
2
and exhibit negative conductance at room temperature. The largest peak-to-
valley ratio is reached for InGaAs/AlAs structures (above 10 at room temperature), but
the highest current densities are, typically, for InAs/AlSb, J
m
4 10
5
Acm
2
. For
GaAs/AlAs structures, these parameters are relatively modest: the peak-to-valley ratio
is about 1.4 and J
m
1.5 10
5
Acm
2
.
Another important parameter of any system exhibiting negative differential resis-
tance is the characteristic time of the processes responsible for the negative differential
resistance. This time determines the physical upper frequency limit at which the nega-
tive differential resistance disappears. For the previously considered resonant-tunneling
device, the frequency limit is not easy to estimate because this type of electron transport
has no classical analogue. Careful analysis of the frequency properties of the tunneling
can be done only by numerical self-consistent calculations involving the time-dependent
Schr odinger equation, the kinetic equations describing processes in the contacts, and
Poissons equation. This complex problem has not yet been solved. However, by modu-
lating the barriers by a small time-dependent voltage and examining the time-dependent
response of the system, it has been shown that the characteristic time of the tunneling
processes can be estimated by the following formula:
tr
= m
_
d
0
dz
_
2m[V(z) E
]

2h
--
I
d
:
g
2h
--
I
. (8.10)
where d is the total thickness of the two barriers and the well, and :
g
can be interpreted
as the electron group velocity. If the perpendicular energy of the incident electron, E
,
is much greater than the resonance width, I, it can be shown that the total transit time
through the structure is given approximately by
tr
. The rst term represents the semi-
classical transit time across the structure and the second termis the so-called phase time.
For a typical example of a symmetric resonant-tunneling structure with 17-

A-thick
AlAs barriers and a 45-
A-thick well, the quasi-bound level has an energy

1
0.13 eV;
hence, 2h
--
, I = 0.45 10
12
s. For a drift velocity :
d
10
7
cms
1
, the rst term gives
only 0.8 10
13
s. Thus,
tr
0.5 10
12
s. Hence these quantum devices are ultra-
fast, with response times in the subpicosecond range.
A resonant-tunneling diode as a microwave oscillator
The application of the resonant-tunneling effect in high-frequency oscillators is based on
the existence of the negative differential resistance. In order to review the principles of
using negative differential resistance for obtaining electrical oscillations, we consider the
simplest electric circuit containing a resistance, R
d
, a capacitance, C, and an inductance,
L, as represented by Fig. 8.8(a). Let us introduce the resistance R
d
as the ratio
R
d
=
L+
LI
.
where LI is the change in the current through the resistor when the voltage drop is
changed by L+. Thus, in fact, R
d
is the differential resistance that can have both positive
(a)
- +
R
d
L C
10
1
10
2
10
3
10
0
10
1
10
2
10
3
10
4

InGaAs/AlAs
InAs/AlSb
GaAs/AlAs
P
o
w
e
r

D
e
n
s
i
t
y

(
W

c
m
2
)
Frequency (GHz)
(b)
Figure 8.8 (a) The simplest electric circuit. (b) The generated microwave power per unit
cross-section of double-barrier resonant-tunneling diodes as a function of the frequency for the
same three devices as in Fig. 8.7. (b) After E. R. Brown, Resonant tunneling in high-speed
double barrier diodes, in Hot Carriers in Semiconductor Nanostructures: Physics and
Applications, J. Shah (Ed.), (Boston, MA, AT&T and Academic, 1992), pp. 469498.
and negative sign. If the alternating current,

I , and voltage,

+, applied to the circuit are
I = I
0
e
it
and

+ = +
0
e
it
.
one has the following relation between the magnitudes I
0
and +
0
:
I
0
=
1
Z()
+
0
. Z() = R
d
i
_
L
1
C
_
. (8.11)
where Z() is the impedance of the circuit. The impedance complex resistance
includes the active resistance, R
d
.
Even if +
0
= 0, oscillations can exist in the circuit at frequencies for which Z(
c
) = 0:
c
= i
R
d
2L

_
1
LC

_
R
d
2L
_
2
. (8.12)
For L,C > (R
d
,2)
2
and for a positive resistance R
d
, there are oscillations with the
frequency
_
1,(LC) [R
d
,(2L)]
2
and damping characterized by
= Im{
c
] = R
d
,(2L).
Here, is positive and Im{] denotes the imaginary part of
c
.
If the real part of the impedance is negative,
Re{Z()] = R
d
- 0. (8.13)
it follows that is negative and the steady state of the circuit is unstable with respect to
the generation of voltage oscillations. Thus, if we want to realize voltage oscillations at
a desired frequency, , the differential resistance at this frequency should be negative.
The fact that double-barrier resonant-tunneling structures have extremely small electron
transit times means that these devices exhibit negative differential resistance up to ultra-
high frequencies. In Fig. 8.8(b), the generated microwave power per unit cross-section
of double-barrier resonant-tunneling diodes is presented as a function of the frequency
for the same three devices as were considered in Fig. 8.7. The generated power decreases
with frequency as a result of the previously discussed nite transit time of tunneling
electrons. Absolute values of the power correlate with the current densities achieved for
these devices. It is seen that oscillations up to 1 THz =1000 GHz (i.e., 10
12
oscillations
per second) have been reached for nanoscale quantum devices of this type.
Finally, the resonant-tunneling structures are the simplest quantum devices. They
exhibit strongly nonlinear currentvoltage characteristics with negative differential resis-
tance. Since these devices have nanoscale sizes, they exhibit extremely short transit
times for carrier transport through the structures. These properties allow one to exploit
resonant-tunneling structures for the generation of ultra-high-frequency electromagnetic
oscillations. Indeed, an oscillation frequency of about 1 THz has already been reached
in resonant-tunneling diodes.
8.3 Field-effect transistors
The previously analyzed two-terminal devices diodes are the simplest electronic
devices, for which the current is controlled by the diode bias and vice versa. A use-
ful function can be performed mainly due to nonlinearity of currentvoltage depen-
dences. In contrast, in three-terminal devices known as transistors there exists the pos-
sibility of controlling the current through two electrodes by varying the voltage or the
current through a third electrode. Depending on the principle of operation, transistors
can be associated with one or other of two large classes: (i) eld-effect transistors and
(ii) potential-effect transistors.
Devices controlled by the eld effect
Devices of the rst group are eld-effect or voltage-controlled devices. Acommon feature
of these devices is that a voltage is applied to a controlling electrode a gate which is
capacitively coupled to the active region of the device. A capacitive coupling means that
by applying a voltage to the gate one creates a transverse electric eld in the conducting
channel, but no useful current ows through the gate. The gate electrode is spatially and
electrically separated from carriers in the active region by an insulator or a depletion
region (a region where electrons are absent). The gate electrode controls the resistance
of the active region and, consequently, the current between two other terminals, which
are known as the source and the drain.
D S
Semi-insulating
substrate
n
l
l
d
L
x
G
D
x
y
Semi-insulating
substrate
n
S D G
D
(a)
(b)
Figure 8.9 The MESFET model. The conducting channel of width l is supposed to be grown on a
semi-insulating substrate. The length of the channel is L
x
. The depleted region, l
d
(x), under the
gate, G, is marked by a shadow region. In case (a) there is no voltage bias applied between the
source, S, and the drain, D; in case (b) a positive voltage is applied to the drain. There are no
electrons in the depletion region.
There exist several different types of eld-effect transistor (FET). Before discussing
the most important types of FET, including those based on nanostructures, we will explain
the operation principles by using the simplest FET, the so-called metalsemiconductor
FET (MESFET).
Figure 8.9 presents a simple model for a eld-effect transistor. It is assumed that the
active region of the device is made of an n channel, which can be fabricated, for example,
by homogeneous doping. The source and drain are heavily doped n
regions, which are

assumed to serve as Ohmic contacts. The gate upon the active layer forms a so-called
Schottky barrier, which produces a depletion region, as discussed in Section 7.2 (see
the subsection on the control of charge transfer). The formation of the depletion region
under the gate is illustrated by Fig. 8.9. Such a design is typical for commonly used
semiconducting materials such as GaAs and InP. From the bottom, the active region is
restricted by a non-conducting (semi-insulating) substrate.
If no voltage is applied to the contacts, the depletion region is almost uniform along
the active region, as shown in Fig. 8.9(a). The characteristics of the depletion region are
determined by the built-in (Schottky) voltage. If a negative voltage is applied to the gate
and there is still no voltage between the source and drain, the depletion region extends
farther into the active region and decreases the width of the channel. At some voltage
the channel is completely pinched off. Let the gate voltage be xed and a small positive
voltage be applied to the drain. A current will ow through the channel in the Ohmic
linear regime. If the drain voltage is increased further, it will affect the distribution of
the potential in the device: the width of the depletion region increases near the drain end
of the channel, as shown in Fig. 8.9(b), as a result of the increasing potential difference
between the gate and the drain. At a certain drain voltage, the channel begins to pinch off
at the drain end. At this voltage the current saturates and remains nearly constant with
further increase in the drain voltage. This behavior is illustrated in Fig. 8.10, where typical
currentvoltage characteristics of a MESFET are presented for various gate voltages. It
is clear that, if the gate voltage is negative, the channel becomes narrow and the current
is suppressed, whereas, for positive gate voltage, the depletion region under the gate
Linear
region
Locus for saturation
drain voltage
Dsat
Saturation region
Breakdown
region
0
Drain voltage
D
I
D
D
r
a
i
n

c
u
r
r
e
n
t
G
= 0
G
< 0

G
=
th
G
> 0
B
Figure 8.10 The drain-current versus drain-voltage characteristics for the MESFET at different
gate voltages. The linear and saturation regions are indicated. The breakdown effect occurs at
high voltages +
B
and is presented schematically.
Field-effect transistors (FETs)
Homogenous FETs MOSFETs Heterostructure FETs
Novel Devices
VMT QUIT
Figure 8.11 The family tree of FETs consists of three groups of conventional FETs that have
much in common in their technologies, properties, and applications and new groups of novel
devices. The abbreviations used in this chart are dened as follows: MOSFET metaloxide
semiconductor FET; VMT, velocity-modulation transistor; and QUIT, quantum-interference
transistor.
becomes narrower, the channel opens wider, and the current has to be larger. Thus, in
comparison with a diode, the more complex conguration of the electrostatic potential
in a FET gives rise to strongly nonlinear sourcedrain currentvoltage characteristics.
Qualitatively, two operational regimes are possible, the linear region and the saturation
region, as indicated in Fig. 8.10. For both regimes, there is effective current control by
the gate voltage. At very large sourcedrain voltages, electric breakdown can occur.
In summary, the phenomenon of controlling the resistance of the conduction channel
by an external voltage (by a eld) is the basic principle of any FET.
The FET-family devices
The FET family can be classied as shown in Fig. 8.11, where we give also short expla-
nations for the abbreviations used to denote these FETs. The FET family of transistors
Figure 8.12 A sketch of an Si-MOSFET.
is divided into three large groups of devices used in practice, as well as a group of
potentially new devices based on recent developments. The rst group includes the vari-
ations of metalsemiconductor FETs discussed previously. The second group consists
of metaloxidesemiconductor FETs (MOSFETs). In contrast to the previously con-
sidered MESFETs, in MOSFETs, the gate and conducting channel are separated by a
dielectric (an oxide) instead of a Schottky barrier. The MOSFETs are typically fabricated
by using a metaloxidesilicon structure, with the oxide layer being silicon dioxide. In
Fig. 8.12 we present a sketch of a Si MOSFET. It should be said that Si MOSFETs are
microelectronic devices of the utmost importance. A third group of FETs in Fig. 8.11
incorporates heterostructures and differs in the design of the doping and conducting
channel. For example, the donor-layer heterostructure FET (HFET) subgroup has one or
several doped layers, which supply electrons to the conducting channel. If the channel
is formed on a single heterointerface (see Section 7.2) using a single donor layer the
device is called a modulation-doped FET (MODFET). Such MODFETs have the largest
electron and hole mobilities. That is why they are also called high-electron-mobility
transistors (HEMTs). Examples of material systems used for HFETs are as follows:
n
-AlGaAs/GaAs (unstrained) heterostructures, strained layer n
-InGaAs channels on
GaAs, and n
-InAlAs/InGaAs heterojunctions grown on InP. Although the latter two

structures are more complex, they offer better transport parameters in the conducting
channel.
Figure 8.13(a) shows schematically one of the variants of a HFET a recessed-gate
n
-AlGaAs/GaAs HFET. The device is fabricated on a semi-insulating GaAs substrate

upon which an undoped buffer layer of GaAs is grown. A thin undoped AlGaAs spacer
layer and an n-doped AlGaAs layer are grown on the buffer layer. Source, drain, and gate
electrodes are fabricated on the top of the structure. Under the electrodes, two heavily
n
-doped regions serve as the contacts to the two-dimensional electron gas formed under
the heterojunction. Such a two-dimensional electron-gas channel was studied in Section
7.2 in detail. The gate length varies from 1.0 m to 0.1 m, or less, depending on the
speed needed for applications.
80
AlGaAs undoped
2DEG
GaAs n+
AlGaAs n+
z
z
(z
V
)
E
F
A
B
(a)
(b)
z
(c)
Source
Gate
Drain
GaAs wafer
GaAs undoped
Figure 8.13 A recessed-gate n
-AlGaAs FET. (a) The cross-section of the HFET. (b) The energy
band diagram along the line AB; the Fermi level, E
F
, and electron conned states are shown.
In the donor layer, there are two depletion regions for a thick barrier: one is due to the Schottky
built-in voltage; the second is due to the conduction-band offset at the junction. (c) The
sequence of layers along the line AB.
The dimensions in the vertical directions can be different. This facilitates the fab-
rication of two types of HFET: normally turned-on and normally turned-off devices.
(Both types of heterostructures were studied in Section 7.2.) Figure 8.13(b) illustrates
the energy-band coordinate dependence for the cross-section of the structure under the
gate. The major element affecting the physical picture is the Schottky depletion layer
under the gate. The depletion layer is different for different thicknesses of the barrier
layer. For a barrier layer of small thickness, the Schottky barrier depletes even the quan-
tum well, leading to the normally turned-off state in the absence of the gate voltage.
For larger barrier thicknesses, the Schottky barrier does not prevent the formation of
the conducting channel, and we obtain the normally turned-on state of the HFET as
illustrated in Figure 8.13(b).
Let us briey review the heterostructures employed for FETs based on Si technology.
Although many combinations of materials are being investigated with the goal of fab-
ricating high-quality Si-based heterostructures, the Si/SiGe system is the most studied
and developed and has already found various device applications. Properties of SiGe
alloys and Si/SiGe heterostructures were studied in Chapter 4. These heterostructures
can be used for the creation of two-dimensional electron and hole gases and for the
improvement of device parameters.
In Si/SiGe systems, a low-dimensional n-type channel can be obtained if a strained
Si layer is grown upon a strained SiGe layer. Figure 8.14(a) depicts the cross-section
of an n-channel modulation-doped Si/SiGe FET. The device is similar to the previously
discussed AlGaAs/GaAs FET with selective doping and improved channel connement.
undoped 10 nm
Ti
Pt
Au
10 nm
Si
0.5
Ge
0.5
10 nm
(a)
Gate
Drain Source
AuSb AuSb
Si
1x
Ge
x
Si
x = 0.5 0
Si 20 nm undoped
Si
0.75
Ge
0.25
0.2 m
Si substrate
10
6
cm
tap layer
graded layer
Sb doping layer
buffer layer
Si
10 nm 10 nm
Si
1x
Ge
x
10 nm
Si
0.5
Ge
0.5
20 nm
Si
200 nm
Si
0.75
Ge
0.25
1.1 eV
0.9 eV
Sb doping spike
2DEG
E
c
E
F
E
v
buffer layer buffer layer
(b)
Figure 8.14 The Si/SiGe-based n-channel MODFET. (a) The cross-section of the device; the
electron channel is formed in the strained Si layer. (b) Energy-band diagram of the device. After
H. Daembkes, W. Goodhue et al., The n-channel SiGe/Si modulation-doped eld-effect
transistor, IEEE Trans. Electron Devices, ED-33, 633638 (1986).
C _
IEEE.
The electronchannel is formedinthe undopedstrainedSi layer situatedbetweentwoSiGe
layers. The lower Si
0.75
Ge
0.25
buffer layer is grown on Si and is strained and undoped.
The upper Si
0.5
Ge
0.5
layer is thin and -doped by donors. Next, the Si
1x
Ge
x
layer is
graded, with x varying from 0.5 to 0. The top thin Si layer is undoped. Such a structure
is chosen to create the electron channel and avoid degradation of the characteristics of
the strained layer. Figure 8.14(b) shows the energy-band diagram for this device. The
electron channel on the Si
0.5
Ge
0.5
/Si heterointerface is marked. The Si
0.5
Ge
0.5
donor
layer is -doped, which leads to the formation of a spike in the potential prole.
There are many other schemes for Si-based HFETs that combine the advantages of
heterostructure bandgap engineering and selective-doping methods. Motivated by the

1
10
100
300
0.1 1 5
10
10
10
10
11
12
t
(
s
)
C
u
t
-
o
f
f

f
r
e
q
u
e
n
c
y
,

f
T
(
G
H
z
)
Si NMOS
GaAs MESFET
AlGaAs/GaAs
MODFET
AlInAs/GaInAs
MODFET
L, L
G
(m)
Figure 8.15 The cut-off frequency in GHz versus the device length L or the gate length L
G
in m
for three groups of FETs: MOSFETs, MESFETs, and MODFETs. The latter are represented by
AlGaAs/GaAs and AlInAs/GaInAs material combinations. The parameters are given at 300 K.
After Fig. 25, S. J. Pearton and N. J. Shah, Heterostructure eld-effect transistors, in S. M. Sze
(Ed.) High-Speed Semiconductor Devices (New York, Wiley, 1990). Reprinted with permission
of John Wiley & Sons, Inc.
current wide spread use of Si-based devices, worldwide developments in these areas are
now in progress.
One of the key parameters of contemporary devices is the maximum frequency of
efcient operation the cut-off frequency. At xed device dimensions, the cut-off fre-
quency depends on material characteristics and device design. As a basis for making
comparisons, Fig. 8.15 illustrates representative parameters of all three of these groups
of FETs. One can see that the HFET class exhibits superb high-frequency performance.
Currently, the highest speed is reached for AlInAs/GaInAs HFETs with short gates.
This type of HFET with a nanoscale gate of the length L
G
= 45 nm has a high cut-off
frequency, above 450 GHz.
Nanowire FETs
The previouslyanalyzed HFETs use heterostructures with two-dimensional electrons.
The next step in reducing the number of dimensions of the device is to exploit nanowires.
Semiconductor nanowires and carbon nanotubes are attractive components for nanoscale
FETs.
For example, nanowire FETs can be congured by depositing the nanomaterial onto
an insulating substrate surface, and making source and drain contacts on the ends of the
(a)
(b)
Figure 8.16 (a) A schematic diagram of a Si FET with the nanowire, the metal source, and drain
electrodes on the surface of a SiO
2
/Si substrate. (b) A scanning electron micrograph of a Si
nanowire FET; the scale bar is 500 nm. The 5-nm diameter Si wire image was obtained by using
a high-resolution transmission electron microscope. Reprinted with permission, from Y. Cui,
Z. Zhong et al., High performance silicon nanowire eld effect transistors, Nano Lett., 3,
149152 (2003).
C _
nanowire. Figure 8.16(a) illustrates this approach. There, we show a schematic diagram
of a Si-nanowire FET with the nanowire, the metal source, and drain electrodes on the
surface of the SiO
2
/Si substrate. The image of the 5-nm-diameter Si wire was obtained
by using a high-resolution transmission electron microscope. In Fig. 8.16(b), a scanning
electron micrograph of a Si-nanowire FETis shown; the scale bar represents 500 nm. This
approach may serve as the basis for hybrid electronic systems consisting of nanoscale
building blocks integrated with more complex planar silicon circuitry.
An extremely small FETmay be built on the basis of carbon nanotubes. Indeed, despite
their nanosize diameters, the carbon nanotubes are sufciently robust and may be long
enough to allow electrical connections to lithographically dened metallic electrodes,
therefore making it possible to probe directly the electrical properties of these nanometer-
size structures. Electrical connections to nanotubes can be achieved either by depositing
a metal electrode over the top of the tubes (end-contacted samples), or by placing the
tubes on the top of prepared metal leads (bulk-contacted samples).
An advanced nanotube-based FETis shown in Fig. 8.17. The depicted nanotube FETis
fabricated on SiO
2
/Si substrates. The silicon is doped heavily to serve as a back gate. The
dielectric SiO
2
of thickness 10 nmcovers most of the area of the substrate and electrically
isolates the Si substrate and the wire. Two metal Pd electrodes were deposited to fabricate
the source and drain. This technology provides the nanotube FET with a back gate shown
in Fig. 8.17(a). Additionally, by using atomic layer deposition, an 8-nm-thick HfO
2
lm
can be deposited to cover the device from above. Then, 20 nm of Al is deposited onto
the dielectric (HfO
2
) lm to create the top gate, as shown in Fig. 8.17(b). Thus, we have
presented nanowire FETs with single and double gates. For both devices, the total tube
length between metal electrodes was 2 m, and the top-gated section length was 0.5 m.
(a)
(b)
(c)
carbon nanotube
Figure 8.17 Schematic diagrams of a carbon-nanotube FET: (a) two metal (Pd) electrodes are
deposited onto a carbon nanotube placed on a 10-nm-thick layer of SiO
2
that insulates it
from the back gate; (b) 8-nm-thick HfO
2
serves as an insulator for the top gate; and (c) a
scanning electron micrograph of the carbon nanotube FET. Reprinted with permission, from
A. Jarvey, J. Guo et al., Carbon nanotube eld-effect transistors with integrated ohmic
contacts and high-k gate dielectrics, Nano Lett., 4, 447450 (2004).
C _
American Chemical
Society.
In Fig. 8.17(c), an electron-scanning micrograph of a carbon-nanotube FET is shown.
This nanotransistor works and has good performance.
Inconclusion, the nanowire devices discussedhere have great potential for applications
in nanoelectronics.
Velocity-modulation transistors
For all of the previously analyzed FETs, the gate voltage controls the number of con-
ducting electrons, and the speed of operation is determined by the transport time of
electrons along the conducting channel. An alternative principle of device operation
consists of modulating electron mobility, or, possibly, electron drift velocity, without
signicant changes in the number of current carriers. To achieve such a modulation
of the drift velocity one can use spatially nonuniform (selective) carrier scattering
across the active channel. Then, a redistribution of the carriers across the channel in
response to a gate voltage would lead to control of the velocity. The time of such a
redistribution across a narrow channel can be considerably smaller than that of the
longitudinal transport. A three-terminal device with modulation of the drift veloc-
ity by an external voltage has been referred to as the velocity-modulation transistor
(VMT).
The characteristic time of charge transfer in a channel of thickness about 100

A can
be estimated to be as short as 10
13
s. Thus, for ultra-high-frequency operation, the
velocity-modulation effect should be the dominant factor controlling electric signals.

E
k
x,y

E
k
x,y

E
k
x,y
Ch 2
Ch1
P
2
P
1
t'
2
t
2
t
1
t'
1
x = 0 x = L
Semi-insulating substrate
Source Drain
Gate
Channel 1
Channel 2
x
y
z
(a)
(b)
(c)
Lowest subband
(populated)
Next subband
(unpopulated)
Barrier
E
F

Figure 8.18 The schematics of QUIT with two parallel channels: (a) the structure of the device;
(b) electron subbands in different regions of the device and the Fermi energy (under
equilibrium); (c) an illustration of the electron paths in the device; t
1
, t
2
and t
/
1
. t
/
2
indicate the
amplitudes of split and interfering waves, respectively. P
1
and P
2
denote factors of different
phase shifts in the channels.
Quantum-interference transistors
The remaining class of novel nanoscale FETs presented in Fig. 8.11 is the so-called
quantum-interference transistor (QUIT). It is based on quantum ballistic electron trans-
port. According to the classication given in Section 6.2, for such a transport regime,
one has to make the coherence length, l
, greater than the characteristic device scale, L:

l
> L; (8.14)
that is, this device is a mesoscopic device. The principle of operation of a QUIT is the
control of the interference pattern of conducting electrons by an external voltage.
Remember that interference is one of the most important effects in the physics of
waves. We studied the interference effect in Sections 2.4. Examples of interference
patterns are given in Figs. 2.3 and 2.7. In a QUIT, these patterns can be arranged as a
result of interference of the waves traveling through two or more channels (arms).
A schematic diagram of a two-channel quantum-interference device is shown in
Fig. 8.18(a). Basically, it is an ordinary FET with a short channel and a barrier par-
allel to the current. The barrier is embedded in the middle region of the device. The
barrier splits the main channel into two channels: 1 and 2. There are also two contacts
to the main channel the source and the drain. The gate is placed on the top of the
device. Owing to the gate, the symmetry of channels 1 and 2 is broken if a gate voltage
is applied. The distance between contacts, L, satises the inequality of Eq. (8.14). The
widths of the main channel and the split channels are small, so that there is a quantization
of transverse electron states in the z direction. Two-dimensional subbands are illustrated
in Fig. 8.18(b); the subbands, E(k
x
. k
y
), and the Fermi energy, E
F
, are plotted for three
major device regions. For simplicity, one can assume that only the lowest subband is
populated in each of the device regions. If a voltage is applied to the gate, in the mid-
dle region, the bottoms of the subbands,
1
and
2
, for channels 1 and 2 are generally
different. The electron energy spectra are of the form
E =
1
h
--2
2m
_
k
2
x.1
k
2
y
_
=
2
h
--2
2m
_
k
2
x.2
k
2
y
_
. (8.15)
where E is the energy of an incident electron, and k
x
and k
y
are the components of its
wavevector in the plane of the device. Obviously, the energy E does not change along
the channel. The k
y
-component is also conserved because it is assumed that there are no
forces along the y direction. Thus, fromEq. (8.15), it follows that only the k
x
-components
in the channels can be different.
Let an electron be injected from the source into the left region of the main channel.
Its wavefunction is
L
=
L
(z)e
i(k
x
xk
y
y)
.
where
L
(z) is the wavefunction of the lowest subband in the left region. We can intro-
duce the amplitudes of the waves transmitted into channels 1 and 2 as t
1
and t
2
. These
amplitudes determine the electron wavefunctions inside the middle device region:
M
=
_
t
1
M.1
e
i(k
x.1
(xx
L
)k
y
y)
. upper channel.
t
2
M.2
e
i(k
x.2
(xx
L
)k
y
y)
. lower channel.
where
M.1
(z) and
M.2
(z) are the wavefunctions of the lowest subbands in channels
1 and 2, respectively, and x
L
is the coordinate of the left end of the channels. Since,
in general, the potential proles in the two channels are different, the wavevectors are
different: k
x.1
,= k
x.2
. Then, let t
/
1
and t
/
2
characterize the waves transmitted from the
channels into the right device region; see Fig. 8.18(c). The wavefunction inside the right
region can be written as
R
=
_
t
1
P
1
t
/
1
t
2
P
2
t
/
2
_
R
(z)e
i(k
x
(x
L
x
R
)k
y
y)
. (8.16)
where P
1
exp(ik
x.1
L). P
2
exp(ik
x.2
L), and
R
(z) is the wavefunction of the trans-
verse motion at the right end of the channels. For the sake of simplicity, we neglect
multiple reections from the ends of the barriers.
The wavefunctions
L
and
R
are normalized to unity; thus, the total transmission
coefcient for the device is
T(E) = [t
/
1
P
1
t
1
t
/
2
P
2
t
2
[
2
. (8.17)
We can assume that both channels, 1 and 2, are symmetric with respect to z = 0 in the
absence of a gate voltage. Thus, for the lowest occupied subbands we nd
t
2
= t
1
and t
/
2
= t
/
1
.
Then, Eq. (8.17) gives us
[T(E)[
2
= 2[t
1
t
/
1
[
2
(1 cos ). (8.18)
where
= (k
x.2
k
x.1
)L (8.19)
is the relative phase shift of the two channels. If we introduce the average electron
velocity,
:
x
=
h
--
(k
x.1
k
x.2
)
2m
.
we can represent the difference between the wavevectors as follows:
k
x.2
k
x.1
=

1
2
h
--
:
x
. (8.20)
Now, the phase shift takes the form
=
L
:
x
2
h
--
.
The origin of the phase shift is obvious: if
1
,=
2
, a difference between the kinetic
energies in the two channels gives rise to different phases of the waves coming into the
right device region; of course, these different phases lead to quantum interference.
For channels that are symmetric with respect to z = 0, in the absence of a gate voltage,
we nd
1
=
2
=
0
, and the phase shift equals zero. If we apply a gate voltage, the
potential energy as a function of the transverse coordinate is modied:
V(z) = V
0
(z) e+(z).
Here, +(z) is the potential induced by the applied voltage. This leads to subband energies
1
=
0
e
M.1
[+[
M.1
).
2
=
0
e
M.2
[+[
M.2
).
Using Eqs. (8.20) and (8.19), we nd the phase shift,
=
L
:
x
e+
12
h
--
. (8.21)
where +
12

M.2
[+(z)[
M.2
)
M.1
[+(z)[
M.1
). The value +
12
represents the dif-
ference between the average potentials in the channels. This difference determines the
transmission through the device.
To calculate the electric current, one can use the results of Section 6.5, where quantum
ballistic transport was analyzed. Equation (6.120) denes the device conductance, G. At
low temperatures, these results lead to the Landauer formula of Eq. (6.121), which, in
the case under consideration, has the form
G =
2e
2
h
T(E) =
4e
2
h
[t
1
t
2
[
2
(1 cos ). (8.22)
The second term in the brackets is due to electron interference. One can see that the
interference controls the device conductance. If the phase shift of split waves = 0
(+
12
= 0), the conductance reaches the maximum, G
max
= 8e
2
[t
1
t
2
[
2
,h. If = , i.e.,
e+
12
=
h
--
:
x
L
. (8.23)
the interference is destructive and the conductance vanishes.
Let us introduce the characteristic transit time through the channels, t
tr
= L,:
x
. If
we set L = 2000

A and :
x
= 2 10
7
cms
1
, we get t
tr
= 1 ps, and for the destruc-
tive potential difference we get +
12
2 mV. The shorter channels require a larger gate
voltage. Obviously, the transit time, t
tr
, determines the cut-off frequency of the device,
cf
= 2,t
tr
. Since the device channels are not doped and can be made quite short, these
devices portend operation up to the terahertz frequency region.
Our simple model of the quantum-interference transistor allows us to compare this
device with a conventional FET. If the latter is in a normally turned-on state with the
Fermi energy of electrons in the conducting channel equal to E
F
, we can estimate the
threshold voltage needed to deplete the FET channel as
e+
th.FET
E
F
=
h
--2
k
2
F
2m
. (8.24)
For quantum devices described by Eq. (8.23), we nd
e+
th.QUIT
e+
12
=
h
--
:
x
L
=
h
--2
k
F
m
L
=
h
--2
k
2
F
2m
2
k
F
L
= e+
th.FET
F
L
. (8.25)
where
F
is the de Broglie wavelength of electrons in the Fermi level. Hence
F
_ L;
one can see that the quantum device can operate with a signicantly smaller controlling
gate voltage.
Another design of a structure for a quantum-interference transistor is sketched in
Fig. 8.19. The structure is T-shaped and consists of a channel connecting the source
(grounded), drain, and a transverse arm (stub). The transverse arm has nite dimensions
and has a gate on the end of the arm; this end is referred to commonly as the top
of the device. If the length, L
2
, and width, L
3
, are less than the coherence length, l
,
the reection of the electron wave from the arm produces de Broglie wave-interference
patterns. If the width of the main channel, L
1
, also is small in comparison with l
,
the pattern extends across the channel and determines the transmission coefcient of
electrons through the device, i.e., the sourcedrain conductance. A voltage applied to
the gate changes the penetration length, L
, of the electron wave into the arm and,

as a consequence, the interference pattern. Estimates show that the gate voltage can
effectively control the conductance of such a T-shaped device when the dimensions L
1
,
L
2
, and L
3
are approximately several hundred

Aor less.
L
*
L
2
L
1
L
3
D
Figure 8.19 The schematics of a T-shaped QUIT. The source (grounded), drain, and gate
electrodes are indicated.
q = 0.69
q = 0.42
(a)
(b)
(c)
Figure 8.20 Experimental results on the QUIT effect: (a) and (b) for the two-channel QUIT T =
0.03 K, (c) for the T-shaped structure, both fabricated from a high-mobility AlGaAs/GaAs
heterostructure. (a) Modulation of the device conductance by the gate voltage. (b) The phase
shift as a function of the gate voltage. (c) The same as in (a) for the T-shaped device at ve
different temperatures. Reprinted with permission from K. Kobayashi, H. Aikawa et al. Fano
resonance in a quantum wire with a side-coupled quantum dot, Phys. Rev. B, 70, 035319
(2004).
C _
As a practical realization of the QUIT effect, we present the results obtained for an
AlGaAs/GaAs structure. The channels were made fromhigh-mobilityGaAs. Inone of the
channels a GaAs quantum dot was embedded under the gate. This device corresponds
to the scheme in Fig. 8.18(a). If one of the channels were blocked, one would obtain
the T-shaped structure, as in Fig. 8.19. Both cases also are illustrated schematically in
the insets of Fig. 8.20. The low-temperature mobility of the two-dimensional electron
gas was as high as 9 10
5
cm
2
V
1
s
1
, and the measured electron concentration was
2 10
11
cm
2
. These parameters correspond to a mean free path of about 8 m. Low
temperatures (0.030.7 K) ensure that the dephasing length l
is even greater. In the

case under consideration, the lengths of the channels are about 1 m, and the quantum-
dot (two-dimensional) size is 0.2 m0.2 m. In Fig. 8.20(a), the QUIT conductance
is presented as a function of the gate voltage (in units of the quantum conductance,
e
2
,h). One can see that the conductance is periodically modulated by the gate voltage, in
accordance with the previously developed model of the QUIT. Special experiments with
applied magnetic eld, B, make it possible to deduce the phase shift of the electrons in
the QUIT. The results are presented in Fig. 8.20(b). The modulation of the conductance
for a T-shaped device is shown in Fig. 8.20(c). Measurements at various temperatures
illustrate the smearing out of the interference patterns due to the temperature dispersion
of the energy of the electrons entering the device.
Summarizing, a pronounced QUIT effect is observed for devices with submicrometer
scale at lowtemperatures. The scaling down of device dimensions will allowsignicantly
increased working temperatures.
In conclusion, the quantum-interference effect provides a new principle of operation
for three-terminal devices. These devices are now in an early stage of development.
However, these approaches portend effective control by a small voltage as well as the
development of very-high-speed transistors.
8.4 Single-electron-transfer devices
Before analyzing examples of the devices based on single-electron transport studied in
Section 6.5, we shall overview briey a relatively simple technique for the formation of
structures with low-dimensional electron gas.
The split-gate technique
Having a two-dimensional electron gas on an interface, or in a quantum well, one can
further lower the electron-gas dimensionality by various methods. One such method
is the so-called split-gate technique. The principles of this technique can be explained
as follows. Typically, two-dimensional electrons are separated from the surface of the
sample by a wide-bandgap dielectric layer. It can be a SiO
2
layer on Si, an AlGaAs barrier
layer onGaAs, etc. Figure 8.21(a) illustrates sucha structure. Amodulation-dopedbarrier
layer decreases electron scattering by the donors and results in high electron mobility. A
thin GaAs layer grown on the top of this structure is used as additional electrical isolation
fromthe metal gates. Let a metal strip a gate be deposited onto the top of this structure.
The distribution of the potential energy for the case of a negative applied gate voltage is
presented in Fig. 8.21(b). According to this energy scheme, two-dimensional electrons
are repelled from the region beneath the metal strip; their Fermi energy, E
F
, is below
the lowest subband energy. As a result, the region under the gate becomes completely
depleted, as depicted in Fig. 8.21. Now it is clear that, by using several gates, possibly
d
sp
l
d
e
G
Thin layer of GaAs to
cover AlGaAs
n
+
-AlGaAs
GaAs
(a)
Thin undoped
layer of AlGaAs
Narrow
metal strip
Z
2DEG
(b)
e
b
l
d
d
sp
0
z
E
F
E
F
2DEG
e
1
Figure 8.21 (a) Two-dimensional electron gas (2DEG, denoted by dashed lines) at the interface of
a modulation-doped GaAs/AlGaAs heterostructure with a depletion region under the negatively
biased narrow metal strip (gate). (b) The band diagram of the AlGaAs/GaAs heterostructure in
the gate region. The lowest subband is shown by the
1
line. +
b
is the built-in Schottky voltage
and +
G
is the gate voltage.
of different forms, one can create various congurations of regions occupied by the
electrons. It is possible to form wires, dots, rings, cavities, etc. for the two-dimensional
electrons. For example, if two closely placed parallel metal strips are fabricated on the
top of the heterostructure, then, by applying negative voltage to these two gates, we can
form two side barriers for the electrons and conne them into a channel. If the channel is
narrow enough, the two-dimensional electrons can be quantized in the second direction
and we obtain a quantum wire; such a wire is shown in Fig. 8.22. The connement of
the electrons to the dots, wires, rings, etc. can be accomplished by a heterojunction on
one side and electrostatics on all other sides. The split-gate technique has successfully
been exploited for measurements of transport regimes in various quantum structures, for
example, in quantum point contacts and electron waveguides.
A considerably more sophisticated design of patterning of a two-dimensional gas to
a shape desirable for single-electron applications is presented in Fig. 8.23. The main
features of the design shown in Fig. 8.23(a) are the following: (i) the Ohmic contacts
(OC) to the two-dimensional electron gas (contacts to two electron reservoirs (R)) and (ii)
a system of gates, which create electrostatic tunnel barriers (TB) and conne electrons
into a quantum dot (QD). The tunnel barriers are formed when the voltages applied to
the gates are negative with respect to the voltages applied to the contacts. The barriers
should be high enough to decouple the quantumdot and the reservoirs. In Fig. 8.23(b), the
l
d
Thin layer of GaAs to
cover AlGaAs
+
-AlGaAs
GaAs
Metal strips
1DEG
Quantum wire (1DEG)
d
sp
Z
n
Figure 8.22 A quantum wire at the interface of a GaAs/AlGaAs heterostructure with two
depletion regions under the negatively biased metal strips.
Gate Gate
E
F
V(z)
TB TB
QD R R
(b) (a)
AlGaAs, n+
AlGaAs undoped
GaAs undoped
GaAs wafer
2DEG in GaAs
Gates
QD TB
R
R
A
OC OC
B TB
GaAs Cap Layer
Figure 8.23 (a) Design patterning of two-dimensional gas to a shape desirable for single-electron
applications: OC, ohmic contacts; R, reservoir of electrons; QD, quantum dot; and TB, tunnel
barrier. (b) The potential prole, V(z), along the line AB. Ohmic contacts, OC, are not shown
since they are outside of the depicted region.
resulting potential prole is depicted schematically. Split-gate techniques and resulting
structures are used to observe single-electron effects. Indeed, if the quantum-dot-like
structure is small enough, it can be considered to be a single-electron box. A voltage
applied to the contacts (the source and the drain) induces transfer of electrons through
the structure one by one, as studied in Chapter 6. In general, the exible split-gate
technique is a powerful method to realize single-electron-transport devices.
Single-electron transistors
Single-electron devices like the one sketched in Fig. 8.23 are two-terminal devices, i.e.,
they are diode-type devices. It is possible to introduce an additional gate and create
(a)
G
1
G
2
G
4
G
3
Drain Source
Gate
(b)
C
1
C
3
C
2
V
G
Figure 8.24 (a) The design of a quantum-dot single-electron transistor; (b) an electric circuit with
a quantum-dot single-electron transistor.
C
o
n
d
u
c
t
a
n
c
e

(
1
0
6
1
)
Sample 1a
Sample 1b
12
8
4
4 0 8 12 12
12
8
8 4
4
0
Gate voltage (mV from threshold)
Figure 8.25 Modulation of conductance in single-electron transistors. Reprinted with permission,
from M. A. Kastner, The single electron transistor, Rev. Mod. Phys., 64, 849 (1992).
a three-terminal device, a single-electron transistor. In Fig. 8.24(a), the design of the
metallic electrodes on the top of a heterostructure with a two-dimensional electron gas
is shown. The gates G
1
, G
2
, and G
4
form a quantum dot as above; the gate G
3
addition-
ally controls the size and shape of the dot, changing its properties. The single-electron
transistor works as follows.
The electron transfer is determined by two factors: the Coulomb charging of the dot and
the quantized energy levels in the dot. If the drain is biased with respect to the source, an
electric current occurs in the regime of single-electron transfer. By applying the voltage
to the gate and changing the quantum-dot parameters, one can change the conditions of
electron tunneling and affect the sourcedrain current. Examples of modulation of the
conductance in single-electron transistors by the gate voltage are presented in Fig. 8.25.
The devices have almost the same geometry. Their dimensions are large enough to have
a number of quantized levels. In Fig. 8.25 each peak in the conductance corresponds to
transfer of one electron, when an energy level enters into resonance with the electron
Figure 8.26 A scanning electron microscope image of a single-electron transistor. From Y. Ono,
A. Fujiwara et al., Manipulation and detection of single electrons for future information
processing, J. Appl. Phys., 97, 031101 (2005). Reused with permission from Yukinori Ono,
Akira Fujiwara, Katsuhiko Nishiguchi, Hiroshi Inokawa, and Yasuo Takahashi, Journal of
Applied Physics, 97, 031101 (2005).
C _
states in the contacts. Though the conductance versus gate-voltage dependences are
different, i.e., not reproducible, the peak spacing is the same for both devices. It is
determined by the change in the gate voltage required to change the charging energy
of the quantum dots by one electron. The gures show clearly that the electric current
is modulated signicantly by the gate voltage. Thus, for transistors with single-electron
transport, strong control of very small electric current may be possible. The problem of
fabrication of reproducible devices requires further improvements in technology.
A single-electron pump and turnstile
In the single-electron transistor considered previously, the barriers which separate the
electrodes and the dot practically do not change when a gate voltage is applied (instead,
quantum-dot parameters are modied). However, variation of the barriers is possible and
it adds a new function to the single-electron devices. As will be shown below, a device
with tunable barrier can work both as a pump and as a turnstile. Consider such a case for
the example of silicon-based devices. A scanning electron microscope image of such a
device is shown in Fig. 8.26; this single-electron transistor is made in a very thin Si layer.
The conducting channel, source, and drain of the device are clearly seen in the gure.
The ne gates are made of polycrystalline Si, which is a good conducting material. Three
gates form a dot. Two of them, gates 1 and 2, tune the barriers between the dot and the
electrodes. Figure 8.27 illustrates the sequence for the pump operation. We start with a
Coulomb-blockade state with an electron in the dot (state I). Then, by applying a negative
bias to gate 1 and making the left barrier higher, we may close the left channel (state
II). During this action, we keep the dot potential nearly constant by applying a positive
control bias to gate 2. Then, we raise the island potential so that an electron in the dot is
ejected to the right channel, ending up with a new Coulomb-blockade state with n 1
electrons (state III). Next, we open the left channel and close the right channel, keeping
(I)
( II)
( III)
(V)
(IV)
Figure 8.27 The sequence for pump operation.
the island potential nearly constant (state IV), and then lower the island potential so that
an electron can enter from the left channel (state V). Thus, during one period only one
electron can be controllably transferred through the device. This system can be viewed
as a single-electron pump, since it transfers the electrons one by one without applying a
bias between the source and drain.
The electron pump can have a number of applications. One of themis a current etalon.
Indeed, the average current owing through the device is
I = ef.
where f is the frequency of variation of the voltage on the gates. Since the frequency can
be measured with very high accuracy, the device can serve as the electric-current stan-
dard. The accuracy of such a standard is much higher than the analogous characteristic
of any other current standards.
The turnstile operationis realizedbymakinga small change inthe previouslydescribed
regime of operation. The detailed procedure for the turnstile is illustrated schematically
in Fig. 8.28. In the turnstile mode, with a non-zero sourcedrain voltage, we rst close
gate 2 so that an electron enters only from the left side (stage I). We next close both gates
(stage II), and then open gate 2 so that an electron is emitted to the right (stage III). Finally,
we again close both gates (stage IV), and open gate 1, reaching the initial state. This
procedure can be accomplished by applying an AC bias to each gate with the phase shift
of . Noteworthy is that, in this turnstile procedure, at least one of the channels is always
closed, which is not the case in the pump operation mode. This is advantageous because
high resistance effectively prevents the tunneling of two electrons one of the major error
sources for single-electron-transfer regimes. Figure 8.29 shows the drain-current versus
drain-voltage characteristics in the aforementioned turnstile mode at several frequencies,
f = 0.001, 0.5, and 1.0 MHz. Staircases quantized in units of ef are observed both for
(I)
E
F
E
F
( II)
E
F
E
F
(IV)
E
F
E
F
( III)
E
F
E
F
Figure 8.28 Schematics of a turnstile operation. From Y. Ono, A. Fujiwara et al., Manipulation
and detection of single electrons for future information processing, J. Appl. Phys., 97, 031101
(2005). Reused with permission from Yukinori Ono, Akira Fujiwara, Katsuhiko Nishiguchi,
Hiroshi Inokawa, and Yasuo Takahashi, Journal of Applied Physics, 97, 031101 (2005).
C _
2005
-120 -80 -40 0 40 80 120
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
1 MHz
0.5 MHz
1 kHz
Drain voltage (mV)
D
r
a
i
n
c
u
r
r
e
n
t
(
p
A
)
Figure 8.29 Drain current versus drain voltage characteristics of a turnstile device. 1 pA =
10
12
A, where pA denotes a picoampere. From Y. Ono, A. Fujiwara et al., Manipulation and
detection of single electrons for future information processing, J. Appl. Phys., 97, 031101
(2005). Reused with permission from Yukinori Ono, Akira Fujiwara, Katsuhiko Nishiguchi,
Hiroshi Inokawa, and Yasuo Takahashi, Journal of Applied Physics, 97, 031101 (2005).
C _
2005
Potential-effect transistors
Bipolar transistor Hot-electron transistor
Silicon bipolar
transistor
Heterojunction bipolar
transistor
Figure 8.30 The family tree of potential-effect transistors. The devices are set apart into two large
groups: bipolar and hot-electron transistors.
positive and for negative drain voltages. This is the evidence of single-electron transfer
by the turnstile operation. The levels of the current plateaus for the positive drain voltages
are exactly equal to ef to within the accuracy of the measurement system of about 10
2
.
In these experiments, the working currents were extremely small.
In conclusion, the main principle of the FETs current control by means of a voltage
applied through capacitively coupled electrode(s) can be successfully applied to single-
electron devices. By combining the eld effect and the single-electron phenomena, one
can manipulate the device states using extremely small electric currents.
To date, single-electron experiments have been conducted at low temperatures (typ-
ically below 20 K). However, with further scaling down of semiconductor technology
to nanosizes, we may expect that single-electron effects and devices will be exploited at
higher temperatures, including under room-temperature conditions.
8.5 Potential-effect transistors
According to the classication given in Section 8.3, a second large family of electronic
three-terminal devices is that of potential-effect transistors. This family is shown in
Fig. 8.30. In contrast to the case of FET devices, potential-effect transistors are current-
controlled. The controlling electrode is resistively coupled to the active region of the
device and the carriers are separated by an energy barrier.
The most important representative of this class is the bipolar transistor, which was
invented in 1947, and has undergone considerable and persistent transformation. Cur-
rently, the bipolar transistor provides high speed of operation in most circuit applications.
Detailed analysis of the physics of bipolar transistors is certainly beyond the scope of
this book. Instead, at the end of this chapter we provide references to books on bipolar
transistors. However, we shall consider the working principles of this transistor, at least
very briey.
pn Junctions
The key element of the transistor is the so-called pn junction. In the simplest case it is
a junction of two regions in a semiconductor: one side is doped by donors and contains
E
v
-
-
Depletion
regions
-
+ + + + + +
Electrons
Holes
E
c
E
F
E
c
BI
e
E
v
+
+ + + + + +
+
-
- - - - - -
Figure 8.31 A pn junction with abrupt coordinate dependence of doping concentration.
e
r
e
f E
Fp
E
Fn
E
Fp
E
Fn
Equilibrium Forward bias Reverse bias
f r
p p p n n n
E
Fn
E
Fp
(a) (b) (c)
Figure 8.32 The scheme of a pn junction (a) at equilibrium; (b) under forward bias +
f
; and
(c) under reverse bias +
r
. For non-equilibrium cases, (b) and (c), instead of one Fermi level,
there are two quasi-Fermi levels, one for holes, E
Fp
, and one for electrons, E
Fn
, as shown.
electrons; the other side is doped by acceptors and contains the holes. In Fig. 8.31, the
scheme of such a pn junction is presented. An abrupt transition between the p- and
n-doped sides of a semiconductor is assumed. The charge transfer across the junction
lines up the various Fermi levels and results in a Fermi level, E
F
. It depletes two narrow
regions in the p- and n-doped sides and creates a built-in electrostatic potential, +
BI
.
The potential barriers of height e+
BI
prevent the penetration of electrons into the p-type
side of the structure and of holes into the n-type side of the structure. No electric current
passes between p and n regions under such an equilibrium condition; see Fig. 8.32(a).
As is clear from this discussion, the potential prole which appears at the pn junction
and the possibility of controlling of it are characteristic for potential-effect devices.
p
n
I
I
S
I
Figure 8.33 The currentvoltage characteristic of the pn junction.

Now, let a voltage be applied to the pn junction. The corresponding electric circuits
are shown in the upper part of Fig. 8.32. There are two quite different cases. The forward
bias of the pn junction corresponds to a negative potential, + = +
f
(+
f
> 0), applied
to the n-type side. Potential barriers both for the electrons and for the holes decrease and
are equal to e(+
BI
+
f
). This case is illustrated by Fig. 8.32(b). A positive potential
(the reverse bias +
r
> 0 of the junction) leads to an increase of the potential barrier
between the n and p parts of the structure, as illustrated by Fig. 8.32(c). Now the barrier
height equals e(+
BI
+
r
). The applied voltage is dropped primarily across the depletion
regions because their resistances are much greater than those of neutral n and p regions
of the junction. The voltage breaks down the equilibrium; in particular, there is no longer
a common Fermi level. Instead of the true Fermi level, quasi-Fermi levels E
Fn
and E
Fp
can be introduced for better understanding of changes in the potential relief, as shown
in Fig. 8.32. As soon as E
Fn
,= E
Fp
, an electric current starts to ow between the n and
p parts of the structure. For a forward applied bias, the current increases exponentially
as the potential barrier decreases, and the junction becomes strongly conducting. For
an applied reverse bias, the current is small and saturates at large biases. The current
voltage characteristic of the pn junction is presented in Fig. 8.33. The main features of
this characteristic are a rectifying behavior and a strong nonlinearity. This nonlinearity
is used in numerous applications of the pn diode.
Bipolar transistors
A homostructure bipolar transistor consists of two pn junctions. Usually one of
them the emitter junction is forward-biased, while the other the collector junc-
tion is reverse-biased. The energy-band diagram for a bipolar transistor is shown
schematically in Fig. 8.34. This diagram clearly illustrates the formation of a potential
for electron and hole motion. The control of this potential is the key element of the
bipolar transistor. The device depicted is of the npn type, though almost all results are
applicable also to pnp devices. This diagram can be easily understood in terms of the
charge-transfer effect between the emitter (left n region), the base (middle p region), and
the collector (right n region). The voltage bias is supposed to be such that an electron
travels from the left n region across the p region to the right n region. An applied forward
n
+
n
p
Emitter Base Collector
Figure 8.34 An energy band diagram of a homostructure npn bipolar transistor along the
direction of electron travel. Quasineutral emitter, base, and collector regions are indicated. The
emitter junction is forward-biased, and the collector junction is slightly reverse-biased.
I
B
I
C
5
4
3
2
1
CB
Figure 8.35 The collector current I
C
in a bipolar transistor for ve different base currents, I
B
.
bias to the emitter junction lowers the energy barrier for an electron owing from the
emitter to the base. Simultaneously, a similar ow of holes from the base to the emitter
appears. Those electrons that overcome the barrier travel across the base. Recombination
of the minority carriers (electrons) occurs in the base, but, if the thickness of the base
region is less than the diffusion length of the minority carriers, these electron losses are
negligible. When the electrons reach the collector junction, they are swept away into the
collector due to the high electric eld within the depletion region of the latter junction
and form the collector current. In the bipolar transistor, each of the conducting regions
(the emitter, the base, and the collector) is provided by an electrical contact. Thus, this is
a three-terminal device. The operation of a bipolar transistor is based on the principle of
controlling the current by injecting minority carriers. For example, in an npn device,
the injection of minority carriers from a forward-biased np junction (emitterbase) into
the base provides the current through the base contact and the controlling function of the
collector current. In Fig. 8.35, the collector current versus the collectorbase voltage,
+
CB
, is shown for various currents, I
B
, through the base contact (compare this with the
currentvoltage characteristics of the FETs presented in Fig. 8.10).
The carriers travel from the emitter to the collector perpendicular to junctions; thus,
the carrier transit time through the base determines the speed of operation and the
cut-off frequency of the device. Device scaling in order to reduce the transit time in a
Collector
n
+
Si substrate
n Si
p
+
SiGe
n
+
Si
n
++
Si Base Base
Emitter
Figure 8.36 The schematic cross-section of an npn Si/SiGe heterojunction bipolar transistor.
n
, n
, p
, and n denote high, intermediate (n
and p
), and relatively low doping,

respectively. The p
-SiGe layer is strained.

homojunction bipolar transistor can not be achieved while simultaneously realizing all
the requirements for high performance; such requirements include high doping levels
both in the emitter and in the base, and a small base region. The use of heterostructures
helps to solve these problems and to improve device performance. Thus, one can divide
this transistor class into two large and important groups: homojunction and heterojunc-
tion bipolar transistors, as shown in Fig. 8.30. As an example of the latter subclass
of bipolar transistors, consider the Si/SiGe device. In Fig. 8.36, the cross-section of a
Si/SiGe heterostructure npn bipolar transistor is shown. The device is fabricated on
an n
-Si substrate (subcollector with doping of about 2 10

19
cm
3
) contacted by an
electrode. The collector layer is doped to approximately 2 10
16
cm
3
and has a width
of about 0.3 m. The p
base is fabricated of a Si
1x
Ge
x
alloy with x ranging from
0.2 to 0.3, and has a width of typically 3050 nm. The base doping is in the range from
2 10
18
to 5 10
19
cm
3
. Two-sided metallic electrodes provide the direct contact with
the base region. The emitter is made of n
-Si with doping of about 5 10

17
cm
3
. A
heavily doped layer, marked as n
-Si, is placed at the top of the emitter to provide a

good contact to the metallic electrode. Owing to the very short base and other pecu-
liarities of the heterojunctions, for such a device an extremely high cut-off frequency of
75 GHz at room temperature is measured, whereas at liquid-nitrogen temperature, a
cut-off frequency as high as 94 GHz is reported for this type of bipolar transistor.
The very recent record results for this type of bipolar transistor were reported by
IBM: the silicongermanium transistor hits 500 GHz. More detailed analysis of het-
erostructure bipolar transistors can be found in references presented at the end of this
chapter.
Hot-electron transistors
Another group of potential-effect devices is represented by hot-electron transistors, as
illustrated in Fig. 8.30. These devices employ an emitterbarrierbasebarriercollector
structure. The hot electrons are injected over, or through, an emitter barrier into a
M
M
M
M M
M
O
O
O
S
S
S
(a) (b) (c)
Figure 8.37 Energy-band diagrams for metal-base ballistic-injection transistors. Metal, oxide,
and semiconductor layers are marked by M, O, and S, respectively. Semiconductor layers are
doped. The diagrams are presented under operational bias conditions.
narrow base region. Injected carriers have a large velocity and transit through the
base almost ballistically; that is, they transit the base without signicant scattering.
The current through the device is controlled by changing the height of a collector bar-
rier. This principle is used in a number of different congurations and combinations of
materials.
Consider a ballistic-injection device, which is primarily a unipolar device; that is,
only one type of carrier, say electrons, is used. A ballistic-injection device also con-
sists of an emitter, base, and collector. The role of the emitter is to inject electrons
with high velocities into the base; the second electrode should collect these electrons.
The input base voltage controls the electron injection and, therefore, the output emitter
collector current. If small changes in the input produce larger changes in the output, the
device exhibits a current gain. Generally, in ballistic devices, the electrons are injected
into the base with a high energy exceeding 0.1 eV. This should lead to a decrease in
the time of ight through the base region. Another advantage of ballistic devices is
related to their unipolar character; this means that it is possible to choose the fastest
type of majority carriers (electrons) and avoid the participation of slower minority
holes.
In order to realize a transistor faster than a bipolar or eld-effect transistor, several
schemes of ballistic-injection devices have been proposed. They differ in terms of the
physical mechanisms of electron injection as well as in the materials used in the devices.
The rst and, perhaps, the simplest device is a metaloxidemetaloxidemetal het-
erostructure. In Fig. 8.37(a), this device is shown under a bias. Other similar device
structures are the metaloxidemetalsemiconductor structure and the semiconductor
metalsemiconductor (SMS) structure presented in Figs. 8.37(b) and 8.37(c), respec-
tively. The principle of operation is the same for these three structures. Consider, for
example, the case of the SMS structure. One can see a close analogy with the bipo-
lar transistor: a forward-biased semiconductormetal junction serves as the emitter, a
second metalsemiconductor junction serves as the collector, and a metal layer is the
base. Both junctions are, in fact, Schottky diodes, one is forward-biased and the other
is reverse-biased. Under such bias conditions, electrons are injected over the Schottky
barrier with energies substantially exceeding the thermal energy in the base. If the base is
narrow, the electrons y across the base region without losing their energy. Their sub-
sequent destiny is dened by the basecollector bias: a lowering of the collector barrier
increases the fraction of electrons coming into the collector electrode and, consequently,
the collector current.
Various materials have been used for the structures shown in Fig. 8.37. In particular,
the Si/CoSi
2
/Si structure was investigated for the SMS scheme. An factor as high as 0.6
was reported for this case. A common disadvantage of metal-base transistors is the high
electron reection fromthe metal. This reection is, mainly, a quantum-mechanical effect
in nature and, therefore, can not be avoided. It occurs even for ideal semiconductormetal
interfaces.
Semiconductor heterostructures may be used in other ways to realize ballistic-injection
devices. Let us consider n-type devices. In semiconductor-based ballistic-injection
devices, the emitter, base, and collector are doped regions separated by two barriers.
The barriers can be formed by growing layers of materials with a positive conduction-
band offset as shown in Fig. 8.38(a). This structure actually uses four heterojunctions.
Another kind of barrier can be produced by planar acceptor doping in a homostructure
as illustrated in Fig. 8.38(b).
Consider the typical parameters of structures for ballistic-injection devices. If the
height of the emitter barrier is V
b
, the velocity of electrons injected into the base can be
estimated as :
B

2V
b
,m
= 5.9 10
7
_
V
b
,(m
,m
0
) cms
1
, where

V
b
is the barrier
energy in eV. For GaAs we can assume

V
b
0.3 eV and m
= 0.067m
0
. For the velocity
we get :
B
= 1.3 10
8
cms
1
. This value is appreciably larger than the characteristic
electron velocity in devices such as FETs and bipolar transistors. Another important
feature is that the injected electrons exhibit velocity spreading in a very narrow velocity
cone. Actually, the average value of the lateral component of the electron momentum in
the emitter is p
[[.E

2m
k
B
T, where T is the device temperature. Owing to the lateral
translation symmetry this component does not change under electron injection through
the barrier. Thus, the characteristic angle for the velocity spreading can be evaluated as
follows:

:
[[.B
:
B
=
p
[[.B
m
:
B
=
p
[[.E
m
:
B
=
_
k
B
T
V
b
.
For example, in the case of liquid-helium temperatures and V
b
0.3 eV, this equation
gives 6
.
Collisions in the base reduce the number of ballistic electrons. If the electron mean
free path in the base is l
e
and the base width is L
B
, the fraction of ballistic electrons
collected by the collector is estimated to be exp(L
B
,l
e
). Thus, the base region
should be quite narrow. In this case, the base region has to be heavily doped to reduce
the resistance for the base current.
A limitation of the base-doping technique comes about as a result of the fact that a
high doping gives rise to additional electron scattering and quenches the ballistic regime
of electron motion. Specically, for IIIVcompounds, if impurity concentrations exceed
10
18
cm
3
electron scattering becomes very strong.
As a result, for AlGaAs/GaAs devices, base regions with doping of about 10
18
cm
3
andwithwidths of 3080nmare used. Optimizationof the structure parameters facilitates
e
BE
n
+
n
+
n
-GaAs
n
-GaAs
AlGaAs
n
-GaAs
emitter
AlGaAs
Depleted
acceptor
layers
n
+
n
-GaAs Graded-gap
Al
x
Ga
1x
As
GaAs
AlGaAs
Undoped
e
CB
(a)
(b)
(c)
collector
base
n
Figure 8.38 Energy-band diagrams for semiconductor ballistic transistors. (a) The AlGaAs/GaAs
structure with four heterojunctions and a doped base. Electrons are injected into the base due to
tunneling. This kind of device is referred to as a tunneling hot-electron-transfer amplier. (b) A
homostructure electron device. Barriers are formed by planar p-doping. The base is n-doped.
(c) A device with an undoped base. After Fig. 12, S. Luryi, Hot-electron transistors, in S. M.
Sze (Ed.), High-Speed Semiconductor Devices (New York, Wiley, 1990). Reprinted with
permission of John Wiley & Sons, Inc.
the realization of a ballistic device with a transfer ratio of = 0.9 at liquid-helium
temperature.
Figure 8.38(c) presents another design for a ballistic device. The emitter barrier is
made with a graded composition and the base region is undoped. The base is induced
by the electric eld of the collector, which leads to the formation of a two-dimensional
electron gas at the undoped interface. This type of ballistic device is referred to as the
induced-base transistor. The advantages of the induced-base transistor are the following.
The two-dimensional electron gas in the base is characterized by a high mobility and
large two-dimensional electron concentrations of up to 2 10
12
cm
2
. Both effects
cause a low base resistance; thus, the base can be chosen to be very narrow, typically
100

A. This results in a high fraction of ballistic electrons coming into the collector.
For example, a well-designed AlGaAs/GaAs induced-base transistor results in 0.96
even at room temperature. Similar induced-base transistors have been fabricated using
InGaAs/AlGaAs and p-doped Ge/SiGe heterostructures.
In conclusion, hot-electron transistors with nanoscale base regions are characterized
by a high operation speed. An electron transit time of the order of 0.10.5 ps corresponds
to a cut-off frequency in the range 210 THz.
8.6 Light-emitting diodes and lasers
So far we have studied electronic nanoscale devices, i.e., a class of devices that
exploits electrical properties of nanostructures and operates with electric input and out-
put signals. Another class is composed of optoelectronic devices, which are based on
both electrical and optical properties of materials and work with optical and electric
signals.
In this section, we will analyze two very important classes of optoelectronic devices:
light-emitting diodes and laser diodes. As their titles imply, the devices were invented
to produce light with certain properties. In particular, the energy of the electric
current owing through these diodes is transformed into light energy. These opto-
electronic devices have a huge number of applications and deserve consideration in
detail.
In Chapters 2 and 3, among various different wave elds, we introduced and studied
electromagnetic elds, of which light waves are a particular example. In addition, we
studied electrons in atoms (Chapters 2 and 3) and solids (Chapter 4). However, it was
supposed that these two physical entities do not interact with each other. In fact, light
matter interaction gives rise to fundamental physical phenomena. Historically, light
matter interactions provided some of the rst evidence for the quantum nature of matter.
Remember that an electromagnetic eld consists of an innite number of modes (waves),
each of which is characterized by a wavevector and a specic polarization. According to
quantum physics (see Chapter 3), each mode may be described in terms of a harmonic
oscillator of frequency . Correspondingly, the energy separation between levels of
this quantum-mechanical oscillator is h
--
; see Eq. (2.41). This oscillator may be in
the non-excited state, which manifests the ground-state or zero-point vibrations of the
electromagnetic eld. The oscillator may be excited to some higher energy level. If the
Nth level of the oscillator is excited, there are N quanta (photons) in the mode under
consideration. The classical description of an electromagnetic wave is valid at large
numbers of photons: N 1.
Besides quantization of the energy, the quantum nature of electromagnetic elds is
revealed in the equilibrium statistics of photons. Indeed, photons obey the so-called
hw = E
2
E
1
E
2
E
1
Figure 8.39 A two-level system interacting with light.
boson statistics, or BoseEinstein statistics, which gives the average number of photons
of some chosen mode under equilibrium in the form
N =
1
exp[h
--
,(k
B
T)] 1
. (8.26)
with T being the ambient temperature. Equation (8.26) is also known as Plancks formula.
Clearly, the BoseEinstein statistics of photons differs drastically from Fermi statistics,
which we used for electrons (see Chapter 6). In particular, it permits the accumulation
of an arbitrary number of photons in any mode. For example, from Eq. (8.26) it follows
that at k
B
T,(h
--
) 1 the phonon number N k
B
T,(h
--
) 1, whereas the number of
electrons occupying any state can not exceed 1.
After these introductory remarks on quantum properties of electromagnetic elds, we
review briey the three major processes involving a quantum material system and light:
absorption, spontaneous emission, and stimulated emission.
Photon absorption and emission
In order to visualize these processes, we consider a simple quantized system with two
energy levels E
1
and E
2
as depicted in Fig. 8.39. The occupancies of the energy levels
of this system correspond to particular states of a system of the electrons. The charged
electrons interact with the electromagnetic eld. This interaction results in transitions
between quantum states of the system. These transitions are frequently referred to as
phototransitions. According to the quantum theory, the system can change its energy as
a result of interaction with electromagnetic waves exclusively of the frequency
= (E
2
E
1
),h
--
. (8.27)
If the lowest energy level E
1
is occupied, the wave can excite the system into the upper
level E
2
and the electromagnetic energy must decrease. One can describe this process as
the absorption of one photon because the energy of the electromagnetic eld decreases
by E
2
E
1
. If the system occupies the upper level E
2
, it can make a transition to level
E
1
as a result of interaction with the electromagnetic eld. Then, the electromagnetic
energy increases by E
2
E
1
. This process represents the emission of a photon with
energy h
--
. When activated by an external electromagnetic wave, the latter process is
called stimulated emission. It is important that, for stimulated emission, each emitted
photon has the energy, direction, polarization, and even phase coinciding precisely with
those of the stimulating wave.
Both processes, absorption and stimulated emission, can be described in terms of
an interaction with a classical electromagnetic wave. The rates of these processes are
proportional to the intensity of the wave, or, according to Table 2.1, proportional to the
number of photons N
q.b
of a given wavevector q and polarization b. These rates can be
written as
R
abs
= B
12
N
q.b
n
1
. (8.28)
R
st.em
= B
21
N
q.b
n
2
. (8.29)
where n
1
and n
2
are the numbers of particles in the system occupying the levels 1 and 2,
respectively; B
12
and B
21
are kinetic coefcients describing these processes. The physical
meanings of these coefcients will be addressed in the subsequent discussion.
The two processes of absorption and emission are insufcient to describe the whole
picture of the interactionbetweenradiationandmatter. For example, let us applythese two
processes only for thermal equilibrium conditions, for which the ratio of the populations
of the two levels is
n
2
n
1
= e
(E
2
E
1
),(k
B
T)
= e
h
--
,(k
B
T)
(8.30)
(see Section 6.3 and particularly Eq. (6.14)). Using Eqs. (8.28)(8.30) one can see that
R
abs
,= R
st.em
at any temperature T. This result is in contradiction with the expected
equilibrium between the system and the eld. According to the Einstein theory, there is
an additional quantum radiative transition in the system with the spontaneous emission
of a photon of the same mode. The rate of this process is
R
sp.em
= A
21
n
2
. (8.31)
where A
21
is the coefcient or rate of spontaneous emission. The spontaneous process
does not depend on the intensity of the electromagnetic wave and takes place even in
the absence of this wave. According to quantum electrodynamics, the excited material
system spontaneously emits a photon as a result of the interaction between the electrons
and the zero-point vibrations of electromagnetic elds. The zero-point vibrations of
electromagnetic elds were introduced briey in Section 3.3.
In contrast to the case of stimulated emission, a photon produced by the spontaneous
process has an arbitrary phase. Moreover, this process produces photons with different
directions of q and polarizations, but the energy is xed, i.e., it produces photons of
different modes of the same frequency.
Now we can apply the results of Eqs. (8.28)(8.31) to thermal equilibrium. Under
equilibrium conditions, the total rate of photon emission has to be equal to the rate of
photon absorption; thus we will have
R
abs
= R
sp.em
R
st.em
. (8.32)
Using the Planck formula of Eq. (8.26) and the ratio of n
2
,n
1
of Eq. (8.30), and sub-
stituting the expressions for R
abs
. R
sp.em
, and R
st.em
into Eq. (8.32), one can nd the
relation
B
21
A
21
= (B
12
A
21
)e
h
--
,(k
B
T)
. (8.33)
Because this relation has to be satised at any arbitrary temperature T, one can obtain
two equalities:
A
21
= B
21
= B
12
. (8.34)
Thus, we have established the existence of three basic processes for the resonant interac-
tion of radiation and matter: absorption, stimulated emission, and spontaneous emission.
Moreover, we have found the relations between the coefcients determining the rates of
these processes. It is worth emphasizing that all three processes are related to interactions
with photons of the same mode.
The sum of the stimulated and spontaneous emission rates, as determined by
Eqs. (8.29) and (8.31), gives the total emission rate of photons for a xed mode:
R
em
= A
21
(1 N
q.b
)n
2
. (8.35)
From this equation, one can see that stimulated emission dominates over spontaneous
emission for a xed mode, {q. b], if the number of photons, N
q.b
, is sufciently larger
than 1. However, there is spontaneous emission of a great number of other modes with
the same frequency but with different directions of q and different polarizations b. This
total spontaneous emission can be the dominant radiative process even if stimulated
emission is the most important process for a particular mode.
Now, let us compare absorption and stimulated emission by calculating the rate of
increase of the number of photons in some xed modes:
_
dN
q.b
dt
_
st
R R
st.em
R
abs
= B
21
N
q.b
(n
2
n
1
). (8.36)
This result shows that, if
n
2
n
1
> 0. (8.37)
stimulated emission dominates over absorption. Evidently, under equilibriumconditions,
the opposite is true.
The inequality of Eq. (8.37) is the criterion for population inversion. If a popula-
tion inversion is achieved, electromagnetic waves with the resonance frequency can be
amplied when passing through the material medium. This process of amplication of
the radiation due to population inversion is the key mechanism underlying the operation
of a laser (light amplication by stimulated emission of radiation). The medium where
the population inversion occurs is often called the active medium.
Actually, Eq. (8.36) describes an increase/decrease in the photon number with
time. Amplication/absorption of light waves can be naturally described as an
increase/decrease of their intensity with spatial coordinate along the direction of their
propagation. The relation between the photon numbers and the intensities of the light
waves was introduced in Section 2.4 (see also Table 2.1). The corresponding equation
for the intensity can be derived easily from Eq. (8.36), if we consider a light pulse prop-
agating through the medium. Let the maximum of the photon number for this pulse be
characterized by the coordinate x(t ). Then, in Eq. (8.36) the derivative with respect to
time can be calculated as
d
dt
=
dx
dt
d
dx
= c
m
d
dx
.
where c
m
is the velocity of light in the medium. Instead of the photon number N
q.b
, we
introduce the light intensity I = c
m
h
--
N
q.b
,V, where V is the volume of the medium.
Now we can rewrite Eq. (8.36) in the form
dI
dx
= I. (8.38)
Here,
=
1
c
m
B
21
N
q.b
(n
2
n
1
)
represents the gain coefcient. The solution of Eq. (8.38) is
I = I
0
e
x
.
with I
0
being the intensity at a point x = 0. Thus, under population inversion the gain
coefcient, , determines the exponential increase of the light intensity. If - 0, then,
instead of amplication, absorption of light occurs.
Amedium with population inversion can be used in two ways. First, an active medium
can amplify an external light beam in accordance with the previously discussed depen-
dence of I(x). Second, an active medium can generate a light beam itself, if the proper
optical feedback is provided. Here, the optical-feedback phenomenon implies that some
portion of the energy of the light which is amplied in the active mediumis returned into
the medium for further amplication. In an active medium with optical feedback, spon-
taneous emission gives birth to the initial photons. These seed photons are amplied
due to the stimulated emission, and then they partially return to be amplied again. As
a whole, this process leads to the generation of coherent laser light.
Typically, optical feedback is realized by placing the active medium in an optical
resonator. In the simplest case, the resonator consists of plane or curved mirrors, which
provide repeated reections, and some kind of trap of the light a cavity in the region
between the mirrors. The optical waves which can be trapped in this cavity compose the
resonator modes. A universal characteristic of a resonator mode is the quality factor, Q,
which can be dened as
Q =
eld energy stored in the cavity
power dissipated in the resonator
. (8.39)
Dissipation of electromagnetic energy is caused by many factors: absorption by the
mirrors or by matter inside the cavity, transmission of light through the mirrors, light
scattering, radiation out of the resonator as a result of the diffraction of light, etc. The
power dissipated and the quality factor may be different for different modes.
In principle, in a resonator there is a large number of modes with different frequencies
and polarizations, and all possible propagation directions. Excitation of many modes
would lead to extremely incoherent emitted light. To avoid such an effect, one can
employ so-called open optical resonators. The simplest open resonator consists of two
L
Z
L
Z
(a) (b)
Figure 8.40 Longitudinal and transverse modes in the planar FabryP erot resonator.
(a) Longitudinal modes. Strictly perpendicular light rays do not escape from the resonator.
(b) Transverse modes. Slightly inclined light rays eventually escape from the resonator and have
poor quality factors.
plane mirrors parallel to each other, which are nite in their transverse dimensions; an
example is given by the so-called FabryP erot etalon shown in Fig. 8.40. In this resonator,
most of the modes propagating through the cavity are lost in a single traversal of the
cavitysince the mirrors are inclinedwithrespect tothe mode-propagationdirections. This
implies that most of the modes, so-called transverse modes, have a very lowquality factor.
Only the waves propagating perpendicular to the mirrors can be reected back and travel
fromone mirror to another without escaping fromthe resonator. These waves correspond
tothe so-calledlongitudinal modes of the resonator. Thus, for nite dimensions of mirrors
only longitudinal modes can have a high quality factor. Their loss and diffraction are
caused by absorption by the mirrors, by transmission through the mirror, and by wave
diffraction on the sides of the mirrors. The wave-diffraction losses can be made much
smaller than can those arising from other loss mechanisms. Thus, the open resonator
provides for strong discrimination between modes. Relatively few of these modes have
a high quality factor, Q. According to the denition of Eq. (8.39), they are capable of
accumulating the light energy. The photons corresponding to these high-quality modes
can be generated effectively by the active medium.
At this point, we end our discussion of the simple two-level model of the optical
medium and start to consider more realistic systems.
Interband emission and absorption in semiconductors
Semiconductors constitute a material system that may be used in practice to realize
controllable light emission. Thus, we shall consider the mechanisms of absorption and
emission of photons in semiconductors. Among these mechanisms, the most important
is interband (band-to-band) phototransitions. Absorption of a photon can result in the
creation of an electron in the conduction band and a hole in the valence band, i.e., an
electronhole pair. The inverse process is radiative electronhole recombination, which
results in the emission of a photon (see Fig. 8.41).
Figure 8.42 illustrates the dependence of the absorption coefcient on photon energy
and wavelength for various semiconductors. From Fig. 8.42, one can see that the
Absorption Emission
Figure 8.41 Absorption and emission of photons in a semiconductor: band-to-band transitions.
absorption increases sharply in the short-wavelength region. Let E
g
be the bandgap of the
semiconductor. Then the material is relatively transparent for h
--
- E
g
. For h
--
> E
g
,
the semiconductor exhibits strong absorption; accordingly, h
--
g
= E
g
corresponds to
the absorption edge. The shape of the absorption edge depends signicantly on the
structure of the electron bands. Remember that in Section 4.4 we dened direct- and
indirect-bandgap semiconductors. Direct-bandgap semiconductors such as GaAs have
a more abrupt absorption edge and a larger absorption value than do indirect-bandgap
materials, of which Si provides an example. We can introduce the so-called bandgap
wavelength, or cut-off wavelength,
g
= 2ch
--
,E
g
. If E
g
is given in eV, the bandgap
wavelength in micrometers is
g
=
1.24
E
g
. (8.40)
The values of E
g
and
g
for various IIIVsemiconductor materials are apparent fromthe
curves plotted in Fig. 8.42. One can see that interband transitions in IIIV compounds
cover a wide range frominfrared to visible spectra. Optical activity in this spectral region
is crucial for optoelectronic applications of these materials.
A photon absorbed during an interband transition excites an electron from the
valence band to the conduction band, i.e., it creates an electronhole pair as depicted in
Fig. 8.43(a). The inverse process the phototransition of an electron fromthe conduction
band to the valence band is referred to as the radiative recombination (annihilation)
of an electron and a hole; this is depicted in Figs. 8.43(b) and 8.43(c). According to the
general properties of phototransitions studied at the begining of Section 8.6, there exist
two such processes: spontaneous and stimulated emission, as illustrated by cases (b) and
(c), respectively, in Fig. 8.43.
From the above analysis it is clear that, to obtain intense interband emission, one
should provide nonequilibrium electrons and holes. Moreover, large concentrations of
Figure 8.42 Absorption coefcient versus photon energy and wavelength for interband
phototransitions in various semiconductors. After G. E. Stillman, V. Robbins et al., IIIV
compound semiconductor devices: optical detectors, IEEE Trans. Electron Devices, ED-31,
16431655 (1984).
C _
IEEE.
E
k k k
E
2
E
1
E
v
E
c
E
g
hw hw hw
hw
hw
(a) (b) (c)
Figure 8.43 (a) The absorption of a photon results in excitation of the electronhole pair.
(b) Spontaneous emission, which can be interpreted as electronhole recombination.
(c) Stimulated emission.
both electrons and holes in the same spatial region correspond to the population inversion
necessary for stimulated emission. Indeed, according to Section 4.4, the presence of the
holes means empty electron states in the valence band. This explains the appearance
of the population inversion between the conduction and valence bands under conditions
of large electron and hole concentrations.
However, nonequilibrium carriers quickly relax due to various interband relaxation
mechanisms. When the electron and hole concentrations are highly nonequilibrium, the
characteristic lifetime of these excess carriers is small. To describe the nonequilibrium
electrons and holes, we introduce the density of the pumping (excitation) rate, R
pump
,
which represents the number of electronhole pairs excited in a unit volume per unit
time. For the case when the radiation mechanism is the decay of electronhole pairs, one
can write
R
pump
= Bn
2
=
n
R
(n)
. (8.41)
where n is the concentration of electronhole pairs, B is a parameter, and
R
= 1,(Bn) is
the radiation lifetime. For the concentrations of about 10
18
cm
3
which are encountered
in practice and are necessary to realize the needed signicant population inversion and
to achieve lasing conditions, this time is less than 10 ns. The pumping rate, R
pump
,
necessary to induce such a concentration is estimated to be
R
pump
10
26
cm
3
s
1
. (8.42)
To obtain an idea of how intense this pumping is, we can perform the following esti-
mates. Suppose that the bandgap, E
g
, is about 1 eV, then the pumping rate of Eq. (8.42)
corresponding to the density of the pumping power given in the previous example is
E
g
R
pump
16 MWcm
3
. This is a huge pumping power!
At this point, an important spectral property of light amplication in semiconductor
materials will be discussed. For the previously analyzed two-level model, phototran-
sitions and light amplication/absorption were possible only at a xed photon energy
h
--
= E
2
E
1
. This implies that in such a two-level systemthe amplication and absorp-
tion occur in a very narrowspectral range near the value (E
2
E
1
),h
--
. On the contrary, in
semiconductors, the spectral range within which amplication and absorption are possi-
ble is restricted only frombelow, > E
g
,h
--
. This results in a wide spectral band of optical
activity. In Fig. 8.44, we show the spectral dependences of the gain/loss coefcient for
GaAs for various electronhole concentrations at room temperature. The left boundary
of these dependences coincides approximately with the bandgap value 1.42 eV. The spec-
tral range for which the gain coefcient, , is positive becomes wider and its maximum
moves to larger photon energies as the electron/hole concentration increases because the
number of inversely populated states increases with the concentration. Amplication in a
wide spectral range allows one to build a laser with considerable tuning of the frequency.
The following two methods can be applied to generate large nonequilibrium carrier
concentrations. The rst method is optical excitation or pumping, whereby external light,
often incoherent, with a photon energy larger than the bandgap is absorbed and creates
nonequilibrium electronhole pairs. Then, because of the short time associated with the
intraband relaxation, the electrons and holes relax to the bottom of the conduction band
and the top of the valence band, respectively, where they are accumulated. If the rate of
optical pumping is sufciently large, the necessary level of interband emission can be
reached. In particular, inversion between the bands can be induced and light amplication
and laser generation become possible. The optical pumping corresponds to a conversion
of one kind of radiation, not necessarily coherent, into coherent radiation with a lower
photon frequency. One employs optical pumping in cases when electric-current pumping
either is not possible or is ineffective. This pumping method is often used to test prototype
laser structures before the design of the current pumping system.
Photon Energy (eV)
1.42 1.46
1.50
80
40
0
40
80
120
160
1
8
c
m
1
8
c
m
(
c
m
1
)
Figure 8.44 The amplication coefcient versus the photon energy in a GaAs light amplier for
ve different carrier concentrations. The temperature is 300 K. The results are given for ve
values of the concentration with equal steps of 0.25 10
18
cm
3
. After J. Singh, Semiconductor
Devices: An Introduction (New York, McGraw-Hill, 1994), Fig. 11.16.
Laser diodes
Amuch more convenient method that can be applied to achieve the interband emission is
electron and hole injection in devices with pn junctions. The physics of a pn junction
was studied in the previous section. There, we found that, under forward bias of a pn
junction, it is possible to inject the electrons from the n part of the junction and the holes
from the p part of the junction into the same spatial region. Such a double injection can
give rise to large nonequilibrium electron and hole concentrations. Often, the n and p
parts are separated by a narrowintrinsic (undoped) i region. The corresponding structure
is called a pin device.
In Fig. 8.45, we show the case of forward biasing of a pin structure the so-called
at-band condition, when there are no potential barriers for the electrons and holes,
and the maximum possible carrier injection into the i region is realized. In the case of
direct-bandgap semiconductors, double injection provides for the intense emission of
light.
It is easy to estimate the electric currents which are necessary to obtain a given
concentration of carriers under double injection. We dene the active region, where both
nonequilibrium electrons and holes are present. In the case of a pin structure, the
active region coincides with the i region. Let the length of this region be l. We introduce
+h
e
p
+
n
+
i
E
c
E
v
Figure 8.45 The double injection of the electrons and holes, which leads to the formation of
interband population inversion in an active region.
a cross-sectional area A, through which the current I is injected into the pn region of a
diode. Then, at steady state, the rate of injection of electrons and holes into a unit volume
per unit time can be expressed as
R =
I
eAl
=
J
el
. (8.43)
where J = I ,A is the injection current density. Injection leads to the accumulation of
nonequilibrium electronhole pairs with concentration
n = R =

el
J. (8.44)
where is the total lifetime of the nonequilibrium pairs in the active region. In general,
due to the small electron and hole lifetimes, the active region is quite narrow. For GaAs,
the length of the active region is estimated to be from 1 m to 3 m, depending on
temperature, crystal quality, etc. Now, we can use Eq. (8.43) to estimate the current
density necessary to achieve the pumping rate of Eq. (8.42): J (1.64.8) 10
3
Acm
2
.
The current densities obtained are very large. These estimates indicate that, in order to
operate with acceptable current levels, the cross-sections of real emitting diodes have to
be very small.
From Eqs. (8.43) and (8.44), it follows that another possible way of decreasing the
pumping electric current is to reduce the thickness of the active region l. This thickness is
one of the critical parameters for injection pumping of light-emitting devices. Indeed, as
understood from the previous discussion and Fig. 8.45, an excess carrier concentration
in the active region always leads to carrier diffusion out of this region: the electrons
diffuse through the active region to the p part of device, while the holes diffuse to the n
part. In general, diffusive leakage of the carriers is limited by their nite lifetime (n).
In Section 6.2 we found that, if the diffusion coefcient D is given, the diffusion length
during a time can be estimated as L
D
=
D. In the case under consideration, is the

excess carrier lifetime and the length L
D
corresponds to the average distance of electron
(hole) transfer before recombination. Obviously, the minority carriers (the electrons in
the p part and the holes in the n part) recombine immediately with the majority carriers
(the holes in the p part and the electrons in the n part). Thus, the width of the region with
the excess carrier concentration can not be less than the diffusion length L
D
, i.e., d > L
D
.
Since the diffusion length of electrons and holes in direct-bandgap materials is of the
order of a few micrometers, it is not possible to make the active region shorter than a
E
g1
h
e
0
E
g2
E
g3
Barrier
E
F
E
F
E
Figure 8.46 The energy-band diagram of a double heterostructure for light amplication/
generation. The applied voltage +
0
induces the at band conditions; E
g1
. E
g2
. and E
g3
are the
bandgaps in different regions of the structure. The energy states lled by the electrons are
denoted by the shaded areas. Thus, in some spatial region there are electrons in the conduction
band and depopulated states (holes) in the valence band.
fewmicrometers. This conclusion is valid for homostructure (bulk-like) semiconductors.
Heterostructure technologies open the way to different solutions to this problem.
In order to localize nonequilibrium electrons and holes in a smaller active region, one
can employ two heterojunctions. The basic idea of using a double heterostructure is to
design potential barriers on both sides of the pn junction this prevents electrons and
holes from diffusing. The potential prole of a double heterostructure is sketched in
Fig. 8.46. The heterostructure consists of three materials with bandgaps E
g1
, E
g2
, and
E
g3
. The band offsets are chosen appropriately to design a structure with a barrier for
electrons in the left part of the structure before the p region and a barrier for holes in the
right part before the n region. The middle i region with the bandgap E
g2
is accessible
by both types of carriers and it serves as the active region. Figure 8.46 corresponds
to the case of the at-band condition of a pn junction with a double heterostructure
embedded in the depletion region. In this case, it is not the diffusion length but the
distance between the barriers that determines the size of the active region. As a result,
the size can be as small as 0.1 m and the critical electric current is smaller by one
order of magnitude or more compared with that of a conventional homostructure pn
junction.
If the thickness of the double heterostructure is decreasedfurther, the inuence of quan-
tum effects on carrier motion becomes important. For the structure shown in Fig. 8.46,
quantum effects do not lead to any advantages because the advantages can be obtained if
heterostructures are designed so that quantum connement applies both to electrons and
to holes. The simplest case of quantum connement can be achieved if a quantum-well
layer is embedded in an active region of a type-I heterostructure. Three possible designs
of active regions exhibiting quantum connement are sketched in Fig. 8.47. Cases (a)
and (b) correspond to the resultant connement in a single quantum well, while case
(c) corresponds to connement in multiple quantum wells. For these designs, electrons
x
i
n

(
A
l
x
G
a
1
x
)
0
.
5
I
n
0
.
5
P
(a) (b) (c)
Figure 8.47 Composition proles of (Al
x
Ga
1x
)
0.5
In
0.5
P heterostructures providing
simultaneously quantum connement of the carriers and optical connement: (a) a single
quantum well and a step-like refractive-index heterostructure, (b) a single quantum well and a
graded-index optical connement structure, and (c) a multiple quantum well and a step-like
index structure. After P. S. Zory Jr., Quantum Well Lasers (Boston, MA, Academic, 1993).
InGaAsP
= 1.3 m
SQW L
z
= 50
GaAs
200
100
0
0 20 40 60
a
0
(
c
m
1
)
j (A cm
2
)
Figure 8.48 The amplication coefcient
0
as a function of the pumping current density for
single InGaAsP and GaAs quantum wells of width 50

A. After P. S. Zory Jr., Quantum Well
Lasers (Boston, MA, Academic, 1993).
and holes that are either generated by external light or injected from p and n regions
move in barrier layers and then are captured in the active region and quantum wells. The
characteristic time of this capture is less than 1 ps. Escape processes require additional
energy and have low relative probabilities of occurrence. Carriers in the quantum wells
relax to the lowest energy states available. This results in the accumulation of both types
of carriers in an extremely narrow active region, which is typically 100

Awide or even
narrower. A similar situation can be realized if quantum wires or dots are embedded in
the active region.
A positive effect of electron connement in quantum wells is illustrated in Fig. 8.48.
In this gure, the results for the gain factor
0
as a function of the current density J
are presented for two material systems: InGaAsP/InP and AlGaAs/GaAs heterostruc-
tures. It can be seen that the gain coefcient always increases with the current. The
current for which
0
becomes positive, i.e., a population inversion is established, can
be dened as the threshold current. The gain coefcient is shown for heterostruc-
tures with 50-

Aquantum wells. This system has low values of threshold currents for
both heterostructures: 20 Acm
2
and 60 Acm
2
for InGaAs/InP and AlGaAs/GaAs,
p n
d
L
x
L
y
Cleaved
surface
Cleaved
surface
Figure 8.49 The scheme of an injection laser with two cleaved facets that act as reectors. After
B. E. Saleh and M. C. Teich, Fundamentals of Photonics (New York, Wiley, 1991). Reprinted
with permission of John Wiley & Sons, Inc.
respectively. For both material systems, the
0
(J) curves increase sharply with the cur-
rent density above the threshold. As discussed previously, lasing can be achieved if a
light amplier is supplied with a path for optical feedback. For an injection laser, the
feedback is usually obtained by cleaving the crystal planes normal to the plane of the
pn junction. Figure 8.49 depicts a device with two cleaved surfaces forming an optical
resonator. For the light reected from the crystal boundaries, we dene the reection
coefcient by
r =
I
r
I
in
.
where I
in
and I
r
are the intensities of incident and reected light, respectively. The
reection coefcient for an airsemiconductor boundary is
r =
_
n
ri
1
n
ri
1
_
2
.
where n
ri
is the refractive index of the semiconductor material. Since semiconductors
usually have large refractive indexes the coefcients r are large enough. The intensity of
the light transmitted through this mirror is
I
out
= (1 r)I
in
.
Let two cleaved surfaces be characterized by two coefcients, r
1
and r
2
. After two passes
through the device, the light intensity is attenuated by the factor r
1
r
2
. We can dene
an effective overall distributed coefcient of optical losses:
r

1
2L
x
ln
_
1
r
1
r
2
_
.
Here, L
x
is the distance between the cleaved surfaces. In principle, there can be other
sources of optical losses in the resonator. Let them be characterized by the absorption
1
2
3
0 2
4 6 8 10 12
Injection current (mA)
O
u
t
p
u
t

p
o
w
e
r

(
m
W
)
(b)
(a)
r ~ 0.8
r ~ 0.3
Figure 8.50 The laser output versus the pumping current for a GaAs single-quantum-well laser.
After B. E. Saleh and M. C. Teich, Fundamentals of Photonics (New York, Wiley, 1991).
Reprinted with permission of John Wiley & Sons, Inc.
coefcient,
s
. Then, the total loss coefcient is
threshold
=
r
s
.
If is the gain coefcient of some light mode in this resonator, we can write the criterion
for laser oscillations as

threshold
=
r
s
. (8.45)
For injection lasers, the criterion of Eq. (8.45) is a condition imposed on the magnitude
of the injection current density, J.
In order to have an idea of the order of magnitude of optical losses in laser diodes, we
consider an example. For light of visible and near-infrared ranges, the typical refractive
index can be estimated as n
ri
3.23.5. Thus, the reection coefcient is r 0.3. Let a
diode have the geometrypresentedinFig. 8.49withdimensions suchthat d L
x
L
y
=
1 m 200 m 200 m. Then, we can estimate the radiation losses:
r
60 cm
1
.
The total current through the laser diode is 12 A. The threshold currents for lasing
for the quantum-well-based structures are considerably smaller and are typically of the
order of tens of mA. In Fig. 8.50, the output light power is presented as a function of the
injection current for an AlGaAs laser with a single 100-

Aquantum well embedded in
the active region. This particular laser design has a threshold current of about 8 mA for
the optical feedback due to the cleaved (uncoated) end facets (the reection coefcient
is r 0.3). The light reection can be improved through the use of special reecting
coatings on the end facets. In the latter case, the threshold current decreases below1 mA,
as shown in Fig. 8.50 for the reection coefcient r 0.8. The same gure illustrates
Light output
metallization
multilayer
active region
implanted region
/n
110 m
Figure 8.51 Surface-emitting lasers.
also a typical output light power in the injection lasers: from 1 mW to tens of milliwatts
depending on the particular diode and resonator design.
Surface-emitting lasers
Thus far, in considering quantum-well lasers we have assumed that the laser is designed
for light propagation along the quantum-well layers. Another possible design uses
light propagating perpendicular to the layers in a so-called vertical geometry. The
amplication of light passing through a quantum-well layer can be dened as
I
out
I
in
I
in
.
Here, I
in
and I
out
are the input and output light intensities, respectively. The quantity
is estimated through the gain coefcient as () ()L, where L is the width
of the quantum well. It is easy to see that is typically very small. For example, if
0
= 100 cm
1
and L = 100

A, we get = 10
4
. In order to obtain laser oscillations in
a vertical-geometry structure, one should employ a multiple-quantum-well structure and
provide near-perfect mirrors with extremely high reectivities. Figure 8.51(a) depicts
schematically such a surface-emitting laser. The laser design includes an active region
providing for high light gain, dielectric multilayers, metallic contacts, and implanted
regions, which form the light output. Layered dielectric mirrors give very high reection
while the active region contains a multiple-quantum-well structure. The lateral sizes
of this laser can be reduced to the 110-m range. A decrease in the surface area of
the diode leads to a considerable decrease in the magnitude of the threshold current.
In the case of the quantum-well structures just considered, we can assume a charac-
teristic current density of about 100 Acm
2
; accordingly, if the lateral sizes are each
5 m, the pumping surface area is 2.5 10
7
cm
2
and the threshold current equals
25 A.
Surface-emitting quantum-well lasers offer a new advantage of high-packing density.
Nowadays, technology allows one to fabricate an array of about 10
6
surface-emitting
0.1 0.5 1 5 10 50 100
Wavelength (m)
(Al
x
Ga
l x
)
y
In
l y
P
Cd
x
Zn
l x
S
CdS
x
Se
l x
In
x
Ga
l x
As
GaAl
l x
P
x
Al
x
Ga
l x
As
(Al
x
Ga
l x`
)
y
In
l y
As
InAs
x
P
1x
GaAs
x
Sb
l x
In
l x
Ga
x
As
y
P
l y
Cd
x
Pb
1-x
S
Cd
x
Hg
l x
Te
InAs
x
Sb
l x
PbS
l x
Se
x
Pb
x
Sn
l x
Te
Pb
x
Sn
l x
Se
Bi
l x
Sb
x
Figure 8.52 Spectral regions of lasers with various semiconductor media. After P. L. Derry, A.
Yariv et al., Ultralow-threshold graded-index separate-connement single quantum well buried
heterostructure (Al, Ga)As lasers with high reectivity coatings, Appl. Phys. Lett., 50,
17731775 (1987). Reused with permission from Pamela L. Derry, Amnon Yariv, Kam Y. Lau,
Nadav Bar-Chaim, Kevin Lee, and Jan Rosenberg, Applied Physics Letters, 50, 1773 (1987).
C _
electrically pumped microlasers. Microlasers can operate at room temperature and
threshold currents are below 0.1 mA.
Blue and ultraviolet quantum-well lasers
As we emphasized previously, the wavelength and the energy quanta of light generated
by a semiconductor laser basically depend on the energy bandgap of the semiconductor
material selected. In Fig. 8.52, the spectral characteristics of the lasers from various
semiconductors are shown. For example, one can see that the most developed lasers
based on IIIV compounds cover the wavelength range from approximately 0.6 m to
3 m.
Operation in the very interesting range of shorter wavelengths requires the use of
semiconductors with larger energy bandgaps. Well-known examples of such wide-
bandgap semiconductors are representative IIVI compounds: CdS, CdSe, ZnCdSe,
etc. By using different quantum-well designs, the blue injection lasers were realized
on the basis of these materials. However, the major problems associated with this class
of quantum-well lasers are the device degradation and short lifetimes of the devices.
The electrical power dissipated in IIVI diodes is too high and causes rapid degra-
dation as a result of the generation of intrinsic defects. Much improvement must be
realized in this eld in order for the group IIVI-based devices to have wide practical
application.
Recently, another class of heterostructure materials has been studied in detail the
nitrides of group III. It is difcult to obtain these materials in the usual single-crystal
form. However, thin nanometer-scale layers can be grown successfully with good quality
by applying the growth methods presented in Chapter 5. The nitrides of group III include
InN, GaN, AlN, and all possible ternary and quaternary alloys. For InN and GaN, the
energy bandgaps are about 0.9 eV and 3.5 eV, respectively, while that for AlN is 6 eV.
Thus, by changing the In and Al content in the alloys In
y
Al
x
Ga
1yx
N, one can increase
the bandgap from about 1 eV to 6 eV and realize lasers with wavelengths spanning the
range from 1.3 m (near-infrared light) to 0.2 m (deep-ultraviolet light). The alloys
In
x
Ga
1x
N produce green, blue, and violet light.
Light-emitting diodes
Althoughstimulatedemissionfromthe injectionlaser diode is veryimportant, practically,
sub-threshold operation of the diode when only spontaneous light is emitted is in
many cases advantageous and has a number of applications. This mode of operation
does not require feedback to control the power output, which facilitates operation over
a wide range of temperatures, and is reliable and inexpensive. Diodes operating with
spontaneous light emission are called light-emitting diodes.
The important characteristic of the light-emitting diode is the spectral distribution
of emission. The spectrum of emission is determined, primarily, by the electron/hole
distributions over energy, which can be approximately described by corresponding Fermi
functions, as studied in Chapter 6. Thus, the ambient temperature, T, denes both the
spectral maximum (the peak) and the spectral width of emission. The peak value of
the spectral distribution can be estimated as
h
--
m
= E
g
1
2
k
B
T.
The full width at half maximum of the distribution is L 2k
B
T,h
--
and is independent
of . In terms of the wavelength, , we obtain
L =
_
2
m
,(2c)
_
L.
or
L = 1.45
2
m
k
B
T. (8.46)
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3
Wavelength (m)
Near infrared
Violet
GaN
Green
GaP:N
Yellow
GaAs
0.14
P
0.86
Orange
GaAs
0.35
P
0.65
Red
GaP:ZnO
GaAs
0.6
P
0.4
GaAs
In
0.83
Ga
0.17
As
0.34
P
0.66
In
0.72
Ga
0.28
As
0.60
P
0.40
Figure 8.53 The spectra of light-emitting semiconductor diodes with different bandgaps. After
Fig. 13 on page 696 of S. M. Sze, Physics of Semiconductor Devices (New York, Wiley, 1981).
Reprinted with permission of John Wiley & Sons, Inc.
where
m
corresponds to the maximum of the spectral distribution, L and
m
are
expressed in micrometers, and k
B
T is expressed in eV. Figure 8.53 shows the spectral
density as a function of the wavelength for light-emitting diodes based on various mate-
rials. The spectral density is normalized so that its maximum equals 1 for all samples.
For these different materials, the spectral linewidths increase in proportion to
2
, in
accordance with Eq. (8.46). From Fig. 8.53, one can see that light-emitting diodes cover
a wide spectral region from the infrared about 8 m for InGaAsP alloys to the near
ultraviolet 0.4 m for GaN. They are, indeed, very universal light sources.
Similarly to the case of lasers, the parameters of light-emitting diodes may be consid-
erably enhanced by using heterostructures, particularly quantum wells. Light-emitting
diodes may be also designed either in a surface-emitting conguration, or in an edge-
emitting conguration. These congurations are illustrated in Figs. 8.54 and 8.55, respec-
tively. Surface-emitting diodes radiate from the face parallel to the pn junction plane.
The light emitted in the opposite direction is either absorbed by a substrate, or reected
by metallic contacts. The edge-emitting diodes radiate from the edge of the junction
region. Usually, surface-emitting diodes are more efcient.
Since the diodes under consideration radiate through spontaneous emission, the spatial
patterns of the emitted light depend only on the geometries of the devices. Different
lenses can improve the emission pattern. Usually, the edge-emitting diodes have narrower
emission patterns.
Light-emitting diodes nd many applications, ranging fromcommon lighting systems
to signal processing and light communications.
Unipolar intersubband quantum-cascade lasers
So far in this chapter, we have considered light emission and laser action based on
interband phototransitions involving both electrons and holes. Another type of photo
Figure 8.54 A surface-emitting diode.
Figure 8.55 An edge-emitting diode.
transition intraband absorption is allowed in an ideal crystal system with heterojunc-
tions. Since the latter phototransitions are drastically different frominterband transitions,
laser action associated with these intraband transitions should differ in a fundamental
way from that studied for the laser schemes considered above. First of all, an intraband-
transition laser should employ only one type of carrier, i.e., it is a unipolar device.
Second, it should be based on electron transitions between conned states arising from
the quantization in semiconductor heterostructures. In order to create a population inver-
sion between two conned states, one needs to provide for (i) electron injection into a
higher lasing state and (ii) depletion of a lower lasing state. For this purpose, a vertical
scheme of electron transport has been proposed. This scheme is illustrated in Fig. 8.56(a).
The proposed heterostructure is a superlattice with a complex design for each period.
Each of the periods consists of four AlInAs barriers, forming three GaInAs quantum
e
3
e
2
e
1
hw
hw
hw
E
Fn
1
2
3
(b)
(a)
Figure 8.56 (a) Two periods of the 25-stage staircase coupled-well region of a quantum-cascade
laser under operational conditions. The laser phototransitions are indicated by arrows. They
occur between levels (subbands) 3 and 2 with the photon energy of 295 meV. Level 2
depopulates through level 1 and subsequent tunneling. The energy separation between levels 2
and 1 is 30 meV. (b) Energy dispersion for subbands 1, 2, and 3, phototransitions, and
intersubband-scattering processes (straight lines). Reprinted with permission from J. Faist,
F. Capasso et al., Quantum cascade laser, Science, 264, 553556 (1994).
C _
1994 AAAS.
wells, and a graded AlInGaAs region, which is doped. Under zero-bias conditions, the
overall band diagram resembles a sawtooth structure. Under an applied electric eld,
the band diagram takes on a staircase structure as shown in Fig. 8.56(a). The barriers
form three coupled quantum wells with three quasi-bound levels. These three levels are
labeled in Fig. 8.56(a) by 1, 2, and 3. Each of the conned states originates from one
of the wells. The structure is chosen so that there is a considerable overlap between the
wavefunctions of the upper state, 3, and the intermediate state, 2. The same holds for
wavefunctions of states 2 and 1. Under a voltage bias, the potential in the doped regions
is almost at, as shown in Fig. 8.56(a). The electrons are injected fromthe doped regions
through the barrier in the conned state 3 of the rst quantum well. From this state, they
relax primarily to state 2. There are two processes of relaxation: phonon emission and
photon emission. In Fig. 8.56(b), these processes are shown for electrons with various
values of the in-plane wavevector, k. The three indicated subbands,
1.2.3
(k), correspond
to the three conned states. The straight arrows represent intersubband phonon relax-
ation. The third conned state, 3, is selected to provide depletion of state 2 as fast as
possible. Thus, in this manner we have a three-level scheme whereby the upper level is
pumped by the direct injection of electrons from the doped region. The second level is
depleted due to strong coupling with the lowest level 1. From level 1, electrons escape to
the next doped region. Then the processes are repeated in each subsequent period of the
superlattice. One can say that the carriers make transitions down through such a cascade
structure.
To describe the processes in the cascade structure we need to dene the numbers of
electrons in three states: n
3
, n
2
, and n
1
. The criterion for population inversion between
levels 2 and 3 should be
n
3
> n
2
.
We can write simple balance equations for n
3
and n
2
:
dn
3
dt
=
1
e
J
n
3
32
. (8.47)
dn
2
dt
=
n
3
32
n
2
21
. (8.48)
where J is the density of the injection current, and
32
and
21
are relaxation times
between the states 3 and 2, and 2 and 1, respectively. In Eq. (8.48), we neglect the
inverse 1 2 process since state 1 can be regarded as almost empty as a result of
fast electron escape to the doped region. For the steady-state conditions, we obtain the
concentrations
n
3
=
1
e
J
32
and n
2
= n
3
21
32
.
and population inversion
Ln n
3
n
2
=
1
e
J
32
_
1

21
32
_
. (8.49)
Thus, to create a population inversion, one should design the laser so that
21
-
32
. (8.50)
In order to fabricate such unipolar laser structures with vertical electron transport, very
precise sophisticated semiconductor technology is necessary.
For a particular device structure (Fig. 8.57) with an optical path of about 700 m and
the mirror reectivity r
1
= r
2
= 0.27, the laser output-current characteristics at various
temperatures are shown in Fig. 8.58. The insets of Fig. 8.58 show the currentvoltage
characteristics and temperature dependence of the laser threshold current. The laser
threshold current can be approximated by I
th
= C exp(T,112), where the constant C is
about 900 mA and T is measured in degrees Kelvin. From Fig. 8.58, it follows that the
output power reaches tens of milliwatts.
The emission energy is in the range 275310 meV. Spectra of the laser output for
various currents at T = 80 K are presented in the right-hand inset of Fig. 8.58. For
this case, the threshold current is about 1.06 A. This inset clearly demonstrates a sharp
narrowing of the emission spectra above the laser threshold: the spectra reduce to a sharp
peak at I = 1.1 A > I
th
.
Thus, the unipolar cascade laser is drastically different from the lasers based on
intersubband phototransitions. The properties of intraband phototransitions and, con-
sequently, the properties of the unipolar laser are determined to a large degree by
quantum connement; accordingly, this novel laser can be tailored for operation in
GaInAs (Sn-doped) n = 2.010
20
cm
3
20.0 nm
GaInAs 1.0 10
18
670.0
AlGaInAs (graded) 1.010
18
30.0
AlInAs 5.010
17
1500.0
AlInAs 1.510
17
1000.0
AlGaInAs (digitally graded) 1.510
17
18.6
Active region undoped 21.1
GaInAs 1.010
17
300.0
17
14.6
17
18.6
Active region undoped 21.1
GaInAs 1.0 10
17
300.0
17
33.2
AlInAs 5.010
17
500.0
Doped n
+
-InP substrate
W
a
v
e
g
u
i
d
e
c
l
a
d
d
i
n
g
C
o
n
t
a
c
t
l
a
y
e
r
W
a
v
e
g
u
i
d
e

c
o
r
e
W
a
v
e
g
u
i
d
e
c
l
a
d
d
i
n
g
25
Figure 8.57 A schematic cross-section of the cascade laser structure. The whole structure
consists of 500 layers. Reprinted with permission from F. Capasso, Quantum cascade laser,
Science, 264, 553556 (1994).
C _
1994 AAAS.
the spectral region from the middle infrared to submillimeter waves. Quantum cas-
cade lasers are very sensitive to the ambient temperature and typically work at reduced
temperatures.
In conclusion, we have shown that nanostructures play a key role not only for down-
scaled electrical devices, but also for optical devices. They facilitate improvements in
bipolar injection lasers and make it possible to realize cascade-laser structures, which
work on the basis of innovative concepts.
8.7 Nanoelectromechanical system devices
So far, we have concentrated on the electron properties of nanostructures and have
shown that electronic effects on the nanoscale can be exploited for electrical devices.
Mechanical properties of nanostructures are very different from those of bulk samples.
One can use both electronic and mechanical properties on the nanoscale to develop a new
class of devices nanoelectromechanical devices. Fabrication of nano-electromechanical
Figure 8.58 The measured optical power P from a single facet of the quantum-cascade laser with
the structure presented in Fig. 8.57 and an optical cavity length of 1.2 mm. The results are given
for four different temperatures. The insets on the left show the dependence of the laser threshold
current as a function of temperature and the currentvoltage characteristics of the device. The
inset on the right of the gure shows the output spectrum for a 1.2-mm-long laser below and
above threshold (I
th
= 1.06 A) at 80 K heat-sink temperature. Above 0.4 A the luminescence
peak was found to grow superlinearly with current due to optical gain causing the line to narrow.
After F. Capasso, Quantum cascade lasers: a unipolar intersubband semiconductor laser, in
Proceedings of the International Conference on the Physics of Semiconductors (Singapore,
World Scientic, 1995), pp. 16361640.
systems (NEMSs) was described in Section 5.9. Here, we will discuss a few particular
NEMS devices.
Resonators. Parametric amplication
Parametric resonator NEMSs serve as mechanical ampliers ina narrowfrequencyrange.
The basic idea of a mechanical parametric resonator can be illustrated by the following
simple example. Consider a simple parallel-plate capacitor, in which one plate is a part
of the mechanical resonator, while the second plate is xed, as shown in Fig. 8.59. A
displacement, z, of the resonator plate changes the spacing in the capacitor and thus the
device capacitance, C = c
0
S,(d z), where S and d are the area of the plates and the
equilibrium distance between them, respectively. Since the displacements are small, we
can use the approximation
C
c
0
S
d
_
1
z
d

z
2
d
2

_
.
If a time-dependent voltage V(t ) is applied to the capacitor, its electrical energy becomes
E
el
=
1
2
CV
2
(t ).
Figure 8.59 A mechanical parametric resonator fabricated by the SIMOX technique.
Characteristic dimensions of the resonator are indicated. From A. N. Cleland, Foundations of
Nanomechanics (Berlin, Springer-Verlag, 2003).
Let M and K be the mass and the spring constant of our resonator, respectively. Then,
Newtons second law for the resonator plate takes the form
M
d
2
z
dt
2
M
dz
dt
Kz = f =
dE
el
dz
. (8.51)
Here, we introduce the term M dz,dt describing the damping of the resonator. The
force acting on the resonator is
f =
c
0
S
d
2
V
2
(t )
_
1
2z
d
_
; (8.52)
i.e., it is proportional to the square of the driving voltage, V(t ). Importantly, the force
contains a term proportional to the displacement z. We can characterize the resonator by
the frequency of vibration,
0
=
K,M, and the quality factor, Q =

0
, . Introducing
the notation
LK(t ) =
2c
0
SV
2
(t )
d
3
(8.53)
and
F
D
(t ) =
c
0
SV
2
(t )
d
2
. (8.54)
we can rewrite the force equation as
d
2
z
dt
2

0
Q
dz
dt

_
2
0
LK(t )
M
_
z = F
D
(t ). (8.55)
The physical signicance of each of the quantities LK(t ) and F
D
(t ) is obvious: LK(t )
describes a parametric modulation of the spring constant by an applied voltage, while
F
D
(t ) is the displacement-independent driving force. The latter equation is the simplest
basic equation for the effect of parametric resonance: the parameters LK and F
D
have the
same time dependence; i.e., they are both proportional to V
2
(t ). If the voltage depends on
time harmonically, V sin(t ), we nd that both the driving force and the spring con-
stant oscillate with the driving frequency, which varies with time as cos(2t ). In general,
a NEMS can be fabricated with two or more capacitor plates; then, different voltages
can be applied to these capacitors. Thus, to generalize the analysis we may assume
that the spring-constant modulation, LK, and the driving force, F
D
, have independent
modulation frequencies,
P
and
D
:
LK = LK
0
sin(
P
t );
F
D
= F
0
cos(
D
t ).
Here, denotes a phase shift between these two dependences. More specically, the
basic equation now becomes:
d
2
z
dt
2

0
Q
dz
dt

_
2
0
LK
0
M
sin(
P
t )
_
z = F
0
cos(
D
t ). (8.56)
This equation is the so-called Mathieu equation with damping. Analysis of this
equation yields the following results. First, let the parametric modulation be absent
(LK
0
= 0), then the equation describes vibrations of a resonator with frequency
0
.
For small damping (Q 1), vibrations will manifest a strong resonance at the driving
frequency,
D

0
:
z(t ) = A cos(
0
t ) B sin(
0
t ). (8.57)
A =
F
0
Q
K
sin . B =
F
0
Q
K
cos . (8.58)
This result shows that, for an oscillator with a quality factor Q , the vibration
magnitude diverges.
If a parametric modulation is tuned by a small driving force, additional strong reso-
nances arise at
P
= 2
0
,n, with n being an integer. That is, the parametric modulation
resonates for all submultiples of the frequency 2
0
. Consider, for example, the case of
n = 1, i.e.,
P
= 2
0
and
D
=
0
. In the limit of high Q, it is possible to nd a solution
in the form of Eq. (8.58) with
A
p
=
F
0
Q
K
sin
1 Q LK
0
,(2K)
. B
p
=
F
0
Q
K
cos
1 Q LK
0
,(2K)
. (8.59)
Thus, the oscillations depend essentially on the parametric coupling parameter LK
0
,
and the amplitudes of the mechanical vibrations can be controlled by these parameters.
By comparing these results with and without parametric coupling, i.e., Eqs. (8.59) and
(8.58), we can dene the parametric gain,
G =
_
A
2
p
B
2
p
A
2
B
2
.
Using the above formulas for the coefcients A. B. A
p
, and B
p
, we nd
G =
_
cos
2
[1 Q LK
0
,(2K)]
2

sin
2
[1 Q LK
0
,(2K)]
2
_
1,2
.
0.01 0.1 1
0.1
1
10

j = p/2
j = 0
Q K
0
/(2K)
G
Figure 8.60 The gain/loss coefcient G as a function of the quality factor Q for different phase
shifts between spring-constant modulation and the driving force. From A. N. Cleland,
Foundations of Nanomechanics (Berlin, Springer-Verlag, 2003).
For LK ,= 0, the gain depends on the relative phase . If = 0. . 2. . . .. the ampli-
tudes of the vibrations are suppressed and G - 1. If = ,2. 3,2. 5,2. . . .. the
vibrations are amplied and G > 1. Formally, if LK
0
2K,Q, the amplitude diverges
with G . This phenomenon is called parametric resonance. Both de-amplication
and amplication regimes are illustrated by Fig. 8.60. We can conclude that, if the quality
factor, Q, is large, an amplication of the amplitude of vibrations can be obtained even
for small modulations of the driving force.
The resonator of a micromechanical parametric oscillator, shown in Fig. 8.59, is
fabricated by the SIMOX technique. The dimensions of the suspended resonator plate
are 4 m 4 m 0.2 m. The substrate is grounded. A metal is deposited onto
the resonator plate. The resonator is driven by a voltage applied between the plate
and the substrate. The fundamental frequency of the resonator is
0
,(2) = 485 kHz.
Thus, the amplied signal is at this frequency, while the parametric drive is at
P
= 2
0
.
These parameters correspond to the lowest resonance in the Mathieu equation (8.56). To
detect and measure displacements, one can use reection of a laser beam from the
substrate and the resonator. Interference of these two reected signals yields good dis-
placement sensitivity. In Fig. 8.61, the measured square of the oscillation amplitude is
presented as a function of the pump amplitude for such a parametric oscillator. The
maximum amplication, G, achieved is about 10.
The resonance properties of NEMSs will undoubtedly be employed in a broad range of
applications. Obviously, one of the principal areas will be signal processing in the very-
high-frequency (VHF), ultra-high-frequency (UHF), and microwave-frequency bands.
Mechanically detected magnetic resonance imaging
Another promising application of NEMSs is mechanically detected magnetic resonance
imaging (MRI). It is well known that the phenomenon of nuclear magnetic resonance is
widely used for diagnostic purposes in medicine. The conventional inductive detection
0.00 0.05 0.10 0.15 0.20
0.1
1
10

Pump amplitude (V)
P
o
w
e
r

(
a
r
b
.

u
n
i
t
s
)
- - -
response
fit
Figure 8.61 The square of the oscillation amplitude (the power of oscillations) versus the
pumping amplitude for the parametric resonance oscillator shown in Fig. 8.59. From A. N.
Cleland, Foundations of Nanomechanics (Berlin, Springer-Verlag, 2003).
displacement sensor
(fiber optic interferometer)
force detector
(NEMS resonator)
excitation field
(rf coil or microstrip)
gradient magnent
(nanoscale magnet)
surfaces
of constant
magnetic field
resonant
slice
H
1
cos(w
0
t)
sample
Figure 8.62 The principal scheme of mechanically detected magnetic resonance imaging. After
M. L. Roukes, Nanoelectromechanical systems, in Technical Digest of the 2000 Solid-State
and Actuator Workshop, Hilton Head Island, SC, pp. 110.
techniques still take about 10
14
10
16
nuclei to generate a measurable signal of the mag-
netic resonance. This means that state-of-the-art MRI in research laboratories attains,
at best, a maximal resolution (minimum voxel size) of the order of 1 m. For standard
clinical MRI, the resolution yielded by commercial instrumentation is much poorer, with
a voxel size of about 1 mm.
Mechanically detected MRI, now commonly called magnetic resonance force
microscopy (MRFM), is signicantly more sensitive than conventional MRI. There are
several principal components of a MRFM instrument, which are presented in Fig. 8.62.
An antenna structure in the formof a coil or a microstripline generates a radio-frequency
eld of frequency
0
. The static magnetic eld created by a miniature magnet splits the
spin states of impurities within a sample and provides a resonance interaction of the
Figure 8.63 The electron shuttle a device with electron transport provided through mechanical
motion of the cantilever. C is a nanosized cantilever, S and D are the source and drain contacts,
and G1 and G2 are the gate contacts, which move the cantilever. The inset depicts the electric
circuit of the structure. From A. N. Cleland, Foundations of Nanomechanics (Berlin,
Springer-Verlag, 2003).
radio-frequency eld with these spins. The magnet is attached to a mechanical resonator
of a cantilever type. The interaction of the resonant spins with the magnet results in
a time-varying back-action force upon this cantilever. This force can be detected by a
sensor with high resolution of displacements, such as an optical interferometer. All of
these elements constitute a resonant force sensor. Such a sensor can detect the extremely
weak forces exerted by the resonant spins upon the mechanical system. Microscopy is
realized by scanning the cantilever with attached magnet over the sample. By correlating
the resonant mechanical response with the cantilever position, one can obtain spatial
imaging of spin density.
The resonance properties of NEMSs provide key advances for MRFM. Indeed, utiliza-
tion of a nanometer-scale ultra-high-frequency mechanical resonator makes it feasible to
couple directly the spin precession and mechanical vibrations and to improve drastically
the resolution of MRI.
The eld of MRFM is still very much in its infancy. Sustained efforts are required in
order to take it from a scientic demonstration to a useful technique for high-resolution
MRI. With its potential for atomic resolution, such efforts are of great potential impor-
tance, especially for biochemical applications.
Coupling of electron transport and mechanical motion. The electron shuttle
The coupling of electron transport and mechanical motion of NEMSs gives rise to new
effects, which can be useful for a number of applications. Consider an example of such
a coupling, which can be called the electron shuttle. The structure contains a metallized
cantilever suspended between two metallic electrodes, as shown in Fig. 8.63, where an
5
0
4
2
1.0 2.0
V (V)
50 100
5
f (MHz)
I

(
n
A
)
I

(
n
A
)
45
300 K
30
15
0 50 100
f (MHz)
I

(
n
A
)
Figure 8.64 Electric current in the electron shuttle versus frequency in the gate circuit. Upper
inset: the cantilever is blocked and there is no current. Lower inset: the shuttle current as a
function of the sourcedrain voltage at a given frequency in the gate circuit. From A. N.
Cleland, Foundations of Nanomechanics (Berlin, Springer-Verlag, 2003).
electron micrograph of the electron shuttle is presented. Two gates are fabricated to
drive the cantilever electrostatically. The cantilever can be driven by the voltage, V
G
1
G
2
,
so that the cantilever can nearly contact each of the spatially separated electrodes. A
voltage V
SD
is applied to the electrodes, which can be considered as a source and a
drain. If the frequency of the voltage applied to the gates, V
G
1
G
2
, does not coincide
with the resonance frequency of the cantilever, the amplitude of cantilever vibrations is
small. Accordingly, the cantilever does not mechanically contact the source and drain
electrodes, and electron transport is suppressed. As soon as the frequency V
G
1
G
2
matches
the mechanical resonance, the cantilever contacts the source and drain during each cycle
of the mechanical motion. When the cantilever contacts the electrodes successively, the
metallized end of the cantilever charges and discharges, and transfers electrons between
the source and the drain. In Fig. 8.64, results of measurements of the sourcedrain current
are presented; several peaks in the current correspond to different resonant vibration
modes of the cantilever. Only for these resonances are the amplitudes of the vibrations
large enough to support electron transfer through the nanostructure. If the motion of the
cantilever is blocked, no current is observed, as seen from the upper inset to Fig. 8.64.
The lower inset shows the magnitude of one of the current peaks as a function of V
SD
at a given resonance frequency. Thus, this NEMS actually presents an electron shuttle.
The number of electrons transferred depends on the dimensions of the metallized island
Figure 8.65 A set of SiC NEMSs. Submicrometer double-clamped SiC beams exhibit
fundamental resonance frequencies from 2 to 134 MHz. From M. L. Roukes,
Nanoelectromechanical systems, in Technical Digest of the 2000 Solid-State and Actuator
Workshop, Hilton Head Island, SC, pp. 110.
on the cantilever. By decreasing the size of the island, it becomes feasible to transfer a
single electron at a time.
Following these examples of NEMSs with different functions, we consider the basic
parameters that determine the performance of any NEMS. Actually, these parameters
are practically the same as for an electronic device: the response time or characteristic
frequency, the quality factor (energy loss), the operating power, the signal-to-noise ratio
(sensitivity), etc.
Frequency
We start by considering a NEMS as a mechanical vibrator at a natural angular frequency,
0
. The frequency can be approximated by
0
=
K,M, where K is an effective spring

constant and M is the mass of the mechanical vibrator. Underlying these simplied
effective terms is a complex set of elasticity equations that govern the mechanical
response of these objects. If we reduce the size of the mechanical device while preserving
its overall shape, then the fundamental frequency,
0
, increases as the vibrators linear
dimension decreases. This is obvious because the mass is proportional to the volume
of the mechanical vibrator, while the effective spring constant for exural deformations
increases with the decrease of its length. This increasing frequency effect is important
because a high frequency response translates directly to a fast temporal response to
applied forces. It also implies that a fast response can be achieved without the expense of
making stiff structures. Moreover, a small spring constant provides very high mechanical
responsivity.
The shape of the vibrations and resulting frequencies depend on the way the beams
are clamped. Three variants can be realized in a particular device: (i) both ends clamped
or free, (ii) both ends pinned, and (iii) a suspended beam clamped at one end (the can-
tilever). In Fig. 8.65, images of double-clamped SiC beams are shown. The beams are of
Table 8.2 Estimated fundamental frequency versus geometry for SiC, Si, and GaAs mechanical resonators
Resonator dimensions
L
x
L
y
L
z
in (m)
Boundary conditions 10 0.2 0.1 1 0.05 0.05 0.1 0.01 0.01
Material SiC, Si, GaAs SiC, Si, GaAs SiC, Si, GaAs
Both ends clamped or free 12, 7.7, 4.2 MHz 590, 380, 205 MHz 12, 7.7, 4.2, GHz
Both ends pinned 5.3, 3.4, 1.8 MHz 260, 170, 92 MHz 5.3, 3.4, 1.8 GHz
Cantilever 1.9, 1.2, 0.65 MHz 93, 60, 32 MHz 1.9, 1.2, 0.65 GHz
After M. L. Roukes Nanoelectromechanical systems, in Technical Digest of the 2000 Solid-State Sensor and
Actuator Workshop, Hilton Head Island, SC, pp. 110.
different dimensions and exhibit fundamental frequencies of exural vibrations varying
from 2 MHz to 134 MHz. Table 8.2 displays frequencies for the fundamental exu-
ral modes of thin beams for various materials (SiC, Si, and GaAs) and dimensions
spanning the domain from micromechanical systems to nanosystems well within the
nanoregime. The last column in Table 8.2 represents the dimensions currently attain-
able with advanced electron-beam lithography. The materials SiC, Si, and GaAs are of
particular interest, because they are available with extremely high purity as monocrys-
talline layers in epitaxially grown heterostructures. The numbers in Table 8.2 are rough
averages for the various commonly used crystallographic orientations. It is particularly
notable that, for structures of the same dimensions, Si yields frequencies of a factor of
two, and SiC of a factor of three, higher than those obtained with GaAs devices. This
increase reects the increased phase velocity of sound as well as the higher stiffness of the
materials.
Thus, mechanical resonators with fundamental frequencies above 10 GHz (10
10
Hz)
can nowbe built using surface nanomachining processes involving state-of-the-art nano-
lithography on the 10-nm scale.
Quality factor
Another important parameter, which characterizes the rate of energy dissipation in a
NEMS, is the so-called quality factor, Q. If we dene a decay time of the exural
vibrations of the beam in a NEMS,
d
, then, Q
0
d
. The quality factor achieved for
a NEMS in a moderate vacuum is in the range from 10
3
to 10
5
. This greatly exceeds the
quality factor typically realized for electrical (microwave) resonators. The small degree
of internal energy dissipation in a NEMS directly relates to the high attainable force
sensitivity and low operating power levels.
Characteristic operating power level
Applications of NEMS resonators involve the use of a specic vibrational mode, typi-
cally one of the lowest exural modes. A rough understanding of the minimum power
Table 8.3 Representative operating power levels for NEMSs
(1 attowatt = 1 aW 10
18
W)
Frequency,
0
Quality factor, Q Minimum power, P
min
100 MHz 10 000 40 aW
100 MHz 100 000 4 aW
1 GHz 10 000 400 aW
1 GHz 100 000 40 aW
After M. L. Roukes (2000) Nanoelectromechanical systems, in
Technical Digest of the 2000 Solid-State Sensor and Actuator Work-
shop, Hilton Head Island, SC, pp. 110.
necessary to operate with a NEMS using the exural mode can be obtained as follows.
At equilibrium the average energy of such a mode is equal to the thermal energy k
B
T.
To perform an operation, the energy of the external input signal should be larger than the
thermal energy. The characteristic time scale for energy exchange between the mode, at
frequency
0
, and its surroundings is just the decay time,
d
= Q,
0
. Thus, the mini-
mum power of the signal which has to be applied to the system to drive it to an amplitude
above the thermal uctuations is
P > P
min
= k
B
T.
d
=
k
B
T
0
Q
.
In Table 8.3, we present values of P
min
for various frequencies and quality factors at room
temperature. As displayed in Table 8.3, this minimum power can be extremely small for
a NEMS. For device dimensions accessible today via electron-beam lithography, the
characteristic level is of the order of tens of attowatts (10
17
W!). This is many orders
of magnitude smaller than the power dissipation in contemporary systems of similar
complexity based on digital devices that work solely with electric signals.
Dynamic range of a NEMS
From Table 8.3 it is clear that NEMSs have the potential to provide new types of ultra-
low-power electromechanical signal processing. However, realization of these potential
advantages is not a simple task. To utilize the full potential of a NEMS, displacement
transduction schemes should be capable of providing resolution of the beam displace-
ment at the level of the thermomechanical uctuations. Indeed, at a nite temperature
there always occur mechanical uctuations and, thus, there exist small random (chaotic)
displacements, Lz
T
, of the beam. These displacements can be readily estimated by
equating the potential energy of our vibrator,
1
2
K(Lz
T
)
2
, to the thermal energy k
B
T,
which gives the amplitude of such thermally induced vibrations as Lz
T
=
2k
B
T,K.
As we indicated above, the spring constant K decreases as the dimensions of the beam
diminish. Thus, thermally induced displacements scale with length, L, as Lz
T
1,
L.
Simultaneously, the signal amplitude of vibrations for a mechanical device (in a linear
operation regime) scales downward in direct proportion to its size. It is important to esti-
mate the acceptable level of operational displacements of the suspended structure, i.e.,
the so-called dynamic range for a linear NEMS. To estimate the characteristic dynamic
range, we need to dene the displacement amplitude corresponding to the onset of non-
linearity. This nonlinearity implies that, in a power-series expansion for elastic potential
energy, the largest term beyond the quadratic term (i.e., beyond the Hookes-law term)
becomes important. For a double-clamped beam, this condition translates into the relation
Lz
N
0.5L
z
. This criterion depends only upon the beam thickness, L
z
, in the direction
of vibrations. The linear dynamic range for a NEMS can be dened as the ratio of the
nonlinearity onset, Lz
N
, to the thermal displacement, Lz
T
: DR = 10 ln(Lz
N
,Lz
T
).
The following example highlights the characteristic displacements and the dynamic
range in a relatively small NEMS. Consider a suspended Si structure of dimensions
0.1 m 0.01 m0.01 m with a quality factor Q = 10
4
at T = 300 K. Then, we
nd Lz
T
0.09 nmand Lz
N
5 nm. For these parameters, a high value of the dynamic
range, DR, of approximately 40 is obtained.
These considerations andestimates indicate twoverycrucial areas for NEMSengineer-
ing necessary to provide femtowatt to picowatt regimes: (i) development of ultra-sensitive
transducers that are capable of enhanceddisplacement resolutionwithincreasinglyhigher
frequencies as device sizes are progressively scaled downward and (ii) development of
techniques tailored to operate over the entire dynamic range of the NEMS. As we studied
previously, nowadays, probe microscopy techniques are capable of probing and measur-
ing quantitatively estimated ultimate displacements.
In conclusion, a nanoelectromechanical system consists of a nanometer-to-
submicrometer-scale mechanical resonator that is coupled to an electronic device of
comparable dimensions. The mechanical resonator may have a simple geometry, such
as a cantilever (a suspended beam clamped at one end) or a bridge (a suspended beam
clamped at both ends) and is fabricated from materials such as silicon using litho-
graphic and other techniques similar to those employed for fabricating integrated cir-
cuits. Because of their submicrometer and nanoscale size, the mechanical resonators can
vibrate at frequencies ranging from a few megahertz up to around a gigahertz. The qual-
ity factors of these resonators greatly exceed those of typical microwave resonators. A
NEMS operates with lowpower dissipation in a wide dynamic range. These properties of
NEMSs openthe waytoa number of applications, rangingfromsignal processingtonovel
detectors.
8.8 Quantum-dot cellular automata
As we discussed previously, the general tendency is that improvement in technology
leads to progressive scaling down of electronic devices and widening of their func-
tionality. However, as more and more devices are placed into the same area, the heat
generated during a switching cycle can no longer be removed and this may result in
- cell - physical interaction
Figure 8.66 A schematic picture of a cellular array with interconnections via physical interaction.
considerable limitation of the speed of operation. In addition, interconnections between
devices can not be scaled in accordance with the device scaling because of the effects
of wire resistance and capacitance. The latter phenomenon can lead to a wiring bot-
tleneck. In such a situation, alternative approaches are needed in order to solve these
problems.
Contemporary nanotechnology can support alternative electronic devices and system
architectures. One such approach is based on quantum dots arranged in locally inter-
connected cellular-automata-like arrays. The fundamental idea of quantum-dot cellular-
automata operation is to encode information using the charge conguration of a set of
quantum dots. Importantly, in the quantum-dot cellular-automata approach, the infor-
mation is contained in the arrangement of charges of the dots, rather than in the ow
of the charges (i.e., current). It can be said that the devices interact by direct Coulomb
coupling rather than via the current through the wires. Figure 8.66 illustrates a locally
interconnected array of cells. Obviously, a dense arrangement of nanometer-scale quan-
tum dots can provide the necessary physical interactions inside the array. These physical
interactions between the elements, together with the topology of the system, determine
the overall functionality of the array.
As discussed in Section 5.5, technologies for the growth and processing of quantum-
dot arrays are already available. They facilitate the fabrication of quantum dots with the
necessary properties, arrangement, etc.
In quantum-dot cellular automata, the building block is called a cell. Figure 8.67(a)
shows such a cell. A cell is composed of (at least) four quantum dots positioned at the
corners of a square. A cell contains two excess electrons, which are allowed to tunnel
between neighboring quantum dots in the cell. Tunneling out of a cell is assumed to be
completely suppressed by the potential barriers between the cells.
P = 1 P = +1
(a) (b)
Figure 8.67 A quantum-dot cell consisting of ve dots with two electrons in the cell: (a) the
Coulomb repulsion causes the electron to occupy the antipodal sites; and (b) two bistable states
result in different polarizations. From S. Craig, P. Lent et al., Bistable saturation in coupled
quantum dots for quantum cellular automata, Appl. Phys. Lett., 62, 714 (1993). Reused with
permission from S. Craig, P. Lent, D. Tougaw, and W. Porod, Applied Physics Letters, 62, 714
(1993).
C _
1.0 0.5 0.0 0.5 1.0
1.0
0.5
0.0
0.5
1.0

P
1
P
2
cell 2 cell 1
Figure 8.68 The cellcell response. A polarized cell 1 implies the polarization of cell 2 due to the
Coulomb repulsion of the electrons populating the cells. The strong nonlinearity of the cellcell
response plays the same role as the gain in a conventional digital device. From S. Craig, P. Lent
et al., Bistable saturation in coupled quantum dots for quantum cellular automata, Appl. Phys.
Lett., 62, 714 (1993). Reused with permission from S. Craig, P. Lent, D. Tougaw, and W. Porod,
Applied Physics Letters, 62, 714 (1993).
C _
The Coulomb repulsion between the electrons in a cell tends to place themat antipodal
sites in the square. For an isolated cell, there are two energetically equivalent arrange-
ments of the extra electrons, which are denoted as cell polarizations, P = 1 and
P = 1. The polarization is used to encode binary information. For example, if P = 1
represents a binary 1, then P = 1 can represent a binary 0. The two polarization states
of a cell will not be energetically equivalent if another cell is nearby. Figure 8.68 shows
how one cell is inuenced by the state of the neighboring cell. The inset illustrates two
cells where the polarization P
1
is determined by the polarization of the neighbor, P
2
,
1 1
wire
fixed driver
(a)
1
1
corner
(c)
0
1
1
1
fan-out
1
(b)
inverter
(d)
Figure 8.69 Examples of the simplest quantum-dot cellular-automata arrays: wire, corner,
fan-out, and inverter. From Y. Ono, A. Fujiwara et al., Manipulation and detection of single
electrons for future information processing, J. Appl. Phys., 97, 031101 (2005). Reused with
permission from Yukinori Ono, Akira Fujiwara, Katsuhiko Nishiguchi, Hiroshi Inokawa, and
Yasuo Takahashi, Journal of Applied Physics, 97, 031101 (2005).
C _
2005 American Institute of
Physics.
through the interaction of the electrons in the cells. Let us assume that the polarization
P
2
has a given value corresponding to a certain arrangement of the charge in the cell 2.
Owing to the charge repulsion, the response of cell 1 is given by the strongly nonlinear
dependence presented in Fig. 8.68: a small asymmetry of charge in cell 2 is sufcient
to break the degeneracy between the two possible states of cell 1 and leads to the same
conguration of cell 1. Thus, in some cell arrangements, xing the polarization of a cell
at an edge by an external bias determines the polarization of the cells at other edges. The
polarization is dependent on the cell conguration.
In the quantum-dot cellular-automata approach, the circuit is built by forming a
tree of cells. Figure 8.69 shows some elements of the quantum-dot cellular automata.
Figure 8.69(a) is a binary wire. The polarization of the leftmost cell is xed, which rep-
resents the input. Then, all the other cells, including the output cell, align with the same
polarity because it is the most energetically favorable. Flipping the polarity of the
input cell results in the ipping of all the other cells. During this procedure, no
direct current ows in the circuit. Cells positioned diagonally from each other tend
to anti-align. This feature is employed to construct other logical elements. A fan-
out, a corner, and an inverter are shown in Figs. 8.69(b), 8.69(c), and 8.69(d),
respectively.
Inconclusion, quantum-dot cellular-automata systems exploit the interactions between
quantumdots onthe nanoscale. Theyare able toperformall processes necessaryfor signal
processing. No current is used in the circuits built on the basis of quantumautomata. The
quantum-dot cellular-automata approach presents a nanoscale alternative to conventional
microelectronics and nanoelectronics.
8.9 Closing remarks
Inthis chapter we have consideredseveral different nanostructure devices. It is convenient
to classify these devices into two categories. The rst are the devices which are based
on usual classical principles of operation but considerably scaled down. This scaling
down facilitates the improvement of device performance and solves some fundamental
difculties, which can not be overcome in microelectronics. These devices include eld-
effect transistors and bipolar transistors.
Another category is nanostructure devices based on new physical principles. Among
these are resonant-tunneling diodes the simplest quantum devices. Since these devices
have nanoscale dimensional features, they have extremely short transit times for carrier
transport through the structures, which results in the possibility of generating ultra-high-
frequency electromagnetic oscillations up to the terahertz frequency range. As examples
of new principles for three-terminal electronic devices (transistors), we analyzed the
quantum-interference and hot-electron transistors, which have great potential for ultra-
high-speed operation.
To illustrate the contribution of nanostructures to optical devices, we have focussed
on the examples of quantum-well and quantum-wire lasers with bipolar injection, and
the multilayered quantum-cascade laser. The latter is a monopolar device, i.e., it uses
only one type of carrier (electrons). Exploitation of nanostructures results in a dramatic
decrease of the threshold current necessary for laser generation and in a widening of
emission spectra.
We have analyzed some nanoelectromechanical systems with mechanical resonators
that can vibrate at frequencies ranging from a few megahertz up to 1 GHz and have
quality factors greatly exceeding those of typical microwave resonators. This implies that
nanoelectromechanical systems operate with low power dissipation. These properties
open the way for a number of applications, ranging from signal processing to novel
sensors/detectors.
Finally, we have described innovative systems for signal processing based on quantum-
dot cellular automata. These automata exploit interactions between quantum dots on the
nanoscale. This approach presents a nanoscale alternative to conventional microelec-
tronics and nanoelectronics.
The description of nanostructured devices which we presented here is not com-
plete. More information on traditional scaling down to nanosize devices, double-barrier
resonant-tunneling diodes, and lasers can be found in the following books:
M. Shur, Physics of Semiconductor Devices (Englewood Cliffs, NJ, Prentice-Hall,
1990).
S. M. Sze, High-Speed Semiconductor Devices (New York, John Wiley & Sons, Inc.,
1990).
Single-electron transport is described in the following references:
K. K. Likharev, Correlated discrete transfer of single electrons in ultrasmall tunnel
junctions, IBM J. Res. Develop., 12, 144 (1988).
M. A. Kastner, The single electron transistor, Rev. Mod. Phys., 64, 849 (1992).
Y. Ono, A. Fujiwara et al., Manipulation and detection of single electrons for future
information processing, J. Appl. Phys., 97, 031101 (2005).
A very detailed review of nanoelectromechanical systems is given in the text
A. N. Cleland, Foundations of Nanomechanics (Berlin, Springer-Verlag, 2003).
Appendix: tables of units
Table 1 SI base units
Unit
Quantity Name Symbol
Length meter m
Mass kilogram kg
Time second s
Electric current ampere A
Temperature kelvin K
Amount of substance mole mol
Table 2 SI derived units
Unit
Quantity Name Symbol Equivalent
Plane angle radian rad m/m = 1
Solid angle steradian sr m
2
/m
2
= 1
Speed, velocity m s
1
Acceleration m s
2
Angular velocity rad s
1
Angular acceleration rad s
2
Frequency hertz Hz s
1
Force newton N kg m s
2
Pressure, stress pascal Pa N m
2
Work, energy, heat joule J N m, kg m
2
s
2
Impulse, momentum N s, kg m s
1
Power watt W J s
1
Electric charge coulomb C A s
Electric potential, emf volt V J C
1
, W A
1
Resistance ohm O V A
1
Conductance siemens S A V
1
, O
1
Magnetic ux weber Wb V s
Inductance henry H Wb A
1
Capacitance farad F C V
1
Electric eld strength V m
1
, N C
1
Magnetic ux density tesla T Wb m
2
, N A
1
m
1
Electric displacement C m
2
Magnetic eld strength A m
1
Celsius temperature degree Celsius

C K
Luminous ux lumen lm cd sr
Illuminance lux lx lm m
2
Radioactivity becquerel Bq s
1
Catalytic activity katal kat mol s
1
324 Appendix
Table 3 Physical Constants
Constant Symbol Value Units
Speed of light in vacuum c 2.9979 10
8
m s
1
3 10
8
Elementary charge e 1.602 10
19
C
Electron mass m
0
9.11 10
31
kg
Electron charge to mass ratio e,m
0
1.76 10
11
C kg
1
Proton mass m
p
1.67 10
27
kg
Boltzmann constant k
B
1.38 10
23
J K
1
Gravitation constant G 6.67 10
11
m
3
kg
1
s
2
Standard acceleration of gravity g 9.807 m s
2
Permittivity of free space c
0
8.854 10
12
F m
1
10
19
,(36)
Permeability of free space j
0
4 10
7
H m
1
Plancks constant h 6.6256 10
34
J s
Impedance of free space
0
=
j
0
,c
0
376.73 120 O
Avogadro constant N
A
6.022 10
23
mol
1
Table 4 Standard prefixes used with SI units
Prex Abbreviation Meaning Prex Abbreviation Meaning
atto- a- 10
18
deka- da- 10
1
femto- f- 10
15
hecto- h- 10
2
pico- p- 10
12
kilo- k- 10
3
nano- n- 10
9
mega- M- 10
6
micro- - 10
6
giga- G- 10
9
milli- m- 10
3
tera- T- 10
12
centi- c- 10
2
peta- P- 10
15
deci- d- 10
1
exa- E- 10
18
Table 5 Conversion of SI units to Guassian units
Quantity Si unit Gaussian units
Length 1 m 10
2
cm
Mass 1 kg 10
3
g
Force 1 N 10
5
dyne = 10
5
g cm s
2
Energy 1 J 10
7
erg = 10
7
g cm
2
s
2
1 eV = 1.602 10
19
J = 1.602 10
12
erg
Index
absorption edge, 290
absorption of one photon, 285
acceptors, 118
active medium, 287
active region, 293
alkanethiol endgroups, 150
alligator clips, 147
alloy, 68
amino acids, 145
amorphous, 68
amphoteric, 119
angular frequency, 16
angular momentum, 57
anisotropic two-dimensional layered
structures, 102
annihilation, 290
antibodies, 145
antibonding state, 60
antigens, 145
approximation, virtual-cathode, 190
approximation, virtual-crystal, 83
armchair, 98
atomic force microscopy, 98
atomic orbitals, 60
ballistic motion, 8
bandgap wavelength, 290
band-to-band, 289
base of diode, 189
basis vectors, 71
biomoleculeinorganic cluster hybrid,
150
bipolar transistors, 278
Bloch, 213
Bloch form, 75
Bloch function, 75
block copolymers, 142
body-centered cubic lattice, 72
Bohr radius, 57
bonding orbital, 60
bonding state, 60
bound states, 36
buckyball fullerenes, 102
carbon nanotubes, 98
carboxyl endgroups, 142
cascade structure, 305
cathode, 189
charge-limited transport, 201
chemical radicals, 117
chemical-vapor deposition, 112
Childs law, 198
chiral angle, 99
classical ballistic regime, 170
classical description, 22
classical regime, 170
classical transport regime, 170
coherence length, 169
coherent tunneling, 245
coherent waves, 19
collector, 245
collisionless motion, 8
complementary, 149
complex conductivity, 187
compound semiconductors, 68
concept of hole, 76
conduction, 70
constructive interference, 19
Coulomb blockade, 212
covalent, 69
criterion for laser oscillations, 298
cross-sectional STM, 122
crystal, 68
crystalline potential, 74
cubic symmetry, 72
cut-off wavelength, 290
CVD, 112
cyclic boundary conditions, 75, 101
dangling bonds, 134
DBRTD, 8
de Broglie wavelength, 28
degeneracy of states, 57
degenerate state, 174
density of states, 180
density of the distribution, 173
dephasing length, 170
326 Index
destructive interference, 19
diagonal components, 91
diamond lattice, 72
differential resistance, 191
diffusion, 168
diffusive classical size effects, 170
diode, 189
dip-pen nanolithography, 155
direct-bandgap semiconductor, 82
discrete energy levels, 41
discrete energy portions, 24
discrete energy spectrum, 36
dispersion relation, 16
distribution function, 173
donors, 118
double-barrier resonant-tunneling diode, 8
drain, 255
DRAM, 3
Drude formula, 188
dynamic range, 317
edge-emitting conguration, 303
effective temperature, 193
elastic collision, 168
elastic constants, 91
elastic moduli, 91
electric current standard, 274
electromagnetic waves in free space, 21
electron afnity, 48
electron energy band, 76
electron mobility, 183
electron shuttle, 313
electron-afnity rule, 85
emission of a photon, 285
emitter, 245
energy band, 66
energy bandgaps, 79
energy splitting, 61
energy subband, 48
energy valleys, 79
enzymes, 145
epitaxial growth, 112
equation, wave, 16
equidistant energy levels case, 46
etching, 117
excess energy, 129
expectation value, 37
extra elastic energy, 92
face-centered cubic lattice, 72
Fermi distribution function, 175
Fermi energy, 175
Fermi level, 175
Fermi statistics, 175
Fermi surface, 177
Fermi wavevector, 177
eld-effect devices,
eld-effect transistors, 255
nite conductance, 206
nite lattice mismatch, 90
nite resistance, 206
rst Brillouin zone, 75
at band, 293
exural deformations, 159
exural modes, 315
forward bias, 278
frequency of single-electron tunneling
oscillations, 213
fullerenes, 98
gated heterostructures, 220
gradient, 12
graphene, 98
ground state, 41
growth chamber, 122
gyromagnetic ratio, 57
Hamiltonian, 12
harmonic oscillator, 45
heavy hole, 81
helicity, 98
Hermite polynomial, 46
heterojunction eld-effect transistor, 93
heterostructure eld-effect transistor, 261
heterostructures, 5
high (classical) frequencies, 172
high-electron-mobility transistors, 258
highest occupied molecular orbital, 153
highly degenerate electron gas, 176
hole, 76
hot-electron effects, 165
hot electrons, 193
hot-electron transistors, 280
hybridized orbitals, 61
impedance, 188
incident waves, 20
incoherent waves, 19
indirect-bandgap semiconductor, 82
induced-base transistor, 283
inelastic, 168
inelastic scattering, 168
inelastic scattering length, 168
injection, 190
inorganic nanotube, 102
insulators, 66
interband phototransitions, 289
intraband absorption, 304
intrinsic density, 118
intrinsic material, 118
isolated islands, 130
isotropic effective mass, 79
Index 327
Landauer formula, 206
laser ablation, 139
laser diodes, 293
lattice, 71
lattice constant, 73
law of momentum conservation, 11
law of energy conservation, 12
light hole, 81
light-emitting diodes, 302
linear electron concentration, 179
longitudinal, 80
longitudinal modes, 289
low frequencies, 172
lowest unoccupied molecular orbital, 152
macromolecular cells, 143
magnetic resonance force microscopy, 312
magnetic resonance imaging, 311
many-electron system, 172
many-valley semiconductors, 79
Mathieu equation with damping, 309
Maxwellian distribution, 174
MBE, 112
mean free path, 168
mesoscopic device, 264
mesoscopic systems, 169
mesoscopic transport regime, 170
metallic island, 209
metaloxidesemiconductor FET, 296
metalsemiconductor FET, 256
microelectronics, 1
microlasers, 300
miniature styrene chain, 142
modes, 47
modulation-doped FET, 258
molecular-beam epitaxy, 112
momentum, 76
Moores law, 2
MottGurney law, 191
nanoelectromechanical systems, 9, 110
nanoelectronics, 1
nanolithography methods, 115
nanoscience, 1
nanotechnology, 1
nanowire FET, 261
NEMS, 9, 110
non-complementary, 150
nonlinear optical media, 25
non-stationary wavefunction, 34
normally off structure, 226
normally on structure, 226
number of quanta, 25
Ohms law, 184
oligonucleotides, 151
one-dimensional electron system, 228
one-dimensional transport, 101
one-dimensional wavevector, 101
one-particle wavefunction, 74
open optical resonators, 288
optical excitation, 292
optoelectronic devices, 9, 284
parametric amplication, 308
parametric resonance, 310
particular potential prole, 202
Pauli exclusion principle, 58
peak-to-valley ratio, 252
penetration of a particle, 45
peptides, 145
period, 16
periodic, 68
phase shift, 187
phase time, 253
phase velocity, 17
photolithography, 115
photons, 24
photoresists, negative, 115
photoresists, positive, 115
phototransitions, 285
physical statistics, 173
Plancks constant, 206
Plancks constant, reduced, 24
pn junction, 276
point particle, 12
point symmetry, 77
polarization charge, 212
polaron, 97
polycrystalline, 69
population inversion, 287
potential-effect transistors, 255
primitive cell, 71
primitive translation vectors, 71
proteins, 145
pseudomorphic heterostructures, 92
pump, 273
pumping, 292
pumping power, 292
quality factor, 316
quanta, 24
quantization of electron motion, 167
quantum ballistic transport regime, 170
quantum ballistics, 206
quantum boxes, 51
quantum dots, 51
quantum number, 41
quantum point contact, 206
quantum regimes of transport, 169
quantum size effects, 168
quantum wire, 49
328 Index
quantum-interference transistor, 264
quasiballistic, 167
quasiballistic transport, 170
quasielectric, 87
quasi-Fermi levels, 278
radially symmetric function, 56
radiative recombination, 290
range of coordinates, 27
range of momenta, 27
reactive-ion etching, 117
reactor, 114
reected waves, 20
relative mismatch, 91
relative positions, 84
resonant-tunneling process, 244
resonators, 308
reverse bias, 278
scanning tunneling microscopy, 120
scattering, 168
Schottky barrier, 256
Schottky depletion region, 225
Schottky voltage, 225
Schottky-gate structures, 225
Schr odinger wave equation, 33
self-consistency problem, 220
semiclassical transit time, 253
separation by implantation of oxygen, 160
sequential tunneling, 250
sheet concentration of electrons, 178
shot noise, 209
simple cubic, 72
single-electron pump, 274
single-electron transport, 209
single-electron devices, 209
single-electron transistor, 271
single-walled nanotube, 98
sites, 71
source, 255
space-charge-limi ted transport, 214
spacer layer, 223
specic conductivity, 184
spherical functions, 52
spin, 58
spin number, 174
spinorbit interaction, 80
split-gate technique, 269
split-off valence band, 81
spontaneous emission of a photon, 286
standing waves, 20
Stark effect, 236
stationary states, 171
stationary wavefunction, 34
stationary-state case, 35
sticky ends, 150
stiff rodlike segment, 142
stimulated emission, 285
strain tensor, 91
superlattice, 93
superposition principle, 21
surface energy, 129
surface reconstruction, 128
surface-emitting conguration, 303
surface-emitting laser, 299
temperature effect, 169
thermal diffusion length, 169
thiol bond, 146
TOPO, 135
total energy, 12
traditional low-dimensional structures,
218
transistors, 255
transit time, 171
translational symmetry, 71
transmission coefcient, 203
transmission electron microscopy, 125
transverse classical size effects, 170
transverse effective masses, 80
transverse modes, 289
traveling plane waves, 17
trial wavefunctions, 222
triblock polymer, 142
tri-n-octylphosphide, 135
tunnel channel, 122
tunneling, 121
tunneling effect, 45
tunneling electric current, 121
turnstile, 273
turnstile operation, 274
two-dimensional, 6
two-dimensional electron concentration, 178
two possible polarizations of the electromagnetic
wave, 22
type-I heterostructures, 85
type-II heterostructure, 86
ultra-high (quantum) frequencies, 172
uncertainty principle, 27
uncertainty relation, 35
ungated heterostructures, 220
unipolar device, 281, 304
unipolar intersubband quantum-cascade laser,
304
unit cell, 71
vacuum level, 84
valence, 70
valence energy, 58
valence-mismatched bonds, 88
valleys, 79
Index 329
variation method, 222
velocity overshoot, 194
velocity-modulation transistor, 263
vertical,
vertical geometry, 299
virtual-crystal approximation, 83
voltage-controlled devices, 255
warm electrons, 193
wave intensity, 18
wave interference, 18
wave mechanics, 26
wave surfaces, 23
wavefunction, 26
wavefunction mapping, 123
wavelength, 17
wavenumber, 20
waveparticle duality, 28
wavevector, 16
wavevector of the electron, 75
wide-bandgap semiconductors, 300
work function, 120, 218
wrapping angle, 99
wrapping vector, 98
zero-dimensional, 6
zero-dimensional system, 51, 232
zero-point vibrations, 47, 284

Nano Electronics

Uploaded by

Copyright:

Available Formats

Nano Electronics

Uploaded by

Document Information

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

Nano Electronics

Uploaded by

Copyright:

Available Formats

This page intentionally left blank

effective mass of electron

(r, t) complex conjugate of wavefunction +(r, t)

are arbitrary constant vectors. The two waves can also be

sin(qz t ). If the walls are rigid there are no displacements

c, where c is the dielec-

c, in a dielectric medium is less than that in free

(r. t ) is the complex conjugate of +(r. t ).

is the so-called effective mass of the electron and 1,E

), have anisotropic parabolic energy dispersion,

, and take [p[ = 1. Consider n-Ge, for

,m t ), with A and being arbitrary

at z . This potential is the most general formof a potential well.

as shown by the line labeled E

there is only one solution satisfying the physical

there exist two independent solutions; see curve 3

when wavefunctions are

A and corresponds to a certain eigenvalue a,

2,L arises as a result of the normalization condition:

c. Here, c and c are the speed of light and

c)q (see Eq. (2.41)). This

= 0, i.e., the wavevector, k, is parallel to the radius, r. On rewriting the radial

Figure 3.15 Eigenenergies of an electron in a hydrogen-atom potential well.

(k), are very complex and can be obtained

(k) always coincide with these high-symmetry points. In particular, this

(k) in series with

(k) is reduced to the following steps: (a) determination of the high-symmetry

(k) dependences is incommensurably simpler than

. Both and the wrapping

) as a result of the hexagonal character

C). Coke is a coal fromwhich most of the

C. Simultaneously with HSiCl

, which is characteristic for the conduction band. By

, is replaced by the effective mass of a

Figure 5.18 An equilibrium phase diagram of a lattice-mismatched heteroepitaxial system as a

in the range (0.011)m

Dt during the time t

, denes the limit belowwhich electron transport has a quantumcharacter.

, the device can be described within the

(also referred to frequently as the dephasing length),

transverse size effects:

). Replacing the summation by an

E(E E). (6.42)

diode. Potential energy proles, V(x),

regions. In such a case, this device is an

diode as shown in Fig. 6.10(a). Let + designate the electrostatic

GaAs diode, where n

, and introduced the maximum mean-free

(y. z) describes the transverse distribution of

are the velocities at the L and R cross-sections,

(r), with being the set

), which determines the probability of the electron occupying the energy

. Thus, the electron charge density is e

). The charge of the

(r) can be factorized as discussed in Section 3.2:

Figure 7.5 Conduction-band diagrams for M/AlGaAs/GaAs heterostructures. The built-in

layer of the AlGaAs

-contact on the left and a metal contact (Schottky contact)

region(onthe left) anda metal contact (the Schottkycontact)

10], [001], and [00