Amorphous Semiconductors - Kugler and Shimakawa

AMORPHOUS SEMI CON DU CTORS
Understanding the structural unit of crystalline solids is vital in determining

their optical and electronic properties. However, the disordered nature of amor-
phous semiconductors, where no long-range order is retained, makes it difficult
to determine their structure using traditional methods. This book shows how
computer modeling can be used to overcome the difficulties that arise in the
atomic-scale identification of amorphous semiconductors.
The book explains how to generate a random structure using computer model-
ing, providing readers with the techniques to construct realistic material struc-
tures. It shows how their optical and electronic properties are related to random
structures. Readers will be able to understand the characteristic features of dis-
ordered semiconductors. The structural and electronic modifications by photon
irradiation are also discussed in detail. This book is ideal for both physicists
and engineers working in solid state physics, semiconductor engineering, and
electrical engineering.
s ándor kugler is an Associate Professor at Budapest University of Tech-

nology and Economics, Hungary, and Guest Professor at Tokyo Polytechnic
University, Japan. He is an overseas editor for the Japanese Journal of Applied
Physics, and is a well-known expert in amorphous semiconductors and chalco-
genide glasses.
koichi shimakawa is a Professor at the University of Pardubice, Czech

Republic, a Fellow and Emeritus Professor of Gifu University, Japan, and Senior
Researcher at Nagoya Industrial Science Research Institute, Japan. He is an
internationally recognized researcher in the field of amorphous solids.
A M O R P H O U S S E M I C O N D U C TO R S
S Á N D O R K U G L E R
Budapest University of Technology and Economics, Hungary
KO I C H I S H I M A K AWA
University of Pardubice, Czech Republic
University Printing House, Cambridge CB2 8BS, United Kingdom
Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781107019348

c S. Kugler and K. Shimakawa 2015
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2015
Printed and bound in the United Kingdom by CPI Group Ltd, Croydon CR0 4YY
A catalog record for this publication is available from the British Library
Library of Congress Cataloging in Publication data
Kugler, Sándor, 1950–
Amorphous semiconductors / Sándor Kugler, Budapest University of Technology and Economics,
Hungary, Koichi Shimakawa, University of Pardubice, Czech Republic.
pages cm
Includes bibliographical references and subject index.
ISBN 978-1-107-01934-8 (hardback)
1. Amorphous semiconductors. 2. Semiconductors – Computer simulation.
I. Shimakawa, Koichi. II. Title.
QC611.8.A5K84 2015
537.6 223 – dc23 2013022108
ISBN 978-1-107-01934-8 Hardback
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.
To my wife, Szabina, and our children, Szilvia and Zsófia
SK
To my wife, Akiko, and our children, Tetsuro, Shoko, and Ryo

KS
Contents
Preface page ix
1 Introduction 1
1.1 Historical overview: science and applications 1
1.2 Definitions 6
References 10
2 Preparation techniques 14
2.1 Growth of thin-film forms 14
2.2 Melt-quenched glasses 18
2.3 Other techniques 19
2.4 The Phillips constraint theory 20
References 22
3 Structure 24
3.1 Differences between amorphous and crystalline
semiconductors 24
3.2 Projection from three-dimensional structures to
one-dimensional functions 33
3.3 Three-dimensional structure derived from one-dimensional
function 41
3.4 Phase change and its applications 73
References 78
4 Electronic structure 86
4.1 Chemical bonds 86
4.2 Electronic density of states 90
viii Contents
4.3 Defects 93
4.4 Optical and electronic properties 104
References 113
5 Photoinduced phenomena 117

5.1 Photoinduced volume changes 117
5.2 Photodarkening and photobleaching 129
5.3 Photoinduced defect creation: the Staebler–Wronski effect 130
5.4 In situ simultaneous measurement of PVE, PD, and PDC:
the dynamics of time evolution and mechanisms 135
5.5 Photocrystallization and photoinduced amorphization 139
5.6 Some applications of photoinduced effects 140
References 141
Index 145
Preface
Understanding the structural unit of crystalline solids is vital in determining their

optical and electronic properties. Determination of the structure of a condensed
phase without periodicity is not an easy task. An important objective of the book
is to provide an introduction to the reader of how to construct computer modeling
of realistic random structures of amorphous IV- and VI-column-element semi-
conductors and their alloys. Both the merits and drawbacks of the techniques
currently used to generate structures using powerful computers are discussed.
Furthermore, the structural, electronic, and optical properties of mostly sigma-
bonded amorphous semiconductors can be learned.
The basis of this monograph was a course given by Sándor Kugler (SK) during
several years at Budapest University of Technology and Economics (and other
universities in Europe and Japan), with an extension by Koichi Shimakawa (KS).
Our common research experience with amorphous semiconductors extends back
more than twenty years. The book is aimed at final-year university students and
PhD students in physics, materials science, and chemistry who have already
completed introductory courses on quantum mechanics and solid state physics.
This book will be useful for both physicists and engineers working in solid state
physics, semiconductor engineering, and electrical engineering. For most of the
text, no high-level mathematics is needed. This book provides a much wider
literature overview than is usual for most handbooks.
A historical overview and a detailed summary of applications are given in
the first part of Chapter 1. Readers are informed how to develop further the
current technology (photovoltaic cells, thin-film transistors, DVDs, and direct
x-ray image detectors for medical use, etc.) using amorphous semiconductors.
The rest of the chapter analyzes and answers one of the most exciting questions
in the field: what are amorphous semiconductors?
x Preface
This is followed by a discussion of preparation techniques in Chapter 2. As

the glass-transition temperature determines most of the physical properties of
glasses, it is briefly discussed, together with Phillip’s constraint theory of glassy
materials, at the end of the chapter.
The third and longest chapter begins with an important topic, namely how
to determine experimentally whether a sample has an amorphous or crystalline
phase. The chapter describes atomic-scale computer modeling, including atomic
interactions, different simulation methods, and models obtained by structure
simulation. The final part of Chapter 3 introduces readers to the most successful
commercialized product of chalcogenide glasses, the phase-change materials.
Chapter 4 deals with the electronic behavior of covalently bonded amorphous
semiconductors, including defect-free systems and deviation from the ideal
networks, i.e. defects. Optical properties of amorphous semiconductors are also
described in this chapter.
Chapter 5 presents experimental results of photoinduced changes. It is shown
that the structural studies by means of atomic-scale computer simulations
are very useful for understanding the experimental results. The photoinduced
changes observed in both amorphous chalcogenides and hydrogenated amor-
phous silicon films are discussed.
We, SK and KS, would like to thank Kazuo Morigaki for introducing us to
the physics of amorphous semiconductors. Thanks are also due to Jai Singh
(University of Charles Darwin), S.O. Kasap (University of Saskatchewan), Keiji
Tanaka (Hokkaido University), Ted Davis (University of Leicester), Stephen
Elliott (University of Cambridge), and Takeshi Aoki (Tokyo Polytechnic Uni-
versity) for their powerful discussions. We must also thank Tokyo Polytechnic
University (formerly the Tokyo Institute of Polytechnics) for allowing us to use
their computing facilities for our large-scale computer simulations.
SK is indebted to many of his colleagues at the Budapest University of Tech-
nology and Economics, but he is especially grateful to his four PhD students,
Krisztina Kádas (Uppsala University), Krisztián Koháry (University of Exeter),
József Hegedüs (University of Helsinki), and Rozália Lukács (Norwegian Uni-
versity of Life Sciences). Furthermore, SK extends his thanks to István László
and László Pusztai, with whom he has had the pleasure of working on diverse
aspects of research on amorphous semiconductors. Special thanks are also due
to Károly Härtlein (Budapest University of Technology and Economics) for his
careful work in drawing some of the figures.
Preface xi
KS wishes to thank his colleagues at Gifu University, Shuichi Nonomura and

Takashi Itoh, for discussions, and also his PhD student, Yutaka Ikeda (YM sys-
tems, Company Executive), who developed a precise measurement system for
the photoinduced volume change observed in amorphous chalcogenides. Discus-
sions with Alex Kolobov (Advanced Institute of Industrial Science, Tsukuba),
Hiroyoshi Naito (Osaka Prefecture University), Hideo Hosono (Tokyo Institute
of Technology), Takashi Uchino (Kobe University), Masaru Aniya (Kumamoto
University), and Sergei Baranovskii (University of Marburg) were also very
fruitful and enjoyable in promoting our subjects.
Tomas Wagner and Miloslav Frumar (University of Pardubice) provided KS
with the opportunity to continue the research work at the Department of General
and Inorganic Chemistry (University of Pardubice), after his retirement from
Gifu University, supported by grant project CZ.1.07/2.3.00/20/0254 ReAdMat
financed by the European Union. KS wishes to express his thanks for this
opportunity.
1
Introduction
We present a brief historical overview of amorphous semiconductors including

their definition.
1.1 Historical overview: science and applications

It is important to include a historical overview of the science and successful
applications of amorphous semiconductors in the market in order to understand
the current situation. Although science and technology are closely connected,
we would venture that some applications have proceeded without secure sci-
entific knowledge. This may be a characteristic feature of the field of material
science. It is thus a good idea to begin by giving some examples of successful
applications.
(1) Chalcogenides. Electrophotography (or so-called xerography, a Greek
word, meaning “dry writing”) has been one of the most successful applications
of amorphous selenium (a-Se). The process was demonstrated by C.F. Carl-
son and O. Kornei in 1938, and modern xerographic processes are the same
as those proposed at that time (Pai and Springett, 1993). The Hungarian sci-
entist Pal Selényi first proposed the concept of the photographic process in the
1930s. His pioneering work in electrostatic picture recording formed the basis
of xerography. In fact, Selényi published and patented several fundamental ideas
of electrography and produced high-quality electrographic copies well before
Carlson’s proposal (Selényi, 1935a, 1935b, 1936). Films of a-Se have the fol-
lowing unique features: (i) high resistivity (it is a good insulator), and (ii) high
photoconductivity. These properties are useful for electronic charging in the
dark state and discharging in the photoilluminated state. These technologies
1
2 Introduction
were later applied to laser printers. More recently, a-Se has been replaced by
organic polymers for xerography (Weiss and Abkowitz, 2006). The main reason
for this is that organic materials cost less.
Most people would recognize the acronym “DVD” (digital versatile disk).
The basic operation of the DVD was first proposed by Feinleib et al. (1971),
although the material Ge2 Sb2 Te5 (also known as GST) has been employed
commercially by the Panasonic group in Japan (Yamada et al., 1991). A DVD
operates via optically induced phase changes (and hence changes in the reflection
coefficient) between amorphous and crystalline states. Using GST, DVDs can be
rewritten in excess of one million times, with a crystallization time of less than
50 ns achieved during each rewriting process. The DVD system currently has a
memory capacity exceeding 50 GB per disk using a blue semiconductor laser.
In the near future, rewritable electrical memory devices will be commercially
available that use a phase-change random access memory (PRAM), following the
memory switching devices proposed by Ovshinsky (1968). In addition, phase-
change materials offer a promising route for the practical realization of new
forms of general-purpose and “brain-like” computers that could learn, adapt,
and change over time (Wright et al., 2011).
As a-Se is a very sensitive photoconductor, especially for x-rays, due to its
high atomic weight, it has been possible to realize a direct x-ray imaging device
for use in the medical field. The image of a human hand by Rowlands and Kasap
(1997) recalls Wilhelm Röntgen’s first x-ray photograph of his wife’s hand. This
device incorporates a large area of thick (1 mm) a-Se evaporated onto a thin-film
transistor (TFT) made from a hydrogenated amorphous silicon (a-Si:H) active
matrix array (AMA). The x-ray-induced carriers in a-Se travel along the electric
field lines and are collected at their respective biased electrode and storage
capacitor. The stored images are then sent directly to the medical specialist’s
computer. This type of x-ray image sensor is used widely in mammography.
The Japan Broadcasting Corporation, NHK, has utilized the “avalanche
photomultiplication” effect in a-Se (Juška, Arlauskas, and Montrimas, 1987;
Tanioka, 2007) to create a powerful broadcasting tool. High-gain avalanche
rushing amorphous photoconductor (HARP) vidicon tubes have been developed
by K. Tanioka and his collaborators, leading to the construction of a HARP
vidicon TV camera that is over 100 times more sensitive than a CCD camera.
(2) Hydrogenated amorphous silicon (a-Si:H). In the early 1970s, the oil
crisis in the Middle East led to the serious consideration of using photovoltaic
(PV) cells as an alternative source of energy. To use PV cells as a viable power
1.1 Historical overview 3
source requires a large area and low cost. Consequently, the first a-Si:H solar
cells were fabricated by Carlson and Wronski (1976) at RCA Laboratories in
Princeton, NJ. Later, Y. Kuwano’s group at Sanyo Co. Ltd. (Japan) was the first to
market the PV devices. Common structures comprise p-i-n type heterojunctions.
There are several types of PV cells available, the most common being tandem
(with dual and triple junctions) a-Si:H configurations, for which more than 10%
efficiency is achieved in large-area commercial devices (Carlson et al., 1996).
People have recognized the importance of developing PV devices following
the Fukushima nuclear power station disaster caused by the earthquake and
subsequent tsunami on 11 March 2011.
Thin-film transistors (TFTs) using a-Si:H were first developed in the form
of field effect transistors (Powell, 1984; Spear and LeComber, 1984). Two of
the most important requirements for TFTs are a high ON/OFF current ratio and
a small gate voltage; these are achieved by using a-Si:H. These characteristics
mean that TFTs are suitable for use as switching transistors in a liquid crystal
display (TFT-LCD), which has completely replaced the former cathode ray
tube. Following subsequent improvements in TFT-LCDs, flat-panel displays
(FPDs) now produce the clear and large (over 100 cm) images used in TVs and
monitors.
(3) Oxides. The high demand for flexible and optically transparent TFTs for
use in the next-generation FPDs led to the realization of transparent conductive
oxides (TCOs) by H. Hosono’s group in 1996 (see, for example, Hosono (2006)).
Transparent TFTs were developed using ionic oxides such as a-InGaZnO4
(known as a-IGZO). The electron mobility of a-IZGO is larger than that of
a-Si:H, and the TFT stability is excellent. Samsung’s group in Korea developed
the a-IGZO TFT-LCD display that is used commercially in the iPad 3 (Apple
Inc.). The high quality and stability of a-IGZO TFT-LCD large-area displays
may dominate the “display world” in the near future.
It should be noted that two great discoveries – the amorphous chalcogenides
(known as a-chalcogenides or a-Chs) used in electrical memory (Ovshinsky,
1968) and switching devices, and device-quality hydrogenated amorphous sil-
icon (a-Si:H) (Spear and LeComber, 1975) – initiated a vast field of science,
which will be briefly reviewed in the following.
The most important scientific issue that resulted from work on amorphous
semiconductors was the structural information gained from experiment and the-
ory (in the form of modeling), because the fundamental physical and chemical
properties are principally determined by their structure. For tetrahedrally bonded
4 Introduction
materials such as amorphous germanium (a-Ge) and a-Si, a hand-built random

network model was first proposed by Polk (1971), in which 440 atoms and bonds
were modeled. Later, computer-generated structures of threefold-coordinated
amorphous arsenic (a-As) and twofold-coordinated a-Se were generated by
Greaves and Davis (1974) and Long et al. (1976), respectively. Molecular
dynamic (MD) simulations are now very popular and are used to understand
microscopic structures (see, for example, Greaves and Sen (2007)). The average
coordination number Z of bonding atoms plays a role in structural properties,
and Phillips (1979) proposed a topological constraint model using Z. The magic
number Zc = 2.4 (Phillips, 1979) or 2.67 (Tanaka, 1989) may dominate opti-
cal and electronic properties in multicomponent glasses. A MD simulation has
been also used for the study of photoinduced structural transformation in a-Se
(Drabold, Zhang, and Li, 2003; Hegedüs et al., 2005).
A model of the electronic density of states (DOS) for non-crystalline semi-
conductors was proposed by Cohen, Fritzshe, and Ovshinsky (1969). The DOS
separated by a bandgap is not sharp, and extends into the bandgap (where it is
known as the band tail). The valence band (VB) originates from the bonding
states in a-Si as it does in crystalline Si. However, the VB in a-Chs is formed
from a lone-pair band. Amorphous chalcogenides are therefore called lone-pair
(LP) semiconductors (Kastner, 1972). In this case, the tailing DOS should be
localized, and hence it is known as the “band tail (localized) states.” The concept
of the mobility edge, which separates the extended and localized states, was then
proposed (see, for example, Mott and Davis, 1979; Mott, 1992). It is believed
that the localized tail states originate mainly from the distortion of the bond
angle, which produces a lack of long-range structural order.
In addition to the localized tail states, there are bonding defects that may pro-
duce midgap states. In a-Si:H, these defects are identified by Si dangling bonds.
These are electronically neutral states. However, in a-Chs, they are believed to
be charged dangling bonds, which may be over- or under-coordinated (Street
and Mott, 1975; Kastner, Adler, and Fritzshe, 1976).
Optical and electronic transport properties are primarily determined by the
electronic DOS. There are no obvious edges to the DOS, and therefore the
bandgap is not easy to define. Tauc (1968) defined the bandgap using optical
absorption with energy space (without using wave vectors), which has led to it
being called the “optical gap” or sometimes the “Tauc gap.” In binary and ternary
a-Chs, a composition-dependent optical gap was found to be well described using
an analogy with mixed crystals (Shimakawa, 1981). Photoluminescence (PL)
1.1 Historical overview 5
is dominated by the DOS. Street (1976) initiated work on PL in this field, and
recently the technique of PL with wide-lifetime distribution (in the nanosecond
to millisecond range) has become well established (Aoki, 2012).
Thermally activated band-type electronic transport occurs in both a-Si:H and
amorphous chalcogenides near room temperature. The Meyer–Neldel compen-
sation law for thermally activated processes has been discussed with reference to
disordered matter; however, this compensation law is not clearly understood from
a theoretical standpoint (Yelon, Movaghar, and Crandall, 2006). At low tempera-
tures, or in defect-dominated materials, transport through localized states is
dominant. Mott (1969) developed the variable-range-hopping (VRH) model for
a single-phonon carrier hopping process. The importance of small polarons
has been pointed out in disordered materials (Emin, 1975), in which multi-
phonon processes dominate transport. Whether multiphonon or single-phonon
processes dominate in amorphous semiconductors is still a matter of debate
(Shimakawa and Miyake, 1988; Emin, 2008). Recent results in hydrogenated
amorphous/nanocrystalline Si suggest that this factor will depend on tempera-
ture (Wienkes, Blackwell, and Kakalios, 2012). Alternating current (ac) transport
studies, initiated by Pollak and Gegalle (1961) in crystalline Si, provide infor-
mation on localized states. Theories involving the continuous time random walk
(CTRW) of carriers have developed well, and apply to hopping systems (Dyre
and Schröder, 2000).
Time-of-flight measurements developed by Spear and Adams (1966) were
applied to crystalline sulfur. In amorphous semiconductors, this technology has
been applied to many systems involving a-Si:H and a-Chs to study the drift
mobility of both electrons and holes. In most cases, non-Gaussian transport
has been discovered. This opened the door for studying dispersive transport in
disordered matter (Pfister and Scher, 1978).
The illumination of amorphous semiconductors, effectively with bandgap
light, induces various changes on structural and electronic properties. In amor-
phous chalcogenides, the bandgap decreases with illumination; this is referred to
as photodarkening, which was first reported by DeNeufville, Moss, and Ovshin-
sky (1973/1974), and it was confirmed that photodarkening accompanied volume
changes (Hamanaka et al., 1976). These effects are of interest in scientific appli-
cations, and hence a huge amount of work has been devoted to these topics
(see the reviews by Tanaka (1990), Pfeiffer, Paesler, and Agarwal (1991), and
Shimakawa, Kolobov, and Elliott (1995)). Photoinduced volume changes in a-Se
films have been discussed using MD simulations (Hegedüs et al., 2005). Giant
6 Introduction
photoexpansion (Hisakuni and Tanaka, 1995) and photodeformations (Tanaka

and Mikami, 2009) have produced the interesting terms photofluidity and optical
force, respectively.
A decrease in the photoconductivity of a-Si:H after photoillumination was first
discovered by Staebler and Wronski (1977), and is called the Staebler–Wronski
(SW) effect. This effect is problematic for applications and hence it is also called
photodegradation. Photodegradation is attributed to defect creation, and under-
standing it is a very important subject (see, for example, Street (1991), Redfield
and Bube (1996), Morigaki (1999), and Singh and Shimakawa (2003)). Similar
effects on photoconductivity have also been found in amorphous chalcogenides
(Shimakawa, 1986).
1.2 Definitions
Crystalline and non-crystalline structure
For the reader unfamiliar with amorphous semiconductors, we present here a
brief explanation of the most essential definitions. Firstly, we must stress that
there is a considerable amount of confusion in the scientific literature concerning
the terms non-crystalline, amorphous, glassy, vitreous, randomness, disorder,
liquid, and even crystalline. A first important question is whether an atomic
structure is crystalline or non-crystalline (amorphous). A perfect crystal is one
in which the atoms, or a group of atoms, are arranged periodically in three
dimensions to an infinite extent and are rigidly fixed at their thermal equilibrium.
This mathematical model of atomic configurations provides us with a relatively
easy method for calculating the different properties of condensed matter, which
may be found in books on theoretical solid state physics.
A more realistic arrangement is an imperfect crystal in which the atoms
form a pattern that repeats periodically only to a finite extent. In fact, this may
seem more realistic, but we still have a serious problem in that this type of
imperfect crystal behaves counter to thermodynamics. This is because, at non-
zero temperatures, defects form in any atomic configuration when the crystals
are in their equilibrium states. Real crystals are, of course, not only finite in
size, but also contain imperfections such as vacancies, interstitial (foreign or
self) atoms, dislocations, impurities, distortions associated with the surface, etc.
Furthermore, another effect must be taken into account at finite temperature. The
random motion of atoms at their equilibrium positions also weakly destroys the
perfect periodicity at any given moment in time. These defects cause distortions
1.2 Definitions 7
in the crystal lattice, but we do not consider these real crystals to be amorphous
solids. Translational symmetry more or less remains.
Until 1992, a crystal was defined by the International Union of Crystallog-
raphy as “a substance, in which the constituent atoms, molecules, or ions are
packed in a regularly ordered, repeating three-dimensional pattern.” In 1984 (two
years after their discovery), Shechtman et al. (1984) published a paper on rapidly
solidified alloys of aluminum with 10–14% manganese that possess icosahedral
symmetry in combination with long-range order, named quasicrystals, in clear
violation of the above definition of a crystal. Since this discovery, hundreds of
similar atomic structures have been synthesized in laboratories around the world.
Recently, the naturally occurring quasicrystalline mineral icosahedrite has been
identified in a sample from the Khatyrka river in Chukotskii Autonomous Okrug,
Far-Eastern Region, Russia. The International Union of Crystallography has had
to modify their declaration, and the new and broader definition of crystal became
“any solid having an essentially discrete diffraction diagram.”
Other types of deviation from perfection occur in crystals. Consider a perfect
lattice, but one in which each atomic site possesses a randomly oriented spin
or other magnetic moment. Another partially ordered arrangement is possible
in materials built up of large near-spherical molecules. Molecular centers of
mass can be found in crystalline sites where the directions of molecular axes
are randomly distributed. A typical example of the so-called plastic crystals is
the solid crystalline phase of C60 molecules. A similar example occurs when
a binary (ternary) alloy has two (three) different, randomly distributed atoms
inside a perfect lattice. Are they crystalline or non-crystalline? This issue does
not occur for amorphous materials.
A liquid crystal is a state of matter that has properties between those of
a conventional liquid and those of a solid crystal. Rigid rod-shaped or disc-
like molecules develop in a well-defined direction. The centers of mass of the
molecules are distributed randomly in two or three dimensions. How does one
distinguish between the condensed phase and the liquid phase? A solid is a phase
whose shear viscosity exceeds 1013.6 Pa s (N s m−2 ), although this threshold is
rather arbitrary. Values for the viscosity of common liquids at room temperature
are on the order of 10–3 Pa s. For some examples, see Table 1.1. Liquid crystals
are anisotropic fluids.
A granular material is a collection of distinct macroscopic particles with
randomly distributed centers of mass, such as sand or peanuts in a container.
There is a weak interaction between macroscopic particles. Their behavior is
8 Introduction
Table 1.1. Shear viscosity values for some materials.
Substance Temperature (°C) Shear viscosity (Pa s)

Water 90 0.31
Water 10 1.31
Honey room temperature 2–10
Bitumen 20 108
Ice –13 15 × 1012
Window glass room temperature 1040
differently from that of solids, liquids, and gases, and this has led many to
characterize granular materials as a new form of matter. A similar topology
can be observed in the “microworld.” Polycrystalline materials (poly is a Greek
word meaning “many”) consist solely of crystalline grains, separated by grain
boundaries, and the interaction between them is strong. Furthermore, in some
cases crystalline grains do not touch each other; rather, they are embedded
into a disordered atomic environment. The grain size determines whether these
materials are named nanocrystalline or microcrystalline materials.
Amorphous, glassy materials

After our brief overview, we can provide the following definition for amorphous
materials: amorphous materials exist in the condensed phase and do not possess
the long-range translational order (periodicity) of atomic structures. By “long
range” we mean over 10 nm. We may also talk about short-range order (also
called “local order,” below 1 nm) and medium-range order (in the range 1–
10 nm). Following the discovery of quasicrystals, another sentence must be
added to the previous definition of amorphous materials.
Amorphous materials exist in the condensed phase and do not possess the long-
range translational order (periodicity) of atomic structures. They do not exhibit a
discrete diffraction pattern.
Using this terminology, the set of amorphous materials has within it a funda-
mental subset glasses, where a glass is an amorphous solid that exhibits a glass
transition. If a liquid (melt) is cooled very rapidly so that crystallization can
be bypassed, the disordered structure can be frozen-in. This disordered con-
densed phase is known as a glass. Such a glass-forming process involves the
1.2 Definitions 9
supercooling of a liquid below its normal freezing point. The freezing of

a crystalline solid is a first-order thermodynamic phase transition; there are
discontinuities in first-order thermodynamic variables such as entropy, S =
– (dGdT )p , or volume, V = (dGdp)T , at the transition (where G is the Gibbs
free energy, G(T, p, N ) = E – TS + pV ). The transformation from a melt to
the glassy phase is a transition in which there is no discontinuity in first-order
thermodynamic variables at the glass-transition temperature, Tg . However, in
second-order thermodynamic variables, such as the calorimetric heat capacity
at constant pressure, Cp = T(dSdT )p = – T(d2 GdT 2 )p , a discontinuity can be
observed.
Groups IV and VI elements and their alloys

For us, the most interesting class of materials comprises the amorphous semi-
conductors. An important task therefore is to define a semiconductor. A simple
definition would be that a semiconductor belongs to a class of materials that have
an energy gap of 1–2 eV in their electronic DOS and whose resistivity values fall
between that of an insulator and a good conductor. A more accurate definition
of crystalline semiconductors is associated with the temperature dependence of
resistivity:
ρ (T ) = ρ0 exp (ε0 /kB T ) , (1.1)
where ρ 0 and ε 0 are constants, kB is the Boltzmann factor, and T is the tempera-
ture. Note that ρ decreases with increasing T, which is opposite to the case for
conventional metals (where ρ is proportional to T). Most chalcogenide glasses
and hydrogenated a-Si follow the above equation, whereas pure a-Si and a-Ge
do not.
Atomic structure is one of the main features that distinguish between different
electronic transport properties. Carbon provides us with a simple example. The
diamond crystal arrangement of carbon atoms creates a good insulator, with
about a 5.5 eV gap in the electronic DOS. However, the non-crystalline config-
uration of carbon displays semiconductor properties, and the graphitic atomic
structure is a conductor. Furthermore, superconductivity at 30 K in caesium-
doped C60 has been observed.
Pure amorphous semiconductor materials are located in the even-numbered
groups of the periodic table, such as group IV (carbon, silicon, and germanium)
10 Introduction
and group VI (selenium and tellurium). In group VI, stable allotropes of sulfur
are excellent electrical insulators.
The basis of this structure was formulated by Sir Nevill Francis Mott (Mott,
1969), who concluded via his famous (8 – N) rule that there are no dangling bonds
inside such a material. This is explained as follows. In a glass, any atom appears
in such a way that it retains its natural coordination. The number of covalent
bonds Z = 8 – N, where N is the number of valence electrons. (We consider
elements only in groups IV, V, and VI.) Further, an additional rule states that
Z = N if N < 4. The most important consequence of this was that amorphous
semiconductors cannot be doped. Today, however, we know this is not the
case.
References
Aoki, T. (2012). Photoluminescence spectroscopy. In Characterization of Materials,
2nd edn., ed. E. N. Kaufmann. Hoboken, NJ: John Wiley and Sons, pp. 1158–
1169.
Carlson, D.E. and Wronski, C.R. (1976). Amorphous silicon solar cell. Appl. Phys. Lett.,
28, 671–673.
Carlson, D.E., Arya, R.R., Bennett, M. et al. (1996). Commercialization of multijunction
amorphous silicon modules, Conf. Record of the Twenty Fifth IEEE, Photovoltaic
Specialists Conf., Crystal City, Washington D.C., May 13–17, 1996, pp. 1023–1028.
Cohen, M.H., Fritzshe, H., and Ovshinsky, S.R. (1969). Simple band model for amor-
phous semiconducting alloys. Phys. Rev. Lett., 22, 1065–1068.
DeNeufville, J.P., Moss, S.C., and Ovshinsky, S.R. (1973/1974). Photostructural trans-
formations in amorphous As2 Se3 and As2 S3 films. J. Non-Cryst. Solids, 13, 191–
223.
Drabold, D.A., Zhang, X., and Li, J. (2003). First-principles molecular dynamics and
photostructural response in amorphous silicon and chalcogenide glasses. In Photo-
Induced Metastability in Amorphous Semiconductors, ed. A.V. Kolobov. Weinheim:
Wiley-VCH, pp. 260–278.
Dyre, J.C. and Schröder, T.B. (2000). Universality of AC conduction in disordered
solids. Rev. Mod. Phys., 72, 873–892.
Emin, D. (1975). Phonon-assisted transition rate: optical-phonon-assisted hopping in
solids. Adv. Phys., 24, 135–218.
Emin, D. (2008). Generalized adiabatic polaron hopping: Meyer-Neldel compensation
and Poole-Frenkel behaviour. Phys. Rev. Lett., 100, 16602–16604.
Feinleib, J., deNeufville, J., Moss, S.C., and Ovshinsky, S.R. (1971). Rapid reversible
light-induced crystallization of amorphous semiconductors. Appl. Phys. Lett., 18,
254–257.
Greaves, G.N. and Davis, E.A. (1974). A continuous random network model with three-
fold coordination. Philos. Mag., 29, 1201–1206.
Greaves, G.N. and Sen, S. (2007). On organic glasses, glass-forming liquids and amor-
phizing solids. Adv. Phys., 56, 1–166.
References 11
Hamanaka, H., Tanaka, K., Matsuda, A., and Iizima, S. (1976). Reversible photo-induced
volume changes in evaporated As2 S3 and As4 Se5 Ge1 films. Solid State Commun.,
19, 499–501.
Hegedüs, J., Kohary, K., Pettifor, D.G., Shimakawa, K., and Kugler, S. (2005). Photo-
induced volume changes in amorphous selenium. Phys. Rev. Lett., 95, 206803, 1–
4.
Hisakuni, H. and Tanaka, K. (1995). Optical microfabrication of chalcogenide glasses.
Science, 270, 217–218.
Hosono, H. (2006). Ionic amorphous oxide semiconductors: material design, carrier
transport, and device application. J. Non-Cryst. Solids, 352, 851–858.
Juška, G., Arlauskas, K., and Montrimas, E. (1987). Features of carriers at very high
electric fields in a-Se and a-Si:H. J. Non-Cryst. Solids, 97–98, 559–561.
Kastner, M.A. (1972). Bonding bands, lone-pair bands, and impurity states in chalco-
genide semiconductors. Phys. Rev. Lett., 28, 355–357.
Kastner, M.A., Adler, D., and Fritzshe, H. (1976). Valence-alternation model for local-
ized gap states in lone-pair semiconductors. Phys. Rev. Lett., 37, 1504–1507.
Long, M., Galison, P., Alben, R., and Connell, G.A.N. (1976). Model for structure of
amorphous selenium and tellurium. Phys. Rev. B, 13, 1821–1829.
Morigaki, K. (1999). Physics of Amorphous Semiconductors. London: Imperial College
Press and World Scientific.
Mott, N.F. (1969). Conduction in non-crystalline materials. Phil. Mag., 19, 835–852.
Mott, N.F. (1992). Conduction in Non-Crystalline Materials, 2nd edn. Oxford: Claren-
don Press.
Mott, N.F. and Davis, E.A. (1979). Electronic Processes in Non-Crystalline Materials,
2nd edn. Oxford: Clarendon Press.
Ovshinsky, S.R. (1968). Reversible electrical switching phenomena in disordered struc-
tures. Phys. Rev. Lett., 21, 1450–1453.
Pai, D.M. and Springett, B.E. (1993). Physics of electrophotography. Rev. Mod. Phys.,
65, 163–211.
Pfeiffer, G., Paesler, M.A., and Agarwal, S.C. (1991). Reversible photodarkening of
amorphous arsenic chalcogens. J. Non-Cryst. Solids, 130, 111–143.
Pfister, G. and Scher, H. (1978). Dispersive (non-Gaussian) transient transport in dis-
ordered solids. Adv. Phys., 27, 747–798.
Phillips, J.C. (1979). Topology of covalent non-crystalline solids I: short-range order in
chalcogenide alloys, J. Non-Cryst. Solids, 34, 153–181.
Polk, D.E. (1971). Structural model for amorphous silicon and germanium. J. Non-Cryst.
Solids, 5, 365–376.
Pollak, M. and Gegalle, T.H. (1961). Low-frequency conductivity due to hopping pro-
cesses in silicon. Phys. Rev., 122, 1742–1753.
Powell, M.J. (1984). Material properties controlling the performance of amorphous
silicon thin film transistors. MRS Symp. Proc., 33, 259–274.
Rowlands, J.A. and Kasap, S.O. (1997). Amorphous semiconductors usher in digital
X-ray imaging. Phys. Today, 50, 24–31.
Redfield, D. and Bube, R.H. (1996). Photoinduced Defects in Semiconductors.
Cambridge: Cambridge University Press.
Selényi, P. (1935a). Elektrographie, ein neues elektrostatisches Aufzeichnungsverfahren
und seine Anwendungen. Elektrotech. Z., 56(35), 961–963.
Selényi, P. (1935b). Methoden, Ergebnisse und Aussichten des elektrostatischen Auf-
zeichnungsverfahrens. (Elektrographie). Z. Tech. Phys., 12, 607–614.
12 Introduction
Selényi, P. (1936). Electrostatic recording (electronography). Electronics, Apr., 44–

46.
Shechtman, D., Blech, I., Gratias, D., and Cahn, J.W. (1984). Metallic phase with
long-range orientational order and no translational symmetry. Phys. Rev. Lett., 53,
1951–1953.
Shimakawa, K. (1981). On the compositional dependence of the optical gap in amor-
phous semiconducting alloys. J. Non-Cryst. Solids, 43, 229–244.
Shimakawa, K. (1986). Persistent photocurrent in amorphous chalcogenides. Phys. Rev.
B, 34, 8703–8708.
Shimakawa, K. and Miyake, K. (1988). Multiphonon tunnelling conduction of localized
π electrons in amorphous carbon films. Phys. Rev. Lett., 61, 994–996.
Shimakawa, K., Kolobov, A.V., and Elliott, S.R. (1995). Photoinduced effects and
metastability in amorphous semiconductors and insulators. Adv. Phys., 44, 475–
588.
Singh, J. and Shimakawa, K. (2003). Advances in Amorphous Semiconductors. London
and New York: CRC Press.
Spear, W.E. and Adams, A.R. (1966). Photogeneration of charge carriers and related
optical properties in orthorhombic sulphur. J. Phys. Chem. Solids, 27, 281–290.
Spear, W.E. and LeComber, P.G. (1975). Substitutional doping of amorphous silicon.
Solid State Commun., 17, 1193–1196.
Spear, W.E. and LeComber, P.G. (1984). The development of the a-Si: H field-effect
transistor and its possible applications. Semicond. Semimetals, 21D, 89–114.
Staebler, D.L. and Wronski, C.R. (1977). Amorphous silicon solar cell. Appl. Phys.
Lett., 28, 671–673.
Street, R.A. (1976). Luminescence in amorphous semiconductors. Adv. Phys., 25, 397–
453.
Street, R.A. (1991). Hydrogenated Amorphous Silicon. Cambridge: Cambridge Univer-
sity Press.
Street, R.A. and Mott, N.F. (1975). States in the gap in glassy semiconductors. Phys.
Rev. Lett., 35, 1293–1296.
Tanaka, K. (1989). Structural phase transitions in chalcogenide glasses. Phys. Rev. B,
39, 1270–1279.
Tanaka, K. (1990). Photoinduced structural changes in chalcogenide glasses. Rev. Solid
State Sci., 4, 641–659.
Tanaka, K. and Mikami, M. (2009). Photo-induced deformations in chalcogenide glass.
J. Opt. Adv. Mater., 11, 1885–1890.
Tanioka, K. (2007). The ultra sensitive TV pickup tube from conception to recent
development. J. Mater. Sci., 18, S321–S325.
Tauc, J. (1968). Optical properties and electronic structure of amorphous Ge and Si.
Mater. Res. Bull., 3, 37–46.
Weiss, D.S. and Abkowitz, M. (2006). Organic photoconductors. In Springer Handbook
of Electronic and Photonic Materials, eds. S.O. Kasap and P. Capper. New York:
Springer, chap. 39.
Wienkes, L.R., Blackwell, C., and Kakalios, J. (2012). Electronic transport in doped
mixed-phase hydrogenated amorphous/nanocrystalline silicon thin films. Appl.
Phys. Lett., 100, 072105, 1–3.
Wright, C.D., Yanwei, L., Kohary, K.I., Aziz, M.A., and Hicken, R.J. (2011). Arithmetic
and biologically-inspired computing using phase-change materials. Adv. Mater., 23,
3408–3413.
References 13
Yamada, N., Ohno, E., Nishiuchi, K., Akahira, N., and Takao, M. (1991). Rapid phase-
transitions of GeTe-Sb2 Te3 pseudobinary amorphous thin films for an optical disk
memory. J. Appl. Phys., 69, 2849–2856.
Yelon, A., Movaghar, B., and Crandall, R.S. (2006). Multi-excitation entropy: its role in
thermodynamics and kinetics. Rep. Prog. Phys., 69, 1145–1194.
2
Preparation techniques
Amorphous semiconductors are prepared via non-equilibrium processes. Many

preparation techniques are possible, depending on what kinds of materials are
required for research and/or application. Quenching from the liquid state (other-
wise known as melt-quenching) is a popular technique used for so-called glasses
and glass fibers. When thin films are required, techniques such as evaporation,
sputtering, and chemical vapor deposition (CVD) are adopted. Ion bombardment
and high-intensity light directed into crystalline solids also produces amorphous
materials. In this chapter, we discuss various methods used in the preparation of
amorphous semiconductors. As the nature of glass and amorphous semiconduct-
ors is closely related to the structural constraints of the materials, the Phillips
constraint theory will also be discussed.
2.1 Growth of thin-film forms

Amorphous Si, Ge, C, and related compounds are prepared via condensation
from the gas phase, which usually produces thin-film forms. We briefly review
in this section the most popular preparation techniques for the growth of thin-
film forms of amorphous semiconductors. These techniques can be roughly
classified into two categories: physical and chemical depositions of thin films.
Thermal evaporation and sputtering are physical vapor deposition (PVD) tech-
niques, and the well-known glow-discharge technique comes under chemical
vapor deposition. We can clearly distinguish between CVD and PVD tech-
niques. In CVD, there is no target (source material) present: the gas is intro-
duced and decomposed in some way, and these decomposed materials are then
deposited onto a substrate as a thin film. The CVD techniques discussed here
include “glow-discharge,” which is called now plasma-enhanced chemical vapor
14
deposition (PECVD), and hot-wire chemical vapor deposition (HWCVD), which

uses thermal energy. We discuss all of these techniques later, but let us start with
PVD.
Thermal evaporation
This is the most conventional method used in preparing amorphous materials.
The starting materials comprise ingots or powders, which are mounted in a
sample boat and melted in a vacuum chamber at around 10–6 Torr (10–4 Pa).
To achieve this low pressure, an oil diffusion pump is used in conjunction with a
liquid-nitrogen-cooled trap. Note that the boat itself acts as the electrical heater.
High-melting-point metal boats, made for example from molybdenum (Mo) or
tungsten (W), are normally used. This technique is suitable for low-melting-
point materials, and the vapor is condensed at arbitrary temperatures onto the
desired substrates. The deposition rate, which is usually about 10 nm s−1 , is
controllable and depends on the boat temperature. The thickness of the film can
be controlled by the deposition time. Well-known amorphous chalcogenides, for
example Se, As2 Se(S)3 , GeSe(S)2 , etc., can be prepared using this technique. The
main disadvantage of thermal evaporation is in ensuring that the composition of
the resulting films is the same as that of the starting materials.
Device-quality a-Se thick (>1 mm!) films, which are commercially utilized
for direct x-ray image detectors (Rowlands and Kasap, 1997) and high-gain
avalanche rushing amorphous photoconductor (HARP) vidicon TV cameras
(Tanioka, 2007), are prepared by the traditional thermal evaporation technique.
Electron-beam and arc evaporation

Materials with higher melting points, for example, Si, C, and SiC, are not easy
to vaporize, and hence a simple evaporation technique is not useful for these
materials. Instead of heating the substances in boats, as in thermal evaporation,
high-energy electron bombardment (from an electron gun placed in the vacuum
chamber) is employed to vaporize such high-melting-point materials.
Arc discharge is also used as a heat source for electrically conducting ma-
terials, such as graphitic carbons (Shimakawa et al., 1991). For example, a high
current is passed through two pieces of graphite, causing the area to heat up.
This process is carried out either in a vacuum or in an atmosphere of Ar or He.
(In He, the deposited films contain soccer-ball-shaped C60 molecules or carbon
nanotubes.) The deposition rate for this method is around 10 nm s−1 .
16 Preparation techniques
Sputtering
The usual sputtering process comprises a radio-frequency (rf, 13 MHz) high-
electric field applied between the target (source material) and the substrate
electrodes. A sputtering gas (0.13–2.7 Pa) (for example, Ar) is introduced into a
chamber and is pumped away by a vacuum system. The substrate temperature is
easy to control. Positively ionized gases produced by the rf field supply kinetic
energy to the atoms on the surface of the source material target. Atoms or
molecules dissociated from the target are deposited onto the substrate in the form
of films. Reactive gases such as H2 and/or N2 are also used for sputtering, in a
technique known as reactive sputtering, in which the gases are incorporated into
the deposited films (for example, a-Si:H, a-C:H, a-Si1–x Nx , etc.). The deposition
rate is lower than that for evaporation techniques, i.e. less than 1 nm s−1, although
this will depend greatly on the sputtering conditions.
The sputtering process is more complicated than thermal evaporation; for
example, we must pay attention to the sputtering gas pressure, the rf power
applied to the target, the bias voltage of the target or substrate, etc. However,
sputtering is superior to evaporation for multicomponent systems, and the com-
position of the films is almost the same as that of the source materials (targets)
because the sputtering rates for different elements are on the same order.
When a magnetic field is applied perpendicular to the electrodes, deposition is
enhanced; this is because the ionized gases are confined in a spiral motion by the
magnetic field (Lorentz force). This method is called magnetron sputtering, and
it significantly improves the deposition rate. Magnetron sputtering is therefore
suitable and cost-effective for large-scale production, and is used for optically
transparent conductive oxides (TCOs) such as indium tin oxide (ITO) and semi-
conducting IGZO for the flat-panel display industry (Kamiya, Nomura, and
Hosono, 2009). Note that device-quality a-IZGO films are also prepared using
pulsed laser deposition, which will be described later. So called phase-change
materials, such as the chalcogenide Ge2 Sb2 Te5 used in optical memory devices
(such as DVDs), are also prepared using the magnetron sputtering technique.
Plasma-enhanced chemical vapor deposition

This technique was previously known as glow-discharge deposition, whereby
hydrogenated amorphous silicon (a-Si:H) films were deposited by means of
the decomposition of SiH4 gas with the help of glow-discharge (Spear and
LeComber, 1975). It is now called plasma-enhanced chemical vapor deposition
(PECVD) (Schiff, Hegedus, and Deng, 2010). The plasma is introduced via the
application of a rf field (usually 13.6 MHz), and is therefore referred to as rf
PECVD. To prepare n or p type Si:H films, PH3 or B2 H6 gas is introduced onto
SiH4 . Radio-frequency PECVD is a very popular technique and provides high-
quality uniform films, although the deposition rate can be as low as 0.3 nm s−1 .
The development of this technique, by controlling the gas pressure (0.05–2 Torr),
rf power (10–100 mW cm−2 ), and substrate temperature (150–350 °C) etc.,
has led to growing numbers of commercial applications, such as large-area
photovoltaics and thin-film transistors.
Radio-frequency PECVD is also widely used in the preparation of hydro-
genated amorphous carbon (a-C:H) thin films and nanoparticles. Hydrocarbons,
such as CH4 or C6 H6 , are the main precursor materials. Changing the pressure
and the rf power allows the preparation of a-C films with remarkably different
characters, ranging from soft and porous polymer-like a-C:H and hard diamond-
like carbon layers to highly conductive graphitic structures (Veres, Tóth, and
Koós, 2008). At low ion energies, intact molecules of the feed gas were found
to be incorporated into the matrix of the amorphous films (Veres, Koós, and
Pócsik, 2002). Hydrogenated amorphous carbon nanoparticles are prepared by
using high gas pressures together with low ion energies and a benzene pre-
cursor (so-called “dusty plasma”) (Pócsik et al., 2003). Specific dusty-plasma
conditions allow amorphous nanoparticles and thin films to be obtained simul-
taneously (Veres et al., 2005).
When PECVD is carried out at very high frequencies (40–100 MHz) it is
called VHF PECVD; this technique improves the deposition rate, which can
reach 2 nm s−1 . VHF PECVD also produces high-quality films, although poor
uniformity is a problem. Microwave PECVD performed at 2.45 GHz (MW
PECVD) significantly enhances the deposition rate (up to 10 nm s−1 ), although
device-quality a-Si:H films are not easy to obtain. Note, however, that VHF
PECVD and MW PECVD are very useful in the preparation of microcrystalline
Si films (µc-Si:H), as the deposition rate of VHF PECVD for µc-Si:H is very
small. A high deposition rate for low-cost mass production can be a necessary
condition.
Hot-wire chemical vapor deposition

If we are interested in achieving a very high deposition rate in a-Si:H and
related films, hot-wire chemical deposition (HWCVD) is a promising technique,
with deposition rates reaching up to 15–30 nm s−1 (Mahan et al., 2001). The
set-up for a HWCVD system is similar to that for rf PECVD, except that the rf
electrode is replaced by a heated filament. The gas introduced into the chamber
is catalytically excited, or decomposed into radicals or ions, by a metal (Pt, W,
Ta, etc.) filament heated to around 1800–2000 °C. Then, for example, Si radicals
diffuse inside the chamber and are deposited onto a substrate that is maintained
at a relatively high temperature (150–450 °C). Although device-quality films
with high deposition rates are obtained using this technique, a drawback is the
poor uniformity of deposited films.
Pulsed laser deposition

Pulsed laser deposition (PLD) is very simple when compared with sputtering
and the various CVD techniques described so far. A pulsed high-power laser
beam is focused onto a target in a vacuum chamber causing evaporation of the
target source material. This resembles electron-beam evaporation, but in PLD a
directional plasma plume is created by the absorption of photons, and hence this
method is distinguished from simple thermal evaporation (Delahoy and Guo,
2010). The pulse energy is 300–600 mJ pulse−1 , and the instantaneous power
density reaches 109 W cm−2 . Excimer gas (KrF or ArF) lasers are used for this
purpose, as are solid state Nd:YAG lasers. A substrate faces the target, and the
ablated source material condenses on it and grows the film. Stoichiometric trans-
fer of atoms from multicomponent source materials to the substrate is performed
by the PLD technique. It is now widely used to create device-quality transparent
and conductive oxide (TCO) films, in particular commercially available IGZO
films (Kamiya et al., 2009).
2.2 Melt-quenched glasses

Most oxide and chalcogenide glasses can be prepared via the melt-quenched
(MQ) method. Techniques used to prepare glass fibers are, in principle, based on
the MQ method. When the temperature of the melt (the liquid state) decreases,
melts become supercooled liquids, and a further decrease in temperature below
the glass-transition temperature Tg produces glassy materials. The terms glass
or glassy material are usually applied to materials prepared via the MQ method.
Thus, glasses prepared by the MQ method undergo the glass transition. It is
of interest to note that the empirical relation Tg 2Tm 3 is reported in most
glasses (Mott and Davis, 1979), where Tm is the melting temperature. When
2.3 Other techniques 19
a glassy material is heated, crystallization occurs at a temperature Tc between

Tg and Tm . In the laboratory, examples of chalcogenide glasses are sealed in
quartz ampoules at 10–6 Torr and heated beyond their melting temperature. The
samples are agitated (through rotation and vibration) then cooled quickly by
immersion in cold water or air (at ambient temperature). Glasses are formed
due to this high cooling rate (Elliott, 1990; Tanaka and Shimakawa, 2011). The
ability under normal conditions of materials to form glasses depends also on the
compositions of the materials (see, for example, Borisova (1981) and Popescu
(2000)).
The nature of the glass transition is an important physical parameter for melt-
quenched glasses (Elliott, 1990; Tanaka and Shimakawa, 2011), and hence Tg
will be discussed in more detail in Chapter 3. It should be also noted that Tg
and Tc are important parameters for phase-change materials, and therefore a
detailed discussion, together with phase-change behavior, will also be given in
Chapter 3.
2.3 Other techniques

There are other techniques that produce amorphous states (Elliott, 1990). One is
irradiation-induced amorphization, which we introduce briefly here. The inter-
action between high-energy ionizing particles and crystalline solids can produce
heavy structural damage, leading eventually to amorphous states. We know, for
example, that almost all a-Si films made by the deposition techniques from gas
states mentioned in Section 2.2 contain significant voids that may extend to
several nanometers. These voids disturb the analysis of physical properties such
as the radial distribution function (RDF), and hence pure and void-free a-Si is
required.
To achieve this, we employ self-ion implantation at MeV energies of crys-
talline silicon (c-Si) at 77 K, i.e. Si itself is implanted into c-Si to produce pure
and void-free a-Si (Laaziri et al., 1999). After bombardment, the Si wafer is
chemically etched from the back surface to remove c-Si; an amorphous layer
of approximately 10 µm is obtained with this method. Similar implantations
have been performed in group III–V compounds such as GaAs, GaSb, InAs,
and InSb, with thicknesses around 2.5 µm obtained (Ridgway et al., 2003).
Amorphization of c-Ge substrates under ion implantation (up to 1016 cm–2 )
has also been studied using transmission electron microscopy (TEM) (Koffel
et al., 2006) and Rutherford backscattering spectroscopy (RBS) (Impellizzeri,
Mirabella, and Grimaldi, 2011). It is known that the Ge surface undergoes a

topology change and the cellular structure appears.
Finally, we introduce a special preparation method that is not popular in
this field: mechanical milling (MM) (Koch et al., 1983; Schwarz and Koch,
1986) in the solid state. The rapid quenching technique is based on the rapid
removal of kinetic energy in an energized state. Mechanical milling comprises
a homogenization process up to the atomic level by chemical reaction in a
system with a negative heat of mixing. In fact, the amorphization of trigonal
Se can proceed by mechanical milling (Fukunaga et al., 1996). The number of
atoms located at the second-nearest-neighbor distance (r = 0.34 nm) in the RDF
is known to be reduced from four to zero, leading to the conclusion that the
bonding between chain molecules in trigonal Se is destroyed by the strain and
defects induced by the milling. Interestingly, a-Se prepared by MM exhibits a
glass transition at the same temperature (320 K) as that occurring in MQ-Se,
whereas crystallization occurs at a lower temperature (370 K) than that for
MQ-Se (400 K).
Mechanical milling proceeds through the destruction of the crystal structure
of an intermetallic compound or pure element due to an accumulation of strains
and defects. Note that as the starting material (before MM) is energetically in the
lowest (crystalline) state, the “uphill” process should be involved in the initial
stage, during which energy is given mechanically to the material system. Chain
molecules will be destroyed first by the strain, and defects are induced at this
stage.
2.4 The Phillips constraint theory

The formation of glasses is highly dependent on network constraints, and hence
we discuss the theory here. According to the Phillips constraint theory (Phillips,
1979), ideal glass-forming conditions occur when the average number of inter-
atomic force-field constraints equals the number of degrees of freedom per atom.
An atom that has all covalent bonds satisfied obeys Mott’s (8 – N) rule (Mott,
1969), i.e. Se has Nc = 2, As and P have Nc = 3, Si has Nc = 4, etc. For a binary
alloy Ax B1–x , the average coordination is given by
Z = xNc (A) + (1 − x) Nc (B) , (2.1)
where Nc (A) is the coordination number of atom A. The number of stretched

bonds is equal to Z2 because every sigma bond belongs to two atoms. A new
2.4 The Phillips constraint theory 21
bond direction is derived by the addition of two bond angles. Therefore the bond-
bending number is 2Z – 3. The total number of linearly independent constraints
Nc = Z2 + (2Z – 3). By the Phillips theory, Nc must be equal to the degrees of
freedom. In the three-dimensional (3D) case, Nc = Nd = 3, so
3 = Z/2 + (2Z − 3) , (2.2)
and we obtain Z = 2.4.

In systems such as a-GeS, a-GeSe, a-SiS, a-SiTe, etc., group IV elements
are connected to four neighbors and group VI elements are connected to two.
For a-AIV x BVI (1–x) , we have 4x + 2(1 – x) = 2.4, so x = 0.2, i.e. a-GeS4 , a-
GeSe4 , a-SiS4 , a-SiSe4 , and a-SiTe4 are the optimum and mechanically most
stable compositions. It is noted again that GeS2 is the chemically most stable
composition. In group V–VI binary alloy systems such as a-AsS, a-AsTe, etc.
(group V elements have three neighbors), we obtain x = 0.4 from 3x + 2(1 – x) =
2.4, and we conclude that a-As2 S3 , a-As2 Se3 , etc. are the optimum compositions.
For amorphous semiconductors with Z < 2.4, the structure is “underconstrained”
or “floppy” and for Z > 2.4 the structure is “overconstrained” or “rigid”.
Some chalcogenide glasses have an average coordination number of Z = 2.67.
The constraint for an atom in the two-dimensional (2D) plane (Nd = 2) is defined
as Nc = Z2 + (Z – 1) = 3 because a new sigma bond is defined by one angle
(Tanaka, 1989). The following general equation may be written for 2D and 3D
cases:
Z (Nd −1) (2Z − Nd )

Nc = + . (2.3)
2 2
Then we obtain a general expression for the average coordination number as

follows:
2Nc + (Nd −1) Nd

Z= . (2.4)
1 + 2 (Nd −1)
The SiO system is a well-known exception to the Phillips theory, with SiO2
instead of SiO4 being the optimum glass composition. This is because the Si–
O–Si bond-angle distribution is rather wide (Kuo, Lee, and Hwang, 2008), and
therefore the constraint associated with oxygen bond angles is weak and may
be neglected. Consider a Six O(1–x) binary alloy. In the equation 3 = Z2 + (2Z –
3), the (2Z – 3) term associated with the bond angles must be modified as there
is no zero-bond-angle constraint for oxygen. The revised term for bond bending
is therefore
(2ZSi −3) x + 0 (1 − x) = 5x
(because ZSi = 4). We must thus solve the following equation:
3 = Z/2 + 5x,
where the average coordination number is again given by Z = 4x + 2(1 – x). We

obtain x = 13, and hence SiO2 is the best glass-forming composition.
References
Borisova, Z.U. (1981). Glassy Semiconductors. New York: Plenum.
Delahoy, A.E. and Guo, S. (2010). Transparent conducting oxides for photovoltaics. In
Handbook of Photovoltaic Science and Engineering, 2nd edn., eds. A. Luque and
S. Hegedus. Chichester: John Wiley and Sons, pp. 716–796.
Elliott, S.R. (1990). Physics of Amorphous Materials, 2nd edn. Harlow: Longman Sci-
ence & Technical.
Fukunaga, T., Utsumi, M., Akatsuka, H., Misawa, M., and Mizutani, U. (1996). Structure
of amorphous Se prepared by milling. J. Non-Cryst. Solids, 205–207, 531–535.
Impellizzeri, G., Mirabella, S., and Grimaldi, M.G. (2011). Ion implantation damage
and crystalline-amorphous transition in Ge. Appl. Phys. A, 103, 323–328.
Kamiya, T., Nomura, K., and Hosono, H. (2009). Origins of high mobility and low
operation voltage of amorphous oxide TFTs: electronic structure, electron transport,
defects and doping. J. Disp. Technol., 5, 468–482.
Koch, C.C., Calvin, O.B., Macklamey, C.G., and Scarbrough, J.O. (1983). Preparation of
‘‘amorphous’’ Ni60 Nb40 by mechanical alloying. Appl. Phys. Lett., 43, 1017–1019.
Koffel, S., Claverie, A., BenAssayag, G., and Scheibin, P. (2006). Amorphization kinetics
of germanium under ion implantation. Mater. Sci. Semicond. Process., 9, 664–667.
Kuo, C-L., Lee, S., and Hwang, G.S. (2008). Strain-induced formation of surface defects
in amorphous silica: a theoretical prediction. Phys. Rev. Lett., 100, 076104, 1–4.
Laaziri, K., Kycia, S., Roorda, S. et al. (1999). High resolution radial distribution
function of pure amorphous silicon. Phys. Rev. Lett., 82, 3460–3463.
Mahan, A., Xu, Y., Nelson, B. et al. (2001). Saturated defect densities of hydrogenated
amorphous silicon grown by hot-wire chemical vapour deposition at rates up to
150 Å. Appl. Phys. Lett., 78, 3788–3790.
Mott, N.F. (1969) Conduction in non-crystalline materials. Phil. Mag., 19, 835–852.
2nd edn. Oxford: Clarendon Press.
Phillips, J.C. (1979). Topology of covalent non-crystalline solids I: short–range order in
chalcogenide alloys. J. Non-Cryst. Solids, 34, 153–181.
Pócsik, I., Veres, M., Füle, M. et al. (2003). Carbon nano-particles prepared by ion-
clustering in plasma. Vacuum, 71, 171–176.
Popescu, M.A. (2000). Non-Crystalline Chalcogenides. Dordrecht: Kluwer.
References 23
Ridgway, M.C., Azevedo, G. de M., Glover, C.J., Yu, K.M., and Foran, G.J. (2003).
Common structure in amorphised compound semiconductors. Nucl. Inst. Meth.
Phys. Res. B, 199, 235–239.
Rowlands, J.A. and Kasap, S.O. (1997). Amorphous semiconductors usher in digital
X-ray imaging. Phys. Today, 50, 24–31.
Schiff, E.A., Hegedus, S., and Deng, X. (2010). Amorphous silicon-based solar cells. In
Handbook of Photovoltaic Science and Engineering, 2nd edn., eds. A. Luque and
S. Hegedus. Chichester: John Wiley and Sons, pp. 487–545.
Schwarz, R.B. and Koch, C.C. (1986). Formation of amorphous alloys by the mechanical
alloying of crystalline powders of pure metals and powders of intermetallics. Appl.
Phys. Lett., 49 146–148.
Shimakawa, K., Hayashi, K., Kameyama, T., and Watanabe, T. (1991). Anomalous
electrical conduction in graphite-vaporized films. Phil. Mag. Lett., 64, 375–378.
Spear, W.E. and LeComber, P.G. (1975). Substitutional doping of amorphous silicon.
Tanaka, K. (1989). Structure phase transitions in chalcogenide glasses. Phys. Rev. B, 39,
1270–1279.
Tanaka, K. and Shimakawa, K. (2011). Amorphous Chalcogenide Semiconductors and
Related Materials. New York: Springer.
Tanioka, K. (2007). The ultra sensitive TV pickup tube from conception to recent
development. J. Mater. Sci., 18, S321–S325.
Veres, M., Füle, M., Tóth, S. et al. (2005). Simultaneous preparation of amorphous solid
carbon films, and their cluster building blocks. J. Non-Cryst. Solids, 351, 981–986.
Veres, M., Koós, M., and Pócsik, I. (2002). IR study of the formation process of hydro-
genated amorphous carbon film. Diam. Relat. Mater., 11, 1110–1114.
Veres, M., Tóth, S., and Koós, M. (2008). New aspects of Raman scattering in carbon-
based amorphous materials. Diam. Relat. Mater., 17, 1692–1696.
3
Structure
A brief summary of the essential differences between amorphous and crystalline

semiconductors is presented at the beginning of this chapter for readers who are
unfamiliar with the amorphous phase. The amorphous phase is not the lowest
energy state, and these materials tend to relax to a crystalline phase, which
is structurally the lowest energy state. Different materials need different time
scales to reach the ground state. Amorphous structures at an atomic scale that
have a large relaxation time are discussed in the main part of this chapter. At the
end of the chapter, we discuss materials which are able to change their phases
within a nanosecond time scale.
3.1 Differences between amorphous and crystalline semiconductors

Charge distributions
Before we enter into a detailed discussion of diffraction patterns, we present a
brief overview of some other significant measurable differences between crys-
talline and amorphous phases, including both non-destructive and destructive
methods. Atoms in monoatomic crystalline semiconductors have no net atomic
charge at their equilibrium positions because of their symmetric arrangement, but
they do carry charges in the amorphous case due to geometric distortions. These
charge accumulations play an important role in the chemical shift of nuclear
magnetic resonance (NMR) measurements for high-resolution core-level spec-
tra, or for infrared absorptivity. In the 1980s, this phenomenon was one of
the most intensively investigated topics in amorphous semiconductor physics
(Guttman, Ching, and Rath, 1980; Klug and Whalley, 1982; Ley, Reichardt,
24
3.1 Classification of semiconductors 25
Figure 3.1. An elementary triad of atoms, denoted by K, L, and M, forming

two bonds (KM and LM), with the bond angle KLM denoted by θ .
and Johnson, 1982; Kramer, King, and Mackinnon, 1983; Brey, Tejedor, and
Verges, 1984; Winer and Cardona, 1986; Kugler and Náray-Szabó, 1987; Bose,
Winer, and Andersen, 1988; Kugler, Surjan, and Náray-Szabó, 1988; Kugler and
Náray-Szabó, 1991a, 1991b).
Net atomic charges are not observable directly, but experimental determin-
ation of their fluctuation is possible. By considering the deviation from the
ideal bond angle, a simple model was developed for the derivation of net
atomic charges in a-Si and diamond-like a-C (da-C) (Kugler et al., 1988;
Kugler and Náray-Szabó, 1991b). Consider an elementary triad of atoms,
denoted by K, L, and M, forming two bonds (KM and LM), with the angle
KML denoted by θ (see Figure 3.1). The net charge on the atoms of the
triad depends linearly on the deviation of θ from the ideal tetrahedral value
(dθ = θ – 109.47°):
qM = 2Adθ and qK = qL = −Adθ , (3.1)

total
where A is a fitting parameter. The total atomic net charge on atom M, qM , is a
sum of the contributions originating from all triads containing atom M. Because
in the distorted tetrahedral model each atom is at the center of 6 triads and at the
end of 12 triads (see Figure 3.2), we obtain the following relation:
⎛ ⎞
6 12
total
qM = A ⎝2 dθi − dθj ⎠ . (3.2)
i=1 j =1
From the semiempirical Hartree–Fock cluster calculation, A = −0.69 milli-

electrons/deg for a-Si (Kugler et al., 1988) (Figure 3.3) and A = −0.51 milli-
electrons/deg for da-C (Kugler and Náray-Szabó, 1991b) have been obtained.
In order to estimate charge fluctuations in a-Si and da-C, the geometric
model of tetrahedrally bonded amorphous semiconductors proposed by Wooten,
Winer, and Weaire (1985) was used. Applying eqn. (3.2) to all 216 atoms of the
26 Structure
3 3
3
3
3 2 L K 2
3
θ
1 M
x θi
y
3 2 θj 2 3
3
3 3
3
z
Figure 3.2. Topological model for net atomic charge calculation.
ΔQ
40
30
20
10
兺
–50 50
–10
–20
–30
–40
Figure 3.3. Comparison of Si atomic net charges calculated by semiempirical

Hartree–Fock calculation (crosses) and by eqn. (3.2) (solid line). Here Ʃ denotes
the quantity in parentheses in eqn. (3.2). Net charge is in millielectrons, Ʃ is in
degrees. (Taken with permission from Kugler et al. (1988). Phys. Rev. B, 37,
9069. Copyright 2013 by the American Physical Society.)
model cluster, dqSi = 0.021 electrons for the root mean square (rms) charge fluc-
tuation in a-Si and dqC = 0.015 electrons in da-C have been obtained. A revised
analysis of a core-level spectroscopic measurement (Ley et al., 1982) yielded the
estimation that the charge fluctuation in a-Si must be lower than 0.030 electrons.
Recently, a similar theoretical model was reported for distorted long disordered
Figure 3.4. Comparison of Se atomic net charges calculated using the Hartree–
Fock ab initio calculation and eqn. (3.4). Each symbol represents an atom.
(From Lukács and Kugler (2010). Chem. Phys. Lett., 494, 287. Copyright
2013 with permission from Elsevier.)
selenium chains (Lukács and Kugler, 2010). The net charge on the atoms of a
triad depends on the deviation of θ from the average value (dθ = θ – 101°):

qM = 2 Adθ + Bdθ 2 and qK = qL = − Adθ + Bdθ 2 . (3.3)
total
The total atomic net charge on atom M, qM , is a sum of the contributions
originating from the three triads containing atom M:
2
total
qM = 2 AdθM + BdθM 2
− A (dθM−1 + dθM+1 ) − B dθM−1 + dθM+1
2
.
(3.4)
Based on the Hartree-Fock ab initio calculation, A = −0.45 millielectron/deg

and B = −0.0089 millielectron/deg2 have been obtained as the best fits
(Figure 3.4).
Heat capacity
In the crystalline case, the heat capacity is easy to derive from the lattice vibra-
tion model. A high-temperature limit applying both classical and quantum-
mechanical models provides us with a value of C = 3NkB , where N and kB are
the number of atoms and the Boltzmann constant, respectively. Low-temperature
heat capacity using a quantum-mechanical calculation is in good agreement with
experiments, whereas the classical counterpart (C = 3NkB , the Dulong–Petit
28 Structure
law) is not applicable. The total energy due to the lattice vibration is obtained
by integrating the energy of a single quantum oscillator as follows:
∞
h̄ω
Eosc = h̄ω/2 + g (ω) dω, (3.5)
eh̄ω/kB T −1
0
where ω is the angular frequency (h̄ω is the phonon energy) and g(ω) represents
the vibration density of states per unit frequency range, i.e. the number of modes
with frequency [ω; ω + dω] is g(ω)dω. Only lattice modes of low frequency
will be excited at low temperatures. The dispersion relation has three acoustic
branches, one longitudinal and two transverse:
V ω2 1 2
g(ω) = 3
+ 3 , (3.6)
2π 2 vL vT
where vL and vT are the sound velocities of two different modes. The heat
capacity due to phonon excitation at low temperature is given by
dEosc 2V π 2 kB 1 2 kB T
C= = + ∼ T 3. (3.7)
dT 15 vL vT h̄
In addition, the contribution due to the electrons must be considered for a
more accurate calculation of heat capacity. The most general model is the free
Fermi-electron-gas description. The low-temperature approach obtained from a
Bethe–Sommerfeld expansion yields
π2 T
Cel = NkB ∼ T, (3.8)
2 TF
where TF is the Fermi temperature. Therefore the total heat capacity of crystalline
materials can be written as
Ctotal = γ T + βT 3 . (3.9)
This model does not work for amorphous semiconductors; there is no low-
frequency collective excitation in amorphous materials. Another problem is that
the free electron model is applicable for metals, but it does not work well for
semiconductors. Near linear heat capacity has been observed at low temperatures
for a large class of different materials, including chalcogenide glasses. The
common property among these structures is the low average coordination number
of atoms. In the same year, Phillips (1972) and Anderson, Halperin, and Varma
(1972) proposed the “two-level system” model for the description of linear
dependence of low-temperature heat capacity. The starting point is a fused
asymmetric double-well potential having two different bottom points. Atoms

are able to occupy one of the two equilibrium positions and to tunnel to the other
position. The free energy of this classical statistical basic problem having two
energy states in a canonical ensemble is given by
F (T , V , N) = U − T S = −kB T ln(cosh( E/2kB T )), (3.10)
where E is the energy difference between the two states. The heat capacity
can be calculated using
∂ 2F
CV = −T . (3.11)
∂T 2 V
For a two-level system,
CV = ( E/2kB T )2 sech2 ( E/2kB T ). (3.12)
At low temperatures (around 1 K), kB T ≈ 10−4 eV, which can be considered as an

upper limit for excitation energies, Emax . As a first approach we use a constant
density of states, i.e. n( E) = n. The total heat capacity can be obtained as
follows:
Emax
2
E E
Ctotal = n sech2 d E. (3.13)
2kB T 2kB T
0
As sech is a rapidly decreasing function (∼e−x ), we put infinity instead of Emax

and the integral calculus can be carried out analytically:
π2 2
Ctotal = k nT ∼ T . (3.14)
6 B
We can conclude that the total heat capacity of amorphous semiconductors at
low temperatures is proportional to T.
Calorimetric properties (glass transition, crystallization)

A calorimetric measurement, such as differential scanning calorimetry (DSC),
monitors a change in the specific heat of a material. When a glass sample is heated
from low temperatures, heat is absorbed at the glass-transition temperature Tg
(an endothermic process) and is emitted at the crystallization temperature Tc
(an exothermic process). Amorphous materials do not show such an endotherm
below Tc in the DSC measurement. An understanding of the mechanism of
crystallization is very important for phase-change materials, and this topic
30 Structure
will be discussed in Section 3.4 in detail. Above Tc an endothermic reaction again

occurs at the melting temperature Tm . As the glass transition and crystallization
are fundamentally important subjects in glass science, details of these phenom-
ena are discussed in the following sections. Note that crystallization is directly
related to phase-change devices, and therefore this will also be discussed in
Section 3.4.
Glass-transition temperature
As discussed in Chapter 2, glasses prepared by the melt-quenched (MQ) method
undergo a glass transition at Tg , which is defined as the transforming temperature
between the glassy and supercooled liquid states. A supercooled liquid describes
a material that remains in a liquid state, even below Tm . Although the definition
of the glass transition itself has a clear meaning, the experimental determination
of Tg is not easy, because the deduced Tg depends on experimental conditions,
such as the cooling rate of the melt in preparation and/or the heating rate in
the DSC measurement itself. It is still unknown which factors dominate Tg
(Elliott, 1990; Greaves and Sen, 2007; Tanaka and Shimakawa, 2011). In spite
of this ambiguity in the determination of Tg , it does correlate to some physical
parameters. A well-known empirical rule is Tg 2Tm 3 (where the temperatures
are measured in kelvin) (Kauzmann, 1948). Recently, it has been suggested that
the quantum effect plays a role in the glass-transition temperature (Novikov and
Sokolov, 2013): quantum effects lead to a significant decrease in Tg with respect
to Tm , so that the ratio Tg Tm can be much smaller than 23 in materials where
Tg is near or below 60 K (lower temperatures), and is given by
Tg A
= , (3.15)
Tm 1 + B/Tg
where A 23 and B depends on the strength of the Boson peak and some other
parameters.
As Tm can be primarily related to a cohesive energy, Tg should also be
related to the cohesive (or bond) energy in covalent glasses; this is discussed
in the following. At the glass-transition temperature, a glassy matrix may be
destroyed; Tg is therefore expected to be related to a network constraint (or
coordination number). Here, we introduce some representative models on this
topic, although there are numerous approaches (see, for example, Tichý and
Tichá, 1995). Tanaka (1985) has proposed the following empirical relation for
Tg (K)
E –0.9 (eV)
Figure 3.5. Correlation between glass-transition temperature Tg and the overall

mean bond energy E of a covalent network of chalcogenide glasses. (From
Tichý and Tichá (1995). J. Non-Cryst. Solids, 189, 141. Copyright 2013 with
permission from Elsevier.)
some covalent bonding glasses involving polymers: ln Tg 1.6Z + 2.3, where

Z is the average coordination number of the atoms. The parameter Z is known
to be important (see Section 2.4) in the description of the physical and chemical
properties of glasses, and there exist the “magic” numbers Z = 2.4 and 2.67 in
the constraint theory of glassy networks (Tanaka and Shimakawa, 2011). For
the vast majority of chalcogenide glasses, however, Tg is not simply related to Z
(Tichý and Tichá, 1995; Freitas, Shimakawa, and Kugler, 2013).
Instead of simply using Z, by taking the overall mean bond energy, E =
I fi Ei Zi , the other empirical relation Tg = 311(E – 0.9) K, as shown in
Figure 3.5, has been proposed for covalent-bonding chalcogenide glasses (Tichý
and Tichá 1995). Here, fi is the atomic fraction of the ith component and Zi is the
coordination number of the ith atom. Note that some special chalcogenides, such
as molecular-type glassy materials, are excluded from the data points because E
does not involve intermolecular interaction. The fact that Tg is well scaled with
E in covalent chalcogenide glasses suggests that Tg is principally determined
32 Structure
Figure 3.6. Correlation between Tg (K) and H(kg/mm2 ) in chalcogenide glasses.

Solid line is least-square fit to the data, producing Tg = 1.6H + 211. (From
Freitas, Shimakawa, and Kugler (2013). Chalcogenide Lett., 10, 39. Copyright
2013.)
by the cohesive energy. The short-range structure of glasses may be an important

property for Tg .
The average bond energy is not a measurable quantity and hence, instead of
scaling with E, the other measurable parameters should be used for scaling.
The hardness H may correlate with E, which is never applied as a scaling
parameter to express Tg . Principally, H can be related to Tm in a covalent-bond
system, and Tg vs. H may have a linear relationship. Figure 3.6 shows a good
correlation between Tg and H in a covalent amorphous chalcogenide system. The
scatter of the experimental data suggests that the intermolecular (van der Waals)
interaction should not be ignored (Freitas, Shimakawa, and Kugler, 2013).
Finally, we mention that computer simulations are also useful in understand-
ing the nature of the glass transition (Langer, 2006; Wilson and Salmon, 2009).
Viscosity
Glasses experience a supercooled liquid state when they are cooled below
their melting points. Glass-forming liquids have high viscosities even at high
3.2 Correlation functions 33
temperatures, and the viscosity of these liquids changes rapidly near the
glass-transition temperature. Thus the temperature at which the viscosity reaches
a certain value (1012 Pa s) is also used as a determination of Tg . The empirical
rule for temperature-dependent viscosity η, called the Vogel–Tammann–Fulcher
equation, in various glasses is given as η = η0 exp[A(T – T0 )], where A and
T0 are constants (Angell, 1988, 1995); T0 is related to the fragility of glasses.
Strong glasses, for example SiO2 , have T0 = 0 (following an Arrhenius relation),
and the fragile structure shows non-Arrhenius behavior. Note that the viscos-
ity is continuous through Tg , whereas the heat capacity shows a discontinuity
(remember the DSC measurement). The above form of the relation appears in
many transport phenomena, such as ionic conduction in liquid states. We will
see this empirical law in Section 3.4 on phase-change phenomena.
The model proposed by Adam and Gibbs (1965) is widely accepted in explain-
ing temperature-dependent viscosity, in which the important factor determining
the relaxation phenomena is the configurational entropy (CE). Alternatively,
Aniya (2002) has suggested that fluctuations in the coordination number Z
and the binding energy E in a glassy system dominate the fragility of glass:
large fluctuations in Z and E produce an increased fragility.
3.2 Projection from three-dimensional structures to

one-dimensional functions
Information about the atomic-scale structure of materials is essential in the
derivation and understanding of different physical properties. The distribution
of atoms as a function of distance, i.e. the probability of finding a particle at dis-
tance r when another particle can be located at the origin, is a possible characteri-
zation of the geometric structure at the microscopic level. In the crystalline case,
diffraction experiments produce diffractograms with the characteristic Bragg
reflections evidenced by well-defined sharp peaks at given angular positions.
As a typical example, peak positions and their coordination numbers Z within a
0–1.0 nm interval for crystalline silicon are displayed in Table 3.1. Pure amor-
phous semiconductors do not possess the long-range translational order of atomic
sites and therefore have no discrete diffraction patterns.
There are two one-dimensional functions that are used to describe three-
dimensional atomic distributions. In the first function, the atomic distribution
in a well-defined direction is investigated. In this case, the pair correlation
(distribution) function g(r) is a convenient function, where g(r)dV is proportional
34 Structure
Table 3.1. Peak positions and their

coordination numbers within a 0–1.0 nm
interval for crystalline silicon.
Peak position
(nm) Z
1 0.235 4
2 0.384 12
3 0.45 12
4 0.543 6
5 0.591 12
6 0.665 24
7 0.705 16
8 0.768 12
9 0.803 24
10 0.858 24
11 0.89 12
12 0.94 8
13 0.969 24
Figure 3.7. Schematic for the pair correlation function. Vector r indicates the
direction along which the atomic distribution is calculated.
to the probability of finding a particle inside a volume dV (see Figure 3.7). The
radial distribution function (RDF) is also a measure of the probability of finding
an atom a distance r away from a reference particle. This function, J(r), represents
the number of atoms within a distance between r and r + dr from a given
particle (see Figure 3.8). In contrast to structures that have translation symmetry,
amorphous materials are isotropic; therefore the pair correlation function and
the RDF contain the same information because of the following relationship:
J(r)dV = g(r)4π r2 dr. Functions must be zero below a given distance, which
is related to the atomic repulsion, and they should exhibit alternating maxima
and minima following the coordination shells. The correlation with the atom
Figure 3.8. Schematic for the radial distribution function (RDF).
at the origin is lost at large r. It follows that g(r) tends to unity (or to the
density, depending on the normalization), whereas the RDF has a parabolic
shape when r goes to infinity. Scientists working with amorphous materials
usually use J(r). The peak positions in J(r) provide the radii of the different
coordination shells of atoms surrounding the average atom. The area under
the peaks derives the coordination numbers of the different shells of atoms.
In a structural model, the RDF may be easily determined by calculating the
distance between all the atomic pairs and putting them into a histogram. An
alternative form of the RDF that is sometimes used is T(r), which is equal
to J(r)r.
The structures of amorphous semiconductors are typically determined by
neutron, electron, or x-ray diffraction experiments. Neutrons, having no charge,
interact with the nucleus of each atom, whereas x-rays or electrons are sensitive
to the electron distribution inside the amorphous materials. A neutron beam is
capable of penetrating well beyond the surface of a sample, to depths on the
order of centimeters in the condensed phases. In a diffraction experiment, we
measure the intensity as a function of the scattering angle. Thermal neutrons
with wavelengths of 0.1– 0.2 nm are very useful for atomic-scale structural
(and magnetic structure) investigations. Neutron diffraction is a non-destructive
technique; it is a form of elastic scattering in which the neutrons exiting the
sample have the same energy as the incident neutrons. In a standard neutron-
diffraction measurement, g(r) and J(r) are derived from the measured scattered
intensity by Fourier transformation of the static structure factor S(Q), where
Q = kinitial – kfinal . Here, kinitial and kfinal represent the wave vectors in the initial
and final states, and
1
ρ (g (r) − 1) = d Q (S ( Q) − 1) e−i Qr . (3.16)
(2π )3
36 Structure
4
3.5
3
2.5
g (r)
2
1.5
1
0.5
0
0 2 4 6 8 10
r (Å)
Figure 3.9. Pair correlation function of graphite-like a-C measured by neutron
diffraction and derived from S(Q) by Fourier transformation.
It is assumed in amorphous materials that the scattering is isotropic, and that

this three-dimensional Fourier transformation can be reduced to the following
one-dimensional sine transformation:
∞
1
ρ (g (r) − 1) = Q [S (Q) − 1] sin (Qr) dQ, (3.17)
2π 2 r
0
where Q = (4π λ)sin θ and r are already scalar values, λ is the de Broglie
wavelength of the neutron, and 2θ is the scattering angle. A special example for
a-C films is shown in Figure 3.9.
Neutron cross-sections can be divided into two parts: coherent and incoherent
cross-sections, σ coh and σ inc , respectively. The elastic part of the coherent scat-
tering provides information about the structure. Measured cross-sections for any
element of the periodic table can be found in Sears (1992), pp. 29–37, or at
http://www.ncnr.nist.gov/resources/n-lengths. The coherent, incoherent, and
absorption cross-sections of isotopes are given in units of barn (1 barn =
10–28 m2 ). In monatomic structure investigation, the absorption cross-section
has no role. Materials with a larger coherent cross-section than their incoherent
counterparts provide a good opportunity to derive the RDF experimentally using
neutron diffraction.
X-rays are scattered by the electrons in atoms. When a covalent bond is
formed, the associated valence electrons are distributed between the two atoms.
It follows that the x-ray diffraction pattern probes the electron distribution around
each atom and between the atoms. In amorphous cases, electrons are placed
asymmetrically, so x-ray scattering may provide a more complex and different
density distribution than that produced by neutron diffraction.
In contrast to x-ray and electron diffraction, neutrons are scattered by nuclei,

and there is no relationship with atomic number, i.e. different isotopes of the
same element can have widely different cross-sections. This fact is a great
advantage for two- and three-component (multicomponent) materials, in which
similar samples having different isotopes are prepared and measured (known as
the isotope substitution method). In essence, the idea is to measure the diffrac-
tion pattern from several identical samples which contain different amounts of
isotopes for one or more of the components. Each diffraction pattern will provide
different partial structure factors, depending on the precise values of the neutron
scattering lengths for each isotope.
Other ways of obtaining information on partial structure factors include the
combined use of x-ray and neutron diffraction, or the joint use of neutron, x-
ray, and electron diffraction data. An excellent review has been published by
Gabriel J. Cuello on the elementary introduction to the method of structure factor
determination by neutron diffraction in non-crystalline materials (Cuello, 2008).
Diffraction measurements on amorphous carbon

A pioneering electron diffraction measurement on evaporated carbon film was
reported more than half a century ago by a Japanese group (Kakinoki et al., 1960).
They concluded that the film is built up by randomly distributed diamond-like
and graphite-like crystalline regions. These regions have no mutual orientation
over the range of several angstroms.
Neutron-diffraction measurements have been performed on amorphous car-
bon at low temperatures (Kugler et al., 1993b) in order to investigate the origin
of the anomalous electrical conduction (Shimakawa et al., 1991). Using an
arc vaporization method, a 1.0 g pure a-C sample was prepared. The experi-
ment was carried out at the 7C2 spectrometer installed on the hot source of the
Orphée reactor at CEA Saclay, in France. A momentum transfer range of 5–
160 nm−1 was covered. In the measured pair correlation function, the positions
of the first four main peaks were 0.142 nm, 0.246 nm, 0.283 nm (third-neighbor
cross-ring distance), and 0.373 nm. In a perfect graphite crystal, the first four
nearest-neighbor distances are 0.142 nm, 0.246 nm, 0.284 nm, and 0.364 nm. It
seems likely therefore that the sample contained a high proportion of threefold-
coordinated carbon atoms, i.e. it is typically a graphite-like material, as shown
in Figure 3.9. Beyond 0.6 nm, there is an absence of atomic structure; g(r)
has no appreciable medium-range order. The derived S(Q) values exhibit small
38 Structure
differences due to the variation of the temperature, but there was no significant
difference between the radial distribution functions.
The structure of another a-C sample (20–30 mg, a very limited specimen mass
for neutron diffraction) prepared by plasma-arc deposition has been determined
by neutron diffraction at the twin-axis diffractometer, D4, at the Institut Laue-
Langevin (ILL), Grenoble, France (Gaskell et al., 1991). The conclusion is in
opposition to the above-mentioned results: the structure factor and reduced radial
distribution functions are similar to those for a-Si and a-Ge, indicating a high
proportion of tetrahedral bonding. The first three main interatomic distances
are very close to bond lengths in diamond crystal (0.152 nm and 0.153 nm;
0.250 nm and 0.252 nm; 0.296 nm and 0.296 nm). A fit to the data yields about
86% tetrahedral bonding.
Thick a-C films were prepared in a high-vacuum system of base pressure
around 10–7 Torr, by rf sputtering on liquid-nitrogen-cooled Cu substrates.
Raman measurements were performed to confirm the amorphous nature of the
deposited films. A 0.8 g sample was used for neutron-diffraction measure-
ments performed at 300 K in the Special Environment Powder Diffractometer
of the Argonne National Laboratory’s Intense Pulsed Neutron Source, IL, USA
(Li and Lannin, 1990). The films were removed from the substrate by using
dilute hydrochloric acid. The RDF was compared to theoretical models, and was
found to exhibit qualitative agreement with a number of basic models, indicating
predominant threefold bonding. The absence of a specific peak at the graphite
intrahexagon distance (third-neighbor peak) indicates that structural models with
such intermediate-range correlations are not correct for a-C. This distinguishes
a-C from local two dimensionally ordered graphite-like materials.
Diffraction measurements on amorphous silicon

The first electron diffraction measurements were carried out by Moss and
Graczyk (1969), and by Barna et al. (1977). At the first important stage of amor-
phous silicon structure computer modeling, the lack of a neutron-diffraction
measurement caused a measured RDF of a-Ge to be applied to make a compar-
ison between theoretical and experimental data (Wooten et al., 1985).
Some years later, a nearly pure, evaporated a-Si sample (0.5 g) was prepared
in the Central Research Institute for Physics, Budapest, Hungary, and a neutron-
diffraction experiment was performed on this sample at the 7C2 spectrometer
installed on the hot source of the Orphée reactor at CEA Saclay, in France
(Kugler et al., 1989). Using an incident wavelength of λ = 0.0706 nm, the

momentum transfer range of 5–160 nm−1 was covered. A second neutron-
diffraction measurement on the same a-Si sample was performed using the
D4 twin-axis diffractometer at the high flux reactor at the ILL, Grenoble (Kugler
et al., 1993a). The incident neutron wavelength of 0.04977 nm and the angular
ranges of 1.5°–65° and 46°–131° covered by the two multi-detectors provide
us with a larger momentum transfer range of 3.3–230 nm−1 . The results of
the experiments show that the covalently bonded a-Si films are not completely
disordered. The bonds between the atoms and the coordination numbers are
similar to those for the crystalline phase. Compared with a perfect crystal, a-Si
shows that the first- and second-neighbor peaks are broadened but the positions
are the same, whereas the third peak disappears in the measured RDF. This fact
is one of the most important properties of group IV amorphous semiconductors.
The measured mean square width of the first peak suggests that the bond-length
fluctuations are around 1–2%. The broadening of the second peak reflects around
10% bond-angle fluctuation. The absence of the third peak confirms that there
is no characteristic dihedral angle.
Neutron-diffraction measurements were performed on an a-Si sample as-
deposited and annealed at 600 °C at the Special Environment Powder Diffract-
ometer of the Argonne National Laboratory Intense Pulsed Neutron Source,
IL, USA (Fortner and Lannin, 1989). It was prepared by rf sputtering at low
Ar pressure. The radial distribution functions indicate values of 108.4° and
108.6° for a tetrahedral angle for as-deposited and annealed films, respect-
ively, which are both smaller than the ideal 109.47° angle. The estimated bond-
angle widths for the samples are 9.7° and 11.3° for as-deposited and annealed
films, respectively. The coordination number in the annealed sample was 3.90,
and a value of 3.55 was estimated for the as-deposited sample, which is smaller
than the expected value of 4.
In addition to neutron-diffraction measurements, the structure factor S(Q)
of high-purity a-Si membranes prepared by self-ion implantation was meas-
ured over a large extended Q range of 0.3–550 nm−1 by x-ray diffraction,
using a Huber six-circle diffractometer at the Cornell High Energy Synchrotron
Source, A2-wiggler beam line, Ithaca, USA (Laaziri et al., 1999). The sample
was kept at 10 K to minimize the thermal effects on the diffraction pattern.
The mean values of the coordination numbers were observed as 3.79 for the
as-implanted sample and 3.88 for the annealed sample. In this nanovoid-free
sample, the density deficit of a-Si relative to crystalline Si (c-Si) is due to a
40 Structure
fundamental under-coordination, i.e. dangling bonds may decrease the local

density.
A combined investigation using the complementary steady-state and pulsed
neutron sources was carried out on hydrogenated amorphous silicon (a-Si:H)
(Bellissent et al., 1989). The aim of this study was to provide accurate data con-
cerning the short-range order in a-Si:H and a-Si:D (where D means deuterium)
samples. The samples were prepared by sputtering and the glow-discharge
method. Samples of mass 1.0–1.5 g were measured on the D4 twin-axis spec-
trometer at steady state in the ILL, Grenoble, and on the LAD spectrometer at the
ISIS spallation neutron source, Rutherford Appleton Laboratory, Chilton, UK.
The Si–H distances were identified as 0.148 nm and 0.321 nm, and those for Si–Si
as 0.234 nm and 0.375 nm for first- and second-neighbor distances, respectively.
Diffraction measurements on amorphous germanium

A large volume, corresponding to 9.6 g, of evaporated amorphous germanium,
was contained in an 11.7 mm diameter thin-walled vanadium can for the neutron-
diffraction experiment. The experiment was carried out at the ILL, Grenoble, on
the D4 twin-axis diffractometer (Etherington et al., 1982). The neutron wave-
length was 0.05057 nm. The observed nearest-neighbor coordinate number,
Z1 = 3.68, is significantly less than the value of 4 expected for tetrahedral
bonding. The second peak yielded a coordination number Z2 = 12.11, which is
somewhat greater than that expected for tetrahedral bonding. The structure of
a-Ge is accurately described in terms of a tetrahedrally bonded random network.
Diffraction measurements on amorphous selenium

Neutron-diffraction measurements have been performed using the SLAD instru-
ment at NFL, Studsvik, Nyköping, Sweden (Jóvári, Delaplane, and Pusztai,
2003). Crystalline selenium powder (3 g) was ball milled under an argon atmos-
phere for 6 hr in a Spex mixer/mill. The milling procedure consisted of milling
for 15 min followed by 45 min of rest to avoid heating, and then the cycle was
repeated. The amorphous selenium powder was contained in a thin-walled van-
adium container. The obtained experimental structure factor S(Q) was compared
with three earlier measured data within the interval 8–120 nm−1 . Furthermore,
pair correlation functions were derived by unconstrained reverse Monte Carlo
computer simulations. The result is presented in Section 3.3.
3.3 Three-dimensional structure 41
3.3 Three-dimensional structure derived from one-dimensional function

Knowledge of the three-dimensional atomic arrangement is an essential pre-
requisite to understand the physical and chemical properties of different materi-
als. In the crystalline case, the determination of the structure means solving the
atomic arrangement within the unit cell containing a few atoms. The structure
of the crystal as a whole is then constructed by repeating the unit cell periodi-
cally. In the non-crystalline case, the lack of periodicity means that the structure
cannot be determined in the same way as for crystalline materials.
Amorphous semiconductor structures have been studied over a long period
of time using different diffraction techniques. Although several experimental
structural studies have been made using x-ray and electron diffraction, the
neutron-diffraction technique yields the most accessible information on these
disordered structures. In recent years, NMR (nuclear magnetic resonance) spec-
troscopy and EXAFS (extended x-ray absorption fine structure) spectroscopy
have become important element-specific tools for the derivation of local atomic
arrangements. The structure factor S(Q) measured in a limited Q range can be
Fourier transformed, which yields the radial distribution function. The obtained
result is only a one-dimensional representation of the spatial atomic distribution
in three dimensions. This projection causes information loss because an unlim-
ited number of different possible disordered atomic structures can display the
same RDF. Unfortunately, no experimental technique has been discovered so far
that can determine the microscopic atomic distributions inside the bulk. A pion-
eering step was taken by Tegze and Faigel (1996) by using x-ray holography.
They demonstrated the efficacy of atomic-scale x-ray holography by obtaining
direct images of the three-dimensional arrangement of strontium atoms in cubic
perovskite, SrTiO3 . To obtain holographic images on an atomic scale, thermal
neutrons were applied by Cser et al. (2002). Disordered atomic arrangements
still remain a problem. Efforts to develop modeling techniques with which to
analyze atomic resolution are therefore continually being made.
Atomic interactions
Classical empirical potentials
In order to prepare three-dimensional atomic structures of amorphous semi-
conductors, a transferable atomic interaction is required. The simplest ver-
sions of atomic interactions are the classical empirical potentials, which are
42 Structure
computationally the cheapest method for structural modeling. The parameters

describing the atomic interaction can be derived either from experimental obser-
vations or from advanced quantum-mechanical calculations. For semiconductor
structures, simple Morse or Lennard–Jones potentials are not suitable because
the directional nature of atomic bonding also needs to be taken into account.
Therefore, the potential needs to be expressed by at least two- and three-body
terms. When a simulation is performed, a truncation of the interaction, known
as the cut-off distance, is usually applied. The term “short range” means that the
total potential energy of a given particle is dominated by the interaction with
neighboring atoms that are closer than the cut-off distance.
Keating potential
The Keating empirical potential (Keating, 1966) is a simple example for a three-
atom local interaction which has macroscopically measurable parameters. It was
originally designed to describe the defect-free diamond type of crystal, such as
silicon or germanium, or diamond-structure carbon, in which the atoms have
four nearest neighbors. The bond-stretching and bond-bending force constants,
α and β, respectively, are determined from the fitting to the elastic properties of
the crystal phase. The number of bonds is equal to 4N2 = 2N, and the number
of atomic triads is equal to (4N × 3)2 = 6N, where N atoms form the geometric
structure. For these materials, the bond length in an equilibrium state is equal
to r0 (0.155 nm, 0.235 nm, and 0.245 nm are for C, Si, and Ge, respectively),
and the bond angle is the tetrahedral angle (Θ = arccos(−13) = 109.47°). In
the Keating potential, bonding geometries which deviate from these equilibrium
values are assigned an energy penalty:
⎡ ⎤
3 ⎣α 2
2N 6N

rij rik + r02 /3 ⎦ ,
2 2
U= 2 r − r02 + β (3.18)
8r0 2 ij ij j ik
where rij is the vector pointing from atom i to atom j, and rij rik is the scalar
product of two vectors. This formula is practically a Taylor expansion where
the linear terms vanish and only the quadratic terms are kept. The constants
α and β are obtained from a fit to the elastic properties of the crystal. For
silicon, values of α = 296.5 eV nm−2 and β = 0.285α were obtained. The
validity of a large deviation from the equilibrium values of bond length and
bond angle is questionable. It is suitable only for small displacements from the
ideal positions. In particular, it should not be used for very disordered systems.
Another disadvantage is that the adjacency matrix of a topological structure must
be known for the total energy calculation. If the adjacency matrix is unknown,
we cannot calculate the diffusion or other dynamical processes.
Several modified versions of the Keating potential, eqn. (3.18), have been
developed. Fullerenes, nanotubes, and graphenes are formed by threefold-
coordinated carbon atoms. Overney, Zhong, and Tománek (1993) reparam-
eterized the potential, i.e. the bond-stretching and bond-bending force constants
(α, β) were determined by applying the density-functional-theory (DFT) ab ini-
tio local-density approximation total energy calculations for specific distortion
modes of graphite. The bond angle is equal to Θ = 120°, and instead of the
last term, r02 /3, we must substitute the term r02 /2. The structural rigidity and
the low-frequency vibration modes of carbon have been calculated using this
modified version of the Keating potential.
Rücker and Methfessel (1995) worked out a generalization of the Keating
model for group IV semiconductors and their alloys. First, the energy of mixing
relative to the pure materials is separated into two terms, namely Uchemical and
Ustrain . The chemical term arises because the strength of an A–B bond is different
from the average of the A–A and B–B bond strengths. Secondly, higher-order
anharmonic terms are included in order to describe strongly distorted systems.
Even for group III–V alloys (Alx Ga1−x As alloys), a modified Keating potential
has been presented in the literature (Sim et al., 2005). The anharmonic model
potential consists of Coulomb interaction terms in the formulas because of the
charge transfer between different types of atoms.
Stillinger–Weber potential
Generally, any empirical potential function could be described by one-, two-,
three-body, etc., contributions. If there is no external force, this expansion begins
with the pair-wise terms. For sigma-bonded materials the three-body interaction
term at least is required. The Stillinger–Weber empirical potential (Stillinger
and Weber, 1985) that has been developed for liquid and crystalline phases of
silicon is based on two- and three-body interactions:
⎧ ⎫
⎨ ⎬
(2) λ (3)
U = Aε vij rij + v rij , rik , (3.19)
⎩ A j ik j ik ⎭
ij
44 Structure
where ε = 2.16826 eV. The summations include all pairs ij and all triples jik of
atoms in the system. For rij /σ < a, the two-body term is chosen to be
r
rij −p rij −q ij
−1
vij(2) rij = B − exp −a , (3.20)
σ σ σ
otherwise vij(2) (rij ) = 0. This definition automatically cuts off at rij /σ = a with-
out discontinuities in any rij /σ derivatives (σ = 0.20951 nm). If rij /σ and rik /σ
are smaller than the cut-off distance a, the three-body part is given by
−1 r −1
(3) rij ik 2
vj ik rij , rik = exp γ −1 + −1 cos θj ik + 1/3 ,
σ σ
(3.21)
otherwise vj(3)ik (rij , rik ) = 0. In practice, only a small subset of these fragments
should actually be computed because atoms interact only when their separation
is less thanσ a. The most “satisfactory” parameter set is the following:
A = 7.049556277, B = 0.6022245584, p = 4, q = 0,
a = 1.80, λ = 21.0, and γ = 1.20.
In its original form, the Stillinger–Weber potential is not expected to describe

accurately the amorphous phase. Parameters were fitted to the crystalline and
liquid phases of silicon. A modified Stillinger–Weber potential was fitted directly
to the amorphous phase of silicon by Vink et al. (2001). As it turned out, the
strength of the three-body interaction was boosted by approximately 50% to
describe this phase correctly. The improved parameters are ε = 1.64833 eV
and λ = 31.5. The rest stay the same. The modified Stillinger–Weber potential
provides a better fit to the measured vibrational density of states spectra. Recently
a new parameterization of this potential has been presented for defects and
plasticity of silicon base materials (Pizzagalli et al., 2013).
Ding and Andersen (1986) derived, with the help of the Stillinger–Weber
potential, a set of parameters to describe germanium. In practice, only three
of the parameters are different from those for silicon, for example ε = 1.93 eV,
σ = 0.2181 nm, and λ = 31. The potential provides a good structural representa-
tion of a-Ge and c-Ge. Furthermore, it gives good results for phonon dispersion
relations of the crystalline phase. Jian, Kaiming, and Xide (1990) presented new
sets of parameters for Si and Ge which provide considerable improvements for
the phonon dispersion relations.
Although the Stillinger–Weber potential has more flexibility to describe dif-

ferent configurations than the (modified) Keating potential, its transferability to
a number of different structures remains in question. To address this issue, a
possible improved version of the potential was suggested by Justo et al. (1998).
Their environment-dependent interatomic potential contains similar terms to
those found in the Stillinger–Weber formalism, for example two- and three-
body terms, which also depend on the local environment of atom i through its
effective coordination number. The effective coordination number is defined by

Zi = f (Rim ), (3.22)
m=i
where f (Rim ) is a cut-off function that measures the contribution of a neighbor

m to the coordination of an atom i in terms of the separation Rim ; f (Rim ) equals
unity when the atomic distance is lower than, or equal to, the first-neighbor
distance, and it gently drops to zero between the first- and second-neighbor
distances. This leads to a considerable improvement in describing the local
bonding for bulk defects and disordered phases of silicon.
Tersoff potentials
A large family of empirical potentials for silicon, carbon, and multicomponent
covalent systems was developed by J. Tersoff (1986, 1988a, 1988b, 1988c).
Sometimes they are named Abell–Tersoff potentials in order to recall the pion-
eering work on empirical pseudopotential theory of G.C. Abell (1985). The pair
potential
1
U= f (rij ) Aexp −λ1 rij − Bij exp(−λ2 rij ) , (3.23)
2 i,j =i
where rij is the distance between atoms i and j, and A, B, λ1 , and λ2 are all
positive; f (rij ) is a cut-off function to restrict the range of the potential. The first
term is repulsive, while the second is interpreted as representing the bonding in
a Morse-type potential. The parameter Bij implicitly includes the bond order
and depends upon local environments. Deviations from the simple pair potential
are ascribed to the dependence of Bij upon the local environment. The bonding
strength Bij should be a monotonically decreasing function of the number of
competing bonds, the strength of the bonds, and the cosines of the angles with
the competing bonds.
46 Structure
Optimized Tersoff potential parameters have been derived by Powell, Miglio-

rato, and Cullis (2007) for group III–V zinc blend semiconductors such as GaN,
AlN, InN, GaAs, InAs, AlAs, GaP, InP, AlP, GaSb, InSb, and AlSb. Mathem-
atical terms contain 64 fitting parameters. This potential was used mostly for
fullerene, nanotubes, and graphenes.
Oligschleger potential
The other important class of amorphous semiconductors is the chalcogenides.
The model element for these materials is selenium. An empirical three-body
potential has been developed by Oligschleger et al. (1996) with the intent of
providing both a realistic and a simple description of a selenium–selenium
interaction. The most common selenium crystals are built up by either selenium
rings or infinite helical chains; in both structures every selenium atom has
two nearest neighbors, and this unique property has to be reproduced by the
interatomic potential. The authors have taken into account both small selenium
clusters and crystalline phases during the parameter fitting of the potential. Its
analytical form is given by

U= V2 rij + h rij , rkj , θij k + cyclic permutations. (3.24)
i<j i<j <k
The three-particle energy is given by

h rij , rkj , θij k = V3 (r) V3 (s) b1 (cos θ − cos β2 )2 + b3 − 0.5b1 cos 4 θ ,
(3.25)
with b1 = 534.4866, b3 = 511.9572, and β 2 = 95.3688°. This three-body term

is purely repulsive. Note that V2,3 (r) has three different terms in the function of
r. If r is lower than 1.6,
V2,3 (r) = a2,3 exp (αr) + b2,3 exp (βr) + c2,3 exp (γ r) . (3.26)
For 1.6 < r < r2,3 ,
V2,3 (r) = d2,3 (r − r2,3 )5 + e2,3 (r − r2,3 )4 + f2,3 (r − r2,3 )3 . (3.27)
Otherwise, V2,3 (r) = 0. Note that in Table I in the original paper (Oligschleger
et al. (1996), p. 6165) the parameter a = 9281.2 in the two-body case. The other
misprint is that α = −7.9284 instead of −7.984. Correct parameters in reduced
units (see the original paper) can be taken from Table 3.2.
Table 3.2. Revised parameters for

selenium potential.
Two-body case Three-body case

a 9281.2 8.8297
b 0.26802 −2.5932
c −16.599 −6.9384
α −7.9284 −0.47601
β −0.000077781 −1.5637
γ −1.8634 −0.37049
d 1.86825 0.22556
e 4.58628 0.12527
f 3.62029 −0.21019
r2,3 2.37 2.35
Altogether, the 23 parameters are fitted to density-functional calculations.

The potential has been carefully tested and compared to experimental results.
The descriptions of atomic geometries and bonding energies are excellent, but
a weak point of the potential lies in the quantitative description of vibration
properties. However, the overall performance of the potential is outstanding
when considering its effectiveness and simplicity.
Tight-binding models
Tight-binding (TB) models offer a semiempirical quantum chemical approach to
describing the interaction between atoms. The TB model is useful in generating
and describing the atomic structure of amorphous materials. We focus only
on orthogonal TB models that have been developed for carbon, silicon, and
selenium. The total energy of the system is written as follows:
Etot = Eb + Erep , (3.28)
where Eb is the sum of electronic eigenvalues εi over all occupied electronic

states, given by

nocc
Eb = 2 εi , (3.29)
i
and Erep is the short-range repulsive energy. The electronic eigenvalues are
obtained by diagonalizing an empirical tight-binding Hamiltonian. The on-site
elements are the atomic orbital energies of the corresponding atoms, whereas
48 Structure
the off-diagonal elements of the tight-binding Hamiltonian are described by a

set of orthogonal sp3 two-center hopping parameters, Vssσ , Vspσ , Vppσ , and
Vppπ , scaled with interatomic separation r as a function s(r). The remainder of
the Etot is modeled by a short-range repulsive term Erep given by
⎛ ⎞

Erep = f⎝ (rij )⎠ , (3.30)
i j
where (rij ) is a pair-wise potential between atoms i and j, and f is expressed

as a fourth-order polynomial.
In order to model the amorphous structure, the TB model has to be transfer-
able. This can be achieved by adopting a suitable functional form, in this case
the one suggested by Goodwin, Skinner, and Pettifor (1989). Thus the scaling
functions s(r) and (r) are given by
nc nc
r r0
s (r) = (r0 /r) exp n −
n
+ (3.31)
rc rc
and
mc mc
r d0
(r) = 0 (d0 /r) exp m −
m
+ , (3.32)
dc dc
where r0 denotes the nearest-neighbor atomic separation in the diamond struc-

ture. Parameters for carbon are found in Xu et al. (1992) and for silicon by
Kwon et al. (1994). A similar formalism has been used for selenium by Molina,
Lomba, and Kahl (1999).
More sophisticated TB models are also available; see, for example, Tang et al.
(1996) for carbon, and Wang, Pan, and Ho (1999) for silicon.
A more accurate approach is to derive the atomic interaction directly from
the electronic ground state, where the total energy functional is calculated using
density-functional theory. The difficulty of DFT is in obtaining an accurate
exchange and correlation interaction inside a many-body system. Nowadays
this is a rapidly developing and most popular method in solid state physics.
The computer cost is much higher than that for TB models, and it is compar-
able to the Hartree–Fock ab initio method, which is not the best for structure
investigation.
Computer simulation methods

During the past decades research has focused on the construction of realis-
tic atomic configurations for amorphous materials using computer simulations.
Computations should not substitute for lack of knowledge of atomic-scale con-
figuration, but instead be a useful tool for constructing atomic-structure models.
There are two main possibilities for structural modeling methods of disordered
configurations at the atomic level: a stochastic method (Monte Carlo (MC) type)
and a deterministic method (molecular-dynamics (MD) simulations).
Monte Carlo methods

The name Monte Carlo arises from the famous casino in Monaco. The method
involves the use of random numbers to govern atomic displacements during
computer simulation processes. Several algorithms fall under the MC method:
traditional MC, reverse MC (McGreevy and Pusztai, 1988), quantum MC, kinetic
MC, path integral MC methods (Herrero, 2000), etc. In this section, only the first
two techniques are discussed. Systems must have a large number of degrees of
freedom, which are investigated by the MC method. The traditional MC system
generating a three-dimensional particle configuration of amorphous materials
searches for a local minimum of total energy on the energy hypersurface. In
order to calculate the energy, we need to describe the influence on each atom
from the atoms that surround it. Two different interactions are usually applied
for relaxation: one is known as the empirical potentials and the other is based
on different quantum-mechanical approaches, such as tight-binding models,
density-functional theory, or Hartree–Fock approximations.
The initial configuration could be any atomic arrangement (even a distorted
crystalline structure), although the perfect crystal provides the absolute minimum
total energy on the energy hypersurface. A randomly chosen atom is displaced
to a new position that is determined with the help of a random-number generator.
In most of the simulations, a maximum displacement is applied as a commonly
used constraint. The energy variation is a crucial parameter for the MC method.
If a MC step provides a downhill motion on the energy hypersurface, then the
new configuration is accepted, and this reordered structure becomes the initial
atomic arrangement for the next MC step. In an uphill case, the new position is
rejected only conditionally. This decision is the essential part of MC simulations.
50 Structure
(a)
(b)
Figure 3.10. Bond transposition process for a-Si (a-Ge) structural model con-
structed by Wooten et al. (1985) using MC simulation. (a) Configuration of
bonds in diamond structure; (b) modified configuration after bond transposition.
A simple explanation of the selection rule is the following. If uphill motion is

always rejected, then the procedure reaches the first shallow energy minimum
and there is no further possibility to search for a deeper minimum which belongs
to a more realistic non-crystalline atomic arrangement. A precise description
of MC philosophy can be learned from any elementary statistical physics
book.
Amorphous silicon model constructed by Monte Carlo simulation

The first state-of-the-art traditional MC simulation was performed for pure
a-Si and a-Ge by Wooten et al. (1985). The main purpose was to construct
a homogeneous structure without any defects, i.e. Si atoms have four nearest
neighbors and there are no voids inside the model structure. The starting atomic
configuration was a simple cubic diamond structure, and a number of bond
transpositions were carried out (see Figure 3.10). This process destroys the
translation and topological symmetries and includes fivefold and sevenfold rings
(Wooten and Weaire, 1984). Similar bond transposition in two-dimensional
carbon systems (fullerenes, nanotubes, and graphenes) is called the Stone–
Wales transformation (Stone and Wales, 1986). After randomization using the
Metropolis MC algorithm (Metropolis et al., 1953), the structure was relaxed by
the Keating empirical potential. A continuous random network with 216 atoms
has periodic boundary conditions in three dimensions in order to preserve local

order on the surface. The topology includes fivefold, sixfold, sevenfold, and
eightfold rings. In the diamond structure, only sixfold and eightfold rings should
be found. Based on a close analogy between silicon and germanium or saturated
carbon compounds, this model can be rescaled by nearest-neighbor distances and
can be applied for a-Si, a-Ge, and tetrahedral a-C (ta-C) configurations. Since
then, several computer-generated models have been constructed using various
classical empirical potentials (Kelires and Tersoff, 1988; Ishimaru, Munetoh,
and Motooka, 1997; Ishimaru, 2001; Vink et al., 2001) or by applying different
quantum-mechanical methods (Drabold et al., 1991; Stich, Car, and Parinello,
1991; Tóth and Náray-Szabó, 1994; Hensel et al., 1996; Yang and Singh, 1998;
Cooper, Goringe, and McKenzie, 2000, Valladares et al., 2001). Despite this,
the WWW model developed by Wooten et al. (1985) is still considered to be
the best defect-free three-dimensional atomic-scale representation of covalently
bonded a-Si (and a-Ge). The WWW or modified WWW models were applied
for the electronic structure calculations, and the results from these works suggest
that the covalently bonded a-Si structures are not completely disordered. The
bonds between atoms and the coordination numbers are similar to those in the
crystalline phase.
A special MC simulation technique, in which the energy is calculated by
quantum-mechanical methods, was proposed by Tóth and Náray-Szabó (1994).
The applied semiempirical fragment self-consistent field technique divides the
periodic simulation cell into two parts. The first is the subsystem where the
important change (the random motions) occurs, and the rest comprises the en-
vironment exerting only secondary effects on the former part. The conventional
self-consistent field equations have to be solved only for the critical subsystem. In
this way, the computational efforts are decreased drastically, as the dependence
on the number of atoms in the environment reduces to quadratic instead of
cubic or quartic, as in conventional semiempirical or ab initio methods. This
quantum MC technique was performed first for a sigma-bonded a-Si structure
investigation (Tóth and Náray-Szabó, 1994).
Reverse Monte Carlo methods

Another version of the MC method – the reverse Monte Carlo (RMC) simula-
tion – has been developed by McGreevy and Pusztai (1988) for the structure
investigation of disordered condensed phases and liquids. Initial results for liquid
52 Structure
argon were presented by the authors. This inverse problem solution method is
convenient for the analysis of diffraction data and at the same time for modeling
the amorphous structures. The RMC simulation is based on the results of diffrac-
tion measurements and is free from the description of the interaction between
atoms. This is the only method generating three-dimensional particle configura-
tions that are consistent with the experimentally derived structure factor or the
radial distribution function.
Theoretically, both functions, the structure factor and the RDF, can comprise
the initial data for RMC simulation, but use of S(Q) has an important advantage.
The main problem that occurs at any diffraction measurement – the measure-
ment of Q in only a limited interval and Fourier transformation in an infinite
interval – is bypassed. To speed up the simulation and/or to reach more realistic
configurations, constraint(s) could usually be applied. To start the RMC simu-
lations, a given number of particles are confined in a well-defined box. The
simulation derives the density that is the most important constraint. Another
constraint that is used several times is the coordination number. It can easily be
prescribed that particles must strictly have a given number of particles within a
fixed distance. A hard core diameter (the lower limit of bond length) can also
be applied in order to avoid physically meaningless configurations. For sys-
tems where atomic interactions are covalent, i.e. bonds are directed, introducing
constraints on bond angles might also be a useful constraint.
The use of constraints modifies the original RMC algorithm in the following
way.
(a) Start with an initial particle configuration. Calculate its RDF or structure
factor. Calculate also the difference between the simulated and experimental
RDFs or structure factors (χ02 ). The comparison with experiment is quantified
(yexp −ysim )2
using a function of the form χ 2 = σ2
, where yexp and ysim are the
measured and calculated quantities; σ is a measure of the accuracy of the
measurement. The sum includes all points in a function such as the RDF or
structure factor.
(b) A new, trial, configuration is generated by the random motion of a particle.
(c) Check whether the new configuration satisfies the constraint(s) applied. If
not, start again from (b). (This is an additional step to the standard RMC
simulation.)
(d) Provided that the constraints are satisfied, calculate χn2 for the new trial
configuration.
2
(e) If χn2 < χ(n−1) , the new configuration becomes the starting configuration, i.e.
the move is accepted; otherwise it is accepted with a probability that follows
the normal distribution.
(f) Repeat the process from (b) until χn2 converges to its “equilibrium” value.
In this way, sets of particle coordinates are generated which are con-
sistent with a given diffraction data set and with the constraint(s) applied.
Figure 3.11 displays three stages of a RMC calculation in two dimensions.
Figure 3.11(a) shows the initial crystalline configuration (right), and the meas-
ured (thick solid line) and calculated (thin solid line) pair correlation func-
tions (left). Figure 3.11(b) displays a middle stage of the RMC simulation, and
Figure 3.11(c) represents a stage where the fit is near optimum (L. Pusztai,
private communication). Free RMC computer code is available at http://www.
szfki.hu/~nphys/rmc++/opening.html. A brief description of the computer code
can be obtained from Gereben et al. (2007) and Gereben and Pusztai (2012).
Constraints
In order to derive constraints, the following question should be answered: what
natural local configurations can be found in nature? During prepartion of the
amorphous phase, the covalently bonded atoms tend to seek out one of the dif-
ferent local configurations. Information about these local orders of the amorph–
ous structures might be provided by analyzing the embedded fragments inside
different large molecules. The environment might be considered as a “white
noise” around the fragments. The Cambridge Structural Database (CSD) (Allen,
Kennard, and Taylor, 1983), the world’s largest experimentally determined crys-
tal structure, has within it the results of x-ray and neutron-diffraction studies.
The CSD is designed as a critically evaluated numerical resource, containing
3D atomic coordinates. It provides a good opportunity to obtain measured lower
and upper limits for bond lengths, bond angles, second-neighbor distances, etc.,
if the CSD contains a large enough number of diffraction records.
A systematic search of structural data can be carried out for X–X and X–X–X
fragments on this database, where X denotes the element forming the disordered
system. We have collected the experimentally determined structural data of
molecules containing Si–Si–Si fragments (Kugler and Varallyay, 2001) and for
Se–Se–Se (Hegedus and Kugler, 2005). Recently, a new search was made1 ,
1 Lukács, R. and Harmath, V. (2011), private communication.

54 Structure
(a)
(b)
(c)
Figure 3.11. Three stages of a two-dimensional RMC simulation. (Provided by

L. Pusztai (Budapest).)
and Figures 3.12 and 3.13 display the results of these bond-angle distributions
for a-Si and a-Se, respectively. The number of Si–Si–Si fragments is equal to
8071, whereas for Se–Se–Se it is less (905 fragments). The results of a search
for a-Si are surprising. The overwhelming majority of these points fall in the
expected region, i.e. around 0.235 nm and 109.47°. The minimum bond length
Angle distribution for Si

0.4
0.35
0.3
0.25
Frequency
0.2
0.15
0.1
0.05
0
40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
Angle (degree)
Figure 3.12. Histogram of bond-angle distribution of Si–Si–Si fragments col-

lated from the CSD.
Angle distribution for Se

0.4
0.35
0.3
0.25
Frequency
0.2
0.15
0.1
0.05
0
40 50 60 70 80 90 100 110 120 130 140 150 160 170 180
Angle (degree)
Figure 3.13. Histogram of bond-angle distribution of Se–Se–Se fragments col-
lated from the CSD.
is about 0.22 nm, whereas the maximum is 0.27 nm. The average bond length
and angle are 0.237 nm and 106.3°, respectively. There are some extrema.
Two well-defined, unexpected regions can also be found, for example there are
some significant angles in an interval of 75°–96°. These angles belong to the
56 Structure
near planar square arrangement. Several squares were found in the database.
Bond angles around 60° can be also found. The second conclusion of bond-
angle analysis is that nearly equilateral triangles of Si3 are present among the
fragments. This is the other unexpected region, and most of the a-Si theoretical
models do not contain this type of structure, i.e. three- and four-membered rings.
A similar conclusion can be delivered for a-Se.
Tetrahedral amorphous carbon model constructed by reverse

Monte Carlo simulations
Tetrahedral a-C (ta-C) is one of several forms where most atomic sites are
fourfold coordinated. A neutron-diffraction measurement by Gilkes and co-
workers (Gilkes, Gaskell, and Robertson, 1995) has concluded that the structure
of ta-C consists predominantly of a disordered tetrahedral network, as does a-Si;
a-C is more complicated because a variety of local arrangements exists. In this
case, 3000 carbon atoms were put into a box with edges of 2.714 nm (Walters
et al., 1998). Four models were constructed with different constraints. Particular
attention was paid to local bonding environments. Walters et al. concluded that
better agreement with the measured data is achieved if the model is allowed to
include threefold and fourfold rings.
Amorphous silicon model constructed by reverse Monte Carlo simulation

The first RMC simulations including constraints were performed for pure a-Si by
Kugler et al. (1993a). A neutron-diffraction measurement on a pure evaporated
a-Si sample was performed at the high flux reactor in the ILL, Grenoble. The
experimentally measured structure factor was used as the input data for the RMC
calculations. Three structural models were constructed. In all the simulations
N = 1728 particles were confined in a cubic box of sides L = 3.2 nm. This
set-up (constraint) yielded the experimental microscopic number density ρ =
0.0505 Å−3 . A hard core diameter (lowest limit of bond length) of 0.22 nm was
also applied in order to efficiently avoid physically meaningless configurations.
About one million accepted steps were completed during these three different
runs.
The first simulation (referred to as Model 1) is the result of the RMC without
any other constraint. Model 2 was produced using a constraint on the coordin-
ation number: every atom in the configuration was required to have exactly four
1.0
(a)
0.5
S(Q)–1
0.0
–0.5
–1.0
0 5 10 15 20 25
Q (Å–1)
1.0
(b)
0.5
S(Q)–1
0.0
–0.5
–1.0
0 5 10 15 20 25
Q (Å–1)
Figure 3.14. (a) Comparison of the RMC solution for Model1 and the experi-
mental S(Q) for a-Si. (b) The same comparison for Model3 for a-Si. (Taken
with permission from Kugler et al. (1993). Phys. Rev. B, 48, 7685. Copyright
2013 by the American Physical Society.)
neighbors in the first coordination shell. The upper limit for the bond length
was 0.27 nm. The g(r) function obtained by the Fourier transformation of the
measured S(Q) almost reached its minimum between the first and second main
peaks at 0.27 nm. Furthermore, the CSD suggests a similar value for the max-
imum first-neighbor distance. Model 3 satisfied the most complicated constraint.
In addition to the atoms having four neighbors, if an attempted move resulted
in newly formed bond angles with an average that did not fit into a normal-
like distribution centered on the tetrahedral angle, the move was immediately
rejected. In systems such as a-Si, it is a normal requirement that most of the
angles should be roughly tetrahedral. Angles far from the tetrahedral value are
less probable, but permitted. Figure 3.14(a) shows the comparison of the RMC
solution for Model 1 and the experimental S(Q), and Figure 3.14(b) gives the
same comparison for Model 3. The overall agreement is quite good. It is thought
58 Structure
4 (a)
g(r)
2
0
0 2 4 6 8 10
r (Å)
5
4 (b)
3
g(r)
0
0 2 4 6 8 10
r (Å)
5
4 (c)
3
g(r)
0
0 2 4 6 8 10
r (Å)
Figure 3.15. (a), (b) Pair correlation functions calculated using RMC with dif-
ferent constraints. (c) Pair correlation function obtained by Fourier transforma-
tion of measured S(Q). (Taken with permission from Kugler et al. (1993). Phys.
Rev. B, 48, 7685–7688. Copyright 2013 by the American Physical Society.)
1.0
Frequency (arb. units)

0.5
0.0
–1.0 –0.5 0.0 0.5 1.0
Cosine of bond angle
Figure 3.16. Cosine of bond-angle distributions for three different models.
(Taken with permission from Kugler et al. (1993). Phys. Rev. B, 48, 7685–
that the greater amplitude of the oscillations at higher Q-values and the small-
angle-scattering part of the S(Q) curve could be fitted more successfully using a
much bigger box.
The RDF for Model 1 (identical to Model 2) is displayed in Figure 3.15(a),
and that for Model 3 is given in Figure 3.15(b). Figure 3.15(c) depicts g(r)
obtained by using a Fourier transform of the measured S(Q). The characteris-
tics of these functions are rather similar. Cosine distributions of bond angles
(cos θ ) lead to significant differences between microscopic structures obtained
using three different constraints (Figure 3.16). This makes clear that a diffraction
experiment by itself provides insufficient information to determine unambigu-
ously the microscopic local structure. Identical S(Q) and/or g(r) functions can
be consistent with substantially different bond-angle distributions. The cosine
distribution for Model 1 has a large peak at around 60°, and also a large propor-
tion of angles were found near 180°. However, an intense peak around 109.5°
is provided by the model. These results are completely different from those of
the other two models. Model 2 reflects probably the most fundamental require-
ment that atoms in any solid form of Si should be fourfold coordinated. This
simple constraint greatly reduces the freedom of moving atoms around. The
main problem with this picture might be the probable over-representation of
angles higher than 130°. Model 3 was introduced to narrow the cosine distri-
bution further, but no essential narrowing could be achieved compared with
Model 2.
60 Structure
Wooten’s model of a-Si contains internal bond angles between about 90° and
150°. The classical empirical Keating potential was applied for the construction
in which the interaction has a quadratic energy term for the difference between
the cosine bond angle and the cosine ideal bond angle. This term avoids the large
deviation from the canonical value of bond angle. Note that the RMC simulation
does not take the energy into account. The existence of angles smaller than
90° or 60° also agrees with the CSD. A Japanese group investigating a-Si:H by
RMC simulation also concluded the existence of small bond angles (Tabuchi
et al., 2004). Furthermore, another Japanese group (Ishimaru, Yamaguchi, and
Hirotsu, 2004) investigated silicon–germanium alloys by a MD method using a
classical empirical potential of Tersoff (Tersoff, 1989), and they also observed
bond angles of 60°.
A comparison between the experimental structure factors of evaporated and
ion-implanted a-Si has been modeled by the RMC simulation (Pusztai and
Kugler, 2005). A detailed comparison, in terms of the pair correlation func-
tion and the cosine distribution of bond angles, has been reported for the two
materials. It is found that for an acceptable RMC reproduction of the measured
structure factors, the evaporated models must contain more “small” bond angles
than their necessary abundance in the implanted models.
RMC modeling for comparison of a-C, a-Si, and a-Ge structures

A series of RMC simulations were carried out for a-C, a-Si, and a-Ge (Gereben
and Pusztai, 1994). A number of constraints were applied during the calculations
to test how many different assumptions are consistent with the given experimen-
tal data sets. The results of the unconstrained calculations seem to be the best
fit to these samples, but these models are not satisfactory from the electronic
structure point of view. The basic constraint that causes a problem is the exact
coordination number of four. The most characteristic common feature of these
amorphous semiconductors is not fourfold coordination.
Amorphous selenium model constructed by reverse

Monte Carlo simulation
Similar RMC work was carried out on the model material (a-Se) of chalcogenide
glasses by Jóvári and co-workers (Jóvári et al., 2003). Neutron-diffraction meas-
urements were performed on the SLAD instrument at NFL, Studsvik, on a
g(r) 2
0
1 2 3 4 5 6
r (Å)
Figure 3.17. Pair correlation functions obtained by RMC simulation applying
different constraints. Solid line: hard-sphere constraints only; dashed line: long
chains; and dots: Se8 rings. (Taken with permission from Jóvári, Delaplane,
and Pusztai (2003). Phys. Rev. B, 67, 172201. Copyright 2013 by the American
Physical Society.)
ball-milled a-Se sample. The obtained structure factor was interpreted by RMC
simulation. Selenium has several crystalline forms, consisting of Se8 rings or
chains. The authors concluded from the diffraction data that it is not possible
to determine whether a-Se consists of long chains, Se8 rings, or a mixture of
both phases. It is possible to construct both of the models whenever the structure
factors agreed quantitatively with measured data (Figure 3.17). The measured
sample was made from crystalline Se consisting of long chains; it is suggested
that the investigated sample contained primarily chain-like structures.
Molecular-dynamics simulation
An alternative 3D structure model construction is the deterministic molecular-
dynamics (MD) method for describing non-crystalline systems. The primary
application of MD is to follow the preparation process in situ and to reach a
condensed phase. In this process, all that is assumed is the validity of classical
mechanics. This approximation is never truly invalid. An indicator of the valid-
√
ity of classical mechanics is the de Broglie wavelength, λ = h/ 2mE. For a
particle moving with a kinetic energy of 1 eV, the de Broglie wavelength of
62 Structure
the particl is 1.2 nm (if it is an electron), 0.3 nm (hydrogen), 0.1 nm (carbon),

and 0.005 nm (silicon). In the condensed phase the characteristic distance is
the interatomic spacing, which is usually equal to 0.1– 0.3 nm. The motion of
a 1 eV electron in the solid state therefore can only be described by including
the quantum effect in the dynamics. On the other hand, classical mechanics is a
good approach to take when considering carbon, silicon, or selenium, etc. The
atoms are treated as point particles because the de Broglie wavelengths are small
enough.
The Newtonian equations of motion can be solved if the position and velocity
of N particles are given at a time t. It should be mentioned that an important
conclusion of quantum mechanics, the Heisenberg uncertainty relation, does not
allow this statement to remain valid. The uncertainty relation forbids knowing
the exact atomic position and momentum simultaneously with arbitrary precision
( px x ≥ h̄/2).
The simplest way of integrating the MD equations of motion is to use the
Verlet algorithm (Verlet, 1967). The atomic motion can be followed by the MD
method. The basic idea is to construct third-order Taylor expansions for r(t + dt)
and r(t – dt), where r(t) is the atomic position at time t:
1 1
r (t + dt) = r (t) + v (t) dt + a(t)dt 2 + b (t) dt 3 + O(dt 4 ), (3.33)
2 6
1 1
r (t − dt) = r (t) − v (t) dt + a(t)dt 2 − b (t) dt 3 + O(dt 4 ). (3.34)
2 6
Adding the two expressions gives
r (t + dt) = 2r (t) − r (t − dt) + a (t) dt 2 + O(dt 4 ), (3.35)
and then we get

r (t + dt) − r(t − dt)
v (t) = + O(dt 3 ). (3.36)
2dt
Acceleration a(t) at time t can be calculated from the atomic interactions as
follows:
a (t) = −(1/m)grad V (r (t)) . (3.37)
The application of the Hellmann–Feynman theorem is required to calculate

the forces in the quantum-mechanical case (Hellmann, 1933; Feynman, 1939).
Equation (3.37) is the basic equation of the Verlet algorithm. On the basis of
these equations, therefore, knowing (i) the positions of each particle at time t
and at the earlier time (t – dt), and (ii) V(r), we can derive the new configuration
at time (t + dt). Knowledge of r (t − dt) provides a little difficulty at the first
MD step, but this is easy to handle as long as we know the initial velocity.
Furthermore, we can calculate the atomic velocities at time t, which is important
for the temperature estimation. The local error in a given position is proportional
to O(dt 4 ). If we calculate the cumulative error for the positions over a given
time period T, we obtain
error (T ) = O(dt 2 ). (3.38)
More commonly, a velocity Verlet algorithm is used (Swope et al., 1982).

This new method uses a similar approach but explicitly incorporates velocity,
solving the first time step problem in the basic Verlet algorithm. The starting
equations are
1
r (t + dt) = r (t) + v (t) dt + a (t) dt 2 (3.39)
2
and
a (t) + a(t + dt)
v (t + dt) = v (t) + dt. (3.40)
2
First, r (t + dt) must be calculated; after that one should compute a(t + dt)
using the new position r(t + dt), and finally v (t + dt) must be derived. The
error for the velocity Verlet is of similar order to that for the basic Verlet
algorithm. The same computer memory is needed for this modified version.
In the leapfrog integration (similar to the velocity Verlet method), the velocity
is calculated in two steps. The new positions are calculated as in the velocity
Verlet algorithm. Then, the velocities at a half-time step are calculated from the
old velocities and accelerations:
1 1
v t + dt = v (t) + a(t)dt. (3.41)
2 2
Then the (t + dt) terms are computed, and finally the velocity can be obtained
in the following way:
1 1
v (t + dt) = v t + + a (t + dt) dt. (3.42)
2 2
It has been shown that this leapfrog algorithm is numerically more stable than
the version in which the velocities are calculated in one step.
64 Structure
There are other methods for the numerical integration of ordinary differential
equations of order 2, such as Beeman’s algorithm (Beeman, 1976), or the well-
known Runge–Kutta method, but in most cases the velocity Verlet algorithm is
applied.
Temperature control
In an amorphous sample, the preparation temperature plays a crucial role. The
temperature must be controlled during quenching from the liquid state and in
the deposition to a substrate. Most of the main preparation techniques utilize
low-temperature-induced amorphization. Each of these preparation techniques
is highly non-equilibrium in nature. A heat bath removes energy from the sample
during the amorphization. At every MD simulation step the temperature must be
monitored. The instantaneous temperature is given by the following well-known
statistical mechanics equation:
3 1N
NkB T = pi2 . (3.43)
2 i=1
2m
To simulate the heat conduction outside the simulation box, and to bring the
structure back to the desired temperature, one is required somehow to control the
temperature. During the process of film growth, particles bombard the growing
film. Deposition to a substrate requires that the substrate temperature is kept
constant. The instantaneous kinetic energies of the substrate atoms (and the
temperature of the substrate), however, may increase during MD simulations due
to particle impact, and they are dependent on the energy of the incoming particle.
At rapid cooling, the characteristic rate of quenching is around 106 K s−1 , or
lower in experiments. A variety of thermostat methods used to control the
temperature can be found in literature, but the Nosé–Hoover thermostat (Nose,
1984; Hoover, 1985; Chen, Wu, and Cheng, 2011) and the simpler velocity-
rescaling methods are the most popular. Equations describing the Nosé–Hoover
thermostat have a free parameter, and sometimes its derivation is difficult.
A MD computer code (ATOMDEP) has been developed to simulate the
preparation procedure of amorphous semiconductors, which are usually grown
by a vapor-deposition technique on a substrate (Kohary and Kugler, 2001). It
should be noted that experimentally no rapid quenching preparation method has
been reported for group IV amorphous materials. To integrate the MD equations
of motion, the velocity Verlet algorithm was included with a time step equal to
about 1 fs. The substrate temperature was fixed by velocity rescaling. During the
growth procedure, an atom started to move toward the target surface. In order
to avoid surface effects, periodic boundary conditions in two dimensions were
adopted. The initial x, y coordinates and the deposition energies were randomly
distributed. The bottom layer of the substrate was fixed at its ideal lattice site in
order to stabilize mechanically the substrate.
A typical deposition rate during sample preparation in the laboratory is around
10 atoms s−1 cm−2 . The surface of the substrate in computer experiments is
14
around 10–14 cm2 , i.e. one incoming atom per second should reach the sub-
strate. This deposition rate presents a difficulty for the simulation, and it is
impossible to handle easily using MD with today’s computer facilities within
a tight-binding atomic interaction model. Despite this disadvantage, the high
theoretical deposition rate applied is low enough to warrant the relaxation of the
previous atom before the next deposition. This developed computer code was
successfully applied to prepare a-C, a-Si, and a-Se, and this will be discussed in
the following section.
Amorphous carbon model construction by molecular-dynamics simulation

MD simulations for a-C were carried out to study the dynamics of the growth
process (Kohary and Kugler, 2001). The tight-binding Hamiltonian of Xu et al.
(1992) was used to calculate the interatomic potential between carbon atoms.
This realistic TB potential has already been successfully applied for different
carbon systems, even for the preparation of fullerenes (Laszlo, 1998, 1999).
The simulation technique took the following form. An ideal diamond film
consisting of 120 carbon atoms was employed to model the substrate. The
rectangular simulation cell was open along the positive z axis (the [111] direction
in our case). Periodic boundary conditions were used in two (x and y) directions.
Carbon atoms in the bottom substrate layer were fixed at their ideal lattice,
whereas the rest of the 96 atoms were allowed to move with full dynamics. To
simulate the constant substrate temperature, the kinetic energy of the movable
atoms in the substrate was rescaled at every time step. The time step was chosen
to be 0.5 fs.
The bombarding atoms were randomly placed in the x and y directions above
the substrate, as near as possible to any other atom, but no closer than the cut-
off distance of the potential. The initial velocities of bombarding atoms were
directed to the substrate. The initial velocities were chosen using the simple
66 Structure
Distance profile
T = 100 K
25 T = 300 K
Density of atoms in z directions
20
15
10
0
–5 –2.5 0 2.5 5 7.5 10 12.5
r (Å)
Figure 3.18. Density profiles of two networks (T = 100 K and T = 300 K)
are displayed perpendicular to the substrate surface. The arrows represent the
layer positions of a perfect diamond crystal. A memory effect is observed near
the surface of the substrate. (Taken with permission from Kohary and Kugler
(2001). Phys. Rev. B, 63, 193404. Copyright 2013 by the American Physical
Society.)
relation v 2 = (2E/m)(1.2 − 0.4p), and the directions were determined using

θ = 120◦ + p60◦ and ϕ = p360◦ , where θ and ϕ are the polar angles, and p
is a uniformly distributed random number between 0 and 1. The frequency of
the atomic injection was, on average, f = 1125 fs−1 . This flux is orders of
magnitude greater than the deposition rate usually applied in real experiments.
The lower substrate temperatures applied in the simulations result in faster
energy dissipation, which compensates for the high deposition rate.
Models were constructed with average bombarding kinetic energies (E) of
1 eV and 5 eV on the substrate with temperatures of 100 K and 300 K. In all
cases the structures were relaxed for 5 ps after deposition times of 25 ps and
40 ps. The density profiles of the two networks (T = 100 K and T = 300 K)
are displayed perpendicular to the substrate surface in Figure 3.18. The arrows
represent the layer positions of a perfect diamond crystal. The difference in
temperature between two substrates causes differences in memory effects. The
structure on the surface at 100 K has a more pronounced layering effect than
the other network on the substrate at room temperature. In the first 0.3–0.4 nm
Partial radial distribution functions

80
‘C–C’
‘C3–C3’
70 ‘C3–C4’
‘C4–C4’
60
50
J(r)
40
30
20
10
0
1 1.5 2 2.5 3
r (Å)
Figure 3.19. Partial radial distribution functions J(r) of an a-C model con-
structed by TBMD simulation. Numbers following letters C refer to the coor-
dination numbers of carbon atoms. (Taken with permission from Kohary and
Kugler (2001). Phys. Rev. B, 63, 193404. Copyright 2013 by the American
Physical Society.)
thick layers over the [111] surface, the sp3 content is high due to the memory
effect. In the rest of the bulk, the sp2 content is dominant. The partial distribution
functions of a model (T = 100 K, E = 1 eV) are shown in Figure 3.19, where
i and j denote the coordination numbers of the connected atoms. The average
first-neighbor distances are 0.143 nm, 0.152 nm, and 0.159 nm for C3–C3, C3–
C4, and C4–C4, respectively. The average coordination numbers for different
models were between 3.1 and 3.4. Coordination numbers larger than four were
not observed.
Angles are analyzed in detail according to the coordination numbers of their
neighbors. Atoms with sp3 local arrangements have nearly the same average
values in the C3–C4–C3, C3–C4–C4, and C4–C4–C4 cases, which are around
the tetrahedral angle. Clear differences are observed for sp2 configurations. The
C4–C3–C4 angles have much lower average values than the others. The latter
average bond angles are close to 120°. This means that threefold-coordinated
central atoms with at least two C3 first neighbors are always near planar struc-
tures, even if the third one is C4. The ring statistics for our models were also
calculated. In the graphitic and diamond networks, sixfold and eightfold rings
68 Structure
are present. The size of a ring is denoted by the number of atoms in a closed path.
In our models, even-membered as well as odd-membered rings can be found,
i.e. 4–7-membered rings.
Amorphous silicon model construction by molecular-dynamics simulation

Amorphous silicon networks were constructed by atomic deposition (and by
rapid quenching) (Kohary and Kugler, 2004). The same molecular-dynamics
computer code (ATOMDEP) as was used for the simulation of atomic bombard-
ment was run. In the simulations, the transferable TB Hamiltonian of Kwon
et al. (1994) was applied to describe the interaction between the silicon atoms.
This group had already developed an excellent TB potential for carbon systems.
All the parameters and functions of the interatomic potential for silicon were
fitted to the results of the local density-functional calculations. The TB model
reproduces the energies of different cluster structures, the elastic constants, the
formation energies of vacancies, and interstitials in crystalline silicon. Accord-
ing to the authors, the only disadvantage of this TB model is that the bond
lengths inside small clusters are slightly longer than those derived from other
theoretical calculations or from experiments.
A surprising result was found in the ring statistics. The networks prepared by
the models have a significant number of squares. Furthermore, triangles are also
present in the atomic arrangements, as shown in Figure 3.20. Most of the theoret-
ical models for a-Si do not contain such structural fractions. This seems to be an
important result. It should be mentioned that a neutron-diffraction measurement
carried out on a pure evaporated a-Si sample and evaluated by the RMC method
led to a similar conclusion (Kugler et al., 1993a). The simulation was repeated
using another TB model (Lenovsky et al., 1997), and similar conclusions were
obtained. The a-Si structure seems to be well understood, but still newer and
newer models appear each year (Gibson, 2012; Treacy and Borisenko, 2012).
Amorphous selenium model construction by

molecular-dynamics simulation
Two different basic preparation methods, liquid quenching and evaporation, for
a-Se are available experimentally in the laboratory. The MD computer code
ATOMDEP was used to model the structure of a-Se. Two different atomic
interactions were used. First, the classical empirical three-body potential of
Oligschleger et al. (1996) was included in the computer code. A crystalline
lattice cell containing 324 selenium atoms was employed to mimic the substrate.
Figure 3.20. An a-Si model constructed by computer simulation of atomic

deposition using the TBMD method. The atomic arrangement contains triangles
(shaded).
There were 108 fixed atoms at the bottom of the substrate. The remaining
atoms could move with full dynamics. The simulation cell was open along the
positive z direction, and periodic boundary conditions were applied in the x and y
directions. The kinetic energy of the atoms in the substrate was rescaled at every
MD step (dt = 1 fs) to keep the substrate at a constant temperature of 100 K.
In the deposition process the frequency of the atomic injection was 300 fs−1
(Figure 3.21).
Three different structures were constructed, with average bombarding ener-
gies of 0.1 eV, 1 eV, and 10 eV. At the end of the deposition, atomic networks
contained about 1000 selenium atoms. Rapid cooling is frequently applied
to construct glassy structures. The system is usually cooled down to room
70 Structure
Figure 3.21. Part of the a-Se structure grown with 1 eV bombarding energy at
100 K. There are 324 substrate atoms at the bottom, of which 108 were held
fixed to mimic the bulk crystal. (From Hegedus, Kohary, and Kugler (2004a).
J. Non-Cryst. Solids, 338–340, 283. Copyright 2013 with permission from
Elsevier.)
temperature at a rate of 1011 –1016 K s−1 in computer simulations, although

this rate is some orders of magnitude smaller in the experimental techniques.
In order to retrieve information on the rapid cooling, we prepared a model in
the following way. The temperature of one of the deposited films (1 eV) was
increased to 900 K as an initial state (liquid phase). After this melting, the
trajectories of the selenium atoms were followed by full dynamics for 100 ps.
The substrate temperature maintained at 100 K leads to the cooling of the film
above the substrate. This technique can be considered as the computer simula-
tion of real splat cooling, where small droplets of melt are brought into contact
with the chill-block, prepared by rapid quenching. This is more homogeneous
y
Figure 3.22. One of the several selenium glassy networks constructed using
the rapid cooling technique. It contains 162 Se atoms with periodic boundary
condition in the x and y directions. (From Hegedus et al. (2004b). J. Non-Cryst.
Solids, 338–340, 557. Copyright 2013 with permission from Elsevier.)
than the deposited counterpart, and is an observable difference between the two
preparation techniques.
Secondly, for the description of the interaction between the selenium atoms,
we used the TB model developed by Molina and co-workers (Molina et al.,
1999). Self-consistency was taken into account via the usual on-site Hubbard
term, and was found to reduce any large charge transfer (Lomba, Molina, and
Alvarez, 2000). The time step was equal to 2 fs. The temperature was controlled
via the velocity-rescaling method. Several different glassy selenium networks
were prepared in a rectangular box with periodic boundary conditions; samples
contained 162 atoms (Figure 3.22). The “cook and quench” sample preparation
is described as follows.
First, the temperature of the system was kept at 5000 K for the first 300 MD
steps to randomize the atomic positions. During the next 2200 MD steps, the
temperature was decreased linearly from 700 K to 250 K, driving the sample
through the glass transition and reaching the condensed phase. Then the final
temperature of 20 K was reached and the sample for 500 MD steps (1 ps)
was relaxed. The periodic boundary conditions were lifted along the z direc-
tion at this point. This procedure provided a slab geometry containing periodic
boundary conditions in only two dimensions. The system was then relaxed
for another 40 000 MD steps (80 ps) at 20 K. Short quenching times in the
simulation compared to those in the experiments might lead to many liquid-
state defects being retained in the amorphous structure. Therefore, the Hubbard
72 Structure
25
Radial distribution function

20
15
10
0
2.0 2.5 3.0 3.5
°
Distance (A)
Figure 3.23. Radial distribution function of selenium glassy networks at T =

20 K (solid line) and T = 300 K (dashed line). (Taken with permission from
Hegedus et al. (2005). Phys. Rev. Lett., 95, 206803. Copyright 2013 by the
American Physical Society.)
parameter U was taken to be 5 eV for the first 4000 MD steps during quenching to
avoid a large number of coordination defects, especially onefold- and threefold-
coordinated atoms. Then U was changed to its accepted value of 0.875 eV for
selenium. The RDF for one of the selenium glassy networks at 20 K is shown in
Figure 3.23. The first main peak at 0.24 nm belongs to covalently bonded
atoms. The crystalline nearest-neighbor distance is 0.237 nm. The second peak
at 0.36 nm corresponds to the intrachain second-nearest-neighbor distance as
expected. The prepeak at 0.33 nm reflects the smallest interchain atomic dis-
tances in a-Se. In simulations at higher temperatures (T = 300 K), these two
peaks merge, as observed in Figure 3.23.
Although simulations based on empirical potentials or TB models have
achieved remarkable success, extension of the simulation methods to more ac-
curate quantum mechanics would be desirable. Unfortunately, DFT within
the most popular local-density approximation (LDA) is very computer time
consuming – it is difficult to perform a separate self-consistent electronic mini-
malization at every MD step. An approach known as the Car–Parrinello method
(Car and Parrinello, 1985) overcomes this difficulty, and it has been successfully
applied to MD simulations of covalently bonded amorphous semiconductors,
yielding valuable structural information for small systems. Structural simula-
tions were carried out on the basis of the Car–Parinello method for a-C (Galli,
Martin, Car, and Parrinello, 1989), a-C:H (Iarloni, Galli, and Martini, 1994),
ta-C (Marks et al., 1996), a-Si (Stich et al., 1991), and a-Si:H having far fewer
than 100 atoms (Buda, Chiarotti, Car, and Parrinello, 1991).
3.4 Phase change and its applications

The optical phase-change disk, i.e. the digital versatile disk (DVD), is the most
successful commercialized product that uses amorphous chalcogenides (a-Chs)
(see Raoux and Wuttig, 2009). It should also be emphasized that the applications
of phase change are fully related to the structural changes of materials. The rep-
resentative phase-change material is Ge2 Sb2 Te5 (stoichiometric composition of
GeTe–Sb2 Te3 ), conventionally called GST. This possesses a pronounced differ-
ence in its optical and electronic properties depending upon the atomic arrange-
ment, i.e. whether it is in the amorphous or crystalline phase. It is therefore very
useful in rewritable optical storage and random access memory devices, since
transitions between two states are performed over a nanosecond time range.
In rewritable optical data storage, a short pulse of a focused high-intensity
laser beam locally heats GST in the crystalline state above its melting tempera-
ture (600 °C). Rapid cooling quenches the melting spots into the amorphous
phase. This “writing” process takes 10 ns, and “reading” (detecting this spot)
is achieved by reflection change using a weak-intensity laser beam. The optical
reflectivity is decreased by more than 5% after reaching the amorphous state.
To erase the memory spots, a laser pulse with medium power for local heating
above the crystallization temperature (150 °C) is applied. Thus the memory
spots return to the original crystalline states in 50 ns.
In electronic memory application (phase-change random access memory,
PRAM), a short pulse with high-intensity voltage (or current) – instead of a
laser pulse (optical recording) – is used to change the phase of materials. The
local heating due to this process converts the materials into the amorphous
state. “Reading” is of course performed electrically by a low-voltage pulse.
The obtained amorphous state has a high resistivity, which is four orders of
magnitude larger than that of the crystalline case. An intermediate moderate
intensity pulse is used to heat the material above its crystallization temperature
and the memory state returns to the original low-resistivity crystalline state. Note
therefore that the process is similar to that of optical recording, and “heating” by
pulse (optical or electric) is a major requirement in optical (DVD) and electronic
(PRAM) processes (Atwood, 2008; Terao, Morikawa, and Ohta, 2009).
We now present an overview of the structural properties during phase changing
in GST films. As the crystallization process limits the rate of the overall writing
and erasing times (i.e. the rate of crystallization is slower than that for the
reverse – crystalline to amorphous states – action), it is important to understand
74 Structure
the crystallization dynamics. It should be noted, however, that the crystallization

process by pulsed heating in devices may not be the same as that by isothermal
heating.
Crystallization kinetics
The dynamics of crystallization in glasses has been fairly well interpreted in
terms of the classical nucleation and growth model developed by Kolmogorov,
Johnson and Mehl, and Avrami (the so-called KJMA model) (see, for ex-
ample, Pineda and Crespo, 1999; Tonchev and Kasap, 2006). The KJMA model
yields the macroscopic evolution of the transformed phase under an isothermal
annealing condition with its simplicity, and it has been extended to analyze
transformations under a constant heating rate. The KJMA formalism may still
present some difficulties, for example fitting the KJMA results to experimental
data produces unreasonable physical parameters for the crystallization, which
have not been well understood so far (Senkader and Wright, 2004).
We briefly discuss now what kind of difficult problems exist in the KJMA
formalism. We consider only the macroscopic properties, without addressing any
specific microscopic view, which is still a matter of debate. These topics will
be described later on. The KJMA theory basically assumes that the nucleation
sites for the new phase are randomly distributed over the volume V, and that the
growth ceases in a region where two neighboring new phase grains impinge on
each other. The transformed volume Vtr is given by

V tr
dV = 1 −
tr
dV ex . (3.44)
V
Here, V ex is called the extended volume, which is the virtual volume of the
particles growing without any impingement. Integration of the above equation
produces the familiar KJMA equation:

f = 1 − exp −f ex , (3.45)
where f = V tr V and f ex = V ex V, and f is the transformed volume fraction. In

the case of instantaneous nucleation of spherical grains with number density N,
the volume growth rate with radius r for an interface-controlled reaction can be
expressed as
dV ex dr
= 4π r 2 NV . (3.46)
dt dt
Here the effective growth rate is given by G = drdt, and then f ex at time t is
expressed by
4π 4π
f ex = Nr 3 = N(Gt)3 . (3.47)
3 3
We then obtain the well-known KJMA equation:

1 − f = exp −(kt)n , (3.48)
with n = 3. Here, n is called the Avrami exponent, and k is an effective rate

constant (growth rate). The Avrami exponent should ideally be an integer pro-
viding information on the dimensionality of the crystallization process: n = 1 for
rod-like crystals; n = 2 for plate-like crystals; and n = 3 for spherical forms.
Many experiments, however, yield numerical values that deviate significantly
from the KJMA prediction. The values of n are, for example, between 1 and 5.8
in GST, and the frequency factor ν, appearing in the reaction-rate k(T), takes
unreasonably large values (1017 −1024 s−1 ) (Senkader and Wright, 2004):
EA
k(T ) = ν exp − , (3.49)
kB T
where EA is the activation energy for the phase change and T is the temperature.
Two problems that result from the KJMA formalism stated above can be
overcome without losing its simplicity and beauty. The first problem is the
Avrami exponent n. The theory predicts that, under heterogeneous nucleation, n
represents the Euclidean dimensions (n = 1, 2, 3). However, the values of n are
reported to be non-integer, and sometimes n is bigger than three. Non-integer
values of n appear as a violation of the fundamental assumptions of the KJMA
formalism, which requires: (i) a random distribution of the potential sites of
nucleation, (ii) instantaneous nucleation, (iii) interface-controlled grain growth,
and (iv) time- (or grain-size-) independent growth rate k.
In the following, we introduce a fractal growth concept to the space dimension
(Shimakawa, 2012). Actually, n = 2.15 is found in supramolecular materials,
and this has been attributed to the fractal growth of network aggregations (Liu
and Sawant, 2002). It is generally expected that the surfaces of most aggregated
materials should have a fractal dimension. The extended volume growth rate,
eqn. (3.46), can then be modified as follows:
dV ex dr
= Cr D NV , (3.50)
dt dt
76 Structure
where C is a constant and D is the fractal dimension of the surface of the grains.
Finally, we obtain f :

1 − f = exp −(kt)D+1 . (3.51)
It is therefore suggested that various values of the experimentally observed

Avrami exponent n can be attributed to the complex shape of the crystalline
grain surfaces with a fractal structure. This fractal-surface-controlled (FSC)
crystallization may be supported by reports, for example, that D 3.0 for silica
gels and D = 2.6–2.9 for sandstones (Shimakawa, 2012).
The second problem is the temperature-dependent rate constant k(T) (see
eqn. (3.49)). The reported activation energies EA , for example in GST, are
around 2 eV (1−3 eV). We recall that unreasonably large frequency factors
(1017 −1024 s−1 ), corresponding to 102 –109 eV, can be deduced from the appli-
cation of the classical KJMA model. These ν values are clearly unacceptable.
What is the origin of these unreasonable values for the frequency factor? Many
thermally activated phenomena, such as electronic and ionic transport, and struc-
tural relaxation in disordered matter, obey the Meyer–Neldel compensation rule
(Yelon, Movaghar, and Crandall, 2006). In eqn. (3.49), k(T) should be modified
as follows:
EA EA
k(T ) = ν0 exp exp − , (3.52)
EMN kB T
where ν is given by
EA
ν ≡ ν0 exp , (3.53)
EMN
where EMN is the Meyer–Neldel energy, and ν 0 is a constant. Equation (3.52) is
called the Meyer–Neldel (MN) rule or compensation law (Yelon et al., 2006).
Introduction of the fractal geometry and the MN rule on thermally acti-
vated processes (extended KJMA) overcomes some of the difficulties with the
classical KJMA model, although the understanding of crystallization from the
microscopic point of view is still lacking. In particular, for GST, due to its
extraordinary fast crystallization rate and small area, the microscopic structural
changes during nucleation and growth are not easy to detect experimentally.
Ab initio molecular-dynamics (AIMD) simulations based on density-
functional theory have been applied to see what happens in the early stage
of crystallization of GST (Lee and Elliott, 2011). The simulations confirm the
stochastic nature of the onset of crystallization as assumed in classical nucleation
theory. An incubation period for crystallization is clearly defined on annealing at

600 K. The critical crystal molecules are estimated to comprise five to ten chem-
ically ordered (Ge,Sb)4 Te4 units. Tellurium atoms in the crystalline phase form a
perfect fcc sublattice of rocksalt structure, and Ge or Sb atoms occupy sites in the
other sublattice. The simulations make clear that the formation of ordered planar
structures in the amorphous phase plays a critical role in lowering the interfacial
energy between crystalline clusters and the amorphous phase, yielding ultrafast
crystallization in GST. The above prediction from the AIMD simulations is very
important, and therefore the simulations will be discussed in the following.
Structural transformations resulting from phase changes in GST

In the previous subsection, we discussed the early stage of the crystallization
process via AIMD simulations (Lee and Elliott, 2011). Now we discuss the
structure of GST, in particular the structural difference between its amorphous
and crystalline states. We need to understand why GST shows fast phase transi-
tions over a nanosecond time scale. Many structural models have been proposed
to interpret the fast phase transitions (Kolobov et al., 2004; Hegedus and Elliott,
2008; Elliott (2009); Yamada, 2009). Although the detailed nature of GST struc-
ture is still disputed, the recently suggested resonance bonding model may be
useful in understanding the difference between the crystalline and amorphous
states in GST (Shportko et al., 2008). It should be noted, however, that there is
no clear evidence of resonance bonding.
We briefly review the principal structural models. The common understanding
of the GST structure is that crystalline GST has a (distorted) rocksalt structure
and that the Ge environment is fourfold coordinated in the amorphous phase.
There is still disagreement on the detailed structure among the modeling studies.
Kolobov et al. (2004) have proposed the so-called umbrella flip model from
the x-ray absorption near-edge structure (XANES) spectra. While an fcc-like
structure is retained in both the crystalline and re-amorphized states, the Ge atom
occupies octahedral and tetrahedral symmetry positions in the re-crystallized and
re-amorphized states. An intense laser pulse causes the longer bonds to break,
and the Ge atom moves into the tetrahedral symmetry position. A molten (liquid)
state as a transit path is not required from the crystalline to the amorphous states
in this model.
Yamada (2009) takes another view in the subject of amorphization. The crys-
talline phases (distorted rocksalt structure) show very large thermal vibrations
78 Structure
at high temperatures. This increases the randomness of the atomic distributions,

which is rather similar to that of the liquid phase. The amorphous structure
reflects the very isotropic atomic configuration in the liquid state. Thus the
atomic arrangements for both crystalline and amorphous states are very close,
and the transit time for phase change should be very short.
Finally, we discuss the very large static dielectric constant ɛ(0) in the crys-
talline state of GST. It is twice as large for the crystalline phase as it is for the
amorphous phase. This difference is attributed to a significant change in bonding
between the two phases (Shportko et al., 2008). The optical dielectric constant
of the amorphous phase is that expected of a covalent bonding (following the
8 − N rule), whereas that of the crystalline phase is expected to be strongly
enhanced by resonant bonding effects. The existence of resonance bonding
requires a longer range order than the conventional covalent bond length, whereas
an electron pair requires only the ordering of the nearest neighbors. Note that
the second- and higher-order neighbors should be aligned in resonance bonding.
As a higher ɛ(0) is also expected for smaller bandgap materials (Phillips, 1973),
the existence of resonant bonds may not be the condition necessary to produce
a higher dielectric constant.
A unique feature of the crystalline state of phase-change materials (not only
GST) is based on a distorted cubic structure. Various novel techniques discussed
in this chapter may clarify this complex issue. A recent monograph (Kolobov and
Tominaga, 2012) can provide a deeper understanding of phase-change materials
for readers.
References
Abell, G.C. (1985). Empirical chemical pseudopotential theory of molecular and metallic
bonding. Phys. Rev. B, 31, 6184–6196.
Adam, G. and Gibbs, J.H. (1965). On the temperature dependence of cooperative relax-
ation properties in glass-forming liquids. J. Phys. Chem., 43, 139–146.
Allen, F.H., Kennard O., and Taylor, R. (1983). Systematic analysis of structural data as
a research technique in organic chemistry. Acc. Chem. Res., 16, 146–153.
Anderson, P.W., Halperin, B.I., and Varma, C.M. (1972). Anomalous low-temperature
thermal properties of glasses and spin glasses. Phil. Mag., 25, 1–9.
Angell, C.A. (1988). Structural instability and relaxation in liquid and glassy phases
near the fragile liquid limit. J. Non-Cryst. Solids, 102, 205–221.
Angell, C.A. (1995). Formation of glasses from liquids and biopolymers. Science, 267,
1924–1935.
Aniya, M. (2002). A model for the fragility of the melts. J. Therm. Anal. Calorim., 69,
971–978.
References 79
Atwood, G. (2008). Phase-change materials for electronic memories. Science, 321,

210–211.
Barna, A., Barna, P.B., Radnoczi, G., Toth, L., and Thomas, P. (1977). A comparative
study of the structure of evaporated and glow discharge silicon. Phys. Stat. Sol. A,
41, 81–84.
Beeman, D. (1976). Some multistep methods for use in molecular dynamics calculations.
J. Comput. Phys., 20, 130–139.
Bellissent, R., Menelle, A., Howells, W.S. et al. (1989). The structure of amorphous
Si:H using steady state and pulsed neutron sources. Physica B, 156–157, 217–
219.
Bose, S.K., Winer, K., and Andersen, O.K. (1988). Electronic properties of a realistic
model of amorphous silicon. Phys. Rev. B, 37, 6262–6277.
Brey, L., Tejedor, C., and Verges, J.A. (1984). Comment on “Static charge fluctuations
in amorphous silicon”. Phys. Rev. Lett., 52, 1840.
Buda, F., Chiarotti, G.L., Car, R., and Parrinello, M. (1991). Structure of hydrogenated
amorphous silicon from ab initio molecular dynamics. Phys. Rev., B44, 5908–5911.
Car, R. and Parrinello, M. (1985). Unified approach for molecular dynamics and density-
functional theory. Phys. Rev. Lett., 55, 2471–2474.
Chen, W-H., Wu, C-H., and Cheng, H-C. (2011). Modified Nosé–Hoover thermostat
for solid state for constant temperature molecular dynamics simulation. J. Comput.
Phys., 230, 6354–6366.
Cooper, N.C., Goringe C.M., and McKenzie, D.R. (2000). Density functional theory
modelling of amorphous silicon. Comput. Mater. Sci., 17, 1–6.
Cser, L., Török, Gy., Krexner, G., Sharkov, I., and Faragó, B. (2002). Holographic
imaging of atoms using thermal neutrons. Phys. Rev. Lett., 89, 175504, 1–4.
Cuello, G.J. (2008). Structure factor determination of amorphous materials by neutron
diffraction. J. Phys.: Condens. Matter, 20, 244109, 1–9.
Ding, K. and Andersen, H.C. (1986). Molecular-dynamics simulation of amorphous
germanium. Phys. Rev. B, 34, 6987–6991.
Drabold, D.A., Fedders, P.A., Klemm, S., and Sankey, O.F. (1991). Finite-temperature
properties of amorphous silicon. Phys. Rev. Lett., 67, 2179–2182.
Elliott, S.R. (1990). Physics of Amorphous Materials, 2nd edn. Harlow: Longman Sci-
entific & Technical.
Elliott, S.R. (2009). Structure of amorphous Ge-Sb-Te solids. In Phase Change Materi-
als: Science and Applications, eds. S. Raoux and M. Wuttig. New York: Springer,
chap. 4.
Etherington, J.H., Wright, A.C., Wenzel, J.T., Dore, J.C., Clarke, J.H., and Sinclair,
R.N. (1982). A neutron diffraction study of the structure of evaporated amorphous
germanium. Non-Cryst. Solids, 48, 265–289.
Feynman, R.P. (1939). Forces in molecules. Phys. Rev., 56, 340–343.
Fortner, J. and Lannin, J.S. (1989). Radial distribution functions of amorphous silicon.
Phys. Rev. B, 39, 5527–5530.
Freitas, R.J., Shimakawa, K., and Kugler, S. (2013). Some remarks on glass-transition
temperature in chalcogenide glasses: a correlation with the microhardness. Chalco-
genide Lett., 10, 39–43.
Galli, G., Martin, R.M., Car, M., and Parrinello, M. (1989). Structural and electronic
properties of amorphous carbon. Phys. Rev. Lett., 62, 555–558.
Gaskell, P.H., Saeed, A., Chieux, P., and McKenzie, D.R. (1991). Neutron-scattering
studies of the structure of highly tetrahedral amorphous diamondlike carbon. Phys.
Rev. Lett., 67, 1286–1289.
80 Structure
Gereben, O. and Pusztai, L. (1994). Structure of amorphous semiconductors: reverse

Monte Carlo studies on a-C, a-Si, and a-Ge. Phys. Rev. B, 50, 14136–14143.
Gereben, O. and Pusztai, L. (2012). RMC_POT, a computer code for reverse Monte Carlo
modeling the structure of disordered systems containing molecules of arbitrary
complexity. J. Comput. Chem., 33, 2285–2291.
Gereben, O., Jóvári, P., Temleitner, L., and Pusztai, L. (2007). A new version of the
RMC++ Reverse Monte Carlo programme, aimed at investigating the structure of
covalent glasses. J. Optoelectron. Adv. Mater., 9, 3021–3027.
Gibson, J.M. (2012). Solving amorphous atructures – two pairs beat one. Science,
335, 929–930.
Gilkes, K.W.R., Gaskell, P.H., and Robertson, J. (1995). Comparison of neutron-
scattering data for tetrahedral amorphous carbon with structural models. Phys.
Rev. B, 51, 12303–12312.
Goodwin, L., Skinner, A.J., and Pettifor, D.G. (1989). Generating transferable tight-
binding parameters: application to silicon. Europhys. Lett., 9, 701–706.
Greaves, G.N. and Sen, S. (2007). Inorganic glasses, glass-forming liquids and amorph-
izing solids. Adv. Phys., 56, 1–166.
Guttman, L., Ching, W.Y., and Rath, J. (1980). Charge-density variation in a model of
amorphous silicon. Phys. Rev. Lett., 44, 1513–1516.
Hegedus, J. and Elliott, S.R. (2008). Microscopic origin of the fast crystallization
ability of Ge-Sb-Te phase-change memory materials. Nature Mater., 7, 399–
405.
Hegedus, J. and Kugler, S. (2005). Growth of amorphous selenium thin films: classical
versus quantum mechanical molecular dynamics simulation. J. Phys.: Condens.
Matter, 17, 6459–6468.
Hegedus, J., Kohary, K., and Kugler, S. (2004a). Comparative analysis of different prep-
aration methods of chalcogenide glasses: molecular dynamics structure simulations.
J. Non-Cryst. Solids, 338–340, 283–286.
Hegedus, J., Kohary, K., Kugler, S., and Shimakawa, K. (2004b). Photo-induced volume
changes in selenium. Tight-binding molecular dynamics study. J. Non-Cryst. Solids,
338–340, 557–560.
Hegedus, J., Kohary, K., Pettifor, D.G., Shimakawa, K., and Kugler, S. (2005).
Photo-induced volume changes in amorphous selenium. Phys. Rev. Lett., 95,
206803, 1–4.
Hellmann, H.G.A. (1933). Zur rolle der kinetischen Elektronenenergie für die zwei-
schenatomaren Krafte. Z. Phys., 85, 180–190.
Hensel, H., Klein, P., Urbassek, H.M., and Frauenheim, T. (1996). Comparison of
classical and tight-binding molecular dynamics for silicon growth. Phys. Rev. B,
53, 16497–16503.
Herrero, C.P. (2000). Quantum atomic dynamics in amorphous silicon; a path-integral
Monte-Carlo simulation. J. Phys.: Condens. Matter, 12, 265–274.
Hoover, W.G. (1985). Canonical dynamics: equilibrium phase-space distributions. Phys.
Rev. A, 31, 1695–1697.
Iarloni, S., Galli, G., and Martini, O. (1994). Microscopic structure of hydrogenated
amorphous carbon. Phys. Rev. B, 49, 7060–7063.
Ishimaru, M. (2001). Molecular-dynamics study on atomistic structures of amorp-
hous silicon. J. Phys.: Condens. Matter, 13, 4181–4189.
Ishimaru, M., Munetoh, S., and Motooka, T. (1997). Generation of amorphous silicon
structures by rapid quenching: a molecular-dynamics study. Phys. Rev. B, 56,
15133–15138.
References 81
Ishimaru, M., Yamaguchi, M., and Hirotsu, Y. (2004). Structural relaxation of amorph-
ous silicon-germanium alloys: molecular-dynamics study. Jpn. J. Appl. Phys., 43,
7966–7970.
Jian, Z., Kaiming, Z., and Xide, X. (1990). Modification of Stillinger-Weber potentials
for Si and Ge. Phys. Rev. B, 41, 12915–12918.
Jóvári, P., Delaplane, R.G., and Pusztai, L. (2003). Structural models of amorphous
selenium. Phys. Rev. B, 67, 172201, 1–4.
Justo, J.F., Bazant, M.Z., Kaxiras, E., Bulatov, V.V., and Yip, S. (1998). Interatomic
potential for silicon defects and disordered phases. Phys. Rev. B, 58, 2539–2550.
Kakinoki, J., Katada, K., Hanawa, T., and Ino, T. (1960). Electron diffraction study of
evaporated carbon film. Acta Cryst., 13, 171–179.
Kauzmann, W. (1948). The nature of glassy state and the behaviour of liquids at low
temperatures. Chem. Rev., 43, 219–256.
Keating, P.N. (1966). Effect of invariance requirements on the elastic strain energy of
crystals with application to the diamond structure. Phys. Rev., 145, 637–645.
Kelires, P.C. and Tersoff, J. (1988). Glassy quasithermal distribution of local geometries
and defects in quenched amorphous silicon. Phys. Rev. Lett., 61, 562–565.
Klug, D.D. and Whalley, E. (1982). Effective charges of amorphous silicon, germanium,
arsenic, and ice. Phys. Rev. B, 25, 5543–5546.
Kohary, K. and Kugler, S. (2001). Growth of amorphous carbon. Low energy molecular
dynamics simulation of atomic bombardment. Phys. Rev. B, 63, 193404, 1–4.
Kohary, K. and Kugler, S. (2004). Growth of amorphous silicon: low energy molecular
dynamics simulation of atomic bombardment. Mol. Simul., 30, 17–22.
Kolobov, A and Tominaga, J. (2012). Chalcogenides: Metastability and Phase Change
Phenomena. Berlin and Heidelberg: Springer.
Kolobov, A.V, Fons, P., Frenkel, A.I., and Ankudinov, A.L. (2004). Understanding of
the phase-change mechanism of rewritable optical media. Nature Mat., 3, 703–708.
Kramer, B., King, H., and Mackinnon, A. (1983). Charge fluctuations in hydrogenated
amorphous silicon. J. Non-Cryst. Solids, 59–60, 73–76.
Kugler, S. and Náray-Szabó, G. (1987). Charge distribution in amorphous silicon clus-
ters: quantum chemical study combined with ring statistics. J. Non-Cryst. Solids,
97–98, 503–506.
Kugler, S. and Náray-Szabó, G. (1991a). Weak bonds and atomic charge distribution in
hydrogenated amorphous silicon. J. Non-Cryst. Solids, 137–138, 295–298.
Kugler, S. and Náray-Szabó, G. (1991b). Atomic charge distribution in diamondlike
amorphous carbon. Jpn. J. Appl. Phys., 30, L1149–L1151.
Kugler, S. and Varallyay, Z. (2001). Possible unusual atomic arrangements in the struc-
ture of amorphous silicon. Phil. Mag. Lett., 81, 569–574.
Kugler, S., Surjan, P.R., and Náray-Szabó, G. (1988). Theoretical estimation of static
charge fluctuation in amorphous silicon. Phys. Rev. B, 37, 9069–9071.
Kugler, S., Molnar, G., Peto, G. et al. (1989). Neutron-diffraction study of the structure
of evaporated pure amorphous silicon. Phys. Rev. B, 40, 8030–8032.
Kugler, S., Pusztai, L., Rosta, L., Chieux, P., and Bellissent, R. (1993a). Structure of
evaporated pure amorphous silicon: neutron-diffraction and reverse Monte Carlo
investigation. Phys. Rev. B, 48, 7685–7688.
Kugler, S., Shimakawa, K., Watanabe, T., Hayashi, K., Laszlo, I., and Bellissent, R.
(1993b). The temperature dependence of amorphous carbon structure. J. Non-Cryst.
Solids, 164–166, 1143–1146.
Kwon, I., Biswas, R., Wang, C.Z., Ho, K.M., and Soukoulis, C.M. (1994). Transferable
tight-binding models for silicon. Phys. Rev. B, 49, 7242–7250.
82 Structure
Laaziri, K., Kycia, S., Roorola, S. et al. (1999). High resolution radial distribution
function of pure amorphous silicon. Phys. Rev. Lett., 82, 3460–3463.
Langer, J.S. (2006). Excitation chains at the glass transition. Phys. Rev. Lett., 97, 11504,
1–4.
Laszlo, I. (1998). Formation of cage-like C60 clusters in molecular dynamics simulations.
Europhys. Lett., 44, 741–746.
Laszlo, I. (1999). Molecular dynamics study of the C60 molecule. J. Mol. Struct.:
THEOCHEM, 463, 181–184.
Lee, T.H. and Elliott, S.R. (2011). Ab initio computer simulation of the early stage of the
crystallization: application to Ge2 Sb2 Te5 phase-change materials. Phys. Rev. Lett.,
107, 145702, 1–5.
Lenovsky, T., Kress, J.D., Kwon, I. et al. (1997). Highly optimized tight-binding model
of silicon. Phys. Rev. B, 55, 1528–1544.
Ley, L., Reichardt, J., and Johnson, R.L. (1982). Static charge fluctuations in amorphous
silicon. Phys. Rev. Lett., 49, 1664–1667.
Li, F. and Lannin, J.S. (1990). Radial distribution function of amorphous carbon. Phys.
Rev. Lett., 65, 1905–1908.
Liu, X.Y. and Sawant, P.D. (2002). Mechanism of the formation of self-organized
microstructures in soft functional materials. Adv. Mater., 14, 421–426.
Lomba, E., Molina, D., and Alvarez, M. (2000). Hubbard corrections in a tight-binding
Hamiltonian for Se: effects on the band structure, local order, and dynamics. Phys.
Rev. B, 61, 9314–9321.
Lukács, R. and Kugler, S. (2010). A simple model for the estimation of charge accumu-
lation in amorphous selenium. Chem. Phys. Lett., 494, 287–288.
McGreevy, R.L. and Pusztai, L. (1988). Reverse Monte Carlo simulation: a new
technique for the determination of disordered structures. Molec. Sim., 1, 359–
367.
Marks, N.A., McKenzie, D.R., Pailthorpe B.A., Bernasconi, M., and Parrinello, M.
(1996). Microscopic structure of tetrahedral amorphous carbon. Phys. Rev. Lett.,
76, 768–771.
Metropolis, N., Rosenbluth, A.W., Rosenbluth, M.N., Teller, A.H., and Teller, E. (1953).
Equation of state calculations by fast computing machines. J. Chem. Phys., 21,
1087.
Molina, D., Lomba, E., and Kahl, G. (1999). Tight-binding model of selenium disordered
phases. Phys. Rev. B, 60, 6372–6382.
Moss, S.C. and Graczyk, J.F. (1969). Evidence of voids within the as-deposited structure
of glassy silicon. Phys. Rev. Lett., 23, 1167–1171.
Nose, S. (1984). A unified formulation of the constant temperature molecular-dynamics
methods. J. Chem. Phys., 81, 511–519.
Novikov, V.N. and Sokolov, A.P. (2013). Role of quantum effects in the glass transition.
Phys. Rev. Lett., 110, 065701, 1–5.
Oligschleger, C., Jones, R.O., Reimann, S.M., and Schober, H.R. (1996). Model
interatomic potential for simulations in selenium. Phys. Rev. B, 35, 6165–
6173.
Overney, G., Zhong, W., and Tománek, D. (1993). Structural rigidity and low frequency
vibrational modes of long carbon tubules. Z. Phys. D, 27, 93–96.
Phillips, J.C. (1973). Bonds and Bands in Semiconductors. New York and London:
Academic Press.
Phillips, W.A. (1972). Tunneling states and the low-temperature thermal expansion of
glasses. J. Low Temp. Phys., 7, 351–360.
References 83
Pineda, E.P. and Crespo, D. (1999). Microstructure development in Kolmogorov,

Johnson-Mehl, and Avrami nucleation and growth kinetics. Phys. Rev. B, 60, 3104–
3112.
Pizzagalli, L., Godet, J., Guénolé, J. et al. (2013). A new parametrization of the
Stillinger–Weber potential for an improved description of defects and plasticity
of silicon. J. Phys.: Condens. Matter, 25, 055801, 1–12.
Powell, D., Migliorato, M.A., and Cullis, A.G. (2007). Optimized Tersoff poten-
tial parameters for tetrahedrally bonded III-V semiconductors. Phys. Rev. B, 75,
115202, 1–9.
Pusztai, L. and Kugler S. (2005). Comparison of the structures of evaporated and ion-
implanted amorphous silicon samples. J. Phys.: Condens. Matter, 17, 2617–2624.
Raoux, S. and Wuttig, M. (2009). Preface. In Phase Change Materials: Science and
Applications, eds. S. Raoux and M. Wuttig. New York: Springer.
Rücker, H. and Methfessel, M. (1995). Anharmonic Keating model far group-IV semi-
conductors with application to the lattice dynamics in alloys of Si, Ge, and C. Phys.
Rev. B, 52, 11059–11072.
Sears, V.F. (1992). Neutron scattering lengths and cross-sections. Neutron News, 3(3),
26–37.
Senkader, S. and Wright, C.D. (2004). Models for phase-change of Ge2 Sb2 Te5 in optical
and electrical memory devices. J. Appl. Phys., 95, 504–511.
Shimakawa, K. (2012). Dynamics of crystallization with fractal geometry: extended
KJMA approach in glasses. Phys. Status Solidi B, 249, 2024–2027.
Shimakawa, K., Hayashi, K., Kameyama, T., Watanabe, T., and Morigaki K. (1991).
Anomalous electrical conduction in graphite-vaporized films. Phil. Mag. Lett., 64,
375–378.
Shportko, K., Kremers, S., Woda, M., Lencer, D., Robertson, J., and Wuttig, M. (2008).
Resonant bonding in crystalline phase-change materials. Nature Mat., 7, 653–
658.
Sim, E., Beckers, J., Leeuw, S., Thorpe, M., and Ratner, M.A. (2005). Parameterization
of an anharmonic Kirkwood-Keating potential for Alx Ga(1−x) As alloy. J. Chem.
Phys., 122, 174702, 1–6.
Stich, I., Car, R., and Parrinello, M. (1991). Amorphous silicon studied by ab ini-
tio molecular dynamics: preparation, structure, and properties. Phys. Rev. B, 44,
11092–11104.
Stillinger, F.H. and Weber, T.A. (1985). Computer simulation of local order in condensed
phases of silicon. Phys Rev. B, 31, 5262–5271.
Stone, A.J. and Wales, D.J. (1986). Theoretical studies of icosahedral C60 and some
related species. Chem. Phys. Lett., 128, 501–503.
Swope, W.C., Andersen, H.C., Berens, P.H., and Wilson, K.R. (1982). A computer
simulation method for the calculation of equilibrium constants for the formation of
physical clusters of molecules: application to small water clusters. J. Chem. Phys.,
76, 637–649.
Tabuchi, N., Kawahara, T., Arai, T., Morimoto, J., and Matsumura, H. (2004). Devel-
opment of structural analysis method based on reverse Monte Carlo simulation
and its application to catalytic chemical vapor deposition hydrogenated amorphous
silicon. Jpn. J. Appl. Phys., 43, 6873–6879.
Tanaka, K. (1985). Glass transition of covalent glasses. Solid State Commun., 54, 867–
869.
84 Structure
Tang, M.S., Wang, C.Z., Chan, C.T., and Ho, K.M. (1996). Environment-dependent
tight-binding potential model. Phys. Rev. B, 53, 979–982.
Tegze, M. and Faigel, Gy. (1996). X-ray holography with atomic resolution. Nature,
380, 49–51.
Terao, M., Morikawa, T., and Ohta, T. (2009). Electrical phase-change memory: funda-
mentals and state of the art. Jpn. J. Appl. Phys., 48, 080001, 1–14.
Tersoff, J. (1986). New empirical model for the structural properties of silicon. Phys.
Rev. Lett., 56, 632–635.
Tersoff, J. (1988a). New empirical approach for the structure and energy of covalent
systems. Phys. Rev. B, 37, 6991–7000.
Tersoff, J. (1988b). Empirical interatomic potential for silicon with improved elastic
properties. Phys. Rev. B, 38, 9902–9905.
Tersoff, J. (1988c). Empirical interatomic potential for carbon, with application to
amorphous carbon. Phys. Rev. Lett., 61, 2879–2882.
Tersoff, J. (1989). Modeling solid-state chemistry: interatomic potentials for multi-
component systems. Phys. Rev. B, 39, 5566–5568. (Erratum: (1990) Phys. Rev. B,
41, 3248.)
Tichý, L. and Tichá, H. (1995). Covalent bond approach to the glass-transition tempera-
ture of chalcogenide glasses. J. Non-Cryst. Solids, 189, 141–146.
Tonchev, D. and Kasap, S.O. (2006). Thermal properties and thermal analysis: funda-
mentals, experimental techniques and applications. In The Springer Handbook of
Electronic and Photovoltaic Materials, eds. S.O. Kasap and P. Capper. Heidelberg:
Springer, chap. 19.
Tóth, G. and Náray-Szabó, G. (1994). Novel semiempirical method for quantum Monte
Carlo simulation: application to amorphous silicon. J. Chem. Phys., 100, 3742–
3746.
Treacy, M.M.J. and Borisenko, K.B. (2012). The local structure of amorphous silicon.
Science, 335, 950–953.
Valladares, A.A., Alvarez, F., Liu, Z., Sticht, J., and Harris, J. (2001). Ab initio studies
of the atomic and electronic structure of pure and hydrogenated a-Si. Eur. Phys. J.
B, 22, 443–453.
Verlet, L. (1967). Computer “experiments” on classical fluids. I. Thermodynamical
properties of Lennard−Jones molecules. Phys. Rev., 159, 98–103.
Vink, R.L.C., Barkema, G.T., van der Weg, W.F., and Mousseau, N. (2001). Fitting the
Stillinger–Weber potential to amorphous silicon. J. Non-Cryst. Solids, 282, 248–
255.
Walters, J.K., Gilkes, K.W.R., Wicks, J.D., and Newport, R.J. (1998). Progress in mod-
elling the chemical bonding in tetrahedral amorphous carbon. Phys. Rev. B, 58,
8267–8276.
Wang, C.Z., Pan, B.C., and Ho, K.M. (1999). An environment-dependent tight-binding
potential for Si. J. Phys.: Condens. Matter, 11, 2043–2049.
Wilson, M. and Salmon, P.S. (2009). Network topology and the fragility of tetrahedral
glass-forming liquids. Phys. Rev. Lett., 103, 157801, 1–4.
Winer, K. and Cardona, M. (1986). Theory of infrared absorption in amorphous silicon.
Wooten, F. and Weaire, D. (1984). Generation of random network models with periodic
boundary conditions. J. Non-Cryst. Solids, 64, 325–334.
Wooten, F., Winer, K., and Weaire, D. (1985). Computer generation of structural models
of amorphous Si and Ge. Phys. Rev. Lett., 54, 1392–1395.
References 85
Xu, C.H., Wang, C.Z., Chan, C.T., and Ho, K.M. (1992). A transferable tight-binding
potential for carbon. J. Phys.: Condens. Matter, 4, 6047–6057.
Yamada, N. (2009). Development of materials for third generation optical strange
media. In Phase Change Materials: Science and Applications, eds. S. Raoux and
M. Wuttig. New York: Springer, chap. 10.
Yang, R. and Singh, J. (1998). Study of the stability of hydrogenated amorphous silicon
using tight-binding molecular dynamics. J. Non-Cryst. Solids, 240, 29–34.
Yelon, A., Movaghar, B., and R.S. Crandall, R.S. (2006). Multi-excitation entropy: its
role in thermodynamics and kinetics. Rep. Prog. Phys., 69, 1145–1194.
4
Electronic structure
The influence of disorder causes essential differences in the electronic density

of states (DOS) of amorphous and crystalline semiconductors. A description of
the electronic properties starts with an understanding of the covalent bonds in
amorphous semiconductors. Defects in these materials also modify their optical
and electronic properties. Different defects and their influences are overviewed
in the second part of this chapter.
4.1 Chemical bonds

In saturated covalently bonded semiconductors, the 8 − N rule states that the
number of sigma bonds or the local coordination number equals eight minus
the relevant column (group) number of the periodic table. This rule is realized
within a valence bond framework by assuming that single saturated covalent
bonds are formed with next neighbors, establishing the closed and stable octet
shell of electrons. The only exception to this rule is graphitic carbon, which
is a good conductor. For N = 4, germanium, silicon, and diamond-like carbon
atoms take the tetrahedral diamond local structure with four sigma bonds, and
for N = 5 the pnictides arsenic, antimony, and bismuth take a puckered layer
structure with three sigma bonds. Furthermore, for N = 6, the chalcogenides
selenium and tellurium take helical chain structures having two sigma bonds.
Finally, group VII elements form diatomic molecules, i.e. atoms having only
one electron for the sigma bond.
Group IV elements
The framework of the linear combination of atomic orbitals (LCAO) can
be used to describe the electronic structure and bonds in these materials.
86
Two-center-localized molecular orbitals are constructed using the LCAO form

of the sigma bonds. We illustrate this methodology for carbon. An isolated car-
bon atom has the electron configuration 1s2 2s2 2p2 . Therefore, one could naively
think the deeper core states (for example the 1s and 2s states) are not involved in
bonding, and that twofold chemical bonds are formed using the 2p states only.
However, in the majority of cases the situation is more complex, and the 2s states
also contribute to bond formation; this mechanism is known as hybridization.
In covalently bonded semiconductors the valence electrons are localized and
form the chemical bonds. Therefore the valence electron wave functions are simi-
lar to the bonding orbitals found in molecules. For a tetrahedrally coordinated
carbon, for example methane (CH4 ), the carbon atom should have four orbitals
with the correct symmetry to bond to the four hydrogen atoms. The solutions,
linear combinations of the 2s and 2p wave functions, are known as hybridized
orbitals. Therefore, in the case of tetrahedral carbon, four orbitals are required.
The 2s orbital is mixed with the three 2p orbitals to form four sp3 hybrids:
1 1 1 1
sp3 = s + px + py + pz ,
2 2 2 2
1 1 1 1
sp3 = s − px − py + pz ,
2 2 2 2 (4.1)
1 1 1 1
sp = s −
3
px + py − pz ,
2 2 2 2
1 1 1 1
sp3 = s + px − py − pz .
2 2 2 2
These four hybrids are orthonormalized states. In CH4 , four sp3 hybridized
orbitals are overlapped by the hydrogen 1s orbitals, yielding four sigma bonds
(i.e. four single covalent bonds). The four equivalent covalent bonds are of
the same length and strength. When the atoms from group IV of the periodic
table combine to combine to form a condensed phase, the interaction splits the
valence states into the electron-bonding and higher-energy antibonding levels.
Between these two levels a gap can be found where no electron states exist.
In a crystal, the sp3 orbitals overlap with similar orbitals of the four adja-
cent atoms, and a well-defined sigma-bond network is formed, as shown in
Figure 4.1. All the bonds are occupied by two electrons. This theory also fits the
fourfold-coordinated group IV elements (such as diamond-like carbon, silicon,
and germanium) in non-crystalline arrangements. However, the ½ prefactors in
eqn. (4.1) have to be modified in order to account for deviations from the ideal
tetrahedral arrangement of four nearest-neighbor atoms.
88 Electronic structure
Figure 4.1. The sp3 orbitals overlap with the similar orbitals of four adjacent
atoms.
Figure 4.2. An sp3 orbital remains as an unpaired and unsatisfied valence state.
This dangling bond (DB) forms an electronic localized state.
It can be concluded from diffraction measurements that the upper limit for
the first-neighbor distance in a-Si is about 0.27 nm. A special case occurs when
only three neighboring atoms are situated within this distance. Three sp3 orbitals
overlap with the orbitals of neighbor atoms and form normal sigma bonds.
The three bonds are usually referred to as “backbonds” and the three atoms
as “backbond atoms.” One sp3 orbital remains as an unpaired and unsatisfied
valence state, as shown in Figure 4.2. This is called the dangling bond (DB),
and it is localized. This single electron has an uncompensated spin, which can
be detected by electron spin resonance (ESR) measurements. Dangling bonds
dominate electronic and optical properties and hence affect the characteristics of
the device. Amorphous semiconductors with lesser defects (DBs, etc.) are often
referred to as device-quality materials.
Exception: graphitic carbon

Let us consider a simple example. Ethylene (C2 H4 ) has a double bond between
the carbon atoms. For this molecule, the wave functions of the carbon atoms will
be sp2 hybrids, because one pi bond is required for the double bond between the
two carbon atoms. Only three sigma bonds can be formed per carbon atom. In
sp2 hybridization, the 2s orbital is mixed with only two (px and py ) of the three
available 2p orbitals. These three hybrids are also orthonormalized states. The
mathematical terms are as follows:
1/2 1/2
1 2
sp 2 = s+ px ,
3 3
1/2 1/2 1/2
1 1 1
sp 2 = s− px + py , (4.2)
3 6 2
1/2 1/2 1/2
1 1 1
sp 2 = s− px − py .
3 6 2
In ethylene the two carbon atoms form a sigma bond by overlapping two sp2
orbitals, and each carbon atom forms two other covalent bonds with hydrogen
via an s–sp2 overlap, all with 120° angles. The pi bond between the carbon atoms
perpendicular to the molecular plane is formed by the 2pz – 2pz overlap. The
hydrogen–carbon bonds are all of equal strength and length (resonant bonds),
which agrees with experimental data. In a honeycomb graphite crystal, such
a local arrangement can be observed as planar. The bond length is shorter by
0.01 nm, and the bond is stronger than in a diamond crystal because of the
additional pi bond. This local arrangement with sp2 hybridization forms the
graphite-like a-C.
Group VI elements
Selenium atoms contain six valence electrons (4s2 4p4 ), and the previously
described sp3 hybridization is no longer valid. From s, px , py , and pz only
four independent orbitals can be constructed, i.e. two out of four must be
doubly occupied. A simple hydrogen selenide molecule has the atomic struc-
ture H–Se–H, with a bond angle of 91°. Note that in hydrogen telluride the
bond angle is equal to 90°. As a first approximation, we can explain the bond-
ing process in the following way. Two near-perpendicular sigma bonds can
be constructed by putting one electron in the px and another in the py atomic
orbitals, which are overlapped by the 1s orbital of hydrogen atoms. The other
two unshared electrons occupy the pz atomic orbital, forming a non-bonding
Figure 4.3. The px and py orbitals overlap with similar orbitals of adjacent
atoms and form sigma bonds. Two unshared electrons occupy the pz atomic
orbital, forming a non-bonding lone-pair (LP) state.
lone-pair (LP) state. The other two electrons remain untouched in the s atomic
orbital.
Pure non-crystalline selenium is the model material for chalcogenide glasses.
In the condensed phase, selenium atoms form a chain-like structure. It can be
concluded that only two electrons in the px and py orbitals form sigma bonds that
overlap with similar orbitals of adjacent atoms (see Figure 4.3). Bond angles
are slightly larger than 90°. This means that the s atomic orbital is not exactly
untouched; a small contribution from the s orbital is required to form near-
perpendicular sigma bonds and a LP. The energy of these non-bonding LP states
lies between the electron-bonding and antibonding energy levels.
Selenium chains have unpaired and unsatisfied valence states (DBs) at their
ends. Unlike a-Si:H, this localized state is not uncompensated, i.e. these orbitals
are doubly occupied by electrons, so we have C− 1 , where the minus superscript
refers to the atomic net charge and the subscript denotes the coordination number.
4.2 Electronic density of states

In crystalline materials the presence of long-range order and perfect translational
symmetry greatly simplifies the mathematical treatment of the electrons. The
electron states can be described by Bloch wave functions extending throughout
the crystal (delocalization) as follows:
ϕ (k, r) = u (k, r) exp (ikr) , (4.3)
where u(k, r) describes the periodicity of the crystal lattice, i.e. u(k, r) =
u(k, r + R), where R is the lattice translation vector, and the exponential term
represents a plane wave. The wave vectors k of the electrons are related to the
translational symmetry of the lattice. It follows that a reciprocal lattice related
to the unit cell parameters is established in k-space. This solid state physics tool
4.2 Electronic density of states 91
is inadequate for non-crystalline materials. The lack of long-range order does

not allow the wave vector k to be defined, and therefore the classification of
electronic band structure E(k) also cannot be applied. Furthermore, this prevents
the traditional use of the definition of effective mass (Singh, 2002).
Instead of E(k), the number of electronic states at energy E per unit energy
N(E) is used as a well-defined expression for amorphous materials. The occu-
pancy is derived by a Fermi distribution function at a given temperature T, which
is nearly a step function at room temperature. As the temperature rises above
absolute zero, there is more energy to spend on lattice vibration and on lifting
some electrons into an energy state of the conduction band (CB). Electrons
excited to the conduction band leave behind holes in the valence band (VB).
Both the CB electrons and the VB holes contribute to electrical conductivity.
The most important property of semiconductors is the existence of a bandgap
which separates the valence and conduction bands. The presence of perfect peri-
odicity in crystalline semiconductors helps to derive this bandgap. However,
the definition of the bandgap is not as straightforward for amorphous semi-
conductors. The first pioneering attempt to define the bandgap was made by
Weaire and Thorpe (1971) for tetrahedrally bonded amorphous semiconductors.
In both crystalline and amorphous semiconductors, bands are desribed by a local
arrangement. The chemical bonding can be described using a tight-binding (TB)
model with the following Hamiltonian:

H = V1 |θik θil | + V2 |θik θj k |, (4.4)
where θ ik denotes the kth sp3 hybrid orbitals associated with the ith atom. The
first term is a sum over the interaction in which the kth and lth wave functions
belong to the same atom, and the second term is a sum associated with the wave
functions with the same bonds. These are the strongest interactions; this is why
the Weaire–Thorpe model is a good starting approximation: V1 V2 = 13 for
silicon. It can be shown that there is no gap for the case of V1 V2 = 12 only
(i.e. the valence and conduction bands touch each other).
Another important law for amorphous semiconductors is Anderson’s theory
of localization (Anderson, 1958). A simple description of this law is that the
increase in the disordered potential causes electron localization, i.e. the wave
function becomes confined into a small volume. This characteristic difference
between the electronic structure of crystalline and non-crystalline solids plays
an essential role in the transport and optical properties, etc.
Figure 4.4. Electronic DOS of tetrahedrally bonded pure amorphous semi-

conductors.
The rapid growth in the number of computer facilities and faster computa-
tional speeds mean that there are many different numerical methods that can
be used to calculate the electronic structure of condensed matter. The transla-
tional symmetry of crystalline materials is beneficial because of the application
of periodic boundary conditions, therefore only a few atoms are needed in the
numerical calculations to obtain an accurate result. For amorphous materials,
a periodic boundary condition using so-called “super cells,” is also applicable,
although there must be a large number of randomly distributed atoms to produce
reasonable results. Here, the term “large number” means at least 100 atoms.
Another possible method is cluster calculation, in which a structural model,
again containing a large number of atoms, is required. At the surface of these
clusters, the DBs should be saturated, usually with hydrogen termination. A
drawback of hydrogen termination is that a large charge transfer occurs in Si–H
bonds. To resolve this problem, a longer Si–H bond length than the average
normal Si–H atomic distance must be substituted into the model structure.
Three basic methods have been developed and used for the calculation of elec-
tronic properties of amorphous semiconductors: tight-binding models, Hartree–
Fock methods, and density-functional theories. For the electronic DOS calcu-
lation, the Hartree–Fock approximation usually overestimates the energy gap,
while the density-functional methods underestimates it.
The main conclusions of a large number of computer calculations and theo-
retical work are displayed in Figure 4.4. In the DOS of a fourfold-coordinated
pure amorphous semiconductor, delocalized and localized regions (tail states)
can be found separately by the mobility edge. Such localized band tails do not
exist in crystalline counterparts. Following Fermi statistics, the lower-energy
4.3 Defects 93
Figure 4.5. Non-bonding LP states are in the middle of the gap in the electronic
DOS.
region is almost completely filled by electrons as a function of temperature. The

pure chalcogenide glasses, however, have a different electronic DOS. As shown
in Figure 4.5, an extra peak appears in the gap; the so-called non-bonding LP
states (bands) are in the middle of the gap between the bonding and antibonding
states (bands). The Fermi energy, EF = μ(T = 0) lies between the top of the LP
band and the conduction band. This characteristic feature in the DOS of chalco-
genides plays a primary role in photoinduced effects, which will be discussed in
Chapter 5.
4.3 Defects
Classification of defects
One possible definition of a defect is something that prohibits or prevents per-
fection. Structures that are devoid of defects are considered ideal. Inclusion of
a structural defect thus refers to a configuration in which an atom, or group
of atoms, does not satisfy the structural rules belonging to the ideal refer-
ence state of the material. Real materials contain structural defects that can
dominate their physical and chemical behaviors. In covalently bonded semi-
conductors, defects usually create states inside the energy gap which control
the electronic and optical properties. Some structural defects can capture elec-
trons or holes. It is not easy to identify structural defects in diffraction mea-
surements. The key defects that occur in crystalline materials comprise the
following:
r point defects due to impurity atoms in either substitutional or interstitial sites;
r complexities associated with special correlations between different point
defects such as impurity–vacancy pairs or donor–acceptor pairs;
r one-dimensional (line) defects, including translational displacements of atoms
(dislocations) and rotational displacements (disclinations);
Figure 4.6. Defect model in a-Si. D0 (D+ ) and D− indicate neutral (positively
charged) and negatively charged dangling bonds. D+ and D− have no spin.
r two-dimensional (planar) defects, including low-angle grain boundaries;

r native point defects, such as vacancies and interstitial atoms;
r defects associated with grain boundaries in polycrystalline materials;
r distortions caused by surfaces or interfaces; and
r voids, i.e. several connecting atoms are missing from the network.
What is the meaning of defects in amorphous semiconductors? At first, this

may seem to be a silly question because amorphous means disorder and so there
is obviously a lack of perfection; however, this statement is incorrect because the
disorder in amorphous semiconductors has order in some sense. As an example,
consider a-Si, in which all the atoms are tetrahedrally coordinated, similar to
the crystalline counterpart. However, there is a lack of long-range order. In this
case, “defects” can be regarded as the deviation from the ideal picture. In this
book, we define a “defect” as a dangling bond, as discussed in Section 4.2. There
are two different classes of defects: natural and photo-created defects. In this
chapter, we discuss natural defects.
In tetrahedrally bonded semiconductors broken bonds (unsaturated bonds) are
observed, as shown in Figure 4.2, that break the local topological order. These
dangling bonds (DBs) release the internal tension inside the continuous random
network. A simple DB normally contains one electron and is electrically neutral,
but under certain circumstances (chemical or electronic doping) the electronic
occupancy changes from D0 (neutral) to D+ (positive) and D− (negative), where
D denotes the DBs, as shown in Figure 4.6. As D− has an extra electron, the
energy level is raised by an amount Uc (the Coulombic repulsive energy at
the DB). Note, however, that D+ should have the same energy level as D0 ;
D− does not have spin. To reduce the number of DBs, the amorphous silicon
(and germanium) are passivated by hydrogen, which reduces the DB density
by several orders of magnitude. These intrinsic defects form electronic states
inside the gap, and this under-coordination defect has a significant ESR signal
4.3 Defects 95
with g = 2.0055 in a-Si, which is the most important experimental evidence

of DBs. The spin density is estimated to be around 1019 cm−3 in a-Si, but is
much less (around 1015 cm−3 ) in a-Si:H, which results from the preparation
techniques and the hydrogen contamination. Such intrinsic defects do not exist
in crystalline Si because removing one atom produces a vacancy with four
DBs. The energy of this arrangement is too high, and bond reconstructions
occur. Pantelides (1986) has suggested other coordination defects, namely over-
coordinated atoms (floating bonds). Whether floating bonds exist or not is still
not clear. If we consider the position of a fourth-nearest-neighbor fourfold-
coordinated atom with respect to that of a threefold-coordinated atom, along
with the direction of the DBs with respect to this atom, we can say that a weak
interaction probably exists between the two atoms. This geometrical arrangement
provides a fivefold-coordinated atom. However, the weak interaction does not
contain two localized electrons; it is not a normal sigma bond, but it can be
considered to be a bond in some sense.
Theoretically, DBs can also consititute a possible defect in chalcogenide
glasses. Group VI elements in the amorphous phase have long chains and rings.
At the ends of the chains are onefold-coordinated atoms, and unpaired spins
are expected. Experiments show different results compared to those with a-Si.
Pure a-Se presents no ESR signal in dark conditions, i.e. there is no neutral
DB with an unpaired spin. Originally, Anderson (1975) proposed a negative-U
concept, and then this idea was applied to chalcogenide glasses (see, for exam-
ple, Mott and Davis 1979). Normally, a Coulomb repulsive interaction should
occur between two electrons. In the Anderson theory, the phonon–electron inter-
action effectively gives rise to an attraction. The Hamiltonian describing the
phenomenon has three parts: electronic, phonon, and electron–phonon. If the
coupling constant in the electron–phonon part is large enough, the effective
correlation between electrons is negative (a negative-U center) and, instead of
C0 , two other charged states, C+ and C− , can be formed, in which there are no
ESR-active spins:
2C0 → C+ + C− . (4.5)
This means that the charged states C+ and C− are more stable than the neutral
state C0 . Kastner, Adler, and Fritzsche (1976) showed that these are threefold
(C+ −
3 ) and onefold (C1 ) coordinated in terms of the chemical bonds, and are
called the valence-alternation pairs (VAPs). Furthermore, another defect, the
threefold-coordinated Se atom (junction), can be formed in the network.
Figure 4.7. Charge distribution of a 162 Se atom cluster calculated using the
DFT GGA method. Larger charge accumulations can be observed in both direc-
tions (see insets). (From Lukács, Hegedus, and Kugler (2009). J. Mater. Sci.:
Mater. Electron., 20, S33–S37. With kind permission from Springer Science
and Business Media.)
A controversial consequence of the Anderson theory has been reported by

Lukács, Hegedus, and Kugler (2009). The charge distribution was calculated in
a-Se using two different calculation methods: a density-functional method and
a tight-binding (TB) model. Figure 4.7 displays the charge distributions using
the DFT GGA (generalized gradient approximations) calculation on 162 atom
clusters. Essentially, the absolute values of the charge accumulations on the
twofold-coordinated atoms are less than 0.1e. Larger charge accumulations in
both positive and negative directions were observed (see the insets in Figure 4.7).
The larger positive charge accumulation belongs to the threefold-coordinated
atoms, and the negatively charged atoms are onefold coordinated. A similar
effect was observed using the TB model. The charges in the TB calculations
are somewhat larger than those calculated using DFT (see Figures 4.7 and
4.8). These calculations do not contain any phonon–electron interactions, and
the results suggest that the threefold-coordinated atoms (which look similar to
VAPs) lose an electron, and that these electrons are transferred to the end of
chains. However, these are not negative-U defects caused by phonon–electron
interactions.
In amorphous semiconductor networks, coordinated impurity atoms having
no unsatisfied bonds are usually electronically neutral. The most important con-
clusion of Mott’s famous (8 − N) rule (1969) is that amorphous semiconductors
4.3 Defects 97
Figure 4.8. Charge distribution calculated using a TB model on the same Se

cluster as in Figure 4.7. (From Lukács, Hegedus, and Kugler (2009). J. Mater.
Sci.: Mater. Electron., 20, S33–S37. With kind permission from Springer Sci-
ence and Business Media.)
Figure 4.9. p doping: schematic representation of spatial distribution of elec-

tron excess around a phosphorus atom in a silicon network.
cannot be doped! Some years later, Spear and LeComber (1975) showed that
plasma-glow-discharge deposition of silicon, using mixtures of silane and
either phosphine or diborane, enabled electrically active pentavalent (P) (see
Figure 4.9) and trivalent (B) (see Figure 4.10) impurities to be incorporated
into the films, making them n type and p type, respectively. These impurities
introduce unsatisfied bonds into the network that are electronically active. An
interesting historical fact is that the first reported, but not successfully exploited,
experiment on doping had been carried out earlier by Chittick, Alexander, and
Figure 4.10. n doping: schematic representation of spatial distribution of elec-

tron deficiency around a boron atom in a silicon network.
Sterling (1969) at the Standard Telephone Laboratories in Harlow, UK. They

stated in the abstract of their paper that “the effects of heat-treatment, ageing,
and doping on the properties of amorphous silicon are reported.”
Estimation of the doping energy levels for a crystalline arrangement is quite
an easy task. A hydrogen molecule type model is a good approach to take (see,
for example, Sólyom, 2009). The core charge of a phosphorus atom is higher
than that of the connecting atom, and the extra electron is weakly bound to the
phosphorus atom by the Coulomb field. For the calculation it must be considered
that a free electron moves in a dielectric media instead of a vacuum and that the
electron inside a periodic potential has an effective mass. Mostly, in heavy-doped
semiconductors two doping atoms can be located nearby, i.e. in first-, second-,
third-, etc. neighbor positions, and the interaction between them is important. A
hydrogen molecule type model should be used in the energy-level calculation
for heavy-doped semiconductors.
In non-crystalline semiconductors the lack of a well-defined effective mass
means that this hydrogen atom/molecule type model is useless for such materials.
The position of the energy levels belonging to dopant pairs was investigated in
fourfold-coordinated a-C (ta-C) and in a-Si by means of a Hartree–Fock ab
initio method (Kádas and Kugler, 1997a,b; Kádas, Ferenczy, and Kugler, 1998).
Models contained 45 to 583 carbon or silicon atoms, and boron, phosphorus,
and nitrogen impurities were incorporated into the amorphous networks. It has
been observed that the positions of the midgap states are primarily determined
by the separation of the impurity atoms. The electronic density of states in ta-C
doped with two nitrogen atoms is shown in Figure 4.11. It should be noted that
the Hartree–Fock method always overestimates the gap energy. Peaks 1 and 2
4.3 Defects 99
Density of states
1 2 3 4
–0.5 –0.3 –0.1 0.1 0.3 0.5 0.7

Energy
Figure 4.11. Electronic DOS obtained from Hartree–Fock ab initio cluster cal-
culations for different ta-C models doped by two nitrogen atoms. (From Kádas,
Ferenczy, and Kugler (1998). J. Non-Cryst. Solids, 227–230, 367. Copyright
2013 with permission from Elsevier.)
display the highest occupied molecular energies when two nitrogen atoms are
at the first- and fourth-neighbor positions, respectively. Peak 3 belongs to a ta-C
cluster containing only one nitrogen dopant. This latter configuration occurs
when the other dopant is placed at an infinite distance. These direct calculations
of the energy levels suggest that a significant difference can occur between
energy levels as a function of doping rate.
The diamond crystal structure consists of sixfold atomic rings, whereas the
corresponding amorphous diamond-like structure is made of five-, six-, and
sevenfold rings, with the threefold and fourfold rings being atypical. Never-
theless, triangle and squares break the local order, i.e. they can be considered
as one type of defect. Such defects were found in a-Si (Kugler et al., 2003).
As a consequence, two characteristic peaks can be found inside the tail of the
band structure (Kugler, 2012). The first larger peak corresponds to the square
atomic arrangements, whereas the peak at the higher energy is due to triangular
arrangements (Figure 4.12). In earlier investigations of electron transport, hop-
ping conductivity, optical properties, etc., the tail was usually considered to be
an exponential or Gaussian decaying function. Triangles and/or squares have
never been considered in any band structure calculations, although they play an
important role in several phenomena.
Normally a two-component (A and B) covalently bonded amorphous net-
work contains A–A, A–B, and B–B bonds. In special stoichiometric compos-
itions (GeSe2 or As2 Se3 , etc.) only A–B bonds are energetically favorable.
In these chalcogenide materials, a few A–A and B–B bonds can be found
Density of states
Squares
Triangles
Energy
Figure 4.12. Electronic density of states of amorphous semiconductors con-

taining two significant peaks belonging to the triangle and square defects.
experimentally. This class of defects is known as “wrong bonds.” Various authors

have worked on this problem, among them Petri, Salmon, and Fischer (2000),
who published an excellent paper on the experimental evidence for wrong bonds.
Neutron-diffraction measurements were made in the D4B instrument at ILL
Laue-Langevin, Grenoble, with GeSe2 as the target material. A significant num-
ber of Ge–Ge or Se–Se homopolar bonds were identified, which demonstrates
the existence of wrong bonds.
In amorphous networks, voids where several connecting atoms are missing
from the network also occur as defects. The properties and number of these
defects strongly depend on the preparation technique. In obliquely deposited
amorphous GeSe3 films, the density decreased when the angle of the evaporate
beam increased, as reported by Rayment and Elliott (1983). Columnar structures
of these materials were also observed in obliquely deposited a-GeS2 . A decrease
in both the refractive index and the microhardness versus angle of incidence has
been reported. These results provide evidence for increasing free volume with
increasing obliqueness (Starbova et al., 1999). Atomic-scale computer simula-
tions for structures of flatly and obliquely deposited amorphous selenium films
have been carried out to understand how the properties of chalcogenide glasses
are influenced by their preparation method (Lukács, Hegedus, and Kugler, 2008).
It was concluded from theoretical structure work that the obliquely deposited
a-Se thin films contain more coordination defects, lower densities, and larger
voids than the flatly deposited ones inside the networks.
Finally, an interesting topological defect is discussed for a-C, which has
different percentages of fourfold- and threefold-coordinated atoms. The pi–
electron network is determined by the threefold-coordinated graphite-like atoms.
These atoms form three localized sigma bonds plus a delocalized pi bond. Graph
4.3 Defects 101
theory can be used to estimate the number of topologically derived defects in

graphite-like a-C (graphene) in relation to the electronic DOS at the Fermi
level (Kugler and Laszlo, 1989). The number of topologically derived energy
levels is determined by the global topological properties of the graphite-like a-C
structures (Laszlo, 1993, 2000).
Defect spectroscopy
Defects produce localized states in the energy gap, and hence they can be
detected by different spectroscopic techniques. Since the defects dominate the
electronic and optical properties, it is important to estimate the number of
defects and their energy levels experimentally. Optical, electrical, and magnetic
measurements are the principal techniques used in defect spectroscopy, which
will be briefly reviewed in this section. The details of defect spectroscopy in
a-Si:H and amorphous chalcogenides (a-Chs) are described elsewhere (Singh
and Shimakawa, 2003).
Optical measurements
The optical absorption (the so-called “midgap absorption”) related to the defect
is usually weak (α 10 cm−1 ) in a-Si:H and a-Chs, and hence several sensitive
techniques are adopted, as described in the following.
(i) Photoinduced absorption (PA) monitors variations in the intensity of probe

light absorbed by trapped carriers in defects, after the carriers are created in
the extended states by illumination. Owing to the increase in absorption, the
transmittance of the probe light changes. The spin-dependent PA (accom-
panying electron spin resonance) is a sensitive technique (Hirabayashi and
Morigaki, 1983).
(ii) Photothermal deflection spectroscopy (PDS) measures the heat absorbed in
the sample and is used to measure the lower optical coefficient of thin films
(Jackson and Amer, 1982). A sample is immersed in liquid CCl4 . As non-
radiative recombination is a dominant channel near room temperature, heat
is generated by phonon emission. This heat generation, with an intensity-
modulation beam of pumping light, induces a change in the refractive index
of CCl4 , which is detected by the deflection of a laser beam passing through
the sample. A He–Ne laser is usually installed to emit the probe light, and
the periodic deflection is measured with a position sensor. If the wavelength
of the pump beam is varied, the deflection of the probe beam becomes
a measure of the optical absorption spectrum of the materials. Some cal-
culations, however, are required to obtain the absolute optical absorption
coefficient. The PDS sensitivity is very high; for example, for a 1000 nm
(1 µm) thickness, values of α as low as 0.1 cm−1 can be obtained.
Electrical measurements
The recombination and emission of carriers at defects are investigated by elec-
tronic transport. As the thermal emission rate is related to the energy level, and
the shift of the Fermi level is related to the number of trapped carriers, these two
parameters are estimated in principle from the electrical transport measurements.
Several techniques are described in the following.
(i) Field-effect measurements were first applied to obtain the localized den-
sity of states in a-Si:H (Madan, LeComber, and Spear, 1976) and in
a-Chs (Marshall and Owen, 1976). A voltage is applied across the dielectric
thin layer deposited on the materials that yield band bending, producing a
space charge near the interface. Note that this is the same principle as for
a thin-film transistor. This produces an excess of carriers in the extended
states (excess conduction), which is due to the shift of the Fermi level
toward the band edge. The DOS near the Fermi level is estimated in this
method. As most of the space charges are 10 nm from the interface, they
are strongly affected by the presence of interface states, which may be a
drawback of this technique.
(ii) Capacitance measurements may be made by considering the depletion layer
of a Schottky barrier under reverse bias. As described in standard text
books, capacitance varies with the bias voltage, reflecting the space-charge
density, and the space charge (i.e. the localized carriers residing at the
defects) can be estimated from the capacitance value. The most familiar
technique that is used to measure deep levels in semiconductors is deep-
level transient spectroscopy (DLTS) in which the transient capacitance of
a Schottky barrier is measured (Lang, 1974). In DLTS, all the traps are
filled by applying a forward bias to the Schottky barrier, and subsequently a
reverse bias. The depletion layer width decreases over time to its steady state
value. Note that the capacitance is inversely proportional to the thickness
of the depletion layer.
4.3 Defects 103
(iii) The space-charge-limited current (SCLC) technique measures the current–

voltage (I–V) characteristics. Deep traps can be filled by current injection,
and hence the Fermi level shifts toward the relevant band states (the con-
duction band for electrons and the valence band for holes). There are sev-
eral ways of analyzing the I–V characteristics (Mackenzie, LeComber, and
Spear, 1982, Shimakawa and Katsuma, 1986). Although SCLC is a simple
technique, good injected electrodes (contacts) are not easy to prepare. If the
number of defects is large, the shift of the Fermi level is small, and hence
the covered energy range of the DOS becomes narrow.
(iv) The constant-photocurrent method (CPM) monitors the illumination inten-
sity G, keeping the photocurrent constant, as a function of photon energy
(Vanecek et al., 1983). In this method, the quasi-Fermi level is kept the
same, independent of photon energy, and hence the recombination time
is unchanged. Because of this condition, it is possible to deduce the opti-
cal absorption coefficient α (around 0.1 cm−1 ). This technique has been
applied to both a-Si:H and a-Chs, and yields information on the weak
optical absorption tails (see, for example, Singh and Shimakawa, 2003).
(v) AC loss occurs during: (a) electric dipole motion, (b) hopping of carriers
between localized states (a-Chs), and (c) inhomogeneous media (a-Si:H).
These will be discussed in detail in Section 4.4. Via the analysis of ac
loss, for example in a-Chs, an acceptable density of charged defects (1017 –
1018 cm−3 ) of reasonable energy at a location in the bandgap can be deduced
(Ganjoo and Shimakawa, 1994).
(vi) The electrophotography (or xerography) method also provides the DOS in
a-Si:H (Imagawa, Akiyama and Shimakawa, 1984) and in a-Se (Abkowitz
and Markovics, 1984). This method is a relatively new candidate with which
to evaluate the DOS; it is able to ignore the surface and interface effects,
which cannot be neglected in the FE and capacitance–voltage (CV) meth-
ods. The principle is the same as that employed in commercial xerographic
equipment.
Magnetic resonance
A quantum state, for example a defect state, occupied by a single electron is
split by a magnetic field; this is known as Zeeman splitting.
(i) Electron spin resonance occurs due to a transition of electrons between

the split energy levels. The transition occurs at microwave frequencies
for a conventional magnetic field (3 T) between two Zeeman levels with
unpaired electrons whose spin states S = 12 (i.e. the paramagnetic state).
ESR signals can be observed by probing the microwave absorption if the
defects have unpaired spins larger than 1015 cm−3 , which of course depends
on the sensitivity of the equipment. A DB in a-Si:H has unpaired spin, and
in the so-called “device-quality” a-Si:H film the spin density is less than
1016 cm−3 (Street, 1991). When an unpaired electron interacts with nuclear
spin, the hyperfine structures are observed in the ESR spectra. Thus the
ESR spectra contain the information of the local configuration of these spin
states.
ESR signals in pure a-Chs are generally not observed under dark condi-
tions, because, unlike for a-Si:H, only the charged defect states (no unpaired
spins) are expected to exist in a-Chs (except for some components). Under
photoillumination at lower temperatures, however, ESR signals appear; we
call this light-induced ESR (LESR) (Bishop, Strom, and Taylor, 1975). Note
that LESR signals appear in a-Si:H at low temperature, which is not the case
for the original ESR signal under dark conditions (Street, 1991). Various
photoinduced effects have been monitored in experiments using LESR (Shi-
makawa, Kolobov, and Elliott, 1995). A pulsed ESR technique, in which
transient signals produced by coherent excitation using microwave pulses
are detected in the time domain, is also useful (Isoya et al., 1993).
(ii) Electron nuclear double resonance (ENDOR) is helpful in obtaining struc-
tural information about defects and their surroundings when hyperfine inter-
actions are weak. Measurements using ENDOR monitor the variation of
the ESR with nuclear magnetic resonance through hyperfine interaction.
The intensity of the ESR signal changes when NMR occurs. The details of
ENDOR measurements on a-Si:H have been described elsewhere (Morigaki,
1999).
4.4 Optical and electronic properties

Fundamental optical absorption
The DOS for both conduction and valence bands is expected to be proportional
to the square root of the energy in 3D materials. It should be noted that the wave
vector k is not a useful quantum number and therefore it cannot be applied in
amorphous solids. Thus the terms used to describe direct and indirect optical
transitions between valence and conduction bands in the study of crystalline

solids are not useful when considering amorphous materials. Furthermore, the
DOS in disordered matter, in general, may be described by taking into account the
fractal concepts (Mandelbrot, 1982) that are known to dominate many physical
properties in amorphous semiconductors (Zallen, 1983).
Based on fractal geometry, i.e. the spherical volume V is proportional to rD ,
where r is the radius and D is the fractal dimension, which is different from the
values defined by the Euclidean space dimension d (= 1, 2, 3), the DOS is given
by
N (E) dE = AE (D−2)/2 dE, (4.6)
where A is a constant (He, 1990). In a normal homogeneous space, D should be

the same as the Euclidean space dimension d. For example, D = d = 3 leads
to N(E) proportional to E12 , as described in any standard text book. When a
material contains a large amount of voids, D can be significantly lower than the
dimensions of the Euclidean space.
Let us discuss the fundamental optical absorption in fractal space with dimen-
sion D. For interband electronic transitions, the optical absorption coefficient
can be written as follows (Mott and Davis, 1979):
Nc (E) Nv (E − h̄ω) dE
α (ω) = B , (4.7)
h̄ω
where B is a constant which includes the square of the transition matrix element
as a factor, and the integration is over all pairs of states in the conduction Nc (E)
and valence Nv (E) bands. Let us express the DOS of the conduction and valence
bands as
Nc (E) = const (E − Ec )α , (4.8)

Nv (E) = const (E − Ev )β , (4.9)
where α = (Dc − 2)2 and β = (Dv − 2)2. Here Dc and Dv are the fractal dimen-
sionalities of the conduction and valence bands, respectively. Then eqn. (4.8)
produces (Nessa et al., 2000)
α (ω) h̄ω = B (h̄ω − Eo )α+β+1 , (4.10)
where Bʹ is another corresponding constant. This yields
[α (ω) h̄ω]n = B 1/n (h̄ω − Eo ) , (4.11)

where 1n = α + β + 1. If the form of both Nc (E) and Nv (E) is parabolic, i.e.
α = β = 12 for 3D materials as in the most of the studies, then eqn. (4.11)
becomes
[α (ω) h̄ω]1/2 = B (h̄ω − Eo ) , (4.12)
which is the well-known Tauc relation.

Tauc’s relation for fundamental optical absorption is applied to many amorph-
ous semiconductors to deduce the optical gap, Eo . There are many exceptions,
for example n deviates from 12 to take values between 0.3 and 1. Note that the
valence band (VB) of a-Chs is formed by lone-pair (LP) electrons (known as LP
semiconductors), and the conduction band (CB) is created by antibonding states.
Thus the energy-dependent DOS form for the CB is expected to be basically
different from that of the VB. A typical example may be n 1 for a-Se. Although
a LP band (VB) may have a 3D nature, the CB (antibonding states) should be
characterized by a chain-like 1D nature, as long as no interaction between the
chains is assumed. We therefore expect that α = −12 and β = 12, producing
n = 1 (Nessa et al., 2000).
As we have already discussed, the energy dependence of the DOS for the CB
and VB is expected to be different in a-Chs because the space dimensionality
for the VB is larger than that for the CB. Therefore, we introduce Dc and
Dv separately, and then n takes non-integer values for fractional-dimensional
systems; 1n = α + β + 1 = (Dc + Dv − 2)2. In some a-Ch films, it is
observed that Dc and Dv depend on the preparation conditions (Shimakawa,
Singh, and O’Leary, 2006; Singh and Shimakawa, 2003).
Optical absorption tail (Urbach tail)

Following the fundamental optical absorption, an optical transition in the (expo-
nential) localized tail states (lower-energy region) is observed. The origin of the
exponential tail states is still a matter of debate, and hence there are many models
available that discuss the origin of such localized tails (see, for example, Singh
and Shimakawa, 2003; Tanaka and Shimakawa, 2011). In earlier investigations
of electronic properties and related effects, the tail was usually considered to be
an exponential or Gaussian decaying function. New types of defects, such as
triangles and squares, have electronic states in the gap. If we accept this result,
we see that the localized states have structure, i.e. two peaks can be found inside
Figure 4.13. Frequency-dependent conductivity (real part) in a-Si:H. Open

circles and crosses show the experimental data in the radio-frequency (rf)
(Shimakawa and Ganjoo, 2002) and THz ranges (Fekete et al., 2009), respect-
ively. The solid line is a prediction using Dyre’s expression (Dyre, 1988; Dyre
and Schroder, 2000).
the tail (Kugler, 2012). The larger peak results from the squares, and the peak
with larger energy is formed by triangles.
Photocarrier thermalization in picosecond time range and

terahertz spectroscopy
The methods classically used to investigate photoconductivity are time-of-flight
(TOF) spectroscopy (called primary photoconductivity) and the more common
steady-state photoconductivity spectroscopy (called secondary photoconduct-
ivity). The TOF experiment, which is completely different from the secondary
photoconductivity experiment, involves photocarriers, far from thermal equilib-
rium, that transit from the illuminated electrode to the counter electrode. As
the details of photoconductivity spectroscopy have been reported extensively in
the literature (for example, Street 1991; Morigaki, 1999, Singh and Shimakawa,
2003), we discuss only photocarrier transport with a subpicosecond time reso-
lution (THz spectra) in a-Si:H. This work has been intensively studied and has
presented excellent information on short-lived photocarriers in the CB (Fekete
et al., 2009; Lloyd-Hughes and Tae-in Jeon, 2012).
Figure 4.13 shows the frequency-dependent conductivity (real part) in a-Si:H.
Open circles and crosses show the experimental data in the radio-frequency (rf)
(Shimakawa and Ganjoo, 2002) and terahertz (THz) ranges (Fekete et al., 2009),
respectively. The solid line is a prediction using Dyre’s expression of the ac loss
C.B.
Epc
E
EF
V.B.
Figure 4.14. Potential fluctuation model in a-Si:H (Overhof and Thomas, 1989;
Shimakawa and Ganjoo, 2002).
induced in inhomogeneous media (Dyre, 1988; Dyre and Schroder, 2000). Both
the rf and THz ranges will be discussed in the following.
Usually, frequency-dependent conductivity (ac conductivity or ac loss) is
considered to be caused by a localized motion of carriers (Mott and Davis, 1979;
Singh and Shimakawa, 2003). A similar ac loss is also induced in macroscop-
ically inhomogeneous media (Dyre and Schroder, 2000; Singh and Shimakawa,
2003). In the following, we briefly summarize the ac loss in inhomogeneous
media, as this mechanism may dominate ac losses of a-Si:H in all frequency
ranges.
A very simple percolation-path method proposed by Dyre (1988) was origin-
ally applied to the random walk (hopping) of localized carriers; now it is also
applied to macroscopically inhomogeneous media. In inhomogeneous media,
“space charges” relax with a relaxation time τ = ɛ0 ɛ σ , where ɛ0 is the electric
constant, ɛ is the background dielectric constant, and σ is the electrical con-
ductivity. A space-charge relaxation induces the complex frequency-dependent
conductivity as (Dyre, 1988)
c iωln (τmax /τmin )

σ ∗ (ω) = −iω + ≡ σR + iσI , (4.13)
6 In [(1 + iωτmax ) / (1 + iωτmin )]
where c is a constant and two cut-offs in the distribution of τ , τ max = ɛ0 ɛ σ min

and τ min = ɛ0 ɛ σ max , are introduced. Here, σ min and σ max are the local-
minimum and local-maximum conductivities in the inhomogeneous media.
Potential fluctuations, shown schematically in Figure 4.14 for an example of
Figure 4.15. Real part of THz conductivity. The open circles and the solid
line show the experimental data and the theoretical prediction, respectively.
(From Shimakawa, Wagner, and Frumar (2013). Phys. Status Solidi B, 250,
1004–1007. Copyright 2013 by John Wiley and Sons Inc.)
inhomogeneity, are suggested in a-Si:H (Overhof and Thomas, 1989). The local
electronic conductivity therefore takes a variety of random values. Using

ln (a + ib) = ln a 2 + b2 + i tan−1 (b/a) (4.14)
in the limit of ω → 0, we obtain

c ln (τmax /τmin )
σ (0) = . (4.15)
6τmax
Then we get the real and imaginary parts of the conductivity, σ R (ω) and σ I( ω),
respectively. Note that c in eqn. (4.15) is a constant and that τ min determines the
high-frequency saturation of the conductivity.
The solid line in Figure 4.13 is the predicted real part of the conductivity, σ R ,
obtained from σ (0) = 3.5 × 10−6 S cm−1 , τ max = 2 × 10−6 s. It is not easy to
predict the value of τ min ; the predicted σ (ω) is not plotted above 1 × 1010 Hz.
Terahertz conductivities, shown by crosses (×) in Figure 4.13, are obtained from
the optical pumping THz spectroscopy (Fekete et al., 2009), and these are very
much higher than those reported in the rf range. This is due to optical pumping,
which produces a large number of free carriers around 1 × 1018 cm−3 just after
photoexcitation (Fekete et al., 2009). We will discuss THz conductivity in more
detail.
The open circles in Figure 4.15 show the experimental data of the real part of
the THz conductivity σ R , as already shown in Figure 4.13. The solid line displays
the predicted values using eqs. (4.13) and (4.15), with σ (0) = 0.35 S cm−1 ,
τ max = 1.0 × 10−11 s and τ min = 4.3 × 10−13 s. The fitting of the predicted
data to the experimental data is good. Note that σ (0) = 0.35 S cm−1 does not
refer to the actual dc conductivity. Rather, it can be regarded as the virtual dc
conductivity, since, in the optical pumping technique, dc conductivity is not a
meaningful concept.
We now know that the value of τ min lies in the region of 10−13 s. Note again
that τ min is not deduced from the rf-range spectroscopy. Let us discuss what
τ min means. The ac loss in the rf range is interpreted under the assumption
of potential fluctuations (Shimakawa and Ganjoo, 2002). Here, the saturated
value of σ (ω) approaches 1 S cm−1 , and then the minimum dielectric relaxation
time τ min is expected to be 9 × 10−13 s. This value is close to that (4.3 × 10−13 s)
deduced from experimental data. Thus, it is suggested that potential fluctuations
which affect the rf conductivity also dominate conduction loss in the THz range
(Shimakawa, Wagner, and Frumar, 2013).
An alternative cause of THz conductivity – carrier hopping – which may
induce behavior similar to that previously described, should also be discussed.
The inverse of τ min (2.5 × 1012 s−1 ) is close to the phonon frequency, and Fekete
et al. (2009) have discussed this behavior in terms of a hopping-type relaxation
in localized tail states. It is not easy to distinguish which mechanism dominates
the THz conductivity: dielectric relaxation or hopping. If τ min changes with
temperature or optical pumping intensity, the dynamic loss may originate from
dielectric relaxation. There is unfortunately no literature available on this subject.
Photoluminescence
We briefly review the photoluminescence in a-Si:H and amorphous chalco-
genides. A radiative photon emission process in condensed matter is called
luminescence, and conceptually photoluminescence (PL) is the inverse pro-
cess of the optical absorption already discussed. As all information on photon
absorption and emission in materials can be involved in PL processes, PL is an
important topic, and there are many reports and reviews discussing PL mech-
anisms in amorphous semiconductors (see, for example, Aoki, 2006a, 2006b,
2012).
Frequency-resolved spectroscopy (FRS) may be used to understand directly
the lifetime distribution. In particular, quadrature frequency-resolved spec-
troscopy (QFRS) is known to be more suitable for studying amorphous semi-
conductors having a broad PL lifetime distribution (Aoki, 2006b). As shown
Figure 4.16. A broad photoluminescence lifetime distribution of a-Si:H. (S):

singlet excitonic recombination; (T): triplet excitonic recombination; (DP):
distant-pair recombination. (From Aoki (2006b). Photoluminescence. In Opti-
cal Properties of Condensed Matter and Applications, ed. J. Singh. Chichester:
John Wiley & Sons, pp. 75–106. Copyright 2013 by John Wiley and Sons Inc.)
in Figure 4.16, Aoki and co-workers (Aoki, 2006a) succeeded in using QFRS
to expand the very wide time (frequency) range 10−9 –102 s in a-Si:H. There
are three main peaks in the lifetime distribution. The shortest lifetime (around
10−6 s) and the medium lifetime (around 10−3 s) can be identified as singlet (S)
and triplet (T) excitonic recombination. The longest lifetime (around 101 s) can
be explained by distant-pair recombination, in which photoexcited electrons and
holes are randomly distributed in band tails.
Similar S and T peaks have been also found in a-Chs (Aoki et al., 2005),
and the principal features in PL have been universally observed in amorphous
semiconductors.
Electronic transport (in dark conditions)

In this section, we discuss the electronic transport phenomena in a-Si:H and
a-Chs. The electronic transport around room temperature is known to be domi-
nated by band transport for both material systems. The central issues concerning
the electronic transport have been widely discussed in the literature, and we do
not need to consider them here because there is nothing new to report to date.
However, we should say that there is an important and interesting issue regard-
ing electronic transport that is still not properly understood. This is called the
107
50
As–Se–S
40 As–Se–I
5
10
As–Se–Te
In(␴0 S cm–1)
30
␴0 (S cm–1)
103 20
10
101 0
–10
10–1
0.0 0.2 0.4 0.6 0.8 1.0 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4
Δ (eV) ΔE (eV)
(a) (b)
Figure 4.17. MN rule in (a) a-Si:H (Overhof and Thomas, 1989), (b) chalco-
genide glasses (Shimakawa and Abdel-Wahab, 1997).
Meyer–Neldel (MN) rule or the compensation rule. A comprehensive review is

given by Yelon, Movaghar, and Crandall (2006), and the current understanding
is reported in Linert and Yelon (2013).
We therefore briefly introduce the MN rule as reported in electronic transport
for a-Si:H and a-Chs, although the MN rule is related to chemical reactions,
the dynamics of electronic and ionic transports, and thermodynamics in many
material systems, when these mechanisms are thermally activated.
The band-like electronic conductivity, for example in a-Si:H and a-Chs, near
room temperature may be empirically described as
σ = σ00 exp ( E/EMN ) exp (− E/kB T ) ≡ σ0 exp (− E/kB T ) , (4.16)
where the prefactor σ 0 is not a constant; it is given by σ 00 exp( EEMN ). Here,

E and EMN are called the activation energy and the Meyer–Neldel energy,
respectively. Examples of the MN rule are shown in Figures 4.17(a) and (b) for
a-Si:H and a-Chs, respectively. The prefactor itself is a function of the activation
energy E. The MN energy, EMN , drops around 40–50 meV.
In a-Si:H, this behavior is well explained by the statistical shift of the Fermi
level (i.e. the temperature variation of the Fermi level) (Overhof and Thomas,
1989). Although, as shown in Figure 4.17(b), a similar effect can be found in
the a-Chs, this behavior is not easy to explain using the statistical shift of the
Fermi level (Shimakawa and Abdel-Wahab, 1997). As the MN rule is universally
observed in a wide class of materials, Yelon et al. (2006) have suggested that
the multiexcitation entropy is important in some kinetics and thermodynamics.
References 113
A recent idea proposed by Okamoto et al. (2010) is based on the universality of

these phenomena.
In our view, however, kinetics and thermodynamics are different topics, and
therefore each phenomenon may have different origins. For example, for ionic
transport in crystalline solids, the same MN rule as in eqn. (4.16) has been found,
and this can be explained in terms of phonon absorption and emission processes
with lattice distortion (Shimakawa and Aniya, 2013). Note that the MN law is
observed in both crystalline and disordered materials, although it is believed
to exist only in disordered matter. The definitive solution to the important and
unresolved problem of the MN rule is therefore extremely necessary, since this
effect may contain unknown principles of physics.
References
Abkowitz, M. and Markovics, J.M. (1984). Evidence of equilibrium native defect popu-
lations in amorphous chalcogenides from analyses of xerographic spectra. Philos.
Mag. B, 49, L31–L36.
Anderson, P.W. (1958). Absence of diffusion in certain random lattices. Phys. Rev., 109,
1492–1505.
Anderson, P.W. (1975). Model for the electronic structure of amorphous semiconductors.
Phys. Rev. Lett., 34, 953–955.
Aoki, T. (2006a). Understanding the photoluminescence over 13-decade lifetime distri-
bution in a-Si:H. J. Non-Cryst. Solids, 352, 1138–1143.
Aoki, T. (2006b). Photoluminescence. In Optical Properties of Condensed Matter and
Applications, ed. J. Singh. Chichester: John Wiley & Sons, p. 75.
Aoki, T. (2012). Photoluminescence spectroscopy. In Characterization of Materials,
2nd edn., ed. E.N. Kaufmann. Hoboken, NJ: John Wiley & Sons, pp. 1–12.
Aoki, T., Saito, D., Ikeda, K., Kobayashi, S., and Shimakawa, K. (2005). Radiative
recombination processes in chalcogenide glasses deduced by lifetime measure-
ments over 11 decades. J. Opt. Adv. Mater., 7, 1749–1757.
Bishop, S.G., Strom, U., and Taylor, P.C. (1975). Optically induced localized paramag-
netic states in chalcogenide glasses. Phys. Rev. Lett., 34, 1346–1350.
Chittik, R.C., Alexander, J.H., and Sterling, H.J. (1969). The preparation and properties
of amorphous silicon. J. Electrochem. Soc., 116, 77–81.
Dyre, J.C. (1988). The random free energy barrier model for ac conduction in disordered
solids. J. Appl. Phys., 64, 2456–2468.
Dyre, J.C. and Schroder, T.B. (2000). Universality of ac conduction disordered solids.
Rev. Mod. Phys., 72, 873–892.
Fekete, L., Kuzel, P., Nemec, H. et al. (2009). Ultrafast carrier dynamics in microcrys-
talline silicon probed by time-resolved terahertz spectroscopy. Phys. Rev. B, 79,
115306, 1–13.
Ganjoo, A. and Shimakawa, K. (1994). Estimation of density of charged defects in
amorphous chalcogenides from a.c. conductivity: random-walk approach for bipo-
larons based on correlated barrier hopping. Philos. Mag. Lett., 70, 287–291.
He, X.-F. (1990). Fractional dimensionality and fractional derivative spectra of interband
optical transitions. Phys. Rev. B, 42, 11751–11756.
Hirabayashi, I. and Morigaki, K. (1983). Light induced metastable effect on the short
lived photoinduced midgap absorption in hydrogenated amorphous silicon. J. Non-
Cryst. Solids, 59 & 60, 433–436.
Imagawa, O., Akiyama, T., and Shimakawa, K. (1984). Localized density of states in
amorphous silicon determined by electrophotography. Appl. Phys. Lett., 45, 438–
439.
Isoya, J., Yamasaki, S., Ohkushi, H., Matsuda, A., and Tanaka, K. (1993). Electron-
spin-echo envelope-modulation study of the distance between dangling bonds
and hydrogen atoms in hydrogenated amorphous silicon. Phys. Rev. B, 47, 7013–
7024.
Jackson, W.B. and Amer, N.M. (1982). Direct measurement of gap-state absorption in
hydrogenated amorphous silicon by photothermal deflection spectroscopy. Phys.
Rev. B, 25, 5559–5562.
Kádas, K. and Kugler, S. (1997a). Midgap states in nitrogen doped diamond-like amorph-
ous carbon. Solid State Commun., 102, 721–723.
Kádas, K. and Kugler, S. (1997b). Impurity levels in phosphorus and boron doped
amorphous silicon. Phil. Mag. B, 76, 281–285.
Kádas, K., Ferenczy, Gy.G., and Kugler, S. (1998). Theory of dopant pairs in four-
fold coordinated amorphous semiconductors. J. Non-Cryst. Solids, 227–230, 367–
371.
Kastner, M., Adler, D., and Fritzsche, H. (1976). Valence-alternation model for localized
gap states in lone-pair semiconductors. Phys. Rev. Lett., 37, 1504–1507.
Kugler, S. (2012). Advances in understanding the defects contributing to the tail states
in pure amorphous silicon. J. Non-Cryst. Solids, 358, 2060–2062.
Kugler, S. and Laszlo, I. (1989). Connection between topology and π -electron structure
in amorphous carbon. Phys. Rev. B, 39, 3882–3884.
Kugler, S., Kohary, K., Kádas, K., and Pusztai, L. (2003). Unusual atomic arrangements
in amorphous silicon. Solid State Commun., 127, 305–309.
Lang, D.V. (1974). Deep-level transient spectroscopy: a new method to characterize
traps in semiconductors. J. Appl. Phys., 45, 3023–3032.
Laszlo, I. (1993). Stable electronic energy levels in the presence of off-diagonal disorder.
Int. J. Quantum Chem., 48, 135–146.
Laszlo, I. (2000). Graph theoretical study of topologically determined electronic energy
levels. J. Mol. Struct.: THEOCHEM, 501–502, 501–508.
Linert, W. and Yelon, A. (eds.) (2013). Isokinetic relationships. Monatsh. Chem. (special
issue), 144, 1–2.
Lloyd-Hughes, J. and Tae-In Jeon (2012). A review of the terahertz conductivity of bulk
and nano-materials. J. Infrared Milli. Terahz. Waves, 33, 871–926.
Lukács, R., Hegedus, J., and Kugler, S. (2008). Structure and photoinduced volume
changes of obliquely deposited amorphous selenium. J. Appl. Phys., 104, 103512,
1–6.
Lukács, R., Hegedus, J., and Kugler, S. (2009). Microscopic and macroscopic models
of photo-induced volume changes in amorphous selenium. J. Mater. Sci: Mater.
Electron., 20, S33–S37.
Mackenzie, K.D., LeComber, L.G., and Spear, W.E. (1982). The density of states
in amorphous silicon determined by space-charge-limited current measurements.
Philos. Mag. B, 46, 377–389.
References 115
Madan, A., LeComber, P.G., and Spear, W.E. (1976). Investigation of the density of
localized states in a-Si using the field effect technique. J. Non-Cryst. Solids, 20,
239–257.
Mandelbrot, B.B. (1982). The Fractal Geometry of Nature. New York: Freeman.
Marshall, J.M. and Owen, A.E. (1976). Field-effect measurements in disordered
As30 Te48 Si12 Ge10 and As2 Te3 . Philos. Mag., 33, 457–474.
Morigaki, K. (1999). Physics of Amorphous Semiconductors. London: Imperial College
Press and World Scientific.
2nd edn. Oxford: Clarendon Press, p. 272.
Nessa, M., Shimakawa, K., Ganjoo, A., and Singh, J. (2000). Fundamental optical
absorption on fractals: a case example for amorphous chalcogenides. J. Optoelec-
tron. Adv. Mater., 2, 133–138.
Okamoto, H., Sobajima, Y., Toyama, T., and Matsuda, A. (2010). Laplace Meyer–Neldel
relation. Phys. Status Solidi A, 207, 566–569.
Overhof, H. and Thomas, P. (1989), Electronic Transport in Hydrogenated Amorphous
Semiconductors. Berlin: Springer-Verlag.
Pantelides, S.T. (1986). Defects in amorphous silicon: a new perspective. Phys. Rev.
Lett., 57, 2979–2982.
Petri, I., Salmon, P.S., and Fischer, H.E. (2000). Defects in a disordered world: the
structure of glassy GeSe2. Phys. Rev. Lett., 84, 2413–2416.
Rayment, T. and Elliott, S.R. (1983). Small-angle neutron scattering study of anisotropic
growth morphology and irreversible photodensification in a-GeSe3 films. Phys. Rev.
B, 28, 1174–1177.
Shimakawa, K. and Abdel-Wahab, F. (1997). The Meyer–Neldel rule in chalcogenide
glasses. Appl. Phys. Lett., 70, 652–654.
Shimakawa, K. and Aniya, M. (2013). Dynamics of atomic diffusion in condensed
matter: origin of the Meyer–Neldel compensation law. Monatsh. Chem., 144, 67–
71.
Shimakawa, K. and Ganjoo, A. (2002). ac photoconductivity of hydrogenated amorphous
silicon: influence of long-range potential fluctuations. Phys. Rev. B, 65, 165213,
1–5.
Shimakawa, K. and Katsuma, Y. (1986). Extended step-by-step analysis in space-charge-
limited current: application to hydrogenated amorphous silicon. J. Appl. Phys., 60,
1417–1421.
588.
Shimakawa, K., Singh, J., and O’Leary, S.K. (2006). Optical properties of disordered
condensed matter. In Optical Properties of Condensed Matter and Applications,
ed. J. Singh. Chichester: John Wiley & Sons, pp. 47–62.
Shimakawa, K., Wagner, T., and Frumar, M. (2013). THz photoconductivity in a-Si:H.
Phys. Status Solidi B, 250, 1004–1007.
Singh, J. (2002). Effective mass of charge carriers in amorphous semiconductors and its
applications. J. Non-Cryst. Solids, 299–302, 444–448.
Singh, J. and Shimakawa, K. (2003). Advances in Amorphous Semiconductors. London:
Taylor & Francis, p. 57.
Sólyom, J. (2009). Fundamentals of the Physics of Solids. Volume 2: Electronic Prop-
erties. Berlin Heidelberg: Springer-Verlag, p. 220.
Spear, W.E. and Le Comber, P.G. (1975). Substitutional doping of amorphous silicon.
Starbova, K., Mankov, V., Dikova, J., and Starbov, N. (1999). The effects of vapour
incidence on the microstructure and related properties of condensed GeS2 thin
films. Vacuum, 53, 441–446.
sity Press.
Vanecek, M., Kocka, J., Stuchlik, J., Kozisek, Z., Stika, O., and Triska, A. (1983). Density
of the gap states in undoped and doped glow discharge a-Si:H. Solar Energy Mater.,
8, 411–423.
Weaire, D. and Thorpe, M.F. (1971). Electronic properties of an amorphous solid. I.
A simple tight-binding theory. Phys. Rev. B, 4, 2508–2520.
Yelon, A., Movaghar, B., and Crandall, R.S. (2006). Multi-excitation entropy: its role in
thermodynamics and kinetics. Rep. Prog. Phys., 69, 1145–1194.
Zallen, R. (1983). The Physics of Amorphous Solids. New York: John Wiley & Sons Inc.
5
Photoinduced phenomena
Reversible and irreversible photoinduced changes are found in amorphous

semiconductors, particularly in amorphous chalcogenides (a-Chs). The term
reversible usually means that the final state of a substance corresponds to its ini-
tial state by means of structural relaxation. The structural relaxation is triggered
by switching off the illumination, thermal annealing near the glass-transition
temperature, or other photon irradiation. Reversible photoinduced phenomena
can be seen as the changes in the physical properties of the semiconductor. On
the other hand, irreversible changes are mostly related to chemical reactions or
to a phase change, such as crystallization. We are interested in physical changes
in a-Chs, and will mostly concentrate on the reversible phenomena observed in
amorphous semiconductors.
5.1 Photoinduced volume changes

In amorphous chalcogenides, photodarkening (PD, in which the bandgap
decreases during and after the illumination) and photoinduced volume changes
(usually photoinduced volume expansion, PVE) were for a long time believed
to be two aspects of the same phenomenon. Therefore, a one-to-one correlation
was expected to exist between PD and PVE. The first systematic study of PVE
was performed by Hamanaka et al. (1976), showing that PD is always accom-
panied by PVE. However, the experiment carried out by Tanaka (1998) showed
that from the time evolution of these two phenomena it follows that the time
constants of the PD and the PVE in a-As2 S3 are different. It was observed that
the evolution of PVE saturates earlier than that of PD, suggesting that these two
phenomena are not directly related to each other.
117
118 Photoinduced phenomena
As most of the measurements of reversible PVE have been made after switch-
ing off the illumination (i.e. in the metastable state) (Hamanaka et al., 1976;
Tanaka, 1990, 1998; Kuzukawa et al., 1999), the correlation during illumination
remains unclear. To understand the dynamics of PVE and PD during illumin-
ation, in situ measurements were performed (Ganjoo, Ikeda, and Shimakawa,
1999; Ganjoo et al., 2000: Ganjoo and Shimakawa, 2002). A real-time in situ
surface height measuring system has been developed using a Twyman–Green
interferometer with image analysis technology (Ikeda and Shimakawa, 2004;
Shimakawa, Ikeda, and Kugler, 2004a).
Figure 5.1(a) shows an example of a surface height map for well-annealed
flatly deposited a-As2 Se3 films (on a Si substrate). The measurements were
carried out at 300 K. Figures 5.1(b) and (c) show the time evolution of
the changes in flatly and obliquely deposited films. In flatly deposited films
(Figure 5.1(b)), the surface height increased by 10 nm ( dd = 2%) dur-
ing 200 s of laser illumination (wavelength of 532 nm and power density of
90 mW cm−2 ). After 600 s, the illumination was turned off. The surface height
started decreasing and settled in 200 s, at which point the height was 2 nm lower
than before the light was switched off. In the case of obliquely deposited films
(Figure 5.1(c)), the surface height decreased by 12 nm ( dd = 2.4%) in 3 ×
104 s. Similar results have been observed in other a-Chs.
As shown in Figure 5.2, a less significant PVE than that seen in a-As2 Se3 was
found in (flatly deposited) a-Se films. The height increased rapidly by 2.5 nm
( dd = 0.5%) with the illumination (wavelength of 532 nm in wavelength and
power density of 90 mW cm−2 ). As soon as the light was turned off, after 800 s
of illumination, the height decreased by 2 nm and then gradually decreased to
the original height in 200 s.
Let us briefly discuss the height variations. As shown in Figure 5.1, in
flatly deposited a-As2 Se3 films PVE was observed during and after the illu-
mination. The transient PVE (PVE occurs only during the illumination) must
be involved in the whole PVE process (Ganjoo et al., 2000). The remain-
ing increase in surface height that occurred after illumination is a so-called
“metastable” PVE (Shimakawa, Kolobov, and Elliott, 1995; Tanaka, 1998). In
obliquely and as-deposited a-As2 Se3 films, as shown in Figure 5.1(c), the sur-
face decreased with time when the light was turned on (volume contraction).
The PVE may have been taken over by photoinduced volume contraction (PVC),
and hence the height decreased with time. This volume contraction can be inter-
preted by considering void corruption, as many voids are involved in obliquely
[mm]
–2.0
–1.0
+2.0 0.0
+1.0
+2.0
+1.0
0.0
–1.0
–2.0 [nm]
[mm] 10
(a) 0
12
Light off
10
Height (nm)
8
6
4
2 Light on
0
–2
0 5 × 102 103 2 × 103 2 × 103 3 × 103 3 × 103
Time (s)
(b)
2
0
–2 Light on
Height (nm)
–4
–6
–8
–10
–12
–14
–16
0 104 2 × 104 3 × 104 4 × 104 5 × 104
Time (s)
(c)
Figure 5.1. (a) Surface height map for flatly deposited a-As2 Se3 . (b) Time
evolution of surface height for flatly deposited a-As2 Se3 . (c) Time evolution
of surface height for obliquely deposited a-As2 Se3 . (Adapted from Ikeda and
Shimakawa (2004). J. Non-Cryst. Solids, 338–340, 539. Copyright 2013 with
permission from Elsevier.)
Figure 5.2. Time evolution of surface height for flatly deposited a-Se.
deposited films which consist of a columnar structure with a large amount

of free space.
Atomic-scale computer simulations of photoinduced volume changes

The photoinduced volume changes in a-Se were simulated using the TBMD
computer code ATOMDEP (see Chapter 3). A tight-binding model developed
by Molina, Lomba, and Kahl (1999) was used to describe the atomic interactions
between selenium atoms. The velocity Verlet algorithm was applied in order to
follow the motion of the atoms with a time step t = 2 fs. The temperature was
controlled via the velocity-rescaling method. Samples containing 162 atoms
were prepared by the “cook and quench” procedure. A temperature of 20 K
was reached at the end of the preparation, and the sample for 500 MD steps
(1 ps) was relaxed. In order to model the photoinduced volume changes, the
periodic boundary conditions were lifted along the z direction at this point. This
procedure gave us a slab geometry with periodic boundary conditions in two
dimensions only. The system was then relaxed for another 40 000 MD steps
(80 ps) to obtain a stable configuration. The sample can expand or shrink in the
z direction. As a measure of sample thickness, the distance between the centers
of masses of 15–15 surface atoms on both sides was used, as indicated by the
the arrow in Figure 5.3 (Lukács, Hegedus, and Kugler, 2009).
Photoexcitation generates an electron–hole pair when a photon is absorbed.
This process can be modeled by transferring an electron from the highest occu-
pied molecular orbital (HOMO) to the lowest unoccupied molecular orbital
(LUMO). In the computer simulations it was assumed that, immediately after
y
Figure 5.3. Snapshot of a glassy selenium network. The sample can move in
the z direction. (Adapted from Lukács, Hegedus, and Kugler (2009). J. Mater.
Sci.: Mater. Electron., 20, S33. With kind permission from Springer Science
and Business Media.)
the photon absorption, the electron and the hole became separated in space
on a femtosecond time scale (Hegedus et al., 2005). The Coulomb interaction
between the electron and the hole was neglected. The influences due to the
excited electron and the hole were treated independently. At first, an extra elec-
tron was put into the LUMO (excited electron creation), then an electron in
the HOMO (hole creation) was annihilated. When an additional electron was
put in the LUMO, a bond-breaking event occurred. In the majority of cases, a
covalent bond between twofold- and threefold-coordinated atoms was broken;
however, bond breaking was sometimes observed between twofold- and twofold-
coordinated atoms. Our localization analysis revealed that the LUMOs are usu-
ally localized at such sites before the bond breaking. The change in bond length
alternates between shrinkage and elongation in the vicinity of the broken bond
due to bond breaking. If the electron in the LUMO is de-excited, then all the
bond lengths are restored to their original value. A characteristic time devel-
opment of a single photoinduced bond-breaking event is shown in Figure 5.4.
Before the excitation at 5 ps, the bond length was approximately 0.255 nm. Dur-
ing illumination, this bond length increased by 10–20%, then decreased to its
original value after the de-excitation of the electron at 15 ps. The volume change
of the a-Se slab followed the bond breaking, and it showed damped oscillations
3.4 25.6
3.2
25.5
bond length (A)

3.0
thickness (A)
°
2.8
25.4
2.6
2.4 25.3
2.2
2.0 25.2
0 5 10 15 20
time (ps)
Figure 5.4. Atomic distance separation of breaking bond (solid line) and thick-
ness of sample (dotted line) as a function of time. Arrows indicate that an
excited electron was created at 5 ps and annihilated at 15 ps. (Taken with per-
mission from Hegedus et al. (2005). Phys. Rev. Lett., 95, 206803. Copyright
2013 by the American Physical Society.)
21.65
Light off Light off
length of the sample (Å)
21.6
21.55
21.5
21.45
Light on Light on
21.4
3 4 5 6
time (ps)
Figure 5.5. Reversible thickness change of a-Se sample during photoexcitation.

(Adapted from Kugler, Hegedus, and Kohary (2007). J. Mater. Sci.: Mater.
Electron., 18, S163. With kind permission from Springer Science and Business
Media.)
on a picosecond time scale, as shown in Figure 5.4. The reversible thickness

change of the a-Se sample was observed during several photoexcitations, as
demonstrated in Figure 5.5 (Kugler, Hegedus, and Kohary, 2007).
Hartree–Fock ab initio Raman spectra calculations and Raman spectroscopic
measurements were carried out on a-Se to identify the characteristic vibrational
mode caused by sigma bonds (Lukács et al., 2010). In the Raman spectra the
peaks are around 250 cm−1 (corresponding to 0.234 nm), which is attributed
Figure 5.6. Difference between the Raman spectra of the 10 min illuminated
a-Se sample and that of the 40 min relaxed a-Se sample. (Taken with permis-
sion from Lukács et al. (2010). J. Appl. Phys., 107, 073517. Copyright 2013,
American Institute of Physics.)
to the vibrational modes of the covalent bonds in a-Se. As the Raman intensity
varies in time due to the illumination, a large number of covalent bonds break;
these form again once the illumination is switched off. The difference between
the Raman spectra of the sample illuminated for 10 min and that for the relaxed
sample after 40 min is shown in Figure 5.6. The Raman spectra measurement
provides strong experimental evidence for the predicted photoinduced bond-
breaking process.
A very different behavior is observed during hole creation (Hegedus et al.,
2005). Interchain weak bonds (see Figure 3.23) are formed after creating a
hole, thereby causing the contraction of the sample, as observed in Figure 5.7.
This always occurs near the atoms where the HOMO is localized. Since the
HOMO is usually localized in the vicinity of a onefold-coordinated atom, the
interchain bond formation often takes place between a onefold-coordinated atom
and a twofold-coordinated atom. Sometimes, the formation of interchain bonds
between two twofold-coordinated atoms is also observed.
3.6 25.7
bond length (A)

°
thickness (A)
3.4 25.6
°
3.2 25.5
3.0 25.4
0 5 10 15 20
time (ps)
Figure 5.7. Atomic distance separation of interchain bond (solid line) and thick-
ness of sample (dotted line) as a function of time. (Taken with permission from
Hegedus et al. (2005). Phys. Rev. Lett., 95, 206803. Copyright 2013 by the
American Physical Society.)
Kinetics of volume change

Volume expansion and shrinkage are additive quantities, i.e. the expansion in
thickness d+ is proportional to the number of excited electrons ne and is given by
d+ = β + ne . Similarly, the shrinkage d− is proportional to the number of created
holes nh and is given by d− = β − nh . The parameter β + (β − ) is the average
thickness change caused by an excited electron (hole). The time-dependent
thickness change is equal to d(t) = d+ (t) – d− (t) = β + ne (t) – β − nh (t). Assuming
that the number of electrons is equal to the number of holes, ne (t) = nh (t) = n(t),
we obtain
d(t) = (β+ − β− )n(t) = βn(t), (5.1)
where β is a characteristic constant of the chalcogenide glass related to photo-

induced volume change, and is a unique parameter for each sample. The sign of
this parameter determines whether the material expands or shrinks.
The number of photoexcited electrons and holes is proportional to the time of a
steady state illumination. Their generation rate G depends on the photon absorp-
tion coefficient and the number of incoming photons. After photon absorption,
the excited electrons and holes randomly diffuse and eventually recombine. A
phenomenological non-linear rate equation for this process is given by
dn (t)
= G − Cn(t)2 . (5.2)
dt
change of thickness (nm)

1
(a)
0
3
(b)
0
0 100 200 300 400
time (s)
Figure 5.8. Measured photoinduced changes in a-Se. (a) Expansion due to

illumination (dotted line) and theoretical (tanh) curve (solid line). (b) Shrinkage
after switching off illumination (dotted line) and fitted theoretical curve (solid
line). (Taken with permission from Hegedus et al. (2005). Phys. Rev. Lett., 95,
The solution for the time-dependent volume change is given by
G √
d(t) = β tanh GC t . (5.3)
C
The factor tanh(x) is measurable after a lengthy illumination (tanh() = 1).
√
Only one fitting parameter remains ( GC). After the light has been turned off,
the rate equation reduces to
dn (t)
= −Cn(t)2 , (5.4)
dt
which has the solution
a
d(t) = , (5.5)
(aCt/β) + 1
where a is equal to d(t) when the illumination is switched off. The photoinduced
expansion of a-Se films was measured in situ using optoelectronic interference,
enhanced by image processing (Ikeda and Shimakawa, 2004). Figures 5.8(a)
and (b) show the measured time evolution and decay, respectively, of the surface
height; the best fits were obtained in the 0–400 s interval of illumination and
after switching off the light.
Some glasses show irreversible changes during illumination (Hegedus,

Kohary, and Kugler, 2006). Furthermore, the measured photoinduced volume
changes in obliquely and flatly deposited a-AsSe were quite different; i.e. the
first one shrunk and the other one expanded (Ikeda and Shimakawa, 2004). To
explain this difference, we must take into account the large number of irre-
versible changes in the local atomic arrangement. The total expansion includes
reversible (transient) and irreversible (metastable) changes:
d(t) = drev (t) + dirr (t). (5.6)
The reversible part follows the previously described time development dur-
ing and after illumination. There is no volume change caused by irreversible
microscopic effects after switching off the light. During the illumination, the
irreversible part of expansion/shrinkage is governed by
dn(t)
= G(N0 − n(t)), (5.7)
dt
where G describes the excited electron–hole generation rate and N0 is the max-
imum number of sigma bonds due to the irreversible bond breaking. Rewriting
the mathematical equation, we obtain
d(dirr (t))
= Girr − Cirr dirr (t). (5.8)
dt
The solution is given by
Girr
dirr(t) = (1 − exp(−Cirr t)). (5.9)
Cirr
The best fits of photoinduced volume changes during and after illumination
for flatly deposited a-AsSe are displayed in Figure 5.9. Both expressions have
positive prefactors; i.e. during the illumination, the sample expands.
The obliquely deposited porous sample shrinks. We can describe this process
using the same mathematical expression, but now both expressions have nega-
tive prefactors (Lukács and Kugler, 2011). The best fit for obliquely deposited
a-AsSe is shown in Figure 5.10. Negative prefactors mean that both reversible
and irreversible bond breaking cause macroscopic shrinkage in porous materials
(Chapter 3).
We should make an additional remark about time-dependent behavior. Instead
of a Debye-type normal exponential relaxation, the stretched exponential
10 10
change of thickness (nm)
8 8
6 6
4 4
2 2
0 0
0 100 200 300 400 500 1000 1500 2000
(a) times (b) times
Figure 5.9. Volume changes as a function of time of flatly deposited a-AsSe

thin film. Solid thin lines are the measured volume changes (a) during and
(b) after illumination; the solid thick line is the best fit; the two dotted
lines represent the reversible and irreversible parts of the theoretical curve.
(From Hegedus, Kohary, and Kugler (2006). J. Non-Cryst. Solids, 352, 1587.
Copyright 2013 with permission from Elsevier.)
–2
–4
Height (nm)
–6
–8
–10
–12
–14
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (s) × 104
Figure 5.10. Volume changes as a function of time of obliquely deposited
a-AsSe thin film. Symbols denote the measured volume changes during the
illumination, and the solid line is the best fit. (From Lukács and Kugler (2011).
Jpn. J. Appl. Phys., 50, 091401.)
functions in many cases are fitted to the measured data in disordered systems.
The mathematical expression is given by
! "
D (t) = A 1 − exp − (t /τ )β , (5.10)
–2
–4
Height (nm)
–6
–8
–10
–12
–14
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (s) × 104
Figure 5.11. Volume changes of the same obliquely deposited a-AsSe thin film
as in Figure 5.10 fitted by a stretched exponential function. (From Lukács and
Kugler (2011). Jpn. J. Appl. Phys., 50, 091401.)
where β (<1), A, and τ are fitting parameters. Usually these curves fit quite well,
yet the physical explanation of this phenomenon still remains unclear. In this
case, the stretched exponential function also fits the exponential data quite well,
as shown in Figure 5.11. This example demonstrates that two different processes
(reversible and irreversible) can also be fitted by a stretched exponential function.
Photoinduced volume changes in a-Si:H films and oxide glasses

Finally, we discuss photoinduced volume changes observed in a-Si:H films and
oxide glasses. Although a volume change after the bandgap illumination on
a-Si:H is quite insignificant, PVE has been reported after employing a special
technique, i.e. one that involves the detection of the bending of an optical lever,
with a detection limit of VV = 2 × 10−7 (Gotoh et al., 1998). The time
evolution of the PVE, together with the photoinduced defect creation (PDC)
by an Ar-ion laser (300 mW cm−2 ) is shown in Figure 5.12. The value of the
PVE, VV, approached 4 × 10−6 , which is much lower than the value of 4 ×
10−3 found in a-As2 S3 . An intimate correlation between the dynamics of PVE
and PDC (see Figure 5.12) suggests that PDC is the origin of PVE. The PDC in
amorphous chalcogenides and a-Si:H will be discussed in Section 5.3.
5.2 Photodarkening and photobleaching 129
5 20
4
15
Nd (×1015 cm–3)
ΔV/V (×10–6)
3
10
2
ΔV/V 5
1
Nd
0 0
0 1 2 3 4 5 6 7
Light soaking time (hr)
Figure 5.12. Time evolution of PVE together with PDC illuminated by Ar-ion
laser (300 mW cm−2 ) in a-Si:H. Open circles: volume change; closed squares:
defect density. (Taken with permission from Gotoh et al. (1998). Appl. Phys.
Lett., 72, 2978. Copyright 2013, American Institute of Physics.)
The opposite effect, i.e. volume compaction, occurs under bandgap illumin-
ation in g-SiO2 . Although the details are not clear, a possible mechanism is
related to PDC (Uchino et al., 2002). A photon creates a defect-like E center,
which may trigger the volume contraction. It is of interest to compare this effect
with volume expansion in a-Si:H.
5.2 Photodarkening and photobleaching

Photodarkening and photobleaching (PB) refer, respectively, to the bandgap
decreasing and increasing with illumination. Reversible photodarkening (RPD)
or photobleaching (RPB) is a characteristic feature of amorphous or glassy
chalcogenide materials. These effects do not occur in crystalline chalcogenides.
Note also that neither PD nor PB is observed in a-Si:H. As the refractive index
n is known to be closely related to the bandgap of a material (Mott and Davis,
1979), the photoinduced changes in the refractive index fall into the category
of PD or PB. Thus we do not discuss here the photoinduced optical constants
n or k.
The PD in amorphous chalcogenides is a very popular subject and is discussed
widely in numerous review works (see, for example, Tanaka, 1990; Pfeifer,
Paesler, and Agarwal, 1991; Shimakawa et al., 1995; Fritzshe, 2000; Kolobov,
2003). We start the discussion with an overall description of the features of the
observed PD.
(1) Among many a-Chs, a-As2 S3 shows the most significant change in the
bandgap, Eg ( Eg 50 meV at around room temperature).
(2) Eg decreases with increasing temperature and is scaled with TTg , where
T is the temperature at which the illumination is made and Tg is the glass-
transition temperature (Tanaka, 1983).
(3) Under the application of a high static pressure over 1.5 GPa (15 kbar), Eg
is known to decrease in a-Se (Tanaka, 1990).
(4) Bandgap illumination is the most effective for PD to occur.
In situ measurements are useful for examining PD dynamics (Ganjoo et al.,

2000). Therefore, we begin with the time evolution of the changes in the optical
absorption coefficient α. Two light beams were used: one was a light excitation
(Ar laser) and the other one was a monitor beam (1.95 eV). The change in optical
absorption coefficient is defined as α = (−1d)ln( T), where d is the film
thickness and T is the change in optical transmission, which is defined here as
T = TT0 (where T is the transmission at any time t and T0 is the transmission
when the illumination is switched on). The details of the experimental set-up
are described elsewhere (Ganjoo et al., 2000).
Figure 5.13 is an example of the time evolution of α for a-As2 Se3 films
during various cycles of excitation at 50 and 300 K (Ganjoo et al., 2002): α
increased rapidly at first for both temperatures, before reaching a state close to
saturation after some time. When the Ar-laser illumination was switched off,
a decrease in α was observed, which quickly reached a constant value. This
portion of the total change is the transient part induced by the illumination,
and the portion which remained after stopping the illumination is the so-called
“metastable” PD. The transient parts of the changes were found to be nearly
60% and 30% of the total changes induced during the illumination at 300 K and
50 K, respectively. The cycle was repeated many times after the metastable state
was reached, and every illumination confirmed the occurrence of transient PD
only.
5.3 Photoinduced defect creation: the Staebler–Wronski effect

Photoinduced degradation in a-Si:H was found during a decrease in photocurrent
after illumination (Staebler and Wronski, 1977), which was then known as
a phenomenon that accompanied defect creation (Street, 1991; Redfield and
Bube, 1996; Morigaki, 1999). Subsequently, photoinduced defect creation has
5.3 Defect creation 131
50 K
2
Δα (cm–1) × 104 OFF
300 K
ON
0
0 2000 4000 6000 8000
Time (s)
Figure 5.13. Time evolution of α for a-As2 Se3 films during various cycles of
excitation at 50 and 300 K. (From Ganjoo et al. (2002). J. Non-Cryst. Solids,
299–302, 917. Copyright 2013 with permission from Elsevier.)
also been found in amorphous chalcogenides (Shimakawa, 1986; Shimakawa,

Inami, and Elliott, 1990; Shimakawa et al., 1992, 1995).
Electron spin resonance is a popular experimental technique that is used
to understand the microscopic nature of defects, and therefore it applies to the
study of the PDC process. It is observed that the number of ESR centers increases
with illumination. However, a more convenient technique with which to obtain
information concerning the change in the number of defects is photocurrent spec-
troscopy, in which a change in photocurrent (conductivity) produces a change
in the number of defects (acting as recombination centers). As in a-Chs, the
principal defects are not ESR active, so spectroscopy using photocurrent can
be one of the more useful methods in the study of a-Chs as well as in a-Si:H
(Shimakawa et al., 1995).
First we discuss PDC in a-Si:H. The breaking of weak Si–Si bonds during
illumination is a generally accepted model in which the breaking is induced by
the non-radiative recombination of electrons and holes. A simple bond break-
ing produces a pair of neighboring dangling bonds (DBs). This model is not
consistent with the ESR results, which evidently suggests the presence of well-
separated DBs. It is now known that the photoenhanced diffusion of H atoms
plays a role in PDC (Street, 1991; Branz, 1999; Morigaki, 1999; Morigaki and
Hikita, 2007). The above models highlight the importance of hydrogen diffu-
sion (H-diffusion) in a Si network which is assisted by photocarrier recombina-
tion energies. We should also consider the primary event of optical excitation;
optical excitation itself may induce bond breaking. We do not have a proper
answer to the question “Is a PDC process dominated by H-diffusion?”. Consider,
for example, the amorphous chalcogenides. Although the structural network is
completely different from that of a-Si:H, PDC does occur in a-Chs. Therefore,
it is suggested that the diffusion of hydrogen atoms is a secondary effect of PDC
processes.
We now discuss PDC in a-Chs. Shimakawa et al. (1990, 1995) have assumed
the existence of strong carrier–phonon interactions and have concluded that
the primary event of photoexcitation is the creation of a self-trapped exciton
(STE) and then its non-radiative recombination, leading either to the creation
of metastable STE (or an intimate pair of charged defects) or to its return to
the ground state through radiative recombination. This process is similar to that
proposed in a-SiO2 (Shugler and Stefanovich, 1990; Itoh and Stoneham, 2001).
According to the STE model, the primary event is an optical excitation, not
a non-radiative recombination. The weakening of a bond due to STE creation
eventually leads to bond breaking. However, it should be noted that the creation
of metastable STE is the result, not the cause, of bond breaking (Shugler and
Stefanovich, 1990).
Let us discuss the model in detail. As already described in Section 4.3, it
is widely accepted that defects in a-Chs are negative-U by their nature, and
hence only the charged defects exist in this class of material (Mott and Davis,
1979). These defects are ESR inactive. However, optical excitation (a change in
electron or hole occupation) changes a-Chs to ESR-active states. This is known
as light-induced ESR (LESR).
A microscopic model for PDC, for example for a-As2 S(Se)3 , is shown in
Figure 5.14. Optical excitation produces P+ − + −
2 − C1 and P4 − C1 , which are
called an intimate pair (IP) of defects, where P and C refer to pnictogen and
chalcogen centers, respectively. The IPs created in this way can be regarded as the
optically created STE (Shimakawa et al., 1995; Song and Williams, 1996). The
subsequent bond-switching reactions can lead to a significant separation between
the charged defects, which are called random pairs (RPs). RPs are responsible
for the decrease in the electronic properties (for example, photoconductivity,
5.3 Defect creation 133
Figure 5.14. A microscopic model for PDC for a-As2 S(Se)3 . X is the ground
state, Y1 and Y2 are the intimate pairs (IPs), and Z1 and Z2 are random pairs
(RPs).
LESR, etc.) as well as the optical properties (Shimakawa et al., 1995). Note that
these IPs and RPs disappear after thermal annealing at low temperature (around
200 K) and at high temperature (near Tg ). The details are described elsewhere
(Singh and Shimakawa, 2003; Tanaka and Shimakawa, 2011).
Let us discuss the quantum efficiency of PDC. As already explained, the use
of a photocurrent measurement during the illumination is useful in defect spec-
troscopy. Now let us focus on the defect creation N(t) during the illumination.
It is assumed that the change in the photocurrent is dominated by PDC itself.
The photocurrent, under conditions of thermal equilibrium, together with the
exposure time can be expressed as follows (Shimakawa et al., 2004b):
C C Ip (0)
Ip (t) = = = , (5.11)
N0 + N(t) N0 (1 + N(t)/N0 ) 1 + N(t)/N0
where C is a constant and Ip (0) (= CN0 ) is the initial photocurrent; N(t)N0 can
be estimated from the ratio Ip (0)Ip (t), with N0 being the initial number of
defects.
Figure 5.15. Quantum efficiency η of defect creation at 300 K for both a-As2 Se3
(solid symbols) and a-Si:H (open symbols).
The photocurrent Ip (t) is written empirically for both a-Chs and a-Si:H as
follows (Shimakawa et al., 1995, 2004b):
Is
Ip (t) = ! ", (5.12)
1 − γ exp − (t /τ )β
where Is is a constant current attained by prolonged illumination, γ = Ns
(N0 + Ns ), τ is the effective creation time, and β is the dispersion parameter
(β < 1.0). Here Ns is the saturated number of the defect density. A change in Nt
is then given by
! "
N (t) = Ns 1 − exp − (t /τ )β . (5.13)
We now define the quantum efficiency (QE) η of PDC as
η = N (t)/Gt = N (t)/np , (5.14)
where G is the rate of absorbed photons (cm−3 s−1 ), t is the illumination time,
and np (= Gt) is the total number of absorbed photons.
For comparison, Figure 5.15 shows the experimental data of η vs. np at 300 K
for both a-As2 Se3 (solid symbols) and a-Si:H (open symbols). The efficiency η
decreased with increasing np and was higher for a-As2 Se3 than for a-Si:H. This
5.4 In situ measurements 135
Figure 5.16. In situ simultaneous measurement system, which monitors the

changes in surface height (for PVE), optical transmittance (for PD), and photo-
current (for PDC). (From Nakagawa et al. (2010). Phys. Status Solidi C, 7, 857.
Copyright 2013 by John Wiley and Sons Inc.)
can be attributed to the high degree of flexibility of the atomic structure in a-Chs.
A reported QE value of η 10−8 is found for a-Si:H, and the estimation here
from the photocurrent spectroscopy gives almost the same value after prolonged
illumination. Note that at the initial stage of the illumination the QE is much
higher than that after prolonged illumination for both a-As2 Se3 and a-Si:H.
5.4 In situ simultaneous measurement of PVE, PD, and PDC: the

dynamics of time evolution and mechanisms
Important issues concerning photoinduced effects on amorphous chalcogenides
are still a matter of debate: is there any direct correlation among PVE, PD, and
PDC? To formulate a proper answer to this question, in situ and simultaneous
measurements of PVE, PD, and PDC have been performed in a-As2 Se3 films.
In the following, we concentrate on the difference in the dynamics of these
photoinduced changes. Note that in previous sections we discussed PVE, PD,
and PDC independently of each other.
Figure 5.16 shows the in situ simultaneous measurement system which moni-
tors the changes of the surface height (for PVE), optical transmittance (for
PD), and photocurrent (for PDC). A phase-shifting interferometer (PSI) was
(a) (b)
Figure 5.17. Time evolution of the change in N(t)N0 , α(t), and surface height
h(t) for thicknesses of (a) d = 700 nm and (b) d = 200 nm in a-As2 Se3 . (From
Nakagawa et al. (2010). Phys. Status Solidi C, 7, 857. Copyright 2013 by John
Wiley and Sons Inc.)
employed to measure the height map, as already described in Section 5.1. A

He–Ne laser (633 nm, 200 mW cm−2 ) was used as the exciting light source,
and a DPSS laser (475 nm) was used to monitor the surface height. For PD, the
transmittance of the He–Ne laser light passing through the sample was used;
for PVC, the photocurrent induced by the He–Ne exciting light was used. The
measurements were performed at 300 K in air (Nakagawa et al., 2010).
Figures 5.17(a) and (b) show the time evolution of the changes in N(t)N0 ,
α(t), and the surface height h(t) for thicknesses d = 700 nm and d = 200 nm
in a-As2 Se3 . Note that the optical penetration depth for a-As2 Se3 is α −1 =
600 nm at λ = 633 nm, and hence 200 nm is significantly lower than α −1 . This
means that for d = 200 nm, the illumination is almost homogeneous throughout
the sample.
All changes are empirically represented by the following function:
! "
y = A 1 − exp − (t /τ )β , (5.15)
where A is a constant (equal to the total change), and τ and β have the same
meanings as described following eqn. (5.13). The fitting of eqn. (5.15) to the
experimental data (thin solid curves in Figures 5.17(a) and (b)) produces the
important physical parameters τ and β, which are indicated in Table 5.1.
In eqn. (5.15), the function with β < 1.0 is called the stretched exponential
function (SEF). As indicated in Table 5.1, the dynamics of PDC (N(t)N0 ) are
clearly represented by a SEF, although the origin of the SEF is not clear. On the
other hand, the dynamics of PVE ( h) are given by the exponential function
(β = 1.0). However, note that for the dynamics of PD ( α) when d = 700 nm,
β is 0.95, which is a lower value than that for d = 200 nm. We know that
PDC and PVE always follow stretched exponential and exponential dynamics,
5.4 In situ measurements 137
Table 5.1. Physical parameters τ and β

obtained for N(t)N0 , α, and h for
each film thickness.
d = 200 nm d = 700 nm
τ (s) β τ (s) β
N(t)N0 18 0.55 40 0.55
h 70 1.0 90 1.0
α 200 1.0 180 0.95
From Shimakawa, Nakagawa, and Itoh (2009).
PDC β<1
PVE β=1
PD β<1, β=1
Time
Figure 5.18. Comparison of the time scales in the evolutions of PDC, PVE,
and PD, together with the parameter β. (From Nakagawa et al. (2010). Phys.
Status Solidi C, 7, 857. Copyright 2013 by John Wiley and Sons Inc.)
respectively, whereas the PD dynamics are expected to be dependent on the film

thickness. The τ value is lowest for PDC and is highest for PD compared with
the τ value for each phenomenon. These features are summarized in Figure 5.18.
It is suggested that there is no direct one-to-one correlation among PDC, PVE,
and PD. Thickness-dependent values of β and τ for PD and the origin of SFE
for PD will be discussed later in this section using simulations.
Let us discuss the models that account for the experimental results so far
observed. There are principally three models to account for the occurrence of
PD: (1) the change in atomic positions (Tanaka, 1983), (2) bond breaking and/or
bond alternation (Elliott, 1986; Kolobov et al., 1997; Hegedus et al., 2005), and
(3) the expansion and slip motions of charged layers (Shimakawa et al., 1998).
This last model can be used to interpret current results of in situ simultaneous
measurements in a unified way, and we discuss this now.
As shown in Figure 5.19, clustered layers, for example As2 Se3 , become nega-
tively charged (the holes diffuse away to the unilluminated region), and hence a
repulsive interlayer Coulombic interaction produces an increase in the interlayer
distance (volume expansion). This process is indicated by the “Expansion”
arrow. A slip motion along the layers, indicated by the “Slip” arrow, should also
take place following the expansion motion. This slip motion is known to cause
Figure 5.19. Schematic of photoinduced effects in layered chalcogenides.
an increase in lone-pair interactions between interlayers, leading to PD. The

PDC corresponds to bond breaking, which occurs principally in intralayers
(for example, the bond breaking between As and Se in Figure 5.19), which is
discussed in Section 5.3. The intralayer bond breaking may not have a significant
role for PD or PVE. However, there are some suggestions that a correlation
between PDC and PD exists (Elliott, 1986; Kolobov et al., 1997; Hegedus et al.,
2005).
Finally, it is of interest to discuss the dynamics of PD, i.e. the time evolution
of PD, since the dynamic function is given by an exponential function (β = 1.0)
in thin films that changes to a stretched exponential function (β < 1.0) in thicker
films. As PD is observed with a change in the optical absorption coefficient (a
change in optical transmittance along the thickness direction), the dynamics of
PD may depend on the thicknesses of the films. If the dynamics changed with
thickness, we would be able to identify the origin of the stretched exponential
function in the PD occurrence and discuss the correlation between PD, PVE, and
PDC. Note that the stretched exponential dynamics for numerous phenomena in
disordered matters has been a perplexing issue for many decades (Palmer et al.,
1984; Shimakawa, 1984, 1985, 1986; Phillips, 1996).
A model calculation for β and τ has been made in a-As2 Se3 by assuming a
series sequence of PD along the thickness direction from the illuminated surface
to the back surface (Shimakawa et al., 2009). We know that the optical absorption
decreases exponentially with the thickness direction toward the non-illuminated
back surface. It is assumed that the reaction rate (the inverse of the response
5.5 Photocrystallization and photoinduced amorphization 139
Figure 5.20. Variations from the model calculation (denoted by ×) in β and τ ,

together with the experimental results (denoted by circles) in a-As2 Se3 films.
(Taken with permission from Shimakawa et al. (2009). Appl. Phys. Lett., 95,
051908. Copyright 2013, American Institute of Physics.)
time) for a PD occurrence is proportional to the number of absorbed photons in

each layer. It is further assumed that in each layer the PD dynamics is dominated
by an exponential function.
Figure 5.20 shows variations from the model calculation (denoted by ×)
in β and τ , together with the experimental results (denoted by circles) in
a-As2 Se3 films. We can see that β and τ obtained from the simulation decreased
and increased with the growing film thickness, respectively, and that they repli-
cate the experimental results well. It is thus concluded that the dynamic response
of the PD itself essentially follows the exponential function (non-dispersive),
and the stretched exponential dynamics occur only in thicker films, due to a
series sequence of PD occurrences layer by layer. Thus, a stretched dynamic
function for PD is not an essential feature here. Instead, as already discussed in
Section 5.3, the dynamics represented by the stretched exponential function is
essential to the nature of PDC processes. The origin of the stretched exponential
function found only in PDC has been discussed recently in the literature (Freitas,
Shimakawa, and Wagner, 2014), in which a special distribution of the reaction
rate in PDC is assumed.
5.5 Photocrystallization and photoinduced amorphization

Unlike the reversible photoinduced changes, photoinduced or photon-assisted
crystallization is an irreversible change. Crystallization sometimes leads to the
damage of optical devices using amorphous states; therefore, we will discuss it
briefly first in this section. We then return to the reversible photoamorphization

observed in some materials. The term “photon assisted” can be used for the
condition when the photon irradiation enhances some effects (for example,
crystallization using thermal energy). Here we use the term photocrystallization
as a photon effect.
In a-Se films, the crystal growth rate increases with the illumination, and
the growth rate is controlled by the hole flux toward amorphous–crystalline
boundaries (Dresner and Stringfellow, 1968). This type of photocrystallization
may be related to the suggestion (in Section 5.1) that holes expand an a-Se net-
work (Hegedus et al., 2005). Inami and Adachi (1999) discovered that oriented
crystallization occurs upon linearly polarized illumination. Photocrystallization
has also been reported in a-GeSe2 (Sakai et al., 2003), Asx Se1−x (Mikla and
Mikhalko, 1995), and AsS (Brazhkin et al., 2007).
A crystallized As50 Se50 film on silica glass substrate becomes amorphous
again when continuous-wave low-intensity light is irradiated. This change is
athermal and is not caused by local melting (and the subsequent quenching)
(Elliott and Kolobov, 1991; Kolobov and Elliott, 1995). The x-ray diffraction
(XRD) patterns of a thermally crystallized and illuminated film clearly show
the difference, and XRD patterns vary markedly with the film thickness, which
implies that the substrate plays a role in the amorphization process. A realgar-
like As4 Se4 molecular structure of the crystallized film is confirmed by the
Raman spectrum. Annealing at a temperature above the glass-transition tempera-
ture results in a different XRD pattern (not the same as that for the originally
crystallized film). It should be noted that this phenomenon seems to be unique
to this composition (As50 Se50 ) (Elliott and Kolobov, 1991).
5.6 Some applications of photoinduced effects

Photoinduced effects are enhanced in the presence of localized plasmon fields,
which can be caused by an increase in the non-equilibrium charge carrier gen-
eration in semiconductors. This has been observed in a-Ch–gold nanoparticle
(GNP) composite structures (Charnovych and Kökényesi, 2011; Charnovych
et al., 2011; Voynarovych et al., 2012). Note also that the localized plasmon
fields enhance the efficiency, for example the performance of solar cells, as well
as the photoinduced structural transformations in amorphous chalcogenides. The
transformation of the GNP plasmon resonance spectrum due to the changes of
complex optical parameters of the surrounding chalcogenide matrix was used
References 141
for a more profound understanding of the photoinduced process in the given

material (Charnovych and Kökényesi, 2011; Charnovych et al., 2011;
Voynarovych et al., 2012).
The photoinduced volume change within the amorphous phase consists of two
components: a rapid but minor one connected with a local structural change, and
a slow one related to a photostimulated mass transport (Kaganovskii et al.,
2011; Trunov et al., 2011), which can also be influenced by plasmon fields
(Charnovych et al., 2013).
The photostimulated mass transport is also a basic effect in chalco-
genide/chalcogenide or metal/chalcogenide nanomultilayers, which results in
an essentially enhanced intermixing of neighboring nanometer-thick layers (for
example, a-Se/As2 S3 , GeS/GeSe, etc.) (Kökényesi, 2006; Naik et al., 2009;
Takats et al., 2009; Charnovych et al., 2013) and provides an exceptionally
interesting basis for direct surface pattern recording. Further development of
photoresist or even electron-beam resist materials based on a one-step recording
process (Cserhati et al., 2012) for the prototype development of optoelectronic
elements, as well as new optical non-linear media, may be expected from these
recent results.
References
Branz, H.M. (1999). Hydrogen collision model: quantitative description of metastability
in amorphous silicon. Phys. Rev. B, 59, 5498–5512.
Brazhkin, V.V., Gavrilyuk, A.G., Lyapin, A.G., Timofeev, Yu.A., Katayama, Y., and
Kohara, S. (2007). AsS: bulk inorganic molecular-based chalcogenide glasses.
Appl. Phys. Lett., 91, 031912, 1–3.
Charnovych, S. and Kökényesi, S. (2011). Plasmon enhanced optical recording in As–Se
layers. Phys. Status Solidi C, 8, 2854–2857.
Charnovych, S., Kökényesi S., Glodan, Gy., and Csik, A. (2011). Enhancement of
photoinduced transformations in amorphous chalcogenide film via surface plasmon
resonances. Thin Solid Films, 519, 4309–4312.
Charnovych, S., Szabó, I.A., Tóth, A., Volk, J., Trunov, M.L., and Kökényesi, S. (2013).
Plasmon assisted photoinduced surface changes in amorphous chalcogenide layer.
J. Non-377Cryst. Solids, 200–204.
Cserhati, C., Charnovych, S., Lytvyn, P.M. et al. (2012). Mechanism of photo induced
mass transfer in amorphous chalcogenide films. Mater. Lett., 66, 159–161.
Dresner, J. and Stringfellow, G.B. (1968). Electronic processes in the photo-
crystallization of vitreous selenium. J. Phys. Chem. Solids, 29, 303–311.
Elliott, S.R. (1986). A unified model for reversible photostructural effects in chalco-
genide glasses. J. Non-Cryst. Solids, 81, 71–98.
Elliott, S.R. and Kolobov, A.V. (1991). Athermal light-induced vitrification of
As50 Se50 films. J. Non-Cryst. Solids, 128, 216–220.
Freitas, R.J., Shimakawa, K., and Wagner, T. (2014). The dynamics of photoinduced
defect creation in amorphous chalcogenides: the origin of the stretched exponential
function. J. Appl. Phys., 115, 013704, 1–4.
Fritzshe, H. (2000). Light-induced effects in glasses. In Insulating and Semiconducting
Glasses, ed. P. Boolchand. Singapore: World Scientific, pp. 653–690.
Ganjoo, A. and Shimakawa, K. (2002). Dynamics of photodarkening in amorphous
chalcogenides. J. Optoelectron. Adv. Mater., 4, 595–604.
Ganjoo, A., Ikeda, Y., and Shimakawa, K. (1999). In situ photoexpansion measurements
of amorphous As2 S3 films: role of photocarriers. Appl. Phys. Lett., 74, 2119–2122.
Ganjoo, A., Shimakawa, K., Kamiya, H., Davis, E.A., and Singh, J. (2000). Percolative
growth of photodarkening in amorphous As2 S3 films. Phys. Rev. B, 62, R14601–
14604.
Ganjoo, A., Shimakawa, K., Kitano, K., and Davis, E.A. (2002). Transient photodark-
ening in amorphous chalcogenides. J. Non-Cryst. Solids, 299–302, 917–923.
Gotoh, T., Nonomura, S., Nishio, M., Nitta, S., Kondo, M., and Matsuda, A. (1998).
Experimental evidence of photoinduced expansion in hydrogenated amorphous
silicon using bending detected optical lever method. Appl. Phys. Lett., 72, 2978–
2980.
Hamanaka, H., Tanaka, K., Matsuda, A., and Iijima, S. (1976). Reversible photo-induced
changes in evaporated As2 S3 and As4 Se5 Ge1 films. Solid State Commun., 19, 499–
501.
Hegedus, J., Kohary, K., Pettifor, D.G., Shimakawa, K., and Kugler, S. (2005). Photo-
induced volume changes in amorphous selenium. Phys. Rev. Lett., 95, 206803, 1–4.
Hegedus, J., Kohary, K., and Kugler, S. (2006). Universal feature of photo-induced
volume changes in chalcogenide glasses. J. Non-Cryst. Solids, 352, 1587–1590.
Ikeda, Y. and Shimakawa, K. (2004). Real-time in situ measurements of photoinduced
volume changes in chalcogenide glasses. J. Non-Cryst. Solids, 338–340, 539–
542.
Inami, T. and Adachi, S. (1999). Structural and optical properties of photocrystallized
Se films. Phys. Rev. B, 60, 8284–8289.
Itoh, N. and Stoneham, A.M. (2001). Materials Modification by Electronic Excitation.
Cambridge: Cambridge University Press.
Kaganovskii, Yu., Beke, D.L., Charnovych, S., Kökényesi, S., and Trunov, M.L. (2011)
Inversion of the direction of photo-induced mass transport in As20 Se80 films: ex-
periment and theory. J. Appl. Phys., 110, 063502, 1–5.
Kökényesi, S. (2006). Amorphous chalcogenide nano-multilayers: research and devel-
opment. J. Optoelectron. Adv. Mater., 8, 2093–2096.
Kolobov, A.V. (ed.) (2003). Photo-Induced Metastability in Amorphous Semiconductors.
Weinheim: Wiley-VCH.
Kolobov, A.V. and Elliott, S.R. (1995). Reversible photo-amorphization of a crystallized
As50 Se50 alloy. Philos. Mag. B, 71, 1–10.
Kolobov, A.V., Oyanagi, H., Tanaka, K., and Tanaka, K. (1997). Structural study of
amorphous selenium by in-situ EXAFS: observation of photoinduced bond alter-
nation. Phys. Rev. B, 55, 726–734.
Kugler, S., Hegedus, J., and Kohary, K. (2007). Modelling of photoinduced changes
in chalcogenide glasses: a-Se and a-As2 Se3 . J. Mater. Sci.: Mater. Electron., 18,
S163–S167.
Kuzukawa, Y., Ganjoo, A., Shimakawa, K., and Ikeda, Y. (1999). Photo-induced struc-
tural changes in obliquely deposited arsenic-based amorphous chalcogenides: a
model for photostructural changes. Philos. Mag. B, 79, 249–256.
References 143
Lukács, R. and Kugler, S. (2011). Photoinduced volume changes of obliquely and flatly
deposited amorphous AsSe films: universal description of the kinetics. Jpn. J. Appl.
Phys., 50, 091401, 1–4.
Lukács, R., Hegedus, S., and Kugler, S. (2009). Microscopic and macroscopic models
of photo-induced volume changes in amorphous selenium. J. Mater. Sci.: Mater.
Electron., 20, S33–S37.
Lukács, R., Veres, M., Shimakawa, K., and Kugler, S. (2010). On photoinduced vol-
ume change in amorphous selenium: quantum chemical calculation and Raman
spectroscopy. J. Appl. Phys., 107, 073517, 1–5.
Mikla, V.I. and Mikhalko, I.P. (1995). Laser-induced structural transformation of
Asx Se1−x thin amorphous films. J. Non-Cryst. Solids, 180, 236–243.
Molina, D., Lomba, E., and Kahl, G. (1999). Tight-binding model of selenium disordered
phases. Phys. Rev. B, 60, 6372–6382.
Morigaki, K. (1999). Physics of Amorphous Semiconductors. London: World Scientific
& Imperial College Press.
Morigaki, K. and Hikita, H. (2007). Modeling of light-induced defect creation in hydro-
genated amorphous silicon. Phys. Rev. B, 76, 085201, 1–17.
2nd edn. Oxford: Oxford University Press.
Naik, R., Adarsh, K.V., Ganesan, R. et al. (2009). X-ray photoelectron spectroscopic
studies on Se/As2 S3 and Sb/As2 S3 nanomultilayered film. J. Non-Cryst. Solids,
355, 1836–1839.
Nakagawa, N., Shimakawa, K., Itoh, T., and Ikeda, Y. (2010). Dynamics of principal
photoinduced effects in amorphous chalcogenides: in situ simultaneous measure-
ments of photodarkening, volume changes, and defect creation. Phys. Status Solidi
C, 7, 857–860.
Palmer, R.G., Stein, D.L., Abraham, E., and Anderson, P.W. (1984). Models of hier-
archically constrained dynamics for glassy relaxation. Phys. Rev. Lett., 53, 958–
961.
Pfeifer, G., Paesler, M.A., and Agarwal, S.C. (1991). Reversible photodarkening of
amorphous arsenic chalcogens. J. Non-Cryst. Solids, 130, 111–143.
Phillips, J.C. (1996). Stretched exponential relaxation in molecular and electronic
glasses. Rep. Prog. Phys., 59, 1133–1207.
Redfield, D. and Bube, R.H. (1996). Photoinduced Defects in Semiconductors. New
York: Cambridge University Press.
Sakai, K., Maeda, K., Yokoyama, H., and Ikari, T. (2003). Photo-enhanced crystallization
by laser irradiation and thermal annealing in amorphous GeSe2 . J. Non-Cryst.
Solids, 320, 223–230.
Shimakawa, K. (1984). Origin of nonsymmetric dielectric relaxation in dipolar materials.
Appl. Phys. Lett., 45, 587–588.
Shimakawa, K. (1985). Exciton recombination in amorphous chalcogenides. Phys. Rev.
B, 31, 4012–4014.
Shimakawa, K. (1986). Persistent photocurrent in amorphous chalcogenides. Phys. Rev.
B, 34, 8703–8708.
Shimakawa, K., Inami, S., and Elliott, S.R. (1990). Reversible photoinduced change
of photoconductivity in amorphous chalcogenide films. Phys. Rev. B, 42, 11857–
11861.
Shimakawa, K., Inami, S., Kato, T., and Elliott, S.R. (1992). Origin of photoin-
duced metastable defects in amorphous chalcogenides. Phys. Rev. B, 46, 10062–
10069.
588.
Shimakawa, K., Yoshida, N., Ganjoo, A., Kuzukawa, Y., and Singh, J. (1998). A model
for the photostructural changes in amorphous chalcogenides. Philos. Mag. Lett.,
77, 153–158.
Shimakawa, K., Ikeda, Y., and Kugler, S. (2004a). Fundamental optoelectronic processes
in amorphous chalcogenides. In Optoelectronic Materials and Devices, Vol.1, eds.
G. Lucovsky and M. Popescu. Bucharest: INOE Publishing House, p. 103.
Shimakawa, K., Mehern-Nessa, Ishida, H., and Ganjoo, A. (2004b). Quantum efficiency
of light-induced defect creation in hydrogenated amorphous silicon and amorphous
As2 Se3 . Philos. Mag. B, 84, 81–89.
Shimakawa, K., Nakagawa, N., and Itoh, T. (2009). The origin of stretched exponential
function in dynamic response of photodarkening in amorphous chalcogenides.
Appl. Phys. Lett., 95, 051908, 1–3.
Shluger, A. and Stefanovich, E. (1990). Models of the self-trapped exciton and nearest-
neighbor defect pair in SiO2 . Phys. Rev. B, 42, 9664–9673.
Singh, J. and Shimakawa, K. (2003). Advances in Amorphous Semiconductors. New
York and London: Taylor and Francis.
Song, K.S. and Williams, R.T. (1996). Self-Trapped Excitons, 2nd edn. Berlin: Springer.
Staebler, D.L. and Wronski, C.R. (1977). Reversible conductivity changes in discharge-
produced amorphous Si. Appl. Phys. Lett., 31, 292–294.
sity Press.
Takats, V., Miller, F., Jain, H., Cserhati, C., and Kökényesi, S. (2009). Direct surface
patterning of homogeneous and nanostructured chalcogenide layers. Phys. Status
Solidi C, 6(suppl. 1), S83–S85.
Tanaka, K. (1983). Mechanisms of photodarkening in amorphous chalcogenides.
J. Non-Cryst. Solids, 59–60, 925–928.
Tanaka, K. (1990). Photoinduced structural changes in chalcogenide glasses. Rev. Solid
State Sci., 4, 641–659.
Tanaka, K. (1998). Photoexpansion in As2 S3 glass. Phys. Rev. B, 57, 5163–5167.
Trunov, M.L., Lytvyn, P.M., Yannopoulos, S.N., Szabo, I.A., and Kökényesi, S. (2011).
Photoinduced mass-transport based holographic recording of surface relief gratings
in amorphous selenium films. Appl. Phys. Lett., 99, 051906, 1–3.
Uchino, T., Takahashi, T., Ichii, M., and Yoko, T. (2002). Microscopic model of photoin-
duced and pressure induced UV spectra changes in germanosilicate glass. Phys.
Rev. B, 65, 172202, 1–4.
Voynarovych, I., Kökényesi, S., Yurkovich, N., Charnovych, S., and Dmitruk, N. (2012).
Plasmon-assisted transformations in metal-amorphous chalcogenide light-sensitive
nanostructures. Plasmonics, 7, 341–345.
Index
8 − N rule 10, 20, 86, 96 continuous random network 50, 94

cook and quenched 71, 120
ab initio 27, 43, 48, 51, 76, 98, 122 coordination defects 72, 94, 95, 100
a-C (amorphous carbon) 17, 25, 36, 37, 38, 51, 56, Coulomb interaction 43, 121, 137
60, 65, 72, 89, 98, 100, 101
a-C:H (hydrogenated amorphous carbon) 16, 17, 72 dangling bond 4, 10, 40, 88, 94, 131
a-Ch (amorphous chalcogenide) 3, 140 dc conductivity 110
ac transport 5 Debye-type relaxation 126
a-Ge (amorphous germanium) 4, 9, 38, 40, 44, 50, deep levels 102
51, 60 deep-level transient spectroscopy (DLTS)
amorphization 19, 20, 64, 77, 139, 140 102
a-Se (amorphous selenium) 1, 40 defect spectroscopy 101, 133
a-Si (amorphous silicon) 4, 99 density-functional 4, 43, 44, 48, 49, 68, 96
a-Si:H (hydrogenated amorphous silicon) 2, 135 density of states (DOS) 4, 106
avalanche 2, 15 dielectric relaxation time 110
average coordination 4, 20, 21, 22, 28, 31, 67 digital versatile disk (DVD) 2, 16, 73
dihedral angle 39
band tail 4, 92, 111 dispersive transport 5
bond angle 4, 90 distant pair recombination 111
bond breaking 121, 126, 131, 132, 137, 138 doping 94, 97, 98, 99
bond energy 31, 32 drift mobility 5
bond length 38, 121
bond strength 43 effective mass 91, 98
Boson peak 30 electron nuclear double resonance (ENDOR) 104
electron–phonon interaction 95
Cambridge Structural Database (CSD) 53, 57, electron spin resonance (ESR) 88, 94, 95, 103, 104,
60 131, 132
capacitance 102, 103 electrophotography (xerography) 1, 103
carrier generation 140 evaporation 14, 15, 16, 18, 68
carrier–phonon interaction 132 excited electron 121, 124
chain structure 86 extended states 101, 102
charged defect 103, 104, 132 extended x-ray absorption fine structure (EXAFS)
charge distribution 24, 96 spectroscopy 41
chemical vapor deposition (CVD) 14, 18
conduction band (CB) 91, 93, 103, 104, 105, 106, Fermi level 101, 102, 103, 112
107 field-effect measurements 102
constant photocurrent method (CPM) 103 field effect transistors 3
constraint 4, 60 fractal structure 76
145
146 Index
generation rate 124, 126 phase-change random-access memory (PRAM) 2,

giant photoexpansion 5 73
glass transition 8, 18, 19, 20, 30, 32, 71 photobleaching (PB) 129
glass transition temperature 9, 18, 29, 30, 33, 117, photoconductivity 1, 6, 107, 132
140 photodarkening (PD) 5, 117, 118, 129, 130, 135,
glow discharge 14, 16, 40, 97 136, 137, 138, 139
photodegradation 6
heat capacity 9, 27, 28, 29, 33 photoinduced absorption (PA) 101
highest occupied molecular orbital (HOMO) 120, photoinduced defect creation (PDC) 128, 139
121, 123 photoinduced volume contraction (PVC) 118, 136
hole creation 121, 123 photoinduced volume expansion (PVE) 117, 118,
hopping 5, 48, 99, 103, 108, 110 128, 135, 136, 137, 138
photoluminescence (PL) 4, 110, 111
impurity 6, 93, 96, 97, 98 photon absorption 110, 121, 124
intimate pair 132 photothermal deflection spectroscopy (PDS) 101,
ionic transport 76, 112, 113 102
irreversible change 117, 126, 139 photovoltaic (PV) 2, 3, 17
isothermal annealing 74 plasma-enhanced chemical vapor deposition
(PECVD) 14, 16, 17, 18
Keating potential 42, 43, 45, 50, 60 polarons 5
potential fluctuations 108, 110
lattice vibration 27, 28, 91 primary photoconductivity 107
light-induced electron spin resonance (LESR) 104,
132, 133 quadrature frequency-resolved spectroscopy
linear combination of atomic orbitals (LCAO) 86, (QFRS) 110, 111
87 quantum efficiency 133, 134
liquid crystal display 3
long-range order 4, 7, 90, 91, 94 radial distribution function (RDF) 19, 72
long-range translational order 8, 33 radiative recombination 132
lone-pair (LP) 4, 90, 93, 106, 138 Raman spectrum 122, 123, 140
lowest unoccupied molecular orbital (LUMO) 120, reactive sputtering 16
121 refractive index 100, 101, 129
ring statistic 67, 68
magnetron sputtering 16
mammography 2 second nearest-neighbor 20, 72
medium-range order 8, 37 self-trapped exciton (STE) 132
melting temperature 18, 19, 30, 73 short-range order 8, 40
melt-quenching 14 solar cells 3, 140
metastable 118, 126, 130, 132 space-charge-limited current (SCLC) 103
Meyer–Neldel (MN) 5, 76, 112, 113 specific heat 29
mobility 3, 5 Staebler–Wronski effect 6, 130
mobility edge 4, 92 stretched exponential function 126, 128, 136, 138,
139
nearest-neighbor 37, 40, 48, 51, 72, 87
negative-U 95, 96, 132 tail states 4, 92, 106, 110
neutron diffraction 35, 37, 38, 39, 40, 53, 56, 60, 68, Tauc relation 106
100 thin-film transistors (TFTs) 3, 17
non-radiative recombination 131, 132 time-of-flight (TOF) 5, 107
nuclear magnetic resonance (NMR) 24, 41, 104
Urbach tail 106
optical absorption 4, 138
optical memory 16 valence-alternation pairs (VAPs) 95, 96
variable-range hopping (VRH) 5
pair correlation 33, 60 Verlet algorithm 62, 63, 64, 120
penetration depth 136 vidicon 2, 15
percolation path 108 viscosity 32, 33
phase-change material 16, 19, 29, 73, 78 Vogel–Tamman–Fulcher equation 33
Index 147
wrong bond 100 x-ray diffraction 35, 36, 39, 140

x-ray image detector 15
xerography 1, 2, 103 x-ray image sensor 2
x-ray absorption near-edge structure
77 Zeeman splitting 77

Amorphous Semiconductors - Kugler and Shimakawa

Uploaded by

Copyright:

Available Formats

Amorphous Semiconductors - Kugler and Shimakawa

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

Amorphous Semiconductors - Kugler and Shimakawa

Uploaded by

Copyright:

Available Formats

AMORPHOUS SEMI CON DU CTORS

Understanding the structural unit of crystalline solids is vital in determining

s ándor kugler is an Associate Professor at Budapest University of Tech-

koichi shimakawa is a Professor at the University of Pardubice, Czech

Cambridge University Press is part of the University of Cambridge.

To my wife, Akiko, and our children, Tetsuro, Shoko, and Ryo

5 Photoinduced phenomena 117

Understanding the structural unit of crystalline solids is vital in determining their

This is followed by a discussion of preparation techniques in Chapter 2. As

KS wishes to thank his colleagues at Gifu University, Shuichi Nonomura and

We present a brief historical overview of amorphous semiconductors including

1.1 Historical overview: science and applications

materials such as amorphous germanium (a-Ge) and a-Si, a hand-built random

photoexpansion (Hisakuni and Tanaka, 1995) and photodeformations (Tanaka

Table 1.1. Shear viscosity values for some materials.

Substance Temperature (°C) Shear viscosity (Pa s)

Amorphous, glassy materials

supercooling of a liquid below its normal freezing point. The freezing of

Groups IV and VI elements and their alloys

ρ (T ) = ρ0 exp (ε0 /kB T ) , (1.1)

Selényi, P. (1936). Electrostatic recording (electronography). Electronics, Apr., 44–

Amorphous semiconductors are prepared via non-equilibrium processes. Many

2.1 Growth of thin-film forms

deposition (PECVD), and hot-wire chemical vapor deposition (HWCVD), which

Electron-beam and arc evaporation

Plasma-enhanced chemical vapor deposition

Hot-wire chemical vapor deposition

Pulsed laser deposition

2.2 Melt-quenched glasses

a glassy material is heated, crystallization occurs at a temperature Tc between

2.3 Other techniques

Mirabella, and Grimaldi, 2011). It is known that the Ge surface undergoes a

2.4 The Phillips constraint theory

Z = xNc (A) + (1 − x) Nc (B) , (2.1)

where Nc (A) is the coordination number of atom A. The number of stretched

3 = Z/2 + (2Z − 3) , (2.2)

and we obtain Z = 2.4.

Z (Nd −1) (2Z − Nd )

Then we obtain a general expression for the average coordination number as

2Nc + (Nd −1) Nd

(because ZSi = 4). We must thus solve the following equation:

where the average coordination number is again given by Z = 4x + 2(1 – x). We

A brief summary of the essential differences between amorphous and crystalline

3.1 Differences between amorphous and crystalline semiconductors

Figure 3.1. An elementary triad of atoms, denoted by K, L, and M, forming

qM = 2Adθ and qK = qL = −Adθ , (3.1)

From the semiempirical Hartree–Fock cluster calculation, A = −0.69 milli-

Figure 3.2. Topological model for net atomic charge calculation.

Figure 3.3. Comparison of Si atomic net charges calculated by semiempirical

Based on the Hartree-Fock ab initio calculation, A = −0.45 millielectron/deg

asymmetric double-well potential having two different bottom points. Atoms

F (T , V , N) = U − T S = −kB T ln(cosh( E/2kB T )), (3.10)

For a two-level system,

CV = ( E/2kB T )2 sech2 ( E/2kB T ). (3.12)

At low temperatures (around 1 K), kB T ≈ 10−4 eV, which can be considered as an

As sech is a rapidly decreasing function (∼e−x ), we put infinity instead of Emax

Calorimetric properties (glass transition, crystallization)

will be discussed in Section 3.4 in detail. Above Tc an endothermic reaction again

Figure 3.5. Correlation between glass-transition temperature Tg and the overall

some covalent bonding glasses involving polymers: ln Tg 1.6Z + 2.3, where

where (rij ) is a pair-wise potential between atoms i and j, and f is expressed