REVIEWS OF MODERN PHYSICS, VOLUME 77, OCTOBER 2005

Physics of thin-film ferroelectric oxides

M. Dawber*
DPMC, University of Geneva, CH-1211, Geneva 4, Switzerland

K. M. Rabe†
Department of Physics and Astronomy, Rutgers University, Piscataway, New Jersey 00854-
8019, USA

J. F. Scott‡
Department of Earth Sciences, University of Cambridge, Cambridge CB2 3EQ, United
Kingdom
共Published 17 October 2005兲

This review covers important advances in recent years in the physics of thin-film ferroelectric oxides,
the strongest emphasis being on those aspects particular to ferroelectrics in thin-film form. The
authors introduce the current state of development in the application of ferroelectric thin films for
electronic devices and discuss the physics relevant for the performance and failure of these devices.
Following this the review covers the enormous progress that has been made in the first-principles
computational approach to understanding ferroelectrics. The authors then discuss in detail the
important role that strain plays in determining the properties of epitaxial thin ferroelectric films.
Finally, this review ends with a look at the emerging possibilities for nanoscale ferroelectrics, with
particular emphasis on ferroelectrics in nonconventional nanoscale geometries.

CONTENTS b. Poole-Frenkel 1098

c. Fowler-Nordheim tunneling 1098
d. Space-charge-limited currents 1098
I. Introduction 1084
II. Ferroelectric Electronic Devices 1084 e. Ultrathin films—direct tunneling 1098
A. Ferroelectric memories 1084 f. Grain boundaries 1098
B. Future prospects for nonvolatile ferroelectric C. Device failure 1099
memories 1086 1. Electrical breakdown 1099
C. Ferroelectric field-effect transistors 1087 2. Fatigue 1100
D. Replacement of gate oxides in DRAMs 1088 3. Retention failure 1102
III. Ferroelectric Thin-Film-Device Physics 1089 IV. First Principles 1102
A. Switching 1089 A. Density-functional-theory studies of bulk
1. Ishibashi-Orihara model 1089 ferroelectrics 1102
2. Nucleation models 1089 B. First-principles investigation of ferroelectric thin
3. The scaling of coercive field with thickness 1090 films 1104
4. Mobility of 90° domain walls 1090 1. First-principles methodology for thin films 1105
5. Imaging of domain-wall motion 1090 2. Overview of systems 1107
B. Electrical characterization 1092 3. Studies of individual one-component
1. Standard measurement techniques 1092 systems 1108
a. Hysteresis 1092 a. BaTiO3 1108
b. Current measurements 1093 b. PbTiO3 1110
c. Dielectric permittivity 1093 c. SrBi2Ta2O9 1111
2. Interpretation of dielectric permittivity data 1093 d. SrTiO3 and KTaO3 1111
a. Depletion charge versus intrinsic response 1093 4. Studies of individual heterostructures 1112
b. Domain-wall contributions 1094 5. First-principles modeling: methods and
c. Dielectric measurements of phase transitions 1094 lessons 1113
3. Schottky barrier formation at 6. Challenges for first-principles modeling 1115
metal-ferroelectric junctions 1095 V. Strain Effects 1116
4. Conduction mechanisms 1097 VI. Nanoscale Ferroelectrics 1121
a. Schottky injection 1097 A. Quantum confinement energies 1121
B. Coercive fields in nanodevices 1121
C. Self-patterned nanoscale ferroelectrics 1121
*Electronic address: [email protected] D. Nonplanar geometries: ferroelectric nanotubes 1122
† VII. Conclusions 1124
Electronic address: [email protected]
‡
Electronic address: [email protected] References 1124

0034-6861/2005/77共4兲/1083共48兲/$50.00 1083 ©2005 The American Physical Society

1084 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

I. INTRODUCTION trics. All ferroelectrics are grown on substrates which

can impose considerable strains, meaning that properties
The aim of this review is to provide an account of the of ferroelectric thin films can often be considerably dif-
progress made in the understanding of the physics of ferent from those of their bulk parent material. The
ferroelectric thin-film oxides, particularly the physics rel- electronic properties also have a characteristic behavior
evant to present and future technology that exploits the in thin-film form. While bulk ferroelectric materials are
characteristic properties of ferroelectrics. An overview traditionally treated as good insulators, as films become
of the current state of ferroelectric devices is followed thinner it becomes more appropriate to treat them as
by identification and discussion of the key physics issues semiconductors with a fairly large band gap. These ob-
that determine device performance. Since technologi- servations are key to understanding the potential and
cally relevant films for ferroelectric memories are typi- the performance of ferroelectric devices, and to under-
cally thicker than 120 nm, characterization and analysis standing why they fail when they do.
of these properties can initially be carried out at compa- In parallel with the technological developments in the
rable length scales. However, for a deeper understand- field, the power of computational electronic structure
ing, as well as for the investigation of the behavior of theory has increased dramatically, giving us new ways of
ultrathin films with thickness on the order of lattice con- understanding ferroelectricity. Over the last 15 years,
stants, it is appropriate to redevelop the analysis at the more and more complex systems have been simulated
level of atomic and electronic structure. Thus, the sec- with more accuracy; and as the length scales of experi-
ond half of this review is devoted to a description of the mental systems decrease, there is now an overlap in size
state of the art in first-principles theoretical investiga- between the thinnest epitaxial films and the simulated
tions of ferroelectric-oxide thin films, concluding with a systems. It is therefore an appropriate and exciting time
discussion of experiment and theory of nanoscale ferro- to review this work, and to make connections between it
electric systems. and the problems considered by experimentalists and
As a starting point for the discussion, it is helpful to engineers.
have a clear definition of ferroelectricity appropriate to Finally, we look at some issues and ideas in nanoscale
thin films and nanoscale systems. Here we consider a ferroelectrics, with particular emphasis on new geom-
ferroelectric to be a pyroelectric material with two or etries for ferroelectric materials on the nanoscale such
more stable states of different nonzero polarization. Un- as ferroelectric nanotubes and self-patterned arrays of
like electrets, ferroelectrics have polarization states that ferroelectric nanocrystals.
are thermodynamically stable, not metastable. Further- We do not attempt to cover some of the issues which
more, it must be possible to switch between the two are of great importance but instead refer readers to re-
states by the application of a sufficiently strong electric views by other authors. Some of the more important
field, the threshold field being designated the coercive applications for ferroelectrics make use of their piezo-
field. This field must be less than the breakdown field of electric properties, for example, in actuators and mi-
the material, or the material is merely pyroelectric and crosensors; this topic has been reviewed by Muralt
not ferroelectric. Because of this switchability of the 共2000兲. Relaxor ferroelectrics in which ferroelectric or-
spontaneous polarization, the relationship between the dering occurs through the interaction of polar nano-
electric displacement D and the electric field E is hyster- domains induced by substitution are also of great inter-
etic. est for a number of applications and have recently been
For thin-film ferroelectrics the high fields that must be reviewed by Samara 共2003兲.
applied to switch the polarization state can be achieved
with low voltages, making them suitable for integrated II. FERROELECTRIC ELECTRONIC DEVICES
electronics applications. The ability to create high-
density arrays of capacitors based on thin ferroelectric A. Ferroelectric memories
films has spawned an industry dedicated to the commer-
cialization of ferroelectric computer memories. The clas- The idea that electronic information can be stored in
sic textbooks on ferroelectricity 共Fatuzzo and Merz, the electrical polarization state of a ferroelectric mate-
1967; Lines and Glass, 1967兲 though good, are now over rial is a fairly obvious one; however, its realization is not
20 years old, and predate the shift in emphasis from bulk so straightforward. The initial barrier to the develop-
ceramics and single crystals towards thin-film ferroelec- ment of ferroelectric memories was the necessity of
trics. While much of the physics required to understand making them extremely thin films because the coercive
thin-film ferroelectrics can be developed from the under- voltage of ferroelectric materials is typically of the order
standing of bulk ferroelectrics, there is also behavior of several kV/ cm, requiring submicron thick films to
specific to thin films that cannot be readily understood in make devices that work on the voltage scale required for
this way. This is the focus of the present review. computing 共all Si devices work at 艋5 V兲. With today’s
One of the points that will become clear is that a deposition techniques this is no longer a problem, and
ferroelectric thin film cannot be considered in isolation, now high-density arrays of nonvolatile ferroelectric
but rather the measured properties reflect the entire sys- memories are commercially available. However, reliabil-
tem of films, interfaces, electrodes, and substrates. We ity remains a key issue. The lack of good device models
also look in detail at the effects of strain on ferroelec- means that the design of ferroelectric memories is ex-

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1085

FIG. 1. 共a兲 1T-1C memory design. When a voltage is applied to both the word and bit line, the memory cell is addressed. Also
shown is the voltage applied to the capacitor and the current output, depending on whether a one or a zero is stored. The current
for the zero state is pure leakage current and by comparison to a reference capacitor can be removed. 共b兲 A 2T-2C memory cell
in which the reference capacitor is part of the memory cell.

pensive and that it is difficult to be able to guarantee An example of a real commercially available memory
that a device will still operate ten years into the future. is the Samsung lead zirconate titanate-based 4 Mbit
Because competing nonvolatile memory technologies 1T-1C ferroelectric memory 共Jung et al., 2004兲. The scan-
exist, ferroelectric memories can succeed only if these ning electron microscopy cross section 共Fig. 2兲 of the
issues are resolved. device gives some indication of the complexity of design
A ferroelectric capacitor, while capable of storing in- involved in a real ferroelectric memory.
formation, is not sufficient for making a nonvolatile Lead zirconate titanate 共PZT兲 has long been the lead-
computer memory. A pass-gate transistor is required so ing material considered for ferroelectric memories,
that a voltage above the coercive voltage is only applied though strontium bismuth tantalate 共SBT兲, a layered
to the capacitor when a voltage is applied to both the perovskite, is also a popular choice due to its superior
word and bit line; this is how one cell is selected from an fatigue resistance and the fact that it is lead-free 共Fig. 3兲.
array of memories. The current measured through a However, it requires higher-temperature processing,
small load resistor in series with the capacitor is com- which creates significant integration problems. Recently
pared to that from a reference cell that is poled in a
progress has been made in optimizing precursors. Until
definite direction. If the capacitor being read is in a dif-
recently the precursors for Sr, Bi, and Ta/ Nb did not
ferent state, the difference in current will be quite large
function optimally in the same temperature range, but
where the displacement current associated with switch-
last year Inorgtech developed Bi共mmp兲3—a 2-methoxy-
ing accounts for the difference. If the capacitor does not
2-propanol propoxide that improves reaction and lowers
switch because it is already in the reference state, the
difference in current between the capacitor being read the processing temperature for SBT, its traditional main
and the reference capacitor is zero. disadvantage compared to PZT. This material also satu-
Most memories use either a 1 transistor–1 capacitor rates the bismuth coordination number at 6. Recently
共1T-1C兲 design or a 2 transistor–2 capacitor 共2T-2C兲 de- several other layered perovskites, for example, bismuth
sign 共Fig. 1兲. The important difference is that the 1T-1C titanate, have also been considered.
design uses a single reference cell for the entire memory
for measuring the state of each bit, whereas in the 2T-2C
there is a reference cell per bit. A 1T-1C design is much
more space effective than a 2T-2C design, but has some
significant problems, most significantly that the refer-
ence capacitor will fatigue much faster than the other
capacitors, and so failure of the device occurs more
quickly. In the 2T-2C design the reference capacitor in
each cell fatigues at the same rate as its corresponding
storage capacitor, leading to better device life. A prob-
lem with these designs is that the read operation is de-
structive, so every time a bit is read it needs to be writ-
ten again. A ferroelectric field-effect transistor, in which
a ferroelectric is used in place of the metal gate on a
field-effect transistor, would both decrease the size of
the memory cell and provide a nondestructive readout;
however, no commercial product has yet been devel-
oped. Current efforts seem to run into serious problems FIG. 2. Cross-sectional SEM image of the Samsung 4 Mbit
with data retention. 1T-1C 3 metal FRAM.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1086 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

FIG. 3. 共a兲 ABO3 cubic perov-

skite structure; 共b兲 strontium
bismuth tantalate 共layered per-
ovskite structure兲.

As well as their applications as ferroelectric random Flash to operate as the silicon logic levels decrease from
access memories 共FRAMs兲, ferroelectric materials have 5 at present to 3.3, 1.1, and 0.5 V in the near future. The
potential use in dynamic random access memories main problem for ferroelectrics is the destructive read
共DRAMs兲 because of their high dielectric constant in operation, which means that each read operation must
the vicinity of the ferroelectric phase transition, a topic be accompanied by a write operation leading to faster
which has been reviewed by Kingon, Maria, and Streif- degradation of the device. The operation principle of
fer 共2000兲. Barium strontium titanate 共BST兲 is one of the MRAMs is that the tunneling current through a thin
leading materials in this respect since by varying the layer sandwiched between two ferromagnetic layers is
composition a transition temperature just below room different depending on whether the ferromagnetic layers
temperature can be achieved, leading to a high dielectric have their magnetization parallel or antiparallel to each
constant over the operating temperature range. other. The information stored in MRAMs can thus be
read nondestructively, but their write operation requires
high power which could be extremely undesirable in
B. Future prospects for nonvolatile ferroelectric memories
high-density applications. We present a summary of the
There are two basic kinds of ferroelectric random ac- current state of development in terms of design rule and
cess memories in production today: 共1兲 the free-standing speed of the two technologies in Table I.
RAMs and 共2兲 fully embedded devices 关usually a CPU, Partly in recognition of the fact that there are distinct
which may be a complementary metal-oxide semicon- advantages for both ferroelectrics and ferromagnets,
ductor electrically erasable programmable read-only there has been a recent flurry of activity in the field of
memory 共CMOS EEPROM兲, the current generation multiferroics, i.e., materials that display both ferroelec-
widely used nonvolatile memory technology, plus a tric and magnetic ordering, the hope being that one
FRAM and an 8-bit microprocessor兴. The former have could develop a material with a strong enough coupling
reached 4 Mbit at both Samsung 共using PZT兲 and Mat- between the two kinds of ordering to create a device
sushita 共using SBT兲. The Samsung device is not yet, as that can be written electrically and read magnetically. In
far as the authors know, in commercial production for general multiferroic materials are somewhat rare, and
real products, but the NEC FRAM is going into full- certainly the conventional ferroelectrics such as PbTiO3
scale production this year in Toyama 共near Kanazawa兲. and BaTiO3 will not display any magnetic behavior as
Fujitsu clearly leads in the actual commercial use of its the Ti-O hybridization required to stabilize the ferro-
embedded FRAMs. The Fujitsu-embedded FRAM is electricity in these compounds will be inhibited by the
that used in the SONY Playstation 2. It consists of 64
partially filled d orbitals that would be required for mag-
Mbit of EEPROM plus 8 kbit of RAM, 128-kbit ROM,
netism 共Hill, 2000兲. However, there are other mecha-
and a 32-kbit FRAM plus security circuit. The device is
nisms for ferroelectricity and in materials where ferro-
manufactured with a 0.5-␮m CMOS process. The ca-
electricity and magnetism coexist there can be coupling
pacitor is 1.6⫻ 1.9 ␮m2 and the cell size is either
27.3 ␮m2 for the 2T-2C design or 12.5 ␮m2 for the 1T- between the two. For example, in BaMnF4 the ferroelec-
1C. tricity is actually responsible for changing the antiferro-
The leading competing technologies in the long term magnetic ordering to a weak canted ferromagnetism
for nonvolatile computer memories are FRAM and 共Fox et al., 1980兲. In addition, the large magnetoelectric
magnetic random access memories 共MRAM兲. These are coupling in these materials causes large dielectric
supposed to replace EEPROMs 共electrically erasable anomalies at the Néel temperature and at the in-plane
programable read-only memories兲 and “Flash” memo- spin-ordering temperature 共Scott, 1977, 1979兲. More re-
ries in devices such as digital cameras. Flash, though cent theoretical and experimental efforts have focused
proving highly commercially successful at the moment, on BiMnO3, BiFeO3 共Seshadri and Hill, 2001; Moreira
is not a long-term technology, suffering from poor long- de Santos et al., 2002; Wang et al., 2003兲 and YMnO3
term endurance and scalability. It will be difficult for 共Fiebig et al., 2002; Van Aken et al., 2004兲.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1087

TABLE I. Some clarification of the numbers presented here is

required. The size of the Fujitsu FRAM memory may seem
small but it is for an actual commercial device in large-scale
use 共in every Playstation 2兲, whereas the others are figures
from internal sampling of unreleased devices that have not
been commercialized. No MRAMs exist in any commercial
device, giving FRAMs a substantial edge in this regard. The
most recent commercial FRAM product actually shipped is a
large-cell-area six-transistor four-capacitor 共6T-4C兲 memory
for smart credit cards and radio-frequency identification tags
共RF-ID兲 and features nondestructive readout 共Masui et al.,
2003兲. A total of 2 ⫻ 108 ferroelectric memories of all types
have been sold industry wide. The Sony MRAM, though small,
has submicron design rules, meaning that in principle a work-
ing device could be scaled up to Mb size.

Design rule Speed FIG. 4. Schematic diagram of an all-perovskite ferroelectric

Company 共feature size兲 共access time兲 FET and measurement circuit. Reprinted with permission
from Mathews et al., 1997. © 1997, AAAS.
MRAMs
NEC/Toshiba 1 Mb
IBM 16 Mb region near the electrode兲 since the ferroelectric makes
Matsushita 4 Mb contact only with a metal 共or metal-oxide兲 electrode. By
Sony 8kb 0.18 ␮m comparison, the ferroelectric gate in an FET contacts
Cypress 256 kb 70 ns the Si substrate directly 共metal-oxide-semiconductor
State of the art 16 Mb 0.09 ␮m 25 ns field-effect-transistor channel兲. Thus it must be buffered
FRAMs from the Si to prevent charge injection. Unfortunately, if
a thin buffer layer of a low-dielectric material such as
Fujitsu 32 kb 100 ns
SiO2 is used, most of the applied voltage will drop across
Samsung 32 Mb 0.18 ␮m 60 ns
the buffer layer and not the ferroelectric gate, making it
Matsushita 4 Mb 60 ns
impossible to switch the gate. As a result, much of the
Laboratory 280 ps ferroelectric FET research has employed buffer layers
with relatively high dielectric constants, or else rather
thick buffer layers, for example, the first BaMnF4 FET
C. Ferroelectric field-effect transistors made at Symetrix 共Scott, 1998兲 used a buffer layer of
⬃40 nm of SiO2. Subsequent studies often used PZT
It has been known for some time that replacing the 共Kalkur et al., 1994兲 although the large remanent polar-
metal gate in a field-effect transistor 共FET兲 by a ferro- ization in this case 共⬃40 ␮C / cm2兲 is actually undesirable
electric could produce a device with nondestructive for a ferroelectric FET gate.
readout in which the polarization of the gate 共+ for “1” As pointed out by Yoon and Ishiwara 共2001兲, the de-
and − for “0”兲 could be sensed simply by monitoring the polarization field in a ferroelectric gate is inevitably gen-
source-drain current magnitude. Thus such a device re- erated when the gate is grounded, and this makes it very
quires no reset operation after each READ and will ex- difficult to obtain ⬎10 year data retention in an FE-FET.
perience very little fatigue in a normal frequent-read, Their solution is to utilize a 1T-2C capacitor geometry in
occasional-write usage. The first mention of the idea of a which this depolarization field is suppressed by poling
ferroelectric FET is in the U.S. patent of Ross 共1957兲 the two capacitors in opposite directions. With this
and the first realization was by Moll and Tarui 共1963兲, scheme Ishiwara and his colleagues achieved an on/off
while the first attempt to fabricate one on silicon was by source-drain current ratio of ⬎1000 for a 150-nm-thick
Wu 共1974兲. The early ferroelectric FETs utilized gates of SBT film in a 5 ⫻ 50 ␮m2 metal-oxide-semiconductor
lithium niobate 共Rice University; Rabson et al., 1995兲 or field-effect transistor channel, with Pt electrodes on the
BaMgF4 共Westinghouse; Sinharoy et al., 1991, 1992, SBT capacitor.
1993兲. An example of a ferroelectric FET device as fab- Note that the direct contact of the ferroelectric onto
ricated by Mathews et al. 共1997兲 is shown in Fig. 4. Si produces a semiconductor junction that is quite differ-
The optimum parameters for such a ferroelectric-gate ent from the metal-dielectric interface discussed above.
material are extremely different from those for pass- The Schottky barrier heights for this case have been cal-
gate-switched capacitor arrays; in particular, the latter culated by Peacock and Robertson 共2002兲. The electron
require a remanent polarization ⬃10 ␮C / cm2, whereas screening length in the Si will be much greater than in
the ferroelectric-gated FETs can function well with 50 the case of metal electrodes; in particular, this will in-
times less 共0.2 ␮C / cm2兲. However, the switched capaci- crease the minimum ferroelectric film thickness required
tor array 共FRAM兲 is very tolerant of surface traps in the to stabilize the device against depolarization instabili-
ferroelectric 共which may be ⬃1020 cm−3 in the interface ties. Although this point was first emphasized by Batra

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1088 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

TABLE II. A number of the most promising gate materials under recent study, together with the
buffer layers employed in each case. Studies of the I共V兲 characteristics of such ferroelectric FETs
have been given by Macleod and Ho 共2001兲 and a disturb-free programming scheme described by
Ullman et al. 共2001兲.

FET gate Buffer layer Reference

LiNbO3 none Rabson et al. 共1995兲

SBT SrTa2O6 Ishiwara 共1993兲,
Ishiwara et al. 共1997兲,
Ishiwara 共2001兲
SBT CeO2 Shimada et al. 共2001兲,
Haneder et al. 共2001兲
SBT SiO2 Okuyuma et al. 共2001兲
SBT ZrO2 Park and Oh 共2001兲
SBT Al2O3 Shin et al. 共2001兲
SBT Si3N4 Han et al. 共2001兲
SBT Si3N4 / SiO2 Sugiyama et al. 共2001兲
SBT poly-Si+ Y2O3 Kalkur and Lindsey 共2001兲
Pb5Ge3O11 none Li and Hsu 共2001兲
YMnO3 Y 2O 3 Cheon et al. 共2001兲,
Choi et al. 共2001兲
Sr2共Ta2xNb2−2x兲O7 none Kato 共2001兲
PZT CeO2 Xiaohua et al. 共2001兲
BST共strained兲 YSZ共zirconia兲 Jun and Lee 共2001兲
BaMnF4 SiO2 Scott 共1998兲,
Kalkur et al. 共1994兲

and Silverman 共1972兲, it has been neglected in the more nearly interchangeable兲 deposited by some form of epi-
recent context of ferroelectric FETs. In our opinion, this taxial growth. This is the technique employed at Mo-
depolarization instability for thin ferroelectric gates on torola, but the view elsewhere is that it is too expensive
FETs is a significant source of the observed retention to become industry process worthy. The second ap-
failure in the devices but has not yet been explicitly proach is to use a material of moderate k 共of order 20兲,
modeled. If we are correct, the retention problem in
with HfO2 favored but ZrO2 also a choice. Hafnium ox-
ferroelectric FETs could be minimized by making the
ferroelectric gates thicker and the Si contacts more con- ide is satisfactory in most respects but has the surprising
ducting 共e.g., p+ rather than p兲. See Scott 共2005兲 for a disadvantage that it often degrades the n-channel mobil-
full discussion of all-perovskite FETs. ity catastrophically 共by as much as 10 000 times兲. Re-
Table II lists a number of the most promising gate cently ST Microelectronics decided to use SrTiO3 but
materials under recent study, together with the buffer with metal-organic chemical-vapor deposition from Aix-
layers employed in each case. Studies of the I共V兲 char- tron, thus combining high dielectric constant and
acteristics of such ferroelectric FETs have been given by cheaper processing.
Macleod and Ho 共2001兲 and a disturb-free programming The specific high-k integration problems are the fol-
scheme described by Ullman et al. 共2001兲. lowing four: 共1兲 depletion effects in the polysilicon gate,
Beyond its use in modulating the current in a semi- 共2兲 interface states, 共3兲 strain effects, and 共4兲 etching dif-
conductor channel the ferroelectric field effect can also
ficulties 共HfO2 is hard to wet etch兲. The use of a poly-Si
be used to modify the properties of more exotic corre-
lated oxide systems 共Ahn, Triscone, and Mannhart, gate instead of a metal gate produces grain-boundary
2003兲. stress in the poly, with resultant poor conductivity. This
mobility degradation is only partly understood. The gen-
eral view is that a stable amorphous HfO2 would be a
D. Replacement of gate oxides in DRAMs
good strain-free solution. Note that HfO2 normally crys-
At present there are three basic approaches to solving tallizes into two or three phases, one of which is mono-
the problem of SiO2-gate oxide replacement for clinic 共Morrison et al., 2003兲. Hurley in Cork has been
DRAMs. The first is to use a high-dielectric 共“high-k”兲 experimenting with a liquid injection system that re-
material such as SrTiO3 共k = 300 is the dielectric con- sembles Isobe’s earlier SONY device for deposition of
stant; ⑀ = k − 1 is the permittivity; for k Ⰷ 1 the terms are viscous precursors with flash evaporation at the target.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1089

C共t,t⬘兲 = CD 冋冕t⬘
t
V共t⬙兲dt⬙ 册 D
, 共1兲

where D is the dimensionality of the growth and CD is a

constant which depends on the dimensionality. It is also
assumed within this model that the nucleation is deter-
ministic and occurs at predefined places; i.e., this is a
model of inhomogeneous nucleation. This is an impor-
tant point since some researchers still use homogeneous
nucleation models. These are completely inappropriate
for ferroelectrics 关where the nucleation is inhomoge-
neous, as is demonstrated by imaging experiments 共Shur,
FIG. 5. The three phases of domain reversal: I, nucleation 1996; Shur et al., 2000; Ganpule et al., 2001兲兴.
共fast兲; II, forward growth 共fast兲; III, sideways growth 共slow兲. The result of the model is that the fraction of switched
charge as a function of field and frequency may be ex-
pressed as
III. FERROELECTRIC THIN-FILM-DEVICE PHYSICS
Q共E,f兲 = 1 − exp关− f−D⌽共E兲兴, 共2兲
We now turn to some of the physics questions which
are relevant to ferroelectric thin-film capacitors. where ⌽共E兲 depends on the wave form used for switch-
ing. After some consideration and the substitution
⌽共E兲 = Ek one obtains a useful relationship for the de-
A. Switching pendence of the field on frequency:
In the ferroelectric phase, ferroelectric materials form Ec = fD/k . 共3兲
domains where the polarization is aligned in the same
direction in an effort to minimize energy. When a field is This relationship has been used to fit data fairly well
applied, the ferroelectric switches by the nucleation of in TGS 共Hashimoto et al., 1994兲, PZT, and SBT 共Scott,
domains and the movement of domain walls and not by 1996兲. More recently, however, Tsurumi et al. 共2001兲 and
the spontaneous reorientation of all of the polarization Jung et al. 共2002兲 have found that over larger frequency
in a domain at once. In contrast to ferromagnets, where ranges the data on several materials is better fitted by
switching usually occurs by the sideways movement of the nucleation-limited model of Du and Chen 共1998a兲.
existing domain walls, ferroelectrics typically switch by Tagantsev et al. 共2002兲 have also found that over large
the generation of many new reverse domains at particu- time ranges the Ishibashi-Orihara model is not a good
lar nucleation sites, which are not random; i.e., nucle- description of switching-current data and that a
ation is inhomogeneous. The initial stage is nucleation of nucleation-limited model is more appropriate. It is quite
opposite domains at the electrode, followed by fast for- possible of course that domain-wall-limited switching
ward propagation of domains across the film, and then 共Ishibashi兲 is operative in one regime of time and field
slower widening of the domains 共Fig. 5兲. In perovskite but that in another regime the switching is nucleation
oxides the final stage of the switching is usually much limited.
slower than the other two stages, as first established by
Merz 共1954兲. In other materials nucleation can be the
2. Nucleation models
slowest 共rate-limiting兲 step.
Some of the earliest detailed studies of switching in
1. Ishibashi-Orihara model ferroelectrics developed nucleation-limited switching
models where the shape of the nucleus of the reversed
For many years the standard model to describe this domain was very important. In the work of Merz 共1954兲
process has been the Ishibashi-Orihara model 共Orihara and Wieder 共1956, 1957兲 a nucleation-limited model was
et al., 1994兲 based on Kolomogorov-Avrami growth ki- used in which when dagger-shaped nuclei were assumed,
netics. In this model one considers a nucleus formed at the correct dependence of the switching current on elec-
time t⬘ and then a domain propagating outwards from it tric field could be derived. This approach leads to the
with velocity V. In the Ishibashi-Orihara model the ve- concept of an activation field for nucleation 共somewhat
locity is assumed to be dependent only on the electric different from the coercive field兲. Activation fields in
field E, and not on the domain radius r共t兲. This makes thin-film PZT capacitors were measured by Scott et al.
the problem analytically tractable but gives rise to un- 共1988兲; very recently, Jung et al. 共2004兲 have studied the
physical fitting parameters, such as fractional dimension- effects of microgeometry on the activation field in PZT
ality D. The fractional D is not related to fractals. It is capacitors.
an artifact that arises because the domain-wall velocity The switching model of Tagantsev et al. 共2002兲 is a
V is actually proportional to 1 / r共t兲 for each domain and different approach in which a number of noninteracting
is not a constant at constant E. The volume of a domain elementary switching regions are considered. These
at time t is given by switch according to a broad distribution of waiting times.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1090 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

wide widths 共hundreds of angstroms兲 and immobile

walls. Some recent papers show experimentally that 90°
domain walls in perovskite ferroelectrics are extremely
narrow 共Tsai et al., 1992; Stemmer et al., 1995; Floquet et
al., 1997; Foeth et al., 1999兲. In PbTiO3 they are
1.0± 0.3 nm wide. This connects the general question of
how wide they are and whether they are immobile. The
review by Floquet and Valot 共1999兲 is quite good. They
make the point that in ceramics these 90° walls are 14.0
nm wide 共an order of magnitude wider than in single
crystals兲. This could be why theory and experiment dis-
agree, i.e., that something special in the ceramics makes
them 10–15 times wider 共and less mobile?兲. The latter
point is demonstrated clearly in experiments on KNbO3,
together with a theoretical model that explains geo-
metrical pinning in polyaxial ferroelectrics in terms of
electrostatic forces. In this respect the first-principles
study of Meyer and Vanderbilt 共2002兲 is extremely inter-
esting. Not only do they show that 90° domain walls in
PbTiO3 are narrow and form much more easily than
180° domain walls, but that they should be much more
mobile as well, the barrier for motion being so low they
predict thermal fluctuation of about 12 unit cells at room
temperature, which could perhaps explain why they ap-
pear to be wide.
Some experimental studies using atomic force micros-
copy have attempted to answer the question of whether
90° domain walls were mobile or not. In certain circum-
FIG. 6. 共Color兲 The scaling of coercive field with thickness in
stances they were immobile 共Ganpule, Nagarajan, Li, et
ferroelectrics; from mm to nm scale. From Dawber, Chandra,
al., 2000; Ganpule, Nagarajan, Ogale, et al., 2000兲, but in
et al., 2003. The bottom is the three sets of data from the upper
all normalized to the same value at 10−6.5 m.
another study 共Nagarajan et al., 2003兲 the motion of 90°
domain walls under an applied field was directly ob-
served. It seems that in principle 90° domain walls can
3. The scaling of coercive field with thickness move, but this depends quite strongly on the sample
For the last 40 years the semiempirical scaling law conditions. Recently Shilo et al. 共2004兲 and Salje and Lee
共Janovec, 1958; Kay and Dunn, 1962兲, Ec共d兲 ⬀ d−2/3, has 共2004兲 have shown that domain wall widths are not a
been used successfully to describe the thickness depen- characteristic of the material per se but vary greatly with
dence of the coercive field in ferroelectric films ranging location in the sample due to nearby impurities. This
from 100 ␮m to 200 nm 共Scott, 2000b兲. In the ultrathin reconciles the diverse values reported.
PVDF films of Bune et al. 共1998兲 a deviation from this
relationship was seen for the thinnest films 共Ducharme
et al., 2000兲. Although they attribute this to a new kind 5. Imaging of domain-wall motion
of switching taking place 共simultaneous reversal of po-
larization, as opposed to nucleation and growth of do- The direct imaging of ferroelectric domain walls is an
mains兲, Dawber, Chandra, et al. 共2003兲 have shown, to excellent method for understanding domain-wall motion
the contrary, that if the effects of a finite depolarization and switching. At first this was carried out in materials
field due to incomplete screening in the electrode are where the domains were optically distinct such as lead
taken into account, then the scaling law holds over six germanate 共Shur et al., 1990兲, but more recently atomic
decades of thickness and the coercive field does not de- force microscopy has become a powerful tool for ob-
viate from the value predicted by the scaling law 共Fig. 6兲. serving domain-wall motion. The polarization at a point
Recently, Pertsev et al. 共2003兲 measured coercive fields can be obtained from the piezoresponse detected by the
in very thin PZT films. Although they have used a dif- tip, and the tip itself can be used to apply a field to the
ferent model to explain their data, it can be seen that in ferroelectric sample and initiate switching. It is thus pos-
fact the scaling law describes the data very well. sible to begin switching events and watch their evolution
over time. Atomic force microscopy domain writing of
ferroelectric domains can also be used to write ex-
4. Mobility of 90° domain walls
tremely small domain structures in high-density arrays
The mobility of domain walls, especially 90° walls, de- 共Paruch et al., 2001兲 or other devicelike structures, such
pends upon their width. In this respect the question has as surface acoustic wave devices 共Sarin Kumar et al.,
been controversial, with some authors claiming very 2004兲.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1091

FIG. 7. Piezoresponse scans of a single cell in

PbZr0.2Ti0.8O3. 共b兲–共d兲 The spontaneous re-
versal of polarization within this region after
wait times of 共b兲 1.01⫻ 103, 共c兲 1.08⫻ 105, 共d兲
1.61⫻ 105, and 共e兲 2.55⫻ 105 s. Faceting can
be seen in 共c兲, 共d兲, and 共e兲. 共f兲 Transformation-
time curve for the data in 共b兲–共e兲. From Gan-
pule et al., 2001.

The backswitching studies of Ganpule et al. 共2001兲 fatigue process due to their ordering 共Park and Chadi,
show two very interesting effects 共Fig. 7兲. The first is the 1998; Scott and Dawber, 2000兲.
finding that reverse domains nucleate preferentially at An unsolved puzzle is the direct observation via
antiphase boundaries. This was studied in more detail atomic force microscopy of domain walls penetrating
subsequently by Roelofs et al. 共2002兲 who invoked a grain boundaries 共Gruverman et al., 1997兲. This is con-
depolarization-field-mediated mechanism to explain the trary to some expectations and always occurs at non-
result. Another explanation might be that the strain is normal incidence, i.e., at a small angle to the grain
relaxed at these antiphase boundaries resulting in favor- boundary.
able conditions for nucleation. Second, the influence of One of the interesting things to come out of the work
curvature on the domain-wall relaxation is accounted for in lead germanate 共where ferroelectric domains are op-
within the Kolomogorov-Avrami framework. The veloc- tically distinct due to electrogyration兲 by Shur et al.
ity of the domains is dependent on curvature, and, as the 共1990兲 is that at high applied electric fields 共15 kV cm−1兲
relaxation proceeds, the velocity decreases and the do- tiny domains are nucleated in front of the moving do-
main walls become increasingly faceted. In Fig. 7 the main wall 共Fig. 9兲. A very similar effect is seen in ferro-
white polarization state is stable, whereas the black is magnets as observed by Randoshkin 共1995兲 in a single-
not. The sample is poled into the black polarization state crystal iron-garnet film.
and then allowed to relax back. However, in ferromagnets the effect is modeled by a
A different kind of study was undertaken by Tybell et spin-wave mechanism 共Khodenkov, 1975兲. This mecha-
al. 共2002兲, in which they applied a voltage pulse to switch nism is based on the gyrotropic model of domain-wall
a region of the ferroelectric using an atomic force mi- motion in uniaxial materials 共Walker, 1963兲. When a
croscopy tip and watched how the reversed domain grew strong magnetic driving field 共exceeding the Walker
as a function of pulse width and amplitude. They were threshold兲 acts upon a domain wall, the magnetization
able to show that the process was well described by a vectors in the domain wall begin to precess with a fre-
creep mechanism thought to arise due to random pin- quency ␥H, where ␥ is the effective gyromagnetic ratio.
ning of domain walls in a disordered system 共Fig. 8兲. By relating the precession frequency in the domain wall
Though the exact origin of the disorder was not clear, it with the spin-wave frequency in the domain, good pre-
suggests that it is connected to oxygen vacancies, which dictions can be made for the threshold fields at which
also play a role in pinning the domain walls during a the effect occurs. We note that the domains nucleated in

FIG. 8. 共a兲 Domain size increases logarithmi-

cally with pulse widths longer than 20 ␮s and
saturates for shorter times as indicated by the
shaded area. 共b兲 Domain-wall speed as a func-
tion of the inverse applied electric field for
290-, 370-, and 810-Å-thick samples. The data
fit well the characteristic velocity-field rela-
tionship of a creep process. From Tybell et al.,
2002.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1092 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

FIG. 9. Nucleation of nanodomains in front of domain wall in FIG. 10. 共a兲 The original Sawyer-Tower circuit, 共b兲 hysteresis
lead germanate at high electric field. Black and white represent in Rochelle salt measured using this circuit by Sawyer and
the two directions of polarization. From Gruverman, 1990. Tower at various temperatures. From Sawyer and Tower, 1930.

front of the wall may be considered as vortexlike Skyr- carried out using commercial apparatus from one of two
mions. The similarity between these effects is thus quite companies, Radiant Technologies and AixAcct. Both
surprising and suggests that perhaps there is more in companies’ testers can carry out a number of tests and
common between ferroelectric domain-wall motion and measurements, and both machines use charge or current
ferromagnet domain-wall motion than is usually consid- integration techniques for measuring hysteresis. Both
ered. However, whereas Democritov and Kreines 共1988兲 machines also offer automated measurement of charac-
have shown that magnetic domain walls can be driven teristics such as fatigue and retention.
supersonically 共resulting in a phase-matched Cherenkov- In measuring P共E兲 hysteresis loops several kinds of
like bow wave of acoustic phonon emission兲, there is no artifacts can arise. Some of these are entirely instrumen-
direct evidence of supersonic ferroelectric domains. Pro- tal, and some arise from the effects of conductive 共leaky兲
cesses such as the nanodomain nucleation described specimens.
above seem to occur instead when the phase velocity of
Hysteresis circuits do not measure polarization P di-
the domain-wall motion approaches the speed of sound.
rectly. Rather, they measure switched charge Q. For an
Of course the macroscopic electrical response to switch-
ideal ferroelectric insulator
ing can arrive at a time t ⬍ v / d, where v is the sound
velocity and d the film thickness, simply from domain
nucleation within the interior of the film between cath- Q = 2PrA, 共4兲
ode and anode.
where Pr is the remanent polarization and A is the elec-
trode area for a parallel-plate capacitor. For a somewhat
B. Electrical characterization
conductive sample
1. Standard measurement techniques
Several kinds of electrical measurements are made on Q = 2PrA + ␴Eat, 共5兲
ferroelectric capacitors. We briefly introduce them here
before proceeding to the following sections where we where ␴ is the electrical conductivity, Ea is the applied
discuss in detail the experimental results obtained by field, and t the measuring time. Thus Q in a pulsed mea-
using these techniques. suring system depends on the pulse width.
The four basic types of apparent hysteresis curves that
a. Hysteresis are artifacts are shown in Fig. 11.
One of the key measurements is naturally the mea-
surement of the ferroelectric hysteresis loop. There are
two measurement schemes commonly used. Tradition-
ally a capacitance bridge as first described by Sawyer
and Tower 共1930兲 was used 共Fig. 10兲. Although this is no
longer the standard way of measuring hysteresis, the cir-
cuit is still useful 共and very simple and cheap兲 and we
have made several units which are now in use in the
teaching labs in Cambridge for a demonstration in which
students are able to make and test their own ferroelec-
tric KNO3 capacitor 共Dawber, Farnan, and Scott, 2003兲.
This method is not very suitable in practice for many
reasons, for example, the need to compensate for dielec- FIG. 11. Common hysteresis artifacts: 共a兲 dead short, 共b兲 linear
tric loss and the fact that the film is being continuously lossy dielectric, 共c兲 saturated amplifier, and 共d兲 nonlinear lossy
cycled. Most testing of ferroelectric capacitors is now dielectric.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1093

Figure 11共a兲 is a dead short in a Sawyer-Tower circuit c. Dielectric permittivity

or modern variant and is discussed in the instruction An impedance analyzer measures the real and imagi-
documentation for both the AixAcct1 and Radiant2 nary parts of the impedance by use of a small-amplitude
testers. ac signal which is applied to the sample. The actual mea-
Figure 11共b兲 shows a linear lossy dielectric. The points surement is then made by balancing the impedance of
where the loop crosses Va = 0 are often misinterpreted as the sample with a set of references inside the impedance
Pr values. Actually this curve is a kind of Lissajous fig- analyzer. From this the capacitance and loss can be cal-
ure. It can be rotated out of the page to yield a straight culated 共all this is done automatically by the machine兲. It
line 共linear dielectric response兲. Such a rotation can be is possible at the same time to apply a dc bias to the
done electrically and give a “compensated” curve. Here sample, so the signal is now a small ac ripple superim-
compensation means to compensate the phase shift posed on a dc voltage. Ferroelectric samples display a
caused by dielectric loss. characteristic “butterfly loop” in their capacitance-
Figure 11共c兲 is more subtle. Here are two seemingly voltage relationships, because the capacitance is differ-
perfect square hysteresis loops obtained on the same or ent for increasing and decreasing voltage. The measure-
nominally equivalent specimens at different maximum ment is not exactly equivalent to the hysteresis
fields. The smallest loop was run at an applied voltage of measurement. In a capacitance-voltage measurement a
Va = 10 V, and yields Pr = 30 ␮C cm−2 and the larger at static bias is applied and the capacitance measured at
Va = 50 V and yields Pr = 100 ␮C cm−2. Note that both that bias, whereas in a hysteresis measurement the volt-
curves are fully saturated 共flat tops兲. This is impossible. age is being varied in a continuous fashion. Therefore it
If the dipoles of the ferroelectric are saturated at Pr is not strictly true that C共V兲 = 共d / dV兲P共V兲 共area taken as
= 30 ␮C cm−2 then there are no additional dipoles to unity兲 as claimed in some textbooks, since the frequency
produce Pr = 100 ␮C cm−2 in the larger loop at high volt- is not the same in measurements of C and P. Impedance
age. What actually occurs in the illustration is saturation spectroscopy 共where the frequency of the ac signal is
of the amplifier in the measuring system, not saturation varied兲 is also a powerful tool for analysis of films, espe-
of the polarization in the ferroelectric. The figure is cially as it can give information on the time scales at
taken from Jaffe, Cook, and Jaffe 共1971兲 where this ef- which processes operate. Interpretation of these results
fect is discussed 共p. 39兲. It will be a serious problem if must be undertaken carefully, as artifacts can arise in
many circumstances and even when this is not the case
conductivity is large in Eq. 共5兲. “Large” in this sense is
many elements of the system 共e.g., electrodes, grain
␴ ⬎ 10−6 共⍀ cm兲−1 and “small” is ␴ ⬍ 10−7 共⍀ cm兲−1. This
boundaries, leads, etc.兲 can contribute to the impedance
is probably the source of Pr ⬎ 150 ␮C cm−2 reports in
in complicated ways.
BiFeO3 where ␴ can exceed 10−4 共⍀ cm兲−1.
Finally, Fig. 11 is a nonlinear lossy dielectric. If it is
phase compensated it still resembles real hysteresis. One 2. Interpretation of dielectric permittivity data
can verify whether it is real or an artifact only by varying
a. Depletion charge versus intrinsic response
the measuring frequency. Artifacts due to dielectric loss
are apt to be highly frequency dependent. Figures 11共b兲 Before looking for ferroelectric contributions to a sys-
and 11共d兲 are discussed in Lines and Glass 共1967, p. 104兲. tem’s electrical properties one should make sure there
No data resembling Figs. 11共a兲–11共d兲 should be pub- are not contributions due to the properties of the system
lished as ferroelectric hysteresis. unrelated to ferroelectricity. Although much is some-
times made of the dependence of capacitance on volt-
age, it is worth noting that metal-semiconductor-metal
systems have a characteristic capacitance voltage which
b. Current measurements arises from the response of depletion layers to applied
voltage 共Fig. 12兲.
Another measurement of importance which is carried
Essentially the problem boils down to the fact that
out in an automated way by these machines is the mea-
there are two possible sources of the dependence of ca-
surement of the leakage current. This is normally dis-
pacitance on applied field, either changes in depletion
cussed in terms of a current-voltage 共I-V兲 curve, where
width or changes in the dielectric constant of the mate-
the current is measured at a specified voltage. It is im- rial, i.e.,
portant, however, that sufficient time is allowed for each
measurement step so that the current is in fact true C共E兲 ⑀共E兲
= . 共6兲
steady-state leakage current and not relaxation current. A d共E兲
For this reason current-time 共I-t兲 measurements can also
Several groups have assumed that the change in the
be important 共Dietz and Waser, 1995兲. Relaxation times
in ferroelectric oxides such as barium titanate are typi- capacitance with field C共E兲 comes from change in deple-
cally 1000 s at room temperature. tion width d共E兲 and that ⑀共E兲 is changing negligibly. The
first to suggest this was Evans 共1990兲, who found d
= 20 nm in PZT. Later Sandia claimed that there was no
1
TF Analyzer 2000 FE-module instruction manual. depletion 共d = 0 or d = ⬁兲 共Miller et al., 1990兲. Several au-
2
http://www.ferroelectrictesters.com/html/specs.html#tut thors have assumed that d共E兲 is responsible for C共E兲

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1094 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

FIG. 12. Current-voltage and capacitance-voltage relationship of Pt-Si-Pt punch-through diode. The characteristics are very
similar to those obtained in metal-ferroelectric-metal systems. Reprinted from Sze et al., 1971, with permission from Elsevier.

共Brennan, 1992; Mihara et al., 1992; Sayer et al., 1992; contributions will vary greatly from sample to sample, or
Scott et al., 1992; Hwang, 1998; Hwang et al., 1998兲. even in the same sample under different experimental
In contrast to the approach of explaining these char- conditions.
acteristics using semiconductor models, Basceri et al.
共1997兲 account for their results on the basis of a Landau-
Ginzburg-style expansion of the polarization 共Fig. 13兲. b. Domain-wall contributions
The change in field due to its nonlinearity has also been Below the coercive field there are also contributions
calculated by both Outzourhit et al. 共1995兲 and Dietz et to the permittivity from domain walls, as first pointed
al. 共1997兲. The real problem is that both pictures are out by Fouskova 共1965; Fouskova and Janousek, 1965兲.
feasible. One should not neglect the fact that the mate- In PZT the contributions of domain-wall pinning to the
rials have semiconductor aspects; but at the same time it dielectric permittivity have been studied in detail by
is not unreasonable to expect that the known nonlinear- Damjanovic and Taylor 共Damjanovic, 1997; Taylor and
ity of the dielectric response in these materials should be Damjanovic, 1997, 1998兲, who showed that the subcoer-
expressed in the capacitance-voltage characteristic. cive field contributions of the permittivity were de-
Probably the best approach is to avoid making any con- scribed by a Raleigh law with both reversible and irre-
clusions on the basis of these kinds of measurements versible components, the irreversible component being
alone, as it is quite possible that the relative sizes of the due to domain-wall pinning.

c. Dielectric measurements of phase transitions

One of the most common approaches to measuring
the transition temperature of a ferroelectric material is
naturally to measure the dielectric constant and loss.
However, in thin films there are significant complica-
tions. In bulk the maximum in the dielectric constant is
fairly well correlated with the transition temperature,
but this does not always seem to be the case in thin films.
As pointed out by Vendik and Zubko 共2000兲, a series
capacitor model is required to extract the true transition
temperature, which in the case of BST has been shown
to be independent of thickness 共Lookman et al., 2004兲, in
contrast to the temperature at which the permittivity
maximum occurs, which can depend quite strongly on
FIG. 13. Capacitance vs applied bias for BST thin films. Re- thickness 共Fig. 14兲. The smearing of phase transitions
printed with permission from Basceri et al., 1997. © 1997, AIP. due to a surface effect or a bulk inhomogeneity has re-

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1095

Typically the regions of the capacitor near the interface

are more oxygen deficient than the bulk. Oxygen vacan-
cies act as donor ions, and this means there can be a
transition from n-type behavior at the interface to
p-type behavior in the center of the film as is evident in
the Kelvin probe study of Nowotny and Rekas 共1994兲,
who found that in bulk BaTiO3 with Pt electrodes a
change in work function from 2.5± 0.3 for surfaces to
4.4± 0.4 eV in the bulk of the material. The nature of the
material near the surface is important since it deter-
mines whether a blocking or Ohmic contact is formed. It
has been shown by Dawber and Scott 共2002兲 that the
defect concentration profile as measured by the
capacitance-voltage technique may be explained by a
model of combined bulk and grain-boundary diffusion
of oxygen vacancies during the high-temperature pro-
FIG. 14. Comparison between the apparent Curie temperature cessing of a film.
in BST taken from Curie-Weiss plots of raw data 共empty
Regardless of the p-type or n-type nature of the ma-
circles兲 and intrinsic data after correction for interfacial capaci-
terial, in most oxide ferroelectrics on elemental metal
tance 共solid squares兲 had been performed. The intrinsic Curie
temperature appears to be independent of film thickness. Re-
electrodes the barrier height for electrons is significantly
printed with permission from Lookman et al., 2004. © 2004, less than the barrier height for holes 共Robertson and
AIP. Chen, 1999兲, and so the dominant injected charge carri-
ers are electrons. As the injected carriers dominate the
conduction, leakage currents in ferroelectrics are elec-
cently been studied theoretically by Bratkovsky and Le- tron currents and not hole currents, contrary to the sug-
vanyuk 共2005兲. gestion of Stolichnov and Tagantsev 共1998兲.
The first picture of barrier formation in semiconduc-
tors is due to Schottky 共1938兲 and Mott 共1938兲. In this
3. Schottky barrier formation at metal-ferroelectric
picture the conduction band and valence band bend
junctions
such that the vacuum levels at the interface are the same
In general, since ferroelectric materials are good insu- and the Fermi level is continuous through the interface,
lators the majority of carriers are injected from the elec- but deep within the bulk of the semiconductor it retains
trode. When a metal is attached to a ferroelectric mate- its original value relative to the vacuum level. This is
rial, a potential barrier is formed if the metal work achieved by the formation of a depletion layer which
function is greater than the electron affinity of the ferro- shifts the position of the Fermi level by altering the
electric. This barrier must be overcome if charge carriers number of electrons within the interface.
are to enter the ferroelectric. On the other hand, if the Motivated by the experimental observation that many
electron affinity is greater than the work function, then Schottky barrier heights seemed to be fairly indepen-
an Ohmic contact is formed. For the usual applications dent of the metal used for the electrode, Bardeen 共1947兲
of ferroelectrics 共capacitors兲 it is desirable to have the proposed a different model of metal-semiconductor
largest barrier possible. If a metal is brought into contact junctions. In this picture the Fermi level of the semicon-
with an intrinsic pure ferroelectric and surface states do ductor is “pinned” by surface states to the original
not arise 共i.e., the classic metal-insulator junction兲, then charge neutrality level. These states, as first suggested by
the barrier height is simply Heine 共1965兲, are not typically real surface states but
rather states induced in the band gap of the semiconduc-
␾b = ␾m − ␹ . 共7兲
tor by the metal.
In this case the Fermi level of the metal becomes the Most junctions lie somewhere between the Schottky
Fermi level of the system, as there is no charge within and Bardeen limits. The metal-induced gap states can
the insulator with which to change it. On the other hand, accommodate some but not all of the difference in the
if there are dopants or surface states, then there can be a Fermi level between the metal and the semiconductor,
transfer of charge between the metal and ferroelectric, and so band bending still occurs to some extent. The
which allows the ferroelectric to bring the system Fermi factor S = d␾b / d␾m is used to define this, with S = 1 being
level towards its own Fermi level. the Schottky limit and S = 0 representing the Bardeen
Single crystals of undoped ferroelectric titanates tend limit. The value of S is determined by the nature of the
to be slightly p type simply because there are greater semiconductor; originally experimental trends linking
abundances of impurities with lower valences than those this to the covalency or ionicity of the bonding in the
of the ions for which they substitute 共Na+ for Pb+2; Fe+3 material were observed 共with covalent materials devel-
for Ti+4兲 共Chan et al., 1976, 1981; Smyth, 1984兲. In reality oping many more metal-induced gap states than ionic
most ferroelectric capacitors are fine-grained polycrys- materials; Kurtin et al., 1969兲. However, better correla-
talline ceramics and are almost always oxygen deficient. tion was found between the effective band gaps 共depen-

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1096 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

FIG. 15. Energy-band diagram of a metal n-type semiconduc-

tor contact. Adapted from Cowley and Sze, 1965. FIG. 16. Schottky barrier height of Pt-SrTiO3 as a function of
oxygen-vacancy concentration. Note that this may explain the
variation of experimental values from ⬃0.7 to 1.0 eV.
dent on the electronic dielectric constant ⑀⬁; Schluter,
1978兲, with 共Monch, 1986兲
out by Rhoderick and Williams 共1988兲, is equivalent to
1 neglecting the charge in the depletion width. In the sys-
− 1 = 0.1共⑀⬁ − 1兲2 . 共8兲
S tems under consideration here this term should not be
Although SrTiO3 was invoked as one of the materials neglected as it can be quite large. To demonstrate the
that violated the electronegativity rule by Schluter effect on the barrier height we calculate the barrier
共1978兲, it is omitted from the plot against ⑀⬁ − 1. The ex- height for a Pt-SrTiO3 barrier over a wide range of va-
perimental value for S in SrTiO3 can be measured from cancy concentrations 共Fig. 16兲.
Dietz’s data as approximately 0.5. This does not agree It can be seen that the effect of vacancies on barrier
well with what one would expect from Monch’s empiri- height becomes important for typical concentrations of
cal relation, which gives S = 0.28 关as used by Robertson vacancies encountered in ferroelectric thin films. Daw-
and Chen 共1999兲兴. Note that the use of the ionic trap- ber et al. 共2001兲 have addressed this issue and also the
effect of introduced dopants on barrier heights. Despite
free value S = 1 for BST gives a qualitative error. It pre-
their omission of the term discussed above, the work of
dicts that BST on Al should be Ohmic, whereas in actu-
Robertson and Chen is valuable because of their calcu-
ality it is a blocking junction; an S value of
lation of the charge neutrality levels for several ferro-
approximately 0.3 predicts a 0.4 eV Schottky barrier
electric materials, an essential parameter for the calcu-
height, in agreement with experiment 共Scott, 2000b兲.
lation of metal-ferroelectric barrier heights.
We can extract the penetration depth for Pt states into
In a ferroelectric thin film this distribution of charges
BaTiO3 from the first-principles calculation of Rao et al.
at the interface manifests itself in more ways than simply
共1997兲 by fitting the density of states of platinum states
in the determination of the Schottky barrier height.
in the oxygen layers to an exponential relationship to
Electric displacement in the system is screened over the
extract the characteristic length as 1.68 Å.
entire charge distribution.
Cowley and Sze 共1965; Fig. 15兲 derived an expression
for the barrier height for junctions between the two ex- In measuring the small-signal capacitance against
tremes. In this approach the screening charges in the thickness there is always a nonzero intercept, which has
electrode and the surface states are treated as delta been typically associated with a “dead layer” at the
functions of charge separated by an effective thickness metal-film interface. However, in most cases this interfa-
cial capacitance can be understood by recognizing that a
␦eff. This effective thickness takes into account both the
Thomas-Fermi screening length in the metal and the finite potential exists across the charge at the interface.
penetration length of the metal-induced gap states, and In the simplest approximation one neglects any charge
is essentially an air-gap approach. in the ferroelectric and uses a Thomas-Fermi screening
The expression for the barrier height is model for the metal. This was initially considered by Ku
and Ullman 共1964兲 and first applied to high-k dielectrics
␾b = S共␾m − ␹兲 + 共1 − S兲共Eg − ␾0兲 + ␨ , 共9兲 by Black and Welser 共1999兲. In their work they use a
large value for the dielectric constant of the oxide metal,

␨=
S 2C
2
冋
− S3/2 C共␾m − ␹兲 + 共1 − S兲共Eg − ␾0兲
C
S
considering it as the dielectric response of the ions
stripped of their electrons. This may seem quite reason-

册
able but is not, however, appropriate. In general we
1/2
C C 2S think of metals not being able to sustain fields, and in
− 共Eg − Ef + kT兲 + . 共10兲 the bulk they certainly cannot, but the problem of the
S 4
penetration of electric fields into metals is actually well
In the above S = 1 / 共1 + q2␦effDs兲, C = 2q␧sND␦eff
2
. When known in a different context, that of the microwave skin
␧s ⬇ 10␧0 and ND ⬍ 10 cm , C is of the order of 0.01 eV
18 −3
depth. It is very instructive to go through the derivation
and it is reasonable to discard the term ␨ as Cowley and as an ac current problem and then find the dc limit
Sze 共1965兲 did. Neglecting this term, as has been pointed which will typically apply for our cases of interest.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1097

We describe the metal in this problem using the 1 / C共d兲 versus thickness d extrapolates to finite values at
Drude free-electron theory: d = 0, demonstrating an interfacial capacitance. However,
whereas the value for the sol-gel films is consistent with
␴0
␴= . 共11兲 the Thomas-Fermi screening approach 共0.05 nm兲, the
1 + i␻␶ value of interfacial thickness 共0.005 nm兲 for the
There are three key equations to describe the charge chemical-vapor deposition films is only 10% of the inter-
distribution in the metal: Poisson’s equation for free facial capacitance that would arise from the known
charges, Fermi-Thomas screening length of 0.05 nm in the Pt
electrodes 共Dawber, Chandra, et al., 2003兲. That is, if this
1 ⳵ E共z兲 result were interpreted in terms of a “dead layer,” the
␳共z兲 = ; 共12兲
4␲ ⳵ z dead layer would have negative width. This result may
arise from a compensating “double layer” of space
the continuity equation, charge inside the semiconducting PZT dielectric; the
⳵ j共z兲 Armstrong-Horrocks 共1997兲 semiconductor formalism
− i␻␳共z兲 = ; 共13兲 form of the earlier Helmholtz and Gouy-Chapman
⳵z
polar-liquid models of the double layer can be used.
and the Einstein transport equation, Such a double layer is unnecessary in PVDF because
that material is highly insulating 共Moreira, 2002兲. This
⳵␳ explains quantitatively the difference 共8 times兲 of inter-
j = ␴E − D . 共14兲
⳵z facial capacitance in sol-gel PZT films compared with
chemical-vapor deposition PZT films of the same thick-
These are combined to give
ness. The magnitude of the electrokinetic potential 共or
⳵2␳ 4␲␴
⳵z 2 =
D
1+ 冉
i␻
4␲␴
␳共z兲. 冊 共15兲
zeta potential兲 ␨ = ␴d⬘ / ⑀⑀0 that develops from the Helm-
holtz layer can be estimated without adjustable param-
eters from the oxygen-vacancy gradient data of Dey for
This tells us that if at a boundary of the metal there a typical oxide perovskite, SrTiO3; using Dey’s surface
exists a charge, it must decay with the metal exponen- charge density ␴ of 2.8⫻ 1018 e / m2, a Gouy screening
tially with characteristic screening length ␭: length in the dielectric d⬘ = 20 nm, and a dielectric con-

␭= 冋 冉4␲␴
D
1+
i␻
4␲␴
冊册 −1/2
. 共16兲
stant of ⑀ = 1300 yields ␨ = 0.78 eV. Since this is compa-
rable to the Schottky barrier height, it implies that much
of the screening is provided internally by mobile oxygen
In the dc limit 共which applies for most frequencies of vacancies. 关Here ␴共␶ , ␮兲 is a function of time ␶ and mo-
our interest兲 this length is the Thomas-Fermi screening bility ␮ for a bimodal 共ac兲 switching process.兴
length,

␭0 = 冉 冊
4␲␴0
D
−1/2
. 共17兲
4. Conduction mechanisms
In general, conduction is undesirable in memory de-
vices based on capacitors, and so the understanding and
So it becomes clear that the screening charge in the minimization of conduction has been a very active area
metal may be modeled by substituting a sheet of charge of research over the years. Many mechanisms have been
displaced from the interface by the Thomas-Fermi proposed for the conduction in ferroelectric thin films.
screening length, but that in calculating the dielectric
thickness of this region the effective dielectric constant
a. Schottky injection
that must be used is 1, consistent with the derivation of
the screening length. Had we used a form of the Poisson Perhaps the most commonly observed currents in fer-
equation that had a nonunity dielectric constant, i.e., roelectrics are due to thermionic injection of electrons
from the metal into the ferroelectric. The current-
⑀ ⳵ E共z兲 voltage characteristic is determined by the image force
␳共z兲 = , 共18兲
4␲ ⳵ z lowering of the barrier height when a potential is ap-
plied. A few points should be made about Schottky in-
then our screening length would be jection in ferroelectric thin films. The first is about the

␭0 = 冉 冊
4␲␴0
⑀D
−1/2
, 共19兲
dielectric constant appropriate for use. In ferroelectrics
the size of the calculated barrier-height lowering de-
pends greatly on which dielectric constant, the static or
which is not the Thomas-Fermi screening length. Thus the electronic, is used. The correct dielectric constant is
the use of a nonunity dielectric constant for the metal is the electronic one 共⬃5.5兲, as discussed by Scott 共1999兲
not compatible with the use of the Thomas-Fermi and used by Dietz and Waser 共1997兲, and by Zafar et al.
screening length. 共1998兲. Dietz and Waser 共1997兲 used the more general
Measurements on both sol-gel and chemical-vapor injection law of Murphy and Good 共1956兲 to describe
deposition lead zirconate titanate 共PZT兲 films down to charge injection in SrTiO3 films. They found that for
⬃60 nm thickness show that reciprocal capacitance lower fields the Schottky expression was valid, but at

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1098 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

higher fields numerical calculations using the general in-

jection law were required. They did not, however, find
that Fowler-Nordheim tunneling was a good description
of any of the experimental data.
It has been shown by Zafar et al. 共1998兲 that in fact the
correct form of the Schottky equation that should be
used for ferroelectric thin films is the diffusion-limited
equation of Simmons. Furthermore, very recently Daw-
ber and Scott 共2004兲 have shown that when one consid-
ers the ferroelectric capacitor as a metal-insulator-metal
system with diffusion-limited current 共as opposed to a
single metal-insulator junction兲, the leakage current is FIG. 17. 共Color兲 Leakage-current data from Au-BST-
explained well; in addition, a number of unusual effects, SrRuO3 film at room temperature and at T = 70 K.
such as the negative differential resistivity observed by
Watanabe et al. 共1998兲 and the PTCR effect observed by
masses considered for these materials. Whereas they use
Hwang et al. 共1997, 1998兲, are accounted for.
effective masses of 1.0, the effective masses in perov-
skite oxides seem to be somewhat larger, 共5 – 7兲me for
b. Poole-Frenkel barium titanate and strontium titanate 共Scott et al.,
One of the standard ways of identifying a Schottky 2003兲. Although the tunneling mass and the effective
regime is to plot log共J / T兲 against V1/2. In this case the 共band兲 mass need not be the same in general, if the tun-
plot will be linear if the current injection mechanism is neling is through thicknesses of ⬎2 nm, they are nearly
Schottky injection. Confusion can arise because carriers so. 关Conley and Mahan 共1967兲 and Schnupp 共1967兲 also
can also be generated from internal traps by the Poole- find that the tunneling mass due to light holes in GaAs
Frenkel effect, which on the basis of this plot is indistin- fits the band mass very well.兴
guishable from Schottky injection. However, if the I-V
characteristic is asymmetric with respect to positive and d. Space-charge-limited currents
negative voltages 共as is usually the case兲 then the injec-
The characteristic quadratic relationship between cur-
tion process is most probably Schottky injection. There
rent and voltage that is the hallmark of space-charge-
are, however, some results that show symmetrical I-V
limited currents is often seen in ferroelectrics. Some-
curves and correctly explain their data on the basis of a
times it is observed that space-charge-limited currents
Poole-Frenkel conduction mechanism 共Chen et al.,
are seen when a sample is biased in one direction,
1998兲.
whereas for the opposite bias Schottky injection domi-
nates.
c. Fowler-Nordheim tunneling
Many researchers have discussed the possibility of e. Ultrathin films—direct tunneling
tunneling currents in ferroelectric thin-film capacitors.
Recently Rodriguez Contreras et al. 共2003兲 have suc-
For the most part they are not discussing direct tunnel-
ceeded in producing metal-PZT-metal junctions suffi-
ing through the film, which would be impossible for typi-
ciently thin 共6 nm兲 that it appears that direct tunneling
cal film thicknesses, but instead tunneling through the
or phonon-assisted tunneling 共in contrast to Fowler-
potential barrier at the electrode. The chief experimen-
Nordheim tunneling兲 through the film may occur, though
tal evidence that it might indeed be possible is from
this result requires more thorough investigation since
Stolichnov et al. 共1998兲, who have seen currents that they
the authors note the barrier heights extracted from their
claim to be entirely tunneling currents in PZT films 450
data using a direct tuneling model are much smaller than
nm thick at temperatures between 100 and 140 K. It
expected. The principal result of this paper is resistive
should be noted, however, that they only observed tun-
switching, which may be of considerable interest in de-
neling currents above 2.2 MV/ cm, below which they vice applications, but also requires more thorough inves-
were unable to obtain data. The narrowness of the range tigation. This very interesting experimental study raises
of fields for which they have collected data is a cause for important questions about the way that metal wave
concern, since the data displayed in their paper go from functions penetrating from the electrode and ferroelec-
2.2 to 2.8 MV/ cm. We conducted leakage-current mea- tric polarization interact with each other in the thinnest
surements on a 70-nm BST thin film at 70 K and found ferroelectric junctions.
that the leakage current, while of much lower magni-
tude, was still well described by a Schottky injection re-
lationship; although if one fitted this data to a similarly f. Grain boundaries
narrow field region, it did appear to satisfy the Fowler- Grain boundaries are often considered to be impor-
Nordheim relationship well 共Fig. 17兲. tant in leakage current because of the idea that they will
The effective masses for tunneling obtained in the provide conduction pathways through the film.
studies of Stolichnov et al. 共1999兲 and Baniecki et al. Gruverman’s results 关private communication, repro-
共2001兲 also seem to be at odds with the normal effective duced in Dawber and Scott 共2001兲兴 suggest that this is

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1099

not the case in SBT. In his experiment an atomic force Even in films for which there is considerable Poole-
microscopy tip is rastered across the surface of a poly- Frenkel limitation of current 共a bulk effect兲, the
crystalline ferroelectric film. The imaged pattern records Schottky barriers at the electrode interfaces will still
the leakage current at each point: white areas are high- dominate breakdown behavior.
current spots; dark areas, low current. If the leakage In general, electrical breakdown in ferroelectric ox-
were predominantly along grain boundaries, we should ides is a hybrid mechanism 共like spark discharge in air兲
see dark polyhedral grains surrounded by white grain in which the initial phase is electrical but the final stage
boundaries, which become brighter with increasing ap- is simple thermal runaway. This makes the dependence
plied voltage. In fact, the opposite situation occurs. This upon temperature complicated.
indicates that the grains have relatively low resistivity, There are at least three different contributions to the
with high-resistivity grain boundaries. The second sur- temperature dependence. The first is the thermal prob-
prise is that the grain conduction comes in a discrete ability of finding a hopping path through the material.
step; an individual grain suddenly “turns on” 共like a light Following Gerson and Marshall 共1959兲 and assuming a
switch兲. Smaller grains generally conduct at lower volt- random isotropic distribution of traps, Scott 共1995兲
ages 关in accord with Maier’s theory of space-charge ef- showed that
fects being larger in small grains with higher surface-
volume ratios 共Lubomirsky et al., 2002兲兴. k BT
EB = G − log A, 共22兲
B
which gives both the dependence on temperature T and
C. Device failure electrode area A in agreement with all experiments on
1. Electrical breakdown PZT, BST, and SBT.
In agreement with this model the further assumption
The process of electrical shorting in ferroelectric PZT of exponential conduction 共nonohmic兲 estimated to oc-
was first shown by Plumlee 共1967兲 to arise from den- cur for applied field E ⬎ 30 MV/ m 共Scott, 2000a兲,

冉 冊
dritelike conduction pathways through the material, ini-
tiated at the anodes and/or cathodes. These were mani- −b
␴共T兲 = ␴0exp 共23兲
fest as visibly dark filamentary paths in an otherwise k BT
light material when viewed through an optical micro-
in these materials yields the correct dependence of
scope. They have been thought to arise as “virtual cath-
breakdown time tB upon field
odes” via the growth of ordered chains of oxygen-
deficient material. This mechanism was modeled in log tB = c1 − c2EB 共24兲
detail by Duiker et al. 共Duiker, 1990; Duiker and Beale,
1990; Duiker et al., 1990兲. as well as the experimentally observed dependence of
To establish microscopic mechanisms for breakdown EB on rise time tc of the applied pulse:
in ferroelectric oxide films one must show that the de- EB = c3t−1/2 . 共25兲
c
pendences of breakdown field EB upon film thickness d,
ramp rate, temperature, doping, and electrodes are sat- Using the same assumption of exponential conduc-
isfied. The dependence for PZT upon film thickness is tion, which is valid for
most compatible with a low power-law dependence or is aeE Ⰶ kBT, 共26兲
possibly logarithmic 共Scott et al., 2003兲. The physical
models compatible with this include avalanche 共logarith- where a is the lattice nearest-neighbor oxygen-site hop-
mic兲, collision ionization from electrons injected via field ping distance 共approximately a lattice constant兲 and e,
emission from the cathode 共Forlani and Minnaja, 1964兲, the electron charge, Scott 共2000a兲 showed that the gen-
which gives eral breakdown field expression
dT
EB = Ad−w , 共20兲 CV − ⵜ共K · ⵜT兲 = ␴EB
2
共27兲
dt
with 1 / 4 ⬍ w ⬍ 1 / 2, or the linked defect model of Gerson in the impulse approximation 共in which the second term
and Marshall 共1959兲, where w = 0.3. The dependence on in the above equation is neglected兲 yields

冋 册 冉 冊
electrode material arises from the electrode work func- 1/2
tion and the ferroelectric electron affinity through the 3CVK b
EB共T兲 = T exp , 共28兲
resultant Schottky barrier height. Following Von Hippel ␴0btc 2kBT
共1935兲 we have 关Scott 共2000a兲, p. 62兴 which suffices to estimate the numerical value of break-
down field for most ferroelectric perovskite oxide films;
eEB␭ = h共⌽M − ⌽FE兲, 共21兲 values approximating 800 MV/ m are predicted and
measured.
where ⌽M and ⌽FE are the work functions of the metal A controversy has arisen regarding the temperature
and of the semiconducting ferroelectric, ␭ is electron dependence of EB共T兲 and the possibility of avalanche
mean free path, and h is a constant of order unity. 共Stolichnov et al., 2001兲. In low carrier-concentration

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1100 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

petitive electrical cycling, is one of the most serious de-

vice failure mechanisms in ferroelectric thin films. It is
most commonly a problem when Pt electrodes, desirable
because of their high work functions, are used.
Importantly fatigue occurs through the pinning of do-
main walls, which pins the polarization in a particular
direction, rather than any fundamental reduction of the
polarization. Scott and Pouligny 共1988兲 demonstrated in
FIG. 18. Change in the polarization hysteresis loop with fa- KNO3 via Raman spectroscopy that only a very small
tigue. Reprinted with permission from Scott and Pouligny, part of the sample was converted from the ferroelectric
1988. © 1988, AIP. to the nonferroelectric phase with fatigue, thus implying
that fatigue must be caused by pinning of the domain
single crystals, especially Si, avalanche mechanisms give walls. They also demonstrated that the domain walls
a temperature dependence that is controlled by the could be depinned via the application of a large field, as
mean free path of the injected carriers. This is physically shown in Fig. 18. The pinning of domain walls has also
because at higher temperatures the mean free path ␭ been observed directly with atomic force microscopy by
decreases due to phonon scattering and thus one must Gruverman et al. 共1996兲 and by Colla et al. 共1998兲.
apply a higher field EB to achieve avalanche conditions, There is a fairly large body of evidence that oxygen

冉 冊
vacancies play some key part in the fatigue process. Au-
EB ger data of Scott et al. 共1991兲 show areas of low oxygen
␭ = ␭0tanh . 共29兲
kT concentration in a region near the metal electrodes, im-
However, this effect is extremely small even for low plying a region of greater oxygen-vacancy data. Scott et
carrier concentrations 共10% change in EB between 300 al. also reproduced Auger data from Troeger 共1991兲 for
and 500 K for n = 1016 cm−3兲 and negligible for higher a film that had been fatigued by 1010 cycles showing an
concentrations. The change in EB in BST between 600 increase in the width of the region with depleted oxygen
and 200 K is ⬎500% and arises from Eq. 共22兲, not Eq. near the platinum electrode 共Fig. 19兲. There is, however,
共29兲. Even if the ferroelectrics were single crystals, with no corresponding change at the gold electrode. Gold
1020 cm−3 oxygen vacancies near the surface, any T de- does not form oxides. This might be an explanation of
pendence from Eq. 共29兲 would be unmeasurably small; the different behavior at the two electrodes. Although
and for the actual fine-grained ceramics 共40 nm grain some researchers believe that platinum also does not
diameters兲, the mean free path is ⬃1 nm and limited by form oxides, the adsorption of oxygen onto Pt surfaces is
grain boundaries 共T independent兲. Thus the conclusion actually a large area of research because of the impor-
of Stolichnov et al. 共2001兲 regarding avalanche is quali- tant role platinum plays as a catalyst in fuel-cell elec-
tatively and quantitatively wrong in ferroelectric oxides. trodes. Oxygen is not normally adsorbed onto gold sur-
faces but can be if there is significant surface roughness
共Mills et al., 2003兲.
2. Fatigue
It has also been found experimentally that films fa-
Polarization fatigue, which is the process whereby the tigue differently in atmospheres containing different
switchable ferroelectric polarization is reduced by re- oxygen partial pressures 共Brazier et al., 1999兲. Pan et al.

FIG. 19. 共a兲 Auger depth profile of PZT thin-film capacitor. 共b兲 Effect of fatigue on oxygen concentration near the electrode.
Reprinted with permission from Scott et al., 1991. © 1991, AIP.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1101

共1996兲 claim to have seen oxygen actually leaving a

ferroelectric sample during switching, though we note
that Nuffer et al. 共2001兲 claim this to be an experimental
artifact. The results of Schloss et al. 共2002兲 are very in-
teresting in that they show directly by O18 tracer studies
that the oxygen vacancies redistribute themselves during
voltage cycling. In their original paper they concluded
that the redistribution of oxygen vacancies was not the
cause of fatigue because they did not see redistribution
of O18 when the sample had been annealed. However, in
a more recent publication they conclude the reason they
could not see the oxygen tracer distribution was more
probably due to a change in the oxygen permeability of
the electrode after annealing 共Schloss et al., 2004兲.
It has been known for some time that the fatigue of
PZT films can be improved by the use of oxide elec-
trodes, such as iridium oxide or ruthenium oxides. de
Araujo et al. 共1995兲 explained the improved fatigue re-
sistance by the fact that oxides of iridium and platinum
can reduce or reoxidize reversibly and repeatedly with-
out degradation. For this same reason iridium is pre- FIG. 20. Atomic force microscopy images 共a兲 before and 共b兲
ferred to platinum as an electrode material for medical after cycling showing evidence of planes of oxygen vacancies in
applications, for which this property was originally stud- the fatigued sample. Image width= 10 ␮m. From Lupascu and
ied by Robblee and Cogan 共1986兲. This property does Rabe, 2002.
make the leakage-current properties of these electrodes
more complicated, and generally films with Ir/ IrO2 or found in barium titanate reduced after an accelerated
Ru/ RuO2 electrodes have higher leakage currents than life test 共Woodward et al., 2004兲.
those with platinum electrodes. Unless carefully an- While most researchers acknowledge that oxygen va-
nealed at a certain temperature RuO2 electrodes will cancies play a role in fatigue, it should be noted that
have elemental Ru metallic islands. Since the work func- Tagantsev et al. 共2001兲 have aggressively championed a
tion for Ru is 4.65 eV and that for RuO2 is 4.95 eV, model of charge injection; however, since this model is
almost all the current will pass through the Ru islands, not developed into a quantitative form, it is very hard to
producing hot spots and occasional shorts 共Hartmann et verify or falsify it. Charge injection probably does play a
al., 2000兲. By contrast, although one expects that there role in fatigue, an idea at least in part supported by the
will be similar issues with mixtures of Ir and IrO2 in detailed experimental study of Du and Chen 共1998b兲,
iridium-based electrodes, when metallic Ir is oxidized to but in the model of Dawber and Scott 共2000兲 关which
IrO2 its work function decreases to 4.23 eV 共Chalamala draws upon the basic idea of Yoo and Desu 共1992兲 that
et al., 1999兲. fatigue is due to the electromigration of oxygen vacan-
The idea that planes of oxygen vacancies perpendicu- cies兴 it is not included; nevertheless, most of the experi-
lar to the polarization direction could pin domain walls mental results in the literature may be accounted for.
is originally due to Brennan 共1993兲. Subsequently, in a The model of Dawber and Scott 共2000兲 basically shows
theoretical microscopic study of oxygen-vacancy defects that in a ferroelectric thin film under an ac field there is
in PbTiO3, Park and Chadi 共1998兲 showed that planes of in fact a net migration of vacancies towards the interface
vacancies are much more effective at pinning domain and it is the high concentration of vacancies in this re-
walls than single vacancies. Arlt and Neumann 共1988兲 gion that results in ordering of the vacancies and pinning
have discussed how under repetitive cycling in bulk fer- of domain walls. The interfacial nature of fatigue in thin
roelectrics the vacancies can move from their originally films has been demonstrated by Dimos et al. 共1994兲 and
randomly distributed sites in the perovskite structure to by Colla et al. 共1998兲.
sites in planes parallel to the ferroelectric-electrode in- To understand fatigue better, what are needed are
terface. We suspect that while this may account for fa- more experiments that try to look at the problem in
tigue in bulk ferroelectrics it is not the operative mecha- novel ways. Standard electrical measurements alone
nism in thin films. Scott and Dawber 共2000兲 have probably cannot shed a great deal of additional light on
suggested that in thin films the vacancies can reach suf- the problem, especially given the problems between
ficiently high concentrations that they order themselves comparing samples grown in different labs using differ-
into planes in a similar way as occurs in Fe-doped bulk ent techniques. Recently, a very interesting study was
samples and on the surfaces of highly reduced speci- undertaken by Do et al. 共2004兲 using x-ray microdiffrac-
mens. Direct evidence that this occurs in bulk PZT was tion to observe fatigue in PZT. Using this technique they
found using atomic force microscope imagery of PZT were able to see how regions of the film stopped switch-
grains by Lupascu and Rabe 共2002; Fig. 20兲. Recently, ing as it fatigued. One of the key findings of this study
evidence of oxygen-vacancy ordering has also been was that there appear to be two fatigue effects opera-

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1102 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

tive, a reversible effect that occurs when low to moder- strain兲. The form of the latter strongly resembles that of
ate fields are used for switching and an irreversible ef- a Landau-Devonshire theory, providing a connection be-
fect which occurs under very high fields. tween first-principles approaches and the extensive lit-
erature on phenomenological models for the behavior of
3. Retention failure
thin-film ferroelectrics. The advantages and disadvan-
tages of using first-principles results rather than experi-
Clearly a nonvolatile memory that fails to retain the mental data to construct models will be considered. In
information stored in it will not be a successful device. addition to allowing the study of systems far more com-
Furthermore, producers of memories need to be able to plex than those that can be considered by first principles
guarantee retention times over much longer periods of alone, this modeling approach yields physical insight
time than they can possibly test. A better understanding into the essential differences between bulk and thin-film
of retention failure is thus required so that models can behavior, which will be discussed at greater length in
be used that allow accelerated retention tests to be car- Sec. IV.B.5. Finally, it will be seen that, despite practical
ried out. The work of Kim et al. 共2001兲 is a step in this limitations, the complexity of the systems for which ac-
direction. curate calculations can be undertaken has steadily in-
It seems that imprint and retention failure are closely creased in recent years, to the point where films of sev-
linked phenomena; i.e., if a potential builds up over a eral lattice constants in thickness can be considered.
period of time, it can destabilize the ferroelectric polar- While this is still far thinner than the films of current
ization state and thus cause loss of information. Further technological interest, concommitent improvements in
comparison of retention-time data and fatigue data sug- thin-film synthesis and characterizaton have made it pos-
gests a link between the two effects. Direct current deg- sible to achieve a high degree of atomic perfection in
radation of resistance in BST seems also to be a related comparable ultrathin films in the context of research.
effect 共Zafar et al., 1999兲. The electromigration of oxy- This progress has led to a true relevance of calculational
gen vacancies under an applied field in a Fe-doped results to experimental observations, opening a mean-
SrTiO3 single crystal has been directly observed via elec- ingful experimental-theoretical dialog. However, this
trochromic effects 共Waser et al., 1990兲. Oxygen-vacancy progress in some ways only serves to highlight the full
redistribution under an applied field has also been in- complexity of the physics of real ferroelectric films: as
voked to explain a slow relaxation of the capacitance in questions get answered, more questions, especially
BST thin films 共Boikov et al., 2001兲. It would seem to about phenomena at longer length scales and about dy-
make sense that whereas fatigue relates to the cumula- namics, are put forth. These challenges will be discussed
tive motion of oxygen vacancies under an ac field, resis- in Sec. IV.B.6.
tance degradation is a result of their migration under an
applied dc field, and retention failure is a result of their
migration under the depolarization field or other built-in A. Density-functional-theory studies of bulk ferroelectrics
fields in the material.
In parallel with advances in laboratory synthesis, the
past decade has seen a revolution in the atomic-scale
IV. FIRST PRINCIPLES theoretical understanding of ferroelectricity, especially
in perovskite oxides, through first-principles density-
With continuing advances in algorithms and computer functional-theory investigations. The central result of a
hardware, first-principles studies of the electronic struc- density-functional-theory calculation is the ground-state
ture and structural energetics of complex oxides can energy computed within the Born-Oppenheimer ap-
now produce accurate, material-specific information rel- proximation; from this the predicted ground-state crystal
evant to the properties of thin-film ferroelectrics. In this structure, phonon dispersion relations, and elastic con-
section, we focus on first-principles studies that identify stants are directly accessible. The latter two quantities
and analyze the characteristic effects specific to thin can be obtained by finite-difference calculations, or,
films. First, we briefly review the relevant methodologi- more efficiently, through the direct calculation of deriva-
cal progress and the application of these methods to tives of the total energy through density-functional per-
bulk ferroelectric materials. Next, we survey the first- turbation theory 共Baroni et al., 2001兲.
principles investigations of ferroelectric thin films and For the physics of ferroelectrics, the electric polariza-
superlattices reported in the literature. It will be seen tion and its derivatives, such as the Born effective
that the scale of systems that can be studied directly by charges and the dielectric and piezoelectric tensors, are
first-principles methods is severely limited at present by as central as the structural energetics, yet proper formu-
practical considerations. This can be circumvented by lation in a first-principles context long proved to be
the construction of nonempirical models with param- quite elusive. Expressions for derivatives of the polariza-
eters determined by fitting to the results of selected first- tion corresponding to physically measurable quantities
principles calculations. These models can be param- were presented and applied in density-functional pertur-
etrized interatomic potentials, permitting molecular- bation theory calculations in the late 1980s 共de Gironcoli
dynamics studies of nonzero temperature effects, or et al., 1989兲. A key conceptual advance was establishing
first-principles effective Hamiltonians for appropriate the correct definition of the electric polarization as a
degrees of freedom 共usually local polarization and bulk property through the Berry-phase formalism of

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1103

King-Smith, Vanderbilt, and Resta 共King-Smith and Singh for PbZrO3 共Singh, 1995兲, in which the correct
Vanderbilt, 1993; Resta, 1994兲. With this and the related energy ordering between ferroelectric and antiferroelec-
Wannier function expression 共King-Smith and Vander- tric structures was obtained and, furthermore, compari-
bilt, 1994兲, the spontaneous polarization and its deriva- sons of the total energy resolved an ambiguity in the
tives can be computed in a post-processing phase of a reported space group and provided an accurate determi-
conventional total-energy calculation, greatly facilitating nation of the oxygen positions.
studies of polarization-related properties. It is important to note, however, that there seem to be
For perovskite oxides, the presence of oxygen and limitations on the accuracy to which structural param-
first-row transition metals significantly increases the eters, particularly lattice constants, can be obtained.
computational demands of density-functional total- Most obvious is the underestimation of the lattice con-
energy calculations compared to those for typical semi- stants within the local-density approximation, typically
conductors. Calculations for perovskite oxides have by about 1% 共generalized-gradient approximation tends
been reported using essentially all of the available first- to shift lattice constants upward, sometimes substantially
principles methods for accurate representation of the overcorrecting兲. Considering that the calculation in-
electronic wave functions: all-electron methods, mainly volves no empirical input whatsoever, an error as small
linearized augmented plane wave 共LAPW兲 and full- as 1% could be regarded not as a failure, but as a success
potential linearized augmented plane wave 共FLAPW兲, of the method. Moreover, the fact that the underesti-
linear muffin-tin orbitals, norm-conserving and ultrasoft mate varies little from compound to compound means
pseudopotentials, and projector-augmented wave- that the relative lattice constants and thus the type of
function potentials. The effects of different choices for lattice mismatch 共tensile/compressive兲 between two ma-
the approximate density functional have been examined; terials in a heterostructure is generally correctly repro-
while most calculations are carried out with the local- duced when using computed lattice parameters. How-
density approximation, for many systems the effects of ever, for certain questions, even a 1% underestimate can
the generalized-gradient approximation and weighted- be problematic. The ferroelectric instability in the per-
density approximation 共Wu et al., 2004兲 have been inves- ovskite oxides, in particular, is known to be very sensi-
tigated, as well as the alternative use of the Hartree- tive to pressure 共Samara, 1987兲 and thus to the lattice
Fock approach. Most calculations being currently constant, so that 1% can have a significant effect on the
reported are performed with an appropriate standard ferroelectric instability. In addition, full optimization of
package, mainly VASP 共Kresse and Hafner, 1993; Kresse all structural parameters in a low-symmetry space group
and Furthmuller, 1996兲, with ultrasoft pseudopotentials can in some cases, PbTiO3 being the most well-studied
and projector-augmented wave function potentials, example, lead to an apparently spurious prediction,
ABINIT 共Gonze et al., 2002兲, with norm-conserving though fixing the lattice constants to their experimental
pseudopotentials and projector-augmented wave func- values leads to good agreement for the other structural
tion potentials, PWscf 共Baroni et al.兲, with norm- parameters 共Saghi-Szabo et al., 1998兲. Thus it has be-
conserving and ultrasoft pseudopotentials, SIESTA 共Soler come acceptable, at least in certain first-principles con-
et al., 2002兲, with norm-conserving pseudopotentials, texts, to fix the lattice parameters or at least the volume
WIEN97 共FLAPW兲 共Blaha et al., 1990兲 and CRYSTAL of the unit cell, to the experimental value when this
共Hartree-Fock兲 共Dovesi et al., 2005兲. value is known.
To predict ground-state crystal structures, the usual In a first-principles structural prediction, the initial
method is to minimize the total energy with respect to choice of space group may appear to limit the chance
free structural parameters in a chosen space group, in a that a true ground-state structure will be found. In gen-
spirit similar to that of a Rietveld refinement in an ex- eral, once a minimum is found, it can be proved 共or not兲
perimental structural determination. The space group is to be a local minimum by computation of the full pho-
usually implicitly specified by a starting guess for the non dispersion and of the coupling, if allowed by sym-
structure. For efficient optimization, the calculation of metry, between zone-center phonons and homogeneous
forces on atoms and stresses on the unit cell is essential strain 共Garcia and Vanderbilt, 1996兲. Of course, this does
and is included now in every standard first-principles not rule out the possibility of a different local minimum
implementation following the formalism of Hellmann with lower energy with an unrelated structure.
and Feynman 共1939兲 for the forces and Nielsen and Mar- For ferroelectrics, the soft-mode theory of ferroelec-
tin 共1985兲 for stresses. tricity provides a natural conceptual framework for the
The accuracy of density-functional theory for predict- identification of low-symmetry ground-state structures
ing the ground-state structures of ferroelectrics was first and for estimating response functions. The starting point
investigated for the prototypical cases of BaTiO3 and is the identification of a high-symmetry reference struc-
PbTiO3 共Cohen and Krakauer, 1990, 1992; Cohen, 1992兲 ture. For perovskite compounds, this is clearly the cubic
and then extended to a larger class of ferroelectric per- perovskite structure, and for layered perovskites, it is
ovskites 共King-Smith and Vanderbilt, 1994兲. Extensive the nonpolar tetragonal structure. The lattice instabili-
studies of the structures of perovskite oxides and related ties of the reference structure can be readily identified
ferroelectric-oxide structures have since been carried from the first-principles calculation of phonon disper-
out 共Resta, 2003兲. The predictive power of first- sion relations 共Waghmare and Rabe, 1997a; Ghosez et
principles calculations is well illustrated by the results of al., 1999; Sai and Vanderbilt, 2000兲, this being especially

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1104 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

efficient within density-functional perturbation theory. tionals, such as the local-density approximation, the fun-
In simple ferroelectric perovskites, the ground-state damental band gaps of insulators and semiconductors,
structure is obtained to lowest order by freezing in the including perovskite ferroelectrics, are substantially un-
most unstable mode 共a zone-center polar mode兲. This derestimated. While for narrow gap materials the system
picture can still be useful for more complex ground-state may even be erroneously found to be metallic, for wider
structures that involve the freezing in two or more gap systems such as most of the simple ferroelectric per-
coupled modes 共e.g., PbZrO3兲 共Waghmare and Rabe, ovskite compounds considered here, the band gap is still
1997b; Cockayne and Rabe, 2000兲, as well as for identi- nonzero and thus the structural energetics in the vicinity
fying low-energy structures that might be stabilized by of the ground-state structure is unaffected. While this
temperature or pressure 共Stachiotti et al., 2000; Fennie error might be considered to be an insuperable stum-
and Rabe, 2005兲. The polarization induced by the soft bling block to first-principles investigation of electronic
polar mode can be obtained by computation of the Born structure and related properties, there is at present no
effective charges, yielding the mode effective charge. widely available, computationally tractable, alternative
The temperature-dependent frequency of the soft polar 关there are, though, some indications that the use of
mode and its coupling to strain are expected largely to exact-exchange functionals can eliminate much of this
determine the dielectric and piezoelectric response of error 共Piskunov et al., 2004兲; it has long been known that
ferroelectric and near-ferroelectric perovskite oxides; Hartree-Fock, i.e., exact exchange only, leads to overes-
this idea has been the basis of several calculations timates of the gaps兴. The truth is that, as will be dis-
共Cockayne and Rabe, 1998; Garcia and Vanderbilt, cussed further below, these results can, with care, aware-
1998兲. ness of the possible limitations, and judicious use of
Minimization of the total energy can similarly be used experimental input, be used to extract useful informa-
to predict atomic arrangements and formation energies tion about the electronic structure and related proper-
of point defects 共Park and Chadi, 1998; Poykko and ties in individual material systems.
Chadi, 2000; Betsuyaku et al., 2001; Man and Feng, 2002; At present, the computational limitations of full first-
Park, 2003; Robertson, 2003; Astala and Bristowe, 2004兲, principles calculations to 70–100 atoms per supercell
domain walls 共Poykko and Chadi, 2000; Meyer and have stimulated considerable interest in the develop-
Vanderbilt, 2002; He and Vanderbilt, 2003兲, and non- ment and use in simulations of effective models, from
stoichiometric planar defects such as antiphase domain which a subset of the degrees of freedom have been
boundaries 共Suzuki and Fujimoto, 2001; Li et al., 2002兲, integrated out. Interatomic shell-model potentials have
in bulk perovskite oxides. The supercell used must ac- been developed for a number of perovskite-oxide sys-
commodate the defect geometry and generally must tems, eliminating most of the electronic degrees of free-
contain many bulk primitive cells to minimize the inter- dom except for those represented by the shells 共Tinte et
action of a defect with its periodically repeated images. al., 1999; Heifets, Kotonin, and Maier, 2000; Sepliarsky
Thus these calculations are extremely computationally et al., 2004兲. A more dramatic reduction in the number
intensive, and many important questions remain to be of degrees of freedom is performed to obtain effective
addressed. Hamiltonians, in which typically one vector degree of
While much of the essential physics of ferroelectrics freedom decribes the local polar distortion in each unit
arises from the structural energetics, the polarization, cell. This approach has proved useful for describing
and the coupling between them, there has been increas- finite-temperature structural transitions in compounds
ing interest in ferroelectric oxides as electronic and op- and solid solutions 共Zhong et al., 1994; Rabe and Wagh-
tical materials, for which accurate calculations of the gap mare, 1995, 2002; Waghmare and Rabe, 1997a; Bellaiche
and dipole matrix elements are important. Furthermore, et al., 2000兲. To the extent that the parameters appearing
as we shall discuss in detail below, the band structures in these potentials are determined by fitting to selected
enter in an essential way in understanding the charge first-principles results 共e.g., structures, elastic constants,
transfer and dipole layer formation of heterostructures phonons兲, these approaches can be regarded as approxi-
involving ferroelectrics, other insulators, and metals. mate first-principles methods. They allow computation
While density-functional theory provides a rigorous of the polarization as well as of the structural energetics,
foundation only for the computation of Born- but not, however, of the electronic states. Most of the
Oppenheimer ground-state total energies and electronic effort has been focused on BaTiO3, though other per-
charge densities, it is also often used for investigation of ovskites, including PbTiO3 and KNbO3, SrTiO3 and
electronic structure. In the vast majority of density- 共Ba, Sr兲TiO3, and Pb共Zr, Ti兲O3, have been investigated
functional implementations, calculation of the ground- in this way and useful results obtained.
state total energy and charge density involves the com-
putation of a band structure for independent electron B. First-principles investigation of ferroelectric thin films
states in an effective potential, following the work of
Kohn and Sham 共1965兲. This band structure is generally The interest in ferroelectric thin films and superlat-
regarded as a useful guide to the electronic structure of tices lies in the fact that the properties of the system as a
materials, including perovskite and layered perovskite whole can be so different from the individual properties
oxides 共Cohen, 1992; Robertson et al., 1996; Tsai et al., of the constituent material共s兲. Empirically, it has been
2003兲. It should be noted that with approximate func- observed that certain desirable bulk properties, such as a

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1105

high dielectric response, can be degraded in thin films, produce a supercell; a few studies have been carried out
while in other investigations there are signs of novel in- with electronic wave-function basis sets that permit the
teresting behavior obtained only in thin-film form. study of an isolated slab. The variables to be specified
Theoretical analysis of the observed properties of thin include the orientation, the number of atomic layers in-
films presents a daunting challenge. It is well known that cluded, choice of termination of the surfaces, and width
the process of thin-film growth itself can lead to non- of vacuum layer separating adjacent slabs. As in the
trivial differences from the bulk material, as observed in first-principles prediction of the bulk crystal structure,
studies of homoepitaxial oxide films such as SrTiO3 choice of a space group for the supercell is usually es-
共Klenov et al., 2004兲. However, as synthetic methods tablished by the initial structure; relaxations following
have developed, the goal of growing nearly ideal, atomi- forces and stresses do not break space-group symme-
cally ordered, single-crystal films and superlattices is tries. The direction of the spontaneous polarization is
coming within reach, and the relevance of first-principles constrained by the choice of space group, allowing com-
results for perfect single-crystal films to experimental parison of unpolarized 共paraelectric兲 films with films po-
observations emerging. larized along the normal or in the plane of the film.
As will be clear from the discussion in the rest of this As we shall see below, in most cases the slabs are very
section, an understanding of characteristic thin-film be- thin 共ten atomic layers or fewer兲. It is possible to relate
havior can best be achieved by detailed quantitative ex- the results to the surface of a semi-infinite system or a
amination of individual systems combined with the con- coherent epitaxial film on a semi-infinite substrate by
struction of models incorporating various aspects of the imposing certain constraints on the structures consid-
physics, from which more general organizing principles ered. In the former case, the atomic positions for inte-
can be identified. In first-principles calculations, there is rior layers are fixed to correspond to the bulk crystal
a freedom to impose constraints on structural param- structure. In the latter, the in-plane lattice parameters of
eters and consider hypothetical structures that goes far the supercell are fixed to the corresponding bulk lattice
beyond anything possible in a real system being studied parameters of the substrate, which is not otherwise ex-
experimentally. This will allow us to isolate and examine plicitly included in the calculation. More sophisticated
various influences on the state of a thin film: epitaxial methods developed to deal with the coupling of vibra-
strain, macroscopic electric fields, surfaces and inter- tional modes at the surface with bulk modes of the sub-
faces, characteristic defects associated with thin-film strate 共Lewis and Rappe, 1996兲 could also be applied to
growth, and “true” finite-size effects, and how they ferroelectric thin films, though this has not yet been
change the atomic arrangements, electronic structure, done.
polarization, vibrational properties, and responses to ap- The local-density approximation underestimate of the
plied fields and stresses. While the main focus of this equilibrium atomic volume will in general also affect
review is on thin films, this approach also applies natu- slab calculations, and similar concerns arise on the cou-
rally to multilayers and superlattices. Extending our dis- pling of strain and the ferroelectric instability. As in the
cussion to include these latter systems will allow us to bulk crystal structure prediction, it may in some cases be
consider the effects of the influencing factors in different appropriate to fix certain structural parameters accord-
combinations, for example, the changing density of in- ing to experimental or bulk information. In the case of
terfaces, the degree of mismatch strain, and the polariza- superlattices and supercells of films on substrates, it may,
tion mismatch. These ideas are also relevant to investi- on the other hand, be a good choice to work consistently
gating the behavior of bulk-layered ferroelectrics, which at the 共compressed兲 theoretical lattice constant since the
can be regarded as natural short-period superlattices. generic underestimate of the atomic volume ensures that
Within this first-principles modeling framework, we the lattice mismatch and relative tensile/compressive
can more clearly identify specific issues and results for strain will be correctly reproduced. This applies, for ex-
investigation and analysis. The focus on modeling is also ample, to the technique mentioned in the previous para-
key to the connection of first-principles results to the graph, in which the effects of epitaxial strain are inves-
extensive literature on phenomenological analysis and tigated by performing slab calculations with an
to experimental observations. This makes the most ef- appropriate constraint on the in-plane lattice param-
fective use of first-principles calculations in developing a eters.
conceptual and quantitative understanding of character- As in first-principles predictions of the bulk crystal
istic thin-film properties, as manifested by thickness de- structure, the initial choice of space group constrains, to
pendence as well as by the dependence on choice of a large extent, the final “ground-state” structure. If the
materials for the film, substrate, and electrodes. supercell is constructed by choosing a bulk termination,
the energy minimization based on forces and stresses
will preserve the initial symmetry, yielding information
1. First-principles methodology for thin films
about surface relaxations of the unreconstructed surface.
The fundamental geometry for the study of thin films, A lower-energy structure might result from breaking ad-
surfaces, and interfaces is that of an infinite single- ditional point or translational symmetries to obtain a
crystalline planar slab. Since three-dimensional period- surface reconstruction. This type of surface reconstruc-
icity is required by most first-principles implementa- tion could be detected by computing the Hessian matrix
tions, in those cases the slab is periodically repeated to 共coupled phonon dispersion and homogeneous strain兲

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1106 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

zero, this inevitably leads to a nonzero electric field in

the vacuum. Physically, this can be interpreted as an ex-
ternal field applied uniformly across the unit cell which
acts partially to compensate the depolarizing field. An
analogous situation arises for an asymmetrically termi-
nated slab when the two surfaces have different work
functions. To eliminate the artificial field in the vacuum,
one technique is to introduce a dipole layer in the mid-
vacuum region far away from the slab 共Bengtsson, 1999兲.
This can accommodate a potential drop up to a critical
value, at which point electrons begin to accummulate in
an artificial well in the vacuum region 共see Fig. 21兲. This
approach can also be used to compensate the depolar-
ization field in a perpendicular polarized film, though it
may happen that the maximum field that can be applied
is smaller than that needed for full compensation. Alter-
natively, by using a first-principles implementation with
a localized basis set, it is possible to perform computa-
tions for isolated slabs and thus avoid not only the spu-
rious electric fields, but also the interaction between pe-
riodic slab images present even for symmetric nonpolar
slabs. Comparison between results obtained with the
two approaches is presented by Fu et al. 共1999兲.
In determining the properties of an ultrathin film, the
film-substrate and/or film-electrode interface plays a
role at least as important as the free surface. In first-
principles calculations, the atomic and electronic struc-
FIG. 21. Schematic picture of the planar-averaged potential
ture of the relevant interface共s兲 is most readily obtained
v̄共z兲 for periodically repeated slabs: 共a兲 with periodic boundary
conditions, 共b兲 potential of the dipole layer, 共c兲 dipole-
by considering a periodic multilayer geometry identical
corrected slabs with vanishing external electric field, and 共d兲 to that used for computing the structure and properties
dipole-corrected slabs with vanishing internal electric field. of superlattices. To simulate a semi-infinite substrate,
From Meyer and Vanderbilt, 2001. the in-plane lattice constant should be fixed to that of
the substrate bulk. The relaxation of a large number of
structural degrees of freedom requires substantial com-
for the relaxed surface. More complex reconstructions
puter resources, and some strategies for efficiently gen-
involving adatoms, vacancies, or both would have to be
erating a good starting structure will be described in the
studied using appropriate starting supercells. Informa-
tion regarding the existence and nature of such recon- discussion of specific systems in Sec. IV.B.3.
structions might be drawn from experiments and/or Calculations of quantities characterizing the electronic
from known reconstructions in related materials. structure are based on use of the Kohn-Sham one-
One very important consideration in the theoretical electron energies and wave functions. Band structures
prediction of stable surface orientations and termina- for 1 ⫻ 1 共001兲 slabs are generally displayed in the 2D
tions and of favorable surface reconstructions is that surface Brillouin zone for ksupercell,z = 0. One way to iden-
these depend on the relative chemical potential of the tify surface states is by comparison with the appropri-
constituents. Fortunately, since the chemical potential ately folded-in bulk band structure. Another analysis
couples only to the stoichiometry, the prediction of the method is to compute the partial density of states pro-
change of relative stability with chemical potential can jected onto each atom in the slab. From these plots, the
be made with a single total-energy calculation for each dominant character of a state at a particular energy can
structure 共see, for example, Meyer et al., 1999兲. Because be found. In addition, an estimate of valence-band off-
of the variation in stoichiometry for different 共001兲 sur- sets and Schottky barriers at an interface can be ob-
face terminations, what is generally reported is the aver- tained by analyzing the partial density of states for a
age surface energy of symmetric AO- and BO2- superlattice of the two constituents. This is done com-
terminated slabs. paring the energies of the highest occupied states in the
A problem peculiar to the study of periodically re- interior of the relevant constituent layers 共because of the
peated slabs with polarization along the normal is the band-gap problem, the positions of the conduction
appearance of electric fields in the vacuum. As shown in bands are computed using the experimental bulk band
Fig. 21, this occurs because there is a nonzero macro- gaps兲. This estimate can be refined, as described in Jun-
scopic depolarizing field in the slab and thus a nonzero quera et al. 共2003兲, by computation of the average elec-
potential drop between the two surfaces of the slab. As trostatic energy difference between the relevant con-
the potential drop across the entire supercell must be stituent layers.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1107

As the computational resources required for full first-

principles calculations even for the simplest slab-vacuum
system are considerable, there is a strong motivation to
turn to interatomic potentials and effective Hamilto-
nians. Interatomic potentials based on shell models fit-
ted to bulk structural energetics are generally directly
transferred to the isolated slab geometry, with no
changes for the undercoordinated atoms at the surface.
As discussed below, this approach seems to be successful
in reproducing the relaxations observed in full first-
principles studies and has been applied to far larger su-
percells and superlattices. In the case of effective Hamil-
tonians, it is, at least formally, possible simply to
perform a simulation by removing unit cells and using
FIG. 22. Schematic illustration of the structure of the first
bulk interaction parameters for the unit cells in the film.
three surface layers. From Meyer and Vanderbilt, 2001.
For a more accurate description, modification of the ef-
fective Hamiltonian parameters for the surface layers is
advisable to restore the charge neutrality sum rule for cluding two or more different materials as well as
the film 共Ruini et al., 1998; Ghosez and Rabe, 2000兲. In vacuum layers have been used to simulate ferroelectric
addition, the effects of surface relaxation would also re- thin-film interactions with the substrate 关e.g.,
sult in modified interactions at the surface. PbTiO3 / SrTiO3 / PbTiO3 / vacuum 共Johnston and Rabe,
2005兲兴, and with realistic electrodes 关e.g.,
Pt/ BaTiO3 / Pt/ vacuum 共Rao et al., 1997兲兴 and
2. Overview of systems
Pt/ PbTiO3 / Pt/ vacuum 共Sai, Kolpak, and Rappe, 2005兲,
In this section, we present a list of the materials and as well as the structure of epitaxial alkaline-earth oxide
configurations that have been studied, followed by a on silicon, used as a buffer layer for growth of
brief overview of the quantities and properties that have perovskite-oxide films 共McKee et al., 2003兲.
been calculated in one or more of the reported studies. In each class of configurations, there are correspond-
A more detailed description of the work on individual ing quantities and properties that are generally calcu-
systems is provided in the following sections. lated. In the single slab-vacuum configuration, for each
The configurations that have been considered to date orientation and surface termination the surface energy is
in first-principles studies can be organized into several obtained. While in initial studies the atomic positions
classes. The simplest configuration is a slab of ferroelec- were fixed according to structural information from the
tric material alternating with vacuum; this can be used to bulk 共Cohen, 1996兲, in most current studies relaxations
investigate the free surface of a semi-infinite crystal, an are obtained by energy minimization procedures. For
unconstrained thin film, or an epitaxial thin film con- the most commonly studied perovskite 共001兲 slab, the
strained to match the lattice constant of an implicit sub- relaxation geometry is characterized by changes in inter-
strate. Specific materials considered included BaTiO3 plane spacings and rumplings quantified as shown in Fig.
关共001兲 surfaces 共Cohen, 1996, 1997; Fu et al., 1999; Heif- 22
ets et al., 2001a; Meyer and Vanderbilt, 2001; Krcmar Most studies assume bulk periodicity in the plane. For
and Fu, 2003兲 and 共110兲 surfaces 共Heifets et al., 2001a兲兴 the study of surface reconstructions, it is necessary to
SrTiO3 共Padilla and Vanderbilt, 1998; Heifets et al., expand the lateral unit cell, leading to a substantial ad-
2001a, 2002a, 2002b; Kubo and Nozoye, 2003兲, PbTiO3 ditional cost in computational resources. Most attention
共Ghosez and Rabe, 2000; Meyer and Vanderbilt, 2001; has been focused on the paraelectric SrTiO3, although
Bungaro and Rabe, 2005兲, KNbO3 共Heifets, Kotonin, recently studies have been carried out as well for
and Jacobs, 2000兲, and KTaO3 共Li, Akhador, et al., 2003兲. PbTiO3. A by-product of the total-energy calculation
Another type of configuration of comparable complexity that is often though not universally presented is the
is obtained by replacing the vacuum by a second mate- band structure and/or density of states; the local density
rial. If this is another insulating perovskite oxide, the of states at the surface is of particular interest.
calculation can yield information on ferroelectric- The two-component superlattice configuration can be
dielectric 关e.g., BaTiO3 / SrTiO3 共Neaton and Rabe, 2003兲 taken to model a film on a substrate and/or with elec-
and KNbO3 / KTaO3 共Sepliarsky et al., 2001, 2002兲兴 or trode layers, or to model an actual superlattice such as
ferroelectric-ferroelectric interfaces and superlattices. that obtained by molecular-beam epitaxy techniques. In
This configuration can also be used to study the inter- these studies, the interface and coupling between the
face between the ferroelectric and a dielectric 共nonfer- two constituents is of primary interest; combinations
roelectric兲 oxide 关e.g., BaTiO3 / BaO and SrTiO3 / SrO most considered to date are ferroelectric+ paraelectric,
共Junquera et al., 2003兲兴. Replacement of the vacuum by a ferroelectric+ dielectric, and ferroelectric+ metal, while
conductor simulates a film with symmetrical top and the combination of two ferroelectrics or ferroelectric
bottom electrodes, e.g., BaTiO3 / SrRuO3 共Junquera and + ferromagnetic materials has been less intensively in-
Ghosez, 2003兲. More complex multilayer geometries in- vestigated. The main questions of interest are the struc-

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1108 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

tural rearrangements at the interface and the change in a. BaTiO3

the structure, polarization, and related properties of in-
For BaTiO3, full first-principles results have been re-
dividual layers relative to the bulk resulting from the
ported primarily for the 共001兲 orientation, with a few
interaction with the other constituents. Analysis of the
results for the 共110兲 and 共111兲 orientations. In the slab-
trends with varying thickness共es兲 of the ferroelectric film
vacuum configuration, systems up to ten atomic layers
and, in the superlattice, other constituents is particularly
have been considered. The unpolarized slab is compared
useful. The electronic structure of these systems can be
with slabs with nonzero polarization, in the plane and/or
most readily characterized by the band offset between
along the normal. After reviewing the results on struc-
the two materials, which should also control the charge
tures, we shall describe the results of extension to larger-
transfer across the interface, formation of a dipole layer,
and the potential difference between the two constitu- scale systems through the use of interatomic potentials.
ents. In the case of a metal, this will determine the type The discussion of BaTiO3 will be concluded by a de-
of contact. The existence of interface states is also very scription of the first-principles results for surface elec-
relevant to the physical behavior of the system. tronic structure.
At a considerable increase in computational expense First-principles FLAPW calculations for the BaTiO3
but also in realism, a system with three or more compo- 共001兲 and 共111兲 slabs were first presented in 1995 共Co-
nents can be studied; e.g., the combination of a sub- hen, 1996兲, and later extended using the LAPW+ LO
strate, a film, and vacuum. The main questions of inter- method 共Cohen, 1997兲. The supercells contained six and
est in the few such studies to date are the analysis of the seven atomic layers, corresponding to asymmetric termi-
ferroelectric instability in the film, and the film-induced nation and two symmetric terminations 共BaO and TiO2兲,
changes in the substrate layers closest to the interface. and an equal vacuum thickness. The central mirror-
As in two-component heterostructures, the partial den- plane symmetry z → −z symmetry is broken for the
sity of states and the layer-average electrostatic potential asymmetric termination even in the absence of ferro-
are also useful in extracting the electronic behavior of electric distortion. The primitive cell lattice constant was
the system. fixed at the experimental cubic value 4.01 Å. Total ener-
In all of these studies, one of the main questions is gies of several selected paraelectric and ferroelectric
that of the structure and polarization of the ferroelectric structures were computed and compared: the ideal
layer compared to that of the bulk. Certainly, the change paraelectric slab and ferroelectric slabs with displace-
of environment 共electrical and mechanical boundary ments along ẑ corresponding to the experimental tetrag-
conditions兲 and the finite dimensions 共film thickness, onal structure. For the asymmetrically terminated slab,
particle size兲 are expected to strongly affect the struc- both the ferroelectric structure with polarization to-
ture and perhaps to eliminate the ferroelectric instability wards the BaO surface 共+兲 and the other towards the
entirely. Relevant quantities to examine include the rela- TiO2 surface 共−兲 were considered. The surface layers
tive stability of lower- and higher-symmetry phases, spa- were relaxed for the ideal and 共+兲 ferroelectric asym-
tial variation in polarization, changes in the average po- metrically terminated slabs and the ferroelectric BaO-
larization magnitude and direction, and the depth and BaO slab. It was found that the depolarization field of
shape of the ferroelectric double-well potential. These the ferroelectric slabs strongly destabilizes the ferroelec-
changes can also be expected to lead to changes in the tric state, as expected, even taking into account the en-
dielectric and piezoelectric response of thin films and ergy lowering due to the surface relaxation. In all slabs
superlattices, which can be studied theoretically and considered, this consists of an inward-pointing dipole
compared with experiments. The implications of the arising from the relative motion of surface cations and
various first-principles studies included in this review oxygen atoms. The average surface energy of the ideal
will be described below. BaO and TiO2 surfaces is 0.0574 eV/ Å2 = 0.923 eV per
surface unit cell.
Padilla and Vanderbilt 共1997兲 reported ultrasoft
3. Studies of individual one-component systems pseudopotential calculations with fully relaxed atomic
In this section, we describe a representative sampling coordinates for symmetrically terminated 共both BaO
of first-principles studies and their results. Most of the and TiO2兲 seven-layer BaTiO3 共001兲 slabs separated by
literature has concentrated on BaTiO3, providing a use- two lattice constants of vacuum. The in-plane lattice
ful comparative test of various first-principles implemen- constant was set equal to the theoretical equilibrium lat-
tations, as well as a benchmark for evaluation and analy- tice constant a computed for the bulk tetragonal phase
sis of results on other systems. We first consider the 共a = 3.94 Å兲. The average surface energy of the ideal
calculations for single slabs of pure material that focus BaO and TiO2 surfaces is 1.358 eV per surface unit cell;
on the properties of surfaces: surface relaxation, surface at least part of the difference relative by Cohen 共1996兲
reconstructions, and surface states. Depending on the could be due to the different lattice constant. Relax-
structural constraints, these calculations are relevant to ations were reported for unpolarized slabs and for po-
the surface of a semi-infinite bulk either for a freestand- larized slabs with polarization along 共100兲 共in the plane
ing thin film or for a thin film epitaxially constrained by of the slab兲. Deviations from the bulk structure were
a substrate. This will be followed by discussions of stud- confined to the first few atomic layers. The surface layer
ies of systems with two or more material components. relaxes substantially inwards, and rumples such that the

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1109

cation 共Ba or Ti兲 moves inward relative to the oxygen comparison between isolated and periodic slabs is pos-
atoms, as in Fig. 22. While the relaxation energy was sible. It was found that the surface charge and surface
found to be much greater than the ferroelectric double- dipoles of isolated slabs converge rapidly as a function
well depth, the in-plane component of the unit-cell di- of slab thickness and can be used, combined with a value
pole moment was relatively insensitive to the surface of ⑀⬁ taken from the bulk, to extract a spontaneous po-
relaxation, with a modest enhancement at the larization of 0.245 C / m2 共corrected to zero field using
TiO2-terminated surface and a small reduction at the the electronic dielectric constant ⑀⬁ = 2.76兲, compared
BaO-terminated surface. The relative stability of BaO with 0.240 C / m2 from a Berry-phase calculation. This is
and TiO2 terminations was compared and both found to only slightly less than the bulk value of 0.263 C / m2
be stable depending on whether the growth was under taken from experiment. The average surface energy for
Ba-rich or Ti-rich conditions. the two terminations of symmetrically terminated slabs
This investigation was extended by Meyer and is 0.85 eV per surface unit cell. Surface longitudinal dy-
Vanderbilt 共2001兲 to seven-layer and nine-layer polar- namical charges differ considerably from bulk values,
ized slabs with polarization along the normal. The prob- satisfying a sum rule that the dynamical charges at the
lem of the artificial vacuum field in this periodically re- surface planes add up to half of the corresponding bulk
peated slab calculation was addressed by the techniques value 共Ruini et al., 1998兲. Convergence of all quantities
of introducing an external dipole layer in the vacuum with slab and vacuum thickness of periodically repeated
region of the supercell described in Sec. IV.B.1. This slabs was found, in comparision, to be slow, with signifi-
technique can also be used to generate an applied field cant corrections due to the fictitious field in the vacuum
that partially or fully compensates the depolarization 共for polarized slabs兲 and the interaction between slab
field for BaTiO3 slabs. As a function of applied field, the images.
change in structure can be understood as arising from The isolated slab was also the subject of a FLAPW
oppositely directed electrostatic forces on the positively study 共Krcmar and Fu, 2003兲. The symmetric
charged cations and negatively charged anions, leading TiO2-TiO2 共nine-layer兲 and asymmetric TiO2-BaO 共ten-
to corresponding changes in the rumplings of the atomic layer兲 slabs were considered in a paraelectric structure
layers and field-induced increases of the layer dipoles. with a fixed to 4.00 Å and a polar tetragonal structure
Analysis of the internal electric field as a function of the with a and c equal to 4.00 and 4.04 Å, respectively. For
applied field allows determination of whether the slab is
the cubic TiO2-terminated slab, displacements in units of
paraelectric or ferroelectric. The BaO-terminated slab is
c for the surface Ti, surface O, subsurface Ba, and sub-
clearly ferroelectric, with vanishing internal electric field
surface O are −0.021, +0.007, +0.022, and −0.009 c, to be
at an external field of 0.05 a.u. and a polarization of
compared with the results 共Padilla and Vanderbilt, 1997;
22.9 ␮C cm−2, comparable to the bulk spontaneous po-
larization. The ferroelectric instability is suppressed in −0.0389,−0.0163,+0.0131,−0.0062兲 for a periodically re-
peated seven-layer slab with lattice constant 3.94 Å. The
the TiO2-terminated slab, which appears to be margin-
ally paraelectric. tetragonal phase was relaxed to a convergence criterion
The Hartree-Fock method was used by Cora and Cat- of 0.06 eV/ Å on the atomic forces; the rumplings of the
low 共1999兲 and by Fu et al. 共1999兲. Cora and Catlow layers follow overall the same pattern as that reported
共1999兲 performed a detailed analysis of the bonding us- by Meyer and Vanderbilt 共2001兲, with an inward-
ing tight-binding parametrization. For the 7-layer BaO- pointing surface dipole arising from surface relaxation,
terminated slab, the reported displacement of selected though the reduction of the rumpling in the interior is
Ti and O atoms is in good agreement with the results of not as pronounced as for the zero-applied field case by
Padilla and Vanderbilt 共1997兲, and these calculations Meyer and Vanderbilt. The energy difference between
were extended to slabs of up to 15 layers. Fu et al. 共1999兲 the paraelectric and ferroelectric slabs was not reported.
performed Hartree-Fock calculations for slabs of two to With interatomic potentials, it is possible to study ad-
eight atomic layers, with symmetric and asymmetric ter- ditional aspects of surface behavior in BaTiO3 thin films
minations. Using a localized basis set, they were able to and nanocrystals. The most important feature of inter-
perform calculations for isolated slabs as well as periodi- atomic studies of thin films relative to full first-principles
cally repeated slabs. Calculations of the macroscopically calculations is the relative ease of extending the super-
averaged planar charge density, surface energy, and sur- cell in the lateral direction, allowing the formation of
face dynamical charges were reported as a function of 180° domains and molecular dynamics studies of finite-
thickness and termination for a cubic lattice constant of temperature effects. Tinte and Stachiotti 共2001兲 studied
4.006 Å. The relative atomic positions were fixed to their a 15-layer TiO2-terminated slab periodically repeated
bulk tetragonal structure values 共note that this polarized with a vacuum region of 20 Å using interatomic poten-
structure in both isolated and periodic boundary condi- tials that had previously been benchmarked against first-
tions has a very high electrostatic energy and is not the principles surface relaxations and energies 共Tinte and
ground-state structure兲. This would significantly affect Stachiotti, 2000兲. The unconstrained 共stress-free兲 slab is
the comparison of the computed surface properties with found to undergo a series of phase transitions with de-
experiment. In particular, it is presumably responsible creasing temperature, from a paraelectric phase to ferro-
for the high value of the average surface energy re- electric phases, first with polarization in the plane along
ported 共1.69 eV per surface unit cell兲. However, a useful 共100兲, and then along 共110兲. Enhancement of the surface

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1110 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

terminated surface, which is found to be positive

共though small兲. The 共110兲 surfaces are found to have
much higher surface energy, except for the relaxed
“asymmetric O-terminated” surface where every
second-surface O atom is removed and the others oc-
cupy the same sites as in the bulk structure. In this struc-
ture, displacements of cations parallel to the surface are
found to lower substantially the surface energy. A simi-
lar study of the KNbO3 共110兲 surface 共Heifets, Kotonin,
and Jacobs 2000兲, which is the surface of most experi-
mental interest for this 1 + / 5+ perovskite, showed strong
relaxations extending deep below the surface, consistent
with suggestions that this surface has a complicated
chemistry.
Next, we consider results on the electronic structure
of BaTiO3 films, particularly the surface states. In the
FLAPW study of Cohen, the band gap of the ideal slab
is found to be reduced from the bulk. A primarily O
p-occupied surface state on the TiO2 surface was identi-
fied at M 共Cohen, 1996兲, with a primarily Ti d surface
state near the bottom of the conduction band. Analysis
of the ferroelectric BaO-terminated slab showed that
the macroscopic field resulted in a small charge transfer
FIG. 23. Cell-by-cell out-of-plane 共top panel兲 and in-plane to the subsurface Ti d states from the O p and Ba p
共bottom panel兲 polarization profiles of a randomly chosen
states at the other surface, making the surfaces metallic.
chain perpendicular to the slab surface for different misfit
Further study of this effect by Krcmar and Fu 共2003兲
strains ␩ at T = 0 K. In the in-plane polarization profiles py
= px. From Tinte and Stachiotti, 2001.
showed that for the symmetric TiO2 nine-layer slab the
ferroelectric distortion similarly shifts the top surface Ti
states and bottom surface O state toward the bulk mid-
polarization at low temperatures appears to be linked to gap, as in Fig. 24, resulting in a small charge transfer and
the existence of an intermediate temperature regime of a metallic character for the surfaces.
surface ferroelectricity. For slabs with a strongly com-
pressive epitaxial strain constraint 共␩ = −2.5% 兲, there is
a transition to a ferroelectric state with 180° domains b. PbTiO3
and polarization along the surface normal. At the sur- In contrast to the numerous papers on calculations on
face, the polarization has a nonzero x component and a the surfaces of BaTiO3, there are relatively few for the
reduced z component, giving a rotation at the surface related material PbTiO3. Regarding surface relaxations
layer. The width of the stripe domains cannot be deter- and energies, it was found by Meyer and Vanderbilt
mined, as it is limited by the lateral supercell size at least 共2001兲 that the two compounds are quite similar. In per-
up to 10⫻ 10. Reducing the compression to ␩ = −1.0% pendicularly polarized films, it seems that both termina-
also gives a 180° domain structure in the z component of tions give ferroelectric films if the depolarization field is
the polarization, combined with a nonzero component compensated, consistent with the stronger ferroelectric
along 关110兴 in the interior of the film as well as the sur- instability of PbTiO3 and the microscopic model analysis
face, as can be seen from Fig. 23. of Ghosez and Rabe 共2000兲, the latter not including the
The thickness dependence of the transition tempera- effects of surface relaxation. Because of the larger spon-
ture to this ferroelectric domain phase was studied at taneous polarization of PbTiO3, it is not possible to com-
␩ = −1.5%, with Tc decreasing from the 15-layer film to pensate fully the depolarization field using the dipole-
the 13-layer and lowest Tc 11-layer film. At and above a layer technique.
critical thickness of 3.6 nm, stress-free films exhibit the There are important differences between A-site Ba
same ferroelectric-domain ground-state structure. and Pb, which are evident even for the bulk. While the
Heifets and co-workers 共Heifets, Kotonin, and Maier, polarization in BaTiO3 is dominated by the Ti displace-
2000; Heifets et al., 2001a兲 studied BTO 共001兲 and 共110兲 ments, Pb off-centering contributes substantially to the
surfaces of an isolated slab using the shell model of spontaneous polarization of PbTiO3; this can be linked
Heifets, Kotonin, and Maier. Between 1 and 16 atomic to the much richer chemistry of Pb oxides compared to
planes were relaxed in the electrostatic potential of a alkaline-earth oxides. One downside is that it is more
rigid slab of 20 atomic planes whose atoms were fixed in challenging to construct accurate interatomic potentials
their perfect 共presumably cubic兲 lattice sites. The re- for perovskites with Pb than with alkaline-earth A-site
laxed structures of the two 共001兲 terminations are in cations 共Sepliarsky et al., 2004兲. In the surface, the char-
good agreement with other calculations, except for the acteristic behavior of Pb leads to an antiferrodistortive
sign of the surface dipole in the relaxation of the BaO- surface reconstruction of the 共001兲 PbO-terminated sur-

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1111

a few representative results drawn from the very exten-

sive literature on this subject.
Many of the studies of BaTiO3 discussed above in-
cluded analogous calculations for SrTiO3. As already
noted in the discussion of PbTiO3, the 共001兲 surface re-
laxations and energies of the nonpolar slab are very
similar for all three materials. First-principles surface re-
laxation for the SrO surface is reported 共in units of a
= 3.86 Å兲 for surface Sr, surface O, subsurface Ti, and
subsurface O as −0.057, 0.001, 0.012, and 0.0, respec-
tively 共Padilla and Vanderbilt, 1998兲, to be compared
with the values −0.071, 0.012, 0.016, and 0.009 共com-
puted using a shell model with the experimental lattice
constant 3.8969 Å兲. Similarly, first-principles surface re-
laxation for the TiO2 surface is reported for surface Ti,
surface O, subsurface Sr, and subsurface O as −0.034,
−0.016, +0.025, and −0.005, respectively 共Padilla and
Vanderbilt, 1998兲, to be compared with the values
−0.030, −0.017, +0.035, and −0.021. The average energy
of the two surfaces is found to be 1.26 eV per surface
unit cell. A detailed comparison of various Hartree-
Fock and density-functional implementations showed
generally good agreement for the surface relaxation
共Heifets et al., 2001b兲. Inward surface dipoles due to re-
laxation are found for both terminations, with the TiO2
termination smaller in magnitude than BaTiO3 共Heifets,
Kotonin, and Maier, 2000兲. The possibility of an in-plane
ferroelectric instability at the surface was examined and
found to be quite weak 共Padilla and Vanderbilt, 1998兲. In
FIG. 24. 共Color online兲 共a兲 Paraelectric-phase energy-band
these studies, the antiferrodistortive instability exhibited
structure of a nine-layer slab of BaTiO3. The surface states in
by bulk SrTiO3 at low temperatures was suppressed by
共or near兲 the band gap are highlighted. 共b兲 As in 共a兲, for the
ferroelectric nine-layer slab. From Krcmar and Fu, 2003. the choice of a 1 ⫻ 1 in-plane unit cell. The surface elec-
tronic band structures show a behavior similar to that of
BaTiO3 described above.
face 共Munkholm et al., 2002; Bungaro and Rabe, 2005; For SrTiO3, there is considerable evidence for a wide
Sepliarsky et al., 2005兲. Specifically, first-principles calcu- variety of surface reconstructions of varying stoichiom-
lations 共Bungaro and Rabe, 2005兲 show that the recon- etry, depending on conditions such as temperature and
struction in the subsurface TiO2 layer occurs only for the oxygen partial pressure as well as the relative chemical
PbO termination and not for TiO2 termination, and also potentials of TiO2 and SrO. Candidate structures can be
that if the Pb in the surface layer is replaced by Ba, the obtained by creating vacancies on the surface 共for ex-
reconstruction is suppressed. ample, missing rows of oxygen兲 and adding adatoms
共Kubo and Nozoye, 2003兲. More drastic rearrangements
of the surface atoms have also been proposed, for ex-
c. SrBi2Ta2O9 ample, a 共2 ⫻ 1兲 Ti2O3 reconstruction 共Castell, 2002兲 and
A first-principles study of an isolated BiO2-terminated a 共2 ⫻ 1兲 double-layer TiO2 reconstruction with edge-
slab of SrBi2Ta2O9 共SBT兲 one lattice constant thick sharing TiO6 octahedra 共Erdman et al., 2002兲. As in the
共composition SrBi2Ta2O11兲 was reported by Tsai et al. case of semiconductor surface reconstructions, first-
共2003兲. Spin-polarized calculations showed that such a principles calculations of total energies are an essential
film would be ferromagnetic as well as ferroelectric. This complement to experimental structural determination
intriguing possibility suggests further investigation. and can also be used to predict scanning tunneling mi-
croscope images for comparison with experiment
共Johnston et al., 2004兲. Even so, there are still many
d. SrTiO3 and KTaO3 open questions about the atomic and electronic struc-
In addition to the work on true ferroelectrics, there ture of SrTiO3 surfaces under various conditions. The
has been interest in first-principles studies of surfaces same applies to KTaO3; the structures and lattice dy-
and heterostructures of incipient ferroelectrics, mainly namics of a variety of 共1 ⫻ 1兲 and 共2 ⫻ 1兲 surface struc-
SrTiO3 and, to a lesser extent, KTaO3. As these have tures were studied by Li, Akhadov, et al. 共2003兲.
closely related properties that can illuminate issues in As previously mentioned, the theoretical and experi-
the ferroelectric perovskites, we include a description of mental literature on SrTiO3 is so extensive that it would

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1112 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

require a review paper in its own right to cover it fully. treated implicitly by constraining the in-plane lattice
Since we expect the ferroelectric instability in related constant of the supercell to that of bulk SrTiO3. For
systems such as BaTiO3 and PbTiO3 to make the physics each BaTiO3 thickness, the system was relaxed assuming
more, not less, complicated, this suggests that we have a nonpolar state for BaTiO3, and the energy of the bulk-
only scratched the surface in developing a complete un- like tetragonal distortion was computed as a function of
derstanding of the surfaces of perovskite ferroelectrics overall amplitude of the distortion. Above a critical
and the resulting effects on thin-film properties. thickness of six unit cells, this distortion lowered the en-
ergy, demonstrating the development of a ferroelectric
instability. As discussed in Sec. IV.B.5, this finite-size ef-
4. Studies of individual heterostructures
fect can be largely understood by considering the imper-
Now we turn to the description of studies of systems fect screening in the metal layers.
with two or more material components. The main struc- Considerable first-principles effort has been devoted
tural issues are the rearrangements at the interface, the to investigating various aspects of epitaxial ferroelectric
change in electrical and mechanical boundary conditions thin films on Si. As perovskite oxides cannot be grown
felt by each constituent layer, and how these changes directly on Si, an approach developed by McKee and
modify the ferroelectric instability exhibited by the sys- Walker 共McKee et al., 1998, 2001兲 is to include an AO
tem as a whole. This geometry also allows the calcula- buffer layer, which apparently also results in the forma-
tion of band offsets and/or Schottky barriers, crucial in tion of a silicide interface phase. The constituent layers
principle to understanding the electronic behavior of this heterostructure should thus be considered to be
共though with the caveat that the measured Schottky bar- Si/ ASi2 / AO/ ABO3. The full system has not been simu-
rier in real systems is influenced by effects such as oxi- lated directly, but first-principles approaches have been
dation of the electrodes that are not included in the used to investigate individual interfaces. The importance
highly idealized geometries studied theoretically兲. The of relaxations, the additional role of the buffer layer in
current state of knowledge, derived from experimental changing the band offset, and the analysis of electronic
measurements, is described in Sec. III.B.3. structure within the local-density approximation are il-
The first combination we discuss is that of a ferroelec- lustrated by the following.
tric thin film with metallic electrodes. Transition-metal McKee et al. 共2003兲 presented first-principles results
interfaces with nonpolar BaO-terminated layers of for the atomic arrangements and electronic structure in
BaTiO3 were studied by Rao et al. 共1997兲, specifically the Si/ ASi2 / AO system, in conjunction with an experi-
systems of three and seven atomic layers of BaTiO3, mental study. A strong correlation is found between the
with the lattice constant set to the bulk value of 4.00Å, valence-band offset and the dipole associated with the
combined with top and bottom monolayers of Ta, W, Ir, A-O bond linking the A atom in the silicide to the O
and Pt representing the electrodes. The preferred ab- atom in the oxide, shown in Fig. 25. It is thus seen that
sorption site for the metal atoms was found to be above the structural rearrangements in the interface are a key
the O site, with calculated metal-oxygen distances rang- determining factor in the band offset.
ing from 2.05 Å for Ta to 2.11 Å for Pt. The BaTiO3 A detailed examination of the interface between the
slabs were assumed to retain their ideal cubic structure. perovskite oxide and the alkaline-oxide buffer layer,
Analysis of the partial density of states of the hetero- specifically BaO/ BaTiO3 and SrO/ SrTiO3, was carried
structure shows that the Pt and Ir Fermi energies lie in out by Junquera et al. 共2003兲. A periodic 1 ⫻ 1 ⫻ 16 su-
the gap of the BaTiO3 layer at 0.94 and 0.64 eV, respec- percell was chosen with stacking of 共001兲 atomic layers:
tively, above the top of the valence band 共this is, fortu- 共AO兲n − 共AO − TiO2兲m, with n = 6 and m = 5. Two mirror-
nately, smaller than the underestimated computed gap symmetry planes were fixed on the central AO and BO2
of 1.22 eV for the BaTiO3 slab兲. Using the experimental layers, and the in-plane lattice constant chosen for per-
gap of 3.13 eV, a Schottky barrier height of 2.19 eV for fect matching to the computed local-density approxima-
Pt and 2.49 eV for Ir is thus obtained. Experimentally, tion lattice constant of Si 共this epitaxial strain constraint
however, the Schottky barrier is known to be substan- is the only effect of the Si substrate included in the cal-
tially lower for Ir than for Pt, illustrating the limitations culation兲. Relaxations within the highest-symmetry te-
mentioned in the previous paragraph. tragonal space group consistent with this supercell were
Robertson and Chen 共1999兲 combined first-principles performed. Analysis of the partial density of states
calculations of the charge neutrality levels with experi- showed no interface-induced gap states. The main effect
mental values of the band gap, electronic dielectric con- observed for relaxations was to control the size of the
stant ⑀⬁, the electron affinity, and the empirical param- interface dipole, which in turn was found to control the
eter S, described in Sec. III.B.3. Values for SrTiO3 and band offsets, shown here in Fig. 26.
PZT were reported for Pt, Au, Ti, Al, and the conduc- As in the studies described above, the conduction-
tion and valence bands of Si. band offset is obtained from the computed valence-band
To explore how the electrodes affect the ferroelectric offset using the bulk experimental band gap. These re-
instability of the film, Junquera and Ghosez 共2003兲 con- sults were combined with offsets reported for other rel-
sidered a supercell of five unit cells of metallic SrRuO3 evant interfaces to estimate band alignments for
and 2–10 unit cells of BaTiO3, with a SrO/ TiO2 interface Si/ SrO/ SrTiO3 / Pt and Si/ BaO/ BaTiO3 / SrRuO3 het-
between SrRuO3 and BaTiO3. A SrTiO3 substrate was erostructures, confirming that the AO layer introduces

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1113

5. First-principles modeling: methods and lessons

As discussed above, the analysis and prediction of the

behavior of ferroelectric thin films and heterostructures
can be carried out with direct first-principles simulations
only for highly idealized configurations. However, it is
possible to consider more complex and realistic situa-
tions by constructing models that incorporate certain
physical ideas about the nature of these systems, with
material-specific parameters determined by fitting re-
sults of first-principles calculations carefully selected for
a combination of informativeness and tractability. This
modeling approach also has the advantage of providing
a conceptual framework for organizing the vast amount
of microscopic information in large-scale first-principles
FIG. 25. 共Color兲 Illustration of three layers of the alkaline- calculations, and communicating those results, particu-
earth oxide on the 共001兲 face of silicon observed in a cross larly to experimentalists. This will not diminish in impor-
section at the 关110兴 zone axis 共blue, alkaline-earth metal; yel- tance even as such calculations become easier with con-
low, oxygen; and green, silicon兲. A distinct interface phase can tinuing progress in algorithms and computer hardware.
be identified as a monolayer structure between the oxide and For successful modeling of measurable physical prop-
the silicon in which the charge density in interface states is
erties, the film must be considered as part of a system
strongly localized around the silicon atoms in the interface
共substrate+ film+ electrode兲 as all components of the
phase. The dipole in the ionic A-O bond between the alkaline-
earth metal in the silicide and the oxygen in the oxide buffers system and their interaction contribute to determine
the junction against the electrostatic polarization of the inter- properties such as the switchable polarization and the
face states localized on silicon. The electron density of this dielectric and piezoelectric responses. We start by con-
valence surface state at the center of the Brillouin zone is sidering the class of first-principles models in which the
shown with the purple isosurface 共0.3⫻ 10−3 e兲. Reprinted with constituent layers 共film, superlattice layers, electrodes,
permission from McKee et al., 2003. © 2003, AAAS. and substrate兲 are assumed to be subject to macroscopic
electric fields and stresses resulting from the combina-
an electrostatic barrier of height greater than 1 eV. This tion of applied fields and stresses and the effects of the
is sufficient to eliminate the carrier injection from the Si other constituents, with the responses of the layers being
into the conduction-band states of the perovskite that given to lowest order by the bulk responses. For systems
would occur if the two were in direct contact 共Robertson with constituent layer thicknesses as low as one bulk
and Chen, 1999兲. lattice constant, it seems at first unlikely that such an

FIG. 26. Schematic representation of the valence-band offset 共VBO兲 and the conduction-band offset 共CBO兲 for 共a兲 BaO/ BaTiO3
expt
and 共b兲 SrO/ SrTiO3 interfaces. Ev, Ec, and Egap stand for the top of the valence band, the bottom of the conduction band, and the
experimental band gap, respectively. Values for Ev, measured with respect to the average of the electrostatic potential in each
material, are indicated. The solid curve represents the profile of the macroscopic average of the total electrostatic potential across
the interface. ⌬V stands for the resulting lineup. The in-plane lattice constant was set equal to the theoretical value of Si 共5.389兲.
The size of the supercell corresponds to n = 6 and m = 5. From Junquera et al., 2003.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1114 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

approach could be useful, but in practice it has been parameters for use in calculations such as those by
found to be surprisingly successful. Speck and Pompe 共1994兲, Alpay and Roytburd 共1998兲,
One simple application of this approach has been Bratkovsky and Levanyk 共2001兲, and Li, Akhadov, et al.
used to predict and analyze the strain in nonpolar thin 共2003兲. The effects of inhomogeneous strain due to misfit
films and multilayers. In the construction of the refer- dislocations that provide elastic relaxation in thicker
ence structure for the AO / ABO3 interfaces by Junquera films have also been argued to be significant.
et al. 共2003兲, macroscopic modeling of the structure with Next we consider the application of these “con-
bulk elastic constants for the constituent layers yielded tinuum” models to analyzing structures in which the
accurate estimates for the lattice constants along the macroscopic field is allowed to be nonzero. Macroscopic
normal direction. In cases of large lattice mismatch, very electrostatics is applied to the systems of interest by a
high strains can be obtained in very thin films and non- coarse graining over a lattice-constant-scale window to
linear contributions to the elastic energy can become yield a value for the local macroscopic electric potential.
important. These can be computed with a slightly more Despite the fact that this is not strictly within the regime
sophisticated though still very easy-to-implement of validity of the classical theory of macroscopic electro-
method that has been developed to study the effects of statics, which requires slow variation over many lattice
epitaixial strain more generally on the structure and constants, this analysis turns out to be remarkably useful
properties of a particular material, described next. for first-principles results. In the simplest example, the
As is discussed in Sec. V, the effects of epitaxial strain polarization of a polar BaTiO3 slab 共periodically re-
in ultrathin films and heterostructures have been identi- peated in a supercell with vacuum兲 is accurately repro-
fied as a major factor in determining polarization-related duced using bulk values for the bulk spontaneous polar-
properties, and have been the subject of intense interest ization and electronic dielectric constant even for slabs
in both phenomenological and first-principles modeling. as thin as two lattice constants 共Junquera and Rabe,
In particular, for ferroelectric perovskite oxides it has 2005兲.
long been known that there is a strong coupling between We have already mentioned in the previous section
strain 共e.g., pressure-induced兲 and the ferroelectric insta- that perpendicular 共to the surface兲 polarization can lead
bility, as reflected by the frequency of the soft mode and to a nonzero macroscopic field that opposes the polar-
the transition temperature. In both phenomenological ization 共the depolarizing field兲. Unless compensated by
and first-principles studies, it has become common to fields from electrodes or applied fields, this strongly de-
study the effects of epitaxial strain induced by the sub- stabilizes the polarized state. In systems with two or
strate by studying the structural energetics of the more distinct constituent layers, this condition in the ab-
strained bulk. Specifically, two of the lattice vectors of a ជ = 0. For ex-
sence of free charge favors states with ⵱ · P
bulk crystal are constrained to match the substrate and
ample, in the first-principles calculations of short-period
other structural degrees of freedom are allowed to relax,
as described in the previous paragraph. In most cases, BaTiO3 / SrTiO3 superlattices, the local polarization
these calculations are performed for zero macroscopic along 关001兴 is found to be quite uniform in the two layers
electric field, as would be the case for a film with perfect 共Neaton and Rabe, 2003; Johnston, Huang, et al., 2005兲,
short-circuited electrodes. Indeed, it is often the case though the in-plane component can be very different
共Pertsev et al., 1998; Junquera et al., 2003; Neaton and 共Johnston, Huang, et al., 2005兲. To the extent that lay-
Rabe, 2003兲 that the strain effect is considered to be the ered ferroelectrics such as SBT can be treated in this
dominant effect of the substrate, which is otherwise not macroscopic framework, one similarly expects that po-
included 共thus greatly simplifying the calculation兲. At larization along c will not tend to be energetically fa-
zero temperature, the sequence of phases and phase vored since the layers separating the polarized perovs-
boundaries can be readily identified as a function of in- kitelike layers typically have low polarizability 共Fennie
plane strain directly through total-energy calculations of and Rabe, 2005兲. This observation provides a theoretical
the relaxed structure subject to the appropriate con- framework for evaluating claims of large ferroelectric
straints. Atomic-scale information can be obtained for polarization along c in layered compounds 共Chon et al.,
the precise atomic positions, band structure, phonon fre- 2002兲; unless there is an unusually high polarizability for
quencies, and eigenvectors. The temperature axis in the the nonperovskite layers, or a strong competing contri-
phase diagram can be included by using effective Hamil- bution to the energy due, for example, to the interfaces
tonian 共or interatomic potential兲 simulations. Results for to help stabilize a high c polarization, other reasons for
selected perovskite oxides are discussed in Sec. V; a the observations need to be considered 共Garg et al.,
similar analysis was reported for TiO2 or by Montanari 2003兲.
and Harrison 共2004兲. These considerations become particularly important
This modeling is based on the assumption that the for ultrathin films with metal electrodes. The limiting
layer stays in a single-domain state. As discussed in Sec. case of complete screening of the depolarizing field by
V, the possibility of strain relaxation through formation perfect electrodes is never realized in real thin-film sys-
of multidomain structures must be allowed for. While tems. The screening charge in real metal electrodes is
this cannot be readily done directly in first-principles spread over a characteristic screening length and is asso-
calculations, first-principles data on structural energetics ciated with a voltage drop in the electrode. For thick
for large misfit strains could be used to refine Landau films, this can be neglected, but the relative size of the

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1115

voltage drop increases as the film thickness decreases. Electronic states associated with surfaces and inter-
This has been identified as a dominant contribution to faces will also contribute to determining the equilibrium
the relation between the applied field and the true field configuration of electric fields and polarizations. In the
in the film for the thinnest films 共Dawber, Chandra, et simple example of periodically repeated slabs separated
al., 2003兲. One way this shows up is in the thickness by vacuum, as the slab gets thicker, a breakdown is ex-
dependence of the apparent coercive field; it is found pected where the conduction-band minimum on one sur-
that the true coercive field scales uniformly down to the face of the slab falls below the valence-band maximum
thinnest films. Effects are also expected on the structure on the other. In this case, charge will be transferred
and polarization. While films with partial compensation across the slab with the equilibrium charge 共for fixed
of the depolarization field may still exhibit a ferroelec- atomic positions兲 being determined by a combination of
tric instability, the polarization and the energy gain rela- the macroscopic electrostatic energy and the single-
tive to the nonpolar state are expected to decrease. This particle density of states. This tends partially to screen
simple model was developed and successfully used by the depolarization field. The role of interface states in
Junquera and Ghosez 共2003兲 to describe the thickness screening the depolarization field in the film has been
dependence of the ferroelectric instability in a BaTiO3 discussed in a model for BaTiO3 on Ge 共Reiner et al.,
film between SrRuO3 electrodes. This analysis identified 2004兲. The presence of surface and interface states can
be established by examination of the band structure and
the thickness dependence of the residual depolarization
partial density of states, as discussed in the previous sec-
field as the principal source of thickness dependence in
tion.
this case. Lichtensteiger et al. 共2005兲 suggest that the re-
Finally, we turn our attention to an analysis of what
duction of the uniform polarization by the residual field
the discussion above tells us about finite-size effects in
and its coupling to tetragonal strain is the cause of the
ferroelectric thin films. We have seen that many factors
decrease in tetragonality with decreasing thickness of
contribute to the thickness dependence of the ferroelec-
PbTiO3 ultrathin films. tric instability: the thickness dependence of the depolar-
It is well known that 180° domain formation provides ization field, the gradual relaxation of the in-plane lat-
an effective mechanism for compensating the depolar- tice constant from full coherence with the substrate to its
ization field, and is expected to be favored when the bulk value, and the changing weight of the influence of
screening available from electrodes is poor or nonexist- surfaces and interfaces. The “true” finite-size effect, i.e.,
ent 关for example, on an insulating substrate 共Streiffer et the modification of the collective ferroelectric instability
al., 2002兲兴. Instability to domain formation is discussed due to the removal of material in the film relative to the
by Bratkovsky and Levanyuk 共2000兲 as the result of a infinite bulk, could possibly be disentangled from the
nonzero residual depolarization field due to the pres- other factors by a carefully designed first-principles cal-
ence of a passive layer. Similarly, a phase transition from culation, but this has not yet been done. We speculate
a uniform polarized state to a 180° domain state with that this effect does not universally act to suppress fer-
zero net polarization is expected to occur with decreas- roelectricity, but could, depending on the material, en-
ing thickness 共Junquera et al., 2003兲. hance ferroelectricity 共Ghosez and Rabe, 2000兲.
Despite the usefulness of macroscopic models, it
should not be forgotten that they are being applied far
outside the regime of their formal validity 共i.e., length 6. Challenges for first-principles modeling
scales of many lattice constants兲 and that atomistic ef- First-principles calculations have advanced tremen-
fects can be expected to play an important role, espe- dously in the last decade, to the point where systems of
cially at the surfaces and interfaces. The structural ener- substantial chemical and structural complexity can be
getics could be substantially altered by relaxations and addressed, and a meaningful dialog opened up between
reconstructions 共atomic rearrangements兲 at the surfaces experimentalists and theorists. With these successes the
and interfaces. These relaxations and reconstructions bar gets set ever higher, and the push is now to make the
are also expected to couple to the polarization 共Meyer theory of ferroelectrics truly realistic. The highest long-
and Vanderbilt, 2001; Bungaro and Rabe, 2005兲 with the term priorities include making finite-temperature calcu-
possibility of either enhancing or suppressing the switch- lations routine, proper treatment of the effects of defects
able polarization. The surfaces and the interfaces will and surfaces, and the description of structure and dy-
also be primarily responsible for the asymmetry in en- namics on longer length and time scales. In addition,
ergy between up and down directions for the polariza- there are specific issues that have been raised that may
tion. For ultrathin films, the surface and interface energy be addressable in the shorter term through the interac-
can be important enough to dominate over elastic en- tion of theory and experiment, and the rest of this sec-
ergy, leading to a possible tradeoff between lattice tion will highlight some of these.
matching and atomic-scale matching for favorable bond- Many applications depend on the stability of films
ing at the interface. These surface and interface energies with a uniform switchable polarization along the film
could even be large enough to stabilize nonbulk phases normal. This stability depends critically on compensa-
with potentially improved properties. This should be es- tion of the depolarization field. Understanding and con-
pecially significant for interfaces between unlike materi- trolling the compensation mechanism共s兲 are thus the
als. subjects of intense current research interest. There are

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1116 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

two main classes of mechanism: compensation by “free” as a whole rather than via a domain-wall mechanism has
charges 共in electrodes/substrate or applied fields兲 and been raised for ultrathin films of PVDF 共Bune et al.,
compensation by the formation of polarization domains. 1998兲, while a different interpretation has been offered
On insulating substrates, this latter alternative has been by Dawber, Chandra, et al. 共2003兲. Some progress has
observed and characterized in ultrathin films 共Streiffer et been made using interatomic potentials for idealized
al., 2002; Fong et al., 2004兲. It has been proposed that defect-free films, though real systems certainly are af-
domain formation occurs in films on conducting sub- fected by defects responsible for such phenomena as im-
strates at very low thicknesses as well as where the print and fatigue. Ongoing comparison of characteristics
finite-screening length in realistic electrodes inhibits that such as coercive fields, time scales, material sensitivity,
mechanism of compensation 共Junquera et al., 2003兲. The and thickness dependence of domain-wall nucleation,
critical thickness for this instability depends on the formation energy, and motion with experimental studies
domain-wall energy. This is expected to be different in promises that at least some of these issues will soon be
thin films than in bulk, one factor being that the bulk better understood.
atomic plane shifts across the domain walls. To conclude this section, we emphasize that it is not
Compensation of the depolarization field by free very realistic to expect first-principles calculations quan-
charges appears to be the dominant mechanism in films titatively to predict all aspects of the behavior of chemi-
on conducting substrates 共even relatively poor conduc- cally and structurally complex systems such as ferroelec-
tors兲 with or without a top electrode. In the latter case tric thin films, although successful predictions should
there must be free charge on the surface; the challenge is continue to become increasingly possible and frequent.
to understand how the charge is stabilized. There are Rather, the quantitative microscopic information and
also unresolved questions about how the charge is dis- the development of a useful conceptual framework con-
tributed at the substrate-film interface and how this tribute in a close interaction with experiment to build an
couples to local atomic rearrangements. Asymmetry of understanding of known phenomena and to propel the
the compensation mechanism may prove to be a signifi- field into exciting new directions.
cant contribution to the overall up-down asymmetry in
the film discussed in the previous section. A better un- V. STRAIN EFFECTS
derstanding could lead to the identification of system
configurations with more complete compensation and Macroscopic strain is an important factor in determin-
thus an enhancement of stability. ing the structure and behavior of very thin ferroelectric
The study of the behavior of ferroelectrics in applied films. The primary origin of homogeneous film strain is
electric fields also promises progress in the relatively lattice mismatch between the film and the substrate. In
near future. Recently, with the solution of long-standing addition, defects characteristic of thin films can produce
questions of principle, it has become possible to perform inhomogeneous strains that can affect the properties of
density-functional-theory calculations for crystalline sol- thicker relaxed films of technological relevance. Because
ids in finite electric fields 共Souza et al., 2002兲. In ferro- of the strong coupling of both homogeneous and inho-
electrics, this allows the investigation of nonlinearities in mogeneous strains to polarization, these strains have a
structure and polarization at fields relevant to experi- substantial impact on the structure, ferroelectric transi-
ments and the possibility of more accurate modeling of tion temperatures, and related properties such as the di-
constituent layers of thin-film and superlattice systems electric and piezoelectric responses, which has been the
subject to nonzero fields. It is also of interest to ask what subject of extensive experimental and theoretical inves-
the intrinsic breakdown field would be in the absence of tigation.
defects, though the question is rather academic with re- The largest effects are expected in coherent epitaxial
spect to real systems. films. These films are sufficiently thin that the areal elas-
The nonzero conductivity of real ferroelectrics be- tic energy density for straining the film to match the
comes particularly important for thinner films, since a substrate at the interface is less than the energy cost for
higher concentration of free carriers is expected to be introducing misfit dislocations to relax the lattice param-
associated with characteristic defects in the film, and eters back towards their unconstrained equilibrium val-
also because a given concentration of free carriers will ues. 关We note that for ultrathin films, the relaxed in-
have a more significant impact as thickness decreases. plane lattice constant will not in general be the same as
Free carriers can at least partially screen macroscopic the bulk lattice constant, and the former is more appro-
electric fields. At the macroscopic level, the concepts of priate for computing lattice mismatch 共Rabe, 2005兲兴.
band bending and space charge arising in semiconductor Very high homogeneous strains, of the order of 2%, are
physics can be applied to thin-film ferroelectrics, while a achievable. For example, barium titanate 共BTO兲 films on
correct atomic-scale treatment of these effects could be strontium titanate 共STO兲, with a bulk mismatch of 2.2%,
important to describing the behavior of ultrathin films. remain coherent in equilibrium up to a critical thickness
The physics of switching presents a significant chal- of 2–4 nm 共Sun et al., 2004兲. With low-temperature
lenge, requiring description of structure and dynamics growth techniques, the formation of misfit dislocations is
on long length and time scales. The questions of what kinetically inhibited and coherent films can be grown to
changes, if any, occur in switching as films become thin- thicknesses of two to three times the critical thickness
ner continue to be debated. The possibility of switching 共Choi et al., 2004兲. Even these films, however, are much

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1117

thinner than the minimum 120 nm thickness for films

used in contemporary applications, and thus much of the
discussion is at present primarily of fundamental rather
than technological interest.
The structure of a coherent film can be a single-crystal
monodomain structure, a polydomain structure, or even
possibly multiphase. We discuss the simplest single-
crystal monodomain case first. The phase diagram as a
function of in-plane strain will in general include lower
symmetry phases due to the symmetry-breaking charac-
ter of the epitaxial constraint. A nomenclature for these
phases of perovskites on a surface with square symmetry
关e.g., a perovskite 共001兲 surface兴 has been established by
Pertsev et al. 共1998兲. For example, a ferroelectric perov-
skite rhombohedral phase will be lowered to monoclinic
symmetry 共called the r phase兲. For highly compressive
FIG. 27. Phase diagram of a 共001兲 single-domain PbTiO3 thin
in-plane strains, coupling between strain and the polar- film epitaxially grown on different cubic substrates providing
ization tends to favor the formation of a tetragonal various misfit strains in the heterostructures. The second- and
phase with polarization along c 共the c phase兲 for highly first-order phase transitions are shown by thin and thick lines,
compressive strains. Conversely, highly tensile strains respectively. From Pertsev et al., 1998.
lead to an orthorhombic phase with polarization along
the cube face diagonal perpendicular to the normal 共the
aa phase兲 or along the in-plane Cartesian direction 共the the constrained dielectric and piezoelectric responses
a phase兲. As a result of the added constraint, it is pos- are clamped, as will be discussed further below兲. The
sible in principle to stabilize perovskite-derived phases nearly vertical morphotropic phase boundary character-
not observed in bulk, for example, the monoclinic r istic of bulk PZT is substantially modified 共Li,
phase and the orthorhombic aa phase in PbTiO3. This Choudhury, et al., 2003; Pertsev et al., 2003兲. In SrTiO3,
mechanism also plays a role in the more general phe- ferroelectricity is found to be induced by both suffi-
nomenon of epitaxial stabilization, discussed, for ex- ciently compressive and tensile strains with a corre-
sponding direction for the spontaneous polarization
ample, by Gorbenko et al. 共2002兲.
As discussed in Sec. IV.B.5, theoretical analysis of the 共c-type and aa-type兲 共Pertsev et al., 2000a; Antons et al.,
in-plane-strain phase diagrams has focused on isolating 2005兲, as shown in Fig. 28.
the effects of strain by computing bulk single-crystal The enhancement of the polarization in the c phase by
monodomain phase diagrams under the epitaxial con- compressive in-plane strain has been noted for BaTiO3
straint and zero macroscopic electric field using phe- 共Neaton and Rabe, 2003兲 and PZT 共Pertsev et al., 2003兲.
nomenological Landau theory or first-principles meth- For both strained SrTiO3 and strained BaTiO3, the
ods. Phenomenological analysis based on Landau- ferroelectric Tc’s are predicted to increase as the strain
Devonshire theory for a number of perovskite oxides magnitudes increase 共Choi et al., 2004; Haeni et al.,
has been presented by Pertsev and co-workers 共Pertsev 2004兲, as shown in Fig. 29.
et al., 1998, 1999, 2000a, 2000b, 2003; Koukhar et al., The use of phenomenological bulk Landau param-
eters yields a very accurate description for small strains
2001兲, with temperature-strain diagrams for BaTiO3,
near the bulk Tc. However, different parameter sets
PbTiO3 共see Fig. 27兲, SrTiO3, and Pb共Zr, Ti兲O3 共PZT兲,
have been shown, for example, in the case of BaTiO3, to
the latter generalized to include nonzero stress.
First-principles methods have been used to construct a
temperature-strain diagram for BaTiO3 共Dieguez et al.,
2004兲 and a zero-temperature–strain diagram for
PbTiO3 and ordered PZT 共Bungaro and Rabe, 2004兲
and SrTiO3 共Antons et al., 2005兲. These theoretical
phase diagrams have some notable features. In particu-
lar, compressive in-plane strain is found to elevate the
ferroelectric 共c兲 -paraelectric transition temperatures in
BaTiO3 and PbTiO3, and tensile in-plane strain elevates
the ferroelectric 共aa兲 -paraelectric transition tempera-
tures. In both cases, the transition is second order, in
contrast to the first-order transition in bulk. To eliminate
a possible source of confusion, we comment that zero
misfit strain as defined by Pertsev et al. 共1998兲 is not FIG. 28. Polarization of SrTiO3 as a function of in-plane
equivalent to an unconstrained film 共the low- strain. Solid circles and squares denote polarization along 关001兴
temperature bulk phases are in general not cubic, and and 关110兴, respectively. From Antons et al., 2005.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1118 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

FIG. 30. Inverses of the eigenvalues of the phonon contribu-

tion to the dielectric tensor as a function of the in-plane lattice
constant for the 关001兴-共PbTiO3兲1共PbZrO3兲1 superlattice. The
lines are a guide to the eye. From Bungaro and Rabe, 2004.

rotate the polarization 共Wu and Krakauer, 1992兲. This

polarization rotation has been identified as a key mecha-
nism in the colossal piezoresponse of single-crystal re-
laxors 共Park and Shrout, 1997; Fu and Cohen, 2000兲. The
sensitivity of the zero-field responses should also be re-
flected in the nonlinear response; thus the electric-field
tunability of the dielectric response can be adjusted by
FIG. 29. Expected Tc of 共a兲 共001兲 BaTiO3 共reprinted with per- changing the misfit strains 共Chen et al., 2003兲.
mission from Choi et al., 2004; © 2004, AAAS兲 and 共b兲 共001兲 While the phase diagrams thus derived are quite rich,
SrTiO3 共from Haeni et al., 2004; © 2004, Nature Publishing even the optimal single-crystal monodomain structure
Group, http: //www.nature.com兲 based on thermodynamic may be unfavorable with respect to formation of poly-
analysis. The range of transition represents the uncertainty in domain 共Speck and Pompe, 1994; Bratkovsky and Le-
the predicted Tc resulting from the spread in reported property vanyk, 2001; Roytburd et al., 2001兲 or multiphase struc-
coefficients. tures, which allow strain relaxation on average and
reduce elastic energy. The evaluation of the energies of
polydomain structures requires taking both strain and
extrapolate to qualitatively different phase diagrams at depolarization fields into account. A recent discussion of
low temperatures 共Pertsev et al., 1998, 1999兲. Quantita-
PbTiO3 using a phase-field analysis 共Li, Choudhury, et
tively, the uncertainty in predicted phase boundaries al., 2003兲 suggested that including the possibility of poly-
produced by the fitting of the Landau theory parameter domain structure formation significantly affects the
increases with increasing misfit strain as shown in Fig. 29 phase diagram for experimentally relevant misfit strains
共Choi et al., 2004兲. The Landau analysis is thus well and temperatures, as shown in Fig. 31.
complemented by first-principles calculations, which can For very thin films, additional effects associated with
provide very accurate results in the limit of zero tem- the interface between the film and substrate are also
perature. In the case of BaTiO3, the ambiguity in the expected to contribute significantly. For example, al-
low-temperature phase diagram 共Pertsev et al., 1998, though polydomain formation can accommodate misfit
1999兲 has been resolved in this way 共Dieguez et al., strain averaged over different variants, each domain will
2004兲, in the case of PbTiO3, the phenomenological re- be mismatched to the substrate at the atomic level, with
sult is confirmed 共Bungaro and Rabe, 2004兲. a corresponding increase in interface energy. Further-
The epitaxial-strain-induced changes in structure and more, the energy of a domain wall perpendicular to the
polarization are also expected to have a substantial ef- substrate will be higher than the energy of the corre-
fect on the dielectric and piezoelectric responses. While sponding wall in the bulk due to the geometrical con-
overall the dielectric and piezoelectric responses should straint on the allowed shifts of the atomic planes across
be reduced by clamping to the substrate 共Canedy et al., the domain wall 关as found for the bulk by Meyer and
2000; Li et al., 2001兲, these responses will tend to diverge Vanderbilt 共2002兲兴 imposed by the planar interface. Dif-
near second-order phase boundaries 共Pertsev et al., 2003; ferent domain walls will in general be affected differ-
Bungaro and Rabe, 2004兲. See Fig. 30. ently by the constraint, possibly changing the relative
Responses will also be large in phases, such as the r energy of different polydomain configurations.
phase, in which the direction of the polarization is not With recent advances in thin-film synthesis, it is pos-
fixed by symmetry, so that an applied field or stress can sible to grow and characterize high-quality films that are

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1119

FIG. 32. Evolution of a⬜ as a function of film thickness

for Ba0.6Sr0.4TiO3 thin films grown on
0.29共LaAlO3兲 : 0.35共Sr2TaAlO6兲 substrates. Also shown is the
theoretical curve, given by the open circles. The straight
dashed line represents the lattice parameter of the ceramic
target 共a = 0.395 05 nm兲. Reprinted with permission from
Canedy et al., 2000. © 2000, AIP.

共2002兲. The strain-induced r phase has been observed in

PZT films thinner than 150 nm on Ir-electroded Si wa-
fers 共Kelman et al., 2002兲. An antiferroelectric to ferro-
electric transition has been observed in thin films of
PbZrO3 共Ayyub et al., 1998兲, though whether it is in-
duced by strain or some other thin-film related effect is
considered an open question. Large polarization en-
hancements have been observed in epitaxially strained
BaTiO3 films 共Choi et al., 2004兲. Most dramatically,
room-temperature ferroelectricity has been achieved for
SrTiO3 under biaxial tensile strain induced by a DyScO3
substrate 共Haeni et al., 2004兲. For thicker films, observa-
FIG. 31. 共a兲 Phase diagram of PZT film under in-plane tensile tion of polydomain structures was reported by Roytburd
strain of 0.005 obtained using thermodynamic calculations as- et al. 共2001兲. It has been observed that domain formation
suming a single-domain state. There are only two stable ferro- may be suppressed by rapid cooling through the transi-
electric phases. The solid lines represent the boundaries sepa- tion 共Ramesh et al., 1993兲.
rating the stability fields of the paraelectric and ferroelectric For thicker films grown at high temperature, misfit
phases, or the ferroelectric-orthorhombic and distorted- dislocations will form at the growth temperature par-
rhombohedral phases. 共b兲 Superposition of the phase diagram tially or completely to relax misfit strain. The degree of
of a PZT film under in-plane tensile strain of 0.005 from the relaxation increases with increasing thickness, until for
phase-field approach 共scattered symbols兲 and from thermody- thick enough films the epitaxial strain is negligible. This
namic calculations assuming a single domain 共solid lines兲. behavior has been studied theoretically 共Matthews and
There are three stable ferroelectric phases: tetragonal, square; Blakeslee, 1974兲 and observed experimentally, for ex-
orthorhombic, circle; and distorted-rhombohedral, triangle ac-
ample, as shown in Fig. 32.
cording to the phase-field simulations. The scattered symbols
Additional strain can arise during cooling from the
represent the ferroelectric domain state obtained at the end of
a phase-field simulation. The shaded portion surrounded by
growth temperature if there is differential thermal ex-
the scattered symbols label the stability regions of a single pansion between the film and the substrate and the for-
ferroelectric phase, and the nonshaded region shows a mixture mation of misfit dislocations is kinetically inhibited. A
of two or three ferroelectric phases. Reprinted with permission detailed theoretical study of strain relaxation in epitaxial
from Li, Choudhury, et al., 2003. © 2003, AIP. ferroelectric films, with discussion of the interplay of
misfit dislocations, mixed domain formation, and depo-
larizing energy, was undertaken by Speck and Pompe
sufficiently thin to be coherent or partially relaxed. Here 共1994兲. It was assumed that for rapid cooling from the
we give a few examples of experimental observations of growth temperature the effect of misfit dislocations can
changes in structure and polarization in very thin films. be incorporated by using an effective substrate lattice
Tc elevation in strained PbTiO3 films has been reported parameter, while in the limit of slow cooling the system
and analyzed by Rossetti et al. 共1991兲 and Streiffer et al. optimally accommodates misfit strain with dislocations.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1120 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

materials, there are some indications that improved

ferroelectric properties and/or very large dielectric con-
stants can be achieved. The most studied system at
present is BaTiO3 / SrTiO3 共Tabata et al., 1994; Ishibashi
et al., 2000; Nakagawara et al., 2000; Shimuta et al., 2002;
Jiang et al., 2003; Neaton and Rabe, 2003; Rios et al.,
2003; Tian et al., 2005兲. In BaTiO3 / SrTiO3 superlattices
lattice matched to a SrTiO3 substrate, the compressive
in-plane strain on the BaTiO3 layer substantially raises
its polarization. Theoretical studies suggest that the
SrTiO3 layer is polarized 共and the polarization in the
BaTiO3 layer is reduced兲 by electrostatic energy consid-
erations, which favor continuity of the component of the
polarization along the normal. Overall the polarization
is enhanced above that of bulk BaTiO3, though not as
high as that of a pure coherent BaTiO3 film if it were
possible to suppress the formation of strain-relaxing de-
fects. While the natural lattice constant of
FIG. 33. Coherent temperature-dependent domain stability
map for PbTiO3 including the cubic lattice parameter for sev-
BaTiO3 / SrTiO3 is intermediate between the two end
eral common single-crystal oxide substrates. The misfit strains points, so that on a SrTiO3 substrate the superlattice is
for epitaxial growth of PbTiO3 at 600 °C are included in the under compressive in-plane stress, it has been suggested
insets along with the critical thickness h for misfit dislocation that the multilayer structure tends to inhibit the forma-
formation. Reprinted with permission from Speck and Pompe, tion of misfit dislocations so that a thicker layer of co-
1994. © 1994, AIP. herent superlattice material can be grown. As the super-
lattice material thickness increases, there will be strain
relaxation via misfit dislocations and the in-plane lattice
共This assumption is valid for films with thickness of or-
constant should increase, putting the SrTiO3 layer under
der 1 ␮m, while the treatment needs to be slightly modi-
in-plane tensile strain. In this case the SrTiO3 layer is
fied for intermediate thicknesses where the equilibrium
observed to have a component of polarization along
concentration of misfit dislocations leads to only partial
strain relaxation.兲 Elastic domains form to relax any re- 关110兴 共Jiang et al., 2003; Rios et al., 2003兲 consistent with
sidual strain. Below Tc, depolarizing energy can change theoretical studies of epitaxially strained SrTiO3 共Pert-
the relative energetics of different arrangements of po- sev et al., 2000a; Antons et al., 2005兲 and of the
larized domains and misfit dislocations. It was suggested BaTiO3 / SrTiO3 superlattice with expanded in-plane lat-
that the electrostatic energy should be more of a factor tice constant 共Johnston, Huang, et al., 2005兲.
for smaller tetragonality systems 共BaTiO3 versus The real appeal of short-periodicity ferroelectric mul-
PbTiO3兲 where the strain energy is less, though this tilayers is the potential to make “new” artificially struc-
could at least be partially balanced by the fact that the tured materials with properties that could open the door
polarization is smaller as well. A typical coherent dia- to substantial improvements in device performance or
gram is shown in Fig. 33. even radically new types of devices. Perovskites are par-
With transmission electron microscopy it is possible to ticularly promising, as individual materials possess a
make detailed studies of the types and arrangements of wide variety of structural, magnetic, and electronic prop-
misfit dislocations in perovskite thin films. Recent stud- erties, while their common structure allows matching at
ies of high-quality films include Suzuki et al. 共1999兲 and the interface to grow superlattices. Beyond the proto-
Sun et al. 共2004兲. While strain-relaxing defects, such as typical example of BaTiO3 / SrTiO3 discussed in the pre-
misfit dislocations, reduce or eliminate the elastic energy vious paragraph, there has been work on other combi-
associated with homogeneous strain, these and other de- nations such as KNbO3 / KTaO3 共Christen et al., 1996;
fects prevalent in films do generate inhomogeneous Sepliarsky et al., 2001, 2002; Sigman et al., 2002兲,
strains. As mentioned at the beginning of this section, PbTiO3 / SrTiO3 共Jiang et al., 1999; Dawber et al., 2005兲,
the inhomogeneous strains couple strongly to the polar- PbTiO3 / PbZrO3 共Bungaro and Rabe, 2002, 2004兲,
ization, and it has been shown by phenomenological La0.6Sr0.4MnO3 / La0.6Sr0.4FeO3 共Izumi et al., 1999兲,
analysis 共Balzar et al., 2002; Balzar and Popa, 2004兲 that CaMnO3 / CaRuO3 共Takahashi et al., 2001兲, LaCrO3-
their effects on Tc can be significant. They have also LaFeO3 共Ueda et al., 1998, 1999a兲, and LaFeO3-
been argued to contribute to the degradation of the di- LaMnO3 共Ueda et al., 1999b兲. In nearly all cases, strain
electric response in thin films relative to bulk values plays an important role in understanding the aggregate
共Canedy et al., 2000兲. properties of these short-period multilayers and super-
Strains and their coupling to polarization are also cen- lattices. In addition to lattice mismatch, the layers also
tral to the properties exhibited by short-period superlat- interact through the mismatch in polarization along the
tices of lattice-mismatched constituents. As the result of layer normal, which leads to mutual influences governed
recent work on artificial superlattices of ferroelectric by considerations of electrostatic energy and nonzero

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1121

macroscopic electric fields. With three or more constitu-

ents, it is possible to break inversion symmetry to obtain
superlattice materials with possibly favorable piezoelec-
tric properties. This idea was first proposed theoretically
共Sai et al., 2000兲 leading to experimental studies of
CaTiO3 / SrTiO3 / BaTiO3 共Warusawithana et al., 2003兲
and LaAlO3 / 共La, Sr兲MnO3 / SrTiO3 共Yamada et al., 2002;
Ogawa et al., 2003; Kimoto et al., 2004兲. Perovskite su-
perlattices combining ferroelectric and ferromagnetic
layers also offer a path to the development of multifer-
roic materials. The identification, synthesis, and charac-
terization of further combinations remains the subject of
active research interest.

VI. NANOSCALE FERROELECTRICS

FIG. 34. Lack of significant dependence of coercive field on
A. Quantum confinement energies lateral area in nanoscale ferroelectrics. From Alexe, 1999; Al-
exe et al., 1999.
Confinement energies are a currently popular topic in
nanoscale semiconductor microelectronics devices
共Petroff et al., 2001兲. The basic idea is that in a system in nanophase ferroelectric cells have generally been mea-
which the electron mean free path is long with respect to sured via atomic force microscopy 共Gruverman et al.,
the lateral dimension共s兲 of the device, a quantum- 1996兲. Domain structures, polarization, and coercive
mechanical increase in energy 共and of the band gap兲 in fields of nanoscale particles of BaTiO3 have been stud-
the semiconductor will occur. In general, confinement ied theoretically using interatomic potentials 共Stachiotti,
energies exist only in the ballistic regime of conduction 2004兲 and a first-principles effective Hamiltonian 共Fu
electrons, that is, where the electron mean free path ex- and Bellaiche, 2003兲.
ceeds the dimensions of the crystal. This usually requires
a high-mobility semiconductor at ultralow temperatures.
C. Self-patterned nanoscale ferroelectrics
Such effects are both interesting and important in con-
ventional semiconductors such as Si or Ge, GaAs and One approach to producing nanoscale ferroelectrics is
other III-V’s, and perhaps in II-VI’s. However, despite to attempt to produce self-patterned arrays of nanocrys-
the fact that the commonly used oxide ferroelectrics are tals, in which ordering is produced by interactions be-
wide-band-gap p-type semiconductors 共3.0⬍ Eg ⬍ 4.5 tween islands through the substrate. This approach
eV兲 共Waser and Smyth, 1996兲, neither their electron nor could be used to produce arrays of metallic nanoelec-
hole mean free paths are sufficiently long for any con- trodes on top of a ferroelectric film or alternatively ar-
finement energies to be measured. Typically the electron rays of crystals from the ferroelectric materials them-
mean free path in an ABO3 ferroelectric perovskite is selves. The first scheme was suggested by Alexe et al.
0.1 to 1.0 nm 共Dekker, 1954兲 depending on applied elec- 共1998兲 who found that a bismuth-oxide wetting layer on
tric field E, whereas the device size d is at least 20 nm. top of a bismuth-titanate film formed an array of metal-
Therefore any confinement energy 共which scales as d−2兲 lic bismuth-oxide nanocrystals on top of the film, which
might be a meV or two, virtually unmeasurable despite a were partially registered along the crystallographic di-
few published claims 共Yu et al., 1997; Kohiki et al., 2000; rections of the underlying substrate 共Fig. 35兲. These
Scott, 2000a兲 reporting extraordinarily large effects. In nanocrystals were used successfully as electrodes to
the case of Bi2O3 and SrBi2Ta2O9 共SBT兲 these effects switch regions of the film 共Alexe et al., 1999兲. In the
may arise from two-phase regions at the sample surfaces second approach one might use a material such as
共Zhou, 1992; Switzer et al., 1999兲. This is theoretically PbTiO3 on a SrTiO3 substrate, which was first demon-
interesting and very important from an engineering de- strated to form islands when grown epitaxially at very
vice point of view; if it were not true the contact poten- thin film thicknesses by Seifert et al. 共1996兲. In the con-
tial at the electrode interface in a 1T-1C device, or at the text of self-patterning of oxide materials, a recent work
ferroelectric-Si interface in a ferroelectric-gate FET, by Vasco et al. 共2003兲 studied the growth of self-
would depend critically on the cell size, which would add organized SrRuO3 crystals on LaAlO3.
a very undesirable complication to device design. When small amounts of materials are deposited on
substrates where there is some degree of mismatch be-
B. Coercive fields in nanodevices tween the two materials, islands form and the repulsive
interactions between them are mediated via strain fields
One of the most pleasant surprises in the research on in the substrate, as first suggested by Andreev 共1981兲.
small-area ferroelectrics is the observation, shown in This idea has been developed into a detailed theory by
Fig. 34, that the coercive field is independent of lateral Shchukin and Bimberg 共1999兲, however, this theory is a
area 共Alexe, 1999; Alexe et al., 1999兲. Coercive fields in zero-temperature theory, whereas a thermodynamic

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1122 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

FIG. 35. Sample from Alexe et al.: 共a兲 TEM

cross section showing underlying layers and
bismuth-oxide nanoelectrodes; 共b兲 semiregis-
tered array of nanoelectrodes taking their ori-
entation from the underlying Si substrate. Re-
printed with permission from Alexe et al.,
1998. © 1998, AIP.

theory is required to describe the crystallization pro- out by Szafraniak et al. 共2003兲. Ruediger et al. 共2004兲
cesses which occur at quite high temperatures. An ex- have recently reviewed size effects in ferroelectric nano-
tension of the theory to finite temperatures has been crystals.
carried out by Williams and co-workers 共Williams et al., Although there is potential to produce self-patterned
2000; Rudd et al., 2003兲. The chief result of this theory is arrays with greater registration by better choice of ma-
the prediction of three different kinds of structures terials and processing conditions, our general conclusion
共pyramids, domes, and superdomes兲, a volume distribu- is that highly registered memory arrays will not occur
tion for a particular species of structure, and a shape spontaneously in the absence of a prepatterned field.
map to describe relative populations of structures as a
function of coverage and crystallization temperature. D. Nonplanar geometries: ferroelectric nanotubes
One interesting result from the experiments is that
similar-shaped structures are observed in both the Almost all recent work on ferroelectric-oxide films
Volmer-Weber and Stranski-Krastanow growth modes, has involved planar geometries. However, both from a
but on different size scales. In the work of Williams the device engineering point of view and from theoretical
thickness above which dome populations occur is of the considerations, it is now appropriate to analyze carefully
order of 4–5 monolayers, corresponding to the critical nonplanar geometries, especially nanotubes.
thickness for misfit dislocations for Ge on Si共100兲. On Nanotubes made of oxide insulators have a variety of
the other hand, Capellini et al. 共1997兲 studied via atomic applications for pyroelectric detectors, piezoelectric ink-
force microscopy the growth of Ge on Si共100兲 in the jet printers, and memory capacitors that cannot be filled
Stanski-Krastanow growth mode and found a much by other nanotubes 共Herzog and Kattner, 1985; Gnade et
larger critical structure height of 50 nm at which dislo- al., 2000; Averdung et al., 2001; Sakamaki et al., 2001;
cations were introduced and the structures changed Sajeev and Busch, 2002兲. In the drive for increased stor-
from being pyramidal in geometry to domelike. The age density in FRAM and DRAM devices, complicated
large increase in critical thickness is due to a substantial
part of the misfit strain being taken up by the substrate
in the Stranski-Krastanow growth mode, as described by
Eaglesham and Cerrulo 共1990兲. The description of self-
patterned ferroelectric nanocrystals by the models of
Schukin and Williams has recently been undertaken by
Dawber, Szafraniak, et al. 共2003兲.
Prior to this two groups have grown PbTiO3 nanocrys-
tals on Pt/ Si共111兲 substrates to measure size effects in
ferroelectricity 共Roelofs et al., 2003; Shimizu et al., 2004兲.
These works both show a lack of piezoresponse in struc-
tures below 20 nm in lateral size 共Fig. 36兲, though we
expect that this is connected to mechanical constraints
rather than any fundamental limiting size for ferroelec-
tric systems. Chu et al. 共2004兲 have highlighted the role
that misfit dislocations can play in hampering ferroelec-
tricity in small structures. Interestingly in the work of
Roelofs et al. 共2003兲 and Shimizu et al. 共2004兲 because of
the 共111兲 orientation of their substrates, instead of
square-based pyramids they obtain triangular-based
structures that display hexagonal rather than cubic reg- FIG. 36. 共a兲 Topographic image of grains from 100 to 20 nm in
istration 关an analogous result is observed when Ge is lateral size. 共b兲 Piezoresponse image of same grains showing
grown on Si共111兲; Capellini et al., 1999兴. The growth and the absence of piezoresponse for grains below 20 nm. From
analysis of PZT nanocrystals on SrTiO3 has been carried Roelofs et al., 2003.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1123

stacking geometries, 3D structures, and trenches with

high aspect ratios are also being investigated to increase
the dielectric surface area. The integration of ferroelec-
tric nanotubes into Si substrates is particularly important
in the construction of 3D memory devices beyond the
present stacking and trenching designs, which according
to the international ULSI schedule3 must be achieved by
2008. Template synthesis of nanotubes and wires is a
versatile and inexpensive technique for producing nano-
structures. The size, shape, and structural properties of
the assembly are simply controlled by the template used.
Using carbon nanotubes as templates, tubular forms of a
number of oxides including V2O5, SiO2, Al2O3, and
ZrO2 have been generated 共Patzke et al., 2002兲. Much
larger 共⬎20 ␮m diameter兲 ferroelectric microtubes have
been made by sputter deposition around polyester fibers
共Fox, 1995; Pokropivny, 2001兲—Fox has made them FIG. 37. 共a兲 SEM micrograph indicating a plan view of a regu-
lar array of SBT tubes in host silicon substrate with diameter
from ZnO and PZT, with a 23 ␮m inside diameter, about
⬃2 ␮m and wall thickness ⬃200 nm. 共b兲 SBT tubes in cross-
1000 times larger than the smallest nanotubes reported
sectional view indicating coating to bottom of pore. 共c兲 Micro-
in the present paper. Porous sacrificial templates as op- graph of free-standing array of tubes with diameter ⬃800 nm;
posed to fibers have also been used. Porous anodic alu- and 共d兲 wall thickness ⬍ 100 nm.
mina has a polycrystalline structure with ordered do-
mains of diameter 1 – 3 ␮m, containing self-organized
2D hexagonal tubular pore arrays with an interpore dis- agonal or orthogonal arrays of pores with diameters 400
tance of 50–420 nm 共Li et al., 1998兲. This nanochannel nm to a few mm and up to 100 ␮m deep can be formed
material can therefore be used as a template for indi- in single-crystal Si wafers 共Ottow et al., 1996; Schilling et
vidual nanotubes but is not suitable for making an or- al., 2001兲. These crystals were originally developed for
dered array of tubes over length scales greater than a application as 2D photonic crystals, but also find appli-
few mm. Many oxide nanotubes, such as TiO2, In2O3, cations as substrates for templated growth and integra-
Ga2O3, BaTiO3, and PbTiO3, as well as nanorods of tion of oxides nanostructures with Si technology. Luo et
MnO2, Co3O4, and TiO2, have been made using porous al. 共2003兲 recently used such crystals to produce indi-
alumina membranes as templates 共Patzke et al., 2002兲. vidual, free-standing PZT and BaTiO3 ferroelectric
Hernandez et al. 共2002兲 used a sol-gel-template synthesis nanotubes by a polymeric wetting technique. Morrison
route to prepare BaTiO3 and PbTiO3 nanotube bundles et al. 共2003兲 described the use of liquid source misted
by dipping alumina membranes with 200-nm pores into chemical deposition to fill such photonic Si crystals with
the appropriate sol. The BaTiO3 and PbTiO3 nanotubes a SBT precursor. During deposition, the SBT precursor
were shown to be cubic 共paraelectric兲 and tetragonal was shown to coat the inside of the pores. After etching
共ferroelectric兲 by x-ray diffraction, although Raman of the photonic crystal with pore diameter 2 ␮m for 30
studies indicated some noncentrosymmetric phase on a sec with aqueous HF/ HNO3, the interface between the
local scale in BaTiO3. Porous silicon materials are also Si substrate and SBT coating is dissolved, exposing the
available as suitable templates. Mishina et al. 共2002兲 used uniform SBT tube, Fig. 37共a兲. The tube walls are very
a sol-gel dipping technique to fill nanoporous silicon uniform with a thickness of ⬃200 nm. The same sample
with a PbZr1−xTixO3 共PZT兲 sol producing nanograins is shown in cross-sectional view after complete removal
and nanorods 10–20 nm in diameter. The presence of the of the host Si walls between pores, Fig. 37共b兲. The result
ferroelectric PZT phase was shown by second-harmonic is a regular array of tubes attached to the host Si matrix
generation measurements. In this instance the porous only at the tube base. Although these tubes have suf-
silicon does not have a periodic array of pores 共Smith fered damage during handling, it is clear that the pores
and Collins, 1992兲 and as in the case for those produced have been filled uniformly to the bottom, a depth of
by Hernandez et al. 共2002兲 we emphasize that those ⬃100 ␮m.
nanotubes are not ordered arrays, but instead “spaghet- The second photonic crystal with pore diameter 800
tilike” tangles of nanotubes that cannot be used for the nm underwent fewer depositions and after etching re-
Si device embodiments. A second type of porous Si tem- vealed a regular array of uniform tubes of diameter 800
plate, however, consists of a very regular periodic array nm, Fig. 37共c兲. The wall thickness is uniform and
of pores with very high aspect ratios. By a combination ⬍100 nm, Fig. 37共d兲. The tubes are ⬃100 ␮m long, com-
of photolithography and electrochemical etching hex- pletely discrete, and are still attached to the host Si ma-
trix, creating a perfectly registered hexagonal array.
Free-standing tubes may be produced by completely dis-
3
International Technology Roadmap for Semiconductors solving the host Si matrix. As yet, no one has applied
共ITRS兲 2002 共available at http://public.itrs.net/Files/ cylindrical electrodes to the tubes; however, Steinhart et
2002Update/Home.pdf兲. al. 共2003兲 recently used porous anodic alumina templates

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1124 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

to grow palladium nanotubes. Using a similar method it REFERENCES

may be possible to alternately deposit Pd or Pt and SBT
to produce a concentric electrode/FE/electrode structure Ahn, C. H., J.-M. Triscone, and J. Mannhart, 2003, Nature
in each nanotube. The use of the photonic crystal tem- 共London兲 424, 1015.
plate with a regular array of pores has significant ben- Alexe, M., 1999, private communication.
efits over other porous substrates in that the coatings or Alexe, M., A. Gruverman, C. Harnagea, N. D. Zakharov, A.
tubes produced are also in a registered array ordered Pignolet, D. Hesse, and J. F. Scott, 1999, Appl. Phys. Lett. 75,
over several mm or even cm. This facilitates addressing 1158.
of such an array for device applications. DRAMs utilize Alexe, M., J. F. Scott, C. Curran, N. D. Zakharov, D. Hesse,
and A. Pignolet, 1998, Appl. Phys. Lett. 73, 1592.
high surface-area dielectrics, and high aspect-ratio SBT
Alpay, S. P., and A. L. Roytburd, 1998, J. Appl. Phys. 83, 4714.
coatings such as these embedded in Si could increase
Andreev, A. F., 1981, JETP Lett. 53, 1063.
storage density. Current state-of-the-art deep-trenched
Antons, A., J. B. Neaton, K. M. Rabe, and D. Vanderbilt, 2005,
capacitors are 0.1 mm diameter by 6 mm deep, aspect
Phys. Rev. B 71, 024102.
ratio 60:1. Using SBT 共or other FE-oxide兲 nanotubes of Arlt, G., and H. Neumann, 1988, Ferroelectrics 87, 109.
wall thickness ⬍100 nm and a trench 共or array of Armstrong, R. D., and B. R. Horrocks, 1997, Solid State Ionics
trenches兲 of 0.1 ␮m diameter and 100 ␮m deep, an as- 94, 181.
pect ratio of ⬎1000:1 is possible. Applying and address- Astala, R. K., and P. D. Bristowe, 2004, Modell. Simul. Mater.
ing electrodes to an array of FE nanotubes could gener- Sci. Eng. 12, 79.
ate 3D FRAM structures offering high storage density Averdung, J., et al., 2001, German Patent No. DE10023456 共2
with improved read/write characteristics compared to January兲.
conventional planar stacks. On removal of the Si walls, Ayyub, P., S. Chattopadhyay, R. Pinto, and M. S. Multani,
the piezoelectric response 共expansion or contraction un- 1998, Phys. Rev. B 57, R5559.
der an applied field兲 of such an array of nanotubes could Balzar, D., and N. C. Popa, 2004, Thin Solid Films 450, 29.
be utilized for a number of microelectromechanical sys- Balzar, D., P. A. Ramakrishnan, P. Spagnol, et al., 2002, Jpn. J.
tems applications. These could include 共1兲 ink-jet Appl. Phys., Part 1 41, 6628.
printing—delivery of subpicoliter droplets for Baniecki, J. D., R. B. Laibowitz, T. M. Shaw, C. Parks, J. Lian,
lithography-free printing of submicron circuits; 共2兲 bio- H. Xu, and Q. Y. Ma, 2001, J. Appl. Phys. 89, 2873.
medical applications—nanosyringes, inert drug delivery Bardeen, J., 1947, Phys. Rev. 71, 717.
implants; and 共3兲 micropositioners or movement sensors. Baroni, S., A. dal Corso, S. de Gironcoli, and P. Giannozzi,
Almost no theoretical work has been published on the unpublished, http://www.pwscf.org
physics of ferroelectric nanotubes. Analytical solutions Baroni, S., S. de Gironcoli, A. D. Corso, and P. Giannozzi,
2001, Rev. Mod. Phys. 73, 515.
for the effects on the dij piezoelectric coefficients of hol-
Basceri, C., S. K. Streiffer, A. I. Kingon, and R. Waser, 1997, J.
low tubes have been given both for the case in which
Appl. Phys. 82, 2497.
polarization P is along the length z 共Ebenezer and Batra, I. P., and B. D. Silverman, 1973, Solid State Commun.
Ramesh, 2003兲 and for P radial 共Ebenezer and Abra- 11, 291.
ham, 2002兲, they did not, however, solve the azimuthal Bellaiche, L., A. Garcia, and D. Vanderbilt, 2000, Phys. Rev.
case where polarization goes around the tube. It it this Lett. 84, 5427.
latter case that has been measured as hysteresis by Luo Bengtsson, L., 1999, Phys. Rev. B 59, 12301.
et al. 共2003兲 with a tube lying on a bottom electrode with Betsuyaku, K., H. Tanaka, H. Katayama Yoshida, and T.
a semicircular sputtered top electrode. Important mat- Kawai, 2001, Jpn. J. Appl. Phys., Part 1 40, 6911.
ters such as the dependence of Tc upon tube diameter Black, C. T., and J. J. Welser, 1999, IEEE Trans. Electron
have also not been examined. Devices 46, 776.
Blaha, P., K. Schwarz, P. Sorantin, and S. B. Trickey, 1990,
Comput. Phys. Commun. 59, 399.
Boikov, Yu. A., B. M. Goltsman, V. K. Yamarkin, and V. V.
VII. CONCLUSIONS Lemanov, 2001, Appl. Phys. Lett. 78, 3866.
Bratkovsky, A. M., and A. P. Levanyuk, 2000, Phys. Rev. Lett.
In this review we have sought to cover the important
84, 3177.
advances in recent years in the physics of thin-film ferro-
Bratkovsky, A. M., and A. P. Levanyuk, 2001, Phys. Rev. Lett.
electric oxides. At present, ferroelectric thin-film
86, 3642.
memory devices have reached a point of maturity where Bratkovsky, A. M., and A. P. Levanyuk, 2005, Phys. Rev. Lett.
they are beginning to appear in real commercial devices. 94, 107601.
At the same time, new directions such as the drive to Brazier, M., S. Mansour, and M. McElfresh 1999, Appl. Phys.
faster, smaller, nanoscale devices and nonplanar geom- Lett. 74, 4032.
etries are evolving and new levels of physical under- Brennan, C., 1993, Ferroelectrics 150, 199.
standing will be required. Over the next years it is ex- Brennan, C. J., 1992, Integr. Ferroelectr. 2, 73.
pected that first-principles computational approaches Bune, A. V., V. M. Fridkin, S. Ducharme, L. M. Blinov, S. P.
will continue to develop, suggesting a new synergy be- Palto, A. V. Sorokin, S. G. Yudin, and A. Zlatkin, 1998, Na-
tween the computational modeling and experimental re- ture 共London兲 391, 1998.
alizations of ferroelectric systems with new and exciting Bungaro, C., and K. M. Rabe, 2002, Phys. Rev. B 65, 224106.
properties. Bungaro, C., and K. M. Rabe, 2004, Phys. Rev. B 69, 184101.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1125

Bungaro, C., and K. M. Rabe, 2005, Phys. Rev. B 71, 035420. Dawber, M., J. F. Scott, and A. J. Hartmann, 2001, J. Eur.
Canedy, C. L., H. Li, S. P. Alpay, L. Salamanca-Riba, A. L. Ceram. Soc. 21, 1633.
Roytburd, and R. Ramesh, 2000, Appl. Phys. Lett. 77, 1695. Dawber, M., I. Szafraniak, M. Alexe, and J. F. Scott, 2003, J.
Capellini, G., L. Di Gaspare, and F. Evangelisti, 1997, Appl. Phys.: Condens. Matter 15, L667.
Phys. Lett. 70, 493. de Araujo, C. A.-Paz, J. D. Cuchiaro, L. D. McMillan, M. C.
Capellini, G., N. Motta, A. Sgarlata, and R. Calarco, 1999, Scott, and J. F. Scott, 1995, Nature 共London兲 374, 627.
Solid State Commun. 112, 145. de Gironcoli, S., S. Baroni, and R. Resta, 1989, Phys. Rev. Lett.
Castell, M. R., 2002, Surf. Sci. 505, 1. 62, 2853.
Chalamala, B. R., Y. Wei, R. H. Reuss, S. Aggarwal, B. E. Dekker, A. J., 1954, Phys. Rev. 94, 1179.
Gnade, R. Ramesh, J. M. Bernhard, E. D. Sosa, and D. E. Democritov, S. O., and N. M. Kreines, 1988, JETP Lett. 48,
Golden 1999, Appl. Phys. Lett. 74, 1394. 294.
Chan, N.-H., R. K. Sharma, and D. M. Smyth, 1976, J. Elec- Dieguez, O., S. Tinte, A. Antons, C. Bungaro, J. B. Neaton, K.
trochem. Soc. 123, 1584. M. Rabe, and D. Vanderbilt, 2004, Phys. Rev. B 69, 212101.
Chan, N.-H., R. K. Sharma, and D. M. Smyth, 1981, J. Am. Dietz, G. W., M. Schumacher, R. Waser, S. K. Streiffer, C.
Ceram. Soc. 64, 556. Basceri, and A. I. Kingon, 1997, J. Appl. Phys. 82, 2359.
Chen, H.-M., S.-W. Tsaur, and Y.-M. Lee J., 1998, Jpn. J. Appl. Dietz, G. W., and R. Waser, 1995, Integr. Ferroelectr. 9, 317.
Phys., Part 1 137, 4056. Dietz, G. W., and R. Waser, 1997, Thin Solid Films 299, 53.
Chen, L., V. Nagarajan, R. Ramesh, and A. L. Roytburd, 2003, Dimos, D., W. L. Warren, M. B. Sinclair, B. A. Tuttle, and R.
J. Appl. Phys. 94, 5147. W. Schwartz, 1994, J. Appl. Phys. 76 4305.
Cheon, C. I., K. Y. Yun, J. S. Kim, and J. H. Kim, 2001, Integr. Do, D-H., P. G. Evans, E. D. Issacs, D. M. Kim, C. B. Eom,
Ferroelectr. 34, 1513. and E. M. Dufresne, 2004, Nat. Mater. 3, 365.
Choi, K. J., M. Biegalski, Y. L. Li, A. Sharan, J. Schubert, R. Dovesi, R., R. Orlando, B. Civalleri, C. Roetti, V. R. Saunders,
Uecker, P. Reiche, Y. B. Chen, X. Q. Pan, V. Gopalan, L.-Q. and C. M. Zichovich-Wilson, 2005, Z. Kristallogr. 220, 571.
Chen, D. G. Schlom, and C. B. Eom, 2004, Science 306, 1005. Du, X., and I. W. Chen, 1998a, J. Appl. Phys. 83, 7789.
Choi, K.-J., W.-C. Shin, and S.-G. Yoon, 2001, Integr. Ferro- Du, X., and I. W. Chen, 1998b, in Ferroelectric Thin Films VI,
electr. 34, 1541. edited by R. E. Treece et al., MRS Symposia Proceedings No.
Chon, U., H. M. Jang, M. G. Kim, and C. H. Chang, 2002, 493 共Materials Research Society, Pittsburgh兲, p. 311.
Phys. Rev. Lett. 89, 087601. Ducharme, S., V. M. Fridkin, A. V. Bune, S. P. Palto, L. M.
Christen, H.-M., L. A. Boatner, J. D. Budai, M. F. Chisholm, L. Blinov, N. N. Petukhova, and S. G. Yudin, 2000, Phys. Rev.
A. Ga, P. J. Marrero, and D. P. Norton, 1996, Appl. Phys. Lett. 84, 175.
Lett. 68, 1488. Duiker, H. M., 1990, Ph.D. thesis 共University of Colorado兲.
Chu, M.-W., I. Szafraniak, R. Scholz, C. Harnagea, D. Hesse, Duiker, H. M., and P. D. Beale, 1990, Phys. Rev. B 41, 490.
M. Alexe, and U. Gosele, 2004, Nat. Mater. 3, 87. Duiker, H. M., P. D. Beale, J. F. Scott, C. A. Paz de Araujo, B.
Cockayne, E., and K. M. Rabe, 1998, in First-Principles Calcu- M. Melnick, and J. D. Cuchario, 1990, J. Appl. Phys. 68, 5783.
lations for Ferroelectrics, edited by R. E. Cohen, AIP Conf. Eaglesham, D. J., and M. Cerullo, 1990, Phys. Rev. Lett. 64,
Proc. No. 436 共AIP, Woodbury, NY兲, p. 61. 1943.
Cockayne, E., and K. M. Rabe, 2000, J. Phys.: Condens. Matter Ebenezer, D. D., and P. Abraham, 2002, Curr. Sci. 83, 981.
61, 305. Ebenezer, D. D., and R. Ramesh, 2003, Curr. Sci. 85, 1173.
Cohen, R. E., 1992, Nature 共London兲 358, 136. Erdman, N., K. R. Poeppelmeier, M. Asta, O. Warschkow, D.
Cohen, R. E., 1996, J. Phys. Chem. Solids 57, 1393. E. Ellis, and L. D. Marks, 2002, Nature 共London兲 419, 55.
Cohen, R. E., 1997, Ferroelectrics 194, 323. Evans, J., 1990.
Cohen, R. E., and H. Krakauer, 1990, Phys. Rev. B 42, 6416. Fatuzzo, E., and W. J. Merz, 1967, Ferroelectricity 共North-
Cohen, R. E., and H. Krakauer, 1992, Ferroelectrics 136, 65. Holland, Amsterdam兲.
Colla, E. L., S. Hong, D. V. Taylor, A. K. Tagantsev, and N. Fennie, C. J., and K. M. Rabe, 2005, Phys. Rev. B 71, 100102.
Setter, 1998, Appl. Phys. Lett. 72, 2763. Feynman, R. P., 1939, Phys. Rev. 56, 340.
Conley, J. W., and G. D. Mahan, 1967, Phys. Rev. 161, 681. Fiebig, M., T. Lottermoser, D. Frohlich, A. V. Goltsev, and R.
Cora, F., and C.R. A. Catlow, 1999, Faraday Discuss. 114, 421. V. Pisarev, 2002, Nature 共London兲 419, 818.
Cowley, A. M., and S. M. Sze, 1965, J. Appl. Phys. 36, 3212. Floquet, N., and C. Valot, 1999, Ferroelectrics 234, 107.
Damjanovic, D., 1997, Phys. Rev. B 55, R649. Floquet, N., C. M. Valot, M. T. Mesnier, J. C. Niepce, L. Nor-
Dawber, M., P. Chandra, P. B. Littlewood, and J. F. Scott, 2003, mand, A. Thorel, and R. Kilaas, 1997, J. Phys. III 7, 1105.
J. Phys.: Condens. Matter 15, L393. Foeth, M., A. Sfera, P. Stadelmann, and P-A. Buffat, 1999, J.
Dawber, M., I. Farnan, and J. F. Scott, 2003, Am. J. Phys. 71, Electron Microsc. 48, 717.
819. Fong, D. D., G. B. Stephenson, S. K. Streiffer, J. A. Eastman,
Dawber, M., C. Lichtensteiger, M. Cantoni, M. Veithen, P. O. Auciello, P. H. Fuoss, and C. Thompson, 2004, Science
Ghosez, K. Johnston, K. M. Rabe, and J.-M. Triscone, 2005, 304, 1650.
Phys. Rev. Lett. 共to be published兲. Forlani, F., and N. Minnaja, 1964, Phys. Status Solidi 4, 311.
Dawber, M., and J. F. Scott, 2000, Appl. Phys. Lett. 76, 1060; Fouskova, A., 1965, J. Phys. Soc. Jpn. 20, 1625.
2000, Appl. Phys. Lett. 76, 3655. Fouskova, A., and V. Janousek, 1965, J. Phys. Soc. Jpn. 20,
Dawber, M., and J. F. Scott, 2001, Integr. Ferroelectr. 38, 161. 1619.
Dawber, M., and J. F. Scott, 2002, Jpn. J. Appl. Phys., Part 1 41, Fox, D. L., D. R. Tilley, J. F. Scott, and H. J. Guggenheim,
6848. 1980, Phys. Rev. B 21, 2926.
Dawber, M., and J. F. Scott, 2004, J. Phys.: Condens. Matter 16, Fox, G. R., 1995, J. Mater. Sci. Lett. 14, 1496.
L515. Fu, H., and L. Bellaiche, 2003, Phys. Rev. Lett. 91, 257601.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1126 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

Fu, H. X., and R. E. Cohen, 2000, Nature 共London兲 403, 281. Herzog, K., and E. Kattner, 1985, U.S. Patent No. 4504845 共12
Fu, L., E. Yaschenko, L. Resca, and R. Resta, 1999, Phys. Rev. March兲.
B 60, 2697. Hill, N. A., 2000, J. Phys. Chem. B 104, 6694.
Ganpule, C. S., V. Nagarajan, H. Li, A. S. Ogale, D. E. Stein- Huffmann, M., 1995, Integr. Ferroelectr. 10, 39.
hauser, S. Aggarwal, E. Williams, R. Ramesh, and P. De Wolf, Hwang, C. S., 1998, Mater. Sci. Eng., B 56, 178.
2000, Appl. Phys. Lett. 77, 292. Hwang, C. S., B. T. Lee, H.-J. Cho, K. H. Lee, C. S. Kang, H.
Ganpule, C. S., V. Nagarajan, S. B. Ogale, A. L. Roytburd, E. Horii, S. I. Lee, and M. Y. Lee, 1997, Appl. Phys. Lett. 71,
D. Williams, and R. Ramesh, 2000, Appl. Phys. Lett. 77, 3275. 371.
Ganpule, C. S., A. L. Roytburd, V. Nagarajan, B. K. Hill, S. B. Hwang, C. S., B. T. Lee, C. S. Kang, J. W. Kim, K. H. Lee, H.-J.
Ogale, E. D. Williams, R. Ramesh, and J. F. Scott, 2001, Phys. Cho, H. Horii, W. D. Kim, S. I. Lee, Y. B. Roh, and M. Y.
Rev. B 65, 014101. Lee, 1998, J. Appl. Phys. 83, 3703.
Garcia, A., and D. Vanderbilt, 1996, Phys. Rev. B 54, 3817. Ishibashi, Y., N. Ohashi, and T. Tsurumi, 2000, Jpn. J. Appl.
Garcia, A., and D. Vanderbilt, 1998, Appl. Phys. Lett. 72, 2981. Phys., Part 1 39, 186.
Garg, A., Z. H. Barber, M. Dawber, J. F. Scott, A. Snedden, Ishiwara, H., 1993, Jpn. J. Appl. Phys., Part 1 32, 442.
and P. Lightfoot, 2003, Appl. Phys. Lett. 83, 2414. Ishiwara, H., 2001, Integr. Ferroelectr. 34, 1451.
Gerson, R., and T. C. Marshall, 1959, J. Appl. Phys. 30, 1650. Ishiwara, H., T. Shimamura, and E. Tokumitsu, 1997, Jpn. J.
Ghosez, P., E. Cockayne, U. V. Waghmare, and K. M. Rabe, Appl. Phys., Part 1 36, 1655.
1999, Phys. Rev. B 60, 836. Izumi, M., Y. Murakami, Y. Konishi, T. Manako, M. Kawasaki,
Ghosez, P., and K. M. Rabe, 2000, Appl. Phys. Lett. 76, 2767. and Y. Tokura, 1999, Phys. Rev. B 60, 1211.
Gnade, B., et. al, 2000, U.S. Patent No. US6033919 共7 March兲. Jaffe, B., W. R. Cook, and H. Jaffe, 1971, Piezoelectric Ceram-
Gonze, X., J.-M. Beuken, R. Caracas, F. Detraux, M. Fuchs, ics 共Academic, London兲.
G.-M. Rignanese, L. Sindic, M. Verstraete, G. Zerah, F. Jol- Janovec, V., 1958, Czech. J. Phys., Sect. A 8, 3.
let, M. Torrent, A. Roy, M. Mikami, Ph. Ghosez, J.-Y. Raty, Jiang, A. Q., J. F. Scott, H. Lu, and Z. Chen, 2003, J. Appl.
and D. C. Allan, 2002, Comput. Mater. Sci. 25, 478. http:// Phys. 93, 1180.
www.mapr.ucl.ac.be/ABINIT Jiang, J. C., X. Q. Pan, W. Tian, C. D. Theis, and D. G. Schlom,
Gorbenko, O. Y., S. V. Samoilenkov, I. E. Graboy, and A. R. 1999, Appl. Phys. Lett. 74, 2851.
Kaul, 2002, Chem. Mater. 14, 4026. Johnston, K., M. R. Castell, A. T. Paxton, and M. W. Finnis,
2004, Phys. Rev. B 70, 085415.
Gruverman, A., 1990, Ph.D. thesis 共Ural State University, Eka-
Johnston, K., X. Huang, J. B. Neaton, and K. M. Rabe, 2005,
terinburg, Russia兲.
Phys. Rev. B 71, 100103.
Gruverman, A., O. Auciello, and H. Tokumoto, 1996, Appl.
Johnston, K., and K. M. Rabe, 2005, unpublished.
Phys. Lett. 69, 3191.
Jun, S., and J. Lee, 2001, Integr. Ferroelectr. 34, 1579.
Gruverman, A., H. Tokumoto, A. S. Prakash, S. Aggarwal, B.
Jung, D. J., M. Dawber, J. F. Scott, L. J. Sinnamon, and J. M.
Yang, M. Wuttig, R. Ramesh, O. Auciello, and T. Venkatesan,
Gregg, 2002, Integr. Ferroelectr. 48, 59.
1997, Appl. Phys. Lett. 71, 3492.
Jung, D. J., H. H. Kim, Y. J. Song, N. W. Jang, B. J. Koo, S. Y.
Haeni, J. H., et al., 2004, Nature 共London兲 430, 758.
Lee, S. D. Park, Y. W. Park, and K. N. Kim, 2000, IEDM 2000
Han, J.-P., X. Guo, C. C. Broadbridge, T. P. Ma, M. Cantoni,
Technical Digest, 801.
J.-M. Sallese, and P. Fazan, 2001, Integr. Ferroelectr. 34, 1505.
Jung, D. J., F. D. Morrison, M. Dawber, H. H. Kim, K. Kim,
Haneder, T. P., W. Hoenlein, H. Bachhofer, H. Von Philips-
and J. F. Scott, 2004, J. Appl. Phys. 95, 4968.
born, and R. Waser, 2001, Integr. Ferroelectr. 34, 1487. Junquera, J., and P. Ghosez, 2003, Nature 共London兲 422, 506.
Hartmann, A. J., M. Neilson, R. N. Lamb, K. Watanabe, and J. Junquera, J., P. Ghosez, and K. M. Rabe, 2005, unpublished.
F. Scott, 2000, Appl. Phys. A 70, 239. Junquera, J., and K. M. Rabe, 2005, unpublished.
Hashimoto, S., H. Orihara, and Y. Ishibashi, 1994, J. Phys. Soc. Junquera, J., M. Zimmer, P. Ordejon, and P. Ghosez, 2003,
Jpn. 63, 1601. Phys. Rev. B 67, 155327.
He, L. X., and D. Vanderbilt, 2003, Phys. Rev. B 68, 134103. Kalkur, T. S., B. Jacob, and G. Argos, 1994, Integr. Ferroelectr.
Heifets, E., R. I. Eglitis, E. A. Kotonin, and G. Borstel, 2001a, 5, 177.
in Fundamental Physics of Ferroelectrics 2001, edited by H. Kalkur, T. S., and J. Lindsey, 2001, Integr. Ferroelectr. 34,
Krakauer, AIP Conf. Proc. No. 582 共AIP, Melville, NY兲, p. 1587.
201. Kato, K., 2001, Integr. Ferroelectr. 34, 1533.
Heifets, E., R. I. Eglitis, E. A. Kotonin, and G. Borstel, 2002b, Kay, H. F., and J. W. Dunn, 1962, Philos. Mag. 7, 1962.
in Fundamental Physics of Ferroelectrics 2002, edited by R. E. Kelman, M. B., L. F. Schloss, P. C. McIntyre, B. C. Hendrix, S.
Cohen, AIP Conf. Proc. No. 626 共AIP, Melville, NY兲, p. 285. M. Bilodeau, and J. F. Roeder, 2002, Appl. Phys. Lett. 80,
Heifets, E., R. I. Eglitis, E. A. Kotonin, J. Maier, and G. Bor- 1258.
stel, 2001b, Phys. Rev. B 64, 235417. Khodenkov, G. E., 1975, Fiz. Met. Metalloved. 39, 466.
Heifets, E., R. I. Eglitis, E. A. Kotonin, J. Maier, and G. Bor- Kim, J., C. J. Kim, and I. Chung, 2001, Integr. Ferroelectr. 33,
stel, 2002a, Surf. Sci. 513, 211. 133.
Heifets, E., E. Kotonin, and P. W. M. Jacobs, 2000, Thin Solid Kim, M., G. Duscher, N. D. Browning, K. Sohlberg, S. T. Pan-
Films 375, 64. telides, and S. J. Pennycock, 2001, Phys. Rev. Lett. 86, 4056.
Heifets, E., E. A. Kotonin, and J. Maier, 2000, Surf. Sci. 462, Kimoto, K., Y. Matsui, H. Yamada, M. Kawasaki, X. Yu, Y.
19. Kaneko, and Y. Tokura, 2004, Appl. Phys. Lett. 84, 5374.
Heine, V., 1965, Phys. Rev. 138, A1689. King-Smith, R. D., and D. Vanderbilt, 1993, Phys. Rev. B 47,
Hernandez, B. A., K.-S. Chang, E. R. Fisher, and P. K. Dor- 1651.
hout, 2002, Chem. Mater. 14, 481. King-Smith, R. D., and D. Vanderbilt, 1994, Phys. Rev. B 49,

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1127

5828. Merz, W. J., 1954, Phys. Rev. 95, 690.

Kingon, A. I., J. P. Maria, and S. K. Streiffer, 2000, Nature Meyer, B., J. Padilla, and D. Vanderbilt, 1999, Faraday
共London兲 406, 1032. Discuss. 114, 395.
Klenov, D. O., T. R. Taylor, and S. Stemmer, 2004, J. Mater. Meyer, B., and D. Vanderbilt, 2001, Phys. Rev. B 63, 205426.
Res. 19, 1477. Meyer, B. E., and D. Vanderbilt, 2002, Phys. Rev. B 65, 104111.
Kohiki, S., S. Takada, A. Shmizu, K. Yamada, and M. Mitone, Miller, S. L., R. D. Nasby, J. R. Schwank, M. S. Rodgers, and
2000, J. Appl. Phys. 87, 474. P. V. Dressendorfer, 1990, J. Appl. Phys. 68, 6463.
Kohn, W., and L. J. Sham, 1965, Phys. Rev. 140, A1133. Mills, G., M. S. Gordon, and H. Metiu, 2003, J. Chem. Phys.
Koukhar, V. J., N. A. Pertsev, and R. Waser, 2001, Phys. Rev. 118, 4198.
B 64, 214103. Mihara, T., H. Watanabe, H. Yoshimori, C. A. Paz de Araujo,
Krcmar, M., and C. L. Fu, 2003, Phys. Rev. B 68, 115404. B. Melnick, and L. D. McMillan, 1992, Integr. Ferroelectr. 1,
Kresse, G., and J. Furthmuller, 1996, Phys. Rev. B 54, 11169. 269.
Kresse, G., and J. Hafner, 1993, Phys. Rev. B 47, 558. Mishina, E. D., K. A. Vorotilov, V. A. Vasil’ev, A. S. Sigov, N.
Krishnan, A., M. M. J. Treacy, M. E. Bisher, P. Chandra, and P. Ohta, and S. Nakabayashi, 2002, J. Exp. Theor. Phys. 95, 502.
B. Littlewood, 2002, Integr. Ferroelectr. 43, 31. Moll, J. L., and Y. Tarui, 1963, IEEE Trans. Electron Devices
Ku, H. Y., and F. G. Ullman, 1964, J. Appl. Phys. 35, 265. ED-10, 338.
Kubo, T., and H. Nozoye, 2003, Surf. Sci. 542, 177. Monch, W., 1986, Phys. Rev. Lett. 58, 1260.
Kurtin, S., T. C. McGill, and C. A. Mead, 1969, Phys. Rev. Lett. Montanari, B., and N. M. Harrison, 2004, J. Phys.: Condens.
22, 1433. Matter 16, 273.
Lewis, S. P., and A. M. Rappe, 1996, Phys. Rev. Lett. 77, 5241. Moreira, R. L., 2002, Phys. Rev. Lett. 88, 179701.
Li, A. P., F. Muller, A. Birner, K. Nielsch, and U. Gosele, 1998, Moriera de Santos, A., S. Parashar, A. R. Raju, Y. S. Zhao, A.
J. Appl. Phys. 84, 6023. K. Cheetham, and C. N. R. Rao, 2002, Solid State Commun.
Li, H., A. L. Roytburd, S. P. Alpay, T. D. Tran, L. Salamanca- 122, 49.
Riba, and R. Ramesh, 2001, Appl. Phys. Lett. 78, 2354. Morrison, F. D., J. F. Scott, M. Alexe, T. J. Leedham, T. Tat-
Li, H., H. Zheng, L. Salamanca-Riba, R. Ramesh, I. Naumov, suta, and O. Tsuji, 2003, Microelectron. Eng. 66, 591.
and K. Rabe, 2002, Appl. Phys. Lett. 81, 4398. Mott, N. F., 1938, Proc. Cambridge Philos. Soc. 34, 568.
Li, J. A., E. A. Akhadov, J. Baker, L. A. Boatner, D. Bonart, Munkholm, A., S. K. Streiffer, M. V. R. Murty, J. A. Eastman,
and S. A. Safron, 2003, Phys. Rev. B 68, 045402. C. Thompson, O. Auciello, L. Thompson, J. F. Moore, and G.
Li, T., and S. T. Hsu, 2001, Integr. Ferroelectr. 34, 1495. B. Stephenson, 2002, Phys. Rev. Lett. 88, 016101.
Li, Y. L., S. Choudhury, Z. K. Liu, and L. Q. Chen, 2003, Appl. Muralt, P., 2000, J. Micromech. Microeng. 10, 136.
Phys. Lett. 83, 1608. Murphy, E. L., and R. H. Good, 1956, Phys. Rev. 102, 1464.
Lichtensteiger, C., J. M. Triscone, J. Junquera, and P. Ghosez, Nagarajan, V., A. Roytburd, A. Stanishevsky, S. Praserth-
2005, Phys. Rev. Lett. 94, 047603. choung, T. Zhao, L. Chen, J. Melngailis, O. Auciello, and R.
Lines, M. E., and A. M. Glass, 1967, Principles and Applica- Ramesh, 2003, Nat. Mater. 2, 43.
tions of Ferroelectrics and Related Materials 共Clarendon, Ox- Nakagawara, O., T. Shimuta, T. Makino, T. Makino, and T.
ford兲. Makino, 2000, Appl. Phys. Lett. 77, 3257.
Lookman, A., R. M. Bowman, J. M. Gregg, J. Kut, S. Rios, M. Neaton, J. B., and K. M. Rabe, 2003, Appl. Phys. Lett. 82,
Dawber, A. Ruediger, and J. F. Scott, 2004, J. Appl. Phys. 96, 1586.
555. Nielsen, O. H., and R. M. Martin, 1985, Phys. Rev. B 32, 3780.
Lubomirsky, I., J. Fleig, and J. Maier, 2002, J. Appl. Phys. 92, Nowotny, J., and M. Rekas, 1994, Ceram. Int. 20, 251.
6819. Nuffer, J., D. C. Lupascu, J. Rodel, and M. Schroeder, 2001,
Luo, Y., I. Szafraniak, N. D. Zakharov, V. Nagarajan, M. Stein- Appl. Phys. Lett. 79, 3675.
hart, R. B. Wehrspohn, J. H. Wendorff, R. Ramesh, and M. Ogawa, Y., H. Yamada, T. Ogasawara, H. Yamada, and H.
Alexe, 2003, Appl. Phys. Lett. 83, 440. Yamada, 2003, Phys. Rev. Lett. 90, 217403.
Lupascu, D. C., and U. Rabe, 2002, Phys. Rev. Lett. 89, 187601. Okuyama, M., H. Sugiyama, T. Nakaiso, and M. Noda, 2001,
Macleod, T. C., and F. D. Ho, 2001, Integr. Ferroelectr. 34, Integr. Ferroelectr. 34, 1477.
1461. Orihara, H., S. Hashimoto, and Y. Ishibashi, 1994, J. Phys. Soc.
Man, Z. Y., and X. Q. Feng, 2002, Solid State Commun. 123, Jpn. 63, 1031.
333. Ottow, S., V. Lehmann, and H. Fo, 1996, Appl. Phys. A 63,
Masui, S., et al., 2003, Proceedings of the IEEE Custom Inte- 153.
grated Circuits Conference 共IEEE, New York兲, p. 403. Outzourhit, A., A. Naziripour, J. U. Trefny, T. Kito, B. Yarar,
Mathews, S., R. Ramesh, T. Venkatesan, and J. Benedetto, R. Yandrofski, J. D. Cuchiaro, and A. M. Hermanna, 1995,
1997, Science 276, 238. Integr. Ferroelectr. 8, 227.
Matthews, J. W., and A. E. Blakeslee, 1974, J. Cryst. Growth Padilla, J., and D. Vanderbilt, 1997, Phys. Rev. B 56, 1625.
27, 118. Padilla, J., and D. Vanderbilt, 1998, Surf. Sci. 418, 64.
McKee, R. A., F. J. Walker, and M. F. Chisholm, 1998, Phys. Pan, M. J., S. F. Park, C. W. Park, K. A. Markowski, S.
Rev. Lett. 81, 3014. Yoshikawa, and C. A. Randall, 1996, J. Am. Ceram. Soc. 79,
McKee, R. A., F. J. Walker, and M. F. Chisholm, 2001, Science 2971.
293, 468. Park, C. H., 2003, J. Korean Phys. Soc. 42, 1420.
McKee, R. A., F. J. Walker, M. B. Nardelli, W. A. Shelton, and Park, C. H., and D. J. Chadi, 1998, Phys. Rev. B 57, R13961.
G. M. Stocks, 2003, Science 300, 1726. Park, J. D., and T. S. Oh, 2001, Integr. Ferroelectr. 34, 1561.
McMillan, L. D., C. A. Paz de Araujo, T. Roberts, J. Cuchiaro, Park, S. E., and T. R. Shrout, 1997, J. Appl. Phys. 82, 1804.
M. C. Scott, and J. F. Scott, 1992, Integr. Ferroelectr. 2, 351. Paruch, P., T. Tybell, and J.-M. Triscone, 2001, Appl. Phys.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1128 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

Lett. 79, 530. Ross, I. M., 1957, U.S. Patent No. 2,791,760.
Patzke, G. R., F. Krumeich, and R. Nesper, 2002, Angew. Rossetti, G. A., L. E. Cross, and K. Kushida, 1991, Appl. Phys.
Chem., Int. Ed. 41, 2446. Lett. 59, 2524.
Peacock, P. W., and J. Robertson, 2002, J. Appl. Phys. 92, 4712. Roytburd, A. L., S. P. Alpay, L. A. Bendersky, V. Nagarajan,
Pertsev, N. A., V. G. Kukhar, H. Kohlstedt, and R. Waser, and R. Ramesh, 2001, J. Appl. Phys. 89, 553.
2003, Phys. Rev. B 67, 054107. Rudd, R. E., G. A. D. Briggs, A. P. Sutton, G. Medeiros-
Pertsev, N. A., J. Rodriguez Contreras, V. G. Kukhar, B. Her- Ribeiro, and R. Stanley Williams, 2003, Phys. Rev. Lett. 90,
manns, H. Kohlstedt, and R. Waser, 2003, Appl. Phys. Lett. 146101.
83, 3356. Ruediger, A., T. Schneller, A. Roelofs, S. Tiedke, T. Schmitz,
Pertsev, N. A., A. K. Tagantsev, and N. Setter, 2000a, Phys. and R. Waser, 2004, Appl. Phys. A 共to be published兲.
Rev. B 61, R825. Ruini, A., R. Resta, and S. Baroni, 1998, Phys. Rev. B 57, 5742.
Pertsev, N. A., A. K. Tagantsev, and N. Setter, 2000b, Phys. Saghi-Szabo, G., R. E. Cohen, and H. Krakauer, 1998, Phys.
Rev. Lett. 84, 3722. Rev. Lett. 80, 4321.
Pertsev, N. A., A. G. Zembilgotov, and A. K. Tagantsev, 1998, Sai, N., A. M. Kolpak, and A. M. Rappe, 2005 Phys. Rev. B 72,
Phys. Rev. Lett. 80, 1988. 020101.
Pertsev, N. A., A. G. Zembilgotov, and A. K. Tagantsev, 1999, Sai, N., B. Meyer, and D. Vanderbilt, 2000, Phys. Rev. Lett. 84,
Ferroelectrics 223, 79. 5636.
Petroff, P. M., A. Lorke, and A. Imamoglu, 2001, Phys. Today Sai, N., and D. Vanderbilt, 2000, Phys. Rev. B 62, 13942.
54 共5兲, 46. Sajeev, J., 2002, U.S. Patent No. US2002074537 共20 June兲.
Piskunov, S., E. Heifets, R. I. Eglitis, and G. Borstel, 2004, Sakamaki, S., et al., 2001, U.S. Patent No. 20010412.
Comput. Mater. Sci. 29, 165. Salje, E. K. H., and W. T. Lee, 2004, Nat. Mater. 3, 425.
Plumlee, R., 1967, Sandia Laboratories Report No. SC-RR-67- Samara, G. A., 1987, Ferroelectrics 73, 145.
730. Samara, G. A., 2003, J. Phys.: Condens. Matter 15, R367.
Pokropivny, V. V., 2001, Physica C 351, 71. Sarin Kumar, A. K., P. Paruch, J.-M. Triscone, W. Daniau, S.
Poykko, S., and D. J. Chadi, 2000, J. Phys. Chem. Solids 61, Ballandras, L. Pellegrino, D. Marr, and T. Tybell, 2004, Appl.
291. Phys. Lett. 85, 1757.
Rabe, K. M., 2005, unpublished. Sawyer, C. B., and C. H. Tower, 1930, Phys. Rev. 35, 269.
Rabe, K. M., and U. V. Waghmare, 1995, Phys. Rev. B 52, Sayer, M., A. Mansingh, A. K. Arora, and A. Lo, 1992, Integr.
13236. Ferroelectr. 1, 129.
Rabe, K. M., and U. V. Waghmare, 2002, Phys. Rev. B 65, Schilling, J., F. Muller, S. Matthias, R. B. Wehrspohn, U. Gos-
214111. ele, and K. Busch, 2001, Appl. Phys. Lett. 78, 1180.
Rabson, T. A., T. A. Rost, and H. Lin, 1995, Integr. Ferroelectr. Schloss, L. F., H. Kim, and P. C. McIntyre, 2004, J. Mater. Res.
6, 15. 19, 1265.
Ramesh, R., T. Sands, and V. G. Keramidas, 1993, Appl. Phys. Schloss, L. F., P. C. McIntyre, B. C. Hendrix, S. M. Bilodeau, J.
Lett. 63, 731. F. Roeder, and S. R. Gilbert, 2002, Appl. Phys. Lett. 81, 3218.
Randoshkin, V. V., 1995, Fiz. Tverd. Tela 共Leningrad兲 37, 3056. Schluter, M., 1978, Phys. Rev. B 17, 5044.
Rao, F., M. Kim, A. J. Freeman, S. Tang, and M. Anthony, Schnupp, P., 1967, Phys. Status Solidi 21, 567.
1997, Phys. Rev. B 55, 13953. Schottky, W., 1938, Naturwiss. 26, 843.
Reiner, J. W., F. J. Walker, R. A. McKee, C. A. Billman, J. Scott, J. F., 1977, Phys. Rev. B 16, 2329.
Junquera, K. M. Rabe, and C. H. Ahn, 2004, Phys. Status Scott, J. F., 1979, Rep. Prog. Phys. 42, 1055.
Solidi B 241, 2287. Scott, J. F., 1995, in Science and Technology of Electroceramic
Resta, R., 1994, Rev. Mod. Phys. 66, 899. Thin Films, edited by O. Auciello and R. Waser, NATO ASI
Resta, R., 2003, Modell. Simul. Mater. Sci. Eng. 11, R69. Series No. 284 共Kluwer, Dordrecht兲.
Rhoderick, E. H., and R. H. Williams, 1988, Metal- Scott, J. F., 1996, Integr. Ferroelectr. 12, 71.
Semiconductor Contacts, 2nd ed. 共Oxford, Clarendon兲. Scott, J. F., 1998, Ferroelectrics 1, 82.
Rios, S., A. Ruediger, J. Q. Jiang, J. F. Scott, H. Lu, and Z. Scott, J. F., 1999, Ferroelectrics 232, 905.
Chen, 2003, J. Phys.: Condens. Matter 15, 305. Scott, J. F., 2000a, Ferroelectric Memories 共Springer-Verlag,
Robblee, L. S., and S. F. Cogan, 1986, in Encyclopedia of Ma- Berlin兲
terials Science and Engineering Supplementary Vol. I, edited Scott, J. F., 2000b, J. Appl. Phys. 88, 6092.
by R. W. Cahn 共Pergamon, Oxford兲, p. 276. Scott, J. F., 2005, Microelectron. Eng. 共to be published兲.
Robels, U., and G. Arlt, 1993, J. Appl. Phys. 73, 3454. Scott, J. F., C. A. Araujo, B. M. Melnick, L. D. McMillan, and
Robertson, J., 2003, J. Appl. Phys. 93, 1054. R. Zuleeg, 1991, J. Appl. Phys. 70, 382.
Robertson, J., and C. W. Chen, 1999, Appl. Phys. Lett. 74, Scott, J. F., M. Azuma, E. Fujii, T. Otsuki, G. Kano, M. C.
1168. Scott, C. A. Paz de Araujo, L. D. McMillan, and T. Roberts,
Robertson, J., C. W. Chen, W. L. Warren, and C. D. Gutleben, 1992, Proceedings of ISAF 1992 共IEEE, New York兲, p. 356.
1996, Appl. Phys. Lett. 69, 1704. Scott, J. F., and M. Dawber, 2000, Appl. Phys. Lett. 76, 3801.
Rodriguez Contreras, J., H. Kohlstedt, U. Poppe, and R. Scott, J. F., A. Q. Jiang, S. A. T. Redfern, M. Zhang, and M.
Waser, 2003, Appl. Phys. Lett. 83, 4595. Dawber, 2003, J. Appl. Phys. 94, 3333.
Roelofs, A., N. A. Pertsev, R. Waser, F. Schlaphof, L. M. Eng, Scott, J. F., L. Kammerdiner, M. Parris, S. Traynor, V. Otten-
C. Ganpule, V. Nagarajan, and R. Ramesh, 2002, Appl. Phys. bacher, A. Shawabkeh, and W. F. Oliver, 1988, J. Appl. Phys.
Lett. 80, 1424. 64, 787.
Roelofs, A., T. Schneller, K. Szot, and R. Waser, 2003, Scott, J. F., B. M. Melnick, L. D. McMillan, and C. A. Paz de
Nanotechnology 14, 250. Araujo, 1993, Integr. Ferroelectr. 3, 225.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides 1129

Scott, J. F., and B. Pouligny, 1988, J. Appl. Phys. 64, 1547. Stolichnov, I., and A. K. Tagantsev, 1998, J. Appl. Phys. 84,
Seifert, A., A. Vojta, J. S. Speck, and F. F. Lange, 1996, J. 3216.
Mater. Res. 11, 1470. Stolichnov, I., A. K. Tagantsev, E. L. Colla, and N. Setter,
Sepliarsky, M., S. R. Phillpot, M. G. Stachiotti, and R. L. 1998, Appl. Phys. Lett. 73, 1361.
Migoni, 2002, J. Appl. Phys. 91, 3165. Stolichnov, I., A. K. Tagantsev, E. L. Colla, and N. Setter,
Sepliarsky, M., S. R. Phillpot, D. Wolf, M. G. Stachiotti, and R. 1999, Ferroelectrics 225, 125.
L. Migoni, 2001, J. Appl. Phys. 90, 4509. Stolichnov, I., A. Tagantsev, N. Setter, S. Okhonin, P. Fazan, J.
Sepliarsky, M., M. G. Stachiotti, and R. L. Migoni, 2005, S. Cross, M. Tsukada, A. Bartic, and D. Wouters, 2001, Integr.
e-print cond-mat/0503524. Ferroelectr. 32, 737.
Sepliarsky, M., Z. Wu, A. Asthagiri, and R. E. Cohen, 2004, Streiffer, S. K., J. A. Eastman, D. D. Fong, C. Thompson, A.
Ferroelectrics 301, 55. Munkholm, M. V. R. Murty, O. Auciello, G. R. Bai, and G. B.
Seshadri R., and N. A. Hill, 2001, Chem. Mater. 13, 2892. Stephenson, 2002, Phys. Rev. Lett. 89, 067601.
Shchukin, V. A., and D. Bimberg, 1999, Rev. Mod. Phys. 71, Sugiyama, H., K. Kodama, T. Nakaiso, M. Noda, and M.
1125. Okuyama, 2001, Integr. Ferroelectr. 34, 1521.
Shilo, D., G. Ravichandran, and K. Bhattacharya, 2004, Nat. Sun, H. P., W. Tian, X. Q. Pan, J. H. Haeni, and D. G. Schlom,
Mater. 3, 453. 2004, Appl. Phys. Lett. 84, 3298.
Shimada, Y., K. Arita, Y. Kato, K. Uchiyama, V. Joshi, and M. Suzuki, T., and M. Fujimoto, 2001, J. Appl. Phys. 89, 5622.
Lim, 2001, Integr. Ferroelectr. 34, 1467. Suzuki, T., Y. Nishi, and M. Fujimoto, 1999, Philos. Mag. A 79,
Shimakawa, Y., Y. Kubo, Y. Nakagawa, T. Kamiyama, H. 2461.
Asano, and F. Izumi, 1999, Appl. Phys. Lett. 74, 1904. Switzer, J. A., M. G. Shumsky, and E. W. Bohannan, 1999,
Shimizu, M., H. Fujisawa, H. Nonomura, and H. Niu, 2004, Science 284, 293.
Integr. Ferroelectr. 62, 109. Szafraniak, I., C. Harnagea, R. Scholz, S. Bhattacharyya, D.
Shimuta, T., O. Nakagawara, T. Makino, S. Arai, H. Tabata, Hesse, and M. Alexe, 2003, Appl. Phys. Lett. 83, 2211.
and T. Kawai, 2002, Jpn. J. Appl. Phys., Part 1 91, 2290. Sze, S. M., D. J. Coleman, and A. Loya, 1971, Solid-State
Shin, C.-H., S. Y. Cha, H. C. Lee, W.-J. Lee, B.-G. Yu, and Electron. 14, 1209.
D.-H. Kwak, 2001, Integr. Ferroelectr. 34, 1553. Tabata, H., H. Tanaka, and T. Kawai, 1994, Appl. Phys. Lett.
Shur, V. Ya., 1996, in Ferroelectric Thin Films: Synthesis and 65, 1970.
Basic Properties, edited by C. Paz de Araujo, G. W. Taylor, Tagantsev, A. K., I. Stolichnov, E. L. Colla, and N. Setter,
and J. F. Scott 共Gordon and Breach, Amsterdam兲, pp. 153– 2001, J. Appl. Phys. 90, 1387.
192. Tagantsev, A. K., I. Stolichnov, N. Setter, J. S. Cross, and M.
Shur, V. Ya., A. Gruverman, V. P. Kuminov, and N. A. Tsukada, 2002, Phys. Rev. B 66, 214109.
Tonkachyova, 1990, Ferroelectrics 111, 197. Takahashi, K. S., M. Kawasaki, and Y. Tokura, 2001, Appl.
Shur, V. Ya, A. L. Gruverman, N. Yu. Ponomarev, E. L. Phys. Lett. 79, 1324.
Rumyantsev, and N. A. Tonkacheva, 1991, JETP Lett. 53, Taylor, D. V., and D. Damjanovic, 1997, J. Appl. Phys. 82,
615. 1973.
Shur, V. Ya., et al., 2000, Taylor, D. V., and D. Damjanovic, 1998, Appl. Phys. Lett. 73,
Sigman, J., D. P. Norton, H. M. Christen, P. H. Fleming, and L. 2045.
A. Boatner, 2002, Phys. Rev. Lett. 88, 079601. Tian, W., J. C. Jiang, X. Q. Pan, J. H. Haeni, D. G. Schlom, J.
Simmons, J. G., 1965, Phys. Rev. Lett. 15, 967. B. Neaton, and K. M. Rabe, 2005, unpublished.
Singh, D. J., 1995, Phys. Rev. B 52, 12559. Tinte, S., and M. G. Stachiotti, 2000, in Fundamental Physics of
Sinharoy, S., H. Buhay, M. H. Francombe, and D. R. Lampe, Ferroelectrics 2000, edited by B. N. Hale and M. Kulmala,
1993, Integr. Ferroelectr. 3, 217. AIP Conf. Proc. No. 535 共AIP, Melville, NY兲, p. 273.
Sinharoy, S., H. Buhay, M. H. Francombe, W. J. Takei, N. J. Tinte, S., and M. G. Stachiotti, 2001, Phys. Rev. B 64, 235403.
Doyle, J. H. Rieger, D. R. Lampe, and E. Stepke, 1991, J. Tinte, S., M. G. Stachiotti, M. Sepliarsky, R. L. Migoni, and C.
Vac. Sci. Technol. A 9, 409. O. Rodriguez, 1999, J. Phys.: Condens. Matter 11, 9679.
Sinharoy, S., D. R. Lampe, H. Buhay, and M. H. Francombe, Troeger, G., 1991, unpublished.
1992, Integr. Ferroelectr. 2, 377. Tsai, F., V. Khiznishenko, and J. M. Cowley, 1992, Ultramicros-
Smith, R. L., and S. D. Collins, 1992, J. Appl. Phys. 71, R1. copy 4, 5.
Smyth, D. M., 1984, Prog. Solid State Chem. 15, 145. Tsai, M. H., Y. H. Tang, and S. K. Dey, 2003, J. Phys.: Condens.
Soler, J. M., E. Artacho, J. D. Gale, A. Garcia, J. Junquera, P. Matter 15, 7901.
Ordejon, and D. Sanchez-Portal, 2002, J. Phys.: Condens. Tsurumi, T., S.-M. Num, Y.-B. Kil, and S. Wada, 2001, Ferro-
Matter 14, 2745. electrics 259, 43.
Souza, I., J. Iniguez, and D. Vanderbilt, 2002, Phys. Rev. Lett. Tybell, T., P. Paruch, T. Giamarchi, and J.-M. Triscone, 2002,
89, 117602. Phys. Rev. Lett. 89, 097601.
Speck, J. S., and W. Pompe, 1994, J. Appl. Phys. 76, 466. Ueda, K., H. Tabata, and T. Kawai, 1998, Science 280, 1064.
Stachiotti, M. G., 2004, Appl. Phys. Lett. 84, 251. Ueda, K., H. Tabata, and T. Kawai, 1999a, Jpn. J. Appl. Phys.,
Stachiotti, M. G., C. O. Rodriguez, C. Ambrosch-Draxl, and N. Part 1 38, 6690.
E. Christensen, 2000, Phys. Rev. B 61, 14434. Ueda, K., H. Tabata, and T. Kawai, 1999b, Phys. Rev. B 60,
Steinhart, M., Z. Jia, A. K. Schaper, R. B. Wehrspohn, U. R12561.
Gsele, and J. H. Wendorff, 2003, Adv. Mater. 共Weinheim, Ullmann, M., H. Goebel, H. Hoenigschmid, and T. Haneder,
Ger.兲 15, 706. 2001, Integr. Ferroelectr. 34, 1595.
Stemmer, S., S. K. Streiffer, F. Ernst, and M. Ruhle, 1995, Phi- Van Aken, B. B., T. T. M. Palstra, A. Filippetti, and N. A.
los. Mag. A 71, 713. Spaldin, 2004, Nat. Mater. 3, 164.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005

1130 Dawber, Rabe, and Scott: Physics of thin-film ferroelectric oxides

Vasco, E., R. Dittmann, S. Karthauser, and R. Waser, 2003, Williams, R. Stanley, G. Medeiros-Ribeiro, T. I. Kamins, and
Appl. Phys. Lett. 82, 2497. D. A. A. Ohlberg, 2000, Annu. Rev. Phys. Chem. 51, 527.
Vendik, O. G., and S. P. Zubko, 2000, J. Appl. Phys. 88, 5343. Woodward, D. I., I. M. Reaney, G. Y. Yang, E. C. Dickey, and
Von Hippel, A., 1935, Ergeb. Exakten Naturwiss. 14, 118. C. A. Randall, 2004, Appl. Phys. Lett. 84, 4650.
Waghmare, U. V., and K. M. Rabe, 1997a, Phys. Rev. B 55, Wu, S.-Y., 1974, IEEE Trans. Electron Devices 21, 8.
6161. Wu, Z., R. E. Cohen, and D. J. Singh, 2004, Phys. Rev. B 70,
Waghmare, U. V., and K. M. Rabe, 1997b, Ferroelectrics 194, 104112.
135. Wu, Z., and H. Krakauer 2003, Phys. Rev. B 68, 014112.
Walker, L. R., 1963, in Magnetism III, edited by G. T. Rado Xiaohua, L., J. Yin, Z. G. Liu, L. Wang, J. Li, X. H. Zhu, and
and H. Suhl 共Academic, New York兲, p. 450. K. J. Chen, 2001, Integr. Ferroelectr. 34, 1571.
Wang, J., et al., 2003, Science 299, 1719. Yamada, H., M. Kawasaki, Y. Ogawa, Y. Ogawa, and Y.
Warusawithana, M. P., E. V. Colla, J. N. Eckstein, and M. B. Ogawa, 2002, Appl. Phys. Lett. 81, 4793.
Weissman, 2003, Phys. Rev. Lett. 90, 036802. Yoo, I. K., and S. B. Desu, 1992, Phys. Status Solidi A 133, 565.
Waser, R., T. Baiatu, and K.-H. Hartdl, 1990, J. Am. Ceram. Yoon, S.-M., and H. Ishiwara, 2001, IEEE Trans. Electron
Soc. 73, 1654. Devices 48, 2002.
Waser, R., and D. M. Smyth, 1996, in Ferroelectric Thin Films: Yu, B., C. Zhu, and F. Gan, 1997, J. Appl. Phys. 82, 4532.
Synthesis and Basic Properties, edited by C. Paz de Araujo, Zafar, S., B. Hradsky, D. Gentile, P. Chu, R. E. Jones, and S.
G. W. Taylor, and J. F. Scott 共Gordon and Breach, Amster- Gillespie, 1999, J. Appl. Phys. 86, 3890.
dam兲, pp. 47–92. Zafar, S., R. E. Jones, B. Jiang, B. White, V. Kaushik, and S.
Watanabe, K., A. J. Hartmann, R. N. Lamb, and J. F. Scott, Gillespie, 1998, Appl. Phys. Lett. 73, 3533.
1998, J. Appl. Phys. 84, 2170. Zhong, W., D. Vanderbilt, and K. M. Rabe, 1994, Phys. Rev.
Wieder, H. H., 1956, J. Appl. Phys. 27, 413. Lett. 73, 1861.
Wieder, H. H., 1957, J. Appl. Phys. 28, 367. Zhou, W., 1992, J. Solid State Chem. 101, 1.

Rev. Mod. Phys., Vol. 77, No. 4, October 2005