1: %%%md.tex%%%
2: % Time-stamp: <2008-06-12 09:46:38 takeshi>
3: %%
4:
5: %\documentclass[aps,preprint,prb,floatfix,showpacs]{revtex4}
6: \documentclass[aps,twocolumn,prb,floatfix,citeautoscript,showpacs]{revtex4}
7: %\documentclass[aps,galley,prb,floatfix,citeautoscript,showpacs]{revtex4}
8: \usepackage[dvips]{graphicx}
9: \usepackage{bm}
10: \usepackage[normalem]{ulem}
11: \usepackage{amsmath,amssymb}
12:
13: \newcommand{\ab}{{\alpha\beta}}
14: \newcommand{\dab}{{\delta_\ab}}
15: \newcommand{\dij}{{\delta_{ij}}}
16: \newcommand{\Rij}{{\bm{R}_{ij}}}
17: \newcommand{\Rijn}{{\bm{R}_{ij}+\bm{n}}}
18: \newcommand{\Gunit}{{\frac{2\pi}{a_0}}}
19: \newcommand{\eInf}{{\epsilon_\infty}}
20: \newcommand{\La}{{L_\alpha}}
21:
22: % % These 5 lines should be comment before submit
23: % \marginparwidth 2.7in
24: % \marginparsep 0.5in
25: % \def\dvm#1{\marginpar{\small DV: #1}}
26: % \def\tnm#1{\marginpar{\small TN: #1}}
27: % \def\uvw#1{\marginpar{\small UVW: #1}}
28:
29: \begin{document}
30: \title{Fast Molecular-Dynamics Simulation for Ferroelectric Thin-Film Capacitors
31: Using a First-Principles Effective Hamiltonian}
32:
33: \author{Takeshi Nishimatsu$^{1,2}$}
34: \author{Umesh V. Waghmare$^{3}$}
35: \author{Yoshiyuki Kawazoe$^{2}$}
36: \author{David Vanderbilt$^{1}$}
37:
38: \affiliation{
39: $^{1}$Department of Physics and Astronomy, Rutgers University,
40: 136 Frelinghuysen Road, Piscataway, NJ 08544-8019\\
41: $^{2}$Institute for Materials Research (IMR), Tohoku University, Sendai 980-8577, Japan\\
42: $^{3}$Theoretical Sciences Unit, Jawaharlal Nehru Centre for Advanced Scientific Research (JNCASR),
43: Jakkur, Bangalore, 560 064, India
44: }
45:
46: \begin{abstract}
47: A newly developed fast molecular-dynamics method is applied to BaTiO$_3$
48: ferroelectric thin-film capacitors with short-circuited electrodes
49: or under applied voltage.
50: The molecular-dynamics simulations based on a first-principles effective
51: Hamiltonian clarify that dead layers (or passive layers) between ferroelectrics
52: and electrodes markedly affect the properties of capacitors,
53: and predict that the system is unable to hop between a
54: uniformly polarized ferroelectric structure and
55: a striped ferroelectric domain structure at low temperatures.
56: Simulations of hysteresis loops of thin-film capacitors are also performed,
57: and their dependence on film thickness, epitaxial constraints,
58: and electrodes are discussed.
59: \end{abstract}
60:
61: \pacs{77.80.Dj, 77.80.Fm, 64.70.Nd}
62: % 77.80.Dj Domain structure; hysteresis
63: % 77.80.Fm Switching phenomena
64: % 64.70.Nd Structural transitions in nanoscale materials
65:
66: %\date{\today}
67:
68: \maketitle
69:
70: \section{Introduction}
71: Ferroelectric thin films are beginning to see wide-ranging applications,
72: for example in multilayer capacitors,
73: nonvolatile FeRAMs~\cite{Scott:Ferroelectric:Memories:2000},
74: and nanoactuators.
75: There is strong pressure to reduce the sizes of such
76: thin-film structures.
77: In recent years,
78: the preparation of oxide thin films by low-temperature
79: non-equilibrium techniques such as
80: molecular beam epitaxy and pulsed-laser deposition
81: have attracted a great deal of attention, as they enable finely
82: controlled growth of epitaxial thin
83: films~\cite{Choi:B:L:S:S:U:R:C:P:G:C:S:E:Science:306:p1005-1009:2004}.
84:
85: It is well known that the properties of ferroelectric capacitors are
86: highly influenced by the properties of the interface between
87: the ferroelectrics and the electrodes. For example,
88: the fatigue of ferroelectric capacitors is associated with the appearance of
89: dead layers (or passive layers) near the electrodes,~\cite{Drougard:L:JAP:30:p1663-1668:1959,
90: Miller:N:S:R:D:JAP:68:p6463-6471:1990,
91: Lemanov:Yarmarkin:PhysSolidState:38:1996,
92: Jin:Z:JAP:92:p4594-4598:2002}
93: and imperfect electrodes cannot fully screen the polarization
94: of ferroelectrics~\cite{MEHTA:S:J:JAP:44:p3379-3385:1973,Dawber:C:L:S:JPhys-CondesMatter:15:pL393-L398:2003},
95: leading to a finite depolarization field in the ferroelectric film.
96: However, the nanosize effects and temperature dependences of
97: ferroelectric capacitor hysteresis, polarization switching, and dynamics
98: of domain wall motion remain poorly known.
99: Experimentally, {\em in situ} observations are difficult.
100: Theoretically, the long-range Coulomb interaction limits the size
101: and time of molecular-dynamics (MD) simulations, and
102: it has been unclear how to include surface effects and
103: depolarization fields caused by interface structures.
104:
105: In 1994, King-Smith and Vanderbilt studied the total-energy surface
106: for zone-center distortions of perovskite-type ferroelectric oxides
107: $AB$O$_3$ ($A$ is a monovalent or divalent cation and
108: $B$ is a penta- or tetravalent metal)
109: at zero temperature
110: using first-principles calculations with ultrasoft-pseudopotentials and
111: a plane-wave basis set.~\cite{King-Smith:V:1994}
112: Starting from the full symmetric cubic perovskite structure,
113: they define the displacements $v_{\alpha}^{\tau}$
114: of atoms $\tau$ (=$A$, $B$, O$_{\rm I}$, O$_{\rm II}$, O$_{\rm III}$)
115: in the Cartesian directions $\alpha (=x,y,z)$
116: along the
117: %
118: $\Gamma_{15}$ soft-mode normalized
119: direction vectors $\bm{\xi}_\alpha$ as
120: \begin{equation}
121: \label{eq:Eigenvector}
122: \bm{v}_\alpha=
123: \left(
124: \begin{array}{c}
125: v_\alpha^A \\
126: v_\alpha^B \\
127: v_\alpha^{\rm O_{\rm I}} \\
128: v_\alpha^{\rm O_{\rm II}} \\
129: v_\alpha^{\rm O_{\rm III}}
130: \end{array}
131: \right)
132: = u_\alpha\bm{\xi}_\alpha
133: = u_\alpha
134: \left(
135: \begin{array}{c}
136: \xi_\alpha^A \\
137: \xi_\alpha^B \\
138: \xi_\alpha^{\rm O_{\rm I}} \\
139: \xi_\alpha^{\rm O_{\rm II}} \\
140: \xi_\alpha^{\rm O_{\rm III}}
141: \end{array}
142: \right)\ ,
143: \end{equation}
144: with the scalar soft-mode amplitude $u_\alpha$.
145: Under the condition that the strain components
146: $\eta_i$~($i=1,\ \dots,\ 6$; Voigt notation; $\eta_1=e_{11}$, $\eta_4=e_{23}$)
147: minimize the total energy for each $\bm{u}=(u_x, u_y, u_z)$,
148: they expressed the total energy as
149: \begin{equation}
150: \label{eq:King-Smith-and-Vanderbilt}
151: E^{\rm tot}
152: = E^{0}
153: + \kappa u^2
154: + \alpha ' u^4
155: + \gamma ' (u_x^2 u_y^2 +
156: u_y^2 u_z^2 +
157: u_z^2 u_x^2)\ ,
158: \end{equation}
159: where $u^2 = u_x^2 + u_y^2 + u_z^2$,
160: $E^{0}$ is the total energy for the cubic structure,
161: $\kappa$ is half the eigenvalue of the soft mode,
162: and $\alpha '$ and $\gamma '$ are the constants determined from
163: coupling constants between
164: atomic displacements and strains.
165: Their expression properly describes the coupling of
166: polar atomic-displacement and strain degrees of freedom.
167:
168: In 1994-1997, Zhong, Vanderbilt,
169: and Rabe~\cite{Zhong:V:R:1994,Zhong:V:R:PRB:v52:p6301:1995}
170: and Waghmare and Rabe~\cite{Waghmare:R:1997PRB}
171: expanded Eq.~(\ref{eq:King-Smith-and-Vanderbilt})
172: from a mean-field framework to a local-mode framework,
173: replacing ${\bm u}$ by $\{\bm{u}\}$,
174: where the braces $\{\}$ denote a {\em set} of $\bm{u}$ in a
175: simulation supercell, as
176: \begin{multline}
177: \label{eq:MC:Hamiltonian}
178: E^{\rm tot} =
179: V^{\rm self}(\{\bm{u}\}) + V^{\rm dpl}(\{\bm{u}\})+V^{\rm short}(\{\bm{u}\})\\
180: + V^{\rm elas}(\eta_1,\cdots\!,\eta_6) + V^{\rm int}(\{\bm{u}\}, \eta_1,\cdots\!,\eta_6)~.
181: \end{multline}
182: Here
183: %$\eta_l$ is the six-component local strain tensor
184: %in Voigt notation ($\eta_1=e_{xx}$, $\eta_4=2e_{yx}$),
185: $V^{\rm self}$, $V^{\rm dpl}$, $V^{\rm short}$, $V^{\rm elas}$, and $V^{\rm int}$ are
186: a local-mode self-energy,
187: a long-range dipole-dipole interaction,
188: a short-range interaction between soft-modes,
189: an elastic energy,
190: and an interaction between the local modes and local strain, respectively.
191: They employed Eq.~(\ref{eq:MC:Hamiltonian})
192: as an effective Hamiltonian for $\{\bm{u}\}$ in
193: the supercell,
194: performed Monte-Carlo simulations,
195: and
196: demonstrated the ability to describe the phase transitions of bulk ferroelectrics.
197: The coarse-graining that reduces the
198: 15-dimensional atomic displacement vector $v_{\alpha}^{\tau}$
199: to a 3-dimensional local soft-mode amplitude vector $u_\alpha$
200: in each unit cell was shown to be a good approximation.
201: However, the computation of $V^{\rm dpl}$ was still time-consuming,
202: owing to the long-range Coulomb interaction, thus
203: limiting system size and simulation time that could be handled in
204: practical simulations.
205:
206: In 2003, Waghmare, Cockayne, and Burton introduced a technique
207: to decrease the computational time
208: for $V^{\rm dpl}$ (or forces exerted on $\{\bm{u}\}$).~\cite{Waghmare:C:B:2003}
209: Direct calculation of the forces in real space
210: requires a computational time proportional to $N^2$, i.e., $O(N^2)$,
211: where $N$ is the supercell size ($N=L_x\times L_y \times L_z$).
212: It decreases to $O(N\log N)$ if one calculates the forces in reciprocal space
213: using fast-Fourier transform (FFT) methods.
214: This acceleration in computational speed enabled us to perform
215: MD simulations on $\{\bm{u}\}$ in a large supercell, and was applied
216: to bulk relaxor ferroelectrics~\cite{Waghmare:C:B:2003,Burton:C:W:PRB:72:p064113:2005}.
217:
218: Here, we explain how the fast MD method for simulating a first-principles
219: effective Hamiltonian can be applied to study ferroelectric thin-film
220: capacitor structures with short-circuited electrodes or external
221: electric fields.
222: This new MD method can simulate perovskite-type ferroelectric thin-film capacitors
223: with dead layers and
224: consequent depolarization fields.
225: The high speed of this MD method enables us to simulate
226: a ferroelectric material
227: for a realistic system size (up to 100~nm)
228: and a realistic time span ($>$ 1~ns).
229:
230: In the next section,
231: we explain the formalism of the new MD-simulation method for thin-film capacitors.
232: Results of simulations of BaTiO$_3$ bulk and thin-film capacitors are shown in Sec.~\ref{sec:results}.
233: In subsection~\ref{subsec:Bulk},
234: we confirm the reliability of our MD program by simulating thermal properties of bulk BaTiO$_3$.
235: The advantage of this MD method compared to the Monte-Carlo method is also discussed.
236: In subsection~\ref{subsec:Capacitors},
237: we perform heating-up and cooling-down simulations for thin-film BaTiO$_3$ capacitors
238: with perfect and imperfect electrodes.
239: Thickness dependence of simulated striped domain structures in thin-film capacitors
240: with imperfect electrodes are analyzed in detail.
241: We have already reported some simulated results
242: of thin-film capacitors of this subsection
243: and determined thermal properties
244: in Ref.~[\onlinecite{Paul:N:K:W:PRL:99:p077601:2007}] briefly.
245: In subsection~\ref{subsec:HysteresisLoops},
246: newly obtained simulated results of hysteresis loops of thin-film capacitors are reported.
247: In Sec.~\ref{sec:summary}, we summarize the paper.
248:
249: We named our MD program {\tt feram} and
250: distribute it as free software through \url{http://loto.sourceforge.net/feram/}.
251:
252: \section{Formalism and method of calculation}
253: \label{sec:Formalism}
254: \subsection{Effective Hamiltonian}
255: \label{sub:Effective:Hamiltonian}
256: The effective Hamiltonian used in the present MD simulations
257: is basically the same as that in Ref.~[\onlinecite{Waghmare:C:B:2003}].
258: Here, we present the Hamiltonian
259: with a notation similar to that in Ref.~[\onlinecite{Zhong:V:R:PRB:v52:p6301:1995}] as
260: \begin{multline}
261: \label{eq:Effective:Hamiltonian}
262: H^{\rm eff}%(\{\bm{u}\},\{\bm{w}\}, \eta_1,\cdots\!,\eta_6)
263: = \frac{M^*_{\rm dipole}}{2} \sum_{\bm{R},\alpha}\dot{u}_\alpha^2(\bm{R})
264: + \frac{M^*_{\rm acoustic}}{2}\sum_{\bm{R},\alpha}\dot{w}_\alpha^2(\bm{R})\\
265: + V^{\rm self}(\{\bm{u}\})+V^{\rm dpl}(\{\bm{u}\})+V^{\rm short}(\{\bm{u}\})\\
266: + V^{\rm elas,\,homo}(\eta_1,\cdots\!,\eta_6)+V^{\rm elas,\,inho}(\{\bm{w}\})\\
267: + V^{\rm coup,\,homo}(\{\bm{u}\}, \eta_1,\cdots\!,\eta_6)+V^{\rm coup,\,inho}(\{\bm{u}\}, \{\bm{w}\})\\
268: -Z^*\sum_{\bm{R}}\bm{\mathcal{E}}\!\cdot\!\bm{u}(\bm{R})~,
269: \end{multline}
270: where
271: $\bm{u}=\bm{u}(\bm{R})$ and $\bm{w}=\bm{w}(\bm{R})$
272: are, respectively,
273: the local soft-mode amplitude vector and the local acoustic displacement vector
274: of the unit cell at $\bm{R}$,
275: the $\alpha$ component of $\bm{R}$ runs over
276: \begin{equation}
277: \label{eq:R}
278: R_\alpha = 0,\ a_0,\ 2 a_0,\ \cdots\ (\La-1)a_0~,
279: \end{equation}
280: $\eta_1,\cdots\!,\eta_6$ are the homogeneous strain components, and
281: $M^*_{\rm dipole}$ and $M^*_{\rm acoustic}$ are the effective masses for $\bm{u}$ and $\bm{w}$, respectively.
282: Note that $\bm{u}$ can also be considered as the optical displacement,
283: in contrast to
284: the acoustic displacement $\bm{w}$, or the
285: dipole moment $Z^*\bm{u}$, where $Z^*$ is the Born effective charge
286: associated with the soft mode.
287: In the effective Hamiltonian (\ref{eq:Effective:Hamiltonian}),
288: external electric field $\bm{\mathcal{E}}$ is taken into account through
289: its vector product with each dipole moment $Z^*\bm{u}$.
290: % This is classical and non-quantum Hamiltonian.
291:
292: To determine the effective mass $M^*_{\rm dipole}$,
293: let $\epsilon_\alpha^\tau(\bm{k},i)$ be a mass-weighted $i$-th eigenvector
294: of the {\em phonon} dynamical matrix at wavevector $\bm{k}$.
295: Its eigenvalue $\{\omega(\bm{k},i)\}^2$ is the corresponding phonon frequency.
296: Moreover,
297: let $d_\alpha^\tau(\bm{k},i)=\epsilon_\alpha^\tau(\bm{k},i)/\sqrt{M_\tau}$
298: be an atomic displacement vector, which is
299: normalized as $\sum_{\alpha, \tau} \{d_\alpha^\tau(\bm{k},i)\}^2 = 1$
300: by adjusting the norm of $\bm{\epsilon}(\bm{k},i)$.
301: Here, $M^\tau$ is the mass of atom $\tau$.
302: Generally,
303: the effective mass of a phonon is $\bm{k}$- and mode-dependent:
304: \begin{equation}
305: \label{eq:EffectiveMassGeneral}
306: M^*\!(\bm{k},i) = \sum_{\alpha, \tau} \{d_\alpha^\tau(\bm{k},i)\}^2 M^\tau~.
307: \end{equation}
308: However, as an approximation,
309: we have to employ a unique effective mass for dipoles in the MD simulation.
310: Thus using the {\em steepest descent} $\Gamma_{15}$ soft-mode normalized
311: direction vectors
312: $\bm{\xi}_z = (0.20,\ 0.76, -0.21, -0.21, -0.53)$ and
313: $\bm{\xi}_x = \bm{\xi}_y = 0$
314: from Ref.~[\onlinecite{Zhong:V:R:PRB:v52:p6301:1995}], for BaTiO$_3$,
315: we set $M^*_{\rm dipole}$ as
316: \begin{equation}
317: \label{eq:EffectiveMassXi}
318: M^*_{\rm dipole} = \sum_\tau \{\xi_z^\tau\}^2 M^\tau = 39.0\,{\rm amu}~.
319: \end{equation}
320: %
321: It should be mentioned that $\xi_z^\tau$ is {\em not} equal to
322: the $d_\alpha^\tau$ of the $\Gamma_{15}$ soft-mode of phonon,
323: because $M^A$, $M^B$, and $M^{\rm O}$ are not identical.
324: %
325: %However, $\bm{\xi}_z$ is approximately equal to the direction
326: %from the full-symmetric cubic atomic configuration
327: %to the atomic configuration which gives the minimum energy
328: %in the case of BaTiO$_3$.~\cite{Hashimoto:N:M:K:S:I:JJAP:43:p6785-6792:2004}
329:
330: The local-mode self-energy $V^{\rm self}(\{\bm{u}\})$ is
331: \begin{multline}
332: \label{eq:V:self}
333: V^{\rm self}(\{\bm{u}\}) =
334: \sum_{i=1}^N
335: \Bigl\{
336: \kappa_2 u^2(\bm{R}_i)
337: + \alpha u^4(\bm{R}_i) + \\
338: \gamma
339: \left[
340: u_y^2(\bm{R}_i) u_z^2(\bm{R}_i) +
341: u_z^2(\bm{R}_i) u_x^2(\bm{R}_i) +
342: u_x^2(\bm{R}_i) u_y^2(\bm{R}_i)
343: \right]
344: \Bigr\}~,
345: \end{multline}
346: where $u^2(\bm{R}_i)
347: = u_x^2(\bm{R}_i)
348: + u_y^2(\bm{R}_i)
349: + u_z^2(\bm{R}_i)$.
350:
351: The long-range dipole-dipole interaction $V^{\rm dpl}(\{\bm{u}\})$ is
352: \begin{equation}
353: \label{eq:V:dpl}
354: V^{\rm dpl}(\{\bm{u}\})=
355: \frac{1}{2}\sum_{i=1}^N \sum_\alpha \sum_{j=1}^N \sum_\beta
356: u_\alpha(\bm{R}_i) \Phi_\ab(\Rij) u_\beta(\bm{R}_j)~,
357: \end{equation}
358: where
359: \begin{equation}
360: \label{eq:Phi}
361: \Phi_\ab(\Rij)
362: = \frac{Z^{*2}}{\eInf}\sum_{\bm{n}}\!'
363: \frac{\dab - 3(\widehat\Rijn)_\alpha(\widehat\Rijn)_\beta}{|\Rij + \bm{n}|^3}~,
364: \end{equation}
365: $\eInf$ is the optical dielectric constant (or refractive index squared),
366: $\dab$ is the Kronecker delta,
367: a hat indicates a unit vector,
368: $\bm{n}$ is the supercell lattice vector
369: \begin{equation}
370: \label{eq:n}
371: n_\alpha = \cdots,\ -2\La a_0,\ -\La a_0,\ 0,\ \La a_0,\ 2\La a_0,\ \cdots\ \ \ ,
372: \end{equation}
373: and $a_0$ is the equilibrium lattice constant.
374: In Eq.~(\ref{eq:Phi}), $\sum'$ indicates that the summation does not include terms
375: for which $\Rij=\bm{n}=0$.
376:
377: We take account of short-range interactions between the optical displacements $\bm{u}(\bm{R})$
378: up to third nearest neighbor (3nn) as
379: \begin{equation}
380: \label{eq:V:short}
381: V^{\rm short}(\{\bm{u}\})=
382: \frac{1}{2}\sum_{i=1}^N \sum_\alpha \sum_j^{\rm 3nn} \sum_\beta
383: u_\alpha(\bm{R}_i) \,J_{ij,\alpha\beta}\, u_\beta(\bm{R}_j)~,
384: \end{equation}
385: where $J_{ij,\alpha\beta}$ is the short-range interaction matrix,
386: which can be classified into
387: 7 independent interaction parameters,~\cite{Zhong:V:R:PRB:v52:p6301:1995}
388: $J_{ij,\alpha\beta} = \pm j_k\ (k=1,\cdots,7)$.
389:
390: In practice,
391: $\kappa_2 u_i^2$ in Eq.~(\ref{eq:V:self}),
392: Eq.~(\ref{eq:V:dpl}), and
393: Eq.~(\ref{eq:V:short}),
394: in which $u_\alpha$ is quadratic,
395: are gathered and calculated in reciprocal space as
396: \begin{equation}
397: \label{eq:V:quad}
398: V^{\rm quad}(\{\bm{u}\})=\frac{1}{2} \sum_{\bm{k}} \sum_{\alpha,\beta}
399: \widetilde{u}_\alpha^*(\bm{k}) \widetilde\Phi_\ab^{\rm quad}(\bm{k}) \widetilde{u}_\beta(\bm{k}),
400: \end{equation}
401: where $\widetilde{u}_\alpha(\bm{k})$ is the Fourier transform
402: \begin{equation}
403: \label{eq:u:FT}
404: \widetilde{u}_\alpha(\bm{k}) = \sum_{\bm{R}}u_\alpha(\bm{R})\exp(-i\bm{k}\cdot\bm{R})~,
405: \end{equation}
406: of $u_\alpha(\bm{R})$, $\widetilde\Phi_\ab^{\rm quad}(\bm{k})$ is similarly the Fourier transform
407: of the quadratic interaction matrix (which is only calculated once
408: at the beginning of simulation\cite{Waghmare:C:B:2003}),
409: %,cond-mat:0512563}
410: and $\bm{k}$ is a reciprocal vector in the first Brillouin zone of the unit cell such as
411: \begin{equation}
412: \label{eq:k}
413: k_\alpha = -\frac{\La-1}{2\La}\Gunit,\ \cdots,\ -\frac{1}{\La}\Gunit,\ 0,\ %
414: \frac{1}{\La}\Gunit,\ \cdots,\ \frac{1}{2}\Gunit\ \ \ .
415: \end{equation}
416:
417: The homogeneous elastic energy $V^{\rm elas,\,homo}(\eta_1,\cdots\!,\eta_6)$ is
418: \begin{eqnarray}
419: \label{eq:V:elas:homo}
420: \nonumber
421: V^{\rm elas,\,homo}(\eta_1,\cdots\!,\eta_6)
422: & = & \frac{N}{2}B_{11}(\eta_1^2+\eta_2^2+\eta_3^2)\\
423: \nonumber
424: & + & N B_{12}(\eta_2\eta_3+\eta_3\eta_1+\eta_1\eta_2)\\
425: & + & \frac{N}{2}B_{44}(\eta_4^2+\eta_5^2+\eta_6^2)~,
426: \end{eqnarray}
427: where $B_{11}$, $B_{12}$, and $B_{44}$ are the elastic constants expressed in energy unit
428: ($B_{11}=a_0^3C_{11}$, $B_{12}=a_0^3C_{12}$, and $B_{44}=a_0^3C_{44}$).
429:
430: The inhomogeneous elastic energy $V^{\rm elas,\,inho}(\{\bm{w}\})$
431: is also calculated in reciprocal space as
432: \begin{equation}
433: \label{eq:V:elas:inho}
434: V^{\rm elas,\,inho}(\{\bm{w}\}) = \frac{1}{2} \sum_{\bm{k}} \sum_{\alpha,\beta}
435: \widetilde{w}_\alpha^*(\bm{k}) \widetilde\Phi_\ab^{\rm elas,\,inho}(\bm{k}) \widetilde{w}_\beta(\bm{k}).
436: \end{equation}
437: For the {\em force constant} matrix $\widetilde\Phi_\ab^{\rm elas,\,inho}(\bm{k})$,
438: we employed the long-wavelength approximation.
439: For instance,
440: the diagonal part is
441: \begin{equation}
442: \label{eq:diagonal}
443: \widetilde\Phi_{xx}^{\rm elas,\,inho}(\bm{k}) = \frac{1}{N}
444: \left[ k_x^2 B_{11} + k_y^2 B_{44} + k_z^2 B_{44} \right]~,
445: \end{equation}
446: and the off-diagonal part is
447: \begin{equation}
448: \label{eq:off-diagonal}
449: \widetilde\Phi_{xy}^{\rm elas,\,inho}(\bm{k}) = \frac{1}{N}
450: \left[ k_xk_yB_{12} + k_xk_yB_{44} \right]~.
451: \end{equation}
452:
453: The coupling between $\{\bm{u}\}$ and homogeneous strain is
454: the same as that given in Ref.~[\onlinecite{King-Smith:V:1994}], i.e.,
455: \begin{equation}
456: \label{eq:V:coup:homo}
457: V^{\rm coup,\,homo}(\{\bm{u}\}, \eta_1,\cdots\!,\eta_6) =
458: \frac{1}{2} \sum_{\bm{R}} \sum_{i=1}^6 \sum_{j=1}^6 \eta_i \, C_{ij} \, y_j(\bm{R})~.
459: \end{equation}
460: Here,
461: $y_1(\bm{R})$ = $u_x^2(\bm{R})$,
462: $y_2(\bm{R})$ = $u_y^2(\bm{R})$,
463: $y_3(\bm{R})$ = $u_z^2(\bm{R})$,
464: $y_4(\bm{R})$ = $u_y(\bm{R})u_z(\bm{R})$,
465: $y_5(\bm{R})$ = $u_z(\bm{R})u_x(\bm{R})$, and
466: $y_6(\bm{R})$ = $u_x(\bm{R})u_y(\bm{R})$,
467: %\begin{widetext}
468: \begin{equation}
469: \label{eq:Cij}
470: {\bf C} = \left(
471: \begin{array}{cccccc}
472: B_{1xx} & B_{1yy} & B_{1yy} & 0 & 0 & 0 \\
473: B_{1yy} & B_{1xx} & B_{1yy} & 0 & 0 & 0 \\
474: B_{1yy} & B_{1yy} & B_{1xx} & 0 & 0 & 0 \\
475: 0 & 0 & 0 & 2B_{4yz} & 0 & 0 \\
476: 0 & 0 & 0 & 0 & 2B_{4yz} & 0 \\
477: 0 & 0 & 0 & 0 & 0 & 2B_{4yz} \\
478: \end{array}
479: \right)~,
480: \end{equation}
481: %\end{widetext}
482: and $B_{1xx}$, $B_{1yy}$, and $B_{4yz}$ are
483: the coupling coefficients defined in Ref.~[\onlinecite{King-Smith:V:1994}].
484:
485: The coupling between $\{\bm{u}\}$ and inhomogeneous strain
486: is also calculated in reciprocal space as
487: \begin{equation}
488: \label{eq:V:coup:inho}
489: V^{\rm coup,\,inho}(\{\bm{u}\}, \{\bm{w}\}) = \frac{1}{2} \sum_{\bm{k}} \sum_{\alpha} \sum_{i=1}^6
490: \widetilde{w}_\alpha(\bm{k}) \widetilde{B}_{\alpha i}(\bm{k}) \widetilde{y}_i(\bm{k})~,
491: \end{equation}
492: where $\widetilde{w}_\alpha(\bm{k})$ and $\widetilde{y}_i(\bm{k})$ are
493: the Fourier transforms of $w_\alpha(\bm{R})$ and $y_i(\bm{R})$, respectively.
494: % \begin{equation}
495: % \label{eq:w:FT}
496: % \widetilde{w}_\alpha(\bm{k}) = \sum_{\bm{R}}w_\alpha(\bm{R})\exp(-i\bm{k}\cdot\bm{R})
497: % \end{equation}
498: % and
499: % \begin{equation}
500: % \label{eq:w:FT}
501: % \widetilde{y}_i(\bm{k}) = \sum_{\bm{R}}y_i(\bm{R})\exp(-i\bm{k}\cdot\bm{R})~,
502: % \end{equation}
503: For the $3\times6$ coupling matrix ${\bf B}(\bm{k})$,
504: we again employed the long-wavelength approximation
505: \begin{widetext}
506: \begin{equation}
507: \label{eq:Balphai}
508: \widetilde{\bf B}(\bm{k}) = \frac{1}{N}\left(
509: \begin{array}{cccccc}
510: k_xB_{1xx} & k_xB_{1yy} & k_xB_{1yy} & 0 & 2k_zB_{4yz} & 2k_yB_{4yz} \\
511: k_yB_{1yy} & k_yB_{1xx} & k_yB_{1yy} & 2k_zB_{4yz} & 0 & 2k_xB_{4yz} \\
512: k_zB_{1yy} & k_zB_{1yy} & k_zB_{1xx} & 2k_yB_{4yz} & 2k_xB_{4yz} & 0 \\
513: \end{array}
514: \right)~.
515: \end{equation}
516: \end{widetext}
517:
518: In the present MD simulations of BaTiO$_3$,
519: the parameters from Refs.~[\onlinecite{Zhong:V:R:1994}] and [\onlinecite{Zhong:V:R:PRB:v52:p6301:1995}],
520: which are determined by first-principles calculations,
521: are employed.
522: As mentioned in Refs.~[\onlinecite{Zhong:V:R:1994}] and [\onlinecite{Zhong:V:R:PRB:v52:p6301:1995}],
523: this parameter set leads to an underestimation of the Curie temperature $T_{\rm C}$.
524: To correct this underestimation, we follow these references in applying a
525: negative pressure of $p=-5.0$~GPa in all simulations.
526:
527: \subsection{Molecular Dynamics}
528: MD simulations with the effective Hamiltonian of Eq.~(\ref{eq:Effective:Hamiltonian})
529: are performed in the canonical ensemble
530: using the Nos\'e-Poincar\'e thermostat.~\cite{Bond:L:L:JComputPhys:151:p114-134:1999}
531: This simplectic thermostat is so efficient that we can set the
532: time step to $\Delta t=2$~fs.
533: In our present simulations,
534: we thermalize the system for 40,000 time steps,
535: after which we average the properties for 10,000 time steps.
536:
537: In Fig.~\ref{fig:flow} we roughly illustrate how to calculate
538: the forces exerted on $u_\alpha(\bm{R})$
539: with $\widetilde\Phi_\ab^{\rm quad}(\bm{k})$ in Eq.~(\ref{eq:V:quad})
540: and how the time evolution is simulated.
541: First, $u_\alpha(\bm{R})$ is FFTed to $\widetilde{u}_\alpha(\bm{k})$,
542: the force $\widetilde{F}_\alpha(\bm{k}) = -\sum_{\beta} \widetilde\Phi_\ab^{\rm quad}(\bm{k}) \widetilde{u}_\beta(\bm{k})$ is calculated in reciprocal space,
543: and then the force in real space is obtained
544: by the inverse FFT of $\widetilde{F}_\alpha(\bm{k})$.
545: %More precisely,
546: In practice, updates of $u_\alpha(\bm{R})$ and
547: $\dot{u}_\alpha(\bm{R})=\frac{\partial }{\partial t}u_\alpha(\bm{R})$ are processed
548: in the manner of the Nos\'e-Poincar\'e thermostat.
549: \begin{figure}
550: \centering
551: \includegraphics[width=60mm]{flow.eps}
552: \caption{Simplified flow chart for calculating forces on $u_\alpha(\bm{R})$.
553: Fast Fourier transform (FFT) and inverse FFT (IFFT) enable
554: rapid calculation of long-range dipole-dipole interactions.}
555: \label{fig:flow}
556: \end{figure}
557:
558: The homogeneous strain components $\eta_1,\cdots\!,\eta_6$ are determined
559: by solving
560: \begin{multline}
561: \label{eq:eta:minimize}
562: \frac{\partial}{\partial\eta_i}\Bigl[V^{\rm elas,\,homo}(\eta_1,\cdots\!,\eta_6)\\
563: + V^{\rm coup,\,homo}(\{\bm{u}\}, \eta_1,\cdots\!,\eta_6)\Bigr] = 0
564: \end{multline}
565: at each time step according to $\{\bm{u}\}$
566: so that $\eta_1,\cdots\!,\eta_6$ minimize
567: $V^{\rm elas,\,homo}(\eta_1,\cdots\!,\eta_6) +
568: V^{\rm coup,\,homo}(\{\bm{u}\}, \eta_1,\cdots\!,\eta_6)$.
569: %
570: While the local acoustic displacement $w_\alpha(\bm{R})$ could be treated
571: as dynamical variables using the effective mass $M^*_{\rm acoustic}$,
572: we have instead chosen to integrate out these variables in a manner
573: similar to the treatment of the homogeneous strain. That is,
574: $w_\alpha(\bm{R})$ is determined
575: so that $V^{\rm elas,\,inho}(\{\bm{w}\})+V^{\rm coup,\,inho}(\{\bm{u}\}, \{\bm{w}\})$
576: becomes minimum at each time step according to $u_\alpha(\bm{R})$.
577: Technically,
578: the minimization is performed by solving the linear set of equations
579: \begin{equation}
580: \label{eq:w:minimize}
581: \widetilde\Phi^{\rm elas,\,inho}(\bm{k}) \widetilde{\bm{w}}(\bm{k}) +
582: \widetilde{\bf B}(\bm{k})\widetilde{\bm{y}}(\bm{k})=\bm{0}
583: \end{equation}
584: for each $\bm{k}$ in reciprocal space.
585:
586: \subsection{Ferroelectric Thin Films}
587: \label{subsec:FerroelectricThinFilm}
588:
589: If a ferroelectric thin film is placed in isolation in vacuum
590: without electrodes as depicted in Fig.~\ref{fig:thinfilmes}(a),
591: its spontaneous polarization $\bm{P}=(P_x, P_y, P_z)$,
592: which is represented by a thick arrow in the figure,
593: induces charges
594: $\pm\sigma_{\rm ind}=\pm P_z$ at both surfaces,
595: and the induced charges cause a full depolarization field in the thin film,
596: $\bm{\mathcal{E}}_{\rm d}=-4\pi\sigma_{\rm ind}\widehat{\bm{z}}
597: =-4\pi P_z \widehat{\bm{z}}$.
598: On the other hand,
599: if the ferroelectric thin film is placed between short-circuited
600: perfect electrodes as depicted in Fig.~\ref{fig:thinfilmes}(b),
601: the induced charges are fully canceled by
602: free charges $\sigma_{\rm free}$ arising at both surfaces of the electrodes,
603: $\bm{\mathcal{E}}_{\rm d}=-4\pi(\sigma_{\rm ind}+\sigma_{\rm free})\widehat{\bm{z}}=0$.
604: This geometric circumstance can be simulated
605: with the doubly periodic supercell as depicted in Fig.~\ref{fig:thinfilmes}(c),
606: because the two electrodes act
607: as two electrostatic mirrors facing each other,
608: and the mirrors make oppositely charged infinite mirror images
609: beyond the electrodes.
610:
611: We can also introduce dead layers of thickness $d$ between the ferroelectric
612: thin film and electrodes by constraining the local soft-mode amplitudes to
613: vanish ($\bm{u}=0$) in these layers, as illustrated
614: in Fig.~\ref{fig:thinfilmes}(d).
615: With the dead layers,
616: the infinite mirror images beyond the electrodes become $\frac{l}{l+d}$ more sparse
617: than images of the without-dead-layer configuration.
618: Consequently, the free charges arising at the electrode surfaces
619: decrease to $\sigma_{\rm free}=-\frac{l}{l+d}\sigma_{\rm ind}$,
620: where $l$ is the ferroelectric film thickness.
621: This simulates short-circuited imperfect electrodes
622: resulting in a depolarization field of
623: \begin{equation}
624: \label{eq:depolarization}
625: \bm{\mathcal{E}}_{\rm d} = -4 \pi \frac{d}{l+d} P_z \widehat{\bm{z}}~.
626: \end{equation}
627: We can also use a doubly periodic supercell with dead layers for this case.
628: %
629: % \dvm{You wrote ``by representing the {\em sum} of the effective screening length
630: % as the dead layer thickness $d$''. I found this confusing. A sum has to be
631: % a sum of something plural. Do you mean ``sum of the effective screening lengths''?
632: % Which effective screening lengths? I tried rewriting this to make it clearer,
633: % but let me know if what I have written is not what you intended.}
634: % \tnm{Thank you. I changed ``the electrode-metal-interface'' to ``the metal electrode''}
635: \uline{Physically, the depolarization field of Eq.~(\ref{eq:depolarization})
636: can arise either from the presence of a dead layer in the ferroelectric
637: near the interface, or from imperfect screening at the metal electrode,
638: or both. We can define an effective screening length for
639: each of these effects, and we interpret the ``dead-layer thickness'' $d$ of
640: our model as corresponding to the {\em sum} of these two physical screening lengths.
641: The screening length associated with the electrode interface appears in
642: Eq.~(16) of Ref.~[\onlinecite{MEHTA:S:J:JAP:44:p3379-3385:1973}] and
643: Eq.~(1) of Ref.~[\onlinecite{Dawber:C:L:S:JPhys-CondesMatter:15:pL393-L398:2003}]
644: and is discussed for the SrRuO$_3$/BaTiO$_3$ interface in
645: Refs.~[\onlinecite{Sai:K:R:PRB:72:p020101:2005}],
646: [\onlinecite{Kim:J:K:C:L:Y:S:N:PRL:95:p237602:2005}], and
647: [\onlinecite{Gerra:T:S:P:PRL:96:p107603:2006}].
648: Therefore, while the model does not explicitly incorporate
649: information about the interface screening, this information is
650: effectively included in the definition of the total
651: screening length $d$ in our model. Thus, for example, simulations
652: at constant $d$ for various film thicknesses can give the thickness
653: dependence of the properties of capacitors with a certain interface
654: structure.}
655:
656: \begin{figure}
657: \centering
658: \includegraphics[width=80mm]{capacitor.eps}
659: \caption{Schematic illustrations of ferroelectric thin films of thickness
660: $l$ unit cells (here $l=2$).
661: (a)~Isolated thin film in vacuum.
662: (b) Thin film sandwiched between short-circuited perfect electrodes.
663: Doubly periodic boundary conditions
664: for simulations of films sandwiched between
665: perfect and imperfect short-circuited electrodes
666: are depicted in (c) and (d), respectively.
667: Horizontal thick lines marked with ``E''
668: represent the electrostatic mirrors used to model electrodes.
669: They are a distance $d/2$
670: away from the ferroelectric film surface ($d=0$ in~(c), $d=1$ in~(d)).
671: Each thin arrow represents a local dipole within
672: a unit cell ($a^3=3.94$~\AA$^3$) of the BaTiO$_3$ crystal.
673: Thick dashed lines enclose the periodic cell used for simulations.}
674: \label{fig:thinfilmes}
675: \end{figure}
676:
677: In the present MD simulations,
678: the local soft-mode amplitude vectors $\bm{u}$
679: in dead layers are fixed to zero by the infinitely large mass.
680: This infinitely-large-mass trick is congenial
681: to the Nos\'e-Poincar\'e thermostat for maintaining the Nos\'e-Poincar\'e Hamiltonian at zero.
682: Moreover, this treatment also has another advantage in that the
683: short-range interactions
684: between the surfaces of ferroelectric thin film and the electrodes
685: are automatically truncated.
686:
687: The depolarization field $\bm{\mathcal{E}}_{\rm d}$ increases the total energy of
688: the ferroelectric thin film by
689: $-\bm{P}\cdot\bm{\mathcal{E}}_{\rm d}=4\pi\frac{d}{l+d}P_z^2$.
690: To avoid forming a depolarization field in ferroelectric thin films,
691: it is known that the films often develop
692: striped domain structures.~\cite{Kittel.PhysRev.70.965,Fong:S:S:E:A:F:T:Science:304:p1650-1653:2004,
693: Bratkovsky:L:PRL:84:p3177-3180:2000,Bratkovsky:L:PRL:87:p179703:2001}
694: % The striped domain structure having the wavelength $\lambda$
695: % may be as depicted in Fig.~\ref{fig:striped},
696: % if surface relaxations are omitted.
697: % \begin{figure}
698: % \centering
699: % \includegraphics[width=70mm]{striped.eps}
700: % \caption{Striped domain structure of ferroelectric thin film.
701: % FE denotes thin ferroelectrics and ER denotes electrodes.
702: % Oppositely charged infinite mirror images are also shown.}
703: % \label{fig:striped}
704: % \end{figure}
705: The introduction of the striped domain structure
706: can eliminate some part of the energy increase $4 \pi \frac{d}{l+d}P_z^2$,
707: because $P_z$ becomes zero on average.
708: However, the striped domain structure involves an energy
709: cost in the short-range interaction $V^{\rm short}(\{\bm{u}\})$,
710: because it has domain boundaries
711: between which $\bm{u}$ has opposite direction $\pm z$.
712: The shorter the wavelength $\lambda$ of the striped domain structure,
713: the weaker the depolarization field,
714: but the higher the short-range interaction energy.
715: The ground state of a ferroelectric thin film
716: will be decided by a competition between
717: the long-range dipole-dipole interactions which favor a short-period
718: domain structure, and domain-wall energy that arises from the
719: short-range interactions and favors a uniformly polarized structure
720: or a longer-period striped structure.
721: In some previous works~\cite{Lai:P:N:K:F:B:S:PRL:96:p137602:2006,
722: Lai:P:K:B:S:PRB:75:p085412:2007,Lai:P:K:B:S:APL:91:p152909:2007},
723: the imperfect screening
724: was mimicked with a parameter.
725: On the other hand,
726: our method with doubly periodic boundary condition
727: does not require any parameters,
728: because the effect of imperfectness of electrodes
729: is automatically and implicitly included
730: in the long-range dipole-dipole interaction $V^{\rm dpl}(\{\bm{u}\})$.
731:
732: \section{Results and Discussion}
733: \label{sec:results}
734: \subsection{Bulk BaTiO$_3$}
735: \label{subsec:Bulk}
736: We first check the reliability of our MD program by comparing results
737: of our simulations for bulk BaTiO$_3$ with earlier work based on
738: the same effective Hamiltonian\cite{Zhong:V:R:1994,Zhong:V:R:PRB:v52:p6301:1995}.
739: We used a system size of $L_x\times L_y\times L_z = 16 \times 16 \times 16$
740: and small temperature steps in heating-up ($+5$~K/step) and cooling-down
741: ($-5$~K/step) simulations, with initial configuration generated randomly:
742: $\langle u_\alpha \rangle = 0.07~{\rm \AA}$ and
743: $\langle u_\alpha^2 \rangle - \langle u_\alpha \rangle^2=(0.02~{\rm \AA})^2$.
744: We have also checked that there was no dependence of results of these simulations
745: on initial configurations.
746: The temperature dependence of the homogeneous strain components (see Fig.~\ref{fig:strain}),
747: which are the secondary order parameters of ferroelectric phase transitions,
748: exhibits the correct sequence of phase transitions in BaTiO$_3$ known experimentally.
749: \begin{figure}
750: \centering
751: \includegraphics[width=75mm]{strain.heating-cooling-5GPa-Color-New.eps}
752: \caption{Average homogeneous strains $e_{xx}$, $e_{yy}$, and $e_{zz}$ as a function
753: of temperature in heating-up ($+5$~K/simulation, solid lines)
754: and cooling-down ($-5$~K/simulation, dashed lines)
755: simulations for a $16\times16\times16$ supercell.
756: Strains are measured relative to the LDA minimum-energy cubic structure
757: with lattice constant 3.948~\AA.}
758: \label{fig:strain}
759: \end{figure}
760: Even under the negative pressure $p=-5.0$~GPa, the paraelectric to ferroelectric
761: transition temperature $T_{\rm C}$ is underestimated at around 320~K in comparison with
762: the experimental value of $T_{\rm C}=408$~K. Our estimates of $T_{\rm C}$'s
763: agree fairly well with the ones reported in Ref.~[\onlinecite{Zhong:V:R:PRB:v52:p6301:1995}].
764: The relatively weak first-order nature of the cubic-to-tetragonal phase transition in
765: comparison with the first-order tetragonal-to-orthorhombic and orthorhombic-to-rhombohedral
766: phase transitions is evident in the width of the temperature intervals of hysteresis
767: (see Fig.~\ref{fig:strain}). We note that the ability to simulate time-dependent phenomena
768: is one of advantages of MD simulations compared to Monte-Carlo simulations.
769:
770: \subsection{BaTiO$_3$ ferroelectric thin-film capacitors}
771: \label{subsec:Capacitors}
772: We now simulate and analyze the behavior of epitaxially grown films of
773: BaTiO$_3$ on GdScO$_3$ substrates.~\cite{Choi:B:L:S:S:U:R:C:P:G:C:S:E:Science:306:p1005-1009:2004}
774: \uline{In our simulations, we represent this with 1\% in-plane biaxial compressive strain
775: by maintaining the homogeneous strain $\eta_1=\eta_2=-0.01$ and $\eta_6=0$.
776: In other words, we maintained the {\em average} lattice constants $a$ and $b$ at $0.99a_0$ and angle $\gamma$ at $90^\circ$.}
777: We use supercell sizes of $L_x\times L_y\times L_z = 32 \times 32 \times 2(l+d)$ and
778: $40 \times 40 \times 2(l+d)$
779: and simulate ferroelectric layers of thickness $l$
780: sandwiched between two short-circuited electrodes
781: with ($d=1$) and without ($d=0$) dead layers.
782: This is accomplished through use of doubly periodic boundary conditions as explained earlier.
783:
784: Both the heating-up and cooling-down simulations are started with an initial
785: configuration of
786: $\langle u_x \rangle = \langle u_y \rangle=0$,
787: $\langle u_z \rangle=0.07~{\rm \AA}$ and
788: $\langle u_\alpha^2 \rangle - \langle u_\alpha \rangle^2=(0.02~{\rm \AA})^2$.
789: In the cooling-down simulations, which start at a sufficiently high temperature,
790: the initial configuration changes to an unpolarized one ($\langle u_z \rangle=0$)
791: during thermalization. We monitor the temperature dependence of $\langle u_\alpha \rangle$
792: and $\langle u_\alpha^2 \rangle$ for thin films with thicknesses $l=15$, 31, 127, and
793: 255 with dead layers $d=1$ and a thin film with thickness $l=32$ without dead layers
794: ($d=0$) (see Fig.~\ref{fig:epit-heat-cool} and animations in the EPAPS\cite{EPAPS}\,). % Do not remove this "\,".
795: The behavior of the film with no dead layer is the same in heating and
796: cooling simulations. In contrast, for the films with a dead layer ($d=1$),
797: the transition behavior exhibited by $\langle u_z \rangle=0$ is rather
798: different in heating and cooling simulations, although
799: the temperature dependence of $\langle u_z^2 \rangle$
800: is almost the same in both the kinds of simulations.
801: %
802: \begin{figure}
803: \centering
804: \includegraphics[width=80mm]{epit32x32-narrow.eps}
805: \caption{$\langle u_z \rangle$ of
806: heating-up (solid lines) and cooling-down (dashed lines)
807: molecular-dynamics simulations of
808: BaTiO$_3$ thin-film capacitors
809: with short-circuited electrodes
810: under 1\% in-plane biaxial compressive strain for
811: (a) thickness $l=15$~layer with dead layer $d=1$,
812: (b) $l=31$ with $d=1$,
813: (c) $l=127$ with $d=1$,
814: (d) $l=255$ with $d=1$, and
815: (e) $l=32$ without dead layer ($d=0$).
816: $\sqrt{\langle u_z^2 \rangle}$ are also plotted in (a)-(e).
817: In (c)-(e), heating-up $\sqrt{\langle u_z^2 \rangle}$
818: and cooling-down $\sqrt{\langle u_z^2 \rangle}$ are almost identical.
819: $\sqrt{\langle u_x^2 \rangle}$ is plotted only in (e),
820: because the behaviors of
821: $\sqrt{\langle u_x^2 \rangle}$ and
822: $\sqrt{\langle u_y^2 \rangle}$ are essentially identical in cases (a)-(e)
823: (for both heating-up and cooling-down).
824: Supercells are of size $32\times 32\times 2(l+d)$.
825: Animations of these cooling-down and heating-up simulations are also available
826: in the EPAPS\cite{EPAPS}.}
827: \label{fig:epit-heat-cool}
828: \end{figure}
829: %
830: % \dvm{The citations to the EPAPS material in the caption of
831: % Fig.~\ref{fig:epit-heat-cool} and elsewhere (other captions and
832: % main text) seems unnecessarily wordy. I would suggest to create an explanatory
833: % footnote that contains the webpage and temporary locations once and for
834: % all, and then cite this from various places in the paper.}
835: % \tnm{Done with {\tt footnote} command.}
836: %
837: In the heating-up simulations,
838: the discontinuity in $\langle u_z \rangle$ as a function of temperature marks a
839: transition from a ferroelectric state with almost uniform out-of-plane polarization
840: (Fig.~\ref{fig:snapshot}(a)) to one with a striped domain
841: structure (Fig.~\ref{fig:snapshot}(b) and (c)). We find that this transition temperature,
842: $T_{\rm S}(l,d=1)$, exhibits a strong dependence on size $l$.
843: We note that this transition is missing in the cooling-down simulations;
844: just above $T_{\rm S}$, striped domain structures appear and the stripes remain and be frozen
845: at $T<T_{\rm S}$.
846: The temperature $T_{\rm C}(l,d=1)$ at which
847: $\sqrt{\langle u_z^2 \rangle}(T)$ exhibits a change in its slope marks another transition,
848: namely from
849: a striped domain phase to a paraelectric phase. $T_{\rm C}(l,d=1)$ depends relatively weakly
850: on the film thickness.
851: \begin{figure}
852: \centering
853: \includegraphics[width=88mm]{epit-32x32-L015-D1-5.0GPa-100K-three.eps}
854: \caption{Snapshots at $T=100$~K in
855: heating-up ((a)) and
856: cooling-down ((b) and (c)) simulations of ferroelectric
857: thin-film capacitors of $l=15$ with $d=1$.
858: (a) and (b) are horizontal slices.
859: (c) is a vertical cross section.
860: Points of snapshots are indicated with ``X'' marks in Fig.~\ref{fig:epit-heat-cool}(a).
861: In horizontal slices,
862: the $+z$-polarized and $-z$-polarized sites are
863: denoted by $\square$ and $\blacksquare$, respectively.
864: In vertical cross sections,
865: the dipole moments of each site are projected onto the $xz$-plane and indicated with arrows.
866: Layers which do not have arrows are dead layers.}
867: \label{fig:snapshot}
868: \end{figure}
869: \uline{It should be mentioned that the results of heating-up simulations
870: with a phase transition from the single domain state to the striped domain state
871: at $T_{\rm S}$ below $T_{\rm C}$
872: agree with thermodynamical treatment
873: of ferroelectric capacitors with dead layers
874: by Chensky and Tarasenko.~\cite{Chensky:Tarasenko:SovPhysJETP:56:p618:1982}}
875:
876: For films with $d=1$,
877: $T_{\rm S}$ is 150 and 210~K for $l=15$ and $l=31$ respectively,
878: which is lower than the bulk transition temperature ($T_{\rm C}\approx320$~K).
879: However, for $d=1$ films with $l=127$ and $l=255$, $T_{\rm S}$ is enhanced to 520 and 610 K
880: respectively, well above the bulk $T_{\rm C}$.
881: In the infinite thickness limit ($l\rightarrow\infty$),
882: it appears that $T_{\rm S}(l,d=1)$ tends to the $T_{\rm C}$ of thick films with no dead layer ($d=0$),
883: since $T_{\rm C}$ is 650 K for $l=32$ and $d=0$.
884: In the $d=1$ cases with $l=127$ and $l=255$, the effect arising from the depolarization field
885: weakens significantly, and the enhancement of $T_{\rm S}$ results from
886: the in-plane biaxial compressive strain.
887: In the $d=0$ case with $l=32$, there is no depolarization field and enhancement
888: of $T_{\rm C}$ by the in-plane biaxial compressive strain is effective even in
889: very thin films. We note that $\sqrt{\langle u_z^2 \rangle}$ and $\sqrt{\langle u_x^2 \rangle}$
890: are distinct even at high temperatures (see Fig.~\ref{fig:epit-heat-cool}e), indicating that
891: the symmetry of the paraelectric phase is broken by
892: the presence of the epitaxial constraint and
893: the electrodes, as well as correlations between local dipoles and their images.
894:
895: For films with a dead layer ($d=1$), the striped domain structures appear in the
896: cooling-down simulations at low temperatures for all values of thicknesses $l$ explored here
897: (see Fig.~\ref{fig:snapshot}(b) and (c) for the case of $l=15$ with $d=1$,
898: and Fig.~\ref{fig:domain} for various $l$).
899: \begin{figure}
900: \centering
901: \includegraphics[width=85mm]{slice32x32.eps}
902: \caption{Horizontal slices of snapshots at 100~K
903: in cooling-down simulations of ferroelectric
904: thin-film capacitors with single dead layer ($d=1$)
905: of various thickness $l=7$, 15, 31, 63, 127, and 255.
906: The $+z$-polarized and $-z$-polarized sites are
907: denoted by $\square$ and $\blacksquare$, respectively.}
908: \label{fig:domain}
909: \end{figure}
910: As shown in Table~\ref{tab:thickness-dep},
911: the wavevector $\bm{k}$ of the striped domain,
912: at which $\widetilde{u}_z(\bm{k})$ has the largest amplitude $|\widetilde{u}_z(\bm{k})|$,
913: exhibits an interesting dependence on thickness $l$.
914: We have determined $\bm{k}$ for two supercell sizes,
915: $32 \times 32 \times 2(l+d)$ and $40 \times 40 \times 2(l+d)$,
916: to identify supercell-size effects.
917: It can be seen that, except for the data for $l=255$,
918: $\bm{k}$ tends to be along the in-plane $\{ 110 \}$ direction,
919: consistent with earlier reports.~\cite{Tinte:Stachiotti:PhysRevB.64.235403,Lai:P:K:B:S:PRB:75:p085412:2007}
920: The simulated striped domain structure for $l=255$,
921: which is parallel to the $\{ 100 \}$ direction, is likely to be an artifact of the
922: finite supercell-size:
923: $L_x\times L_y=32\times 32$ or even $40\times 40$ are too small
924: to allow for the formation of a sufficiently thick $\{ 110 \}$ striped domain.
925: \begin{table}
926: \caption{Dependence of the wavevector $\bm{k}/2\pi$ of the striped domain structure
927: on thickness $l$ in the thin-film BaTiO$_3$ capacitor with a dead layer ($d=1$).}
928: \label{tab:thickness-dep}
929: \centering
930: \begin{tabular}{rrccclrcccl}
931: \hline
932: $l$ & \multicolumn{5}{c}{$32 \times 32 \times 2(l+d)$} & \multicolumn{5}{c}{$40 \times 40 \times 2(l+d)$} \\
933: \hline
934: 7 & \ \ \ \ \{ & 4/32 & 3/32 & 0 & \} \ \ \ \ \ & \ \ \ \ \{ & 5/40 & 5/40 & 0 & \} \ \ \ \ \ \\
935: 15 & \ \ \ \ \{ & 3/32 & 3/32 & 0 & \} \ \ \ \ \ & \ \ \ \ \{ & 4/40 & 3/40 & 0 & \} \ \ \ \ \ \\
936: 31 & \ \ \ \ \{ & 2/32 & 2/32 & 0 & \} \ \ \ \ \ & \ \ \ \ \{ & 3/40 & 2/40 & 0 & \} \ \ \ \ \ \\
937: 63 & \ \ \ \ \{ & 2/32 & 1/32 & 0 & \} \ \ \ \ \ & \ \ \ \ \{ & 2/40 & 2/40 & 0 & \} \ \ \ \ \ \\
938: 127 & \ \ \ \ \{ & 1/32 & 1/32 & 0 & \} \ \ \ \ \ & \ \ \ \ \{ & 1/40 & 1/40 & 0 & \} \ \ \ \ \ \\
939: 255 & \ \ \ \ \{ & 1/32 & 0/32 & 0 & \} \ \ \ \ \ & \ \ \ \ \{ & 1/40 & 0/40 & 0 & \} \ \ \ \ \ \\
940: \hline
941: \end{tabular}
942: \end{table}
943: The wavelength $\lambda=2\pi/|\bm{k}|$ of dominant periodicity of the domain pattern
944: is shown as a function of thickness $l$ in Fig.~\ref{fig:thickness-dep}, where it is
945: evident that the thinner films have smaller $\lambda$
946: to avoid the stronger depolarization field, Eq.~(\ref{eq:depolarization}).
947: The fitting shown in Fig.~\ref{fig:thickness-dep} suggests a square-root dependence~\cite{Kittel.PhysRev.70.965}
948: on $l$ (the result for $l=255$ is not included in the fit).
949: Extensive simulations at larger length scales would probably be required to clarify
950: further the dependence of the domain period of these striped structures on film thickness
951: and dead-layer thickness.
952: \begin{figure}
953: \centering
954: \includegraphics[width=55mm]{k.eps}
955: \caption{Calculated thickness $l$ dependence of wavelength $\lambda$ of striped
956: domain structures in thin film BaTiO$_3$ capacitors with a dead layer ($d=1$).
957: $+$ marks are from $32 \times 32 \times 2(l+d)$ supercell
958: calculations and $\times$ are those of $40 \times 40 \times 2(l+d)$.
959: Data of $l<=127$ are fitted with $\lambda=c\sqrt{l}$ (dotted line). $l=255$ data are
960: omitted, because of their large supercell-size dependence.}
961: \label{fig:thickness-dep}
962: \end{figure}
963:
964: %It can be said that
965: The stark difference in the behavior of $\langle u_z \rangle$ in heating-up and
966: cooling-down simulations hints that
967: the (almost) uniformly polarized state and the
968: $\langle u_z \rangle=0$ striped domain states
969: are frozen and
970: thermal hopping between them may be almost impossible
971: at low temperatures.
972: To understand why both uniformly polarized and striped domain states
973: are stable and thermal hopping between them are difficult,
974: we investigated the effective potential-energy surfaces for
975: striped domain structures of various stripe wavevectors $\bm{k}$
976: and various $l$ for thin-film ferroelectric capacitors with and
977: without the dead layer (see Fig.~\ref{fig:potential:surface}(a)-(e)).
978: \begin{figure}
979: \centering
980: \includegraphics[width=88mm]{all-potential-surface-110.eps}
981: \caption{Effective potential surfaces of BaTiO$_3$ thin-film capacitors
982: with short-circuited electrodes:
983: (a)-(e), under 1\% in-plane biaxial compressive strain arising from epitaxial constraints;
984: (f)-(j), without epitaxial constraints (i.e., for ``free'' films).
985: The thicknesses of ferroelectric films and dead layers are indicated in each panel
986: with $l$ and $d$ respectively.
987: Total energies as functions of $u_z$
988: are compared among striped domain structures with wavevectors $\bm{k}$ parallel to $(110)$.
989: $\bm{k}=(000)$ corresponds to the uniformly polarized structure.
990: The zero of the energy scale is placed at the total energy of the non-polarized $u_z=0$ structure.
991: A negative pressure $p=-5$~GPa is applied to correct the underestimation in $T_{\rm C}\,$.}
992: \label{fig:potential:surface}
993: \end{figure}
994: Omitting surface relaxations in this analysis may be reasonable
995: because the surface relaxations are confined to the surface region
996: of the ferroelectric thin films as shown in Fig.~\ref{fig:snapshot}(c).
997: It can be seen that thinner ferroelectric film have
998: a shorter stripe wavelength $\lambda=2\pi/|\bm{k}|$
999: in their ground states. As the thickness $l$ is increased to
1000: $l \approx 127$, the ground state changes from
1001: the striped domain structure to the out-of-plane uniformly-polarized ferroelectric
1002: structure ($\bm{k}=(000)$). However, on the time scale of our simulations ($\approx 1$~ns),
1003: even at $l\approx255$ there is no hopping from the striped domain metastable
1004: state to the uniformly polarized ground state (Fig.~\ref{fig:epit-heat-cool}(d)).
1005: It can also be seen in Fig.~\ref{fig:potential:surface}(a)-(e) that the
1006: magnitude of $u_z$ which gives the minimum-energy ground state becomes larger,
1007: and the minimum energy gets deeper, as $l$ increases, in
1008: good correspondence with the thickness dependence of $T_{\rm C}$.
1009: The trend of $\bm{k}$ with $l$ also shows
1010: good agreement with the simulated values shown in Table~\ref{tab:thickness-dep}.
1011: \uline{The simulated stability of the out-of-plane uniformly-polarized states
1012: against the energetically lower striped-domain states
1013: in thinner ($l<127$) films at low temperature seems
1014: to give support to the recent idea of elastic stabilization of a
1015: homogeneously polarized state in strained ultrathin films.\cite{Pertsev:K:PRL:98:p257603:2007}}
1016: As shown in Fig.~\ref{fig:epit-vs-free}(a), the polarization switching in the epitaxially
1017: constrained film may be suppressed by the presence of a potential barrier that prevents
1018: hopping between the uniformly-polarized and striped-domain states.
1019: \begin{figure}
1020: \centering
1021: \includegraphics[width=55mm]{epit-vs-free.eps}
1022: \caption{Schematic comparison between the epitaxially constrained film and the ``free'' film.
1023: In the epitaxially constrained film, switching may have to climb over a potential barrier,
1024: but, in the ``free'' film,
1025: dipoles can be easily rotated and
1026: switching can go around a valley of the potential.}
1027: \label{fig:epit-vs-free}
1028: \end{figure}
1029: %From Fig.~\ref{fig:potential:surin the cooling-down simulationsface},
1030: For $l \le 127$ with $d=1$, it is expected that a uniformly polarized film would evolve
1031: into a striped domain state, or vice versa, over a sufficiently long time at $T<T_{\rm S}$.
1032: However, the time scale of the evolution might be very much longer than the
1033: present simulation time scale ($\sim$1~ns).
1034: It might also be expected that, in the cooling-down simulations of films with
1035: $l \le 127$ and $d=1$,
1036: the uniformly polarized state is obtained at $T<T_{\rm S}$.
1037: Instead, however, we find that stripes appear. A close inspection
1038: of the simulations shows that the stripes form slightly above
1039: $T_{\rm S}$, initially in a somewhat disordered fashion, presumably
1040: because such a structure provides a good compromise between
1041: energetic and entropic considerations. The stripes then get frozen
1042: into place, and become better ordered, as the temperature is reduced
1043: below $T<T_{\rm S}$.
1044: Conversely,
1045: in the case of $d=0$ (Fig.~\ref{fig:epit-heat-cool}(e),
1046: depolarization field $\mathcal{E}_{\rm d}=0$),
1047: the striped domain stricture does not appear during
1048: the heating-up and cooling-down simulations.
1049: This may be because, when $\mathcal{E}_{\rm d}=0$,
1050: there is no reason or chance to form a striped domain structure
1051: even just above $T_{\rm C}$.
1052: At $T_{\rm C}$,
1053: direct phase transition form paraelectric phase to
1054: uniformly polarized ferroelectric phase occurs.
1055: Then, below $T_{\rm C}$,
1056: the system tends to be in its ground state,
1057: the uniformly polarized ferroelectric structure.
1058:
1059: \subsection{Hysteresis loops}
1060: \label{subsec:HysteresisLoops}
1061: A measurement of polarization typically involves use of a
1062: triangle-wave electric field for recording the ferroelectric hysteresis
1063: loops (inset of Fig.~\ref{fig:step}).
1064: The hysteresis loops and coercive fields $\mathcal{E}_{\rm c}$
1065: depend on the amplitude $\mathcal{E}_0$ and frequency $f$ of the applied fields.
1066: %To simulate that and to use the Nos\'e-Poincar\'e thermostat,
1067: We simulate hysteresis here using triangle-wave with steps (width $\Delta t\, n_{\rm steps}$
1068: and height $\Delta \mathcal{E}$) as sketched schematically in Fig.~\ref{fig:step}.
1069: Thus, the frequency of the applied field in our simulations is
1070: $f=\Delta \mathcal{E}/4\, \Delta t\, n_{\rm steps}\mathcal{E}_0$.
1071: \begin{figure}
1072: \centering
1073: \includegraphics[width=60mm]{step.eps}
1074: \caption{Schematic illustrations of triangle-wave electric field used
1075: to measure ferroelectric hysteresis loops experimentally (inset)
1076: and in the present simulations.}
1077: \label{fig:step}
1078: \end{figure}
1079: We used supercell sizes of $L_x\times L_y\times L_z = 16 \times 16 \times 2(l+d)$ in
1080: simulations of hysteresis loops for ferroelectric thin-film capacitors
1081: with 1\% in-plane biaxial compressive strain and
1082: without constraints of strain (namely, the ``free'' film)
1083: (see Fig.~\ref{fig:epit-vs-free-hysteresis-box}).
1084: \begin{figure}
1085: \centering
1086: \includegraphics[width=88mm]{epit-vs-free-hysteresis-box.eps}
1087: \caption{Calculated hysteresis loops for capacitors with
1088: (a) epitaxially constrained films, and (b) ``free'' films
1089: of various thickness $l$ and with dead layer $d$.
1090: Supercell sizes were $16\times 16\times 2(l+d)$.
1091: $\Delta t=2$~fs and $n_{\rm steps}=$~50,000.
1092: For epitaxially constrained films,
1093: $\mathcal{E}_0 =$~4,000~kV/cm and $\Delta \mathcal{E} =$~100~kV/cm are employed.
1094: For ``free'' films,
1095: $\mathcal{E}_0 =$~400~kV/cm and $\Delta \mathcal{E} =$~10~kV/cm are employed.}
1096: \label{fig:epit-vs-free-hysteresis-box}
1097: \end{figure}
1098: \uline{The temperature is maintained at 100~K through the simulations.}
1099: For both the epitaxially constrained and ``free'' films, our simulations confirm
1100: that the imperfect screening of the electrodes decreases the coercive field
1101: as the film thicknesses decreases, as described
1102: phenomenologically in Ref.~[\onlinecite{Dawber:C:L:S:JPhys-CondesMatter:15:pL393-L398:2003}].
1103: There is a large (order-of-magnitude) difference in the
1104: coercive field $\mathcal{E}_{\rm c}$ between
1105: the epitaxially constrained film and the ``free'' film.
1106: This may be because the compressive strain arising from epitaxial constraints
1107: prevents the polarization switching, while the inclusion of
1108: inhomogeneous strain (i.e., acoustic displacements) eases the switching,
1109: as depicted in Fig.~\ref{fig:epit-vs-free}.
1110: The potential barriers themselves are lower in the ``free'' films than
1111: in the epitaxially constrained films (see Fig.~\ref{fig:potential:surface}).
1112: We note that hysteresis loops for ``free'' film capacitors with $l=63$ and $l=127$
1113: are very similar to the experimentally observed hysteresis loops
1114: of a ferroelectric capacitor with damaged electrodes
1115: that have ``steps'' and ``plateaus'' during polarization switchings.~\cite{Scott:Ferroelectric:Memories:2000}
1116: This is because,
1117: in the ``free'' film capacitors with imperfect electrodes ($d=1$),
1118: the configuration with out-of-plane polarization is no longer the ground state.
1119: In fact, the ground state has a nonzero in-plane polarization.
1120: Thus, the dipoles $Z^*\bm{u}(\bm{R})$
1121: have large in-plane components $Z^*u_x(\bm{R})$ and $Z^*u_y(\bm{R})$
1122: in the hysteresis-loop simulations (and experiments), as evident in the snapshot
1123: shown in Fig.~\ref{fig:damaged}.
1124: \begin{figure}
1125: \centering
1126: \includegraphics[width=55mm]{damaged.eps}
1127: \caption{Vertical cross section of a simulated ferroelectric ``free'' film capacitor
1128: with a single dead layer; $16 \times 16 \times (l=63,~d=1)$.
1129: The snapshot was taken at the point marked ``x'' in Fig.~\ref{fig:epit-vs-free-hysteresis-box}(b).
1130: The projection of the dipole moments onto the $xz$-plane are indicated with arrows.}
1131: \label{fig:damaged}
1132: \end{figure}
1133:
1134: \uline{Unfortunately, attempts
1135: to fit our results to the usually-assumed Kay-Dunn scaling of the coercive field $\mathcal{E}_{\rm c}$
1136: with film thickness $l$ of thicker %($l>15$~nm)
1137: films\cite{KayDunn1962} were unsuccessful, as were attempts
1138: to emulate the relatively weak dependence of $\mathcal{E}_{\rm c}$
1139: on $l$ for epitaxially grown high-quality ultrathin %($l<15$~nm)
1140: films.\cite{Kim:J:K:C:L:N:S:Y:C:B:K:J:APL:88:p072909:2006,
1141: Jo:K:N:Y:S:APL:89:p232909:2006,Petraru:P:K:P:W:S:K:JAP:101:p114106:2007}
1142: The experimentally observed values of coercive fields $\mathcal{E}_{\rm c}$ for ultrathin BaTiO$_3$ capacitors
1143: range from 200 to 500~kV/cm,\cite{Kim:J:K:C:L:N:S:Y:C:B:K:J:APL:88:p072909:2006,
1144: Jo:K:N:Y:S:APL:89:p232909:2006,Petraru:P:K:P:W:S:K:JAP:101:p114106:2007}
1145: while simulations of epitaxially constrained films largely overestimate $\mathcal{E}_{\rm c}$,
1146: and those of ``free'' films slightly underestimate $\mathcal{E}_{\rm c}$.
1147: This may be because the switching in real thin-film capacitors is a
1148: large-scale ($>$~100~nm) phenomenon involving defect-mediated
1149: nucleation mainly at ferroelectrics-electrodes
1150: interfaces,\cite{KayDunn1962,Janovec1958,
1151: Chandra:coercive:field:Ferroelectrics:313:p7-13:2004,
1152: Kim:J:K:C:L:Y:S:N:PRL:95:p237602:2005,
1153: Tagantsev:G:JAP:100:p051607:2006,
1154: Jo:K:K:C:S:Y:N:PRL:97:p247602:2006}
1155: as well as the possibility that the strain conditions may be intermediate
1156: between the cases of epitaxially constrained and ``free'' films.}
1157: \uline{Such intermediate strain conditions may be achieved and will be simulated with MD in the future
1158: by introducing a mechanical boundary condition such as presented in Ref.~[\onlinecite{Pertsev:K:PRL:98:p257603:2007}].}
1159: \uline{In contrast to our case of ultrathin BaTiO$_3$ capacitors,
1160: it is well known that for ultrathin PbZr$_x$T$_{1-x}$O$_3$ (PZT) capacitors
1161: the coercive fields $\mathcal{E}_{\rm c}$ increase with decreasing
1162: film thickness $l$, and there is an argument whether
1163: this strong increase of $\mathcal{E}_{\rm c}$ is coming from
1164: compressive substrate-induced lattice strain\cite{Pertsev:C:K:H:K:W:APL:83:p3356-3358:2003} or
1165: not.\cite{Lee:N:C:C:R:V:PRL:98:p217602:2007}
1166: Constructing a first-principles Hamiltonian for PZT and simulations with this MD method
1167: will help us to understand this difference between BaTiO$_3$ and PZT.}
1168: %
1169: % \tnm{David, I cite your paper (Ref.~[\onlinecite{Lee:N:C:C:R:V:PRL:98:p217602:2007}]) above. My understanding is OK?}
1170: % \dvm{I think so. The discussion about coercive fields in that paper came
1171: % from the experimental side, not the theoretical side, so I am not so
1172: % familiar with the story. But I think what you said is probably OK.}
1173:
1174: \section{Summary}
1175: \label{sec:summary}
1176: We have developed a robust and highly efficient molecular-dynamics scheme, based on a
1177: first-principles effective Hamiltonian formulation, for
1178: simulating the behavior of the polarization in perovskite-type ferroelectrics.
1179: We have applied this approach to study BaTiO$_3$ ferroelectric thin-film capacitors,
1180: with special attention to the dependence on film thickness and choice of electric boundary
1181: conditions. We find that striped domain structures tend to form on cooling-down simulations
1182: when a ferroelectric dead layer is present near the electrodes, and we study the dependence
1183: of the domain period on the conditions of formation. We also study the hysteresis loops
1184: for capacitor structures, both with and without such dead layers, and we find dramatic
1185: differences in the hysteretic behavior for the cases of elastically constrained or
1186: ``free'' films. Our MD simulator {\tt feram} will be a powerful tool for further
1187: investigations of the physical properties of ferroelectric nanostructures that are
1188: relevant for a variety of potential device applications.
1189:
1190:
1191: \section*{Acknowledgments}
1192: Computational resources
1193: were provided by the Center for Computational Materials Science,
1194: Institute for Materials Research (CCMS-IMR), Tohoku University.
1195: We thank the staff at CCMS-IMR for their constant effort.
1196: This research was done when T.N. stayed at JNCASR and Rutgers University
1197: under the support from
1198: JNCASR, Rutgers University,
1199: the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan,
1200: and the Japan Society for the Promotion of Science (JSPS).
1201: D.V. acknowledges support of ONR Grant N00014-05-1-0054.
1202:
1203: % Bibliography Start
1204: \bibliographystyle{apsrev}
1205: \bibliography{biblio/ferroelectrics,biblio/Bratkovsky-Levanyuk,biblio/FerroelectricCapacitor,biblio/MD}
1206: \end{document}
1207: