1: \documentstyle[aps,prb,epsfig,amssymb,floats]{revtex}
2: %\documentstyle[preprint,aps,prb]{revtex}
3:
4: \begin{document}
5: \draft
6:
7: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname
8: @twocolumnfalse\endcsname
9:
10: \title{Amorphous ZrO$_2$ from {\em Ab-initio} molecular dynamics:\\
11: Structural, electronic and dielectric properties}
12:
13: \author{Xinyuan Zhao \cite{pad}, Davide Ceresoli, and David Vanderbilt}
14: \address{Department of Physics and Astronomy, Rutgers University,
15: Piscataway, NJ 08854-8019}
16: \date{October 13, 2004}
17: \maketitle
18: \begin{abstract}
19: Realistic models of amorphous ZrO$_2$ are generated in a
20: ``melt-and-quench'' fashion using {\em ab-initio} molecular
21: dynamics in a plane-wave pseudopotential formulation of
22: density-functional theory. The structural properties of the
23: resulting amorphous models are analyzed, with special attention to
24: coordination statistics. The vibrational and dielectric properties
25: of one of these models are then investigated from first principles
26: using linear-response methods. The electronic dielectric constant
27: and Born effective charges are found to be very similar to those
28: of the crystalline phases. Encouragingly, the predicted total static
29: dielectric constant is $\sim$22, comparable to that of the monoclinic
30: phase. This work is motivated by the
31: search for improved gate dielectric materials for sub-0.1\,$\mu$m CMOS
32: technology, and may also have implications for HfO$_2$ and for
33: silicates of ZrO$_2$ and HfO$_2$.
34: \end{abstract}
35:
36: \pacs{PACS numbers: 77.22.-d, 61.43.Bn, 63.50.+x, 71.23.Cq}
37:
38: \vskip2pc]
39:
40: % \marginparwidth 3.1in
41: % \marginparsep 0.5in
42:
43: \columnseprule 0pt
44: \narrowtext
45:
46: %%========================================================================
47: \section{Introduction}
48: \label{sec:intro}
49: %%========================================================================
50:
51: The projections in the International Technology Roadmap for
52: Semiconductors call for an effective gate dielectric thickness of
53: approximately 1\,nm for 0.1\,$\mu$m CMOS technology by the year
54: 2006. However, the use of current SiO$_2$ and silica nitride
55: materials in this regime will become problematic because of
56: intolerably high leakage currents. A possible solution, currently
57: under intensive exploration, is to replace SiO$_2$ as the CMOS gate
58: dielectric by an oxide having a much higher dielectric constant.
59: Use of such a ``high-$K$ dielectric'' would allow one to make
60: physically thicker films, hence reducing the leakage current, while
61: at the same time maintaining or even increasing the gate
62: capacitance. High-$K$ metal oxides are at the focus of this
63: effort, with HfO$_2$, ZrO$_2$, and their mixtures with SiO$_2$ showing
64: great promise.\cite{wilk,gusev}
65:
66: In previous work, we studied the structural, electronic, and
67: lattice dielectric propterties of crystalline phases of ZrO$_2$ and
68: HfO$_2$ (Refs.~\onlinecite{zro2,hfo2,ortho,zrhf}) using a
69: first-principles density-functional approach. We found that the
70: lattice dielectric constant depends strongly on crystal phase; for
71: example, for ZrO$_2$, we find orientationally-averaged total
72: $\epsilon$ values of 37 and 38 for cubic and tetragonal phases, and
73: smaller values of 20, 20, and 19 for the monoclinic and for two
74: orthorhombic phases, respectively. We also found strong
75: anisotropies, most notably a large in-plane susceptibility for the
76: tetragonal phase. Differences in the electronic density of states
77: and in band gaps were also studied,\cite{ortho,zrhf} but these are
78: not profound enough to have much effect on the purely electronic
79: dielectric constant, which remains approximately 5 for all crystal
80: phases. These results suggest that novel structural modifications
81: of ZrO$_2$ or HfO$_2$ might possibly provide a route to enhanced
82: dielectric constants.
83:
84: One of the great advantages of using SiO$_2$ for CMOS technology
85: has been the fact that it forms an {\it amorphous} oxide, thus
86: allowing it to conform to the substrate with enough freedom to
87: eliminate most electrical defects at the interface. Materials
88: like ZrO$_2$ and HfO$_2$ tend to crystallize much more readily than
89: SiO$_2$, raising questions about whether equally smooth and clean
90: interfaces can be formed from such materials. While it is possible to
91: form such oxides in an amorphous state by low-temperature processes,
92: they tend to recrystallize during the thermal treatments needed for
93: later stages of device processing. One current
94: avenue of investigation is to consider crystalline oxides, such as
95: perovskites, that can be grown epitaxially on Si.\cite{mckee}
96: Another is to search for ways to raise the recrystallization
97: temperature of ZrO$_2$- or HfO$_2$-based materials, e.g., by
98: incorporating Si, Al, or N into the random network structure.
99: However, very little theoretical guidance is available for such
100: initiatives, since almost nothing is known about the structure and
101: atomic-scale properties even for pure amorphous ZrO$_2$ or HfO$_2$.
102:
103: With these motivations, we have embarked on a theoretical study of
104: the structure and properties of amorphous ZrO$_2$ ($a$-ZrO$_2$). In
105: this paper, we report the construction of realistic models of
106: $a$-ZrO$_2$ through ``melt-and-quench'' {\em ab-initio}
107: molecular-dynamics (MD) simulations. This approach, first
108: pioneered by Car and Parrinello,\cite{cpmd} combines
109: density-functional theory (DFT) with MD into a powerful tool for
110: investigating the physics of large systems, especially for liquids
111: and amorphous structures where the atomic coordinates cannot be
112: obtained from diffraction experiments. Using the
113: generated models as prototypes for $a$-ZrO$_2$, we investigate the
114: atomistic structure, especially the bonding and coordination statistics,
115: and then focus in detail on the
116: lattice dielectric properties. While the calculations have been
117: carried out only for $a$-ZrO$_2$, it can be expected that many
118: conclusions would hold also for $a$-HfO$_2$ in view of the close
119: chemical and crystallographic similarity of these two materials.
120:
121: The paper is organized as follows. Section \ref{sec:thy} describes the
122: theoretical approach of
123: our first-principles simulations, the supercells used for modeling $a$-ZrO$_2$,
124: and the linear-response method used to analyze the lattice dielectric
125: properties. In Sec.~\ref{sec:md} we present
126: and discuss the structural properties of the amorphous models obtained
127: from the MD simulations, with special attention to bonding and coordination.
128: The results of the linear-response calculations, such as the Born effective
129: charges, phonon modes, and dielectric tensors, are then presented and
130: discussed in Sec.~\ref{sec:diel}. Section \ref{sec:summary} concludes the paper.
131:
132: %%===================================================================
133: \section{Theory}
134: \label{sec:thy}
135: %%===================================================================
136:
137: %%-----------------------------------------------------
138: \subsection{Details of static structural calculations}
139: \label{sec:thy_struct}
140: %%-----------------------------------------------------
141:
142: In this work, we perform {\em ab-initio} constant-temperature MD
143: as implemented in the VASP simulation package, \cite{vasp} in which
144: the electronic structure is described within DFT \cite{hk64,ks65}
145: in the local-density approximation (LDA), \cite{lda} using a
146: plane-wave basis and ultrasoft pseudopotentials. \cite{uspp} At
147: each molecular-dynamics step, the instantaneous Kohn-Sham electronic
148: wavefunctions are obtained using the RMM-DIIS (residual
149: minimization with directive inversion in the iterative subspace)
150: method,\cite{pulay80,wood85} which is found to be
151: particularly efficient for diagonalizing the Kohn-Sham Hamiltonian for
152: large systems such as the 96-atom supercell considered here. The
153: forces at each MD step are calculated as the derivatives of
154: the generalized free energy with respect to the ionic
155: positions based on the Hellmann-Feynman theorem.\cite{hf-thm} The
156: calculated forces are then used to integrate Newton's equations of motion
157: for the ionic degrees of freedom via the Verlet algorithm.\cite{verlet}
158:
159: %%
160: \begin{table}
161: \caption{Structural parameters obtained for the three ZrO$_2$
162: phases from the present theory, compared with a previous theoretical
163: calculation of higher accuracy (see text). Lattice parameters $a$, $b$,
164: and $c$ in ${\rm \AA}$; volume $V$ in ${\rm \AA}^{3}$; monoclinic angle
165: $\beta$ in degrees. }
166: \begin{center}
167: \begin{tabular}{ccdd}
168: Phase & Parameter & This Work & Ref.~\protect\onlinecite{zro2}
169: \\ \hline
170: monoclinic & $a$ & 5.098 & 5.108 \\
171: & $b$ & 5.171 & 5.169 \\
172: & $c$ & 5.264 & 5.271 \\
173: & $\beta$ & 99.49 & 99.21 \\
174: & $V$ & 136.77 & 137.40 \\ \hline
175: tetragonal & $a$ & 5.037 & 5.029 \\
176: & $c$ & 5.113 & 5.101 \\
177: & $d_z$ & 0.041 & 0.042 \\
178: & $V$ & 129.73 & 129.03 \\ \hline
179: cubic & $a$ & 5.034 & 5.037 \\
180: & $V$ & 127.57 & 127.80 \\
181: \end{tabular}
182: \end{center}
183: \label{table:md_uspp}
184: \end{table}
185: %%
186:
187: Because MD simulations on large supercells are quite computer-intensive,
188: we have taken some care to make the calculations as efficient as
189: possible. In particular, we have reduced the plane-wave cutoff energy
190: to 15 Ry, and the pseudopotential for Zr includes only the outermost
191: shells (4$d$, 5$s$) in the valence. To test the adequacy of these
192: approximations, we tabulate in
193: Table~\ref{table:md_uspp} the lattice parameters
194: calculated with VASP using the current settings and those determined
195: in Ref.~\onlinecite{zro2}, where the cutoff was 25 Ry and the
196: 4$s$ and 4$p$ Zr levels were included in the valence.
197: It can be readily seen that they agree
198: very well. In Fig.~\ref{fig:md_ev}, we show the calculated ground-state
199: energies of the three ZrO$_2$ phases (monoclinic, tetragonal, and cubic).
200: The energy ordering (monoclinic lowest, cubic highest) is qualitatively
201: and semiquantitatively correct. We thus conclude that the chosen
202: pseudopotentials and plane-wave cut-off are adequate to correctly
203: reproduce the energetics of the three ZrO$_2$ phases.
204:
205: %%----------------------------------
206: \subsection{Supercell structures}
207: \label{sec:thy_cell}
208: %%----------------------------------
209:
210: %%
211: \begin{figure}[t]
212: \begin{center}
213: \epsfig{file=fig1.ps,width=2.8in}
214: \end{center}
215: \caption{Relative energies per formula unit versus lattice constant
216: $a$ (see Table \protect\ref{table:md_uspp}) for the monoclinic ($\Box$),
217: tetragonal ($\bigtriangleup$), and cubic ($\circ$) phases of ZrO$_2$.}
218: \label{fig:md_ev}
219: \end{figure}
220: %%
221:
222: In our simulation, $a$-ZrO$_2$ is modeled to have periodic boundary
223: conditions with a cubic supercell containing 96 atoms. This is the
224: same content as a 2$\times$2$\times$2 repetition of the monoclinic
225: unit cell, since each monoclinic cell contains 4 chemical formula
226: units (4 Zr atoms and 8 oxygen atoms). In fact, the cubic lattice
227: vectors are only slightly distorted from the 2$\times$2$\times$2
228: monoclinic ones ($\beta=99.21^{\circ}$ and the lengths of three
229: lattice vectors differ only within $3\%$).\cite{zro2} Thus,
230: we use as a starting point a structure in which the lattice vectors
231: are perfectly cubic but the fractional coordinates are those of the
232: relaxed monoclinic structure. (This monoclinic structure remains
233: stable in this environment; that is, when it is relaxed under fixed
234: cubic lattice vectors, only small further displacements occur and
235: the symmetry of the atomic coordinates remains monoclinic.) All
236: subsequent MD simulations are done using fixed cubic lattice
237: vectors for the supercell, and using single $k$-point sampling at the
238: $\Gamma$ point of the Brillouin zone.
239:
240: %%-------------------------------------------------
241: \subsection{{\em Ab-initio} molecular dynamics}
242: \label{sec:thy_md}
243: %%--------------------------------------------------
244:
245: The ``melt-and-quench'' simulation scheme proceeds by carrying out a
246: series of MD simulations, each at a fixed temperature
247: (i.e., in the canonical ensemble).
248: Temperature $T$ is sequentially increased, step by step, from
249: room temperature ($T_m$) to some hypothetical temperature ($T_{\rm max}$)
250: high enough to liquefy the system. At each step, the structure obtained
251: from the previous MD simulation is used as the starting point for the
252: next one.
253: The process is then reversed to quench the system back to $T_m$.
254: For each MD simulation, the system is given enough
255: time to reach its thermal equilibrium in order to eliminate its correlation
256: to the previous structure. In the case of $a$-ZrO$_2$,
257: we let the system evolve at temperatures ranging from zero
258: to 4000\,K, well above the experimental melting point of 2980\,K.
259: The system is then rapidly quenched back to 300\,K, followed by a
260: relaxation to zero temperature of both atomic coordinates and lattice
261: vectors.
262:
263: A canonical ensemble is realized using a Nos\'{e}
264: thermostat, which modifies the Newtonian MD
265: by introducing an additional degree of freedom such that
266: the total energy of the physical system is allowed to fluctuate.
267: \cite{nose,hoover} In the algorithm of Nos\'{e}, the extra degree of
268: freedom --- the Nos\'{e} mass $Q$ --- controls the frequency of the
269: temperature fluctuations. In principle, the Nos\'{e} mass should be chosen
270: so that the induced temperature fluctuations show approximately the
271: same frequencies as the ``typical'' phonon frequencies of the considered
272: system. \cite{vguide}
273: We know from Ref.~\onlinecite{zro2} that the phonon frequencies at the zone
274: center in monoclinic ZrO$_2$ are in the range of 100\,--\,748\,cm$^{-1}$,
275: corresponding approximately to 40\,--\,300\,fs in period.
276: Our experience shows that $Q$ is rather insensitive to the
277: designated temperature. In the case of ZrO$_2$, $Q$ =
278: 0.15\,--\,0.20 has been found to be suitable.
279: For $Q$ in this range, the characteristic period of the temperature
280: fluctuations is approximately 40\,fs, which corresponds the high end of the
281: phonon spectrum for monoclinic ZrO$_2$.
282:
283: It has to be emphasized that VASP currently supports only constant-volume
284: MD simulations, i.e., the volume of a unit cell is fixed during a
285: ``melt-and-quench'' simulation. In order to study the influence of volume on
286: the resultant structure, we carry out a series of similar MD simulations
287: on supercells of different
288: volumes, as will be discussed in details in Sec. \ref{sec:md_details}.
289:
290: %%-------------------------------------
291: \subsection{Details of the linear-response calculations}
292: \label{sec:thy_rf}
293: %%-------------------------------------
294:
295: The dielectric properties of the MD-simulated amorphous models of ZrO$_2$
296: are calculated by the specialized linear-response techniques as implemented
297: in ABINIT, \cite{abinit} a simulation package that is also based
298: on DFT with pseudopotentials and plane-wave expansion. ABINIT
299: has the capability to calculate response functions such as phonon
300: dynamical matrices, dielectric tensors, and Born effective charge
301: tensors via a variational approach to density-functional perturbation
302: theory \cite{dfpt1,dfpt2} in which phonon displacements and
303: static homogeneous electric fields are treated as perturbations.
304: Our linear-response calculations are performed within the
305: LDA \cite{lda} using the Perdew-Wang's parameterization. \cite{pw92}
306: The Brillouin zone is sampled only at the $\Gamma$ point.
307: Extended norm-conserving pseudopotentials \cite{teter} with
308: valence configurations of Zr(4$s$,4$p$,4$d$,5$s$) and O(2$s$,2$p$),
309: and a cutoff energy of 35.0 Hartree, are found to provide satisfactory
310: convergence.
311:
312: %%====================================================================
313: \section{Amorphous Models of Z\lowercase{r}O$_2$}
314: \label{sec:md}
315: %%====================================================================
316:
317: %%----------------------------------------------
318: \subsection{Generation of amorphous structures}
319: \label{sec:md_details}
320: %%----------------------------------------------
321:
322: %%
323: \begin{figure}
324: \begin{center}
325: \epsfig{file=fig2.ps,width=3.2in}
326: \end{center}
327: \caption{Programmed application of temperature vs.~time
328: in the ``melt-and-quench'' simulation. Temperature is increased from 300 to
329: 4000 K in steps of 1.5 ps, then decreased in a reversed sequence, for
330: a total simulation time of 22.5\,ps.}
331: \label{fig:mq}
332: \end{figure}
333: %%
334:
335: The ``melt-and-quench'' MD simulations are schematically illustrated
336: in Fig. \ref{fig:mq}. At each selected temperature, the ZrO$_2$ system
337: is allowed to run for 1.5\,ps to reach its thermodynamical equilibrium.
338: The time step for the simulations is set to be 3 fs
339: for $T=$ 300\,--\,2500\,K, and is reduced to 2 fs
340: when $T$ is in the range of 3000\,K to 4000\,K in order to
341: suppress numerical error at higher temperature. The total
342: simulation time is 22.5\,ps (see Fig. \ref{fig:mq}).
343:
344: As mentioned in Sec.~\ref{sec:thy_md}, several supercells with
345: different volume are used in our simulation. The prototypical
346: supercell, designated as the $p$-cell, is adjusted to have its volume
347: equal to $8\,V_{\rm mono}$, where $V_{\rm mono}$ is the volume of a unit
348: cell of monoclinic ZrO$_2$. The side length of
349: the $p$-cell is thus determined as 10.320\,a.u.. The other cells,
350: designated as the 2-, 4-,
351: 7-, and $11$-cell, have their side lengths increased by 2\%, 4\%, 7\%,
352: and 11\% over that of the $p$-cell respectively. This information is
353: summarized in Table~\ref{table:md_cells}.
354:
355: %%
356: \begin{table}
357: \caption{Supercells used in the simulations. The side length
358: $a$ is in atomic units (a.u.). The percentage in the second column
359: indicates the increment of $a$ from the p-cell.}
360: \begin{tabular}{ccc}
361: Supercell Type & Increment (\%) & Side Length ($a$) \\
362: \tableline
363: $p$-cell & 0 & 10.320 \\
364: 2-cell & 2 & 10.527 \\
365: 4-cell & 4 & 10.727 \\
366: 7-cell & 7 & 11.043 \\
367: 11-cell & 11 & 11.467 \\
368: \end{tabular}
369: \label{table:md_cells}
370: \end{table}
371: %%
372:
373: %%
374: \begin{table}[b]
375: \caption{Summary of structures resulting from the candidate
376: supercells.
377: $V_i$ is the volume at which the MD simulation is carried out;
378: $V_f$ is the volume after the final structural relaxation.
379: Fourth column shows the coordination number (CN),
380: or range of CNs, for Zr and O respectively. Last column
381: characterizes the resulting structure.}
382: \begin{center}
383: \begin{tabular}{rcccc}
384: & $V_i$ (\AA$^3$) & $V_f$ (\AA$^3$) & CN (Zr/O) & Descrip. \\ \hline
385: $p$-cell & 1099.2 & 1041.7 & 8 / 4 & Crystal \\
386: 2-cell & 1166.5 & 1091.9 & \phantom{0--}7 / 3, 4 & Crystal \\
387: 4-cell & 1234.4 & 1040.8 & 8 / 4 & Crystal \\
388: 7-cell & 1346.1 & 1230.8 & 5--8 / 2--5 & Amorph. \\
389: 11-cell & 1507.9 & 1329.6 & -- & Collapsed \\
390: \end{tabular}
391: \end{center}
392: \label{table:md_volume}
393: \end{table}
394: %%
395:
396: The ``melt-and-quench'' simulations are performed on these five candidate
397: supercells. The results are summarized in Table~\ref{table:md_volume},
398: where $V_i$ is the initial volume of the supercell that was kept fixed
399: during the MD simulation, and $V_f$ denotes the volume resulting from
400: the final zero-temperature structural relaxation of coordinates and
401: lattice vectors at the end of the simulation.
402: The coordination number of each atom is determined by counting the
403: number of atoms within a cutoff radius of 3.00\,\AA, which can be
404: compared with the range of Zr--O bond lengths, 2.035 -- 2.233\,\AA,
405: that characterizes monoclinic ZrO$_2$.\cite{zro2}
406: %
407: It can be readily seen that the volume has a significant effect on
408: the final structure. For the 11-cell, we find that the
409: atomistic structure collapses into smaller clusters accompanied with
410: ``spatial voids'' even at room temperature ($T_m$), suggesting that
411: the supercell volume is too large to sustain the atomic structure.
412: For the $p$-, 2- and 4-cell, during the ``melt-and-quench'' simulation
413: the coordination numbers suggest some degree of disorder, but
414: the final structures recrystallize when the systems are quenched back and
415: relaxed to the ground state.
416:
417: Fortunately, a reasonable amorphous model of ZrO$_2$ (called ``Model
418: I'') is realized from the 7-cell. Although it has an initial volume of
419: 1346.1\,\AA$^3$, a subsequent relaxation to zero temperature
420: reduces the volume to 1230.8\,\AA$^3$, which is approximately
421: 12\% bigger than $V_i$ of the $p$-cell. Using Model I as the starting
422: structure, we perform a second ``melt-and-quench'' simulation and
423: obtain another model of $a$-ZrO$_2$, which is called ``Model II''
424: hereafter. In the remainder of this section, we will mainly focus on
425: analyzing the structural properties of two models.
426:
427: %%--------------------------------------------
428: \subsection{Analysis of amorphous structures}
429: \label{sec:md_models}
430: %%--------------------------------------------
431:
432: %%----------------------------------
433: \subsubsection{Amorphous Model I}
434: \label{sec:md_model1}
435: %%----------------------------------
436:
437: %%
438: \begin{figure}
439: \begin{center}
440: \epsfig{file=fig3.ps,width=3.1in}
441: \end{center}
442: \caption{Distribution of coordination numbers during the
443: ``melt-and-quench'' simulation giving rise to Model I. ``Melt'' and
444: ``quench'' processes are shown at left and right, respectively.
445: Simulation temperature is indicated in each panel; Zr and O atoms
446: are indicated by solid and dashed bars, respectively.}
447: \label{fig:md_7mc_coor}
448: \end{figure}
449: %%
450:
451: %%
452: \begin{figure}
453: \begin{center}
454: \epsfig{file=fig4.ps,width=3.0in}
455: \end{center}
456: \caption{Structure of amorphous Model I. Zr atoms are black; O atoms
457: are white. Coordination number is indicated on each atom (for atoms
458: near front or back of view, some neighbors may not appear).}
459: \label{fig:md_q7mc}
460: \end{figure}
461: %%
462:
463: Figure \ref{fig:md_7mc_coor} tracks the atomic coordination number
464: (CN) of the Zr and O atoms during the formation of Model I, starting
465: from the 7-cell in which oxygen atoms have CN=3 or 4 and Zr
466: atoms have CN=7. Interestingly, the system retreats at $T$=1500\,K to a
467: higher-symmetry structure with CN=3 and 6 for O and Zr atoms
468: respectively. A hint of disorder is displayed at $T$=3000\,K via the
469: incidence of the 5-coordinated Zr atoms and the 2-coordinated O
470: atoms. When $T$ is increased to 4000\,K,
471: the distribution of coordination numbers suggests that the system
472: has become strongly disordered, which we take as a sign that it has
473: melted. The system is then quenched quickly (about 12\,ps, see
474: Fig.~\ref{fig:mq}) back to $T_m$. From the variation
475: of the coordination numbers, one can readily see that the amorphous character
476: has survived to $T_m$. A snapshot of this amorphous structure (Model I)
477: is shown in Fig.~\ref{fig:md_q7mc}, from which we can directly see
478: that a reasonable amorphous structure appears to have been generated.
479: The numbers
480: on the atoms in Fig.~\ref{fig:md_q7mc} indicate the coordination numbers.
481:
482: %%
483: \begin{figure}
484: \begin{center}
485: \epsfig{file=fig5.ps,width=3.1in}
486: \end{center}
487: \caption{Bond length distributions in Model I. Each panel shows
488: the number of bonds (frequency) {\em vs.}\ the bond length (\AA) for
489: a particular species (``O'' or ``Zr'') with a particular
490: coordination number (e.g., ``O$\,$4'' indicates 4-coordinated oxygen).}
491: \label{fig:md_bondstat_q7mc}
492: \end{figure}
493: %%
494:
495: The bond distributions of Zr and O atoms in Model I are plotted in
496: Fig.~\ref{fig:md_bondstat_q7mc}. While there are 5- and
497: 8-coordinated Zr atoms in Model I, it is obvious that the 6- and
498: 7-coordinated Zr atoms predominate. We can see from
499: Fig.~\ref{fig:md_bondstat_q7mc} that most of the oxygen atoms have
500: a CN of 3 or 4, except for a few 2- or 5-coordinated oxygen atoms.
501:
502: %%
503: \begin{table}
504: \caption{Bond (Zr-O) statistics in the amorphous models I and
505: II. Subscripts of Zr and O denote coordination numbers;
506: $N$ is the number of atoms having the specified coordination number.
507: The minimum ($L_{\rm min}$ ), maximum ($L_{\rm max}$), and
508: average ($\overline{L}$) bond length for each type of atom
509: is given in \AA.}
510: \begin{center}
511: \begin{tabular}{ccrccccrccc}
512: & &\multicolumn{4}{c}{Model I} & &\multicolumn{4}{c}{Model II} \\
513: & & $N$ & $L_{\rm min}$ & $L_{\rm max}$ & $\overline{L}$ & & $N$ &
514: $L_{\rm min}$ & $L_{\rm max}$ & $\overline{L}$ \\ \hline
515: Zr$_5$ & & 2 & 1.90 & 2.12 & 2.04 & & - & - & - & - \\
516: Zr$_6$ & & 12 & 1.96 & 2.32 & 2.09 & & 10 & 1.92 & 2.66 & 2.11 \\
517: Zr$_7$ & & 16 & 1.97 & 2.96 & 2.17 & & 18 & 1.95 & 2.98 & 2.16 \\
518: Zr$_8$ & & 2 & 2.05 & 2.82 & 2.25 & & 4 & 2.02 & 2.94 & 2.24 \\ \hline
519: O$_2$ & & 2 & 1.94 & 2.02 & 1.98 & & 3 & 1.92 & 2.05 & 1.98 \\
520: O$_3$ & & 43 & 1.91 & 2.33 & 2.09 & & 32 & 1.92 & 2.95 & 2.10 \\
521: O$_4$ & & 18 & 2.01 & 2.92 & 2.23 & & 29 & 1.99 & 2.98 & 2.22 \\
522: O$_5$ & & 1 & 2.18 & 2.96 & 2.47 & & - & - & - & -\\
523: \end{tabular}
524: \end{center}
525: \label{table:md_bondlength}
526: \end{table}
527: %%
528:
529: The number of atoms having each CN, together with the minimum
530: ($L_{\rm min}$), maximum ($L_{\rm max}$), and average
531: ($\overline{L}$) bond length for each CN, are reported in
532: Table~\ref{table:md_bondlength} for both Model I and Model II. The
533: bond lengths are found to be in the range of 2.04 -- 2.25\,\AA\ for
534: $\overline{L}$, generally similar to the range of 2.035 --
535: 2.233\,\AA\ that we found previously for monoclinic
536: ZrO$_2$.\cite{zro2} We find that the average bond length
537: $\overline{L}$ decreases monotonically as the coordination number
538: increases, consistent with the usual expectation, based on chemical
539: principles, of an inverse correlation between CN and bond
540: strength. The minima ($L_{\rm min}$) and maxima ($L_{\rm max}$) of
541: the bond lengths manifest similar behavior.
542:
543: %%------------------------------------
544: \subsubsection{Amorphous Model II}
545: \label{sec:md_model2}
546: %%------------------------------------
547:
548: As mentioned earlier, a second amorphous
549: structure, Model II, is obtained using Model I as the starting
550: input structure. Figures \ref{fig:md_rel7_coor} and
551: \ref{fig:md_bondstat_rel7} present results for Model II in a
552: parallel fashion as for Figs.~\ref{fig:md_7mc_coor} and
553: \ref{fig:md_bondstat_q7mc} for Model I. A significant change in
554: going from Model I to Model II is the lack of the 5-coordinated Zr
555: and O atoms (Zr$_5$ and O$_5$ in Table~\ref{table:md_bondlength}).
556: Interestingly, in Model II, the numbers of O$_3$ and O$_4$ atoms
557: have become nearly equal, while in Model I there was a
558: substantial difference ($N$=43 for O$_3$, 18 for O$_4$). The
559: relative populations of Zr$_6$ and Zr$_7$ do not change very much
560: from Model I to Model II, although a slight shift from Zr$_6$ to
561: Zr$_7$ has occurred. The average lengths of the bonds in the two
562: amorphous models are approximately equal. Finally, we point out
563: that the density of the system has increased by $\sim$6\% in going
564: from Model I to Model II.
565:
566: %%---------------------------
567: \subsection{Discussion}
568: \label{sec:md_discussion}
569: %%---------------------------
570:
571: %%
572: \begin{figure}
573: \begin{center}
574: \epsfig{file=fig6.ps,width=3.1in}
575: \end{center}
576: \caption{Distribution of coordination numbers during the
577: ``melt-and-quench'' simulation giving rise to Model II. ``Melt'' and
578: ``quench'' processes are shown at left and right, respectively.
579: Simulation temperature is indicated in each panel; Zr and O atoms
580: are indicated by solid and dashed bars, respectively.}
581: \label{fig:md_rel7_coor}
582: \end{figure}
583: %%
584:
585: We have presented, in this section, two models of amorphous ZrO$_2$
586: obtained via a ``melt-and-quench'' scheme using {\em ab-initio} MD.
587: Since the volume in our simulations is not allowed to vary in any single
588: MD run, certain candidate supercells of different volume have been constructed
589: in order to study the effects of volume on the final amorphous
590: structures (see Table~\ref{table:md_cells}). One candidate cell
591: (i.e. the 7-cell in Table~\ref{table:md_cells}) successfully leads
592: to an amorphous model of ZrO$_2$ (Model I). The density of the
593: 7-cell is about 20\% less than the density of monoclinic
594: ZrO$_2$ ($\rho_{\rm mono}$), but the final zero-temperature relaxation
595: of atomic coordinates and lattice vectors reduces the density somewhat,
596: such that the final Model I has a density $\sim$12\% smaller than
597: $\rho_{\rm mono}$. A second amorphous structure, Model II, is generated
598: using Model I as the starting structure. The final mass density
599: increases $\sim$6\% in going from Model I to Model II.
600:
601: Although the crystalline phases of ZrO$_2$ have been extensively investigated
602: both experimentally and theoretically (Ref.~\onlinecite{zro2} and
603: references therein), there has been very little work to characterize
604: the amorphous phase. Experimental efforts have mainly involved
605: studies of $a$-ZrO$_2$ in the form of powders \cite{tanaka90,landron94} or
606: thin films. \cite{winterer00}
607: As for a-ZrO$_2$ thin films, the resulting structure may be expected
608: to depend strongly on deposition and processing conditions and
609: on stoichiometric variations and impurities.
610: Moreover, while ZrO$_2$ films grown at high temperature are
611: typically polycrystalline, there are hints that ZrO$_2$ films grown at low
612: temperature may be a mixture of amorphous and polycrystalline
613: phases. One might also expect that the amorphous structure in thin films
614: could be quite different from the ``bulk'' amorphous state that our
615: model attempts to simulate.
616: Thus, a direct comparison of our results with experiments may
617: not presently be feasible. In particular, Winterer \cite{winterer00}
618: used a thin-film sample (presumably a-ZrO$_2$) with a density
619: of 4.2\,g/cm$^3$, approximately 71\% of the density of
620: monoclinic ZrO$_2$ ($\rho_{\rm mono}$=5.89 g/cm$^3$). The density
621: of his sample corresponds to that of the 11-cell in our simulation.
622: According to our calculations, the volume at this density is too large to
623: sustain a void-free atomistic structure. On the
624: contrary, the densities of two amorphous structures we simulated
625: --- Model I and II --- are about 88\% and 94\% of $\rho_{\rm mono}$.
626: Such a large difference in density would naturally result in important
627: differences in the local structure (e.g., coordination numbers) and
628: the overall physical properties.
629: In any case, in view of the potential importance of ZrO$_2$ and
630: HfO$_2$ in CMOS gate dielectrics, it can be hoped that more
631: experimental work in this area may soon emerge.
632:
633: %%
634: \begin{figure}
635: \begin{center}
636: \epsfig{file=fig7.ps,width=3.1in}
637: \end{center}
638: \caption{Bond length distributions in Model II. Each panel shows
639: the number of bonds (frequency) {\em vs.}\ the bond length (\AA) for
640: a particular species (``O'' or ``Zr'') with a particular
641: coordination number.}
642: \label{fig:md_bondstat_rel7}
643: \end{figure}
644: %%
645:
646: %%======================================================================
647: \section{Dielectric Properties of \lowercase{$a$}-Z\lowercase{r}O$_2$}
648: \label{sec:diel}
649: %%======================================================================
650:
651: %%-----------------------------
652: \subsection{Introduction}
653: \label{sec:diel_intro}
654: %%-----------------------------
655:
656: The dielectric properties of amorphous ZrO$_2$ are calculated using the
657: linear-response features of the ABINIT package. Since the structural
658: models generated in Sec.~\ref{sec:md_models} are
659: quite large and have essentially no symmetry,
660: the computation of their dielectric properties is quite
661: time-consuming. Therefore, we could afford to carry out a full study of
662: the dielectric properties of only one of these models, and we somewhat
663: arbitrarily chose to focus on Model I.
664:
665: Before computing the dielectric properties with ABINIT, the ground-state
666: structure of Model I obtained in VASP is first relaxed again using ABINIT
667: for consistency. The lattice constant increases slightly from
668: 10.717\,{\rm \AA} to 10.917\,{\rm \AA}, corresponding to a 5.6\% increase
669: in volume with respect to the ground-state volume of 1230.76\,${\rm \AA}^3$
670: obtained in VASP, while the atomic coordinates change very little
671: (less than 1\%). We use the newly relaxed structural
672: coordinates in our subsequent linear-response calculations.
673:
674: %%-----------------------------
675: \subsection{Results}
676: \label{sec:diel_results}
677: %%----------------------------
678:
679: The purely electronic dielectric tensor is computed to be
680: %%
681: \def\phm{\phantom{-}}
682: \[\epsilon_{\infty} =
683: \pmatrix{ \phm4.76 & -0.03 & \phm0.03 \cr
684: -0.03 & \phm4.62 & \phm0.00 \cr
685: \phm0.03 & \phm0.00 & \phm4.54 \cr}
686: . \]
687: %
688: Clearly this tensor is approximately isotropic and diagonal,
689: as expected for any large supercell containing an amorphous material.
690: We obtain an orientationally-averaged dielectric constant of
691: $\epsilon_\infty$=4.6.
692: This is only slightly smaller than the values obtained previously
693: for the various crystalline phases of ZrO$_2$
694: ($\sim\,$5.3--5.7),\cite{zro2,ortho,detraux98,rignanese01}
695: confirming our previous conclusion that $\epsilon_\infty$ is
696: fairly insensitive to the structural phase.
697:
698: In order to obtain the lattice contribution to the dielectric tensor,
699: we first need to compute the phonon mode frequencies and effective
700: charges. Beginning with the mode frequencies, we present in
701: Fig.~\ref{fig:diel_dos}(a) a histogram plot of the phonon DOS.
702: (To be more precise, this is the DOS of modes at the Brillouin zone
703: center of the supercell; this accounts for the absence of very
704: low-frequency modes in the plot.)
705: As can be seen in Fig.~\ref{fig:diel_dos}(a), the phonon modes extend up
706: to about 850 cm$^{-1}$, and aggregate roughly into three groups with
707: frequencies in the range 70 -- 300, 300 -- 510, and 510 -- 850 cm$^{-1}$.
708: The overall DOS spectrum does not show the kind of discrete features
709: expected for crystalline solids.
710:
711: Next, we use the computed mode effective charges
712: ${\widetilde{Z}}^{*}_\lambda$ and frequencies ${\omega}_{\lambda}$
713: of zone-center modes $\lambda$ to construct the ``infrared activity''
714: shown in Fig.~\ref{fig:diel_dos}(b) and defined as the phonon DOS weighted by
715: $\widetilde{Z}_{\lambda}^{*\,2} / {\omega}_{\lambda}^{2}$. (That
716: is, the frequency integral of this function gives the lattice
717: contribution to the isotropically averaged dielectric constant. Here
718: the scalar mode effective charge ${\widetilde{Z}}^{*}_{\lambda \alpha}$,
719: defined via ${\widetilde{Z}}^{*\,2}_{\lambda} =
720: \sum_\alpha {\widetilde{Z}}^{*\,2}_{\lambda \alpha}$,
721: is essentially the projection
722: of the atomic $Z^*$ tensors onto the dynamical matrix eigenvectors as
723: described in Eqs.~(6-9) of Ref.~\onlinecite{zro2}.)
724: We find that the largest contribution to the lattice dielectric
725: response comes from modes of low frequency, $\sim$100-250$\,$cm$^{-1}$.
726:
727: %%
728: \begin{figure}
729: \begin{center}
730: \epsfig{file=fig8.ps,width=3.1in}
731: \end{center}
732: \caption{(a) Histogram of phonon density of states (DOS) vs.~frequency.
733: (Height of each bar is the number of phonon modes in the
734: bin divided by the bin width of 10\,cm$^{-1}$.)
735: (b) Histogram of DOS weighted by $\widetilde{Z}_{\lambda}^{*\,2} /
736: {\omega}_{\lambda}^{2}$.}
737: \label{fig:diel_dos}
738: \end{figure}
739: %%
740:
741: From these ingredients, we calculate the total lattice contribution
742: to the dielectric tensor, which is
743: %%
744: \[\epsilon_{\rm latt} =
745: \pmatrix{ \;20.3 & -1.2 & \;\;\;0.5 \cr
746: -1.2 & \;17.1 & -0.4 \cr
747: \;\;\;0.5 & -0.4 & \;15.5 \cr }
748: . \]
749: %%
750: Combining with the purely electronic dielectric tensor, the total
751: dielectric tensor is then
752: %%
753: \[\epsilon_{0} =
754: \pmatrix{ \;25.1 & -1.2 & \;\;\;0.5 \cr
755: -1.2 & \;21.7 & -0.4 \cr
756: \;\;\;0.5 & -0.4 & \;20.0 \cr }
757: . \]
758: %%
759: Once again, we find that these tensors are roughly diagonal and
760: isotropic, which tends to confirm that our structural model is
761: indeed amorphous. Taking orientational averages, we can
762: summarize our results by observing that the purely electronic
763: dielectric constant $\epsilon_\infty$$\simeq$4.6 is augmented by a lattice
764: contribution $\epsilon_{\rm latt}$$\simeq$17.6 for a total dielectric
765: constant $\epsilon_0$$\simeq$22. For the crystalline phases, we found
766: that $\epsilon_0$ and $\epsilon_{\rm latt}$ (in contrast to
767: $\epsilon_\infty$) are strong functions of crystal structure, with
768: $\epsilon_{\rm latt}$ ranging from about 14 to 33, or
769: $\epsilon_0$ ranging from about 19 to 39.\cite{zro2,hfo2,ortho,zrhf}
770: Here, we find that the dielectric constant of $a$-ZrO$_2$
771: ($\epsilon_0$=22) is similar to that calculated previously for the
772: monoclinic ($\epsilon_0$=20) and two orthorhombic phases
773: ($\epsilon_0$=20 and 19).\cite{zro2,ortho}
774:
775:
776: %%-------------------------------------------
777: \subsection{Decomposition by atom type}
778: \label{sec:diel_decomp}
779: %%-------------------------------------------
780:
781: Clearly it is desirable to understand more fully the various
782: contributions to the lattice dielectric response of the amorphous
783: form of ZrO$_2$. To this end, we now decompose various lattice
784: properties by ``atom type'' (that is, by chemical species and
785: coordination number) in the hope that such an analysis may provide
786: further insight into our numerical results.
787:
788: We begin by decomposing the total phonon DOS $g(\omega)$ in
789: Fig.~\ref{fig:diel_dos}(a) into a local DOS on each type of atom
790: corresponding to the rows of Table \ref{table:md_bondlength}.
791: The total DOS
792: $g(\omega) = \sum_{\lambda} \delta(\omega_{\lambda} - \omega)$
793: can be decomposed as $g(\omega)=\sum_\tau g_\tau(\omega)$ where
794: the local DOS for atoms of type $\tau$ is
795: %%
796: \begin{equation}
797: g_\tau(\omega) = \sum_{j \in \tau} \sum_{\lambda\alpha}
798: |e_{j \alpha}^{\lambda}|^2 \delta(\omega_{\lambda} - \omega).
799: \end{equation}
800: %%
801: Here $\omega_{\lambda}$ is the eigenfrequency of phonon mode $\lambda$,
802: $\tau$ runs over atom types (O2--O5, Zr5--Zr8),
803: and $e_{j \alpha}^{\lambda}$ is the component of the eigenvector of
804: phonon mode $\lambda$ for atom $j$ along Cartesian direction
805: $\alpha$. The results are plotted in Fig.~\ref{fig:ldos}.
806:
807: %%
808: \begin{figure}
809: \begin{center}
810: \epsfig{file=fig9.ps,width=3.3in}
811: \end{center}
812: \caption{Site-projected phonon density of states $g(\omega)$ vs.~phonon
813: frequency for different atoms types (`O2' indicates 2-fold
814: coordinated oxygen, etc.) in amorphous Model I.}
815: \label{fig:ldos}
816: \end{figure}
817: %%
818:
819: We were hopeful that such a decomposition might help us understand
820: whether the modes associated
821: with certain atom types are systematically much softer or harder.
822: Instead, no clear trends emerge from Fig.~\ref{fig:ldos}.
823: (The different appearance of the spectra for O5 and Zr8 most
824: probably results from the lack of statistics for these atoms,
825: which occur in the supercell only once and twice, respectively.)
826: A slight reweighting of the spectra towards softer mode frequencies
827: for increasing atomic coordination numbers can be observed, but
828: this is not a very strong effect.
829:
830: %%
831: \begin{figure}
832: \begin{center}
833: \epsfig{file=fig10.ps,width=2.6in}
834: \end{center}
835: \caption{(a) Scatterplot of isotropically-averaged atomic $Z^*$
836: values (vertical axis) vs.~atom type and coordination number
837: (horizontal axis) for amorphous Model I. Circles and diamonds
838: denote O and Zr atoms, respectively. (b) Same but with `dielectric
839: activity' (see Eq.(\protect\ref{eq:activ})) plotted vertically.}
840: \label{fig:cn_diel_zstar}
841: \end{figure}
842: %%
843:
844: Next we test how the atomic Born effective charges correlate with
845: atom type. Recall that the Born effective charge tensor
846: {\bf Z}$_{i,\alpha \beta}^*$ quantifies the macroscopic polarization
847: along Cartesian direction $\alpha$ induced by a displacement of
848: sublattice $i$ along $\beta$. Due to the amorphous nature of our model,
849: the calculated $Z^*$ tensors have no symmetry, although we find that
850: for most atoms the diagonal elements are dominant. To reduce the
851: large quantity of $Z^*$ data to manageable proportions, we present in
852: Fig.~\ref{fig:cn_diel_zstar}(a) a scatterplot of just the isotropic
853: averages (i.e., one third of the trace) of the atomic $Z^*$ tensors
854: sorted by atom type for all 96 atoms.
855:
856: We find that the $Z^*$ values are fairly tightly clustered about their
857: average values of 5.0 and $-$2.5 for Zr and O, respectively.
858: These are just slightly smaller than the typical values of 5.4 (Zr)
859: and $-$2.7 (O) computed for the crystalline phases,\cite{zro2,hfo2}
860: but still clearly larger than the nominal chemical valences of 4 and
861: $-$2, indicating that significant partial covalent character survives
862: in the amorphous phase.
863:
864: To investigate the roles of the various atom types further,
865: we have carried out an analysis
866: in which we also make an atomic decomposition of the lattice dielectric
867: response.\cite{zro2} To do this, we first decompose the lattice dielectric
868: into contributions from pairs of atoms $i$ and $j$,
869: %
870: \begin{equation}
871: \epsilon_{\alpha\beta}^{\rm latt} = \sum_{ij}
872: \tilde{\epsilon}^{\,ij}_{\alpha \beta},
873: \end{equation}
874: %
875: where $\alpha$ and $\beta$ label the Cartesian directions. Here
876: %
877: \begin{equation}
878: \tilde{\epsilon}^{\,ij}_{\alpha \beta} =
879: \frac{4 \pi e^{2}}{V} \sum_{\lambda}
880: \frac{1}{\kappa_{\lambda}}R^{\lambda}_{\alpha i} R^{\lambda}_{\beta j},
881: \end{equation}
882: %
883: where $\kappa_{\lambda}$ and $e^{\lambda}_{j \beta}$ are the eigenvalue
884: and eigenvector of the force constant matrix $\Phi_{ij}^{\alpha \beta}$
885: for the phonon mode $\lambda$, $V$ is the volume of unit cell, and
886: $R^{\lambda}_{\alpha j} = \sum_{\beta} Z^{*}_{j, \alpha \beta}\,
887: e^{\lambda}_{j \beta}$. We then heuristically define the contribution
888: coming from atom $i$ to be
889: %%
890: \begin{equation}
891: \bar{\epsilon}^{\,(i)}_{\alpha \beta}=\sum_{j} \frac{1}{2}
892: \left( \tilde{\epsilon}^{\,ij}_{\alpha \beta} +
893: \tilde{\epsilon}^{\,ji}_{\alpha \beta} \right) \;.
894: \end{equation}
895: %%
896: Finally, we will refer to the trace
897: %
898: \begin{equation}
899: \bar{\epsilon}_i=\sum_\alpha \bar{\epsilon}^{\,(i)}_{\alpha\alpha}
900: \label{eq:activ}
901: \end{equation}
902: %
903: as the ``dielectric activity" of atom $i$.
904:
905: This quantity is plotted versus coordination number in the scatterplot of
906: Fig.~\ref{fig:cn_diel_zstar}(b). The results indicate that a large
907: number of atoms contribute, but that a notable subpopulation of
908: mostly 6-fold Zr atoms are particularly strong contributors.
909:
910: %%------------------------
911: \subsection{Discussion}
912: \label{sec:discuss}
913: %%------------------------
914:
915: The dielectric constant ($\epsilon_0 = 22$) of our
916: amorphous model is much larger than that of SiO$_2$ ($\epsilon_0 =
917: 3.5$) and quite comparable to the average dielectric constant of the
918: monoclinic phase. Thus, we find that from the point of view of this
919: one criterion alone, $a$-ZrO$_2$ is indeed a promising high-$K$
920: dielectric for next-generation gate dielectrics. Of course,
921: many other issues need to be addressed, not the least of which is
922: the stability (e.g., as measured by the recrystallization temperature)
923: of the amorphous phase.
924:
925: Direct comparison with experiment is difficult, both because few
926: experimental measurements on amorphous ZrO$_2$ are available, and
927: because the sample preparation procedures may vary and may result
928: in rather different amorphous samples.\cite{devine01,liu04} We
929: compare here with the
930: recent work of Koltunski and Devine, \cite{devine01} focusing
931: mainly on the sample deposited at room temperature using no rf bias
932: (for which spectra are shown in Figs.~1(c) and 2(c) of
933: Ref.~\onlinecite{devine01}) since the authors judged this sample to
934: be their ``most amorphous'' one. (Upon annealing, the samples
935: tended to recrystallize partially to monoclinic and tetragonal
936: phases.) These authors reported a dielectric constant $\epsilon_0$
937: (at 100\,kHz) of $\sim$\,15 -- 18 as measured electrically on the
938: as-deposited ZrO$_2$ thin films. The optical refractive index
939: measured at 632.8\,nm was in the range $n$ = 1.83 -- 1.85. For this
940: most amorphous sample, the absorbance spectrum, obtained using a
941: Nicolet Fourier-transform infrared spectrometer, shows some TO
942: modes beyond 1000\,cm$^{-1}$. However, since our calculation does
943: not show any phonon modes in this range, we suspect these may be
944: defect-related features. The sample shows broad TO and LO
945: features centered at frequencies identified as 410\,cm$^{-1}$ and
946: 693\,cm$^{-1}$, respectively.\cite{devine01} (The authors also
947: pointed out that use of the Lyddane-Sachs-Teller relation would yield
948: an estimate of $\epsilon_{0} \simeq 10$ for the amorphous phase,
949: but this method substantially underestimates $\epsilon_{0}$
950: because the IR measurements did not extend to very
951: low frequency.) While our value of 22 certainly exceeds the
952: experimentally measured one of 15--18, it is roughly the same order
953: of magnitude, and better agreement might not be expected in view
954: of the fact that the nanostructure and stoichiometry of the
955: experimental film are imperfectly characterized.
956:
957:
958: %%====================================================
959: \section{Summary}
960: \label{sec:summary}
961: %%====================================================
962:
963: We have generated two realistic models of amorphous ZrO$_2$ via a
964: ``melt-and-quench'' scheme using {\em ab-initio} MD simulations.
965: Candidate supercells of different volume were constructed in order
966: to study the effects of volume on the simulated structures. An
967: amorphous model (Model I) was obtained from one candidate structure
968: (i.e., the 7-cell), avoiding recrystallization or void formation
969: during the ``melt-and-quench'' sequence. The equilibrium density of this
970: model is $\sim$12\% less than that of the monoclinic phase. The
971: second amorphous structure (Model II) was generated from Model I
972: with further processing. We expect both models to be reasonable
973: representatives of $a$-ZrO$_2$. The structural properties and the
974: bond-length distributions of Model I and II were analyzed, and both
975: models were found to be composed mainly of 6- and 7-fold
976: coordinated Zr atoms and 3- and 4-fold coordinated O atoms.
977:
978: The Born effective charges, phonon mode vectors and frequencies,
979: and electronic and lattice dielectric tensors were then calculated
980: for Model I using linear-response methods. The phonon DOS was
981: found to be relatively featureless, and the infrared activity
982: spectrum (i.e., the DOS weighted by $\widetilde{Z}_{\lambda}^{*\,2}
983: / {\omega}_{\lambda}^{2}$) showed a broad peak in the range of
984: 100--250$\,$cm$^{-1}$ in
985: the phonon spectrum. The calculations show that the Born
986: effective charges of the Zr and O atoms are fairly narrowly distributed
987: about 5.0 and $-$2.5, respectively, rather similar to what was
988: found previously for the crystalline phases.
989: Both the electronic and lattice contributions
990: to the dielectric tensor were found to be fairly isotropic,
991: as expected for an amorphous structure. The calculated electronic
992: contribution was about 4.6, slightly less than for the
993: crystalline phases. On the other hand, the lattice contribution to
994: $\epsilon_{0}$ was calculated to be 17.6, rather similar to the
995: average value of monoclinic ZrO$_2$, though still significantly
996: smaller than for the cubic and tetragonal phases.
997:
998: It is difficult to know just how similar our ``theoretical {\it
999: a}-ZrO$_2$'' is to ``experimental {\it a}-ZrO$_2$''. There are
1000: difficulties on both sides of the comparison.
1001: %
1002: On the theoretical side, several approximations have been
1003: introduced. First, the modest size of our model supercell
1004: (2$\times$2$\times$2) is a serious approximation, even though it is
1005: essentially at the limit of current computational
1006: capabilities. However, the nearly isotropic form of the lattice
1007: ($\epsilon_{\rm latt}$) and electronic ($\epsilon_{\infty}$)
1008: dielectric tensors tends to confirm that we have lost the memory of
1009: the initial crystalline structure and obtained an amorphous one, in
1010: spite of the supercell size limitation. Second, recall that the
1011: 4$s$ and 4$p$ electrons were included in the valence in the Zr
1012: pseudopotential for the MD simulations, but not for the
1013: linear-response calculations. This additional approximation could
1014: affect the accuracy of the computed dielectric properties of our
1015: amorphous structure. However, we believe this approximation is not
1016: very significant since the structural calculations on the
1017: monoclinic phase of ZrO$_2$, which is rather ``disordered'' itself,
1018: give nearly identical results using the two Zr pseudopotentials.
1019: %
1020: Experimentally, it is possible that the amorphous state manifests
1021: itself differently in bulk and thin-film forms. Indeed, because
1022: ZrO$_2$ is essentially a poor glass-former, it can only be prepared
1023: in amorphous form by low-temperature deposition or in other
1024: non-equilibrium conditions, and it is therefore to be
1025: expected that the properties of such samples may vary strongly with
1026: preparation conditions. Motivated by applications, much work is
1027: now under way to explore whether chemical substitution (e.g., by
1028: addition of Si, Al, or N) may help stabilize the amorphous phase
1029: and raise its recrystallization temperature, and ultimately it will
1030: be important to extend the theory to such substitutions in the
1031: future. Nevertheless, we believe that the present computational
1032: investigation can serve as an important
1033: first step in answering some of the many open questions about the
1034: structural and chemical properties of this class of materials.
1035:
1036:
1037: \section*{Acknowledgments}
1038:
1039: This work was supported by NSF Grant DMR-0233925. X.Z. thanks Jeff
1040: Neaton for help with the VASP software package.
1041:
1042: \begin{references}
1043:
1044: \bibitem[*]{pad} Present address: School of Physics, Georgia Institute of
1045: Technology, Atlanta, GA 30332.
1046:
1047: \bibitem{wilk} G. D. Wilk, R. M. Wallace, and J. M. Anthony, J. Appl.
1048: Phys. {\bf 89}, 5243 (2001).
1049:
1050: \bibitem{gusev} E. Cartier, D. A. Buchanan, M. Gribelyuk, M. Copel,
1051: H. Okorn-Schmidt, and C. D'Emic, Microelectron. Eng. {\bf 59}, 341 (2001).
1052:
1053: \bibitem{zro2} X. Zhao and D. Vanderbilt, Phys. Rev. B {\bf 65},
1054: 075105 (2002).
1055:
1056: \bibitem{hfo2} X. Zhao and D. Vanderbilt, Phys. Rev. B. {\bf 65},
1057: 233106 (2002).
1058:
1059: \bibitem{ortho} X. Zhao and D. Vanderbilt, {\sl Novel Materials and
1060: Processes for Advanced CMOS,} edited by M.I. Gardner, J.-P. Maria, S. Stemmer,
1061: S. De Gendt, Proceeding of the 2002 MRS Fall Meeting, Volume 745, p. N7.2.1.
1062:
1063: \bibitem{zrhf} X. Zhao, S. Sayan, E. Garfunkel, and D. Vanderbilt (unpublished).
1064:
1065: \bibitem{mckee} R.A.~McKee, F.J.~Walker, and M.F.~Chisholm,
1066: Phys.\ Rev.\ Lett.\ {\bf 81}, 3014 (1998).
1067:
1068: \bibitem{cpmd} R. Car and M. Parrinello, Phys. Rev. Lett. {\bf 55},
1069: 2471 (1985).
1070:
1071: \bibitem{vasp} G. Kresse and J. Hafner, Phys. Rev. B {\bf 47}, RC558
1072: (1993); G. Kresse and J. Furthm\"uler, {\bf 54}, 1169 (1996).
1073:
1074: \bibitem{hk64} P. Hohenberg and W. Kohn, Phys. Rev. {\bf 136}, B864 (1964).
1075:
1076: \bibitem{ks65} W. Kohn and L. J. Sham, Phys. Rev. {\bf 140}, A1133 (1965).
1077: \bibitem{lda} D. M. Ceperley and B. J. Alder, Phys. Rev. Lett. {\bf 45},
1078: 566 (1980).
1079:
1080: \bibitem{uspp} D. Vanderbilt, Phys. Rev. B {\bf 41}, 7892 (1990).
1081:
1082: \bibitem{pulay80} P. Pulay, Chem. Phys. Lett. {\bf 73}, 393 (1980).
1083:
1084: \bibitem{wood85} D. M. Wood and A. Zunger, J. Phys. A {\bf 18}, 1343
1085: (1985).
1086:
1087: \bibitem{hf-thm} R. P. Feynman, Phys. Rev. {\bf 56}, 340 (1939).
1088:
1089: \bibitem{verlet} D. Levesque and L. Verlet, J. Stat. Phys. {\bf 72},
1090: 519 (1993).
1091:
1092: \bibitem{nose} S. Nos\'{e}, Molec. Phys. {\bf 52}, 255 (1984);
1093: Prog. Theor. Phys. Suppl. {\bf 103}, 1 (1991).
1094:
1095: \bibitem{hoover} W. G. Hoover, Phys. Rev. A {\bf 31}, 1695 (1985).
1096:
1097: \bibitem{vguide} G. Kresse and J. Furthm\"uler, {\em VASP the
1098: Guide} (March 23, 2001).
1099:
1100: \bibitem{abinit} X. Gonze, {\em et al}, Comp. Mat. Sci. {\bf 25},
1101: 478 (2002); also see http://www.abinit.org.
1102:
1103: \bibitem{dfpt1} X. Gonze, Phys. Rev. B {\bf 55}, 10337 (1997).
1104:
1105: \bibitem{dfpt2} X. Gonze and C. Lee, Phys. Rev. B {\bf 55}, 10355 (1997).
1106:
1107: \bibitem{pw92} J. P. Perdew and Y. Wang, Phys. Rev. B {\bf 45},
1108: 13244 (1992).
1109:
1110: \bibitem{teter} M. Teter, Phys. Rev. B. {\bf 48}, 5031 (1993).
1111:
1112: \bibitem{tanaka90} T. Tanaka, T. M. Salama, T. Yamguchi, and
1113: K. Tanabe, J. Chem. Soc., Faraday Trans. {\bf 86}, 467 (1990).
1114:
1115: \bibitem{landron94} C. Landron, A. Douy, and D. Bazin, Phys. Status
1116: Solidi B {\bf 184}, 299 (1994).
1117:
1118: \bibitem{winterer00} M. Winterer, J. Appl. Phys. {\bf 88}, 5635
1119: (2000).
1120:
1121: \bibitem{detraux98} F. Detraux, Ph. Ghosez, and X. Gonze, Phys. Rev.
1122: Lett. {\bf 81}, 3297 (1998).
1123:
1124: \bibitem{rignanese01} G.-M. Rignanese, F. Detraux, X. Gonze, and A. Pasqua\-rello,
1125: Phys. Rev. B {\bf 64}, 134301 (2001).
1126:
1127: \bibitem{devine01} L. Koltunski and R. A. B. Devine, Appl. Phys. Lett.
1128: {\bf 79}, 320 (2001).
1129:
1130: \bibitem{liu04}
1131: J. Zhu and Z.G. Liu, Appl. Phys. A: Mater. Sci. and Proc.
1132: {\bf 78}, 741 (2004).
1133:
1134: \end{references}
1135:
1136: \end{document}
1137: