1: \documentstyle[prb,aps,eqsecnum,manuscript]{revtex}

2: %\documentclass[prb,aps,manuscript,showpacs]{revtex}

3: %\usepackage{epsfig}

4: %\usepackage{latexsym}

5: %\usepackage{amssymb}     

6: %\usepackage{graphicx}

7: 

8: \begin{document}

9: 

10: \title{Temperature effects on dislocation core energies \\ in silicon and 

11: germanium}

12:  

13: \author{Caetano R. Miranda}

14: 

15: \address{Instituto de F\'{\i}sica ``Gleb Wataghin'', 

16: Universidade Estadual de Campinas,\\

17: CP 6165, CEP 13083-970, Campinas, SP, Brazil}       

18: 

19: \author{R. W. Nunes }

20: 

21: \address{Departamento de F\'{\i}sica, Universidade Federal de Minas Gerais, \\

22: Belo Horizonte, Minas Gerais , CEP 30123-970, Brazil}       

23: 

24: \author{A. Antonelli}

25: 

26: \address{Instituto de F\'{\i}sica ``Gleb Wataghin'', 

27: Universidade Estadual de Campinas,\\

28: CP 6165, CEP 13083-970, Campinas, SP, Brazil} 

29: 

30: \date{\today}

31: \maketitle

32:    

33: \begin{abstract}

34: 

35: Temperature effects on the energetics of the 90$^\circ$ partial

36: dislocation in silicon and germanium are investigated, using

37: non-equilibrium methods to estimate free energies, coupled with Monte

38: Carlo simulations. Atomic interactions are described by Tersoff and

39: EDIP interatomic potentials. Our results indicate that the vibrational

40: entropy has the effect of increasing the difference in free energy

41: between the two possible reconstructions of the 90$^\circ$ partial,

42: namely, the single-period and the double-period geometries. This effect

43: further increases the energetic stability of the double-period

44: reconstruction at high temperatures. The results also indicate that

45: anharmonic effects may play an important role in determining the

46: structural properties of these defects in the high-temperature regime.

47: 

48: \end{abstract}

49: 

50: %\maketitle

51: 

52: \pacs{61.72.Lk, 61.72.Bb, 62.20.Fe}

53: 

54: 

55: 

56: %\begin{multicols}{2} 

57: 

58: %

59: \section{Introduction}              

60: %

61: 

62: The study of the atomic structure of dislocation cores in

63: semiconductors is not free from outstanding

64: issues.~\cite{hirsch,duesbery,alexan,bulatov} At low deformations,

65: dislocations in these systems are known to occur along $<$110$>$

66: directions on $\{111\}$ slip planes of the diamond (or zinc-blend)

67: lattice. Due to the presence of a two-atom basis, this lattice

68: supports two possible $\{111\}$ dislocation slip planes, the so-called

69: glide and shuffle sets.  In the glide set, the slip plane lies between

70: atomic layers separated by a third of the nearest-neighbor bond

71: length, and geometrically it would appear that dislocation glide

72: should require breaking three covalent bonds.  In the case of the

73: shuffle set, the slip plane lies between atomic layers separated by a

74: full bond length, and glide appears to involve breaking only a single

75: covalent bond. This would suggest that lattice friction should be

76: smaller in the shuffle set, and hence that dislocations would glide in

77: that configuration. However, the prevalent opinion is in favor of the

78: glide set, based primarily on the experimental evidence showing that

79: dislocations in Si are dissociated into Shockley partials bounding

80: stacking-fault ribbons, and on the fact that stacking faults are not

81: stable in the shuffle plane.~\cite{hirsch,duesbery,alexan,bulatov}

82: Nevertheless, this conclusion has been disputed by Louchet and

83: Thibault-Desseaux,~\cite{louchet} who argued that partials in the

84: shuffle position could occur in conjunction with glide-plane stacking

85: faults.  The mechanism controlling dislocation glide is also a matter

86: of debate. Many theoretical and experimental works have either

87: supported or assumed the Hirth and Lothe mechanism,~\cite{hl} based on kink

88: nucleation and migration, but an impurity-pinning

89: mechanism proposed by Celli et al.~\cite{celli} has been shown to be

90: consistent with recent experimental results.~\cite{kolar} This issue

91: is far from being resolved.

92: 

93: Another unresolved issue has recently arisen, regarding the core

94: structure of the 90$^\circ$ partial dislocation. In $\{111\}$ planes,

95: dislocations lining along $<$110$>$ directions have their Burgers

96: vectors either parallel (the screw orientation) or making a 60$^\circ$

97: angle with the dislocation line. Both dissociate into Shockley

98: partials, the former into two 30$^\circ$ partials, and the latter into

99: a 30$^\circ$ partial and a 90$^\circ$ partial.  Both partials are

100: believed to undergo core reconstruction, as indicated by the low

101: density of unpaired spins found in EPR

102: experiments.~\cite{hirsch,duesbery,alexan,bulatov}

103: 

104: In the glide set, the nature of the reconstruction of the 90$^\circ$

105: partial appears straightforward. The symmetrically-reconstructed

106: geometry shown in Fig.~\ref{cores}(a) displays mirror symmetries along the

107: $[110]$ direction. A symmetry-breaking reconstruction shown in

108: Fig.~\ref{cores}(b) lowers the core energy by $\sim$200 meV/\AA, according to

109: {\it ab initio} and tight-binding (TB)

110: calculations.~\cite{Bigger,nbv,hansen} This structure retains the

111: periodicity of the lattice along the dislocation direction, thus being

112: called the single-period (SP) structure. Until recently, it had been a

113: matter of consensus that the SP core was the ``ground-state''

114: structure of the 90$^\circ$ partial, since the earliest independent

115: works of Hirsch and Jones.~\cite{hirsch2,jones} However, the structure

116: shown in Fig.~\ref{cores}(c) has been recently proposed by Bennetto and

117: collaborators,~\cite{Nunes_2} and shown to have lower internal energy

118: than the SP core, by means of total-energy tight-binding (TETB), and

119: {\it ab initio} local-density-approximation (LDA) calculations. In

120: this new geometry, along with the mirror symmetry, the translational

121: symmetry of the lattice by a single period is also broken, and the

122: periodicity along the core is doubled. For that reason, this

123: reconstruction has been called the double-period (DP) structure.

124: 

125: Both the SP and DP geometries are consistent with all available

126: experimental information about the 90$^\circ$ partial. EPR

127: measurements in Si indicate a rather small density of dangling bonds

128: in the core of the dislocation.~\cite{hirsch,duesbery,alexan} The two

129: structures are fully reconstructed, meaning that neither would give

130: rise to deep-gap states which would show an EPR signal. Moreover, both

131: cores consist entirely of fivefold, sixfold, and sevenfold rings, both

132: being consistent with images produced by transmission electron

133: microscopy, at the current level of resolution of this

134: technique.~\cite{kolar} So far, the experiments appear unable to

135: decide clearly on the issue. Recent experimental work by

136: Batson~\cite{batson} indicates a DP-derived structure (called the

137: ``Extended DP structure'' by this author) to be more consistent with

138: STEM and EELS experiments.

139: 

140: At the theoretical level, several calculations have addressed the

141: energetics of the SP and DP cores in diamond (C), silicon (Si), and

142: germanium (Ge), using different

143: methods.~\cite{Nunes_2,Vall,nv,Blase,Nunes_1,Nunes_3,Lehto} Virtually

144: all of the more accurate {\it ab initio} and TETB

145: studies~\cite{Nunes_2,nv,Blase,Nunes_1,Nunes_3} agree that the DP core

146: is more stable at 0 K, with the exception of the {\it ab initio}

147: cluster calculations for Si in Ref.~\onlinecite{Lehto}. For example,

148: LDA and TETB calculations in Refs.~\onlinecite{Nunes_2} and

149: \onlinecite{Nunes_3} indicate that the internal energy of the DP

150: structure is lower than that of the SP core in all three materials (C,

151: Si, and Ge). A possible influence of supercell boundary conditions on

152: these results was raised in Ref.~\onlinecite{Lehto} (using

153: Keating-potential calculations) and later refuted by supercell-size

154: converged TETB calculations in Ref.~\onlinecite{nv}. The issue was

155: further revisited in Ref.~\onlinecite{Blase}, with {\it ab initio}

156: calculations showing that the DP core remains more stable in C even

157: under relatively severe stress regimes.  Ideally, supercell-size

158: converged calculations like the TETB ones in Ref.~\onlinecite{nv}

159: would properly address the issue for isolated dislocations in the

160: bulk. However, in that study very large supercells (containing up to

161: 1920 atoms) were employed, which makes such calculations prohibitive

162: for the more accurate {\it ab initio} techniques.

163: 

164: Closely related to the present study is the work of Valladares and

165: co-workers,~\cite{Vall} in which temperature effects on the energetics

166: of the SP and DP cores in Si were investigated using a Tersoff

167: potential,~\cite{Ters} within a harmonic approximation treatment of

168: vibrational entropic effects in the dislocation free energy, at finite

169: temperatures. They find that the free-energy difference between the

170: two cores, at a temperature of 800 K, becomes smaller than the

171: zero-temperature internal energy split, suggesting the possibility of

172: the two cores playing a role in the plasticity of Si, in the

173: temperature range of most plastic deformation experiments ($\sim$700-1000

174: K). However, the calculations in Ref.~\onlinecite{Vall} did not

175: include anharmonic effects, which are likely to be relevant in that

176: range of temperatures.

177: 

178: In this paper, we present the results of a study of anharmonic

179: vibrational effects in the free-energy difference between the SP and

180: DP cores in Si and Ge. As discussed in the following, we arrive at

181: different conclusions with respect to the work of Valladares {\it et

182: al.}, with our results indicating that the effect of increasing

183: temperature is to enhance the thermodynamical stability of the DP core

184: with respect to the SP geometry.  In our calculations, anharmonic

185: effects are fully taken into account by using a methodology that

186: allows to estimate the free energy through non-equilibrium

187: simulations.  The details of this methodology are discussed in

188: Sec. II. In Sec. III our results are presented and discussed. We end

189: with a summary and conclusions in Sec. IV.

190: 

191: 

192: %

193: \section{Methodology}              

194: 

195: \subsection{Interatomic Potentials}

196: \label{secA}

197: 

198: There are several empirical potentials available in the literature to

199: model the interactions between Si atoms.~\cite{Potential1} In the

200: present work, our aim is to account for vibrational entropic effects

201: on the energetics of the SP and DP reconstructions of the 90$^{\circ}$

202: partial dislocation in Si. One key ingredient for this purpose is that

203: both the SP and DP structures be a local minimum of the potential. It

204: is essential then that the potential be able to describe the

205: reconstruction-driven symmetry breaking that leads to the SP

206: geometry.~\cite{Bigger,nbv,hansen} This is an important aspect, since

207: both core structures consist of fully reconstructed bonds, and a

208: realistic description of vibrational properties requires a proper

209: treatment of bonding for the minima in question. For example,

210: potentials such as the Stillinger-Weber(SW) and the Kaxiras-Pandey(KP)

211: fail in that regard.~\cite{Potential2} Both the Tersoff~\cite{Ters}

212: and the Environment Dependent Interatomic Potential

213: (EDIP)~\cite{edip1,edip2} models describe correctly the energetics of

214: the SP reconstruction with respect to a symmetric

215: (so-called quasi-fivefold) structure.~\cite{Bigger,edip2}

216: 

217: The focus of our paper is on the effects of the dislocation

218: vibrational modes on the free energies of the two core models at

219: finite temperatures.  In a previous work we have used these two

220: potentials to estimate the free energy of Si in several different

221: phases (amorphous, liquid, and crystalline, as well as clathrate

222: structures), obtaining results which are in good agreement with those

223: of experimental and other theoretical works.~\cite{Caetano} For a more

224: specific analysis of the description of anharmonic effects by the EDIP

225: and Tersoff potentials, we show in Fig.~\ref{bulk} the results we obtained for

226: the free energy as a function of temperature, in a 216-atom bulk

227: supercell.  The figure includes also the Tersoff-potential harmonic

228: approximation values from Ref.~\onlinecite{harm} and the experimental

229: results from Ref.~\onlinecite{fexp}. As expected, the experimental

230: numbers start deviating appreciably from the harmonic approximation

231: curve around the Debye temperature of Si ($\sim$640 K), where

232: anharmonic effects begin manifesting themselves. From the figure, we

233: see that both the Tersoff and the EDIP potentials can account for

234: these anharmonic effects in the free-energy curve at a quantitative

235: level.

236: 

237: Moreover, it is clear that the EDIP results are in better agreement

238: with the experimental ones. Of particular relevance to the present

239: work is the scale of the difference in free energy between

240: harmonic-approximation and experimental results, which is

241: $\sim$10~meV/atom at 640 K. As a first approximation, we may consider

242: ten atoms per period as constituting the core in the SP and DP

243: structures (those taking part in the fivefold and sevenfold rings,

244: seen on [110] projections, surrounding the geometric center of the

245: core), which gives an initial estimate of $\sim$30~meV/\AA\ for the

246: scale of the anharmonic effects, from the bulk results. From this, one

247: might expect that anharmonic effects should be important in the case

248: of the 90$^\circ$ partial, since the difference in enthalpy at 0 K

249: between the SP and DP cores is $\sim$60~meV/\AA.

250: 

251: On the basis of these results, we argue that both potentials give a

252: realistic description of vibrational effects, in particular of the

253: anharmonic contributions which become important in the range of

254: temperatures of our study. For the Ge case, only a Tersoff model was

255: employed, since there are no available parameters for Ge in the EDIP

256: form. The functional form and the parameters we use for the two models

257: are given in Refs.~\onlinecite{Ters} (Tersoff) and \onlinecite{edip2}

258: (EDIP).

259: 

260: \subsection{Simulations and Free Energy Calculations}

261: \label{secB}

262: 

263: In our statistical computation of dislocation free energies, we use a

264: Metropolis-algorithm Monte-Carlo method. Two types of simulations were

265: performed, one employed a canonical-ensemble (constant-NVT) treatment,

266: and the other simulated an isobaric-isothermal ensemble

267: (constant-NPT).~\cite{Frenkel} A simulation cell with 192 atoms was

268: used, containing a 90$^\circ$ partial-dislocation dipole which was

269: introduced by displacing the atoms according to the

270: continuous-elasticity solution for the displacement field of a

271: dislocation dipole.~\cite{hl} The introduction of a dipole enables the

272: use of periodic boundary conditions to avoid surface effects.  It

273: should be pointed out that the free energy calculations presented in

274: this work are far more demanding in terms of computer time than

275: harmonic-approximation calculations, which only require one

276: diagonalization of the dynamical matrix, in order to determine the

277: phonon frequencies. Therefore, the complexity of our calculations

278: imposes a limitation on the size of the cell used in the

279: simulations. The cell size chosen reflects the best compromise between

280: a good description of the essential physics and computational

281: feasibility.

282: 

283: The free energy calculations were performed using a non-equilibrium

284: method, namely the Reversible-Scaling method (RS).~\cite{RS_1} This

285: method consists of simulating a quasi-static process where the

286: potential-energy function of the system is dynamically scaled. It

287: allows for an accurate and efficient evaluation of the free energy

288: over a wide range of temperatures, using a single Monte Carlo

289: simulation, starting from a reference temperature for which the free

290: energy of the system is known. In order to obtain the free energy at

291: this reference temperature, we used the Adiabatic Switching method

292: (AS).~\cite{AS_1} Using the AS approach, the absolute free energy can

293: be computed by determining the work done to switch the potential

294: energy of the system from the configuration of interest to that of a

295: reference system for which the free energy is known analytically. The

296: systematic error in the free energy, caused by dissipation, can be

297: estimated by reversing the switching procedure and computing the

298: overall work performed in this round-trip procedure. The statistical

299: error can be estimated by performing several trajectories in the

300: configuration space of the system.

301: 

302: In this work, the reference system was an Einstein crystal, i.e., a

303: collection of independent harmonic oscillators. The angular frequency

304: of the oscillators ($\omega$) was taken to be approximately the

305: vibration modes of Si ($\omega = 28$~Trad/s) and Ge ($\omega = 15$~

306: Trad/s), which correspond to the acoustic phonon modes from the

307: edge of the Brillouin zone that give rise to a pronounced peak in the 

308: phonon density of states of these materials.~\cite{Kane,Nilsson}

309: It should be pointed out that the final value for the Gibbs free energy

310: is not sensitive to the choice of the oscillator frequency,

311: provided that the chosen frequency is not very far from the

312: relevant frequencies of the physical system.

313: Monte Carlo constant-NVT simulations were used to generate

314: switching trajectories, with a typical "switching time" of 3.0

315: $\times$ 10$^{5}$ MC/steps. The volume in these simulations was chosen

316: to be that obtained from a constant-NPT simulation for the given

317: reference temperature at zero pressure. In general, 25 switching

318: trajectories were used to estimate the statistical error. The

319: estimated error was $\pm$ 1.0 $\times$ 10$^{-3}$ eV/atom. RS

320: calculations were carried out at zero pressure, covering a range from

321: T=100 K to T=2000 K, using a scaling time of 5.0 $\times$ 10$^{5}$

322: MC/steps. Five different switching trajectories were sampled, in order

323: to estimate the statistical errors. In this way, we were able to

324: estimate the absolute Gibbs free energy (G$_{abs}$) at zero pressure

325: of the SP and the DP reconstructions in the interval from 300K to 1200

326: K using EDIP and Tersoff models for Si, and for Ge in the same range

327: of temperatures using a Tersoff potential. The absolute vibrational

328: entropy was calculated by taking the numerical derivative of the

329: absolute Gibbs free energy with respect to temperature: $S_{abs}=

330: -(\frac{\partial G_{abs}}{\partial T})$. 

331: 

332: 

333: \section{Results and Discussion}              

334: \label{enthal}

335: \subsection{Enthalpies at zero Kelvin}

336: 

337: Imposing periodic boundary conditions in a dislocation calculation

338: using supercells requires the net Burgers vector in the cell to

339: vanish. This is done by using cells which contain two dislocations

340: with opposite Burgers vectors. Hence, besides the core enthalpies that

341: interest us, supercell calculations also include elastic interactions

342: between the two dislocations, and between the dislocations and their

343: periodic images. In order to extract meaningful core energy values,

344: these elastic interactions must be properly accounted for. Previous

345: calculations have shown that dislocation cores in covalent

346: semiconductors are very narrow, with the elastic fields reaching the

347: linear behavior predicted by the continuum linear elasticity solution

348: already at a distance of about two lattice periods from the center of

349: the core.~\cite{hansen,Blase} This indicates that, to the extent that

350: one is interested in differences in energies between different core

351: models, long-range elastic effects should mostly cancel out for

352: sufficiently large cells. This is because these long-range fields are

353: defined by the dislocation slip system which is obviously the same in

354: our case, since we are comparing different core structures of the same

355: dislocation. We must be careful, however, in providing numerical

356: evidence that the supercells we use in our calculations are

357: sufficiently large, in the above sense.

358: 

359: In the particular case of the SP geometry, it is known that a

360: relatively strong dipolar elastic interaction is present, which

361: depends on the relative senses of reconstruction of the two

362: dislocations in the cell, with respect to the broken mirror

363: symmetry. Depending on the size of the cell, this interaction may be

364: of the same order of the core-energy difference between the SP and DP

365: geometries. (A similar effect is present in the DP core, but this is a

366: weaker quadrupolar interaction which can be neglected.~\cite{Nunes_2}) As

367: discussed in the literature,

368: ~\cite{Nunes_2,Vall,nv,Blase,Nunes_1,Nunes_3} this dipolar interaction

369: can be properly handled by considering the SP-core energy as the

370: average energy between two supercell calculations, one in which the

371: two SP dislocations are reconstructed in the same sense, and the other

372: where they have opposite reconstruction

373: senses.~\cite{nv,Nunes_1,Nunes_3} In order to further investigate the

374: accuracy of this procedure and also to address the numerical

375: convergence level of our calculations for the cell sizes we use, we

376: show in Table I the 0 K enthalpies computed with the Tersoff and EDIP

377: potentials for several different cell sizes. For comparison, we also

378: include the TETB results from Ref.~\onlinecite{nv}, to which we add

379: numbers for a 240-atom supercell. In the notation introduced in

380: Ref.~\onlinecite{Lehto}, the relevant supercell parameters are $D$,

381: the height of the cell in the $<111>$ direction (perpendicular to the

382: slip plane), and $L$ the width of the cell in the $<11\bar{2}>$

383: direction (perpendicular to the dislocation line) on the slip

384: plane. We consider here only supercells with ``dipolar'' boundary

385: conditions,~\cite{bc} which were shown by Lehto and \"Oberg~\cite{Lehto} to

386: minimize the strains associated with the stacking of the infinite

387: array of dislocations. We will comment below on the comparison of our

388: results with those from Ref.~\onlinecite{Vall}, which were obtained

389: with ``quadrupolar'' boundary conditions.~\cite{bc}

390: 

391: Table I shows our results for $E_{SP-DP} = \overline{E}_{SP} -

392: E_{DP}$, the energy difference between the average energy of the two

393: SP cells ($\overline{E}_{SP}$) and the energy of the DP cell, and for

394: $\Delta E_{SP} = E_{SP}^{same} - E_{SP}^{opposite}$, the energy

395: splitting between the two SP cells (as described in the previous

396: paragraph), for four different cell sizes. Note that $E_{SP-DP}$

397: converges for relatively small cell sizes, and faster than $\Delta

398: E_{SP}$, due to the cancellation of elastic effects that happens when

399: we take the average of the two SP calculations. For the free-energy

400: calculations we discuss in the next section, we used the smaller cell,

401: with $L = 13.3$~\AA\ and $D = 18.8$~\AA, for a total of 192 atoms. In

402: the 240 atom cell, $D$ is the same as in the 192-atom cell, and $L$ is

403: only 25\% larger.  We can see from Table I that, already for this

404: 240-atom cell, the TETB result for $E_{SP-DP}$ is converged to within

405: the numerical accuracy of the method. This is also what can be

406: observed for the EDIP potential. The EDIP result for $E_{SP-DP}$ in

407: the 192-atom cell is within 13\% of the EDIP converged value, and for

408: the 240-atom cell $E_{SP-DP}$ is converged to within 3\%.  The Tersoff

409: numbers converge more slowly, with the 192-atom cell numbers being

410: more than twice as large as the converged value.  In the case of Ge,

411: we only calculated $E_{SP-DP}$ using the 192-atom cell. We obtained

412: $E_{SP-DP}$~= -8~meV/\AA\ in disagreement with the LDA and

413: Keating-potential results from Ref.~\onlinecite{Nunes_1}. However, as

414: discussed in Sec.~\ref{free-ener}, the behavior of the free energy as

415: a function of temperature for Ge is the same as that for Si, i.e.,

416: $E_{SP-DP}$ increases as the temperature is raised.  If we were to

417: extrapolate to the free-energy calculations these convergence trends

418: observed for the 0 K enthalpy results in Table I, we would expect our

419: EDIP free-energy values to be converged to within $\sim$10\%, and the

420: Tersoff values to be about twice as large as what we would obtain with

421: larger cells.

422: 

423: The TETB calculations in Table I were done at the fixed cell volume

424: (corresponding to the experimental lattice constant). As a test, we

425: also performed calculations for volumes 1\% larger and smaller,

426: obtaining the same convergence trends, i.e., energies were essentially

427: converged for the 240-atom cell. More importantly, the convergence

428: calculations with the Tersoff and EDIP potentials in Table I were done

429: at constant pressure, resulting in very small changes from the initial

430: volumes.

431: 

432: Before proceeding to the discussion of the free-energy results, let us

433: put the results in Table I in perspective. Previous studies have shown

434: that the Tersoff potential does give the right ordering of 0 K

435: enthalpies, despite underestimating the difference in enthalpies

436: between the two cores. Our converged Tersoff values of 6~meV/\AA\ are

437: in very good agreement with the converged results obtained in

438: Ref.~\onlinecite{chrzan} using supercell and cylindrical cluster

439: calculations. However, they are smaller by one order of magnitude,

440: when compared with the converged TETB values. The EDIP potential has

441: the shortcoming of predicting the wrong ordering of 0 K enthalpies for

442: the SP and DP cores. When comparing our results with those obtained by

443: Valladares {\it et al.}, we must observe that they used cells with $D

444: = 9.4$~\AA, which is half the height of our cells, and quadrupole

445: boundary conditions. As a check, we did run Tersoff calculations with

446: the same cell parameters and obtained the same results as in their

447: work. The value for $E_{SP-DP}$ in this case is 32 meV/\AA, which is

448: more than five times as large as the converged Tersoff value in Table

449: I.  This shows that, at least for the 0 K enthalpy, their results are

450: not converged at the same level as our calculations.

451: 

452: \subsection{Free energies}

453: \label{free-ener}

454: 

455: In order to test whether the empirical potentials support the SP and

456: DP reconstructions at high temperatures, we performed constant-NPT

457: simulations of all structures in a broad range of temperatures, from 0

458: K up to 2000 K (1500 K) for Si (Ge). The structures were further

459: characterized for several temperatures (at every 500 K between 0 K and

460: the highest temperature) by evaluating structural properties such as

461: the pair-correlation function, the bond-angle distribution, and the

462: atomic-coordination number. Only the natural thermal broadening of

463: these functions was observed, with no structural changes and all atoms

464: remaining fourfold coordinated. From these results, we conclude that

465: for both the Tersoff and the EDIP potentials, the SP and DP

466: reconstructions remain stable and do not undergo restructuring over

467: the entire range of temperatures we tested for each of the two

468: semiconductors. 

469: 

470: The metastability of the SP core, even at temperatures as high as 2000

471: K, is not surprising. Unlike the spontaneous transformation of the

472: symmetric quasi-fivefold core into the SP core, which happens at 0 K

473: due to the absence of an energy barrier,~\cite{Bigger,nbv,hansen} the

474: SP $\rightarrow$ DP transformation involves breaking the strong

475: reconstructed covalent bonds in the SP core. The same bond breaking

476: mechanism regulates the kink migration barriers.~\cite{nbv} We have

477: computed an energy barrier of $\sim$1.5 eV for the process involving

478: the conversion of a segment of two SP periods into a segment of one DP

479: double period.

480:     

481: Let us now turn to the focus of this work, the Gibbs free-energy

482: difference, $\Delta G$, between the SP and the DP cores as a function

483: of temperature, obtained by the Reversible Scaling Method within the

484: Monte Carlo method (RS-MC). In the previous section we discussed the

485: numerical convergence of our calculations with respect to the two

486: dimensions of the supercell which are perpendicular to the dislocation

487: line. When we introduce thermal effects, periodicity along the line is

488: disrupted by atomic vibrations, and one must be careful about the

489: sampling of the phonon modes along the dislocation direction. In our

490: free-energy calculations, we used 192-atom supercells with twice the

491: lattice period along this direction. To test the phonon sampling in

492: this cell, we also calculated free energies for a supercell with four

493: times the lattice period along the line, using the Tersoff

494: potential. In Fig.~\ref{tamanho} we show the results for the

495: free-energy difference $\Delta G$ between the SP and DP cores, for the

496: two types of SP cells and also for the average between the two

497: cells. Note that, while the results for $\Delta G$ for the individual

498: SP calculations differ quite appreciably for a given cell size and

499: vary substantially when we go from the smaller (192 atoms and twice

500: the lattice period along the dislocation line) to the larger cell (384

501: atoms and four times the period along the line), the results for

502: $\Delta G$ between the average of the SP cells and the DP cell are

503: very similar for the two cell sizes. This shows that our simulations

504: with 192 atoms capture most of the relevant differences between the

505: vibrational modes of the two cores and also lends further confirmation

506: of the cancellation of elastic interaction effects that occurs when

507: the average between the two SP cells is considered, as discussed in

508: the previous section.

509: 

510: In Fig.~\ref{G-si} we show the results for $\Delta G_{SP-DP}$ for the

511: Tersoff and EDIP models in a 192-atom supercell, showing now only the

512: average of the SP calculations. Fig.~\ref{G-ge} shows the results for

513: Ge using the Tersoff potential. In order to analyze the entropic

514: effects, Figs.~\ref{G-si} and \ref{G-ge} also display the entropic

515: contribution ($-T \Delta S$) to $\Delta G$. The Tersoff results for Si

516: in Fig.~\ref{G-si} show $\Delta G$ increasing with temperature from

517: 19~meV/\AA\ at 300 K to 37~meV/\AA\ at 1200 K.  The entropic term also

518: increases with temperature, being the dominant contribution for

519: $\Delta G$ in the high temperature regime.  For Si using the EDIP

520: model, we observe the same behavior, with $\Delta G$ increasing with

521: temperature from -32~meV/\AA\ at 300 K to -23~meV/\AA\ at 1200 K,

522: notwithstanding the fact that for this potential the SP core remains

523: more stable over the entire temperature range. This is mostly due to

524: the fact that the EDIP potential does not describe properly the

525: energetics of the two reconstructions at 0 K, as discussed above. But

526: even for this potential, the behavior of the entropic term is

527: determinant of the temperature trend observed for $\Delta G$, which is

528: qualitatively the same trend observed with the Tersoff model. As

529: indicated by the convergence trends we discussed in the previous

530: section for the 0 K enthalpies, a quantitative comparison between the

531: changes in $\Delta G$ produced by the Tersoff and EDIP potentials

532: would suffer form the fact that, for the 192-atom cell, the Tersoff

533: results could be overestimated by as much as a factor of two. Note

534: that, for silicon, the overall change in $\Delta G$ over the 300 K -

535: 1200 K interval is 18~meV/\AA\ for Tersoff and 9~meV/\AA\ for

536: EDIP. While speculative, it seems reasonable to expect that the overall

537: change in $\Delta G$ for the two potentials, over the temperature

538: range of our calculations, would be in good agreement if we used cells

539: with larger values of $D$ and $L$.

540: 

541: Regarding the contribution of the entropic term depicted in

542: Figs.~\ref{G-si} and \ref{G-ge}, two points have to be

543: emphasized. First of all, the entropy is calculated by taking the

544: numerical derivative of the free energy, and the numerical data for

545: this quantity presents some statistical fluctuations. Most certainly,

546: these fluctuations are bound to be enhanced when one takes numerical

547: derivatives. This explains some of the oscillations observed in the

548: entropic term.  Thus the qualitative behavior of the total free-energy

549: is more representative than the behavior of the isolated entropic term

550: shown in the figures.  Another point has to do with the behavior of

551: the entropic term as a function of temperature, in particular for the

552: computations using the Tersoff potential. The free energy difference

553: $\Delta G$ is quite small, when compared with the total free energy of

554: each system, and the same is true for the entropic contribution to

555: this difference.  These energy differences will be affected by

556: anharmonic effects, and the discrepancy between the entropic term

557: computed using the two models (EDIP and Tersoff) above 600 K is

558: related to the way anharmonic effects are accounted for in each

559: model. It is clear from Fig.~\ref{bulk} that the results obtained

560: using the Tersoff potential tend to deviate more from the experimental

561: data than those obtained using the EDIP, and that the difference

562: between the two potentials increase with increasing temperature. The

563: more accentuated increase in the entropic term computed using the

564: Tersoff potential has to do with its tendency to overestimate the

565: anharmonic effects, and represents a small amount of energy that is

566: not related to any changes in the structure of the defect (which would

567: imply in much bigger changes in the free energy).

568: 

569: In the case of Ge, $\Delta G$ increases with temperature faster than

570: in Si (compared to either of the Si potentials used), from 2~meV/\AA\

571: at 300 K to 33~meV/\AA\ at 1200 K. While the enthalpy difference at 0

572: K marginally favors the SP reconstruction (at variance with the LDA

573: result in Ref.~\onlinecite{Nunes_3}), the behavior of the free energy

574: suggests that at room temperature the DP reconstruction would be more

575: stable than the SP structure. It is important to emphasize the

576: similarity between our results for Si and Ge, i.e., the free energy

577: difference $\Delta G$ increases with temperature in both cases (the

578: increase rate is larger in Ge than in Si), meaning that the entropic

579: term leads to an enhancement of the thermodynamical stability of the

580: DP structure with respect to the SP, as the temperature increases.

581: 

582: At this point it is interesting to compare the RS-MC results for Si

583: with the harmonic-approximation calculations of Valladares {\it et

584: al.}.~\cite{Vall} In Fig.~\ref{vallad}, we show our RS-MC calculations

585: using the Tersoff (full line) and EDIP (dashed line) models, and the

586: Tersoff harmonic-approximation results from Ref.~\onlinecite{Vall}

587: (dotted line). For a better comparison of the temperature trends in

588: each calculation, our values for $\Delta G$ for both potentials have

589: been shifted such that they extrapolate to the same zero-temperature

590: $\Delta G$ value obtained by Valladares {\it et al.}. The same

591: qualitative behavior is observed in Fig.~\ref{vallad} for the RS-MC

592: calculations using both the Tersoff and the EDIP models. In both

593: cases, $\Delta G$ {\it increases} with temperature, while the opposite

594: is observed for the harmonic-approximation calculations, which show

595: $\Delta G$ {\it decreasing} with increasing temperature.  This points

596: to an important role of anharmonic effects in describing thermal

597: effects on dislocation cores in semiconductors. 

598: 

599: It should also be pointed out that at low temperatures our

600: results cannot be directly compared to those in

601: Ref.~\onlinecite{Vall}, because their calculations take into account

602: quantum effects which are absent in our approach. While quantum

603: effects may be relevant at low temperatures, we are mostly concerned

604: with the anharmonic effects, which become more relevant in the

605: high-temperature regime. Valladares {\it et al.} point out that the

606: order of magnitude of the difference in free energies between the SP

607: and the DP reconstructions is smaller than the thermal energy

608: ($k_{B}T$) over the entire range of temperature in their study. This

609: would suggest that the two structures would be nearly equally

610: stable. Our results, however, indicate that at high temperatures the

611: free energy difference between the SP and the DP cores is of the same

612: order of the thermal energy. Therefore, considering that the thermal

613: fluctuations are smaller than the thermal energy itself, our results

614: suggest that the DP reconstruction should be dominant in the high

615: temperature regime. In other words, while thermal fluctuations may

616: account for the creation of local defects such as kinks, it is

617: unlikely that these fluctuations may cause substantial portions of the

618: dislocation to switch from the DP to the SP reconstruction (barring a

619: possible dependence of the core energies on the stress acting on the

620: dislocation,~\cite{Blase} an issue that has not yet been investigated

621: in Si and Ge).

622: 

623: %

624: \section{Conclusions}              

625: % 

626: In this work, we studied the differences in free energies and

627: vibrational entropies between the SP and DP core reconstructions of

628: the 90$^{\circ}$ partial dislocation as a function of temperature in

629: silicon and germanium, using a Reversible Scaling Method and Monte

630: Carlo simulations, with Tersoff and EDIP models for the

631: energetics. This methodology is a fully-classical simulation which

632: includes all anharmonic vibrational effects. Our results indicate that

633: the difference in the free energies (and in the vibrational-entropy

634: contribution to this quantity) between the two core structures

635: increases with temperature in both materials. In the case of Si, this

636: behavior occurs for both potentials used in the calculations. This is

637: in contrast with the free-energy calculations by Valladares and

638: co-workers~\cite{Vall} which incorporates vibrational-entropy effects

639: only at the harmonic-approximation level. Our calculations indicate

640: that the DP reconstruction, which is lower in energy at 0 K, becomes

641: even more stable with respect to the SP structure in the high

642: temperature regime. Moreover, our results also suggest that the

643: anharmonic effects may play an important role in the description of

644: the thermal behavior of extended defects in semiconductors.

645:   

646: \begin{acknowledgments} 

647: 

648: C. R. Miranda and A. Antonelli acknowledge the support from the 

649: Brazilian funding agencies: FAPESP, CNPq, and FAEP.

650: R. W. Nunes acknowledges support from the Brazilian agencies: CNPq and

651: FAPEMIG.

652: 

653: \end{acknowledgments}  

654: 

655: %\begin{thebibliography}{99}

656: \begin{references}

657: 

658: \bibitem{hirsch} P.~B.~Hirsch, Mater. Sci. Tech. {\bf 1}, 666 (1985).

659: 

660: \bibitem{duesbery} M.~S.~Duesbery and G.~Y.~Richardson,

661: Crit. Rev. Solid State Mater. Sci. {\bf 17}, 1 (1991).

662: 

663: \bibitem{alexan} H.~Alexander and H.~Teichler, in {\it Materials

664: Science and Technology}, edited by R.~W.~Cahn, P.~Hassen, and

665: E.~J.~Kramer (VCH Weiheim, Cambridge, 1993), Vol.~4, p.~249.

666: 

667: \bibitem{bulatov} V.~V.~Bulatov, J.~F.~Justo, W.~Cai, S.~Yip,

668: A.~S.~Argon, T.~Lenosky, M.~de~Koning, and T.~Diaz de la Rubia,

669: Philos. Mag. A {\bf 81}, 1257 (2001).

670: 

671: \bibitem{louchet} F.~Louchet and J.~Thibault-Desseaux,

672: Rev. Phys. Appl. {\bf 22}, 207 (1987).

673: 

674: \bibitem{hl} J. P. Hirth and J. Lothe, {\it Theory of Dislocations}

675: (Wiley, New York, 1982).

676: 

677: \bibitem{celli} V.~Celli, M.~Kabler, T.~Ninomiya, and R.~Thomson,

678: Phys. Rev. {\bf 131}, 58 (1963).

679:  

680: \bibitem{kolar} H.~R.~Kolar, J.~C.~H.~Spence, and H.~Alexander,

681: Phys. Rev. Lett. {\bf 77}, 4031 (1996).

682: 

683: \bibitem{Bigger} J. R. K. Bigger, D. A. McInnes, A. P. Sutton, M. C. Payne, I. 

684: Stich,

685: R. D. King-Smith, D. M. Bird, and L. J. Clarke,

686: Phys. Rev. Lett {\bf 69}, 2224 (1992).

687: 

688: \bibitem{nbv} R. W. Nunes, J. Bennetto, and David Vanderbilt, 

689: Phys. Rev. Lett {\bf 77}, 1516 (1996).

690: 

691: \bibitem{hansen} L.~B.~Hansen, K.~Stokbro, B.~I.~Lundqvist,

692: K.~W.~Jacobsen, and D.~M.~Deaven, Phys. Rev. Lett {\bf 75}, 4444

693: (1995).

694: 

695: \bibitem{hirsch2} P.~B.~Hirsch, J. Phys. (Paris), Colloq. {\bf 40},

696: C6-27 (1979).

697: 

698: \bibitem{jones} R.~Jones, J. Phys. (Paris), Colloq. {\bf 40},

699: C6-33 (1979).

700: 

701: \bibitem{Nunes_2} J. Bennetto, R. W. Nunes, and David Vanderbilt, 

702: Phys. Rev. Lett {\bf 79}, 245 (1997).

703: 

704: \bibitem{batson} P.~E.~Batson, Phys. Rev. Lett {\bf 83}, 4409 (1999).

705: 

706: \bibitem{Vall} A. Valladares, A. K. Petford-Long, and A. P. Sutton,

707: Phil. Mag. Lett. {\bf 79}, 9 (1999).

708: 

709: \bibitem{nv} R. W. Nunes and David Vanderbilt, Phys. Rev. Lett {\bf 85}, 3540 

710: (2000).

711: 

712: \bibitem{Blase} X. Blase, Karin Lin, A. Canning, S. G. Louie, and D. C. Chrzan,

713: Phys. Rev. Lett {\bf 84}, 5780 (2000).

714: 

715: \bibitem{Nunes_1} R. W. Nunes and David Vanderbilt, J. Phys.:

716: Condens. Matter {\bf 12}, 10021 (2000).

717: 

718: \bibitem{Nunes_3} R. W. Nunes, J. Bennetto, and David Vanderbilt, 

719: Phys. Rev. B {\bf 58}, 12563 (1998).

720: 

721: \bibitem{Lehto} Niklas Lehto and Sven \"{O}berg,

722: Phys. Rev. Lett {\bf 80}, 5568 (1998).

723: 

724: \bibitem{Ters} J. Tersoff, Phys. Rev. B {\bf 39}, 5566 (1989).

725: 

726: \bibitem{Potential1} H. Balamane, T. Halicioglu, and W. A. Tiller,

727: Phys. Rev. B {\bf 46}, 2250 (1992).

728: 

729: \bibitem{Potential2} M. S. Duesbery, B. Joos, and D. J. Michel,

730: Phys. Rev. B {\bf 43}, 5143 (1991).

731: 

732: \bibitem{edip1} M. Z. Bazant, E. Kaxiras, and J. F. Justo, 

733: Phys. Rev. B {\bf 56}, 8542 (1997).

734: 

735: \bibitem{edip2} J. F. Justo, M. Z. Bazant, E. Kaxiras, V. V. Bulatov, 

736: and S. Yip, Phys. Rev. B {\bf 58}, 2539 (1998).

737: 

738: \bibitem{Caetano} Caetano R. Miranda and A. Antonelli, unpublished.

739: 

740: \bibitem{harm} K. Moriguchi, S. Munetoh, A. Shintani, T. Motooka, 

741: Phys. Rev. B {\bf 64}, 195409 (2001).

742: 

743: \bibitem{fexp}  D. R. Lide and H. V. Kehiaian, {\it CRC Handbook of Thermophysical and

744: Thermochemical Data}, (CRC Press, Boca Raton, 1994).

745: 

746: \bibitem{Frenkel} D. Frenkel and B. Smit, {\it Understanding Molecular

747: Simulations}, (Academic Press, San Diego, 2001).

748: 

749: \bibitem{RS_1} Maurice de Koning, A. Antonelli, and Sidney Yip,

750: Phys. Rev. Lett {\bf 83}, 3973 (1999).

751: 

752: \bibitem{AS_1} Maurice de Koning and A. Antonelli

753: Phys. Rev. E {\bf 53}, 465(1996); Phys. Rev. B {\bf 55}, 735 (1997).

754: 

755: \bibitem{Kane} E. O. Kane, Phys. Rev. B {\bf 31}, 7865 (1985).

756: 

757: \bibitem{Nilsson} G. Nilsson and G. Nelin, Phys. Rev. B {\bf 6}, 3777 (1972).

758: 

759: \bibitem{bc} The issue of supercell boundary conditions in dislocation

760: calculations has been properly addressed in

761: Ref.~\onlinecite{Lehto}. When the choice of supercell vectors is such

762: that when the cells are stacked in the direction perpendicular to the slip

763: plane, the dislocation with Burgers vector {\bf b} ({\bf -b}) in

764: one cell lies on top of the dislocation with the same Burgers vector

765: {\bf b} ({\bf -b}), in the next cell, we have a dipolar

766: configuration. On the other hand, when the cell vectors are chosen

767: such that the dislocation with Burgers vector {\bf b} ({\bf -b}) lies

768: on top of the dislocation with the opposite Burgers vector {\bf -b}

769: ({\bf b}), in the next cell, we have a quadrupolar configuration. As

770: shown in Ref.~\onlinecite{Lehto}, the former choice minimizes the spurious

771: strains associated with the periodic array of dislocations, in the

772: case of the 90$^{\circ}$ partial.

773: 

774: \bibitem{chrzan} K.~Lin and D.~C.~Chrzan, Mater. Sci. and Eng. {\bf

775: A319-321}, 115 (2001).

776: \end{references}

777: 

778:            

779: \begin{table}

780: \centering{

781: \caption{Calculated enthalpy differences at 0 K between core

782: reconstructions of the 90$^{\circ}$ partial dislocation, for four

783: different cell sizes, in meV/\AA. Results are shown for TETB (from

784: Ref.~17), Tersoff, and EDIP. Cells are defined by their height $D$ and

785: width $L$ (in \AA), as defined in the text. The number of atoms

786: ($N_{at}$) in each cell is also included. $\Delta{E}_{SP}$ is the

787: difference in enthalpy between the cells with the same senses and

788: opposite senses of the SP reconstruction (see text).  $\Delta E_{SP -

789: DP}$ is the difference in enthalpy between the average of the SP cells

790: and the DP structure.}

791: \label{tab1}

792: \begin{tabular}{lcccccccc}  

793: & & &\multicolumn{2}{c}{Tersoff} &\multicolumn{2}{c}{EDIP} &\multicolumn{2}{c}{TBTE} \\

794: $~N_{at}$ &$L$ &$D$ &$E_{SP-DP}$ &$\Delta E_{SP}$ &$E_{SP-DP}$ &$\Delta E_{SP}$ &$E_{SP-DP}$ &$\Delta E_{SP}$ \\

795: \hline

796: ~192 &13.3 &18.8 &~13 &24 &-35 &~6 &~62 &39 \\

797: ~240 &16.6 &18.8 &~~9 &16 &-39 &~4 &~55 &26 \\

798: ~576 &26.6 &28.2 &~~7 &~6 &-39 &~2 &~55 &~9 \\

799: 1920 &53.2 &47.0 &~~6 &~2 &-40 &~0 &~55 &~4 

800: \end{tabular}

801: }

802: \end{table}         

803:            

804:  

805: \begin{figure}

806: \caption{Atomic structure of the 90$^{\circ}$ partial dislocation

807: viewed from above the $\{111\}$ slip plane. (a) symmetrically

808: reconstructed core; (b) SP reconstruction; and (c) DP

809: reconstruction.

810: \label{cores}}

811: \end{figure} 

812:           

813: \begin{figure}

814: \caption{Gibbs free-energy for the bulk of silicon. Full and dashed

815: lines show our RS-MC calculations, including anharmonic effects, for

816: the EDIP and Tersoff potentials, respectively. The dotted line shows the

817: Tersoff-potential harmonic-approximation results from Ref.~28,

818: and the squares are the experimental results from Ref.~29.

819: \label{bulk}}

820: \end{figure} 

821:           

822: \begin{figure}

823: \caption{Gibbs free-energy difference per unit length $\Delta G$,

824: between the SP and the DP geometries of the 90$^\circ$ partial

825: dislocation in silicon, with different samplings of the phonon modes

826: along the dislocation direction. For the 192-atom calculations, the

827: results for the SP cores with same reconstruction senses, with

828: opposite reconstruction senses, and the average between the two (as

829: explained in the text) are shown by the squares, circles, and

830: triangles, respectively. For the 384-atom cell, same reconstruction

831: senses, opposite reconstruction senses, and the average are

832: represented by the solid line, the dashed line, and the dotted line,

833: respectively.

834: \label{tamanho}}

835: \end{figure} 

836: 

837: \begin{figure}

838: \caption{Gibbs free-energy difference per unit length ($\Delta G$) and

839: the entropic contribution $-T\Delta S$, between the SP and the DP

840: geometries in silicon, in $\rm{meV/\AA}$, using the Tersoff

841: potential (full line for $\Delta G$ and squares for $-T\Delta S$), and

842: the EDIP potential (dashed line for $\Delta G$ and circles for

843: $-T\Delta S$).

844: \label{G-si}}

845: \end{figure} 

846: 

847: \begin{figure}

848: \caption{Gibbs free-energy difference per unit length $\Delta G$, and

849: the entropic contribution $-T\Delta S$, in $\rm{meV/\AA}$, between the

850: SP and the DP geometries in germanium, using the Tersoff

851: potential (full line for $\Delta G$ and squares for $-T\Delta S$).

852: \label{G-ge}}

853: \end{figure} 

854: 

855: \begin{figure}

856: \caption{Difference in free energy between SP and DP cores in silicon for

857: RS-MC calculations using the Tersoff (full line) and EDIP (dashed

858: line) potentials, compared with the harmonic-approximation results

859: from Ref.~10. RS-MC energies at 0 K were shifted, for better

860: comparison (see text).

861: \label{vallad}}

862: \end{figure} 

863: 

864: \end{document}

865: