1 \chapter{Silicon carbide precipitation simulations}
3 The molecular dynamics (MD) technique is used to gain insight into the behavior of carbon existing in different concentrations in crystalline silicon on the microscopic level at finite temperatures.
4 Both, quantum-mechanical and classical potential molecular dynamics simulations are performed.
5 Since quantum-mechanical calculations are restricted to a few hundreds of atoms only small volumes composed of three unit cells in each direction and small carbon concentrations are simulated using the VASP code.
6 Thus, investigations are restricted to the diffusion process of single carbon interstitials and the agglomeration of a few dumbbell interstitials in silicon.
7 Using classical potentials volume sizes up to 31 unit cells in each direction and high carbon concentrations are realizable.
8 Simulations targeting the formation of silicon carbide precipitates are, thus, attempted in classical potential calculations only.
10 \section{Ab initio MD simulations}
12 No pressure control, since VASP does not support this feature in MD mode.
13 The time step is set to one fs.
14 Explain some more parameters that differ from the latter calculations ...
16 Molecular dynamics simulations of a single, two and ten carbon atoms in $3\times 3\times 3$ unit cells of crytsalline silicon are performed.
18 \section{Classical potential MD simulations}
20 In contrast to the quantum-mechanical MD simulations the developed classical potential MD code is able to do constant pressure simulations using the Berendsen barostat.
21 The system pressure is set to zero pressure.
22 Due to promising advantages over the Tersoff potential the bond order potential of Erhard and Albe is used.
23 A time step of one fs is set.
25 \subsection{Simulations at temperatures used in ion beam synthesis}
26 \label{subsection:initial_sims}
28 In initial simulations aiming to reproduce a precipitation process simulation volumes of $31\times 31\times 31$ unit cells are utilized.
29 Periodic boundary conditions in each direction are applied.
30 The system temperature is set to $450\, ^{\circ}\mathrm{C}$, the temperature for which epitaxial growth of 3C-SiC films is achieved by ion beam synthesis (IBS).
31 After equilibration of the kinetic and potential energy carbon atoms are consecutively inserted.
32 The number of carbon atoms $N_{\text{Carbon}}$ necessary to form a spherical precipitate with radius $r$ is given by
34 N_{\text{Carbon}}=\frac{4}{3}\pi r^3 \cdot \frac{4}{a_{\text{SiC}}^3}
35 =\frac{16}{3} \pi \left( \frac{r}{a_{\text{SiC}}}\right)^3
36 \label{eq:md:spheric_prec}
38 with $a_{\text{SiC}}$ being the lattice constant of 3C-SiC.
39 In IBS experiments the smallest precipitates observed have radii starting from 2 nm up to 4 nm.
40 For the initial simulations a total amount of 6000 carbon atoms corresponding to a radius of approximately 3.1 nm is chosen.
41 In separated simulations these 6000 carbon atoms are inserted in three regions of different volume ($V_1$, $V_2$, $V_3$) within the simulation cell.
42 For reasons of simplification these regions are rectangularly shaped.
43 $V_1$ is chosen to be the total simulation volume.
44 $V_2$ approximately corresponds to the volume of a minimal 3C-SiC precipitate.
45 $V_3$ is approximately the volume containing the necessary amount of silicon atoms to form such a precipitate, which is slightly smaller than $V_2$ due to the slightly lower silicon density of 3C-SiC compared to c-Si.
46 The two latter insertion volumes are considered since no diffusion of carbon atoms is expected within the simulated period of time at prevalent temperatures.
47 This is due to the overestimated activation energies for carbon diffusion as pointed out in section \ref{subsection:defects:mig_classical}.
48 For rectangularly shaped precipitates with side length $L$ the amount of carbon atoms in 3C-SiC and silicon atoms in c-Si is given by
50 N_{\text{Carbon}}^{\text{3C-SiC}} =4 \left( \frac{L}{a_{\text{SiC}}}\right)^3
52 N_{\text{Silicon}}^{\text{c-Si}} =8 \left( \frac{L}{a_{\text{Si}}}\right)^3 \text{ .}
55 Table \ref{table:md:ins_vols} summarizes the side length of each of the three different insertion volumes determined by equations \eqref{eq:md:n_prec} and the resulting carbon concentrations inside these volumes.
56 Looking at the carbon concentrations simulations can be distinguished in simulations occupying low ($V_1$) and high ($V_2$, $V_3$) concentrations of carbon.
59 \begin{tabular}{l c c c}
62 & $V_1$ & $V_2$ & $V_3$ \\
64 Side length [\AA] & 168.3 & 50.0 & 49.0 \\
65 Carbon concentration [$\frac{1}{\text{c-Si unit cell}}$] & 0.20 & 7.68 & 8.16\\
70 \caption{Side lengthes of the insertion volumes $V_1$, $V_2$ and $V_3$ used for the incoorperation of 6000 carbon atoms.}
71 \label{table:md:ins_vols}
74 The insertion is realized in a way to keep the system temperature constant.
75 In each of 600 insertion steps 10 carbon atoms are inserted at random positions within the respective region, which involves an increase in kinetic energy.
76 Thus, the simulation is continued without adding more carbon atoms until the system temperature is equal to the chosen temperature again, which is realized by the thermostat decoupling excessive energy.
77 Every inserted carbon atom must exhibit a distance greater or equal than 1.5 \AA{} to present neighboured atoms to prevent too high temperatures.
78 Once the total amount of carbon is inserted the simulation is continued for 100 ps followed by a cooling-down process until room temperature, that is $20\, ^{\circ}\mathrm{C}$ is reached.
79 Figure \ref{fig:md:prec_fc} displays a flow chart of the applied steps involved in the simulation sequence.
82 \begin{pspicture}(0,0)(15,17)
84 \psframe*[linecolor=hb](3,11.5)(11,17)
85 \rput[lt](3.2,16.8){\color{gray}INITIALIZIATION}
86 \rput(7,16){\rnode{14}{\psframebox{Create $31\times 31\times 31$
88 \rput(7,15){\rnode{13}{\psframebox{$T_{\text{s}}=450\,^{\circ}\mathrm{C}$,
89 $p_{\text{s}}=0\text{ bar}$}}}
90 \rput(7,14){\rnode{12}{\psframebox{Thermal initialization}}}
91 \rput(7,13){\rnode{11}{\psframebox{Continue for 100 fs}}}
92 \rput(7,12){\rnode{10}{\psframebox{$T_{\text{avg}}=T_{\text{s}}
93 \pm1\,^{\circ}\mathrm{C}$}}}
98 \ncbar[angle=0]{->}{10}{11}
100 \nbput*{\scriptsize false}
102 \psframe*[linecolor=lbb](3,6.5)(11,11)
103 \rput[lt](3.2,10.8){\color{gray}CARBON INSERTION}
104 \rput(3,10.8){\pnode{CI}}
105 \rput(7,10){\rnode{9}{\psframebox{Insertion of 10 carbon aoms}}}
106 \rput(7,9){\rnode{8}{\psframebox{Continue for 100 fs}}}
107 \rput(7,8){\rnode{7}{\psframebox{$T_{\text{avg}}=T_{\text{s}}
108 \pm1\,^{\circ}\mathrm{C}$}}}
109 \rput(7,7){\rnode{6}{\psframebox{$N_{\text{Carbon}}=6000$}}}
113 \trput*{\scriptsize true}
114 \ncbar[angle=180]{->}{7}{8}
115 \psset{fillcolor=lbb}
116 \naput*{\scriptsize false}
117 \ncbar[angle=0]{->}{6}{9}
118 \nbput*{\scriptsize false}
119 \ncbar[angle=180]{->}{10}{CI}
120 \psset{fillcolor=white}
121 \nbput*{\scriptsize true}
123 \rput(7,5.75){\rnode{5}{\psframebox{Continue for 100 ps}}}
125 \trput*{\scriptsize true}
127 \psframe*[linecolor=lachs](3,0.5)(11,5)
128 \rput[lt](3.2,4.8){\color{gray}COOLING DOWN}
129 \rput(3,4.8){\pnode{CD}}
130 \rput(7,4){\rnode{4}{\psframebox{$T_{\text{s}}=T_{\text{s}}-
131 1\,^{\circ}\mathrm{C}$}}}
132 \rput(7,3){\rnode{3}{\psframebox{Continue for 100 fs}}}
133 \rput(7,2){\rnode{2}{\psframebox{$T_{\text{avg}}=T_{\text{s}}
134 \pm1\,^{\circ}\mathrm{C}$}}}
135 \rput(7,1){\rnode{1}{\psframebox{$T_{\text{s}}=20\,^{\circ}\mathrm{C}$}}}
139 \trput*{\scriptsize true}
140 \ncbar[angle=0]{->}{2}{3}
141 \psset{fillcolor=lachs}
142 \nbput*{\scriptsize false}
143 \ncbar[angle=180,arm=1.5]{->}{1}{4}
144 \naput*{\scriptsize false}
145 \ncbar[angle=180]{->}{5}{CD}
146 \trput*{\scriptsize false}
148 \rput(7,-0.25){\rnode{0}{\psframebox{End of simulation}}}
150 \trput*{\scriptsize true}
153 \caption[Flowchart of the simulation sequence used in molecular dnymaics simulations aiming to reproduce the precipitation process.]{Flowchart of the simulation sequence used in molecular dnymaics simulations aiming to reproduce the precipitation process. $T_{\text{s}}$ and $p_{\text{s}}$ are the preset values for the system temperature and pressure. $T_{\text{avg}}$ is the averaged actual system temperature.}
154 \label{fig:md:prec_fc}
157 The radial distribution function $g(r)$ for C-C and Si-Si distances is shown in figure \ref{fig:md:pc_si-si_c-c}.
160 \includegraphics[width=12cm]{sic_prec_450_si-si_c-c.ps}
162 \caption[Radial distribution function of the C-C and Si-Si distances for 6000 carbon atoms inserted into the three different volumes $V_1$, $V_2$ and $V_3$ at a temperature of $450\,^{\circ}\mathrm{C}$ and cooled down to room temperature.]{Radial distribution function of the C-C and Si-Si distances for 6000 carbon atoms inserted into the three different volumes $V_1$, $V_2$ and $V_3$ at a temperature of $450\,^{\circ}\mathrm{C}$ and cooled down to room temperature. The bright blue graph shows the Si-Si radial distribution for pure c-Si. The insets show magnified regions of the respective type of bond.}
163 \label{fig:md:pc_si-si_c-c}
167 \includegraphics[width=12cm]{sic_prec_450_energy.ps}
169 \caption[Total energy per atom as a function of time for the whole simulation sequence and for all three types of insertion volumes.]{Total energy per atom as a function of time for the whole simulation sequence and for all three types of insertion volumes. Arrows mark the end of carbon insertion and the start of the cooling process respectively.}
170 \label{fig:md:energy_450}
172 It is easily and instantly visible that there is no significant difference among the two simulations of high carbon concentration.
173 The first C-C peak appears at about 0.15 nm, which is compareable to the nearest neighbour distance of graphite or diamond.
174 The number of C-C bonds is much smaller for $V_1$ than for $V_2$ and $V_3$ since carbon atoms are spread over the total simulation volume.
175 These carbon atoms are assumed to form strong bonds.
176 This is supported by figure \ref{fig:md:energy_450} displaying the total energy of all three simulations during the whole simulation sequence.
177 A huge decrease of the total energy during carbon insertion is observed for the simulations with high carbon concentration in contrast to the $V_1$ simulation, which shows a slight increase.
178 The difference in energy $\Delta$ growing within the carbon insertion process up to a value of roughly 0.06 eV per atom persists unchanged until the end of the simulation.
180 The excess amount of next neighboured strongly bounded C-C bonds in the high concentration simulations make these configurations energetically more favorable compared to the low concentration configuration.
181 However, in the same way a lot of energy is needed to break these bonds to get out of the local energy minimum advancing towards the global minimum configuration.
182 Thus, such conformational chamges are very unlikely to happen.
183 This is in accordance with the constant total energy observed in the continuation step of 100 ps inbetween the end of carbon insertion and the cooling process.
184 Obviously no energetically favorable relaxation is taking place at a system temperature of $450\,^{\circ}\mathrm{C}$.
186 The C-C peak at about 0.31 nm perfectly matches the nearest neighbour distance of two carbon atoms in the 3C-SiC lattice.
187 {\color{red}Todo: Mention somewhere(!) that the distance is due to neighboured differently oriented C-Si \hkl<1 0 0> dumbbells!}
188 As can be seen from the inset this peak is also observed for the $V_1$ simulation.
189 In 3C-SiC the same distance is also expected for nearest neighbour silicon atoms.
190 The bottom of figure \ref{fig:md:pc_si-si_c-c} shows the radial distribution of Si-Si bonds together with a reference graph for pure c-Si.
191 Indeed non-zero $g(r)$ values around 0.31 nm are observed while the amount of Si pairs at regular c-Si distances of 0.24 nm and 0.38 nm decreases.
192 However, no clear peak is observed but the interval of enhanced $g(r)$ values corresponds to the width of the C-C $g(r)$ peak.
193 In addition the abrupt increase of Si pairs at 0.29 nm can be attributed to the Si-Si cut-off radius of 0.296 nm as used in the present bond order potential.
194 The cut-off function causes artificial forces pushing the Si atoms out of the cut-off region.
195 Without the abrubt increase a maximum around 0.31 nm gets even more conceivable.
196 For low concentrations of carbon, that is the $V_1$ simulation and early stages of the $V_2$ and $V_3$ simulations, analyses of configurations in which Si-Si distances around 0.3 nm appear and which are identifiable in regions of high disorder, which especially applies for the high concentration simulations, identify the \hkl<1 0 0> C-Si dumbbell to be responsible for stretching the Si-Si next neighbour distance.
197 This excellently agrees with the calculated value $r(13)$ in table \ref{tab:defects:100db_cmp} for a resulting Si-Si distance in the \hkl<1 0 0> C-Si dumbbell configuration.
201 \includegraphics[width=12cm]{sic_prec_450_si-c.ps}
203 \caption{Radial distribution function of the Si-C distances for 6000 carbon atoms inserted into the three different volumes $V_1$, $V_2$ and $V_3$ at a temperature of $450\,^{\circ}\mathrm{C}$ and cooled down to room temperature together with Si-C bonds resulting in a C-Si \hkl<1 0 0> dumbbell configuration.}
204 \label{fig:md:pc_si-c}
206 Figure \ref{fig:md:pc_si-c} displays the Si-C radial distribution function for all three insertion volumes together with the Si-C bonds as observed in a C-Si \hkl<1 0 0> dumbbell configuration.
207 The first peak observed for all insertion volumes is at approximately 0.186 nm.
208 This corresponds quite well to the expected next neighbour distance of 0.189 nm for Si and C atoms in 3C-SiC.
209 By comparing the resulting Si-C bonds of a C-Si \hkl<1 0 0> dumbbell with the C-Si distances of the low concentration simulation it is evident that the resulting structure of the $V_1$ simulation is dominated by this type of defects.
210 This is not surpsising, since the \hkl<1 0 0> dumbbell is found to be the ground state defect of a C interstitial in c-Si and for the low concentration simulations a carbon interstitial is expected in every fifth silicon unit cell only, thus, excluding defect superposition phenomena.
211 The peak distance at 0.186 nm and the bump at 0.175 nm corresponds to the distance $r(3C)$ and $r(1C)$ as listed in table \ref{tab:defects:100db_cmp} and visualized in figure \ref{fig:defects:100db_cmp}.
212 In addition it can be easily identified that the \hkl<1 0 0> dumbbell configuration contributes to the peaks at about 0.335 nm, 0.386 nm, 0.434 nm, 0.469 nm and 0.546 nm observed in the $V_1$ simulation.
213 Not only the peak locations but also the peak widths and heights become comprehensible.
214 The distinct peak at 0.26 nm, which exactly matches the cut-off radius of the Si-C interaction, is again a potential artifact.
216 For high carbon concentrations, that is the $V_2$ and $V_3$ simulation, the defect concentration is likewiese increased and a considerable amount of damage is introduced in the insertion volume.
217 The consequential superposition of these defects and the high amounts of damage generate new displacement arrangements for the C-C as well as for the Si-C pair distances, which become hard to categorize and trace and obviously lead to a broader distribution.
218 Short range order indeed is observed but only hardly visible is the long range order.
219 This indicates the formation of an amorphous SiC-like phase.
220 In fact the resulting Si-C and C-C radial distribution functions compare quite well with these obtained by cascade amorphized and melt-quenched amorphous SiC using a modified Tersoff potential \cite{gao02}.
222 \subsection{Limitations of conventional MD and short range potentials}
223 \label{subsection:md:limit}
225 At first the formation of an amorphous SiC-like phase is unexpected since IBS experiments show crystalline 3C-SiC precipitates at prevailing temperatures.
226 On closer inspection, however, reasons become clear, which are discussed in the following.
228 The first reason is a general problem of MD simulations in conjunction with limitations in computer power, which results in a slow and restricted propagation in phase space.
229 In molecular systems, characteristic motions take place over a wide range of time scales.
230 Vibrations of the covalent bond take place on the order of $10^{-14}\,\text{s}$ of which the thermodynamic and kinetic properties are well described by MD simulations.
231 To avoid dicretization errors the integration timestep needs to be chosen smaller than the fastest vibrational frequency in the system.
232 On the other hand, infrequent processes, such as conformational changes, reorganization processes during film growth, defect diffusion and phase transitions are processes undergoing long-term evolution in the range of microseconds.
233 This is due to the existence of several local minima in the free energy surface separated by large energy barriers compared to the kinetic energy of the particles, that is the system temperature.
234 Thus, the average time of a transition from one potential basin to another corresponds to a great deal of vibrational periods, which in turn determine the integration timestep.
235 Hence, time scales covering the neccessary amount of infrequent events to observe long-term evolution are not accessible by traditional MD simulations, which are limited to the order of nanoseconds.
236 New methods have been developed to bypass the time scale problem like hyperdnyamics (HMD) \cite{voter97,voter97_2}, parallel replica dynamics \cite{voter98}, temperature acclerated dynamics (TAD) \cite{sorensen2000} and self-guided dynamics (SGMD) \cite{wu99} retaining proper thermodynmic sampling.
238 In addition to the time scale limitation, problems attributed to the short range potential exist.
239 The sharp cut-off funtion, which limits the interacting ions to the next neighboured atoms by gradually pushing the interaction force and energy to zero between the first and second next neighbour distance, is responsible for overestimated and unphysical high forces of next neighboured atoms \cite{tang95,mattoni2007}.
240 This is supported by the overestimated activation energies necessary for C diffusion as investigated in section \ref{subsection:defects:mig_classical}.
241 Indeed it is not only the strong C-C bond which is hard to break inhibiting carbon diffusion and further rearrengements.
242 This is also true for the low concentration simulations dominated by the occurrence of C-Si dumbbells spread over the whole simulation volume.
243 The bonds of these C-Si pairs are also affected by the cut-off artifact preventing carbon diffusion and agglomeration of the dumbbells.
244 This can be seen from the almost horizontal progress of the total energy graph in the continuation step, even for the low concentration simulation.
245 The unphysical effects inherent to this type of model potentials are solely attributed to their short range character.
246 However, since valueable insights into various physical properties can be gained using this potentials, modifications mainly affecting the cut-off were designed.
247 One possibility is to simply skip the force contributions containing the derivatives of the cut-off function, which was successfully applied to reproduce the brittle propagation of fracture in SiC at zero temperature \cite{mattoni2007}.
248 Another one is to use variable cut-off values scaled by the system volume, which properly describes thermomechanical properties of 3C-SiC \cite{tang95} but might be rather ineffective for the challange inherent to this study.
250 To conclude the obstacle needed to get passed is twofold.
251 The sharp cut-off of the used bond order model potential introduces overestimated high forces between next neighboured atoms enhancing the problem of slow phase space propagation immanent to MD simulations.
252 Thus, pushing the time scale to the limits of computational ressources or applying one of the above mentioned accelerated dynamics methods exclusively will not be sufficient enough.
254 Instead the first approach followed in this study, is the use of higher temperatures as exploited in TAD to find transition pathways of one local energy minimum to another one more quickly.
255 Since merely increasing the temperature leads to different equilibrium kinetics than valid at low temperatures, TAD introduces basin-constrained MD allowing only those transitions that should occur at the original temperature and a properly advancing system clock \cite{sorensen2000}.
256 The TAD corrections are not applied in coming up simulations.
257 This is justified by two reasons.
258 First of all a compensation of the overestimated bond strengthes due to the short range potential is expected.
259 Secondly there is no conflict applying higher temperatures without the TAD corrections, since crystalline 3C-SiC is also observed for higher temperatures than $450\,^{\circ}\mathrm{C}$ in IBS \cite{lindner01}.
260 It is therefore expected that the kinetics affecting the 3C-SiC precipitation are not much different at higher temperatures aside from the fact that it is occuring much more faster.
261 Moreover, the interest of this study is focused on structural evolution of a system far from equilibrium instead of equilibrium properties which rely upon proper phase space sampling.
262 On the other hand, during implantation, the actual temperature inside the implantation volume is definetly higher than the experimentally determined temperature tapped from the surface of the sample.
264 \subsection{Increased temperature simulations}
265 \label{subsection:md:inct}
267 Due to the limitations of short range potentials and conventional MD as discussed above elevated temperatures are used in the following.
268 The simulation sequence and other parameters aside system temperature remain unchanged as in section \ref{subsection:initial_sims}.
269 Since there is no significant difference among the $V_2$ and $V_3$ simulations only the $V_1$ and $V_2$ simulations are carried on and refered to as low carbon and high carbon concentration simulations.
270 Temperatures ranging from $450\,^{\circ}\mathrm{C}$ up to $2050\,^{\circ}\mathrm{C}$ are used.
272 A simple quality value $Q$ is introduced, which helps to estimate the progress of structural evolution.
273 In bulk 3C-SiC every C atom has four next neighboured Si atoms and every Si atom four next neighboured C atoms.
274 The quality could be determined by counting the amount of atoms which form bonds to four atoms of the other species.
275 However, the aim of the simulation on hand is to reproduce the formation of a 3C-SiC precipitate embedded in c-Si.
276 The amount of Si atoms and, thus, the amount of Si atoms remaining in the silicon diamond lattice is much higher than the amount of inserted C atoms.
277 Thus, counting the atoms, which exhibit proper coordination is limited to the C atoms.
278 The quality value is defined to be
280 Q = \frac{\text{Amount of C atoms with 4 next neighboured Si atoms}}
281 {\text{Total amount of C atoms}} \text{ .}
284 By this, bulk 3C-SiC will still result in $Q=1$ and precipitates will also reach values close to one.
285 However, since the quality value does not account for bond lengthes, bond angles, crystallinity or the stacking sequence high values of $Q$ not necessarily correspond to structures close to 3C-SiC.
286 Structures that look promising due to high quality values need to be further investigated by other means.
290 \includegraphics[width=12cm]{tot_pc_thesis.ps}\\
291 \includegraphics[width=12cm]{tot_ba.ps}
293 \caption[Si-C radial distribution and quality evolution for the low concentration simulations at different elevated temperatures.]{Si-C radial distribution and quality evolution for the low concentration simulations at different elevated temperatures. All structures are cooled down to $20\,^{\circ}\mathrm{C}$. The grey line shows resulting Si-C bonds in a configuration of substitutional C in c-Si (C$_\text{sub}$) at zero temperature. Arrows in the quality plot mark the end of carbon insertion and the start of the cooling down step. A fit function according to equation \eqref{eq:md:fit} shows the estimated evolution of quality in the absence of the cooling down sequence.}
294 \label{fig:md:tot_si-c_q}
296 Figure \ref{fig:md:tot_si-c_q} shows the radial distribution of Si-C bonds for different temperatures and the corresponding quality evolution as defined earlier for the low concentration simulaton, that is the $V_1$ simulation.
297 The first noticeable and promising change in the Si-C radial distribution is the successive decline of the artificial peak at the Si-C cut-off distance with increasing temperature up to the point of disappearance at temperatures above $1650\,^{\circ}\mathrm{C}$.
298 The system provides enough kinetic energy to affected atoms, which are able to escape the cut-off region.
299 Another important observation in structural change is exemplified in the two shaded areas.
300 In the grey shaded region a decrease of the peak at 0.186 nm and the bump at 0.175 nm and a concurrent increase of the peak at 0.197 nm with increasing temperature is visible.
301 Similarly the peaks at 0.335 nm and 0.386 nm shrink in contrast to a new peak forming at 0.372 nm as can be seen in the yellow shaded region.
302 Obviously the structure obtained from the $450\,^{\circ}\mathrm{C}$ simulations, which is dominated by the existence of \hkl<1 0 0> C-Si dumbbells transforms into a different structure with increasing simulation temperature.
303 Investigations of the atomic data reveal substitutional carbon to be responsible for the new Si-C bonds.
304 The peak at 0.197 nm corresponds to the distance of a substitutional carbon to the next neighboured silicon atoms.
305 The one at 0.372 is the distance of the substitutional carbon atom to the second next silicon neighbour along the \hkl<1 1 0> direction.
306 Comparing the radial distribution for the Si-C bonds at $2050\,^{\circ}\mathrm{C}$ to the resulting Si-C bonds in a configuration of a substitutional carbon atom in crystalline silicon excludes all possibility of doubt.
307 The resulting bonds perfectly match and, thus, explain the peaks observed for the increased temperature simulations.
308 To conclude, by increasing the simulation temperature, the \hkl<1 0 0> C-Si dumbbell characterized structure transforms into a structure dominated by substitutional C.
310 This is also reflected in the quality values obtained for different temperatures.
311 While simulations at $450\,^{\circ}\mathrm{C}$ exhibit 10 \% of fourfold coordinated carbon simulations at $2050\,^{\circ}\mathrm{C}$ exceed the 80 \% range.
312 Since substitutional carbon has four next neighboured silicon atoms and is the preferential type of defect in elevated temperature simulations the increase of the quality values become evident.
313 The quality values at a fixed temperature increase with simulation time.
314 After the end of the insertion sequence marked by the first arrow the quality is increasing and a saturation behaviour, yet before the cooling process starts, can be expected.
315 The evolution of the quality value of the simulation at $2050\,^{\circ}\mathrm{C}$ inside the range in which the simulation is continued at constant temperature for 100 fs is well approximated by the simple fit function
317 f(t)=a-\frac{b}{t} \text{ ,}
320 which results in a saturation value of 93 \%.
321 Obviously the decrease in temperature accelerates the saturation and inhibits further formation of substitutional carbon.
322 Conclusions drawn from investigations of the quality evolution correlate well with the findings of the radial distribution results.
326 \includegraphics[width=12cm]{tot_pc2_thesis.ps}\\
327 \includegraphics[width=12cm]{tot_pc3_thesis.ps}
329 \caption[C-C and Si-Si radial distribution for the low concentration simulations at different elevated temperatures.]{C-C and Si-Si radial distribution for the low concentration simulations at different elevated temperatures. All structures are cooled down to $20\,^{\circ}\mathrm{C}$. Arrows with dashed lines mark C-C distances of \hkl<1 0 0> dumbbell combinations and those with solid lines mark C-C distances of combinations of substitutional C. The dashed line corresponds to the distance of a substitutional C with a next neighboured \hkl<1 0 0> dumbbell.}
330 \label{fig:md:tot_c-c_si-si}
332 The formation of substitutional carbon also affects the Si-Si radial distribution displayed in the lower part of figure \ref{fig:md:tot_c-c_si-si}.
333 Investigating the atomic strcuture indeed shows that the peak arising at 0.325 nm with increasing temperature is due to two Si atoms directly bound to a C substitutional.
334 It corresponds to the distance of second next neighboured Si atoms along a \hkl<1 1 0>-equivalent direction with substitutional C inbetween.
335 Since the expected distance of these Si pairs in 3C-SiC is 0.308 nm the existing SiC structures embedded in the c-Si host are stretched.
337 In the upper part of figure \ref{fig:md:tot_c-c_si-si} the C-C radial distribution is shown.
338 With increasing temperature a decrease of the amount of next neighboured C pairs can be observed.
339 This is a promising result gained by the high temperature simulations since the breaking of these diomand and graphite like bonds is mandatory for the formation of 3C-SiC.
340 A slight shift towards higher distances can be observed for the maximum above 0.3 nm.
341 Arrows with dashed lines mark C-C distances resulting from \hkl<1 0 0> dumbbell combinations while the arrows with solid lines mark distances arising from combinations of substitutional C.
342 The continuous dashed line corresponds to the distance of a substitutional C with a next neighboured \hkl<1 0 0> dumbbell.
343 By comparison with the radial distribution it becomes evident that the shift accompanies the advancing transformation of \hkl<1 0 0> dumbbells into substitutional C.
344 Next to combinations of two substitutional C atoms and two \hkl<1 0 0> dumbbells respectively also combinations of \hkl<1 0 0> dumbbells with a substitutional C atom arise.
345 In addition, structures form that result in distances residing inbetween the ones obtained from combinations of mixed defect types and the ones obtained by substitutional C configurations, as can be seen by quite high g(r) values to the right of the continuous dashed line and to the left of the first arrow with a solid line.
346 For the most part these structures can be identified as configurations of one substitutional C atom with either another C atom that practically occupies a Si lattice site but with a Si interstitial residing in the very next surrounding or a C atom that nearly occupies a Si lattice site forming a defect other than the \hkl<1 0 0>-type with the Si atom.
347 Again, this is a quite promising result, since the C atoms are taking the appropriate coordination as expected in 3C-SiC.
348 However, this is contrary to the initial precipitation model proposed in section \ref{section:assumed_prec}, which assumes that the transformation into 3C-SiC takes place in a very last step once enough C-Si dumbbells agglomerated.
350 {\color{red}Todo: Summarize again! Mention, that the agglomeration necessary in order to form 3C-SiC is missing.}
354 \includegraphics[width=12cm]{12_pc_thesis.ps}\\
355 \includegraphics[width=12cm]{12_pc_c_thesis.ps}
357 \caption[Si-C and C-C radial distribution for the high concentration simulations at different elevated temperatures.]{Si-C (top) and C-C (bottom) radial distribution for the high concentration simulations at different elevated temperatures. All structures are cooled down to $20\,^{\circ}\mathrm{C}$.}
360 Figure \ref{fig:md:12_pc} displays the radial distribution for Si-C and C-C pairs obtained from high C concentration simulations at different elevated temperatures.
361 Again, in both cases, the cut-off artifact decreases with increasing temperature.
362 Peaks that already exist for the low temperature simulations get slightly more distinct for elevated temperatures.
363 This is also true for peaks located past distances of next neighbours indicating an increase in the long range order.
364 However this change is rather small and no significant structural change is observeable.
365 Due to the continuity of high amounts of damage atomic configurations remain hard to identify even for the highest temperature.
366 Other than in the low concentration simulation analyzed defect structures are no longer necessarily aligned to the primarily existing but succesively disappearing c-Si host matrix inhibiting or at least hampering their identification and classification.
367 As for low temperatures order in the short range exists decreasing with increasing distance.
368 The increase of the amount of Si-C pairs at 0.186 nm could be positively interpreted since this type of bond also exists in 3C-SiC.
369 On the other hand the amount of next neighboured C atoms with a distance of approximately 0.15 nm, which is the distance of C in graphite or diamond, is likewise increased.
370 Thus, higher temperatures seem to additionally enhance a conflictive process, that is the formation of C agglomerates, instead of the desired process of 3C-SiC formation.
371 This is supported by the C-C peak at 0.252 nm, which corresponds to the second next neighbour distance in the diamond structure of elemental C.
372 Investigating the atomic data indeed reveals two C atoms which are bound to and interconnected by a third C atom to be responsible for this distance.
373 The C-C peak at about 0.31 nm, wich is slightly shifted to higher distances (0.317 nm) with increasing temperature still corresponds quite well to the next neighbour distance of C in 3C-SiC as well as a-SiC and indeed results from C-Si-C bonds.
374 The Si-C peak at 0.282 nm, which is pronounced with increasing temperature is constructed out of a Si atom and a C atom, which are both bound to another central C atom.
375 This is similar for the Si-C peak at approximately 0.35 nm.
376 In this case, the Si and the C atom are bound to a central Si atom.
377 To summarize, the amorphous phase remains though sharper peaks in the radial distributions at distances expected for a-SiC are observed indicating a slight acceleration of the dynamics due to elevated temperatures.
379 Regarding the outcome of both, high and low concentration simulations at increased temperatures, encouraging conclusions can be drawn.
380 With the disappearance of the peaks at the respective cut-off radii one limitation of the short range potential seems to be accomplished.
381 In addition, sharper peaks in the radial distributions lead to the assumption of expeditious structural formation.
382 The increase in temperature leads to the occupation of new defect states, which is particularly evident but not limited to the low carbon concentration simulations.
383 The question remains whether these states are only occupied due to the additional supply of kinetic energy and, thus, have to be considered unnatural for temperatures applied in IBS or whether the increase in temperature indeed enables infrequent transitions to occur faster, thus, leading to the intended acceleration of the dynamics and weakening of the unphysical quirks inherent to the potential.
384 In the first case these occupied states would be expected to be higher in energy than the states occupied at low temperatures.
385 Since substitutional C without the presence of a Si self-interstitial is energetically more favorable than the lowest defect structure obtained without removing a Si atom, that is the \hkl<1 0 0> dumbbell interstitial, and the migration of Si self-interstitials towards the sample surface can be assumed for real life experiments \cite{}, this approach is accepted as an accelerated way of approximatively describing the structural evolution.
386 {\color{red}Todo: C sub and Si self-int is energetically less favorable! Maybe fast migration of Si (mentioned in another Todo)? If true, we have to care about Si removal in simulations? In any way these findings suggest a different prec model.}
388 \subsection{Valuation of a practicable temperature limit}
389 \label{subsection:md:tval}
393 \includegraphics[width=12cm]{fe_and_t.ps}
395 \caption{Free energy and temperature evolution of plain silicon at temperatures in the region around the melting transition.}
396 \label{fig:md:fe_and_t}
398 The assumed applicability of increased temperature simulations as discussed above and the remaining absence of either agglomeration of substitutional C in low concentration simulations or amorphous to crystalline transition in high concentration simulations suggests to further increase the system temperature.
399 So far, the highest temperature applied corresponds to 95 \% of the absolute silicon melting temperature, which is 2450 K and specific to the Erhard/Albe potential.
400 However, melting is not predicted to occur instantly after exceeding the melting point due to additionally required transition enthalpy and hysteresis behaviour.
401 To check for the possibly highest temperature at which a transition fails to appear plain silicon is heated up using a heating rate of $1\,^{\circ}\mathrm{C}/\text{ps}$.
402 Figure \ref{fig:md:fe_and_t} shows the free energy and temperature evolution in the region around the transition temperature.
403 Indeed a transition and the accompanying critical behaviour of the free energy is first observed at approximately 3125 K, which corresponds to 128 \% of the silicon melting temperature.
404 The difference in free energy is 0.58 eV per atom corresponding to $55.7 \text{ kJ/mole}$, which compares quite well to the silicon enthalpy of melting of $50.2 \text{ kJ/mole}$.
405 The late transition probably occurs due to the high heating rate and, thus, a large hysteresis behaviour extending the temperature of transition.
406 To avoid melting transitions in further simulations system temperatures well below the transition point are considered safe.
407 Thus, in the following system temperatures of 100 \% and 120 \% of the silicon melting point are used.
409 \subsection{Long time scale simulations at maximum temperature}
411 As discussed in section \ref{subsection:md:limit} and \ref{subsection:md:inct} a further increase of the system temperature might help to overcome limitations of the short range potential and accelerate the dynamics involved in structural evolution.
412 A maximum temperature to avoid melting was determined in section \ref{subsection:md:tval}, which is 120 \% of the Si melting point.
413 In the following simulations the system volume, the amount of C atoms inserted and the shape of the insertion volume are modified from the values used in the first MD simulations to now match the conditions given in the simulations of the self-constructed precipitate configuration for reasons of comparability.
414 To quantify, the initial simulation volume now consists of 21 Si unit cells in each direction and 5500 C atoms are inserted in either the whole volume or in a sphere with a radius of 3 nm.
415 Since the investigated temperatures exceed the Si melting point the initial Si bulk material is heated up slowly by $1\,^{\circ}\mathrm{C}/\text{ps}$ starting from $1650\,^{\circ}\mathrm{C}$ before the C insertion sequence is started.
416 The 100 ps sequence at the respective temperature intended for the structural evolution is exchanged by a 10 ns sequence, which will hopefully result in the occurence of infrequent processes.
417 The return to lower temperatures is considered seperately.
421 \includegraphics[width=12cm]{fe_100.ps}
422 \includegraphics[width=12cm]{q_100.ps}
424 \caption[Evolution of the free energy and quality of a simulation at 100 \% of the Si melting temperature.]{Evolution of the free energy and quality of a simulation at 100 \% of the Si melting temperature. Matt colored parts of the graphs represent the C insertion sequence.}
425 \label{fig:md:exceed100}
429 \includegraphics[width=12cm]{fe_120.ps}
430 \includegraphics[width=12cm]{q_120.ps}
432 \caption[Evolution of the free energy and quality of a simulation at 120 \% of the Si melting temperature.]{Evolution of the free energy and quality of a simulation at 120 \% of the Si melting temperature. Matt colored parts of the graphs represent the C insertion sequence.}
433 \label{fig:md:exceed120}
435 Figure \ref{fig:md:exceed100} and \ref{fig:md:exceed120} show the evolution of the free energy per atom and the quality at 100 \% and 120 \% of the Si melting temperature.
437 {\color{red}Todo: Melting occurs, show and explain it and that it's due to the defects created.}
439 {\color{red}Todo: Due to melting, after insertion, simulation is continued NVE, so melting hopefully will not occur, before it will be cooled down later on.}
441 {\color{red}Todo: In additions simulations at 95 \% of the Si melting temperature are started again for longer times.}
443 \subsection{Further accelerated dynamics approaches}
445 {\color{red}Todo: self-guided MD!}
447 {\color{red}Todo: other approaches?}
451 Also, how about forcing a migration of a $V_2$ configuration to a constructed prec configuration, detrmine the maximum saddle point and let the simulation run.
projects / lectures / latex.git / blob