Previous theoretical modelling studies [8, 9, 10, 11] has found that the ground state configuration is the [100] C-Si split-interstitial, with symmetry, as illustrated in Fig. 1. We therefore take this structure as the starting point, and relaxed all atoms of the 72 atom cluster containing the defect shown in Fig. 1 in various charge states.
Details of the structure found for C are given in Table 1. and compare favorably with a previous ab initio calculation [11]. For the neutral defect, the C atom made strong bonds with the Si atom, labeled Si , along the [100] axis and slightly weaker bonds with its neighboring Si atoms, labeled Si , along [011]. Nevertheless, the C atom pulls in the Si neighbors along this direction leaving the Si -Si bonds along [011] slightly stretched at 2.352 Å. The Si -C-Si and Si -C-Si bond angles are 137 and 125 .
There are two high frequency modes at 922.2 ( ) and 866.9 ( ) cm , which decrease by 27.6 and 27.8 cm respectively with C. These frequencies are in fair agreement with the experimental results (Table ) of 921 and 930 cm - although the ordering of the modes is given incorrectly. The calculated isotope shifts are in very good agreement with the experimental results, with a decrease of 28.8 and 25.8 cm for the and the modes respectively. These results are similar to those of [9], where again, the ordering of the LVMs were given incorrectly.
The separation of the two vibrational modes is much greater than experiment and is very sensitive to the C-Si bond lengths [24]. If the length of the C-Si bond is reduced by 2% then the modes become 926 cm ( ) and 950 cm ( ) and the ordering is now in agreement with experiment. In fact, the 1% difference in the experimental frequencies suggests that the C-Si bond lengths can differ by no more than 0.03% [25].
We then investigated the structure of the negatively charged defect. Adding an electron and relaxing all the atoms of the cluster, without any restriction, causes the three C-Si bond lengths to become almost equal: the C-Si and C-Si bonds becoming 1.788 and 1.805 (2) Å respectively. The Si -C-Si bond angle decreased slightly to 136.0 , and the Si -Si bond length decreased slightly to 2.238 Å . The resulting vibrational modes have now crossed over with an ordering in agreement with experiment. The mode has decreased to 840.98 cm with a corresponding increase of the mode to 901.78 cm . So, once again the near degeneracy of these modes seems to arise from an equality in the three Si-C bonds. In this case, if the C bond angles were 120 , the C atom and its three Si neighbors would then make a molecule with a axis along [011]. This would then automatically give a two-fold degenerate E vibrational mode [24]. This symmetry would, however, also imply that the infra-red intensities of the two modes are equal which is contrary to experiment. The effective charges for the highest two local modes of the fully relaxed neutral defect were then calculated from the change in the cluster dipole moment with the atoms displaced according to the normal coordinates. The effective charges were found to be =0.44|e| and =0.34|e|. The IR intensity ratio of the local modes is given by , and results in a ratio in the range 1.5-1.7:1, in reasonable agreement with the experimental ratio of 2:1 with the mode being the more intense in agreement with observation. For the negatively charge defect, we find the intensity ratio to be in the range 1.4-1.5:1. Thus although the symmetry of the core of the defect is almost trigonal and the two LVMs transform as E, the departure from symmetry is sufficient to lead to quite distinct induced dipole moments.
We now discuss the electronic structure of the defect. Although the energy levels are only given approximately by the theory, we use them along with the pseudo-wavefunctions for a qualitative analysis of the zero phonon transition. The Kohn Sham eigenvalues indicated that there are at two, possibly three defect related levels in the band gap. The highest occupied state has symmetry, and in agreement with EPR [4] it corresponds to a non-bonding p- orbital on carbon (Fig. 4b). This lies above an level whose wave-function, Fig. 4a, has little amplitude on C, and is mainly localized over the Si atoms neighboring C atoms. The first unoccupied state shown in Fig. 4c has symmetry, and is diffusely localized over the three Si neighbors of C, predominantly on Si but with little amplitude on carbon. Again, this is in agreement with the EPR data on the ionized donor and acceptor states [5]. The level above the level has symmetry, and is delocalized and probably part of the conduction band. The 856 meV zero phonon line transition might be either due to an electron in the upper state recombining with a hole trapped in the state associated with the non-bonding p-orbital on C, or, alternatively a electron in the upper state recombining with a hole in the lower state. Most workers have considered it to be the former process, involving the state localized on carbon [26]. This seems to us as unlikely because the PL has no fine structure, and the zero phonon line is unaffected by substitution of C [27]. This implies that the bonding of carbon is unchanged as a result of the transition. If the C atom relaxed as a consequence of a hole occupying the C-related state - as indicated by the calculations - then an isotope shift would be expected. We therefore suggest that the observed PL involves the latter transition, which involves the diffusely localized states. This would imply that the dipole for the PL transition is at right angles to the p-orbital on the carbon atom.
The energies for migration/reorientation of the defect were also calculated for the trajectories considered by Capaz et al using a relaxation procedure with constraints. In contradiction to this work, we find that both the and bond centered saddle points have almost equivalent barrier heights, with an energy of 1.10 eV. We therefore conclude that both are viable paths for the migration/rotation of C . The 1.10 eV barrier is in fair agreement with the experimental results of between 0.73-0.87 eV for both migration and re-orientation.
In summary, the calculations yield a structure which has approximately equal C-Si bonds and gives two almost degenerate modes around 920 cm . The near degeneracy in modes being attributed to an approximate trigonal symmetry. However, the deviation from this symmetry cause the two modes to have different absorption strengths. The migration energy of the defect is low with the C atom being either two- or four-fold coordinated at the saddle point.