We propose that a scheme consistent with the available experimental evidence for later thermal donors is one whereby all the later TDs contain a common thermal donor core which is not modified in the later species. Instead they could develop through nearby weak dimer binding to the core, forming a `cloud' of Oi around the donor core. This binding will be increasingly weak with increasing defect size, and these atoms will not significantly perturb the core. Thus there will be a common set of vibrational modes as observed, and any symmetry idenfication will only measure the C2v symmetry of the core. The defect levels will become increasingly shallow as these aggregate atoms gradually increase the stress exerted on the defect core. This model is also consistent with the rapid observed reorientation under stress.
Such a picture does not attempt to tackle the final stages of TD formation. After 900 hours annealing at 420C the 999 cm-1 thermal donor vibrational absorption drops away, while three new bands appear at 1006, 1012 and 1015 cm-1. It seems likely that this stage represents the thermal donors finally transforming into quartz precipitates, with these new bands being associated with other electrically active defect complexes that form at this point, possibly shallow thermal donors. In addition once quartz precipitates form there are other `new donors' formed, possibly associated with the quartz boundaries.
There are other explanations for these higher order modes, apart from shallow thermal donors vibrations. They could be caused by dimer / trimer trapping near to the quartz. Alternatively if quartz precipitates are being produced in these regimes, the associated lattice strain punches out prismatic dislocation loops in the Si. These dislocations can be pinned by Oi, and so it seems reasonable they may also be pinned by dimers and trimers. Hence these broadened modes could be associated with dimers and trimers pinned at dislocations, the electrical activity coming from the dislocations themselves.
The best that theory can do at this stage is to propose various models, as has been done above, and therefore suggest discriminatory experiments to select between them. We are involved in on-going discussions with several experimental groups on this subject (Lindström and Hallberg, McQuaid, Newman and Markevich amongst others), and hope to be able to reach some firmer conclusions soon.