Diffusion and Defects in Elementary and Compound Semiconductors
Global parameterization of multiple point-defect dynamics models in silicon
The task of determining globally robust estimates for the thermophysical properties of intrinsic point defects in
crystalline silicon remains challenging. Previous attempts at point-defect model regression have focused on the
use of a single type of experimental data but as of yet no single parameter set has produced predictive models
for a variety of point-defect related phenomena. A stochastic optimization technique known as simulated
annealing is used to perform simultaneous regression of multiple models. Specifically, zinc diffusion in silicon
wafers and the dynamics of the so-called interstitial-vacancy boundary during Czochralski crystal growth are
used to systematically probe point-defect properties. A fully transient model for point-defect dynamics during
crystal growth is presented which employs a sophisticated adaptive mesh refinement algorithm to minimize the
computational expense associated with each optimization. The resulting framework leads to a quantitatively
coherent picture for both experimental systems, which are modeled with a single set of point-defect
thermophysical properties. Our results are entirely consistent with other recent model-fitting estimates and
indicate that as the number of experiments considered simultaneously within this framework increases it should
be possible to systematically specify these properties to higher precision.