No abstract

Growth of high quality InGaSb crystals by Vertical Gradient Freezing (VGF) under microgravity was numerically simulated. Machine learning tools were used to optimize the growth conditions. The study focuses on controlling the growth interface shape which directly affects the quality and homogeneity of the grown crystals. Initially, Bayesian optimization was utilized to search for the most favorable growth conditions that promote a desirable flatter growth interface shape. Afterwards, a reinforcement learning model was developed. The system was subjected to a lower temperature gradient near the feed crystal and to crucible rotation with a rate ranging according to the obtained optimal strategy. Results showed that the interface deformation is considerably reduced, and a flatter growth interface could be maintained. The growth rate and solute concentration uniformity were also improved. This adaptive control recipe proves to hold a great potential in the continuous and rapid optimization of other crystal growth processes.

No abstract