Process optimization for Mg2Si single crystal growth using crystal growth simulation and Bayesian optimization with feasibility constraints

We propose an efficient framework for optimizing the growth process of large-diameter Mg 2 Si single crystals using the Bridgman method. This approach combines crystal growth simulation and Bayesian optimization with feasibility constraints. As the crystal diameter increases, the risk of cracking due to thermal stress rises, and partial polycrystallization can occur because of undesirable shapes at the solid–liquid interface. To tackle these challenges, we formulate a multi-objective optimization problem to minimize the maximum von Mises stress while maximizing the convexity of the solid–liquid interface across different growth stages. In our optimization process, we introduce a feasibility-aware Bayesian optimization framework that integrates Gaussian process regression and Gaussian process classification to avoid conditions that could lead to crystal remelting. The proposed method effectively identifies process parameters that minimize thermal stress and promote a convex interface shape. This framework serves as a general strategy for designing efficient crystal growth processes within constrained parameter spaces.