A design methodology of crystal growth furnace and process aided by two-step optimization using machine learning models and genetic algorithm

We have developed an innovative design method for crystal growth furnaces and processes involving two-step optimization. The first step focuses on finding the ideal temperature transition around the crucible without assuming a specific crystal growth furnace. The design of the crystal growth furnace and process is then optimized to replicate the ideal temperature transition. We utilized a deep neural network model in each optimization step to substitute crystal growth simulation and genetic algorithm. A proof-of-concept optimization is performed for the directional solidification of a crystalline silicon ingot in a crucible. Since our method does not rely on a predetermined furnace, we can achieve more flexible temperature distribution transitions than conventional approaches by implementing adaptable temperature boundary conditions. This allows us to refine the design of the crystal growth furnace and process, which has significant potential to advance the production of a wide range of materials and improve materials production environments and equipment design.