[19a-D103-1~12] 15.6 Group IV Compound Semiconductors (SiC)

△ [19a-D103-5] Suggestion of optimization method using thermo-fluid analysis and machine learning

〇Yosuke Tsunooka^1,2, Nobuhiko Kokubo^1,2, Shunta Harada¹, Miho Tagawa¹, Toru Ujihara^1,2 (1.Nagoya Univ., 2.AIST GaN-OIL)

Keywords:machine learning, simulation, optimization

In SiC solution growth, flow and supersaturation are important, thus thermo-fluid analysis is frequently conducted. We have successfully predicted the results of thermo-fluid analysis of solution by the neural network. In this study, we established a method of parameter optimization of crystal growth in prediction model by neural network. Firstly, we performed 800 kinds of fluid analysis with growth parameters decided randomly, and trained prediction model by neural network. Using the prediction model, we searched optimum parameters with target function determined representing ideal growth conditions. We discovered various optimum parameters that satisfy the conditions.