講演情報
[17a-K505-7]Enhancing an Autoregressive Generative Wafer Polishing U-net Model
〇Kevin Operiano1, Roberto Iaconi1, Riku Tanaka1, Fumiya Kawate1, Sepasy Saeed2, Yoshifumi Watanabe2 (1.Aixtal Corporation, 2.Mipox Corporation)
キーワード：
Deep Learning、SiC、Semiconductor
Deep learning models are susceptible to prediction errors when used in simulating manufacturing processes. One of the errors, as we have observed during the prediction of the polishing amount profiles of the SiC wafer and its pad, which requires the use of output profile as a model input recursively, leads to exponential compounding errors caused by exposure bias. To mitigate this problem, we propose methods (i.e., dataset preprocessing and post-processing, and autoregressive training) to promote model robustness and output stability.