[3O1-GS-10v] AI応用：実世界システム

9:00 〜 9:15

[3O1-GS-10v-01] Machine learning model for surface profile prediction of Silicon Carbide semiconductor wafers in the Chemical-Mechanical Planarization process

〇Kevin Operiano¹, Roberto Iaconi¹, Fumiya Kawate¹, Saeed Sepasy², Yoshifumi Watanabe² (1. Aixtal Corporation, 2. Mipox Corporation)

キーワード：

Machine Learning、Chemical-Mechanical Planarization、Silicon Carbide、Semiconductor、Wafer Polishing

Planarization is an important step in semiconductor manufacturing to ensure wafer flatness, which is critical in circuit layer-by-layer construction. A process called chemical-mechanical planarization (CMP) is typically employed to achieve this. However, the complexity of the process, strictness of wafer specifications, and diverse target material properties are problematic and necessitate efficient process optimization. Currently, experimentation based on human expertise, traditional statistical methods, and numerical simulations are employed to optimize the process, but nonetheless require considerable time and cost. To address these concerns, leveraging the U-Net architecture, this work proposes a wafer surface profile prediction model for 6-in. Silicon Carbide (SiC) semiconductor wafers, which are known for their high material hardness and polishing difficulty. The model achieves a remarkably high R² score and low RMSE, demonstrating accurate and efficient prediction. In addition, the model can perform iterative prediction at one-ninth timestep intervals of the CMP process, helping optimize towards desired profiles.