Neural Network-based Classification of Breakdown Mechanisms and Prediction of breakdown Voltage and On-resistance for 4H-SiC Trench Gate MOS Devices

Since neural networks do not face the convergence issues like TCAD simulations, we attempted to construct two types of neural networks for numerical prediction and mechanism classification, respectively. After training with the SiC trench gate MOS dataset obtained from different structural parameters using TCAD simulations, the accurate prediction results were achieved. During the validation process, the breakdown type classifier achieves an accuracy of 96.5%, and the numerical predictors for breakdown voltage and on-resistance have the average errors within 10%.