An Area-Efficient SiC-MOSFET Termination Optimization Method Based on Deep Learning and Specified MOPSO Algorithm

The breakdown and reliability performance of SiC power devices is greatly affected by edge termination design. However, considering the complexity of parameters and the tradeoff between multiple objectives, it is a challenging process to achieve high BV and area efficiency simultaneously in typical termination structures like field-limit ring (FLR). To address the problem, this paper proposes a novel and efficient termination design optimization method based on physically characterized variables. With the utilization of auxiliary parameter γ specified multiple objective particle swarm optimization (MOPSO) algorithm, a 110 V improvement in BV and more than 26% reduction in termination area has been achieved respectively. Importantly, the design scheme provided by this method demonstrates superiority over the dataset utilized during training. The results are validated by TCAD simulation, and we also successfully predicted the performance gain of the actual manufactured devices. Therefore, this paper offers a potential solution approach for multi-objective optimization problems. With the diverse distribution of Pareto solutions, this method proves broad applicability in the design of power devices involving multi-objective optimization.