PSO-Based Volterra Tensor Network for Predicting Short Circuit Degradation of p-GaN HEMT

In recent years, p-GaN high electron mobility transistors (HEMTs) have become the most promising devices of the third-generation semiconductors for its excellent performance. However, the reliability problems of p-GaN HEMTs under repetitive short circuit stresses are significant issues for both researchers and engineers working on electronic science and technology. Due to the lack of sufficient knowledges, it is difficult to build physic-of-failure (PoF) solutions to predict the degradation processes. Nevertheless, existed measurement procedures can collect enough information of the degradation process of the devices. Hence, data-driven (DD) techniques are promising tools to monitor the degradation states of p-GaN HEMTs. However, state-of-art DD methods cannot depict the relationship of high-dimension features under the degradation processes. To overcome the weakness, we proposed a prediction model called PSO-based Volterra tensor network (PSO- VTN). The proposed model builds tensor network respect to a Volterra system and take PSO algorithms to optimize the corresponding parameters. Experiments show that PSO-VTN can successfully predict the output performance and transfer performance of the target devices with the historical data and provide support information for reliability decisions.