42 / 2023-08-29 21:26:38

High-Dimension Feature-Informed Learning Machine for Degradation Prediction of p-GaN HEMT

GaN HEMT,Degradation Prediction,High-dimension Time Series,Machine Learning

The reliability problems of p-GaN high electron mobility transistors (HEMTs) have become a great concern in recent years. Although condition monitoring tasks are important, the existed methods may have poor performance since the degradation process of p-GaN HEMTs is complicated. Besides, the degradation characteristics of GaN HEMTs not only contain single-dimension time series characteristics, but also include high-dimension characteristics. Thus, the state-of-art methods may not suit for p-GaN HEMTs since those methods only take single-dimension characteristics as the degradation features of the models. To integrate high-dimension characteristics of p-GaN HEMTs during the degradation process, we designed a high-dimension feature informed learning machine to predict the degradation tendency of p-GaN HEMT. We combine particle swarm optimization based Volterra tensor network (PSO-VTN) and multi-step robust prediction machine (MRPM) to depict the relationship of high dimension features and provide the prediction value of V_th. the proposed method has better prediction performance comparing with most of state-of-art CM methods, especially when conducting multi-step-ahead prediction. Hence, the proposed method can support the CM designing of p-GaN HEMTs for reliability decisions.