Algorithm prediction of lifetime under high-power microwave based on T-type field plate HEMTs

To predict the reliability of T-type field plate HEMTs under high-power microwave (HPM) stress, this study constructed a deep learning model and adopted TCAD technology to explore HPM's influence on the devices. The model takes pulse width, power and duty cycle as input features, with its prediction relative error below 15% and most below 5%. Comparative experiments with five traditional machine learning (ML) algorithms (SVM, decision tree, KNN, ridge regression, linear regression) show the deep learning algorithm achieves the minimum average relative error of 6.654% and optimal performance. This data-driven model allows researchers without semiconductor expertise to quickly obtain device lifetime data under arbitrary conditions, and can reflect individual device HPM reliability for device design, boasting extremely broad application prospects in device lifetime prediction.