An Improved Backpropagation Neural Network Method for Characterizing Performance Degradation of LDMOS Device Under Transmission Line Pulse Interference

With the wide applications of Lateral Diffused Metal-Oxide-Semiconductor (LDMOS) transistors in power conversion systems, their reliability has become a growing concern. It is also well known that electrostatic discharge (ESD) has long been the primary sources of electromagnetic interference, and its cumulative effects can result in the performance degradation of devices significantly. However, existing research on the characterization of performance degradation of LDMOS devices are mainly focused on their physical models, and accurately quantifying the degradation under diverse conditions still remain a challenge. Here, transmission line pulse (TLP) is used to simulate the ESD interference in LDMOS devices. One performance degradation prediction method, based on a multi-head incremental i.e. improved backpropagation (BP) neural network (IBP-NN), is developed, where the input parameters as well as network structure are optimized, and therefore, the prediction accuracy of device degradation is enhanced greatly. Further, the degradation mechanisms of MOS devices are analyzed, followed by the validation of independence of different parameters affecting the degradation process. In particular, our experimental results show that the proposed method can perform well in predicting the safety or reliability of LDMOS devices under different electromagnetic pulse loading cases.