An interface trap density evaluation method for SiC MOSFET based on neural network

Abstract

The high density of interface defects at the gate oxide interface is closely related to the device characteristics and reliability of silicon carbide metal-oxide-semiconductor field effect transistors. However, an efficient and reliable characterization method for interface defects remains to be developed. In this work, we propose a neural network method for evaluating the interface trap density distribution and fixed oxide charge density using the transfer characteristics of devices. The neural network utilizes a long short-term memory structure to capture the mapping relationship between the transfer characteristic and the interface defect parameters. The numerical simulation data are used to form the training dataset of the network, and an effective preprocessing method is also presented. The proposed method was successfully verified by comparing simulated transfer characteristics using the evaluated results of commercial devices with corresponding measurements. Also, the interface defect parameters were evaluated by the subthreshold current method for comparison. The result shows that the evaluated results of the proposed method are close to the experimental evaluated results, with relative errors of 3.3 %, 6.6 %, and 28.2 % for the three devices under threshold voltage, respectively. Further, the proposed method was successfully applied during the high temperature gate bias tests to detect the degradation trend of the gate oxide interface. The result reflects its practicality for the interface reliability analysis of silicon carbide metal-oxide-semiconductor field effect transistors.