A Nonsegmented Behavioral Modeling for 1.2 kV SiC MOSFETs in PSpice based on Neural Network

Abstract

This article introduces an advanced modeling method for the silicon carbide (SiC) metal-oxide-semiconductor field-effect transistors (MOSFETs). This method based on artificial neural network (ANN) to address the problem of existing modeling being unable to describe the current-voltage characteristics (I - V) of the first and third quadrant in a nonsegmented model. Additionally, the voltage-dependent capacitance characteristics (C - V) of intrinsic parasitic elements-gate-source capacitance (CGS), gate-drain capacitance (CGD), and drain-source capacitance $(C_{\text{DS}})$ — are modeled using the same ANN framework, enabling comprehensive representation of the device's dynamic behavior. In order to verify the effectiveness of the proposed model method, the SiC MOSFET E3M0075120D (1200 V/75 mΩ) is modeled by ANN. ANN training utilizes $I - V$ and $C - V$ data from datasheet. The optimized ANN architecture comprises a three-layer feedforward network for both the $I - V$ and $C - V$. Then, the obtained ANN structure is described using the circuit description language in personal simulation program with integrated circuit emphasis (PSpice) to realize the MOSFET characteristics, thus completing the establishment of the model. To verified the accuracy of the model, the modeling results is compared with the measured results. Based on the comparison results, the modeling results are largely consistent with the measurement results which means this ANN-based modeling approach is suitable for power SiC MOSFETs.