Faster Junction Temperature Estimation Incorporating Power SiC MOSFET Mismatch

Abstract

Estimating the power semiconductor device junction temperature in high-power converters is critical to study their durability and monitor their degradation. This paper presents a faster simulation workflow and its application on a 12-SiC MOSFET single-phase power module for automotive traction inverter applications. First, an electrical model of the power module estimates the on-state current, the on and off state energies and on-state resistance, for each device, based on their threshold voltages and electron mobilities. Then, an analytical model translates these values to power losses. Finally, the Foster network thermal model, constructed using a finite element method, estimates the corresponding junction temperatures. The workflow speed is increased by replacing the double-pulse electrical model by a novel neural network trained on less than 500 simulation points and achieving 1% mean absolute percentage error. This allows to simulate one 39-minute mission profile under 100 seconds. The results demonstrate that, over 100 simulations, accounting for the device mismatch could lead to a 6% higher degradation estimation compared to the ideal case where all devices are equal.