Verification of the die Temperature of a SiC Transistor Based on the Thermal Imaging and Machine Learning

The article presents a method for the automatic verification of the thermograms of the transistor packages, enabling the differentiation between those in which the die temperature is correct and those in which it is too high. This method is based on an indirect thermographic measurement of the transistor die temperature. The paper describes a measurement methodology that provides a reliable temperature assessment, based on simulations using the Finite Element Method (FEM). A transistor operating in pulsed mode (switching at a specific frequency, rather than in the active state) was used. The procedure for determining the coefficients required for these simulations is also presented. Additionally, the article includes an algorithm for analyzing thermograms of the investigated transistor packages and the results of this analysis, which allow for distinguishing components operating at the correct and excessive temperatures.