Novel Approach of Determining and Predicting SiC MOSFET’s On Resistance From Device Case Temperature Using Machine Learning

Silicon-carbide (SiC) MOSFET devices are increasing in popularity in high-power converter applications. Device on-resistance (R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dson</inf>) is an important indicator for SiC MOSFET health status. Increments in R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dson</inf> over device lifetime indicate imminent device failure and result in decreased system efficiency. Direct and accurate measurement of SiC MOSFET device R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dson</inf> in high-power applications is difficult. Another approach is to estimate/predict the device R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dson</inf> from other, more easily measurable quantities, however, little work has been done on this approach in the literature. This leaves a significant technical gap in measuring/predicting device R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dson</inf> and slows down the device health status monitoring and power converter reliability research. To address the technical gaps, this work proposes a novel approach to predicting device R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dson</inf> from thermal cycle count and instantaneous temperature using machine learning regression models. The actual hardware data collected from accelerated lifetime tests of high-power SiC MOSFETs are used to train, test, and validate the proposed machine-learning regression models. The developed models, coupled with cycle counting algorithms, and device case thermal measurements, provide accurate live estimates of R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dson</inf> and can be used to predict changes in R<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">dson</inf> over expected mission profiles during power converter design.