There is a need for a foundation of a research study aimed at investigations on near real-time reliability awareness of Gallium Nitride devices in high-frequency power converters for which we need advanced hardware and algorithms. This dissertation is moving beyond traditional reliability analysis and looking to more applicable and accurate analytical tools by introducing deep learning techniques and advanced sensing solutions. The computational structures will be applied at the edge of the power converter through online sensing and data processing units as well as on a remote server. They will provide an iterative ability to predict the time until the device may fail or reach a pre-defined degradation threshold. With the availability of the most granular information deduced from advanced devices, a new data-driven scheme is proposed for system monitoring and possible lifetime extension Gallium Nitride power converters. The approach relies on the real-time on-resistance data extraction from the power converter, and calibration of an adaptive model using multi-physics co-simulations under power cycling. More specifically, the focus is on deploying machine learning algorithms to exploit for the parameter estimation in power electronics engineering reliability. The proposed techniques in this work are quite new and have not yet been developed and analyzed for high-frequency power converters specifically with Gallium Nitride power semiconductor devices.
