Machine Learning Assisted Visual Design Space Exploration for GaN Half-Bridges with Output Filter

Abstract:
This contribution proposes a design space exploration method for GaN half-bridges with output filter, combining analytical and machine learning techniques to achieve a top-down estimate of power loss and system volume. Without explicitly formulating a design strategy, the rapid nature of the method allows designers to approximate an optimal design point or to visualize the causality of different design decisions based on high-level voltage and current system requirements. The presented examples include the estimation of an optimum switching frequency for minimum power loss considering current ripple and the analysis the scaling effect of interleaving parallelization.