Machine Learning Assisted Magnetic-Core Coupled Inductor Design for Interleaved Buck Converter

The increased demand for on-chip inductors due to advancements in heterogeneous integration necessitates compact and efficient on-chip inductors. Currently, magnetic material-based on-chip inductors have become a potential alternative to realize compact and efficient on-chip inductors due to their higher inductance density. Many on-chip circuit applications require coupled inductor implementations. The major bottleneck with the coupled on-chip magnetic inductors is the accurate characterization of the inductor metrics since the analytical expressions are not accurate due to significant higher-order effects such as proximity effects, hysteresis and, eddy losses. Towards this, we propose to investigate machine learning algorithms to design the on-chip magnetic-core coupled inductor. Our study on different machine learning algorithms suggests that the estimation of inductor metrics specifically, inductance, saturation current, losses and, coupling factor are below 10%. In addition, we also investigated the design of an interleaved buck converter, where an optimization framework is developed for efficient on-chip magnetic-core coupled inductor design by utilizing machine learning algorithms. Simulation results suggest that the interleaved buck converter design obtained from the proposed optimization framework has 83.48% efficiency at <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$350mA$</tex> load current with the on-chip inductor area of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$0.45mm^{2}$</tex>.