Exploring the Potential of FPGA in High-Frequency Switching DC-DC Boost Converters Using Model Predictive Control

Model Predictive Control (MPC) has demonstrated significant potential in the control of power electronics; however, the trade-off between the online computational demands and the storage requirements for offline explicit control laws presents a persistent challenge for industrial controller implementation. This work introduces an innovative FPGA-based neural network controller tailored for high-switching frequency DC/DC Boost converters. Multi-Layer Perceptron (MLP) is used to approximate parametrized Explicit MPC controller capable of adapting to fluctuations in the systems variables and parameters. The high-speed implementation of the MLP using High-Level Synthethis (HLS) on the FPGA enables operation at higher switching frequencies and allows for higher power density in power converters design with wide bandgap (WBG) devices such as GaN. This method offers substantial improvements over conventional techniques by delivering low-latency computation. Experimental validation on a 500W, 5MHz GaN-based Boost converter prototype demonstrates that the MLP-based controller closely replicates the performance of the original Explicit MPC (EMPC) controller, achieving superior transient response.