An AI-Based Pareto-Driven Cost-Dimension Optimization of EMI filters for DC-DC Converters

This paper presents an AI-based and data-oriented optimization algorithm for the synthesis of passive Electro-Magnetic Interference (EMI) filters in DC–DC converters. The algorithm minimizes both cost and occupied area (or volume) targeting specific EMI attenuation levels and ensuring system stability. It operates on commercial component datasets and exploits the mathematical properties of Pareto dominance to explore feasible design configurations. Through dominance preserving set operators, the proposed approach guarantees exact Pareto-optimality while drastically reducing computational complexity. A neural network assists the optimization process by identifying computationally feasible ranges of component parameters, thus accelerating convergence. The method is validated on two commercial DC–DC converter evaluation boards from STMicroelectronics and Texas Instruments. It demonstrates its ability to identify solutions that outperform existing reference designs in both cost and area/volume.