PCB Thermal Layout Optimization for Power Electronics: Integrating Large Language Models with NSGA-II for Enhanced Energy Efficiency

Paper in proceeding, 2025

This paper proposes an LLM-enhanced PCB thermal layout optimization method for power electronics, integrating the Non-dominated Sorting Genetic Algorithm II (NSGA-II) multi-objective evolutionary algorithm with large language models (LLMs) to improve thermal management. Our approach leverages LLMs to guide the optimization process through intelligent population generation and solution selection based on design knowledge. Experimental validation using a multivibrator circuit demonstrates a 7.0% reduction in maximum PCB temperature compared to conventional NSGA-II methods, while simultaneously improving routing quality and area utilization. The proposed method generates more diverse Pareto solutions and achieves better trade-offs between conflicting design objectives. This approach addresses limitations of traditional optimization methods and provides an effective framework for thermal-aware PCB layout design in power electronic applications.