Optimization of PDN decoupling capacitors for EMI Reduction based on Deep Reinforcement Learning

Abstract

The reinforcement learning (RL) is applied to the optimization of decoupling capacitors on power distribution network (PDN) for reduction of radiated emissions (REs). A small-size parallel-plates PDN structure containing two ICs is modeled as equivalent lumped-circuits, and far-field REs due to the structure are calculated using closed-form expressions. The closed-form expressions are validated with the full-wave simulation results. The environment with a proper reward system for RL is proposed by using the closed-form REs expressions. The proposed RL environment is tested with two design examples for Q-learning and deep reinforcement learning (DRL). The learning results are converged to optimal policies very efficiently, which satisfy the RE regulation with minimum number of decaps for the given PDN structures.