Impedance and Cost based PDN Decoupling Optimization using Reinforcement Learning

Abstract

PDN optimization involves selection of capacitors to meet the target impedance. This paper uses reinforcement learning to solve decoupling stuffing problem based on impedance-based reward and then with both impedance & cost-based reward. It is shown how the agent can be biased when trained only on impedance-based reward. Key results including attainment of target impedance and overall achieving cost and impedance optimized solution are reported.