Multi-Agent Proximal Policy Optimization Applications in Low-Dropout Regulator Design

In analog and mixed-signal integrated circuits (ICs), low-dropout regulators (LDOs) are crucial for maintaining a stable power supply throughout the IC. As such, designing LDOs with both time and quality efficiency has attracted substantial research interest. This paper presents an implementation of multi-agent proximal policy optimization (MAPPO) in both separated-parameter and parameter-sharing configurations to address the challenges of multi-objective, multi-variable LDO design. Our experiments show that parameter-sharing MAPPO outperforms both separated-parameter MAPPO and single-agent PPO in exploration and convergence, benefiting from cooperative learning via parameter sharing, which accelerates the identification of optimal design configurations. In summary, our findings indicate that parameter-sharing MAPPO efficiently manages complex specifications and variables.