Advanced Analog Design Optimization: Comparison Between Reinforcement Learning and Heuristic Algorithms
Michel Chevalier, Severin Trochut, Roberto Guizzetti, Pascal Urard, Lioua Labrak, John Samuel, Rémy Cellier, Nacer Abouchi · 2024
Last decades have seen a lot of research on Analog Design Automation. The most recent approaches are based on Reinforcement Learning (RL). This paper describes a new learning strategy enhancing the most recent Proximal Policy Optimization RL approach, applied to analog design. The method is compared to more classical heuristic approaches such as Ant Colony Optimization, Particle Swarm Optimization or Differential Evolution algorithm. This study is done using a gymnasium environment embedding a spice-like electrical simulator. The experiments are done under equivalent calculation conditions. The paper highlights convergence properties and demonstrates the RL ability to avoid local minimum traps.