Reinforcement learning in integrated circuits: Design, synthesis, layout, and hardware security

Abstract

The growing complexity of semiconductor technology has led to an increased demand for advanced optimization and automation techniques in IC design. Among these, RL has emerged as a promising approach, offering the ability to explore vast design spaces and optimize performance metrics in ways traditional methods struggle to achieve. RL techniques have demonstrated their potential to complement and even outperform conventional methods in analog and RF IC design by autonomously learning optimal design strategies, from circuit synthesis and layout generation to performance optimization. Moreover, RL-based EDA tools have shown significant promise in addressing challenges related to process variation, power efficiency, and design security. In this context, recent progress in the utilization of RL for analog/RF circuit design is reviewed, with coverage of optimization, layout automation, and security, and emphasis placed on its pivotal role in advancing modern IC development.