Reinforcement-Learning-Based Foggy-Aware Optimal Placement Method for Analog and MixedSignal Circuits

Despite advancement in artificial intelligence (AI) and subsequent successful applications in a vast variety of areas, addressing progressive need in electronic design automation for high performance and fine precision remains a significant challenge. Optimal layout placement design, a notably time-consuming task, often poses an issue to the conventional design process with an aim of maintaining high circuit performance. To mitigate this problem, in this paper we propose an AI-based optimization method to automate this process with better accuracy. We utilize a reinforcement learning (RL) method, advantage actor critic (A2C), to implement a full automation procedure for analog circuit layout design. A topological representation is employed to decrease the size of states in the optimization process. In addition, we have considered the foggy effect caused by electron-beam lithography (EBL) technology. Our simulation results demonstrate the remarkable efficacy of our approach, which can achieve 44 times smaller foggy effect variation and reduce the run time by 42 times in comparison with the analytical and another RL-based (DQN) method, without compromising the wire length and chip area minimization.