Auto-sizing of Multi-stage Complementary Metal Oxide Semiconductor Operational Amplifiers by Deep Q-Network and Particle Swarm Optimization

Abstract

This paper presents a study on the application of a deep Q-network combined with particle swarm optimization in the automatic sizing of multi-stage complementary metal oxide semiconductor operational amplifiers. Our main novelty is to combine deep Q-network and particle swarm optimization to achieve fast convergence and optimal circuit performance. The combined method takes advantage of the global search capability by deep Q-network and the local search capability by particle swarm optimization, enabling shortened training time cost needed by applying a deep Q-network only. It is demonstrated that the combined method can work out better operational amplifier device sizes in our experiment. The two-stage simple Miller compensation circuit reaches all the design targets in an average time of 0.32 h, and the three-stage circuits in this paper reach all the design targets within 0.68 h. Comparisons of using our method with other sizing methods such as genetic algorithm, particle swarm optimization only, and deep Q-network only are reported.