| Title |
Analog Circuit Design Optimization Algorithm based on DQN with Success Memory and Regeneration Set Training |
| Authors |
김도희(Dohee Kim) ; 이수훈(Soohoon Lee) ; 송익현(Ickhyun Song) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.3.126 |
| Keywords |
Circuit optimization; Deep reinforcement learning; Design automation; Operational amplifier; Q-learning |
| Abstract |
In this paper, we propose Success Memory and Regeneration-set Training Deep Q-Network (SMART-DQN), an advanced algorithm based on DQN integrating deep learning and reinforcement learning. This new algorithm is designed to efficiently optimize target specifications and was applied to a two-stage operational amplifier. Through extensive experiments, it is demonstrated that SMART-DQN, using success memory and regeneration-set training, significantly improves the convergence, speed, and accuracy of the optimization process, especially in complex design spaces where traditional methods struggle. By employing this algorithm, analog-circuit design process can be greatly accelerated, allowing for more efficient optimization while maintaining high performance. Furthermore, it enables parallel simulations by generating virtual data sets, thus further accelerating the process. This emphasizes the effectiveness of SMART-DQN in enhancing the optimization process of analog circuit design. |