Improved Algorithm of Dueling DQN for BSIM Parameter Extraction Task

Traditional Berkeley Short-channel IGFET Model (BSIM) parameter extraction methods are inefficient, time-consuming, and heavily dependent on the experience of specialists. To tackle these issues, this paper proposes a BSIM parameter extraction algorithm based on deep reinforcement learning, combining the Dueling Deep Q-Network (Dueling DQN) architecture with Prioritized Experience Replay (PER). The algorithm also enhances the traditional ε-greedy strategy by implementing optimal step exploration, significantly improving exploration efficiency. We achieve an average error of below 2.5%. Moreover, we have automated the parameter extraction process, offering a promising alternative to existing methods. This advancement aims to streamline and enhance the efficiency of BSIM parameter extraction in semiconductor modeling. Code is available at https://github.com/zouningmu/DuelingDQNforBSIM.