Presentation Information
[9a-N106-5]Reinforcement learning-enabled voltage scheduling for high-efficiency and linear resistive switching of analog memristor
〇ZHUO DIAO1, Tetsuya Tohei1, Akira Sakai1 (1.Osaka Univ.)
Keywords:
analog memristor,Measurement Informatics,Machine Learning
Analog memristors, capable of multi-level resistance states, are promising for applications such as AI computing. However, nonlinear resistance changes during programming remain a key challenge. This study proposes a voltage scheduling optimization framework based on reinforcement learning (RL) to address this issue by simultaneously improving linearity and write efficiency. By combining simulation-based pretraining with transfer learning on actual devices, the framework enables autonomous and efficient resistance control without manual tuning. Results demonstrate that the proposed method achieves high-precision resistance setting and effective voltage scheduling for analog memristors.