A Reinforcement Learning-Based Method for Automated Repair of Scanning Probe Tips in a Simulated Environment

Scanning tunneling microscopy (STM) enables the manipulation of atomic structures on sample surfaces under ultra-high vacuum conditions. However, a major bottleneck in continuous STM operation is the frequent need to repair the probe tip. In this work, we present an automated STM tip repair framework utilizing a simulation system based on molecular dynamics. A Deep Q-Network (DQN) reinforcement learning algorithm is employed to learn optimal repair strategies by iteratively interacting with the simulator through tip repair actions and STM image feedback. The trained algorithm demonstrates a 30% higher success rate in tip repair compared to randomly selected actions.