Machine learning-enhanced optical tweezers for defect-free rearrangement

Abstract

Optical tweezers constitute pivotal tools in Atomic, Molecular, and Optical (AMO) physics, facilitating precise trapping and manipulation of individual atoms and molecules. This process affords the capability to generate desired geometries in both one-dimensional and two-dimensional spaces, while also enabling real-time reconfiguration of atoms. Due to stochastic defects in these tweezers, which cause catastrophic performance degradation especially in quantum computations, it is essential to rearrange the tweezers quickly and accurately. Our study introduces a machine learning approach that uses the Proximal Policy Optimization model to optimize this rearrangement process. This method focuses on efficiently solving the shortest path problem, ensuring the formation of defect-free tweezer arrays. By implementing machine learning, we can calculate optimal motion paths under various conditions, resulting in promising results in model learning. This advancement presents new opportunities in tweezer array rearrangement, potentially boosting the efficiency and precision of quantum computing research.