Atomic-scale manipulation using scanning tunneling microscopy (STM) has enabled the creation of quantum states of matter and miniaturization of computational circuitry. This paper uses deep reinforcement learning (DRL) to control atom manipulation, addressing challenges like unknown manipulation parameters and tip apex changes. A DRL agent learns to manipulate Ag adatoms on Ag(111) surfaces with high precision and is integrated with path planning for autonomous atomic assembly. The results demonstrate the effectiveness of DRL in real-world nanofabrication and complex atomic-scale experiments.

I-Ju Chen, Markus Aapro, Abraham Kipnis, Alexander Ilin, Peter Liljeroth, Adam S. Foster