Exploring electron beam induced atomic assembly via reinforcement learning in a molecular dynamics environment

Area of Science:

• Nanotechnology
• Materials Science
• Artificial Intelligence

Background:

• Atom-by-atom assembly is a key goal in nanotechnology for creating functional materials and devices.
• Scanning transmission electron microscopy enables individual atom manipulation, but control is challenging due to dynamic processes.

Purpose of the Study:

• To explore the use of artificial agents trained with reinforcement learning for automated atomic manipulation.
• To investigate the potential of AI in nanoscale fabrication and understand the physics learned by these agents.

Main Methods:

• Utilizing a simplified molecular dynamics environment with graphene and silicon (Si) dopants.
• Training artificial agents with reinforcement learning, focusing on reward function engineering.

Main Results:

• Demonstrated successful training of artificial agents for atomic manipulation in a simulated environment.
• Showed that reward function engineering can guide agents to form local dopant clusters under constraints.
• Confirmed that learned dynamics by agents encode relevant physical principles.

Conclusions:

• Reinforcement learning holds significant potential for advancing automated nanoscale fabrication.
• AI agents can learn and apply fundamental physics principles for precise atomic control.
• This work is a foundational step towards fully automated atom-by-atom assembly.