Zero-Shot Autonomous Microscopy for Scalable and Intelligent Characterization of 2D Materials.

Area of Science:

• Materials Science
• Artificial Intelligence
• Nanotechnology

Background:

• Atomic-scale materials characterization demands extensive expert training and time.
• Analyzing novel materials like 2D structures presents significant challenges for human experts.
• Current methods are often data-intensive and lack scalability for new material discovery.

Purpose of the Study:

• To develop a fully autonomous system for zero-shot characterization of 2D materials.
• To integrate foundation models for automated microscope control, scanning, and intelligent analysis.
• To eliminate the need for large training datasets in materials characterization.

Main Methods:

• Integration of vision foundation models (Segment Anything Model) and large language models (ChatGPT).
• Utilizing unsupervised clustering and topological analysis for automated image segmentation and data interpretation.
• Employing prompt engineering for intelligent control and analysis without additional training.

Main Results:

• Achieved 99.7% segmentation accuracy for single-layer MoS2 identification, matching human expert performance.
• Successfully detected challenging grain boundary slits, often missed by human observation.
• Demonstrated robust accuracy across variable conditions (defocus, lighting, exposure) and diverse 2D materials (graphene, MoS2, WSe2, SnSe).

Conclusions:

• ATOMIC provides a scalable, data-efficient paradigm for autonomous nanoscale materials research.
• Foundation models enable transformative, automated characterization of 2D materials.
• The system significantly accelerates the discovery and analysis of novel atomic-scale materials.