Bridging electron microscopy and materials analysis with an autonomous agentic platform.

Area of Science:

• Materials Science
• Artificial Intelligence
• Electron Microscopy

Background:

• Electron microscopy (EM) is crucial for atomic-scale materials analysis but involves fragmented, time-consuming expert workflows.
• Existing AI tools address individual EM tasks but struggle with complex, multi-step analyses across diverse modalities.
• Current methods require weeks of expert effort, hindering rapid materials discovery.

Purpose of the Study:

• To develop an integrated, automated platform for connecting electron microscopy data to materials insights.
• To overcome the limitations of siloed AI approaches in complex EM analysis workflows.
• To significantly reduce the time and expertise required for materials characterization using EM.

Main Methods:

• Developed EMSeek, a modular, multiagent platform orchestrating five key units: segmentation, reconstruction, property prediction, literature retrieval, and consistency checks.
• Utilized large language models (LLMs) to automate the planning, invocation, and execution of these tools.
• Incorporated uncertainty calibration and audit-ready reporting for rigorous analysis.

Main Results:

• EMSeek achieved over 90% structural similarity on STEM2Mat and outperformed Segment Anything in segmentation speed and accuracy.
• The platform matched or surpassed expert performance on property prediction benchmarks with minimal calibration.
• Automated workflows reduced analysis time to 2-5 minutes per image, a ~50x speedup compared to traditional methods.

Conclusions:

• EMSeek significantly accelerates materials discovery by automating complex, multi-step electron microscopy workflows.
• The platform provides rigorous, actionable guidance through integrated uncertainty calibration and audit signals.
• EMSeek demonstrates the potential of LLM-orchestrated multiagent systems for advancing materials science research.