AI turns electron microscopy into materials insights in minutes

An electron microscopy image can capture atoms arranged in a crystal lattice or defects threading through a semiconductor material, but turning that image into materials insight can take weeks of careful analysis. Now, an autonomous artificial intelligence platform developed at Cornell can do that work in minutes.

The EMSeek platform, reported April 1 in Science Advances, streamlines materials research by identifying key features in a microscopy image, determining the crystal structure, predicting material properties, comparing results with existing scientific literature, and generating a report within a single, integrated workflow.

"Electron microscopy produces incredibly rich information, but the bottleneck is often turning those images into usable scientific understanding," said corresponding author Fengqi You, the Roxanne E. and Michael J. Zak Professor in Energy Systems at the Cornell Duffield College of Engineering. You is also co-director of the Cornell University AI for Science Institute.

"Our goal," You said, "was to build an autonomous AI platform that helps bridge that gap and makes advanced materials analysis faster, more integrated and more reproducible."

EMSeek employs an "agentic" architecture, in which multiple AI agents handle different parts of the workflow and are coordinated by a central system. The platform plans tasks, selects tools, and verifies results, mimicking how a human researcher might approach a complex analysis.

The researchers demonstrated that EMSeek can process a microscopy image into a structured scientific output in just two to five minutes, roughly 50 times faster than conventional expert workflows. The system was tested across 20 different materials and five tasks typically performed by researchers, showing strong performance across a range of conditions.

Beyond speed, the platform also emphasizes scientific rigor. Each step in EMSeek's process includes checks for consistency and accuracy, helping to ensure that results are transparent and reproducible. The platform can also draw on published literature to provide context for its interpretations, reducing the risk of unsupported claims.

"This is not just AI for one isolated step," said Guangyao Chen, first author of the study and a postdoctoral associate. "It is a system that connects microscopy, materials analysis and scientific reasoning in a way that can help researchers spend less time stitching together workflows and more time interpreting results."