Argonne builds AI-guided X-ray microscope that follows spoken instructions
Researchers at Argonne National Laboratory connected an agentic AI system to an X-ray nanoprobe beamline, allowing natural-language requests to guide where the instrument focuses. A test located the boundary between two regions on an integrated circuit, but the team says the approach could aid many scientific fields. The AI can reconstruct ptychographic images in real time, avoiding the hours or days that conventional analysis can require.
Argonne researchers linked an AI agent to a hard X-ray nanoprobe at beamline 26-ID. In a trial, it found the border between two areas of an integrated circuit after being asked in ordinary language.
The work is phase two of a project. Phase one, reported in April, examined whether AI could shorten ptychographic image processing. Ptychography gathers many diffraction patterns; experiments can yield millions of images in hours, and traditional reconstruction may take hours or days. The AI reconstructs in real time. The effort sits within SYNAPS-I, part of the Department of Energy's Genesis Mission, with Berkeley Lab leading and several national labs participating.
If such systems mature, they may let researchers spend less time on manual beam alignment and image reconstruction, potentially speeding experiments in materials science, semiconductor inspection, and other fields. Scientists at national laboratories and user facilities could benefit first, while industry may eventually see faster defect detection. The technology could also raise questions about reliability and oversight, since AI would help decide where to collect data and how to interpret results.