ALCF: SYNAPS-I Brings Agentic AI to Experimental Workflows via Genesis Mission

Argonne National Laboratory Demonstrates Agentic AI Platform to Revolutionize Scientific Experimentation

Argonne National Laboratory, in collaboration with the U.S. Department of Energy’s (DOE) Lawrence Berkeley National Laboratory (LBNL) and other national labs, has unveiled a new agentic AI platform designed to automate complex experimental workflows through natural-language instructions, according to a report by the Argonne Leadership Computing Facility (ALCF). The technology, part of the SYNAPS-I project under the DOE’s Genesis Mission, replaces traditional data analysis with autonomous experiment control, enabling scientists to interact with AI systems using everyday language.

Agentic AI Transforms Scientific Workflow Automation

The SYNAPS-I project’s second phase introduces an agentic AI layer that allows researchers to direct beamlines and experimental setups by simply describing their goals, such as “map this area at high resolution” or “locate the interface between two regions.” This capability eliminates the need for technical coding or using complex control panels, as the AI autonomously handles data collection, processing, and analysis.

Ming Du, an Argonne computational scientist and SYNAPS-I team member, emphasized the platform’s potential to free scientists from repetitive tasks. “By automating these tasks, researchers have more time to focus on the science and on operating the beamline itself,” he said. The system also incorporates Meta’s Segment Anything Model 3 (SAM3) for image segmentation, enabling precise feature isolation and labeling within experimental data.

Agentic AI Enables Autonomous X-Ray Experiment Adjustments

The platform builds on the first phase of SYNAPS-I, which demonstrated real-time AI-driven ptychographic image reconstruction at the Advanced Photon Source (APS). Ptychography, an X-ray technique that reconstructs high-resolution images from overlapping diffraction patterns, now benefits from agentic AI’s ability to autonomously adjust experiments.

Aileen Luo, an Argonne assistant computational scientist, highlighted the impact of real-time reconstruction. “This enables researchers to adjust measurements on the fly instead of discovering key features after the experiment has ended,” she said. During a recent demonstration, the AI located a critical interface in a microelectronics sample by scanning, analyzing, and refining its search without human intervention.

Collaborative Effort Across National Laboratories

The SYNAPS-I project involves a coalition of national laboratories, including Brookhaven National Laboratory (BNL), SLAC National Accelerator Laboratory, and Oak Ridge National Laboratory (ORNL), using computing resources from the ALCF, the Oak Ridge Leadership Computing Facility, and the National Energy Research Scientific Computing Center. The ALCF integrated SAM3 into its Inference Service, making the model accessible to DOE researchers.

Mathew Cherukara, an Argonne computational scientist and leader of the Computing and AI group and the Argonne SYNAPS-I team, described the project as a “next-generation, AI-enhanced microscope” with applications beyond microelectronics. “The real achievement is developing a next-generation, AI-enhanced microscope that can benefit many scientific fields,” he said.

Implications for Self-Driving Laboratories

The demonstration moves the field closer to autonomous AI-driven laboratories, where experiments operate, adapt, and interpret results with minimal human oversight. The DOE’s Genesis Mission, led by LBNL, aims to accelerate scientific discovery through such innovations.

Cherukara noted the potential for closed-loop experimentation and automated defect detection, which could streamline research in semiconductor manufacturing and advanced materials. Together, these capabilities establish a general framework for accelerating materials research across many domains.

Source and Citation Details

The findings were reported by the ALCF, with contributions from researchers at Argonne, LBNL, and other institutions.

U.S. Department of Energy

Argonne Leadership Computing Facility

DOE Office of Science