The Lawrence Berkeley National Laboratory is deploying Meta’s open-source artificial intelligence models to resolve massive data bottlenecks at its Advanced Light Source (ALS) facility. The ALS uses beamlines—instruments producing intense X-ray light—that have seen data generation jump from one image every six seconds to 100,000 images per second following recent hardware upgrades. This volume of information, reaching tens of petabytes annually, exceeds the processing capacity of traditional manual analysis methods.

To address this, the laboratory is utilizing the Segment Anything Model 3 (SAM 3) and DINOv3 within the SYNAPS-I (SYnergistic Neutron And Photon Science – Intelligence) project. Launched as part of the White House’s 2025 Genesis Mission, SYNAPS-I integrates these models to automate image segmentation—the process of identifying and outlining distinct structures in visual data. DINOv3 provides global structural context, while SAM 3 draws precise, pixel-level boundaries. The team fine-tuned these models on scientific imaging data and deployed them across 300 A100 GPUs at national supercomputing facilities like NERSC to enable real-time processing.

A demonstration of this pipeline focused on grapevine drought resilience. By analyzing micro-CT scans, the system identifies xylem vessels—the plant's water-transporting tubes—and reconstructs 3D volumes. This process, which previously required one month of manual annotation per time step, now delivers fully reconstructed and labeled volumes to researchers in approximately 15 minutes. This speed allows scientists to study dynamic biological processes as experiments occur rather than waiting for post-experiment analysis.

The open-source nature of Meta’s models is critical for the project, as national laboratories must maintain prepublication data and models on secure, government-managed infrastructure rather than external cloud services. SYNAPS-I involves 60 researchers across five national labs, including Argonne, Brookhaven, and Oak Ridge. DOE Under Secretary Dario Gil emphasized that by integrating AI with experimental systems, the initiative shifts scientific inquiry toward adaptive, automated decision-making. As the Genesis Mission expands, the team aims to build a future where intelligent discovery platforms not only process data but also generate hypotheses and transfer knowledge across national facilities.

The Lawrence Berkeley National Laboratory is deploying Meta’s open-source artificial intelligence models to resolve massive data bottlenecks at its Advanced Light Source (ALS) facility. The ALS uses beamlines—instruments producing intense X-ray light—that have seen data generation jump from one image every six seconds to 100,000 images per second following recent hardware upgrades. This volume of information, reaching tens of petabytes annually, exceeds the processing capacity of traditional manual analysis methods.

To address this, the laboratory is utilizing the Segment Anything Model 3 (SAM 3) and DINOv3 within the SYNAPS-I (SYnergistic Neutron And Photon Science – Intelligence) project. Launched as part of the White House’s 2025 Genesis Mission, SYNAPS-I integrates these models to automate image segmentation—the process of identifying and outlining distinct structures in visual data. DINOv3 provides global structural context, while SAM 3 draws precise, pixel-level boundaries. The team fine-tuned these models on scientific imaging data and deployed them across 300 A100 GPUs at national supercomputing facilities like NERSC to enable real-time processing.

A demonstration of this pipeline focused on grapevine drought resilience. By analyzing micro-CT scans, the system identifies xylem vessels—the plant's water-transporting tubes—and reconstructs 3D volumes. This process, which previously required one month of manual annotation per time step, now delivers fully reconstructed and labeled volumes to researchers in approximately 15 minutes. This speed allows scientists to study dynamic biological processes as experiments occur rather than waiting for post-experiment analysis.

The open-source nature of Meta’s models is critical for the project, as national laboratories must maintain prepublication data and models on secure, government-managed infrastructure rather than external cloud services. SYNAPS-I involves 60 researchers across five national labs, including Argonne, Brookhaven, and Oak Ridge. DOE Under Secretary Dario Gil emphasized that by integrating AI with experimental systems, the initiative shifts scientific inquiry toward adaptive, automated decision-making. As the Genesis Mission expands, the team aims to build a future where intelligent discovery platforms not only process data but also generate hypotheses and transfer knowledge across national facilities.