Lawrence Berkeley National Laboratory (LBNL), a flagship research lab of the U.S. Department of Energy (DOE) known for Nobel Prize–winning work, operates the Advanced Light Source (ALS), a large X‑ray facility whose beamlines produce huge volumes of imaging data. Recent upgrades have dramatically increased detectors’ resolution and speed, and the resulting data deluge has outpaced traditional, manual analysis methods.
Key figures
- DOE light and neutron source facilities now generate tens of petabytes of data per year — equivalent to millions of gigabytes or roughly 2 million hours of HD video.
- Detector throughput has gone from one image every six seconds to as much as 100,000 images per second.
The challenge is not only data volume: domain experts are limited and overburdened, and modern in‑situ experiments — which observe dynamic processes such as chemical reactions or material failures as they happen — require near real‑time interpretation that human teams cannot provide manually. Central to the analysis workload is segmentation: identifying and drawing precise boundaries around distinct structures in images so that raw X‑ray frames become labeled maps of meaningful features.
SYNAPS-I within The Genesis Mission
In late 2025 the White House launched The Genesis Mission, a DOE‑led initiative to accelerate scientific discovery and technological leadership using advanced AI. SYNAPS-I (SYnergistic Neutron And Photon Science – Intelligence), working with Argonne, Brookhaven, Oak Ridge, and other laboratories, seeks to turn months‑long analysis bottlenecks in X‑ray and neutron science into real‑time discovery workflows. Image segmentation, where extracting structures from experimental data could previously consume weeks of expert effort per dataset, is a major target.
SAM 3 and DINOv3 form the analysis pipeline
SYNAPS-I’s segmentation pipeline centers on two open‑source foundation models from Meta: Segment Anything Model 3 (SAM 3) and DINOv3.
- DINOv3 is a self‑supervised vision model that learns visual structure from raw images without human labels and provides global context to identify different structures.
- SAM produces precise pixel‑level boundaries around objects in an image, effectively automating the expert’s manual outlining.
The two models complement each other: DINO supplies contextual identification, while SAM delivers accurate boundaries. SYNAPS-I fine‑tuned both models on scientific imaging data gathered at DOE beamlines and deployed them on national supercomputing resources — including NERSC — running across 300 A100 GPUs.
From data to labeled 3D volume in minutes
As deployed, the pipeline returns a fully reconstructed, semantically labeled 3D volume to the scientist at the beamline while the experiment is still running. End‑to‑end turnaround is approximately 15 minutes. That represents a reduction from what formerly required about a month of expert annotation per time step to a quarter‑hour processing window, enabling scientists to follow dynamic processes at acquisition speed.
Demonstration: tracking drought responses in grapevines
The team demonstrated the pipeline on a practical agricultural problem: mapping how grapevine tissues respond to drought at the cellular level. Using micro‑CT scans collected at the Advanced Light Source, the system reconstructs 3D volumes of vine stems and automatically identifies xylem vessels — the microscopic tubes that transport water inside the plant. Tracking changes in these vessels over time provides insights useful for developing drought‑resilient crops.
Security, open source, and the path forward
National laboratories keep prepublication research data and AI models on government infrastructure rather than external cloud services; in‑progress work must run on secure platforms. Meta’s open‑source release of SAM and DINO allowed SYNAPS-I to download, fine‑tune, and deploy the models inside their secure computing environments, adapting models trained on natural images to scientific domains.
SYNAPS-I involves about 60 researchers across five national labs and aims to turn user facilities into intelligent discovery platforms: systems that not only process data faster but also help generate hypotheses, recommend next experiments, and transfer knowledge across sites. As The Genesis Mission expands from seed projects to broader programs, that open foundation is positioned to accelerate discovery across multiple national priorities.
At the Trillion Parameter Consortium, DOE Under Secretary Dario Gil summarized the effort: “By seamlessly combining AI, advanced computing, and experimental systems, SYNAPS-I analyzes data as it's produced and guides experiments in real time, replacing slow manual steps with adaptive, automated decision-making. This compresses discovery time from days to moments and establishes a continuous, self-improving model of science that will be essential to realizing the full potential of the Genesis Mission.”
