From the Research Universities edition of November 4, 2026
SYNAPS-I: Intelligent Computing Transforms X-ray Imaging, from Data Collection to Discovery
Brookhaven National Laboratory (NSLS-II), with Berkeley Lab ALS, Argonne, Oak Ridge and SLAC; DOE Genesis Mission ModCon seed project · National laboratory user facilities (AI-driven experimental workflows) · United States (multi-laboratory DOE)
- Publisher
- Brookhaven National Laboratory Newsroom
- Original publication
- 4 November 2026 (BNL newsroom)
- Source retrieved
- 2026-11-05
What happened
Brookhaven reports SYNAPS-I demonstrations at NSLS-II. One automated experiment analyzed images as acquired and autonomously selected scientifically interesting regions for high-resolution measurement without human intervention. A foundation model trained at Argonne replaced iterative ptychographic reconstruction and, integrated with NVIDIA Holoscan, ran real-time reconstruction at 200 images per second. A model trained at NSLS-II reconstructed a microchip from nearly one tenth of the usual data, and fine-tuned Meta SAM3 segmented the 3D volumes. Next steps are real-time tomography and more general autonomous workflows.
Why it matters
Relevant to research universities that use DOE light sources and to campus imaging cores: AI is shifting beamlines from collect-then-analyze to data-guided experiments, with implications for user data pipelines and GPU needs. Limited direct relevance outside imaging-intensive research.
Evidence and measured results
Laboratory press release describing demonstrations; no error rates, validation against conventional reconstruction, or beam-time savings quantified beyond the 200 images per second and roughly one-tenth data claims. NVIDIA and Meta tools named. Classified as a self-reported institutional claim.
Limitations and uncertainty
Demonstrations only; quantitative claims self-reported; sample types (thin, flat microchips) favour the reduced-data approach; generalization unproven.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-11-04; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Imaging experiments produce data faster than staff can reconstruct and analyze, wasting scarce instrument time.
- Stakeholders
- Imaging core director
- Research computing director
- Beamline users and PIs
- Facility data science lead
- Discovery questions
- Where are reconstruction bottlenecks in your imaging workflows?
- Is GPU capacity available near instruments?
- How would learned reconstructions be validated?
- Who decides when models select measurement regions?
- Value hypothesis
- Streaming learned reconstruction may cut data collection and analysis time for suitable sample types, as BNL reports for microchips.
- Potential engagement
- A workflow assessment for one imaging instrument covering streaming reconstruction, data paths and validation.
- Unsupported claims
- Beam-time or cost savings
- Accuracy across sample types
- Autonomous experiments without human oversight
Pre-sales engineering
Role takeaway
- Fit
- Strong for ptychography and tomography with reconstruction bottlenecks.
- Architecture and integration
- Edge GPU streaming inference, foundation reconstruction models, segmentation models, facility data infrastructure.
- Prerequisites
- Edge GPUs
- High-bandwidth data paths
- Trained models and validation datasets
- Constraints
- Demonstrations on thin, flat samples
- Self-reported metrics
- Export control for some samples
- Security
- Control access to sample data; separate proprietary user data; log model versions used per experiment.
- Proof of value
- Compare learned vs iterative reconstruction fidelity
- Measure data reduction on local samples
- Audit autonomous region selection against expert choices
Delivery
Role takeaway
- Work
- Integrate streaming reconstruction; Validate models and expose user controls
- Dependencies
- Facility data infrastructure
- GPU procurement
- Model training partners
- Ownership
- Facility data science group with beamline or core staff.
- Skills and adoption
- Staff need ML operations and validation skills; users need guidance on model-guided experiments.
- Governance checkpoints
- Model validation before scientific use
- User consent for autonomous region selection
- Data-handling review for proprietary samples
- Acceptance criteria
- Proposed: reconstruction fidelity documented versus baseline
- Proposed: data reduction measured per sample type
- Proposed: model version recorded for every dataset
- Risks
- Model drift across sample types
- Missed features in unselected regions
- Vendor platform dependence
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Edge streaming reconstruction (NVIDIA Holoscan) at the beamline, foundation models trained at partner labs, integration with facility data infrastructure; moves compute into the experiment loop.
Governance
Who approves, reviews and stays accountable for outcomes?
Autonomous selection of measurement regions shifts decisions from users to models; facilities need validation and user consent norms. Not discussed.
Security and privacy
What data, permissions and controls need testing?
Microelectronics samples may be proprietary or export-controlled; data handling not described.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Facility staff need ML engineering and data infrastructure skills; collaboration across labs is emphasized.
Procurement
What should contracts, pricing and exit terms secure?
Named NVIDIA Holoscan and Meta SAM3; facilities should plan for edge GPU purchases and open-weight model use.
Operating model
Which teams own the service once it runs?
Multi-lab collaboration model with facility data science teams integrating tools for users.
Publication history
- 2026-11-04Research Universities · Issue 573 resources
Stable resource ID: doe-synaps-i-intelligent-xray-imaging-nsls2-20261104