From the Research Universities edition of October 28, 2026
PNNL uses a hierarchy of AI agents (SciLink) to combine radiation-damage measurements on chips, with scientists checking results for hallucinations
Pacific Northwest National Laboratory (PNNL), with experimental data from the University of Tennessee · National laboratory materials research and autonomous science · United States (Washington, Tennessee)
- Publisher
- Pacific Northwest National Laboratory
- Original publication
- 26 October 2026 (PNNL news release; republished by HPCwire 27 Oct)
- Source retrieved
- 2026-10-29
What happened
PNNL researchers applied SciLink, a multi-agent system built at the lab, to data from University of Tennessee ion-beam experiments on aluminum gallium nitride and gallium nitride chips. SciLink organizes agents in a hierarchy: three lead agents for experimental analysis, planning and computational simulation delegate to specialist agents that analyze data from methods such as atom probe tomography and scanning transmission electron microscopy. The lead agents then discussed the combined results, drew conclusions and proposed new experiments. PNNL says the agents did the analyses in hours instead of days or weeks, and most conclusions carried a confidence score (for example 90% that ions dominated the observed damage, 60% for proton damage). Scientists designed the experiments, did the lab work and reviewed the agent output for hallucinations and with their own calculations; PNNL states human oversight is critical. The work was presented at an American Chemical Society regional meeting and funded by PNNL.
Why it matters
A concrete national-lab pattern for multi-agent analysis across several instruments, with explicit human checking, that university core facilities and materials labs can compare against their own data-integration bottlenecks.
Evidence and measured results
Read the full release. Results are lab-reported; no paper, accuracy measure or comparison against human analysis was provided.
Limitations and uncertainty
Institutional news release about a conference talk; no peer-reviewed results, no measured accuracy, and the hours-versus-weeks comparison is not quantified. Confidence scores are the system's own and were not shown to be calibrated. Published 26 Oct, just before the last completed run, but not previously archived.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-28; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Labs generate data from several instruments that is slow to combine into one coherent interpretation.
- Stakeholders
- Core facility director
- Materials science department chair
- Research computing lead
- Export-control or research security officer
- Discovery questions
- How long does it take today to combine results from multiple instruments?
- Which instrument data is accessible in structured form?
- Do any datasets carry export-control restrictions?
- Value hypothesis
- A supervised multi-agent analysis layer may shorten time to a combined interpretation while scientists retain judgment.
- Potential engagement
- A pilot reanalyzing past multi-instrument experiments with agents and comparing against expert conclusions.
- Unsupported claims
- Measured speedups
- Accuracy of agent conclusions
- Calibrated confidence scores
- Unsupervised operation
Pre-sales engineering
Role takeaway
- Fit
- Fits materials and characterization labs with digital outputs from several instruments.
- Architecture and integration
- Lead agents for analysis, planning and simulation delegating to instrument-specific agents, with confidence scores and logs returned to scientists.
- Prerequisites
- Structured instrument data
- Simulation tool access
- Model service approved for the data class
- Constraints
- No published evaluation
- Possible export-control limits
- In-house PNNL system not described as available
- Security
- Classify datasets before agent access and keep sensitive data on approved local infrastructure.
- Proof of value
- Compare agent conclusions with expert analysis on archived experiments
- Check calibration of confidence scores
- Time the analysis against the current process
Delivery
Role takeaway
- Work
- Connect instrument data, configure agents, run on past experiments, collect expert review and report agreement and time.
- Dependencies
- Instrument data access
- Expert reviewer time
- Compute for agents and simulation
- Ownership
- Core facility or lab group lead with research computing support.
- Skills and adoption
- Scientists learn to direct agents and review their conclusions for hallucinations.
- Governance checkpoints
- Data classification review
- Expert sign-off on conclusions
- Pilot review before use on new experiments
- Acceptance criteria
- Agreement with expert conclusions on a test set (proposal)
- All agent conclusions reviewed by a scientist
- Analysis time measured
- Risks
- Hallucinated conclusions accepted
- Overconfidence in uncalibrated scores
- Sensitive data exposure
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?
A hierarchical agent design maps agents to instruments and to planning and simulation roles, requiring access to heterogeneous instrument data.
Governance
Who approves, reviews and stays accountable for outcomes?
Human verification of agent conclusions is presented as required, which suits lab policies that keep scientists accountable for findings.
Security and privacy
What data, permissions and controls need testing?
Radiation-hardening work for satellites and reactors may involve export-controlled or sensitive data; the release does not discuss data controls.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Shifts scientist time toward checking and directing analysis; requires training in reviewing agent output.
Procurement
What should contracts, pricing and exit terms secure?
Built in-house at PNNL; universities would need comparable tooling or open releases.
Operating model
Which teams own the service once it runs?
Agents act as an integration layer across instrument teams, with scientists retaining experiment design and final judgment.
Publication history
- 2026-10-28Research Universities · Issue 505 resources
Stable resource ID: pnnl-scilink-agent-team-rad-hard-chip-analysis-20261026