AI Is Rewriting How We Design and Run Particle Accelerators
Particle accelerators sit at the core of discovery in physics, chemistry, materials science, and biology. They also deliver practical wins in medicine, national security, and manufacturing. The Multi-Office particle Accelerator Team (MOAT) is bringing artificial intelligence into this stack to optimize design and operations, with a simple goal: more science per hour, fewer bottlenecks.
This effort is part of the Department of Energy's Genesis Mission, a national initiative to advance AI for science, energy, and national security. A key pillar is the Transformational AI Models Consortium (ModCon), which will build self-improving models using DOE data, facilities, and expertise.
What MOAT Will Deliver
MOAT will expand digital twins, intelligent assistants, and advanced AI models to simulate complex accelerator physics and day-to-day operations. These tools will be platform agnostic so improvements can move across national labs, universities, and industry without friction.
The project taps into extensive experimental data, simulations, and expert knowledge from DOE Office of Science accelerators and light sources-facilities that account for half of the agency's user infrastructure. That scale matters for training and validating AI systems that have to work under real-world constraints.
Early Results: An AI Assistant at the Advanced Light Source
At Berkeley Lab's Advanced Light Source (ALS), a large language model-based "Accelerator Assistant" demonstrated that an AI system can autonomously prepare and run a multi-stage physics experiment on a synchrotron light source. Setup ran roughly 100x faster than human-only workflows, even with a system that has hundreds of thousands of variables.
Engineers describe their goal in natural language. The AI locates relevant variables, generates and executes analysis code, visualizes results, and interacts with the accelerator within safety boundaries.
"By making intelligent AI tools that continuously learn and work across facilities and fields, we're accelerating discoveries that particle accelerators can make in key applications such as fundamental physics, fission and fusion energy, advanced materials, and advanced medical technologies," said Jean-Luc Vay, MOAT's lead and head of the Advanced Modeling Program in the Accelerator Technology & Applied Physics (ATAP) Division at Berkeley Lab.
"We're witnessing the emergence of a new layer of scientific infrastructure: AI systems that can interpret intent, plan actions, and safely operate complex instruments," said Thorsten Hellert, ATAP staff scientist and lead author on the recent milestone. "This approach can accelerate experiments at the ALS today and form the foundation for connected, AI-enabled facilities across the DOE complex."
Why This Matters for Your Research
- Throughput: Faster experiment setup and tuning means more beam time for science.
- Reproducibility: Shared AI workflows reduce drift across runs, instruments, and sites.
- Safety and stability: Guardrails keep operations within validated limits while exploring optimal settings.
- Portability: Platform-agnostic tools move across labs and vendors without heavy rework.
From Pilot to Network
MOAT's first phase is about generalizing the ALS framework and proving it works across multiple accelerator facilities. The team is demonstrating cross-site interoperability and shared AI infrastructure so gains at one site can propagate to the rest.
Collaborators currently include Argonne, Brookhaven, Fermi, Jefferson, Oak Ridge, and SLAC national laboratories, along with industry partners. The objective is a connected ecosystem where models, digital twins, and assistants evolve with each facility's feedback and operating data.
How It Fits Into the Genesis Mission
Genesis is the DOE's push to apply AI where it moves the needle on science and national priorities. Through ModCon, the mission will develop self-improving AI models grounded in DOE-grade data, facilities, and expert oversight-exactly the environment needed for safe deployment on complex instruments.
For background on the DOE Office of Science and its user facilities, see the official overview at energy.gov/science.
About Berkeley Lab
Lawrence Berkeley National Laboratory advances discovery science and solutions for reliable, abundant energy. Founded in 1931 and managed by the University of California for the DOE Office of Science, the lab supports researchers worldwide through its user facilities and has been recognized with 17 Nobel Prizes.
If You're Building AI Skills for Lab Work
For curated training that can help technical teams get fluent with modern AI tools and workflows, explore these AI courses.
Your membership also unlocks:
