DIII-D Integrates AI and Supercomputers to Analyze Fusion Shots in 10 to 20 Minutes
Updated
Updated · HPCwire · Jul 15
DIII-D Integrates AI and Supercomputers to Analyze Fusion Shots in 10 to 20 Minutes
2 articles · Updated · HPCwire · Jul 15
Summary
DIII-D and Argonne’s ALCF have linked live fusion experiments with remote supercomputing so researchers can analyze plasma shots during the 10-to-20-minute gap before the next run.
That setup uses on-demand queues, automated job execution and secure data transfers to run heavy workflows such as CAKE and IonOrb on DOE supercomputers and return results fast enough to guide machine settings.
A parallel effort is building an AI digital twin of DIII-D using ALCF and NERSC systems, with surrogate models delivering feedback in milliseconds instead of the 30 minutes to 1 hour needed for full CAKE runs.
The 25-year Argonne-DIII-D collaboration now feeds broader DOE programs including the Genesis Mission and American Science Cloud, aiming to connect experiments, data platforms and exascale computing into a unified fusion research system.
Is this project creating an AI scientist that will revolutionize all complex research forever?
With global rivals also building fusion digital twins, can America's AI strategy win the race to clean energy?
Can an AI truly master the chaotic physics of a star in a jar and prevent catastrophic plasma events?
DIII-D’s AI-Driven Superfacility: Real-Time Plasma Analysis and the Future of Fusion Research
Overview
The DIII-D National Fusion Facility has recently advanced fusion research by integrating its experiments with powerful DOE supercomputers at ALCF and NERSC. This collaboration accelerates fusion experiment data analysis, leading to faster and more accurate plasma analysis. The joint effort between DIII-D and ALCF researchers sets a new standard for cross-facility science under the DOE’s Integrated Research Infrastructure program. By enabling on-demand high-performance computing during experiments, this integration transforms research workflows, allowing scientists to gain immediate insights and paving the way for more dynamic and responsive fusion research.