US system to cut nuclear fusion simulation time from months to real-time

The Princeton Plasma Physics Laboratory (PPPL) has introduced STELLAR-AI, a new computing platform designed to drastically reduce the time required for nuclear fusion simulations from months to real-time. By integrating artificial intelligence (AI) with high-performance computing, STELLAR-AI connects computing resources directly to experimental devices, enabling real-time data analysis during fusion experiments. The platform’s hardware architecture combines CPUs for standard tasks, GPUs for AI model training, and quantum processing units (QPUs) to handle complex calculations beyond the capabilities of traditional computers. A key experimental partner is the National Spherical Torus Experiment-Upgrade (NSTX-U), which will benefit from a digital twin model to simulate experiments virtually before physical testing. STELLAR-AI supports the U.S. Department of Energy’s Fusion Science and Technology Roadmap, aiming to accelerate the commercialization of fusion power plants through AI-driven design and optimization. Projects under this initiative include StellFoundry, which uses AI to speed up the design of stellarators