Multi-institutional team selected for three-year grant program; will develop machine learning platform to address high-priority research related to the design and operation of fusion pilot plants
San Diego, Sept. 1. – The U.S. Department of Energy (DOE) has selected a multi-institutional team of data scientists led by General Atomics (GA) to develop a Fusion Data Platform (FDP) for advancing high-priority fusion research. The FDP will provide researchers with access to high-quality fusion data for the creation of reproducible Artificial Intelligence (AI) and Machine Learning (ML) models. These advanced models will support the design and operation of Fusion Pilot Plants (FPPs) within a decadal timescale.
GA is partnering with the San Diego Supercomputer Center (SDSC) at the University of California, San Diego (UC San Diego), Hewlett Packard Enterprise (HPE), and Sapientai to create the FDP on behalf of the U.S. Department of Energy. Once complete, the tool will be made available to the scientific community to support rapid advancements in FPP designs.
“Creating a robust AI/ML platform with very large, curated datasets and efficient processing tools will be transformational for fusion energy,” said Brian Sammuli, Deputy Director of the Advanced Computing Center of Excellence at General Atomics and Principal Investigator. “By advancing AI/ML research in fusion, we will be able to rapidly address some key remaining challenges in fusion science and reactor development. We look forward to leading this team to provide an outstanding platform for the scientific community to advance fusion research and support the deployment of the first generation of fusion energy power plants.”
“A key mission of the FDP is to accelerate AI/ML research by expanding access to high-quality fusion data and the tools needed to process the data at scale,” said Dr. Raffi Nazikian, Senior Director and leader of the ITER Research Hub at General Atomics. “The FDP will include experimental and simulated data in an integrated platform. We are talking many petabytes of data that will be easily accessible on the platform. The success of the FDP will be measured by how well we serve the needs of the fusion and broader data science community, including students and researchers from universities, national laboratories, and industry.”
“The FDP is an important step towards harnessing the power of fusion data to advance the development of fusion energy. GA and SDSC have a long history going back almost 40 years, and this is the beginning of a new chapter in our cooperation to advance fusion energy science and education,” said Prof. Frank Würthwein, Director of SDSC.
“Among the FDP unique capabilities will be the ability for users to access, understand and leverage prior data and AI pipelines to advance their research and build reproducible, certifiable AI/ML models, said Paolo Faraboschi, HPE Fellow and AI Research Lab Director at Hewlett Packard Labs. “We fully understand the unprecedented societal impact that fusion power can one day provide, and we look forward to working with the scientific community on the FDP to help realize the decadal vision for fusion energy development.”
“We’re very excited to participate in the FDP project,” said Craig Michoski, Founder and CEO of Sapientai. “This is a phenomenal set of collaborative institutions, and we have high aspirations for the success and impact the FDP project will have across the fusion landscape. We think the era of data-driven science and technology advancement is well upon us, and we are extremely excited to see how these tools applied to the treasure trove of DOE’s fusion data can advance the field and accelerate progress towards commercial fusion energy.”
Supporting Data-Informed FPP Designs
To achieve fusion conditions relevant for energy production, an FPP must sustain plasmas at temperatures exceeding 100 million degrees Celsius—approximately ten times the temperature at the center of the sun. In magnetic confinement fusion, plasmas are controlled using powerful electromagnets that shape and confine the superheated gas. At such extreme temperatures, the plasmas may exhibit instabilities that may cause them to momentarily breach the magnetic fields and interact with the inner walls of the fusion machine, which could decrease efficiency or even cause damage. Successfully designing FPPs that account for these and other types of instabilities requires robust data sets to model and predict plasma behaviors across designs.
The FDP will help address this need by making large-scale fusion data easier to access and analyze. The multi-institutional team will draw from its significant AI/ML industry expertise to develop the FDP as a resource capable of being collectively utilized across distributed computational facilities.
The FDP will leverage GA’s scalable, fusion-specific data processing tool, TokSearch, to process and curate the data sets at the required scale. The team will also draw from HPE’s Common Metadata Framework to create reproducible workflows that include metadata tracking, source code integration, and data version control. A publishing portal will be incorporated into the system to facilitate search and discovery of these curated datasets. A suite of AI/ML modeling capabilities developed by SapientAI and UCSD will be integrated with the platform, allowing it to serve as a powerful data and analysis tool that meets the growing needs of the fusion science community.
The FDP will initially be deployed at SDSC at UC San Diego. Once completed, the research team will utilize the FDP to develop new ML/AI models that address high-priority research areas relevant to a broad range of FPP designs and plasma configurations.
About the Team
The SDSC was established in 1985 as one of the nation’s first supercomputer centers under a cooperative agreement by the National Science Foundation in collaboration with UC San Diego and GA. The SDSC provides resources, services, and expertise to the national research community, including industry and academia, and features the Expanse and Voyager supercomputers. Expanse supports SDSC's theme of “Computing without Boundaries” with a data-centric architecture, public cloud integration, and state-of-the art GPUs for incorporating experimental facilities and edge computing. The first-of-its-kind experimental system with training and inference accelerators to provide high-performance, high-efficiency AI compute, Voyager supports AI research across a range of science and engineering domains.
Hewlett Packard Enterprise is the global edge-to-cloud company that helps organizations accelerate outcomes by unlocking value from all their data, everywhere. Built on decades of reimagining the future and innovating to advance the way people live and work, HPE delivers unique, open, and intelligent technology solutions as a service.
Sapientai LLC combines ML and AI with data-intensive science, notably in nuclear fusion and plasma physics. They provide versatile software solutions, including off-the-shelf applications as well as tailored services. With a firm belief in collaboration, Sapientai encourages innovative research partnerships. Their work aligns with the Department of Energy's mission, committed to advancing scientific frontiers.
About General Atomics: Since the dawn of the atomic age, General Atomics innovations have advanced the state of the art across the full spectrum of science and technology – from nuclear energy and defense to medicine and high-performance computing. Behind a talented global team of scientists, engineers, and professionals, GA’s unique experience and capabilities continue to deliver safe, sustainable, economical, and innovative solutions to meet growing global demands.
For more information contact: Evan Polisar, Strategic Communications Manager – 858-455-3474 – Evan.Polisar@ga.com