Lawrence Livermore National Laboratory (LLNL) researchers have reached a significant milestone by integrating artificial intelligence (AI) with fusion target design. They deployed AI agents on two of the world’s most powerful supercomputers to automate and accelerate inertial confinement fusion (ICF) experiments. Part of an AI framework called the Multi-Agent Design Assistant (MADA), LLNL scientists and collaborators combine Large Language Models (LLMs) with advanced simulation tools. This system interprets natural language prompts from human designers and generates complete physics simulation decks for LLNL’s next-generation 3D multiphysics code, MARBL. MARBL excels at designing and analyzing high-energy-density experiments, including ICF. In ICF experiments at LLNL’s National Ignition Facility (NIF), fusion energy is produced when 192 laser beams converge on a tiny target of deuterium and tritium, triggering a fusion chain reaction. The MADA team uses the exascale El Capitan supercomputer—the world’s fastest at 2.79 exaFLOPs peak—and its smaller sibling Tuolumne to test the AI system. The framework includes an Inverse Design Agent (IDA) that engineers new ICF targets. Jon Belof, LLNL physicist and principal investigator, notes the project began in 2019, initially exploring the combination of AI with shockwave physics. As LLMs advanced, semi-autonomous AI systems collaborating with humans for ICF design became a natural progression. The MADA team, including collaborators from Los Alamos (LANL) and Sandia National Laboratories (SNL), has transformed this concept into a sophisticated AI-driven design workflow. Recently, an open-source LLM fine-tuned on MARBL documentation successfully converted a hand-drawn capsule diagram and natural language request into a full simulation deck, running thousands of simulations to explore new ICF capsule geometries. This AI-driven design approach comes at a critical time for fusion research. Following LLNL’s historic ignition achievement at NIF in December 2022, the lab aims to develop a robust ignition platform with potential national security applications. Belof explains that MADA drastically compresses design cycles and explores vast design spaces, enabling the pursuit of hundreds or thousands of ICF design concepts simultaneously. This shift from human-run simulation ensembles to AI-driven idea ensembles could be transformative. At the core of MADA are AI “agents”—autonomous software entities with two main components: an LLM that understands human language and a specialized tooling interface that performs domain-specific tasks. For MADA, this tooling generates structured simulation input files and launches them on high-performance computing (HPC) systems. “We are putting AI in the driver’s seat of a supercomputer, which has never been done before,” said Belof. Supporting the IDA is the Job Management Agent (JMA), which manages large-scale simulation workflows across LLNL’s supercomputers. It interacts with the Flux scheduler and workflow tools like Merlin to queue jobs, allocate resources, and efficiently collect simulation outputs. JMA Team Lead Giselle Fernandez highlights that this coordination between AI and HPC offers a critical advantage in advancing toward a robust fusion ignition platform. This iterative workflow enables unprecedented interactivity between designers and simulations. Instead of manually coding and launching jobs—a process that could take days or weeks—researchers can explore thousands of design variations in parallel by conversing with an AI agent. Belof elaborates, “The agent can take a capsule diagram and a plain-language prompt like, ‘Explore the effect of changing a certain part of the geometry,’ translate that into a valid MARBL simulation deck, run it, collect results, and build training datasets for surrogate models.” MADA leverages HPC to run massive ensembles—tens of thousands of ICF simulations in a single study—on Tuolumne, the world’s 12th fastest supercomputer. The simulation outputs train a machine learning model called PROFESSOR, which provides instant feedback for designers exploring new capsule geometries. “Once trained, PROFESSOR generates implosion time histories that update instantaneously when designers change input geometry,” said Belof. “It’s a powerful new tool enabled by AI, machine learning, and HPC.” By integrating natural language interaction, image interpretation, and rapid simulation-to-model pipelines, MADA demonstrates how AI can be embedded directly into high-stakes scientific workflows. This collaborative AI augmentation replaces slow, manual iteration with accelerated design cycles. The implications extend beyond ICF. As more exascale systems like El Capitan come online, MADA offers a blueprint for AI agents acting as digital collaborators in fields ranging from materials discovery to weapons certification. Belof concludes, “It’s about enhancing human productivity through AI in transformative ways. This project shows we’re just beginning to tap AI’s potential to optimize resources and understand tradeoffs for next-generation fusion facilities.” The work is funded by NNSA’s Advanced Simulation & Computing program. LLNL MADA team members include deputy principal investigator Charles Jekel, MARBL Project Lead Rob Rieben, and researchers Will Schill, Meir Shachar, and Dane Sterbentz. Nathan Brown of SNL and Ismael Djibrilla Boureima of LANL also contributed.

Source: Originally published at Newswise on August 4, 2025.

Frequently Asked Questions (FAQ)

AI in Fusion Target Design

Q: What is the primary goal of integrating AI into fusion target design at LLNL? A: The primary goal is to automate and accelerate inertial confinement fusion (ICF) experiments by deploying AI agents on supercomputers for advanced target design and simulation. Q: What is the MADA framework? A: MADA stands for Multi-Agent Design Assistant. It's an AI framework developed by LLNL researchers that combines Large Language Models (LLMs) with simulation tools to interpret natural language prompts and generate physics simulation decks. Q: How does MADA work in practice? A: MADA uses AI agents, including an Inverse Design Agent (IDA), to interpret human design inputs (like diagrams and natural language) and generate simulation decks for codes like MARBL. These agents can then run simulations to explore design variations. Q: Which supercomputers are being used for this research? A: The research utilizes two powerful supercomputers: El Capitan, the world's fastest at 2.79 exaFLOPs peak, and its sibling, Tuolumne. Q: What are the key components of an AI agent within the MADA framework? A: Each AI agent consists of a Large Language Model (LLM) for understanding human language and a specialized tooling interface for performing domain-specific tasks, such as generating simulation input files. Q: How does MADA help compress design cycles? A: MADA drastically compresses design cycles by allowing researchers to explore thousands of design concepts simultaneously through AI-driven workflows, replacing slower, manual simulation processes. Q: What is the role of the Job Management Agent (JMA)? A: The JMA manages large-scale simulation workflows across LLNL's supercomputers, interacting with schedulers and workflow tools to queue jobs, allocate resources, and collect outputs efficiently. Q: How does the PROFESSOR model contribute to the design process? A: PROFESSOR is a machine learning model trained on simulation outputs. Once trained, it provides instant feedback to designers on implosion time histories as they modify capsule geometries, enabling rapid iteration. Q: What are the broader implications of this AI-driven approach beyond fusion research? A: The MADA framework offers a blueprint for AI agents to act as digital collaborators in various scientific fields, including materials discovery and weapons certification, as more exascale systems become available.

Crypto Market AI's Take

The integration of advanced AI agents and supercomputing power for complex scientific simulations, as demonstrated by LLNL's MADA project, mirrors the rapid advancements we're seeing in the application of AI within the financial markets. Just as AI is accelerating fusion target design by automating complex tasks and exploring vast parameter spaces, AI-powered tools are revolutionizing cryptocurrency trading. At Crypto-Market.AI, we leverage similar principles to develop sophisticated AI trading bots and AI analysts that process market data, identify patterns, and execute trades with unprecedented speed and efficiency. This parallels the LLNL approach of using AI agents to manage complex workflows, enabling our users to navigate the volatile crypto landscape with data-driven strategies and enhanced productivity.

LLNL Pushes Frontier of Fusion Target Design with AI

Frequently Asked Questions (FAQ)

AI in Fusion Target Design

Crypto Market AI's Take

More to Read: