This article covers a fascinating study from the University of New Hampshire. Researchers there used large language models and custom parsing tools to build NEMAD, a public database that maps magnetic materials at scale.
They pulled data from 100,000 article identifiers across major journals. That’s how they created a Curie-temperature–focused resource that might speed up the hunt for high-temperature ferromagnets and help move magnets beyond rare-earth elements in clean-energy tech.
What is NEMAD and why it matters
NEMAD (the public magnetism database) holds 67,573 entries spanning 84 elements and 15 material features tied to magnetism. The team harvested data from 100,000 article identifiers in Elsevier and American Physical Society journals, capturing magnetic, chemical, and structural details on a scale no single lab could hope to match.
The core of NEMAD is the Curie temperature—that’s the point where materials lose magnetic order. This number matters a lot for magnets in electric vehicle (EV) motors and wind-turbine generators, which run hot and need to stay reliable. By organizing this info, NEMAD helps guide researchers toward materials that can handle real-world heat and stress.
AI-enabled curation and Curie-temperature prediction
The team trained large language models and parsing tools to classify magnetic ordering and predict Curie temperatures. With this, they built a scalable, detailed map of magnetic materials that can point experimentalists toward promising directions—and help them skip the dead ends.
Key performance metrics
- Magnetic ordering classifier accuracy: ~90%
- Curie-temperature predictor: R² = 0.87; MAE = 56 K
- They flagged 32 high-temperature candidates, like GaFe2Co4Si, which has a predicted Curie temperature near 1,005 K (~732 °C)
- Out of those, seven high-probability compounds have experimental ordering temperatures reported in the literature. The other 25 still need to be tested in the lab.
Right-sized data to shrink rare-earth dependence
Most permanent magnets depend on rare-earth elements, and that’s a big supply-chain headache for EVs and wind energy. This project gives researchers a fast, data-driven way to spot robust candidates that use less rare earth—something the field really needs right now.
Industry and policy data make the stakes clear. According to the IEA, the top three refining nations controlled about 86% of the market by 2024. China alone could supply around 80% of refined rare earths by 2035. With recycling gains limited and supply shocks always possible, finding new magnets that need fewer rare earths feels like a smart move for energy security and costs.
Implications for industry and research
AI won’t ever replace the lab, but this LLM-based approach works like a high-speed map, pointing experimentalists toward promising, lower-risk magnet candidates. That could help diversify supply chains, reduce geopolitical and price risks, and maybe lower clean-energy costs—once we figure out scalable synthesis and validation.
What researchers and policymakers can take away
- The NEMAD framework speeds up discovery by highlighting magnets with the most promise for experiments.
- It helps move us away from relying so much on rare-earth elements, which could make EVs and wind-energy systems more resilient.
- To really see the benefits, scientists need to confirm results in the lab and figure out how to make new compounds at scale—and in ways that won’t hurt the planet.
This study shows that AI-assisted data mining and predictive modeling can work hand-in-hand with lab research. NEMAD turns scattered literature into a searchable catalog, which could seriously cut down development times and open up new options for magnets. Maybe it’ll even help us get to cheaper, more reliable magnets for future clean-energy tech—here’s hoping.
Here is the source article for this story: It’s not about lithium or batteries: the problem driving up the cost of electric cars and wind power might lie in a tiny magnet, and a new AI has already found a way to do without rare earth elements