AI breakthrough could replace rare earth magnets in electric vehicles
Scientists at the University of New Hampshire built an AI-powered database of 67,573 magnetic compounds and surfaced 25 materials that remain magnetic at high temperatures. That matters because the strongest commercial magnets depend on rare earth elements that are costly, import-dependent, and hard to secure. This work points to cheaper, sustainable magnets for EVs, generators, and modern electronics.
"By accelerating the discovery of sustainable magnetic materials, we can reduce dependence on rare earth elements, lower the cost of electric vehicles and renewable-energy systems, and strengthen the U.S. manufacturing base," said Suman Itani, a doctoral student in physics. Co-author and physics professor Jiadong Zang added, "We are tackling one of the most difficult challenges in materials science - discovering sustainable alternatives to permanent magnets."
What the team built
- A comprehensive, searchable resource - the Northeast Materials Database - covering 67,573 magnetic compounds.
- 25 newly recognized compounds that keep their magnetism at high temperatures, a key requirement for motors and power systems.
- An AI pipeline that reads papers, extracts experimental data, predicts whether a material is magnetic, and estimates the temperature at which it loses magnetism (Curie temperature).
Why this is a big deal for EVs and energy
Permanent magnets sit at the heart of traction motors, turbines, sensors, and power electronics. Today's top performers rely on rare earths, which add cost and supply risk. A validated set of high-temperature magnetic candidates widens the materials funnel and could reduce rare-earth demand across critical systems.
Despite thousands of documented magnetic compounds, a truly new permanent magnet hasn't emerged from that pool. A data-driven map of the space changes how fast researchers can rule in (or rule out) candidates for device-grade performance.
How the AI workflow works
- Literature ingestion: A large language model reads papers and pulls out experimental measurements and material identifiers.
- Labeling and cleaning: Extracted data is standardized for composition, structure, and magnetic properties.
- Prediction models: Classifiers flag likely magnetic compounds and regressors estimate the critical temperature for losing magnetism.
- Searchable database: Results are unified so researchers can filter by element set, predicted temperature, and application needs.
What you can do with it
- Screen for rare-earth-free candidates that meet your temperature window and processing limits.
- Prioritize synthesis by predicted high-temperature stability and known manufacturability pathways.
- Design experiments that validate magnetism and temperature thresholds first, then move to microstructure and device tests.
- Feed your lab results back into models to tighten predictions for your chemistry domain.
Limits and open questions
- Predictions need experimental confirmation: phase stability, coercivity, corrosion resistance, and scalability remain to be shown.
- Supply and processing: even without rare earths, some elements may face cost or safety constraints.
- Device integration: motor-grade performance depends on grain alignment, anisotropy, and processing routes beyond composition alone.
Beyond materials discovery: education
The team notes the same large language model could support higher education and libraries by converting image-heavy documents into modern rich text formats. That improves access to legacy collections and preserves hard-to-digitize scientific records.
Funding and publication
The study was published in Nature Communications and supported by the Office of Basic Energy Sciences, Division of Materials Sciences and Engineering, U.S. Department of Energy (BES). Co-author Yibo Zhang contributed across physics and chemistry.
If you're building with AI in materials R&D
- For methods, tooling, and case studies: AI for Science & Research
- On paper-reading models and data extraction: Generative AI and LLM
Bottom line: a vetted, AI-built map of magnetic materials shortens the distance between paper data and lab-grade candidates. Faster screening means faster synthesis queues - and a realistic path to rare-earth-free magnets for next-generation EVs and clean energy systems.
Your membership also unlocks:
