UNH’s AI pipeline speeds up discovery of rare-earth-lean magnets for decarbonisation

Researchers at the University of New Hampshire have assembled an AI-driven pipeline and a large experimental database intended to accelerate the search for permanent-magnet materials that could reduce reliance on constrained rare-earth elements critical to decarbonisation technologies.

The project produced the Northeast Materials Database, a searchable compilation of 67,573 experimentally reported magnetic entries that integrates composition, magnetic phase-transition temperatures, structural details and other measured properties. Machine-learning classifiers trained on the dataset can distinguish ferromagnetic, antiferromagnetic and non-magnetic materials with about 90% accuracy, according to a paper reported in Nature Communications. The models also provide estimates of the temperatures at which candidate materials lose their magnetism, an essential metric for applications such as electric-vehicle motors and offshore-wind generators. ScienceDaily and the University of New Hampshire note that the database flagged 25 compounds not previously recognised as retaining magnetism at elevated temperatures.

The platform couples automated literature reading with structured data extraction. Large language models parse published papers to pull composition and experimental parameters; those entries are organised into a continuously updatable resource that supports downstream machine-learning screening. According to UNH researchers, the automation is intended to shorten the early-stage screening cycle that otherwise would require impractical laboratory testing across millions of chemical permutations.

“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 UNH physics Ph.D. student and lead author. UNH postdoctoral researcher Yibo Zhang added that the same LLM-based tooling could be applied more broadly in academia, for example to convert legacy images into modern rich-text formats and modernise library holdings.

The work has attracted federal and industry-aligned support. ScienceDaily reports the project received backing from the U.S. Department of Energy’s Office of Basic Energy Sciences, Division of Materials Sciences and Engineering. Separately, a UNH researcher involved in related AI-for-science initiatives is a sub-award recipient from a $152 million National Science Foundation–NVIDIA partnership to advance AI for scientific research, a programme that will also fund interface and community work linked to materials-discovery tools.

The research underscores two practical aims for industry: diversifying supply chains away from expensive, geopolitically sensitive rare-earth magnets and reducing development time and cost for magnets suitable for high-temperature and high-frequency service. Sandia National Laboratories and other agencies have described similar AI-driven efforts to shorten the design and synthesis cycle for novel magnetic materials used in resilient grid and power-conversion hardware.

For industrial decarbonisation stakeholders, the principal value lies in better prioritisation of laboratory resources. By surfacing experimentally grounded candidates and predicting critical operating temperatures, the UNH database and its models can help manufacturers and materials suppliers focus synthesis and testing on the most promising compositions, potentially accelerating deployment of lower-cost, rare-earth-lean magnets in vehicles, turbines and power electronics.

The UNH team presents the database and accompanying machine-learning models as an evolving infrastructure: the automated pipeline is intended to ingest new literature as it appears, enabling continual refinement of predictions and discovery efforts.