Artificial intelligence (AI) is moving beyond making drawings or photos to being used as a tool to discover new materials. Recently, a U.S. research team succeeded in finding new quantum materials using AI.
A Massachusetts Institute of Technology (MIT) team said on the 22nd that it developed AI that finds materials with quantum properties, such as superconductors or special magnetic states. The findings were published the same day in the international journal Nature Materials.
Global corporations such as Google, Microsoft (MS), and Meta have designed tens of millions of new materials with AI for years. But there were limits to finding materials with quantum properties, such as superconductors or special magnetic states.
In fact, only a little over 10 candidate materials have been identified for "quantum spin liquids," a core material for next-generation quantum computers, despite 10 years of research. A quantum spin liquid is a state in which "spin," the electron's intrinsic quantum-mechanical property, moves fluidly like a liquid without following specific rules; using this, it is possible to implement qubits that are resistant to errors.
To solve this challenge, the team developed a new AI method called "SCIGEN." SCIGEN imposes "design rules" on existing generative models so that only materials with specific atomic array structures are created.
For example, the model is forcibly applied with lattice rules favorable for expressing quantum properties, such as the kagome lattice or Archimedean lattice. The kagome lattice is a distinctive pattern of overlapping triangles, and the Archimedean lattice is a motif in which different polygons are arranged repeatedly. These lattices make electrons move in special ways, increasing the likelihood of "superconductivity," where electrical resistance drops to 0 at ultralow temperatures, or other rare quantum phenomena.
Using this approach, the team generated millions of candidate materials, selected 26,000 of them for precise supercomputer simulations, and then synthesized two compounds that had not been discovered before. The materials showed the distinctive properties predicted by AI.
MIT Professor Mingda Li said, "Models from large corporations generate materials optimized for stability, but to change the world, you don't need 10 million materials—just a single outstanding one is enough," summarizing the significance of the study.
The team said, "This achievement could greatly speed up the development of new materials for quantum computers," adding, "If AI shortens experimental processes that would otherwise take decades and presents numerous candidates, researchers can use actual experiments to select the most promising materials."
References
Nature Materials (2025), DOI: https://doi.org/10.1038/s41563-025-02355-y