An international research consortium has demonstrated that machine learning can dramatically accelerate the discovery of superconducting materials, using AI to screen vast numbers of elemental combinations and identify the most promising candidates for detailed quantum analysis.
The SuperC consortium, led by Aalto University Professor Päivi Törmä, used the approach to identify two previously unknown superconductors, YRu3B2 and LuRu3B2, both of which derive their properties from electrons forming flat bands within a kagome lattice structure. A machine-learning algorithm first screened enormous numbers of possible material combinations, with the strongest candidates then subjected to targeted quantum calculations. Collaborators at Rice University, led by Professor Emilia Morosan, subsequently synthesised and experimentally verified both materials. The findings were published in Physical Review Research.
The significance lies in the scale the method makes possible. Of more than 7,000 known superconductors, researchers have only been able to theoretically predict the viability of around 20, due to the computational demands involved. Törmä said the AI-driven approach could push the number of materials that can be screened into the billions.
The consortium’s broader ambition is to find a room-temperature superconductor by 2033. Such a material, Törmä argued, could fundamentally reduce global energy consumption, particularly in computing and data centre infrastructure where heat generation represents a significant and growing cost. SuperC was established in 2023 and receives funding from sources including The Kavli Foundation and the Jane and Aatos Erkko Foundation.
