Researchers have discovered two new superconductors using a machine learning-based screening method, demonstrating a faster way to identify materials that could one day enable room-temperature superconductivity.
The international team, led by researchers at Aalto University, combined machine learning with quantum physics calculations to identify two previously unknown superconductors, YRu3B2 and LuRu3B2. The approach significantly reduces the time needed to search through the enormous number of possible material combinations.
Superconductors can carry electricity with zero resistance, but only at extremely low temperatures. They are already used in technologies such as quantum computers, MRI scanners, fusion reactors, and maglev trains. Scientists have long sought materials that retain superconductivity at room temperature, a breakthrough that could transform power transmission and computing.
The newly discovered materials derive their superconducting properties from electrons arranged in a kagome lattice, a geometric pattern inspired by traditional Japanese basket weaving. After machine learning identified promising candidates, researchers verified them through theoretical calculations before synthesizing and experimentally confirming the materials.
AI narrows search
According to the researchers, the new workflow addresses one of the biggest challenges in superconductor research: the overwhelming number of possible material combinations.
“Superconductive materials that can operate at room temperature would forever change the way we consume energy,” explained Aalto University Professor Päivi Törmä. “If such a material could replace regular conductors in applications like computers and data centres, global energy consumption could be slashed and the heat footprint of the ICT sector vastly reduced.”
The work is part of the SuperC consortium, an international collaboration launched in 2023 with the goal of discovering a room-temperature superconductor by 2033.
After the computational screening, collaborators at Rice University synthesized the candidate materials into real samples. The experimental team then confirmed that both compounds exhibited superconductivity, providing proof that the machine learning-guided discovery process works.
Discovery process accelerates
For decades, scientists have relied largely on trial and error to discover superconducting materials.
“Over the decades researchers have recognised over 7,000 superconductors, but mostly serendipitously,” explained Törmä. “The process of identifying possible materials is so computationally heavy that, in fact, researchers have only been able to theoretically predict the viability of about 20 of these.”
The researchers say their approach could dramatically expand the number of materials that can be evaluated.
“Our method uses machine-learning-based pre-screening followed by targeted calculations on the promising candidates. This approach will greatly speed up superconductor discovery in the future. With machine learning, we may be able to push the number of materials we can process into the billions,” said Törmä. “This will take us a critical step closer to finding a room-temperature superconductor.”
Rather than replacing traditional physics calculations, the machine learning system acts as a filter, allowing researchers to focus computational resources on the most promising candidates. The team believes the approach could unlock thousands of new superconductors and accelerate the search for materials suitable for large-scale energy and computing applications.
The study was published in Physical Review Research.
Leave a comment