An international team of physicists applied machine learning to accelerate the search for superconductors. Instead of lengthy laboratory trial and error, the algorithm quickly sifted through thousands of untested chemical structures.
The SuperC project, launched in 2023, coordinated the work of researchers from Aalto University in Finland. The machine learning model identified two previously unknown compounds: YRu₃B₂ and LuRu₃B₂.
Scientists from Rice University in Texas synthesized these materials in the laboratory and experimentally confirmed their superconducting properties.
Both compounds have a kagome structure — a triangular lattice reminiscent of Japanese basket weaving. This geometry creates flat energy bands where electrons slow down and more easily form the pairs necessary for superconductivity.
Superconductors allow current to flow without loss to resistance, but most known materials only work at extremely low temperatures. The new compounds are no exception: they require strong cooling.
The main outcome of the work is not the materials themselves, but the proven effectiveness of the method. The algorithm filters out unpromising candidates in seconds, leaving only the most promising ones for quantum calculations and synthesis.
Can such an approach now be scaled to billions of structures in the search for superconductors that work at room temperature?
The researchers note that the validated methodology opens the way to systematic search for materials for energy, electronics, and other fields where energy losses remain a serious problem.
