Two New Superconductors Discovered Thanks to Artificial Intelligence: Is It a Breakthrough?
The search for superconducting materials may be entering a new phase thanks to the use of artificial intelligence. An international group coordinated by Aalto University has demonstrated how a machine learning system can drastically reduce the number of material combinations to analyze, allowing for the identification of promising candidates much faster than traditional methods.
The result has already been validated with the discovery of two new superconductors, named YRu3B2 and LuRu3B2, and represents concrete evidence of the effectiveness of this approach. Superconductors are materials capable of transporting electric current without any resistance, thus eliminating energy losses. However, this property only emerges at extremely low temperatures, close to absolute zero, making the use of expensive cryogenic systems necessary. Despite this limitation, such materials are already used in numerous sectors, including quantum computers, magnetic resonance imaging, fusion reactors, and magnetic levitation trains.
For decades, the scientific community has been engaged in the search for a room-temperature superconductor, considered one of the most important goals in materials physics. Achieving such a result would have profound implications for electricity transmission and the efficiency of information systems and digital infrastructures.
YRu3B2 and LuRu3B2 owe their superconductivity to electrons forming flat bands in a kagome lattice. This lattice takes its name from a hexagonal pattern typical of Japanese basket weaving. According to Päivi Törmä, a professor at Aalto University and head of the international consortium SuperC, materials of this type could radically change the way energy is utilized.
If used in computers and data centers, for example, they could significantly reduce both electricity consumption and the heat generated by ICT infrastructures. This work falls under the activities of SuperC, an international project launched in 2023 that brings together physicists and researchers with the declared aim of identifying a room-temperature superconductor by 2033. To achieve this goal, the consortium combines quantum physics models with machine learning tools, using artificial intelligence not as a substitute for theoretical calculations but as a filter capable of quickly identifying the most promising combinations.
The developed method follows a complex procedure. The machine learning algorithm conducts an initial screening among a potentially vast number of chemical element combinations, identifying only those with the highest likelihood of exhibiting superconductive properties. Subsequently, detailed quantum calculations are performed on the selected candidates to theoretically verify their behavior.
The two identified materials, YRu3B2 and LuRu3B2, share a specific electronic structure known as a kagome lattice, a geometry inspired by traditional Japanese basket weaving. In this configuration, electrons form flat energy bands that favor the emergence of superconductivity according to the theoretical models developed by the group. After theoretical validation, the experimental phase was entrusted to researchers at Rice University, led by Professor Emilia Morosan. The group synthesized the new compounds from the basic chemical elements and subsequently experimentally verified that both indeed exhibited superconductive properties, thereby confirming the reliability of the entire discovery process.
The result mainly represents a proof of principle. Historically, the discovery of superconductors has largely occurred through trial and error or by chance observations. Although more than 7,000 superconducting materials are known today, only about twenty have been theoretically predicted prior to their synthesis, due to the enormous computational complexity required in analyzing possible atomic combinations.
The new approach aims to radically alter this scenario. By using machine learning as a pre-selection system, researchers can concentrate computing resources only on truly promising candidates, thereby reducing the time and costs of the entire process. According to the team, this strategy could allow for the evaluation of billions of different materials in the future, paving the way for the discovery of thousands of new superconductors.