Unveiling future superconductors through machine learning

• Published in: Materials today physics Vol. 43; p. 101384
• Main Authors: Bai, Zihao, Bhullar, Mangladeep, Akinpelu, Akinwumi, Yao, Yansun
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-04-2024
• Subjects: First principles calculations; High-throughput computing; Machine learning; Superconductivity; Superhydrides; First principles calculations; High-throughput computing; Superhydrides; Superconductivity; Machine learning
• ISSN: 2542-5293; 2542-5293
• DOI: 10.1016/j.mtphys.2024.101384

The recent discovery of superconductivity above 200 K in hydrides of sulfur and lanthanum under high pressure marked a significant advance toward the realization of room-temperature superconductivity. While binary hydrides have almost been completely studied theoretically, experimental evidence suggests that the next breakthrough in finding high-temperature and low-pressure limits is likely connected with ternary and higher hydrides. Unlike the traditional synthesis-test-repeat approach, experimental discovery of superhydrides under high pressure often follows prior theoretical predictions. In this Minireview, we describe how various artificial intelligence schemes enable and enrich each stage of the discovery cycle of superhydrides and new developments made toward predicting ternary and higher hydrides. As a new enabling tool, machine learning-informed material simulation is still making its way into this field but is already playing an essential role in augmenting the prediction of new superhydrides through automated and iterative machine-learning processes. The review concludes with a perspective on outstanding challenges and possible future developments in the field.