Artificial intelligence guided study of superconductors

Superconductors have zero electrical resistance and perfect diamagnetism below a critical temperature. They have the potential to spark industrial innovation in areas such as power transmission, transportation, and computing. Nevertheless, the search for new superconducting materials is a slow and laborious process. We leverage artificial intelligence (AI) to accelerate the discovery of superconductors.

We present a data-driven framework for accelerating the discovery of novel BCS superconductors. We leverage a carefully curated superconductor database and AI models for prediction and inference. The database is based on the Crystallographic Open Database (over 500,000 entries) and the Handbook of Superconductivity (over 800 entries). Semi-supervised learning is leveraged to mitigate the scarcity of labelled data.

AI is first used to determine if a material is superconducting or not. Next, AI predicts the critical temperature of superconductors. We report the prediction accuracy for both tasks.

AI predictions rapidly identify both known and novel superconductors and their properties. We harness AI to identify ‘hidden’ patterns in high-dimensional data that describe the relationships between crystal structure and the electronic properties of superconductors.