Superconductors and Quantum Materials

The search for high-temperature superconductors has been one of the defining challenges in materials science for decades. Despite enormous research investment, the discovery of new superconducting families has been sporadic and largely serendipitous — cuprates in the 1980s, iron-based superconductors in 2008, nickelate superconductors recently. The fundamental problem is generative: the space of possible crystal structures, compositions, and electronic configurations that could support superconductivity is vast, while the physical intuition guiding human researchers is anchored to a small number of known mechanisms. Entire regions of materials space that could host novel superconducting phases remain unexplored because there is no systematic method to generate plausible candidates outside established chemical families.

Conventional approaches to superconductor discovery rely on chemical intuition, analogy to known superconducting families, or high-throughput DFT screening of structural databases. Chemical intuition produces incremental modifications — doping variations, pressure studies, isovalent substitutions — that refine known families but rarely discover new ones. Database screening is limited to structures already cataloged in repositories like ICSD or the Materials Project, missing the vast space of synthesizable but unreported compositions. Machine learning models trained on existing superconductor data suffer from severe class imbalance and cannot reliably generate candidates in compositional regions far from training data, precisely where the most impactful discoveries would occur.