Machine-Guided Design of Oxidation-Resistant Superconductors for Quantum Information Applications

Decoherence in superconducting qubits has long been attributed to two-level systems arising from the surfaces and interfaces present real devices. A recent significant step reducing decoherence was replacement of niobium by tantalum, resulting a tripling transmon qubit lifetimes (T1). The identity, thickness, quality native surface oxide, is thought play major role, as tantalum only one oxide whereas several. Here we report development thermodynamic metric rank materials based on their potential form well-defined, thin, oxide. We first computed this for known binary ternary metal alloys using data available Materials Project experimentally validated strengths limits through preparation controlled oxidation eight alloys. Then trained convolutional neural network predict value atomic composition properties. This allowed us compute that are not Project, including large selection superconductors, and, when combined with Tc, identify new candidate superconductors quantum information science engineering (QISE) applications. tested resistance pair these predictions experimentally. Our results expected lay foundation tailored rapid improved QISE.