Beyond Diamond: Interpretable Machine Learning Discovery of Coherent Quantum Defect Hosts in Semiconductors

Abstract: Quantum point defects in wide bandgap semiconductors-such as the nitrogen-vacancy (NV) center in diamond are leading candidates for solid-state spin qubits due to their long coherence times and optically addressable spin states. Yet, identifying host materials capable of supporting such deep-level defects remains a significant challenge, owing to the complex interplay between chemical composition, crystal structure, and electronic environment. Here, we present a scalable, interpretable machine learning framework that combines density functional theory (DFT)-informed descriptors with a structure-agnostic ensemble model to predict quantum-compatible defect-host materials. Trained on a curated dataset from the Materials Project and ICSD, our model achieves a high Matthews correlation coefficient (MCC > 0.95) and assigns confidence scores to guide prioritization of candidates. First-principles calculations of coherence-relevant properties, including static dielectric constants and defect formation energetics, validate key predictions. While high dielectric response is a necessary but not sufficient condition for spin coherence, our model successfully recovers known hosts such as diamond and SiC, and reveals previously overlooked candidates such as WS2 , MgO, CaS and TiO2 . This approach establishes a robust path for discovering next-generation quantum materials, bridging data-driven screening with physically interpretable design principles.