Unveiling stable and efficient antiperovskite semiconductors via high-throughput computation and interpretable machine learning

Nitride antiperovskites have recently been theoretically identified as promising optoelectronic materials, yet their chemical space remains largely unexplored. Here, we employ a high-throughput first-principles screening workflow to systematically investigate the X₃BA antiperovskite family. Six candidates that exhibit both structural and dynamical stability together with desirable bandgaps are identified. Electronic-structure calculations reveal that the alkaline-earth-based compounds (e.g., Ca₃AsSb, Sr₃AsSb, Ba₃AsSb) not only possess suitable direct bandgaps and strong optical absorption, but also exhibit favorable ambipolar carrier mobilities and low exciton binding energies (< 45 meV). Notably, Sr₃AsSb and Ba₃AsSb are predicted to achieve theoretical maximum power-conversion efficiencies of 28.1% and 29.4%, respectively. Finally, an interpretable machine-learning model demonstrates that the electronegativity of the A-site anion is the single most influential descriptor governing bandgap trends across the chemical space. This work establishes a data-driven design heuristic and provides a predictive framework for the accelerated discovery of efficient and stable antiperovskite-based optoelectronic materials.