Theoretical prediction of semiconductors by data driven light-element substitution in topological materials

The rational design of high-performance semiconductors-particularly those with tunable bandgaps, high carrier mobility, and stability-remains a fundamental challenge in materials science, as traditional trial-and-error approaches struggle to explore vast chemical spaces efficiently. Guided by insights from band-inverted topological materials, we present an integrated high-throughput calculation and machine learning-based workflow that rapidly uncovers previously uncharted semiconductors. The workflow homes in on the non-parabolic band structures produced by inversion to reveal compounds with high mobility and defect tolerance. Starting from topological materials containing heavy IVA-VIIA elements, we generated thousands of crystal structures via light-element substitution-a strategy designed to engineer new semiconductors beyond conventional compositions. Using high-throughput calculations, we computed key electronic and stability properties, training a classification machine learning model on a 50% subset of the data to predict formation energies for the remaining candidates. The developed workflow enabled us to identify 14 new stable semiconductor candidates with promising potential for photovoltaic and thermoelectric applications. These findings demonstrate that the data-driven approach to light-element substitution in topological materials enables the design of promising semiconductors beyond conventional chemical spaces.