Bayesian Optimization-Guided Discovery of Gallium-Containing Semiconductors with Targeted Band Gaps

Gallium-based semiconductors provide tunable electronic and optical properties across different oxidation states and compositional environments. However, rational discovery of gallium-containing materials with targeted band gaps is difficult, given the vast chemical space and expensive first-principles screening. In this study, we present a machine-learning-guided Bayesian optimization (BO) framework that enables inverse design of Ga-based compositions with predefined electronic properties while maintaining chemical plausibility. Multiple regression models were evaluated, and KNN emerged as the optimal surrogate (R² = 0.812). Using the KNN predictor, BO proposes novel Ga-containing compositions targeting band gaps from 0.5–3.5 eV. Chemical plausibility is enforced via SMACT screening (charge balance, elemental feasibility, physical plausibility). BO adaptively steers the search toward regions of highly expected improvement. Postoptimization analyses show 100% uniqueness and novelty relative to training data, with increased SMACT validity near 1.5–2.5 eV. The framework accelerates inverse design under realistic chemical constraints, offering an alternative to purely DFT-centric screening.