ML Screening and DFT Validation of BeA 2 X 4 (A = Al, Ga, In; X = Se, Te) Ternary Compounds: Design and Optimization for Infrared Optoelectronic Materials

Machine-learning-accelerated discovery of monoclinic BeA2X4 (A = Al, Ga, In; X = Se, Te) yields six infrared-absorbing chalcogenides with thickness-tunable absorptance. A 3814-compound prototype library was stability- and gap-screened to 1268 candidates; XGBoost (R2 = 0.939, MAE = 0.241 eV) nominated BeAl2Se4, BeGa2Se4, BeIn2Se4, BeAl2Te4, BeGa2Te4, and BeIn2Te4. Hybrid-DFT (HSE06) and 10,000 fs ab initio molecular dynamics (AIMD) simulations at 300 K confirm the monoclinic Pm structure with thermal robustness. BeGa2Se4 exhibits the highest Vickers hardness (1.59 GPa), and BeAl2Te4 exhibits the largest Poisson ratio (0.44). The strong anisotropic absorption spans 0.5–1.6 eV, peaking at 1.21 × 105 cm–1 (BeIn2Se4) and 1.16 × 105 cm–1 (BeAl2Te4, 0.5–1 eV). Absorptance scales linearly with thickness: 0.747 (500 nm BeAl2Se4), 0.726 (450 nm BeGa2Se4), and 0.650 (300 nm BeIn2Se4), defining a 300–500 nm design window that balances absorption gain against carrier transport loss. Direct band gaps eliminate phonon bottlenecks, while their systematic narrowing with A/X cation radii enables continuous spectral tuning across the near- to mid-infrared. The quantitative structure–absorption map provides experimentally accessible guidelines for next-generation infrared absorption coatings and photodetector absorber layers.