Accelerated screening for high-efficiency inorganic photovoltaic materials: from ensemble voting regression to device performance

Abstract

Spectral limited maximum efficiency (SLME), a theoretical measure of photoelectric conversion efficiency, has been widely used in solar cell research. A Voting Regressor model, integrating three base algorithms, is utilized to rapidly predict the SLME of 51 472 inorganic materials. The predicted result indicates that transition metals and chalcogens are identified as key elements with high SLME potential due to their complex and tunable band structures. SHAP analysis combined with DFT calculations reveals that characteristics such as the band gap, crystal structure, and space group significantly impact SLME by affecting light absorption and carrier mobility. The dipole-allowed direct optical transitions, combined with low electron and hole effective masses, not only enhance the efficiency of photogenerated charge production but also promote carrier transport and separation, thereby significantly improving the SLME. The final selected candidates all exceed an SLME of 30%, with Ba₃(InP₂)₂ achieving a J_sc of 425.75 A m⁻², Ba₂PCl and NaSrP having V_oc values over 0.95 V, and NaSrP reaching a fill factor of 88.31%. Specifically, Ba₈P₅Br and Mg(InTe₂)₂ significantly surpass the classical GaAs semiconductor in all performance metrics. This work demonstrates the enormous potential of five new materials in photovoltaic applications, offering fresh insights for the design and optimization of efficient photovoltaic materials.