Machine learning-driven optimization of lead-free Rb2PtxPd1-xBr6 dual-absorber photodetectors

Abstract

Lead-free halide perovskites are a promising class of materials for environmentally friendly optoelectronics, although the optimization of their complex device architectures presents a considerable challenge. Here, a high-performance photodetector is engineered using a dual-absorber layer architecture composed of Rb 2 Pt 0.75 Pd 0.25 Br 6 and Rb 2 Pt 0.25 Pd 0.75 Br 6 . An integrated methodology combining simulations using SCAPS-1D with machine learning and SHAP interpretability analysis enables the systematic optimization of absorber layer thickness, doping concentration, and defect density. The optimized device exhibits a responsivity of 0.582 A/W, a specific detectivity of 8.14 × 10 13 Jones at 767 nm, and a broadband response from 300 to 860 nm. The SHAP analysis reveals the dominant influence of one absorber layer's thickness on responsivity. This computational approach further uncovers synergistic effects and opposing contributions from identical parameters across the two layers, insights not accessible through conventional simulation alone. The combination of physics-based simulation with interpretable machine learning therefore establishes a powerful paradigm for the rational design and optimization of advanced photodetectors.