Eco-friendly Cs2SnGeCl6 perovskite absorber: A combined numerical simulation and machine learning analysis for high efficiency solar cells

This study presents a comprehensive numerical analysis of eco-friendly Cs 2 SnGeCl 6 -based double perovskite solar cells using the SCAPS-1D simulation framework. A total of 64 device architectures, incorporating different combinations of ETLs and HTLs, were examined. Among them, the ITO/Ws 2 /Cs 2 SnGeCl 6 /CuSbS 2 /Ni structure demonstrated the highest performance, achieving a peak power conversion efficiency (PCE) of approximately 30.80%. Subsequent optimization of the absorber layer identified the ideal parameters as a 0.9 µm thickness, an acceptor density (N a ) of 1 × 10 18 cm⁻ 3 , and a defect density (N t ) of 1 × 10 15 cm⁻ 3 , accompanied by fine-tuning of the transport layer thicknesses and doping levels. The device was further analyzed under variations in temperature, series resistance, and shunt resistance, with detailed evaluations of J–V characteristics, quantum efficiency, recombination and generation dynamics, capacitance, and Mott–Schottky behavior. In addition to simulation, a machine learning pipeline was developed using 2187 high-fidelity SCAPS-1D configurations to predict key photovoltaic parameters. Among ten tested algorithms, the ExtraTrees model achieved outstanding predictive performance (R 2 = 0.999945) following optimal hyperparameter tuning. Feature importance analysis indicated that N t is the most influential factor, and the model predicted a PCE of 29.79% under the optimized conditions of absorber thickness, defect density, and temperature (0.9 µm, 1 × 10 15 cm⁻ 3 , and 300). Lead-free Cs 2 SnGeCl 6 double-halide perovskite solar cells are optimized using SCAPS-1D simulations and machine learning. Screening 64 device architectures identifies a WS 2 /CuSbS 2 -based configuration achieving 30.8% efficiency. Machine-learning validation highlights absorber defect density as the dominant performance-limiting factor.