First-principles-informed device modeling and machine learning-guided sensitivity analysis of a lead-free Ba₃SbBr₃/CsGeCl₃ dual-absorber perovskite solar cell

This work provides a theoretical multiscale analysis of a lead-free dual-absorber perovskite solar cell based on density functional theory (DFT) calculations, SCAPS-1D simulations, and machine learning-based sensitivity analysis. First-principles calculations show that both Ba 3 SbBr 3 and CsGeCl 3 are cubic compounds (space group Pm-3m) with direct HSE06 band gaps of 1.711 and 1.373 eV, respectively. Cohesive energy, elastic constants, and phonon dispersion curves support the structural, mechanical, and dynamical stability of the absorbers. The absorption coefficients of both absorbers are of the order of 10 4 cm −1 in the visible range, which is suitable for photovoltaic applications. SCAPS-1D numerical simulations were performed for the Al/FTO/TiO 2 /Ba 3 SbBr 3 /CsGeCl 3 /V 2 O 5 /Au device architecture under standard AM 1.5 G illumination. The values of optimized open-circuit voltage (V oc ), short-circuit current density (J sc ), fill factor (FF), and power conversion efficiency (PCE) are 1.035 V, 32.02 mA/cm 2 , 84.22%, and 27.90%, respectively, higher than the single-absorber devices made from Ba 3 SbBr 3 (23.02%) and CsGeCl 3 (25.09%) under identical simulation conditions. The dual-absorber device retained approximately 73% of its simulated maximum efficiency at 450 K, indicating relatively stable electrical performance at high temperature. To investigate the parameter space further, four machine learning regression models (XGBoost, Random Forest, Support Vector Regression, and Gaussian Process Regression) were trained and tested using 2000 simulated device configurations. On the held-out test set, all four models predicted the PV output parameters accurately, with R² ≥ 0.994 on the held-out test set, and RMSE ≤ 0.2519 and MAE ≤ 0.1278 across all outputs. On an independent dataset comprising 192 SCAPS-1D simulations, prediction accuracy was lower, with R² ≥ 0.7217, RMSE ≤ 0.3863, and MAE ≤ 0.3070. Feature importance, SHAP, and Pearson correlation analyses also reveal that the key parameters governing PCE are CsGeCl₃ defect density and operating temperature, while CsGeCl₃ thickness mainly controls J sc . The results provide theoretical support to design the future lead-free dual-absorber solar devices.