Performance Optimization of Lead-Free KGeCl3 Solar Cells Using SCAPS-1D and Stacked Machine Learning Models

The development of high-efficiency and environmentally benign photovoltaic devices has stimulated growing interest in lead-free perovskite and halide-based absorber materials. In this work, we systematically investigate and optimize the performance of a novel solar cell architecture, FTO/WO3/KGeCl3/HTL (GQD and MoS2)/Au, employing two different hole transport layers (HTLs), namely graphene quantum dots (GQDs) and MoS2. Using SCAPS-1D numerical simulations, key device parameters including absorber layer thickness, HTL thickness, total defect density of the absorber, bandgap dependence, and operating temperature are comprehensively optimized. The GQD-based device exhibits a superior power conversion efficiency (PCE) of 23.35%, outperforming the MoS2-based device, which achieves 20.07%. The enhanced performance of the GQD-based architecture is attributed to improved energy band alignment, stronger internal electric fields, reduced carrier recombination, and enhanced charge extraction efficiency. Detailed analyses of the energy band diagram, electric field distribution, carrier generation and recombination profiles, and external quantum efficiency (EQE) further elucidate the underlying physical mechanisms governing device operation. To accelerate device optimization and performance prediction, a machine-learning-assisted framework is implemented by stacking Random Forest (RF) and XGBoost (XGB) models and feeding their outputs into a Neural Network (NN) meta-learner. The stacked model demonstrates excellent predictive accuracy, achieving R2 values of 99.80%, 99.77%, and 99.83% for RF, XGB, and NN models, respectively, with a remarkably low mean squared error (MSE) of 0.001 for the NN. These results highlight the effectiveness of integrating physics-based simulations with machine learning for the design and optimization of next-generation, high-efficiency solar cells.