AI-assisted optimization of lead-free Rb2SiX6 (X = F, Cl, Br, and I) perovskite solar cells with a 2D buffer layer design using DFT and SCAPS-1D

Abstract

In this work, the potential of the environmentally benign double halide perovskite Rb₂SiX₆ (X = F, Cl, Br, and I) as a solar cell absorber is investigated through the combined use of density functional theory (DFT), SCAPS-1D modeling, and machine learning techniques. Detailed DFT analysis indicates that Rb₂SiBr₆ crystallizes in a stable cubic structure with a direct bandgap of ∼1.65 eV, making it the most suitable among Rb₂SiF₆, Rb₂SiCl₆, Rb₂SiBr₆, and Rb₂SiI₆ for visible light absorption. SCAPS-1D modeling reveals that device performance depends strongly on absorber thickness, doping concentration, and interface defect density. In the absence of a back surface field (BSF) layer, the device achieved a V_OC of 1.13 V, a J_SC of 21.28 mA cm⁻², an FF of 85.66%, and a PCE of 20.56% with a two-dimensional (2D) WS₂ buffer layer. Introducing a Spiro-OMeTAD BSF layer further boosts performance, increasing V_OC to 1.25 V and J_SC to 22.93 mA cm⁻² with a slight change in FF to 84.64%, resulting in a PCE of 24.27%. The BSF layer also enhances thermal stability, suppresses recombination losses, and improves carrier extraction under varied operating conditions. Machine learning, particularly Random Forest with SHAP analysis, highlights electron affinity and conduction band density of states as the most influential factors for device efficiency. While simulations predict efficiencies above 24%, experimental validation is essential to address fabrication and long-term stability challenges, supporting Rb₂SiBr₆ as a non-toxic, scalable photovoltaic candidate.