Hybrid advanced computational modeling of RbGeX3 solar cells with NZO/GO charge transport layers

This is the authors' abstract. We don't add key points for this paper.

This study involves a multiscale DFT + SCAPS-1D + ML framework, which provides a powerful toolset for optimizing perovskite solar cells. Atomic‐scale simulations reveal that RbGeI 3 forms a cubic perovskite lattice with a direct band gap of ≈1.55 eV, whereas RbGeBr 3 and RbGeCl 3 adopt distorted orthorhombic structures with much larger gaps (∼2.58 eV and ∼3.2 eV). The RbGeI 3 gap lies near the ideal single‐junction range (∼1.2–1.7 eV), and its electronic structure exhibits delocalized Ge–I orbitals with mixed ionic–covalent bonding, implying high carrier mobility. In contrast, the Br/Cl compounds exhibited stronger structural distortion and reduced orbital overlap, consistent with lower mobility. These DFT findings indicate that RbGeI 3 is intrinsically best suited for efficient light absorption and charge transport. SCAPS-1D device simulations of FTO/Nd:ZnO (NZO)/RbGeX 3 /graphene‐oxide (GO)/Pt or C-Cu cells confirm that RbGeI 3 yields the highest PV performance. Under optimized conditions (absorber thickness ∼ 500–600 nm, bulk defect density <10 15 cm − 3 , and a back-contact work function >=5.0 eV), the RbGeI 3 ‐based cell achieves Voc ≈ 1.054 V, Jsc ≈ 27.94 mA/cm 2 , FF ≈ 80.4 %, and PCE ≈ 23.68 %. Efficiency falls sharply for thinner/thicker layers or higher trap densities. The inclusion of a GO hole-transport layer enhances the hole extraction and Voc. In contrast, the wider-gap RbGeBr 3 and RbGeCl 3 devices show much lower Jsc and efficiency under the same conditions. Machine-learning regression models (linear regression, SVR, random forests, XGBoost) trained on the SCAPS dataset accurately reproduce these device outputs. XGBoost provides the best predictions (for PCE, MSE ≈ 0.072, R 2 > 0.999), consistently outperforming simpler models. SHAP feature analysis reveals that absorber defect density, series/shunt resistances, and doping concentrations are the most influential variables for Voc, Jsc, FF, and PCE. This confirms that defect and resistive losses dominate the device performance in these cells.