Lead-free Sr3BiI3 heterojunction anti-perovskite solar cells with band-aligned WS2 and Cu2O transport layers: SCAPS-1D optimization and machine-learning-assisted efficiency prediction

Energy demands and lead toxicity concerns are driving the development of stable, eco-friendly perovskite alternatives. This study presents the design, simulation, and comprehensive parametric optimization of a Sr 3 BiI 3 anti-perovskite solar cell, with the final optimized Al/FTO/WS 2 /Sr 3 BiI 3 /Cu 2 O/Ni heterojunction architecture being Pb-free and Cd-free. Systematic screening of ETL candidates (In 3 Se 4 , CdS, SnS 2 , and WS 2 ) and HTL candidates (P3HT, SnS, CuSCN, Spiro-MeOTAD, CuI, and Cu 2 O) confirmed that interface band-alignment engineering is the principal determinant of performance. Device performance was investigated by systematically tuning perovskite thickness, carrier concentration, bulk defect states, and defect densities at both interfaces, along with analysis of the influence of the conduction-band offset (CBO) and valence-band offset (VBO) using Nyquist plots. The impact of operating temperature, back-contact work function, internal parasitic resistive losses, carrier lifetime, and diffusion length was also evaluated to determine their impact on overall device performance. The WS 2 /Sr 3 BiI 3 interface formed a favorable spike-type conduction-band offset of +0.17 eV, effectively suppressing minority-carrier recombination compared with cliff-type alternatives, whereas Cu 2 O provided a near-zero valence-band offset of +0.014 eV that enabled selective hole extraction while blocking electron back-transport. The optimized device achieved a PCE of 30.53%, Voc of 0.87 V, Jsc of 40.54 mA/cm 2 , and FF of 86.29%, with near-unity EQE sustained above 95% across 350-1000 nm. Moreover, among eight Machine learning (ML) and deep learning (DL) regressors applied to a 1024-point full-factorial SCAPS-1D dataset, Gaussian process regression (GPR) achieved the highest predictive accuracy (R 2 = 0.9999, RMSE = 0.044%), with subsequent importance and partial-dependence analysis identifying acceptor concentration (7.23%) and defect density (5.05%) as the dominant optimization levers. The GPR response surface revealed a peak theoretical PCE of 39.85% at N A of 10 20 cm −3 and N t = 10 12 cm −3 , establishing Sr 3 BiI 3 as a promising computational candidate warranting experimental investigation for next-generation lead-free anti-perovskite photovoltaics.