Machine Learning‐Assisted Optimisation of Sb ₂ S ₃ Hole Transport Layer for Enhanced Efficiency in MoTe ₂ ‐Based Solar Cells

In this work, we present a machine learning–augmented simulation study of a MoTe 2 ‐based solar cell incorporating Sb 2 S 3 (antimony sulfide) as the hole transport layer (HTL). MoTe 2 was selected as the absorber due to its strong optical absorption, low toxicity, and compatibility with low‐cost fabrication methods. Using SCAPS‐1D, we optimised the heterojunction structure (Al/FTO/CdS/MoTe 2 /Sb 2 S 3 /Pt) by varying absorber and HTL thickness, doping concentration, defect density, and temperature. The best simulated device (MoTe 2 thickness 0.5 μm, N A ≈ 10 17 cm −3 ) achieves V oc ≈ 1.05 V, J sc ≈ 40.8 mA/cm 2 , FF ≈ 87.6%, and a power conversion efficiency (PCE) of 40.33%. We clarify that this extremely high efficiency represents a theoretical upper bound under idealised assumptions (e.g., negligible nonradiative losses), rather than an experimentally demonstrated result. For comparison, the baseline cell without Sb 2 S 3 yields V oc ≈ 0.95 V, J sc ≈ 38.15 mA/cm 2 , FF ≈ 81.09%, and η ≈ 29.35%. The Sb 2 S 3 layer significantly suppresses back‐surface recombination and improves carrier extraction, thereby enhancing V oc and FF. To streamline design, we generated a dataset of approximately 6735 SCAPS simulations spanning key input variables (thickness, doping density, defect density, temperature) and trained five regression models for performance prediction. Among these, Random Forest regression achieved the highest accuracy ( R 2 ≈ 0.98), effectively capturing nonlinear dependencies. Feature‐importance analysis confirmed that absorber thickness, defect density, and doping are the dominant performance drivers, consistent with the physics‐based trends. This hybrid SCAPS–ML framework provides a fast, data‐driven tool for optimising next‐generation solar cells. Our study advances previous work by explicitly identifying robust parameter ranges, introducing predictive modelling, and clarifying the theoretical bounds of simulated efficiency.