Machine Learning-Assisted Design Optimization of CdS/Zn(O,S) Window Layer CdTe Solar Cells Using SCAPS-1D

Abstract

Cadmium telluride (CdTe) is a scalable, high-absorption thin-film photovoltaic technology, but a trade-off at the front interface limits its efficiency. Thinner CdS window layers cut parasitic short-wavelength absorption, while rough commercial substrates need a thick window stack of at least 60 nm to prevent micro-shunting. A two-stage SCAPS-1D framework was used. Gaussian Process Regression (GPR) first optimized a discrete CdS/ZnO/ZnS stack within physically motivated thickness bounds. The resulting 10/40/10 nm geometry satisfied the 60 nm floor on its own this constraint was never given to GPR directly. A dilute linear compositional gradient (6-% sulfur) then reduced the resulting conduction band offset below 0.07 eV. This kept the alloy in the oxygen-rich regime, away from the bandgap collapse seen in high-sulfur Zn(O,S) compositions. The graded architecture reached 16.86% efficiency and a 79.06% fill factor against a 15.39% baseline and retained 16.26% efficiency at 307 K versus 14.84% for the baseline. Confining the CdS seed layer to 10 nm cuts CdS consumption by 83.3% relative to the conventional 60 nm baseline. Efficiency gains and reduced CdS use did not trade off against each other in this geometry.