Numerical simulation and machine learning frameworks for efficiency improvement of CuInSe2 thin-film solar cells with WS2 ETL and MoO3 HTL

The copper indium selenide (CuInSe 2 , or CIS) absorber has garnered significant research interest as a cost-effective, durable, and highly effective material for thin-film solar cells (TFSCs). A SCAPS-1D simulator is utilized to evaluate the cell performance with various electron transport layers (ETLs) and hole transport layers (HTLs). This work is among the first investigations to incorporate WS 2 as an ETL and MoO 3 as an HTL to improve the efficiency of CuInSe 2 -based TFSCs and optimize the Al/FTO/WS 2 /CuInSe 2 /MoO 3 /Au heterojunction structure. Additionally, the present research compares the photovoltaic (PV) outcome of CuInSe 2 -based TFSCs with different ETL and HTL materials. The proposed configuration effectively reduce carrier recombination losses at the front and rear interfaces, appreciably improving the PV performance. Several physical characteristics, such as layer thickness, carrier concentration, defect density, interfacial defect, temperature, and cell resistances, have been carefully adjusted to evaluate the performance of the CIS-based solar cell (SC). An optimal PCE of 25.34%, along with a Voc of 0.99 V, a Jsc of 32.91 mA/cm 2 , and an FF of 77.31%, is achieved using a 0.10 μm MoO 3 HTL, a 1.0 μm CuInSe 2 absorber, and a 0.05 μm WS 2 ETL. Furthermore, eight machine learning algorithms and eight deep learning models are evaluated to identify the most effective approach for PV performance prediction. Notably, Gradient Boosting achieves the highest performance among all evaluated models, achieving a low MAE of 0.0311, an RMSE of 0.0450, and a high R 2 value of 0.9998. Additionally, the study investigates the impact of material-level variations on prediction accuracy, providing insights that may support the design of optimized and cost-effective solar cell prototypes.