Predictive modeling and optimization of WS2 thin-film solar cells: A comprehensive study integrating machine learning, deep learning and SCAPS-1D approaches

In this study, the SCAPS-1D simulator is employed to design and examine the photovoltaic characteristics of the WS2 absorber-based thin-film solar cell with various electron and hole transport layers. Among the various structures, the Al/FTO/TiO2/WS2/Zn3P2/Ni photovoltaic device provides excellent performances due to the proper energy band alignment at interfaces with favorable material properties. Here, the outputs of the proposed structure are assessed by varying the thickness, carrier concentration, and defect density of the absorber layer. The thermal stability of the device is also determined by analyzing the impact of temperature on the cell performance. The proposed device reveals outstanding performances, including Voc of 1.03 V, Jsc of 35.04 mA/cm2, fill-factor of 87.73 %, and admirable efficiency of 31.78 %. Furthermore, three machine learning and eight deep learning methods are introduced to compare and determine the most efficient algorithm for photovoltaic devices. Among the eleven algorithms, the multi-layer perceptron model demonstrates outstanding performance including a lower MSE of 0.0213 and an outstanding R2 of 0.9992, making it the most accurate model for outcome prediction. Here, the impact of material properties on model output is also investigated. In general, these results may help experimental researchers to design highly efficient and low-cost solar devices.