Machine learning-aided optimization for transport layer parameters of low lead inorganic Zn-based mixed-halide perovskite solar cell

Abstract

Perovskite-based solar cells have captivated researchers, due to their outstanding photovoltaic (PV) performance. All inorganic halide perovskite solar cells (AIHP-SC) exhibited excellent stability against environmental conditions with improved lifetime. In this investigation, low lead CsPb0.625Zn0.375I2Cl based AIHP-SC is designed with the utilization of a SCAPS-1D simulator. Further, the impact of ETL/HTL parameters viz. mobility and doping on the PV performance is analyzed. Thereafter, machine learning (ML) models are trained, tested, and verified using artificial intelligence (AI) algorithms. Utilization of ML models shortens experiment time and eliminates the requirement for extensive resources in designing and predicting the PV performance of solar cells. In this research, the PV performance of AIHP-SC layered as TiO2/ CsPb0.625Zn0.375I2Cl/ Spiro-MeOTAD at different ETL/HTL mobility (ranging from 0.004 cm2/Vs to 4 cm2/Vs) and doping density (ranging from 1015 cm−3 to 1019 cm−3) has been evaluated. The influence of these variations leads to the generating 2500 PV performance datasets including current density (JSC), fill factor (FF), open circuit voltage (VOC), and power conversion efficiency (PCE). These generated data set attributes are fed to train the ML models viz. linear regression (LR), random forest (RF), support vector regression (SVR), eXtreme gradient boosting (xGB), and artificial neural network (ANN). For validation, the performance of ML models is verified against SCAPS-1D generated performance results by using mean square error (MSE) and R square (R2) as the performance metrics. Execution of ML algorithms revealed that performance of RF and xGB exhibits high correlation with SCAPS-1D generated data set as the prediction made by these two algorithms are best matched to actual outcome. Additionally, Shapley additive explanations (SHAP) analysis is performed to examine the influence of input variables (also known as independent variables) on target performance parameters (VOC, JSC, FF and PCE). A significant rise in PCE is observed from 9.5 % to 21.90 % while optimizing ETL/HTL mobility and doping density of AIHP-SC. The investigated results provide supervision to researchers in the design of highly stable AIHP-SC, further integration of ML in AIHP-SC is time time-efficient approach to design and performance prediction.