Data-Driven Optimization of Lead-Free Perovskite Solar Cell Architecture Using Machine Learning and SCAPS Simulations

Significant advancements have been made in the study, research, and optimization of electrical devices and materials using machine learning (ML) models. Data regression and prediction provide for accurate estimates of device performance across a variety of factors by utilizing machine learning algorithms and input information. This study utilizes an ML algorithm to analyze the impact of 9 parameters on the output characteristics of methylammonium tin halide perovskite (CH<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf>NH<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf>SnX<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf>) solar cells. The perovskite device structure proposed here is FTO/ETL/CH3NH3SnX3/HTL/Au. Distinct parameters like doping levels, defect densities, and thickness of the diverse materials are analyzed to understand their impacts on device performance. A dataset of 2565 points was created using validated numerical simulations for training and testing the model. The PCE variation across the two metal halides was investigated, yielding a maximum PCE of 24.31% for CH<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf>NH<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf>SnBr<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf> and $27.37 \%$ for CH<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf>NH<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf>SnI<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</inf>. Using the SCAPS-1D simulation, the effects of several HTLs are examined. The device’s PCE rose from $27.37 \%$ to $39.68 \%$ by adjusting intrinsic material factors like absorber layer thickness, HTL thickness, doping, and defect concentration. A trained ML model further predicts PSC performance metrics with 91% accuracy. Consequently, the Al/FTO/PCBM/CH3NH3SnI3/V2O5/Au configuration is identified as the most suitable for manufacturing highly efficient photovoltaic devices.