Predictive modeling and optimization of CIGS thin film solar cells: A machine learning approach

Abstract

This study employs Machine Learning (ML) techniques to optimize the fabrication of Copper Indium Gallium Selenide (CIGS) thin-film solar cells and enhance their efficiency. An extensive dataset encompassing over 5000 data points from CIGS solar cell fabrication experiments is analyzed using various ML algorithms such as Artificial Neural Network (ANN), and Random Forest (RF). RF emerge as the most effective model, achieving adjusted R-squared values exceeding 0.87 for all the outputs, predicting key solar cell performance metrics, while ANN with R² less than 0.68 for all the outputs, underperformed. Feature importance analysis based on RF revealed that compositional ratios of precursor materials, particularly Ga/(In + Ga) and Cu/(In + Ga), followed by RTA temperature and i-ZnO thickness, are the most critical factors influencing device performance. A decision tree model provide detailed insights into optimal compositional ratios and fabrication conditions, suggesting RTA temperatures around 475 °C and i-ZnO thicknesses of approximately 50 nm for maximizing efficiency. This machine learning-driven approach offers a powerful tool for guiding CIGS solar cell fabrication, potentially accelerating the optimization process and advancing thin-film photovoltaic technology.