Key factors governing the device performance of GIGS solar cells: Insights from machine learning

Cu(In(1-x)Gax)Se-2 (CIGS) solar cells are a kind of highly efficient thin film solar cells, further breakthrough in their device efficiency relies on the development of advanced methods and/or deep insights into the factors governing the device efficiency. Herein, we use the machine learning (ML) algorithms to explore the key factors governing the device performance of the CIGS solar cells and the underlying correlations. The datasets for ML are obtained from the experimental reports, which enables the results more referable for experimental optimization. Key factors governing the device performance are screened based on the correlation studies. The ML algorithms including linear regression, neural network, random forest (RF) and extreme gradient boosting are employed, among which RF performs best in predicting the efficiency of the CIGS solar cells with high accuracy (mot mean square error of 0.9% and1.8%, Pearson coefficient r of 0.9 and 0.88 for validation and test sets, respectively). Furthermore, the factors and their optimal scales for high device efficiency are predicted, which provides essential guidance for experimental device optimization.