Investigation of the open-circuit voltage of non-fullerene acceptors-based ternary organic solar cells based on interpretable machine-learning approach and chemically inspired descriptors

Abstract

Ternary strategy to enhance the photovoltaic performances (e.g., open-circuit voltage (Voc), current density (Jsc), fill factor (FF), and power conversion efficiency (η)) of non-fullerene acceptors (NFAs)-based organic solar cells (OSCs) has exhibited alluring potential for next generation photovoltaic technology. However, the prediction and optimization of Voc in NFA-based ternary OSCs through expensive and time-consuming experiments or a theoretical perspective (e.g., an empirical rule for Voc prediction of the donor–acceptor binary system) is still an open challenge, mainly because of the complicated ternary photoactive layers. In this study, the prominent predictive performance (R2 > 0.7) of Voc is obtained by utilizing the technique of machine-learning combined with specific descriptors (e.g., frontier molecular orbital theory or electrophilicity index (ɷ)), for the first time forNFA-based ternary OSCs. Furthermore, the Shapley additive explanation (SHAP) model is employed to provide both global and local interpretable explanations for exploring the interpretability of the (eXtreme Gradient Boosting (XGBoost) model prediction and extracting the complex correlation between the specific descriptors and the Voc of NFA-based ternary OSCs. The proposed interpretable machine-learning strategy can contribute to the predictive modeling of highly efficient ternary OSCs based on multicomponent photoactive layers.