Accelerating the generation and discovery of high-performance donor materials for organic solar cells by deep learning

Abstract

Organic solar cells are a potential solution for sustainable energy. However, it is time-consuming and costly to discover high-efficiency materials for organic solar cells. In this study, a deep learning-based framework (DeepDonor) has been developed to find high-performance donor materials. Specifically, a small molecule (SM) dataset and a polymer molecule (PM) dataset were collected from the literature. Then, the quantum deep field (QDF) model and transfer learning method were used to predict the power conversion efficiency (PCE). Results show that DeepDonor significantly outperforms other common methods with r = 0.82, MAE = 1.67% on the SM test set, and r = 0.77, MAE = 1.59% on the PM test set. These models can predict PCE with great accuracy, extrapolation performance, and transferability. Then, a molecule generation and screening strategy was proposed to discover high-performance polymer donors with the help of DeepDonor. Two discovered candidates were further validated by experiments, and their PCE reached 16.27% and 15.07%, respectively. The released online interface of DeepDonor substantially boosts the accessibility and efficiency of designing and discovering high-performance donors. In sum, DeepDonor is a promising method to predict the PCE and speed up the discovery of high-performance donor materials.