Machine learning-assisted performance prediction and molecular design of all-small-molecule organic solar cells based on the Y6 acceptor

Abstract

Organic solar cells (OSCs) have received considerable attention as a promising photovoltaic technology. However, it is time-consuming and laborious to design and synthesize high-performance material molecules for OSCs by conventional trial-and-error methods. Data-driven machine learning (ML) can leverage abundant amounts of trustworthy materials data to extract meaningful information, mine potential relationships, and construct scientific models to make reasonable predictions. In this work, 2 decision tree-based models were constructed for predicting power conversion efficiency (PCE) of binary all-small-molecule OSCs based on Y6 acceptor, which both exhibited satisfactory performance. And then, 9673 potential small molecular donor molecules were automatically generated by combination of molecular scaffolds and molecular fragments for virtual screening. The donor molecules with the highest predicted PCE were further analyzed by density functional theory (DFT), including UV–Vis absorption and energy level. Quantum chemical calculations analyzed and verified that the virtual screened 15 donor molecules with high PCE. This work provides a systematic framework for the design and discovery of innovative and promising donor molecules, thereby being expected to accelerate the development of OSCs.