A high-throughput screening frame of polymer donor for efficient organic solar cell constructed by machine learning with encoded frequency molecular fingerprint

Abstract

The improvement in the performance of organic solar cells (OSCs) is constrained by the challenges of traditional trial-and-error methods in a vast and high-cost chemical space for the novel donor molecule design. In this work, a high-throughput screening frame with machine learning (ML) based on the frequency molecular fingerprint (FreMFp) is proposed to address the limitation of insufficient expression of polymer donor structures via traditional molecular fingerprints. The constructed CatBoost model with FreMFp of substructure function classification and frequency quantization exhibits excellent prediction performance on power conversion efficiency (PCE), which is significantly superior to the traditional molecular fingerprints as the descriptor. Interpretability analysis revealed the positive effects of key sub-units, liking 4-fluoro-2-methylthiophene (Sub-unit 45), dithieno [3 ',2':3,4; 2 '',3":5,6]benzo[1,2-c][1,2,5]thiadiazole (Sub-unit 178), 2-ethylhexyl (Sub-unit 196) in fused-ring skeleton, monocyclic-ring linker, and side-chain modification group on PCE. Further approximately six million virtual structures are generated by the screened key sub-units, and finally the D-it-A-it type and D-A type candidate donors that matched both acceptors of Y6 and L8-BO are screened out. More importantly, the typical representatives of D-it-A-it type (D18-D2Cl) and D-A type (PTQ10-EH) with L8-BO can obtain the predicted PCE of binary OSCs of 19.59 % and 17.79 %, respectively. And density functional theory (DFT) further demonstrates the regulation of energy level arrangement and charge transfer characteristics by the electron-induced effect and steric hindrance caused by halogen substitution in the efficient candidate molecule of D18-D2Cl. The built ML frame in this work provides an efficient and low-cost solution to design the donor materials for accelerating the development process of high-performance OSCs.