Computational design of new polymers having low exciton binding energy for organic solar cells fabrication: Chemical generation and visualization

Abstract

Conventional organic solar cell (OSC) fabrication, based on the hit and trial methodology, is quite laborious and lethargic, requiring extensive human workforce and instrumental resources. Recent advancements in machine learning (ML) approaches have revolutionized selection of the suitable active polymeric layer for OSC devices. Among numerous aspects impacting efficacy of the OSC fabrication, the lower exciton binding energy (Eb) was found one of the most crucial factors. The dataset containing Eb values was prepared from already documented literature on polymeric active layers of OSC devices. Various ML models were applied and tested to predict Eb values accurately, however, the random forest regressor model was proved to be the most efficient one. The ML analysis was further utilized to develop a new and novel chemical database that could be utilized as an active layer for the fabrication of OSCs. The synthetic accessibility (SA) score was also investigated to explore the relative ease by which a proposed polymeric material could be synthesized in practical lab. Interestingly, analysis of the top thirty generated polymeric materials exhibited extremely decent SA score, and remarkably lower Eb values; both properties are highly desirable for the fabrication of efficient OSCs. Further, we also applied deep learning approach to generate and screen new monomers with lower Eb and SA score. This effort reveals significance of the computational data-driven approaches for real-life applications. Our current contribution of the novel polymeric materials database having extremely low Eb values could be valuable for the fabrication of efficient OSC devices.