Generative Design and Experimental Validation of Non-Fullerene Acceptors for Photovoltaics

Abstract

The utilization of non-fullerene acceptors (NFA) in organic photovoltaic (OPV) devices offers advantages over fullerene-based acceptors, including lower costs and improved light absorption. Despite advances in small molecule generative design, experimental validation frameworks are often lacking. This study introduces a comprehensive pipeline for generating, virtual screening, and synthesizing potential NFAs for high-efficiency OPVs, integrating generative and predictive ML models with expert knowledge. Iterative refinement ensured the synthetic feasibility of the generated molecules, using the diketopyrrolopyrrole (DPP) core motif to manually generate NFA candidates meeting stringent synthetic criteria. These candidates were virtually screened using a predictive ML model based on power conversion efficiency (PCE) calculations from the modified Scharber model (PCE_MS). We successfully synthesized seven NFA candidates, each requiring three or fewer steps. Experimental HOMO and LUMO measurements yielded calculated PCE_MS values from 6.7% to 11.8%. This study demonstrates an effective pipeline for discovering OPV NFA candidates by integrating generative and predictive ML models.