Accelerating the discovery of N-annulated perylene organic sensitizers via an interpretable machine learning model

Abstract

A data-driven strategy was proposed by combining machine learning (ML) model and first-principles verification to rapidly quarry efficient and synthesizable candidate organic sensitizers from the vast chemical space of dye-sensitized solar cell (DSSC) sensitizers. We built interpretable machine learning models using easy-to-obtain descriptors and obtained 50 candidates which could be synthesized and have the overall power conversion efficiency (PCE) greater than 13 %. Especially, the density functional theory (DFT) and time-dependent density functional theory (TD-DFT) showed that top 2 candidates had better open-circuit Voltage (Voc) and PCEs than C281. And the PCE predicted by the ML model was roughly consistent with the ones calculated by quantum chemistry. Model analysis revealed that Cobalt-based electrolyte was the better option. Paid attention to the electronegativity and molecular weight of doner of N-annulated perylene organic sensitizers (N-P sensitizers), appropriately increased the electronegativity group of accepter fragment. Hopefully, this data-driven approach is expected to provide further creativity for the practical application of other advanced energy materials.