Multidimensional modeling assisted mining of GDB17 chemical database: A search for polymer donors for organic solar cells and machine learning assisted performance prediction

Indeed, the designing of efficient material for organic solar cells is challenging and time-consuming task. In this work, we have proposed a design pipeline to screen efficient polymer donors. Machine learning based models are trained to predict the power conversion efficiency (PCE). Structural descriptors are used as input to train machine learning models. The data of organic molecules is extracted from Photovoltaics database. Best buildings blocks are selected on the basis of similarity analysis using PBT7-Th and MP6 as reference (standard). New polymers are designed using building blocks having high similarity. The PCE is predicted with the help of high-performance machine learning model. Top three candidates are studied by molecular dynamics (MD) simulations. Packing behavior of polymer donors and their blend with fullerene acceptor is studied using radial distribution function (RDF). Density functional theory analysis is also performed on selected polymers. Electrostatic potential analysis has indicated the highly polar behavior that can result higher charge generation. Our proposed pipeline has ability to design and screen the polymer donors, as well as can predict their PCE without any experimentation. More importantly computational cost is marginal.

Automated summary

This work proposes a design pipeline to screen efficient polymer donors for organic solar cells, using machine learning models to predict power conversion efficiency. Molecular dynamics simulations and density functional theory analysis are employed to study the top three candidates. The pipeline has the ability to design and screen polymer donors and predict their efficiency with minimal computational cost.