Artificial Intelligence in Organic Photovoltaics: Predicting Power Conversion Efficiency From the Molecular Chemical Structure of (Donor/Acceptor) Pairs

Abstract

Organic solar cells (OSCs) can achieve power conversion efficiencies around 20%. Yet, further improvements in efficiency and long-term stability are necessary to rival the dominant silicon technology. Key factors influencing OSC performance include device architecture and the active-layer semiconducting organic materials. In this study, we utilize artificial intelligence (AI) techniques to analyze an experimental dataset of organic semiconductors used in the active layer of OSCs. We propose an AI-based methodology to predict the performance of OSCs using the chemical structure of Donor-Acceptor (D/A) pairs. The method employs Simplified Molecular Input Line Entry System (SMILES) representations to extract molecular features. These features, selected according to maximum relevance and minimum redundancy criteria, are used by supervised machine learning regression algorithms to predict the main photovoltaic parameters. Our AI model demonstrates significant predictive power. Further, we use our model to predict the photovoltaic parameters of (D/A) pairs that were not included in our initial dataset. These findings highlight the potential of AI-driven analysis to accurately estimate the photovoltaic potential of new (D/A) pairs before synthesizing them and therefore to accelerate the development of commercially viable OPV devices and to lower the materials research cost.