Machine Learning in Non-fullerene Organic Solar Cells: Accelerating Discovery, Design, and Understanding.

Area of Science:

• Materials Science
• Organic Electronics
• Computational Chemistry

Background:

• Nonfullerene acceptors (NFAs) have significantly improved organic solar cell (OSC) efficiencies, exceeding 19%.
• The vast number of potential donor-acceptor materials makes experimental exploration challenging.
• Machine learning (ML) and artificial intelligence offer powerful solutions for accelerating materials discovery and design in organic photovoltaics.

Purpose of the Study:

• To systematically review and assess predictive and generative ML approaches for organic solar cells (OSCs).
• To evaluate the effectiveness of ML in predicting device performance metrics and guiding molecular design.
• To identify limitations and future directions in ML-driven organic photovoltaic research.

Main Methods:

• Review of various ML methods including random forests, gradient boosting, graph neural networks, transformers, variational autoencoders, generative adversarial networks, and genetic algorithms.
• Examination of molecular representation and feature engineering techniques.
• Evaluation of model performance in predicting power conversion efficiency, open-circuit voltage, short-circuit current density, and fill factor.

Main Results:

• ML models show promise in predicting key device metrics for organic solar cells.
• Advances in ML are enabling faster materials discovery and molecular design for organic photovoltaics.
• Key limitations such as data bias, chemical validity, and synthetic accessibility persist.

Conclusions:

• Machine learning is becoming integral to the rational design and accelerated development of advanced organic photovoltaic materials.
• Emerging ML directions like physics-informed learning and active learning are poised to drive future progress.
• Bridging the gap between predicted and realized performance remains a critical challenge.