Machine Learning-Assisted Development of Organic Solar Cell Materials: Issues, Analyses, and Outlooks

Abstract

Nonfullerene, a small molecular electron acceptor, has substantially improved the power conversion efficiency of organic photovoltaics (OPVs). However, the large structural freedom of π-conjugated polymers and molecules makes it difficult to explore with limited resources. Machine learning, which is based on rapidly growing artificial intelligence technology, is a high-throughput method to accelerate the speed of material design and process optimization; however, it suffers from limitations in terms of prediction accuracy, interpretability, data collection, and available data (particularly, experimental data). This recognition motivates the present Perspective, which focuses on utilizing the experimental data set for ML to efficiently aid OPV research. This Perspective discusses the trends in ML-OPV publications, the NFA category, and the effects of data size and explanatory variables (fingerprints or Mordred descriptors) on the prediction accuracy and explainability, which broadens the scope of ML and would be useful for the development of next-generation solar cell materials.

A Theoretical Perspective on the Thermodynamic Stability of Polymer Blends for Solar Cells: From Experiments to Predictive Modeling

“…The efficiency and the role of strategies based on artificial intelligence are expected to grow rapidly in next years, also considering likely improvements of ML algorithms, which is a very active rising research field. [195,[209][210][211][212][213][214] Furthermore, several freely available databases have been made easily accessible through web interfaces, such as AFLOW, Materials Project, NOMAD, Materials Cloud. [215][216][217][218] Nevertheless, when dealing with stability, the complexity of the blends and entropic effects plays an important role and it is difficult to include them in the atomistic models affordable by only ab initio methods.…”

Designing Efficient Tandem Organic Solar Cells with Machine Learning and Genetic Algorithms

“…To expand the training data, a new dataset containing 1,225 donor/NFA pairs, of which 1,001 are unique, is prepared from hundreds of publications, containing photovoltaic performance such as opencircuit voltage (V OC ), short-circuit current (J SC ), fill factor (FF), and PCE, and optical and electronic data such as ionization en-ergy, electron affinity, optical bandgap, and wavelength of maximum absorption. Unlike other datasets [19][20][21] , the donors and NFAs include both polymers and small molecules. A distribution of the experimental PCE in this data set (ESI Fig.…”

“…However, this dataset did not take into account the Y-series acceptors, a new class of materials incorporated in many high-performing OSCs. In a followup paper 20 , they expanded this dataset to 1,318 pairs, the current largest dataset containing NFAs, including multiple classes of NFAs such as A-D-A, subphthalocyanines, perylene bisimides (PBI), fullerene-appended molecules, and a few others. They trained a random forest model on this data set with a predictive performance of R 2 of 0.71.…”

Improved Predictions of Organic Photovoltaic Performance through Machine Learning Models Empowered by Artificially Generated Failure Data

“…In this study, we investigated the impact of data selection and the introduction of failure data on ML-based predictions of NFA OPV polymer properties. The experimental data (1318 entries taken from 558 papers) reported in our previous study 27 were subjected to a random forest regression. Although the data selection directly improved the prediction accuracy (r = 0.97), the applicability of the ML model to unknown polymers was slightly lower.…”