The La FREMD Team, led by sun Kuan, a researcher of the School of Energy and Power Engineering, Chongqing University, and a number of research groups inside and outside Chongqing University have recently succeeded in screening organic photovoltaic materials through machine learning. This achievement was published in Science Advances (impact factor = 12.8), a sub-journal of Science under the title "machine learning assisted molecular design and efficiency prediction for high performance organic photovoltaic materials”, with Chongqing University as the first corresponding organization. Sun Kuan, a researcher of Chongqing University, Lu Shirong and Xiao Zeyun, researchers of Chongqing Institute of Green and Intelligent Technology, Chinese Academy of Sciences are co-authors, and Sun Wenbo, Dr. Zheng Yujie and Yang Ke, a doctoral candidate, are first co-authors. So far, Researcher Sun Kuan has published four original papers as the first or corresponding author in the sub-journals of Cell, Nature and Science (CNS) since he started to work for Chongqing University five years ago.

Organic solar cell is a direct and economical way to convert solar energy into electric energy. In recent years, the research of organic solar cells has experienced rapid development, and the photoelectric conversion efficiency (PCE) has exceeded 17%. At present, the research of organic photovoltaic mainly focuses on the relationship between the molecular structure of new materials and their photovoltaic properties. This process usually involves the design and synthesis of photovoltaic materials, the characterization of photoelectric properties of materials, and the assembly and optimization of photovoltaic cells. This traditional method includes fine control and optimization of chemical synthesis and device preparation, which requires a lot of resources and implicates a long research cycle, so the development of organic photovoltaic materials has been slow. Since the organic solar cell was first reported in 1973, there have been less than 2,000 kinds of donor materials synthesized and tested in photovoltaic devices. However, the experimental data generated in the past half century through exploration are valuable. So far, however, the potential value of the data has not been fully utilized in the search for high-performance organic photovoltaic materials. In order to extract useful information from these data, we need a program that can scan a large number of data sets and extract the relationship between features. Machine learning is an algorithm that meets the requirements. It provides a set of calculation tools, which are able to learn and identify patterns or relationships, predict results or make decisions according to the minimization of errors (or loss functions) or probability rules (such as maximum likelihood). This data-driven approach enables machine learning to predict a wide range of material properties without the need for an in-depth understanding of the chemical or physical principles behind them.

Recently, the La FREMD Team from the School of Energy and Power Engineering of Chongqing University, in collaboration with many teams from the School of Automation, the School of Computer Science, the School of Economics and Business Administration, North China University of Science and Technology, and Chongqing Institute of Green and Intelligent Technology, Chinese Academy of Sciences, used machine learning to establish chemical structure and relationship between the photovoltaic properties before synthesizing new materials in the process of developing high-performance organic solar cell donor materials. And the efficiency of new materials was predicted. In this Project, a database covering more than 1,700 kinds of organic solar cell donor materials has been established. Through supervised learning, the machine learning model of project design was able to establish the "structure performance" relationship, so as to realize the rapid screening of organic photovoltaic materials. The research team probed into several expression forms of molecular structure, such as molecular structure diagram, ASCII code string, molecular descriptor and molecular fingerprint, as the input of various machine learning algorithms. It was found that the molecular fingerprint with a length of more than 1,000 bits was able to obtain high prediction accuracy, which was the best expression form for such machine learning. In addition, the author used machine learning model to predict 10 kinds of new designed donor materials. The predicted results of the model were consistent with the experimental results, which further verified the reliability of machine learning method. The results showed that machine learning was a powerful tool to pre-evaluate and screen new organic photovoltaic materials, and this method could accelerate the development of organic solar cell field.

The research work was supported by the National Natural Science Foundation of China, the National Special Instrument Project, the One-hundred-Talent Program of Chinese Academy of Sciences and the project of Chongqing Science and Technology Bureau. At the same time, the author would like to thank Academician Li Yongfang of the Institute of Chemistry, Chinese Academy of Sciences for his valuable suggestions on the research work.

Link of the full paper: https://advances.sciencemag.org/content/5/11/eaay4275