Achieving better photovoltaic materials faster with AI
How machine learning greatly accelerates the search for new semiconducting molecules for perovskite solar cells
perovskite solar cells are considered a flexible and sustainable alternative to conventional silicon-based solar cells. Within a period of a few weeks, researchers have now found new organic molecules that can be used to increase the efficiency of perovskite solar cells. The international team includes scientists from the Helmholtz Institute Erlangen-Nürnberg for Renewable Energy (HI ERN), a branch office of Forschungszentrum Jülich. The scientists skilfully combined the use of AI with fully automated high-throughput synthesis. The strategy developed can be transferred to other areas of materials research, such as the search for new battery materials. The results of the research work were published in the journal Science.
If you want to find the one million molecules that make perovskite solar cells particularly efficient as conductors of positive charge, you have to produce and test these million molecules - or proceed as researchers led by tenure-track professor Pascal Friederich from the Institute of Nanotechnology at KIT and Professor Christoph Brabec from HI ERN have done. "With just 150 targeted experiments, a breakthrough was achieved that would otherwise have required hundreds of thousands of tests. The workflow developed opens up new possibilities for the rapid and cost-effective discovery of high-performance materials in a variety of application fields," says Brabec. With one of the materials discovered in this way, they increased the efficiency of a reference solar cell by around 2 percent to 26.2 percent. "This success shows that a clever strategy can save an enormous amount of time and resources when developing new energy materials," says Friederich.
The starting point at HI ERN was a database with the structural formulas of around one million virtual molecules that could be produced from commercially available substances. The researchers at KIT used established quantum mechanical methods to calculate the energy levels, polarity, geometry and other characteristics of 13,000 of these randomly selected virtual molecules.
AI training with data from just 101 molecules
From these 13,000 molecules, the researchers then selected 101 molecules that differed as much as possible in their characteristics. These were automatically produced at HI ERN with the help of a robotic system and used to manufacture otherwise identical solar cells. They then measured their efficiency. "Thanks to our highly automated synthesis platform, it was crucial for the success of our strategy that we produced truly comparable samples and thus determined reliable values for the efficiency," says Christoph Brabec, who led the work at HI ERN.
The KIT researchers used the efficiency values obtained and the characteristics of the associated molecules to train an AI model. The model then suggested a further 48 molecules for synthesis, based on two criteria: an expected high efficiency and unpredictable properties. "If the machine learning model is uncertain about predicting the efficiency, it is worth producing the molecule to study it in more detail," says Pascal Friederich, explaining the second criterion. "It could surprise us with a high degree of efficiency."
In fact, the molecules proposed by the AI could be used to build solar cells with above-average efficiency, including those that outperform other state-of-the-art materials. "We can't be sure that we have really found the best of a million molecules, but we are certainly close to the optimum," says Friederich.
AI versus chemical intuition
The researchers can follow the AI's molecular suggestions to a certain extent, as the AI used indicates which characteristics of the virtual molecules were decisive for its suggestions. It turned out that the AI suggestions were also partly based on characteristics that chemists had previously paid less attention to, for example the presence of certain chemical groups such as amines.
Christoph Brabec and Pascal Friederich are convinced that their strategy is also promising for materials research in other areas of application or can be extended to the optimization of entire components.
The research results, which were obtained in collaboration with researchers from the University of Erlangen-Nuremberg, the South Korean Ulsan National Institute of Science, the Chinese Xiamen University and the University of Electronic Science and Technology in Chengdu, China, were recently published in the journal "Science".
Note: This article has been translated using a computer system without human intervention. LUMITOS offers these automatic translations to present a wider range of current news. Since this article has been translated with automatic translation, it is possible that it contains errors in vocabulary, syntax or grammar. The original article in German can be found here.
Original publication
Jianchang Wu, Luca Torresi, ManMan Hu, Patrick Reiser, Jiyun Zhang, Juan S. Rocha-Ortiz, Luyao Wang, Zhiqiang Xie, Kaicheng Zhang, Byung-wook Park, Anastasia Barabash, Yicheng Zhao, Junsheng Luo, Yunuo Wang, Larry Lüer, Lin-Long Deng, Jens A. Hauch, Dirk M. Guldi, M. Eugenia Pérez-Ojeda, Sang Il Seok, Pascal Friederich, Christoph J. Brabec; "Inverse design workflow discovers hole-transport materials tailored for perovskite solar cells"; Science, Volume 386
Other news from the department science
![[Fe]-hydrogenase catalysis visualized using para-hydrogen-enhanced nuclear magnetic resonance spectroscopy](https://img.chemie.de/Portal/News/675fd46b9b54f_sBuG8s4sS.png?tr=w-712,h-534,cm-extract,x-0,y-16:n-xl)
