Machine Learning For Virtual Solar Cells Trial

Engineers at Osaka University apply machine learning to develop and virtually experiment with molecules for organic solar cells. It can guide to higher productivity working materials for renewable energy applications.
Osaka University Engineers applied machine learning to create unique polymers for use in solar devices. After virtually testing over 200,000 candidate materials, they synthesized one of the most promising and found its characteristics were consonant with their predictions. This research may lead to a change in the way working materials are discovered.

Machine Learning  and Solar Cells properties
Chemical compositions of a polymer (left) and a non-fullerene acceptor (right). Source: Osaka University

Machine learning algorithms enable computers to execute predictions regarding complicated situations, as long as the algorithms are provided with enough example data. It is beneficial for complex problems, such as developing molecules for organic photovoltaic cells. It depends on a wide range of factors and unknown molecular structures.
It would need humans years to sort through the data to detect the underlying patterns—and even hard to test all of the potential candidate sequences of donor polymers and acceptor molecules that give an organic solar cell. Thus, the process of improving the productivity of solar cells to be competitive in the renewable energy space has been slow.
Now, engineers at Osaka University employed machine learning to test millions of donor: acceptor pairs based on an algorithm trained with data from earlier published experimental research. Analyzing all potential sequences of 382 donor molecules and 526 acceptor molecules resulted in 200,932 pairs that were virtually examined by predicting their energy conversion efficiency.

Machine Learning  and Solar Cells
Process for the evolution of the machine learning model, virtual generation of polymers, and selection of polymers for synthesis. Source: Osaka University

“Basing the development of our machine learning model on an experimental dataset enhanced the prediction correctness,” prime author Kakaraparthi Kranthiraja says.
To verify this process, one of the polymers predicted to have high productivity was synthesized in the lab and tested. Its characteristics were discovered to fit with predictions, which gave the engineers more trust in their approach.
“This project may contribute not only to the evolution of highly effective organic solar cells but also can be modified to material informatics of other working materials,” senior author Akinori Saeki says.
We may notice this kind of machine learning, in which an algorithm can immediately test millions of candidate molecules based on machine learning predictions, used in other fields, such as catalysts and functional polymers.

Reference: “Experiment‐Oriented Machine Learning of Polymer: Non‐Fullerene Organic Solar Cells” by Kakaraparthi Kranthiraja and Akinori Saeki, 25 February 2021, Advanced Functional Materials.

DOI: 10.1002/adfm.202011168

Funding: Japan Society for the Promotion of Science, Ministry of Education, Culture, Sports, Science and Technology, Japan Science and Technology Agency.

Leave a Reply

Your email address will not be published.