 |
A major challenge in the development of organic photovoltaic (PV) cells is obtaining the polymer materials that meet the design requirements. Because there are millions of possible combinations of molecules that go into polymer materials, researchers generally have to painstakingly synthesize and test each of the polymers they have created through trial-and-error experimentation. Recently, material scientists at Osaka University in Japan have found a way to shorten and optimize the process for determining the right combinations for polymers that will be used in organic PV cells. Instead of relying on the traditional experimentation and simulation, these scientists now use a computer-powered artificial intelligence (AI) system that automates the search for the suitable combinations. Their findings reveal that AI has the potential to radically improve the designs of organic PV cells and speed up their commercialization.
Organic PV cells have a chance to become the next big breakthrough in solar power because they have several important advantages over the conventional PV cells that are made of crystalline silicon. First, organic PV cells can be made on flexible substrates, thus offering new design possibilities. Second, a very efficient light absorption capability can be derived from a thin layer of organic material, while additional conductive polymers or molecular units can assist in this process and in the transfer of electric charges. Lastly, organic PV cells can be mass produced using a simple printing process with low-cost materials.
However, the conversion efficiency rates of the current generation of organic PV cells are within the range of 11-12%. The overall performance of the organic cells is therefore still not good enough for full-scale commercialization, which requires an efficiency rate of at least 15%. The main reason why researchers have been unable to substantially raise the conversion efficiency of organic cells is the challenge of devising, synthesizing, and testing polymers for solar materials. Shinji Nagasawa, a member of the research team at Osaka University, pointed out that the properties of polymers can affect the flow of the short-circuit current, which in turn can have an impact on the cell’s conversion efficiency.
Akinori Saeki, another member of the research team, added that a polymer has a complex structure composed of many sub-units (e.g. a donor unit, an acceptor unit, a spacer, and alkyl chains). Saeki said that even if the number of possible choices of each sub-unit is limited to 20, the number of possible combinations for researchers to synthesize will still exceed 1 million. Saeki also noted that there are many other factors affecting the conversion efficiency of an organic PV cell besides the properties of the polymer materials. These factors include the film morphology, the p-n junction, and the solubility of the materials. The applications of advanced modeling, even ones based on quantum chemistry, cannot accurately predict the efficiency of a solar cell.
Since testing individual polymer samples is time and resource consuming, the team at Osaka University hit upon the idea of using AI to accelerate the searching and screening of samples.
 |
(Credit: Osaka University)
The researchers first created a data set that included the properties of 1,200 organic cells from 500 studies. Then, they used a type of machine learning algorithm called Random Forest to build a custom model that uses the data set as a reference to predict the theoretical efficiency of a potentially new polymer sample. This AI-based model takes account of important polymer attributes such as band gap, molecular weight, chemical structure, and electronic properties.
With the incorporation of the Random Forest algorithm, the model has an enhanced ability to find the connections between properties of materials and their contributions to the efficiency of an organic PV cell. Researchers can screen prospective polymers based on their theoretical efficiency rates that are calculated by the model. This faster filtering process also allows researchers to discover new polymers that have not been tested previously.
The AI-based model did not yield the expected results in the actual trials. Nevertheless, it has given researchers a greater understanding of the relationship between the structure and properties of a material. The research team stated that more parameters such as the polymer’s solubility in water will be added into the model to improve its applicability.
Saeki said that the model is far from perfect as its accuracy is around 20-50%, but it can immediately forecast similar results that would take months to produce with experiments and simulations in a laboratory setting. Hence, this AI-based tool can dramatically accelerate the design and development of solar cells. While the Random Forest algorithm cannot fully replace the human thought process, it can be used to assist molecular designers in deciding the suitable combination for a polymer and lighten their workload.
The description of the model and the findings of the research team at Osaka University were published in the Journal of Physical Chemistry Letters on 7 May 2018.
(The above article is an English translation of a Chinese article written by Daisy Chuang. The credit of the photo at the top of the article goes to Katy Warner via Flickr and falls under the license of CC BY 2.0.)