Environmentally friendly solar panels using organic dyes aren’t commercially viable yet. But artificial intelligence and chemistry can make them a reality, explains Professor Feng Wang.
Organic solar cells are closer to use than ever before, and Professor Feng Wang believes artificial intelligence (AI) can speed up the search for the right material.
Although harnessing solar energy is more sustainable than burning coal, almost all conventional rooftop solar panels capture the sun’s energy with a layer of silicon, which is expensive and energy-intensive to make, demanding temperatures well above 1,000°C. Organic solar cells, which, instead of silicon, have a light-capturing layer made of carbon-based molecules known as dyes, are cheaper, easier and much less energy-intensive to make - the dye layer can be applied as an ink, using rapid roll-to-roll machines like those used to print newspapers.
“Currently, the big issue with commercialising organic solar cells is that their conversion rate for turning solar energy into electricity is very low compared to silicon,” explains Wang. Whereas rooftop silicon cells convert solar energy into electricity at almost 20% efficiency, organic solar cells have recently been closer to 10%.
In the past few years, a handful of organic solar materials have been discovered that achieved efficiencies in the high teens. In 2018, one group co-led by researchers from Nankai University and the National Center for Nanoscience and Technology in China published on a material in Science that claimed an efficiency of 17% — in the right ballpark for potential commercial viability, if the material proves durable enough. These discoveries, she adds, have inspired new optimism in the sector, suggesting there could be even more efficient organic solar materials waiting to be discovered.
Molecule pick and mix
One problem is that there are millions of dyes with potential for use in solar cells — making and testing even a small fraction of these in the lab is a huge task. Until now, organic solar materials have typically been discovered by accident, or by researchers applying their chemical intuition.
However, the reason some dyes can capture more energy from sunlight, is simply down to the combination and arrangement of their atoms, says Wang, who leads research in intelligent atomic design at Swinburne’s Centre for Translational Atomaterials (CTAM). The dyes typically used in organic dye sensitised solar cells are a combination of two units, an electron donor unit and an electron acceptor unit, which are physically connected by an electrically conductive ‘bridge’.
The bridging component contributes to sunlight absorption, and helps ensure the electric charges generated when light is absorbed can easily flow through the material to be gathered by the solar panel’s electrodes. The amount of solar energy captured depends on the exact combination of donor, acceptor and bridge components.
However, even just understanding the best molecules from which to build the bridge is daunting. For example, 26 building blocks have previously been identified that can be combined to make bridges of different lengths, but, just like the 26 letters of the alphabet fill a dictionary with different words, the number of possible bridge configurations is overwhelming. “With 26 building blocks, you have millions of possible dyes. We can’t calculate millions yet,” says Wang.
This is where AI could help, she says. Some bridge structures have already been synthesised and tested in the lab, others have been computationally simulated. “If we can use this existing data, the computer can learn. We can use machine learning to pick up the trends and give us suggestions,” says Wang. The AI will generate a shortlist of promising potential bridge structures that can be made and tested.
Together with Dr Rui Zhou, from Swinburne’s Department of Computer Science and Software Engineering, and his student, Minh Tai Nguyen, Wang is developing a computer program to do this called Dyemaker.
“We have a basic prototype, a graphical user interface that can already start to help us design new compounds,” Wang says. A chemist can use the prototype to manually try different combinations of electron donor, electron acceptor and bridge components, and the program will calculate the dye’s sunlight absorbing properties.
The next step will be to incorporate the AI into this shell, and to plug data that the AI algorithm can learn from. “Once we learn the strategy, we can tap into other solar material databases, such as the Harvard clean energy database,” explains Wang.
The Harvard clean energy dataset is one of the world’s most comprehensive and lists millions of organic solar structures that have had their properties measured in the lab or predicted by computer modelling. At that point, the computer should be able to automatically and intelligently combine different dye components to predict combinations with optimal light harvesting performance.
Identifying the best bridge components will be just the start of AI’s contribution to organic solar cell design, Wang predicts. This project will be the proof of concept she says. Then other aspects of organic solar cell design can be optimised using the same approach, by plugging into different datasets. (The larger the dataset the AI has to learn from, the more useful its predictions should be.)
Wang’s ambitions to combine AI and chemistry don’t end with solar harvesting materials. She hopes to use the same AI technologies to optimise organic compounds in areas such as materials and drug design.
“In chemistry, we have a golden rule: structure dictates properties and functionality,” says Wang. Better predictive mechanisms and computer modelling of chemical components could have implications for many of the projects underway in Wang’s lab, which range from designing more active chemical catalysts to reporting on anti-cancer drug-protein binding. So AI should help to plug a huge gap in chemistry research, explains Wang. Theoretical chemists like herself, she explains, take a bottom-up approach, using computers to simulate different combinations of atoms to create new materials and predict their properties. Experimental chemists take a top-down approach, making a series of materials with particular structures, then testing their properties to try to identify performance trends. But the two approaches don’t quite meet.
“Between the top-down and bottom-up approach, there’s a big gap,” Wang says. Experimental chemists can’t fully drill down to probe the properties of each individual molecule within a substance to establish clear structure-property relationships. Whereas, computers often don’t have the power to simulate the behaviour of large assemblies of molecules interacting as they would in real materials for theoretical chemists.
But as AI becomes more sophisticated, it’s becoming ever-more capable of simulating the complexities of real-world materials, says Wang: “We are getting better and better at closing this gap.” She predicts, effective applications of AI to chemistry will bring with them a trove of world-changing discoveries.