WHY THIS MATTERS IN BRIEF
Trying to find the right formulations to create new materials is a long and arduous process, so scientists are turning to AI to help them speed the process up.
As we see several ways to increase solar panel efficiency from today’s meagre record breaking 27 percent and up to 80 percent efficiency, researchers in the US are now suggesting that Artificial Intelligence (AI), which is increasingly being used to help accelerate scientific discoveries, and create new drugs and new materials, may be just the thing to accelerate the development of spray on solar cells that, according to the team behind the idea, “could revolutionize how consumers use energy.”
The team from the University of Central Florida used Machine Learning to optimize the materials used to make Perovskite Solar Cells (PSC’s) that today are already being used to plot a path to 32 percent solar panel efficiency and, when combined with “cyborg bacteria” even 50 percent efficiencies.
The Organic-Inorganic perovskite material used in PSC’s convert photovoltaic power into consumable energy can be processed in solid or liquid state so once the researchers and their new AI Robo-Scientists hit the magic formula imagine, for example, being able to spray or paint bridges, houses and skyscrapers with the material which would then capture light, turn it into energy and feed it into the electrical grid.
Until now, the solar cell industry has relied on silicon because of its efficiency but that’s old technology with significant limits, but using perovskites, however, has several barriers, they’re brittle to make and difficult to turn into usable and stable materials so scientists spend a lot of time trying to find just the formulas that have the right combination of flexibility, stability, efficiency and low cost. And the team’s most recent work has been so promising that their findings are the cover story in this months Advanced Energy Materials journal.
During their research they reviewed more than 2,000 peer reviewed publications about perovskites and collected more than 300 data points then fed it all into their new AI model. The system was able to analyze the information and predict which perovskites recipe would work best.
“Our results demonstrate that machine learning tools can be used for crafting perovskite materials and investigating the physics behind developing highly efficient PSCs,” says Jayan Thomas, the study’s lead author and an associate professor at the NanoScience Technology Center with multiple affiliations. “This can be a guide to design new materials as evidenced by our experimental demonstration.”
If this model bears out, it means researchers could identify the best formula to create a world standard, and that spray-on solar cells “may happen in our lifetime,” the researchers say.
“This is a promising finding because we use data from real experiments to predict and obtain a similar trend from the theoretical calculation, which is new for PSCs. We also predicted the best recipe to make PSC with different bandgap perovskites,” says Thomas and his graduate student, Jinxin Li, who is the first author of this paper. “Perovskites have been a hot research topic for the past 10 years, but we think we really have something here that can move us forward.