Perovskite materials have the potential to replace silicon in the production of solar cells that are more thinner, lighter, and less expensive. However, converting these materials into a product that can be made competitively has been a lengthy process. A new approach based on machine learning could accelerate the development of improved manufacturing procedures and contribute to the realization of this next generation of solar power.
Perovskites are a class of materials that are now the frontrunners in the race to replace silicon-based solar photovoltaics. They carry the promise of panels that are much smaller and lighter, that can be manufactured with ultra-high-throughput at room temperature rather than hundreds of degrees, and that are less expensive and easier to transport and install. But bringing these materials from controlled laboratory experiments into a product that can be manufactured competitively has been a long struggle.
Production of perovskite-based solar cells requires optimizing at least a dozen or so variables at the same time, even within a single manufacturing process among numerous options. However, a new system based on a revolutionary approach to machine learning could accelerate the development of improved production methods and contribute to the realization of the next generation of solar power.
The method, created over the last few years by researchers at MIT and Stanford University, allows for the integration of data from previous experiments as well as information based on personal observations by experienced workers into the machine learning process. This improves the accuracy of the results and has already resulted in the production of perovskite cells with an energy conversion efficiency of 18.5 percent, which is competitive in today’s market.
The research is reported in the journal Joule, in a paper by MIT professor of mechanical engineering Tonio Buonassisi, Stanford professor of materials science and engineering Reinhold Dauskardt, recent MIT research assistant Zhe Liu, Stanford doctoral graduate Nicholas Rolston, and three others.
There’s always a tremendous barrier when you’re trying to take a lab-scale technique and then move it to something like a startup or a production line. The real breakthrough with this platform is that it would allow us to scale in a way that no other material has allowed us to do. Even materials like silicon require a much longer timeframe because of the processing that’s done.
Tonio Buonassisi
Perovskites are a class of layered crystalline substances characterized by the arrangement of their atoms in the crystal lattice. There are dozens of such chemicals and numerous methods for creating them. While most lab-scale research of perovskite materials involves a spin-coating approach, this is not practicable for larger-scale manufacturing, thus corporations and labs throughout the world have been looking for ways to translate these lab materials into a practical, manufacturable product.
“There’s always a tremendous barrier when you’re trying to take a lab-scale technique and then move it to something like a startup or a production line,” says Rolston, who is now an assistant professor at Arizona State University. The team looked at a process that they felt had the greatest potential, a method called rapid spray plasma processing, or RSPP.
The manufacturing process would involve a moving roll-to-roll surface, or series of sheets, on which the precursor solutions for the perovskite compound would be sprayed or ink-jetted as the sheet rolled by. The material would then move on to a curing stage, providing a rapid and continuous output “with throughputs that are higher than for any other photovoltaic technology,” Rolston says.
“The real breakthrough with this platform is that it would allow us to scale in a way that no other material has allowed us to do,” he adds. “Even materials like silicon require a much longer timeframe because of the processing that’s done. Whereas you can think have like spray painting.”
At least a dozen variables may influence the result of the process, some of which are more controlled than others. These include the beginning material composition, temperature, humidity, the speed of the processing path, the distance of the nozzle used to spray the material onto a substrate, and curing procedures. Many of these elements can interact with one another, and if the process is carried out in the open air, humidity, for example, may be uncontrollable. Because it is impossible to evaluate all conceivable combinations of these factors through testing, machine learning was required to help guide the experimental procedure.
But while most machine-learning systems use raw data such as measurements of the electrical and other properties of test samples, they don’t typically incorporate human experience such as qualitative observations made by the experimenters of the visual and other properties of the test samples, or information from other experiments reported by other researchers. So, the team found a way to incorporate such outside information into the machine learning model, using a probability factor based on a mathematical technique called Bayesian Optimization.
“Having a model that comes from experimental data, we can figure out trends that we weren’t able to identify previously,” he adds of the system. For example, they originally had difficulty compensating for uncontrollable humidity swings in their ambient setting. However, the model demonstrated that “we could overcome our humidity difficulties by changing the temperature, for example, and by adjusting some of the other knobs.”
The system now allows experimenters to much more rapidly guide their process in order to optimize it for a given set of conditions or required outcomes. In their experiments, the team focused on optimizing the power output, but the system could also be used to simultaneously incorporate other criteria, such as cost and durability — something members of the team are continuing to work on, Buonassisi says.
The Department of Energy, which funded the research, encouraged the researchers to commercialize the technology, and they are currently focused on technology transfer to existing perovskite producers. “We’re reaching out to organizations right now,” Buonassisi adds, adding that the code they created is publicly available via an open-source site. “It’s now on GitHub, everyone can get it, anyone can execute it,” he explains. “We’re delighted to assist businesses in getting started with our code.”
“The problem is that they can’t agree on which production technique to utilize,” Liu explains. According to him, the Stanford-developed RSPP approach “still has a decent chance” of being competitive. And the machine learning system the team developed could prove to be important in guiding the optimization of whatever process ends up being used.
