Perovskites are a family of materials that are currently the leading candidates to potentially replace today’s silicon-based solar photovoltaics. They promise much thinner and lighter panels, which could be manufactured with ultra-high throughput at room temperature instead of hundreds of degrees, and which are cheaper and easier to ship and install. But turning these materials from controlled laboratory experiments into a product that can be produced competitively has been a long struggle.
Manufacturing perovskite-based solar cells involves optimizing at least a dozen variables at once, even within a particular manufacturing approach among many possibilities. But a new system based on a new approach to machine learning could accelerate the development of optimized production methods and help make the next generation of solar power a reality.
The system, developed by researchers at MIT and Stanford University over the past few years, allows data from previous experiments and information based on personal observations of experienced workers to be integrated into the machine learning process. . This makes results more accurate and has already led to the fabrication of perovskite cells with an energy conversion efficiency of 18.5%, a competitive level for today’s market.
The search is reported today in the review Joulein an article by MIT Professor of Mechanical Engineering Tonio Buonassisi, Stanford Professor of Materials Science and Engineering Reinhold Dauskardt, recent MIT Research Assistant Zhe Liu, Stanford Ph.D. Nicholas Rolston, and three others .
Perovskites are a group of layered crystalline compounds defined by the configuration of atoms in their crystal lattice. There are thousands of these possible compounds and many different ways to make them. While most lab-scale perovskite material developments use a spin-coating technique, this is impractical for large-scale fabrication, so companies and labs around the world have sought ways to translate these laboratory materials into a practical, manufacturable product.
“There’s always a big challenge when you’re trying to take a lab-scale process and then transfer it to something like a startup or a manufacturing line,” says Rolston, who is now an assistant professor at Arizona State University. The team looked at a process they believed 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, onto which the precursor solutions of the perovskite compound would be sprayed or ink-jetted as the sheet rolled. The material would then go through a curing stage, providing fast, continuous output “with higher flow rates than any other photovoltaic technology,” Rolston says.
“The real breakthrough of this platform is that it would allow us to evolve in a way that no other material allowed us to do,” he adds. “Even materials like silicon require a much longer lead time due to the processing done. So that you can think of [this approach as more] like spray painting.
In this process, at least a dozen variables can affect the outcome, some of them more controllable than others. These include the composition of the starting materials, temperature, humidity, speed of the processing path, the distance from the nozzle used to spray the material onto a substrate, and the methods of curing the material. Many of these factors can interact with each other, and if the process is in the open, humidity, for example, can get out of control. Assessing all possible combinations of these variables through experimentation is impossible, so machine learning was needed to help guide the experimental process.
But while most machine learning systems use raw data such as measurements of electrical and other properties of test samples, they generally do not incorporate human experience such as qualitative observations made by experimenters of the properties. visual and other findings from test samples, or information from other experiments reported by other researchers. So the team found a way to feed this outside information into the machine learning model, using a probability factor based on a mathematical technique called Bayesian optimization.
Using the system, he says, “along with a model from experimental data, we can uncover trends that we couldn’t see before.” For example, they initially found it difficult to adapt to uncontrolled variations in humidity in their surrounding environment. But the model showed them “we could overcome our humidity issues by changing the temperature, for example, and changing some of the other buttons.”
The system now allows experimenters to guide their process much more quickly to optimize it for a given set of conditions or required results. In their experiments, the team focused on optimizing power output, but the system could also be used to simultaneously incorporate other criteria, such as cost and durability, which members of the team continue to work, explains Buonassisi.
The researchers were encouraged by the Department of Energy, which sponsored the work, to commercialize the technology, and they are currently focused on transferring technology to existing perovskite manufacturers. “We’re reaching out to businesses now,” says Buonassisi, and the code they’ve developed has been made freely available through an open-source server. “It’s now on GitHub, anyone can download it, anyone can run it,” he says. “We are happy to help companies start using our code.”
Already, several companies are gearing up to produce perovskite-based solar panels, though they’re still working out the details of how to make them, says Liu, who currently works at Northwestern Polytechnical University in Xi’an, China. He says the companies there are not yet doing large-scale manufacturing, but are instead starting with smaller, high-value applications such as building-integrated solar tiles where appearance is important. Three of these companies “are on the right track or are being pushed by investors to manufacture rectangular modules of 1 meter by 2 meters [comparable to today’s most common solar panels]within two years,” he said.
“The problem is that they don’t have a consensus on what manufacturing technology to use,” Liu says. The RSPP method, developed at Stanford, “still has a good chance” of being competitive, he says. And the machine learning system developed by the team could prove important in guiding the optimization of any process ultimately used.
“The main goal was to speed up the process, so it took less time, less experiments and less human hours to develop something that was immediately usable, free of charge, for the industry,” he says.
“Existing work on machine learning-based perovskite PV fabrication largely focuses on spin-coating, a lab-scale technique,” says Ted Sargent, a university professor at the University. of Toronto, who was not associated with this work, which he says demonstrates “a workflow that easily adapts to the deposition techniques that dominate the thin film industry. Only a handful of groups have the simultaneous engineering and computational expertise to drive such breakthroughs. Sargent adds that this approach “could be an exciting advance for the fabrication of a broader family of materials,” including LEDs, other photovoltaic technologies, and graphene, “in short, any industry that uses some form of deposition in the vapor phase or under vacuum.
The team also included Austin Flick and Thomas Colburn at Stanford and Zekun Ren at the Singapore-MIT Alliance for Science and Technology (SMART). In addition to the Department of Energy, the work was supported by a fellowship from the MIT Energy Initiative, the National Science Foundation Graduate Research Fellowship Program, and the SMART program.