Process Guided by Machine Learning to Boost Perovskite Solar Cells Production
MIT, Stanford researchers fed observations of experienced workers into ML algorithm for analysis
Researchers at the Massachusetts Institute of Technology (MIT), in collaboration with Stanford University, have developed a method called Rapid Spray Plasma Processing (RSPP), which is a process guided by machine learning (ML) algorithms to optimize the manufacturing of perovskite solar cells.
The ML-guided perovskites manufactured by the researchers attained an energy efficiency of 18.5%. While small-scale perovskite manufacturing in the lab uses spin-coating techniques, large-scale production becomes an arduous process.
The process discovered by MIT and Stanford’s team of experts involves a moving roll-to-roll surface or sheet on which the precursor solutions for the perovskite compound are sprayed as the surface rolled. The sheet is further moved to a curing stage, providing a rapid and continuous output with higher throughputs than any other photovoltaic technology, said one of the researchers and Stanford doctoral graduate Nicholas Rolston.
The ML-guided process published in the journal Joule allows scaling up the production by reducing the processing time. The researchers fed data from prior experiments and observations of experienced workers into a machine learning algorithm. The algorithm nullifies human intervention in analyzing a dozen variables that go into the production of perovskites.
Controlling the variables with ML
The team attempted to control variables like the temperature, humidity, and pace of the processing path. Many such variables are prone to factors like humidity when the process is in the open air.
Researchers used a mathematical technique called Bayesian Optimization (sequential design strategy for global optimization of black-box functions that do not assume any functional forms) to study the probability factors in all the previous data later fed into the algorithm.
The experts used the mathematical technique in ML to optimize the power output besides incorporating other criteria like cost and durability.
The Department of Energy sponsored the research to commercialize the technology, while the Stanford team is now aiming at tech transfer to existing perovskite manufacturers. MIT professor of mechanical engineering Tonio Buonassisi observed, “We are reaching out to companies now. The code we developed is now on the open-source server GitHub. Anyone can download it and run it. We’re happy to help companies get started in using our code.”
According to former MIT assistant professor Zhe Liu, many components manufacturing companies worldwide have started with smaller, high-value applications like building-integrated solar tiles. He points out that appearance is the key. “Three such manufacturing companies are on track or are being pushed by investors to manufacture 1 meter by 2-meter rectangular modules [comparable to today’s most common solar panels] within two years. The problem is that they don’t have a consensus on what manufacturing technology to use. The RSPP method still has a good chance to be competitive.”
Perovskite is being studied by various institutions across the globe for its potential to replace silicon-based photovoltaic solar panels. Earlier this month, researchers from Dartmouth Engineering in New Hampshire developed a rapid printing method to produce perovskite solar cells. The engineers claimed that the printing technique accelerated the total processing time of solar charge transport layers (CTLs) by 60 times, with efficient sunlight-to-electricity conversion.
Mercom also reported that researchers from the University of Toronto claimed to have developed an alternative solar technology using an inverted perovskite cell structure. The team pursued quantum mechanics to channel the active layer in an inverted perovskite solar cell and achieved a power conversion efficiency of 23.9%.