Using a digital twin, roll-to-roll platform to design high-performance organic photovoltaic devices – pv magazine International

July 8, 2024 Valerie Thompson

An international team of researchers has demonstrated a high-throughput platform to help discover high-performance organic solar cells suitable for scalable production, using digital twin technology and roll-to-roll (R2R) closed-loop printing.

The new platform, called MicroFactory, was used to fabricate, characterize, and analyze 11,800 non-fullerene (NFA) organic PV devices in 24 hours. After analyzing the initial devices, the team used a large dataset of fabrication and characterization parameters to train and optimize machine learning models.

In the next iteration, 1,200 devices were developed with improved PCEs based on “inverse parameter generation machine learning,” according to Leonard Ng Wei Tat, co-author of the study, who noted that a masterful efficiency of up to 9.35% was observed, which translates to “an improvement of 1% in just one cycle.”

Off-the-shelf equipment had to be modified to create the processing and characterization systems needed for the study. “R2R equipment is commercially available, but most of it is designed for graphic printing purposes and is largely unsuitable for printing solar cells.” Ng he said magazine pv.

“Basically, what we were doing was applying an age-old, mature technology toward high-throughput solar cell production. The concept of printing solar cells is simple, depositing layer upon layer of functional material until you build a heterostructure that can act as the different components required in a solar cell,” Ng said.

The equipment consisted of coating and slot annealing subsystems with integrated sensors. “We fabricate the photovoltaic cells by depositing functional layers on a polyethylene terephthalate (PET) strip with a patterned transparent conducting electrode (TCE),” the research team said. The functional layers consisted of a layer of a conducting polymer, PEDOT:PSS, a layer of a silver mesh, and zinc oxide nanoparticles.

Multiple sensors collected data on 36 process parameters, which are stored in a database on a remote data server for use in digital twin models. “These models suggested specific changes to key manufacturing parameters, especially the donor-to-acceptor (D:A) ratio, and also enabled the incorporation of new, reported scientific knowledge, which included the introduction of new interface layers,” the research team said.

The ability to collect a large amount of data allowed for analysis and identification of trends and performance factors. As an example, Ng pointed to the finding that humidity control was much more important than temperature control in ensuring good device quality. “This largely correlates with the observed trends that our manufactured solar cells perform better in low-humidity conditions in winter than in summer, despite being manufactured in the same air-conditioned environment in both seasons,” Ng said.

Popular content

The researchers emphasized that the iterative approach, based on machine learning insights, provides strategic optimization as a digital twin-driven alternative to traditional design of experiments. “For example, large-scale solar module manufacturers can quickly create simple digital twins of their processes to build large data sets to identify factors that really move the needle to increase productivity and efficiency,” Ng said.

The research team required interdisciplinary skills, including materials science, hardware and software development skills, and machine learning expertise. “Most researchers only know one field, and contextualizing things for each other takes a lot of coordination and effort,” Ng said.

Details of the study are discussed in the article “A printing-inspired digital twin for the self-driving, high-throughput, closed-loop optimization of roll-to-roll printed photovoltaics” published in Cell Reports – physical sciences. Research team members come from Australia’s Commonwealth of Scientific and Industrial Research Organisation (CSIRO), South Korea’s Pukyong National University and Singapore’s Nanyang Technological University.

Looking ahead, the researchers are exploring new materials and device architectures for more efficient flexible solar cells, and are also continuing to apply artificial intelligence (AI) technology, digital twins and inverse parameter generation capabilities to other processes, such as batch processing and traditional solar panel manufacturing. “Ultimately, we hope to develop a unified system that can connect multiple machines, factories and labs around the world, enabling more advanced AI,” Ng said.