ACAP collaborators pioneer a fundamental shift to predictive data-driven design for high performance organic solar cells

A collaborative research team has demonstrated a step change in how organic photovoltaic (OPV) research can be conducted – moving the field from slow, trial-and-error experimentation to a predictive, data-driven paradigm.

By combining high-throughput manufacturing-style experimentation with machine learning (ML), researchers from CSIRO and Monash University have achieved both unprecedented data scale and record device performance, as published in Energy & Environmental Science. The project’s first author Na Gyeong (Korean Research Foundation, hosted by Monash University) was overseen by Professor Udo Bach (Monash University), working closely with Dr Doojin Vak and researchers at CSIRO operating ACAP’s high-throughput roll-to-roll OPV MicroFactory platform.

A long-standing barrier to applying artificial intelligence in experimental energy research is the lack of large, consistent, high-quality datasets. In OPVs, this problem is particularly acute: device fabrication is typically manual, low-throughput, and highly researcher-dependent. Through ACAP’s industrially relevant MicroFactory infrastructure, the team overcame this bottleneck by mimicking real manufacturing conditions inside the lab.

“Machine learning in photovoltaics has been talked about for years, but without the right kind of data it simply can’t deliver,” said Dr Doojin Vak.

"ACAP’s high-throughput roll-to-roll platform allowed us to generate manufacturing-grade data at a scale that changes what’s possible. Instead of guessing what might work, we can now predict performance before a device is even made.”

The high-throughput roll-to-roll coater is part of ACAP’s MicroFactory platform at CSIRO and is enabling data driven design of organic photovoltaic solar cells.

Using this approach, more than 26,000 unique OPV devices were fabricated and characterised in just four days – two to three orders of magnitude faster than conventional research methods. ML models trained on this dataset successfully predicted full current–voltage behaviour and guided optimisation of materials and processing conditions.

The result was a record 11.8% power-conversion efficiency for fully roll-to-roll-fabricated organic solar cells, the highest reported globally.

A video link in the paper shows the interactive  J–V prediction ML model, which provides a whole new method of device optimisation. The team can browse experimental parameters with the ML tool, instantly showing J–V curves of virtually planned experiments.

“We can now conduct experiments with predictive guidance, where decisions about materials and processing are informed by data-driven models before fabrication,” said Dr Vak.

“Seeing machine learning move from post-analysis to an active part of the experimental workflow is particularly exciting.”

The interactive  J–V prediction ML tool lets the team browse experimental parameters, instantly showing J–V curves of virtually planned experiments.

ACAP’s collaborative edge drives leading innovation

The project exemplifies the strength of ACAP’s collaborative model. The combination of roll-to-roll–compatible fabrication, automated characterisation, and large-scale data production in a laboratory setting is still rare internationally. By demonstrating not only accurate machine-learning prediction but also ML-guided experimental optimisation, this work moves beyond proof-of-concept studies and establishes a practical framework that few groups worldwide currently possess. It positions ACAP firmly at the international leading edge and lays the groundwork for future data-rich solar manufacturing innovations.

“We’ve shown that large, high-quality experimental datasets – once a major bottleneck – can be generated routinely, allowing machine learning to become a reliable tool rather than an exploratory add-on,” said Dr Vak.

“This fundamentally changes how device research and optimisation can be approached.”

Reference:

An, N. G., Ng, L. W. T., Liu, Y., Song, S., Gao, M., Zhou, Y., Ma, C., Wei, Z., Kim, J. Y., Bach, U., & Vak, D. (2025). Mass-customization of organic photovoltaics and data production for machine learning models precisely predicting device behavior. Energy & Environmental Science, 18, 9524–9537. https://doi.org/10.1039/D5EE02815A