A research team at Penn State has introduced a faster, more accessible way to design metasurfaces (engineered materials that can precisely control light and other electromagnetic waves using their structure alone).
Image Credit: Chaosamran_Studio/Shutterstock.com
The team's AI-driven approach promises to simplify the development of next-generation optical technologies, including compact camera lenses, VR headsets, and holographic imagers.
Their method, featured on the cover of the October issue of Nanophotonics, makes use of large language models (LLMs) to accurately predict how a metasurface will interact with light.
LLMs are a form of artificial intelligence (AI) capable of learning and refining tasks over time using training data and repeated behavior.
This approach bypasses traditional simulation techniques that require deep domain knowledge and significant time, allowing engineers to quickly design these nanoscopic materials and predict their optical behavior using simple AI prompts.
According to Doug Werner, John L. and Genevieve H. McCain Chair Professor of Electrical Engineering and corresponding author on the study, metasurfaces provide far more flexibility and capability than conventional materials in nanophotonic devices - systems that manipulate light at scales smaller than the wavelength of visible light.
You can only go so far using naturally occurring materials when trying to manipulate light or other types of electromagnetic waves. Through the structure of the sub-wavelength unit cells that make up the materials, metasurfaces can manipulate the way light behaves at a nanoscopic level, allowing us to slim down optical systems that are traditionally very bulky.
Douglas H. Werner, John L. and Genevieve H. McCain Chair Professor, Electrical Engineering, Penn State
Despite their advantages, metasurfaces have been difficult to develop.
Huanshu Zhang, a third-year electrical engineering doctoral student and first author of the paper, noted that while AI has been part of the design process for several years through deep-learning neural networks (which mimic the brain’s nonlinear processing), designers still had to run time-intensive simulations and build custom networks for each new design.
This limitation led Zhang to introduce LLMs into the workflow.
“The main limitation of current neural-network-based methods is that you must try many neural network configurations in order to find one that accurately predicts how a metasurface will interact with light. By training LLMs, we can accurately predict how a metasurface will interact with light in seconds compared to the hours, days, or even months it previously took, without needing specialized AI expertise or countless trials,” said Zhang.
To test the method, the team compared LLM-generated predictions with conventional computer-simulated metasurfaces.
The LLMs predicted how light would behave when interacting with a metasurface, adjusting through designated “control points” that shaped the design. These predictions were then compared to a dataset of more than 45,000 randomly generated metasurface designs. The results showed that the LLMs delivered highly accurate forecasts of light behavior, while eliminating the time-consuming neural network setup.
This efficiency boost makes it easier for researchers to explore what Lei Kang, associate research professor of electrical engineering and co-author on the paper, referred to as “arbitrarily shaped” metasurface elements.
Compared to standardized shapes like cylinders or cubes, these free-form designs can significantly improve performance and efficiency, but have historically been difficult to optimize.
Arbitrary designs allow researchers to create application-specific metasurfaces that vastly outperform designs based on traditional shapes. However, these designs couldn’t be optimized and tested effectively because traditional simulation methods would take an impractically long time to complete. By integrating LLM predictions, we can see how the metasurfaces will influence light at unprecedented speeds.
Lei Kang, Associate Research Professor and Study Co-Author, Electrical Engineering, Penn State
According to Sawyer Campbell, associate professor of electrical engineering and co-author on the paper, the new method also makes metasurface engineering much more approachable. He pointed out that LLMs are particularly effective at “inverse design,” starting from the desired outcome and working backward to identify the optimal system, material, or structure to produce it.
While inverse design has been possible in the past, the simulation time often stretched into weeks or even months.
Looking ahead, the team plans to continue refining this method.
According to Werner, their primary goal is to dramatically reduce the time and complexity involved in designing metasurface-enabled devices, helping bring them to market more quickly in industries ranging from healthcare and defense to energy and consumer electronics.
“We believe this approach could usher in a new standard for how industry engineers and researchers approach developing nanophotonic devices. With this new method, researchers unfamiliar with the complex metasurface design process can approach the LLMs with an explanation of what they need and effectively generate it,” said Werner.
This research was supported by the John L. and Genevieve H. McCain Endowed Chair Professorship at Penn State.
Journal Reference:
Zhang, H., et al. (2025) Chat to chip: large language model based design of arbitrarily shaped metasurfaces. Nanophotonics. DOI: 10.1515/nanoph-2025-0343. https://www.degruyterbrill.com/document/doi/10.1515/nanoph-2025-0343/html