Physicists have long relied on spectroscopic ellipsometry to measure the thickness and optical properties of materials far thinner than a human hair, but interpreting the raw data has remained a slow, labor-intensive process. Alex Bordovalos, a physics Ph.D. student at The University of Toledo, is developing an AI tool to automate that interpretation step, a bottleneck that currently slows both research and manufacturing quality control in the semiconductor and solar energy sectors.
Bordovalos began working on the problem during an internship with a solar cell manufacturer, part of his master's degree in photovoltaics at UToledo. On a production line, nonuniformities in ultra-thin materials are invisible to the naked eye and difficult to identify even with specialized equipment. The standard technique works, but analyzing the data requires physicists to continually adjust mathematical models to account for irregularities in the materials being measured.
"It's not like you can take a ruler and see it," Bordovalos said. "You need specialized tools."
Training AI on half a million simulations
Bordovalos has built an AI model that can perform the analysis more efficiently than a human physicist. To train it, he generated roughly 500,000 simulations covering a wide range of material variations.
"The simulations cover all these different possibilities," he said. "For example, I'll start with the thickness that I expect, and I might vary it by about 50% in either direction."
His preliminary results have already been published. He was first author on a paper in the peer-reviewed Journal of Applied Physics, presented at the IEEE Photovoltaics Specialists Conference, and received the J.A. Woollam Outstanding Student Poster Award from the Spectroscopic Ellipsometry Technical Group in 2024.
From simulation to real-world testing
Bordovalos is now nearing the end of his doctoral program and preparing to test the model on real cadmium telluride solar cells, a thin-film photovoltaic technology that is a research specialty at UToledo's Wright Center for Photovoltaics Innovation and Commercialization. The cells are readily available on campus, making the final validation step straightforward.
"That's the last hurdle," Bordovalos said. "I need to see if the model can make the jump handling the simulated stuff very well to handling the real stuff very well."
Dr. Nik Podraza, Bordovalos's advisor and interim dean of the College of Natural Sciences and Mathematics, said the tool addresses a practical need. "Alex is doing great work that addresses a major bottleneck in industry and research," Podraza said. "The tool he's developing promises huge decreases in analysis time in measurements of materials that may be many times thinner than a human hair. This has significant potential for manufacturers in the semiconductor and energy sectors."
The work sits at the intersection of two fields that are increasingly intertwined. For professionals tracking AI for Science & Research, the project is a concrete example of machine learning applied to a measurement problem that has resisted automation. For those following AI for IT & Development, it demonstrates how AI models can be trained on synthetic data to handle real-world variability.
Why this matters for science and IT professionals
For physicists and materials scientists, Bordovalos's approach points to a future where spectroscopic ellipsometry data analysis is no longer the rate-limiting step in characterizing thin films. If the model performs on real materials as it does on simulations, it could cut analysis time from hours to minutes, accelerating both R&D cycles and production-line quality checks.
For developers and IT professionals, the project is a useful case study in training AI on simulated data when real-world labeled data is scarce or expensive to obtain. The 500,000-simulation training set is a reminder that synthetic data generation is often the practical path forward in specialized technical domains, and that the hardest part of deploying an AI tool is often the final bridge from simulation to reality.
Your membership also unlocks:
