AI-assisted spectroscopic ellipsometry turns slow, expert-heavy thin-film analysis into fast, automated insight. A University of Toledo physics Ph.D. student trained neural networks on vast simulated datasets to read nonuniform layers more quickly. The result can cut a key bottleneck in solar and semiconductor lines and improve quality control without touching the product. Nonuniform layers can stall a production line. When parts are only nanometers thick, workers cannot see problems with the naked eye. They need tools that read how light changes when it reflects from a surface. That is where spectroscopic ellipsometry shines. Now, a new AI approach aims to make this trusted method much faster and easier to use on the factory floor.

What AI-assisted spectroscopic ellipsometry means

Spectroscopic ellipsometry measures thin-film thickness and optical properties by tracking tiny changes in the polarization of light. Skilled scientists fit the measured spectra to a model of the stack. This step takes time, judgment, and many manual tweaks. AI-assisted spectroscopic ellipsometry replaces much of that manual fitting with a trained model. A neural network learns the link between spectra and film parameters across many possible real-world variations. It can then interpret new measurements more quickly, even when layers vary across a surface.

Why this breakthrough matters

The bottleneck: from raw light data to decisions

– Traditional workflows require experts to adjust models again and again – Nonuniform films make fits unstable and slow – Long analysis loops delay feedback to production teams – Throughput suffers, and scrap risk rises

How AI speeds the path

– Trained networks read spectra and return thickness and optical values fast – Automated outputs improve consistency shift-to-shift – Faster analysis enables denser mapping across large panels or wafers – Inline or near-line feedback helps catch problems early Dr. Nik Podraza, professor of physics at The University of Toledo, calls this a major step for research and industry. He notes that cutting analysis time for films many times thinner than a human hair can pay off in semiconductors and energy manufacturing.

From classroom to cleanroom: a student’s journey

Alex Bordovalos earned a physics bachelor’s degree in 2020 and a Professional Science Master’s in Photovoltaics in 2022 at UToledo. During an industry internship, he saw how thin-film nonuniformities slow production decisions. He leaned into his long interest in computer science and built an AI tool to accelerate ellipsometry analysis. His early results earned a first-author paper in the Journal of Applied Physics and a J.A. Woollam Outstanding Student Poster Award at the AVS International Symposium and Exhibition. He also presented at the IEEE Photovoltaic Specialists Conference.

Inside the model training

Bordovalos trained the network on about half a million simulations. Each run varied expected thickness and other film properties across wide ranges, including ±50% swings. This gave the model exposure to many “messy” conditions that real devices show on a line. – Large simulated datasets cover edge cases before real data arrive – The model learns to handle irregular, nonuniform layers – Training supports materials used in thin-film solar, such as cadmium telluride (CdTe) The next step is real-world testing on cadmium telluride solar cells produced and studied at UToledo’s Wright Center for Photovoltaics Innovation and Commercialization. If results match the simulated performance, the tool can move closer to production use.

Impact across solar and semiconductors

Solar manufacturing

– Map thickness across large CdTe glass panels more quickly – Link optical signatures to process drifts in near real time – Reduce rework and stabilize yield

Semiconductor fabs

– Speed metrology for ultra-thin gate oxides and dielectrics – Support uniformity checks across 200–300 mm wafers – Feed rapid data to process control loops

Research labs

– Free scientists from repetitive fitting – Let teams scan more samples in less time – Standardize analysis across instruments and sites

What to watch next

– Validation: Does the AI match expert fits on real devices within tight error bounds? – Robustness: Can it handle new materials, stacks, and process shifts without retraining? – Integration: How easily does it plug into existing ellipsometers and factory software? – Governance: Clear versioning, drift checks, and human-in-the-loop review to maintain trust The University of Toledo’s strength in thin-film photovoltaics and materials science provides an ideal testbed. With cadmium telluride cells on hand and deep metrology expertise, the team can compare AI outputs to gold-standard methods and drive rapid iteration. In short, AI-assisted spectroscopic ellipsometry offers a practical way to shrink analysis time while keeping accuracy high. By moving expert knowledge into a trained model, teams can see nonuniformities sooner, adjust processes faster, and ship better products. As real-world testing progresses, expect broader adoption of AI-assisted spectroscopic ellipsometry in both research labs and advanced manufacturing lines.

(Source: https://news.utoledo.edu/index.php/08_26_2026/physics-ph-d-student-develops-ai-tools-to-tackle-industry-bottleneck)

For more news: Click Here

FAQ