Machine-learning based inference of critical dimensions of gate-all-around transistors from reflectance spectroscopy measurements
Machine Learning: Engineering · 01.12.2026 · 2 (2) · 025003
DOI: 10.1088/3049-4761/ae87a4 ↗Kurz gefasst
Fachartikel in der Zeitschrift Machine Learning: Engineering, online erschienen am 23.07.2026. Verfasst von James Wang und 1 weiteren Autor:innen. Die Arbeit nennt 47 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: NIST.
Abstract (Originalsprache)
This paper presents a method for scatterometry-based inspection of large area gate-all-around (GAA) transistor arrays using a machine learning (ML) based inference of the underlying critical dimensions (CDs). A surrogate ML model was trained on a library of reflectance spectra obtained using physics-based finite-difference time-domain simulations for a set of nanosheet CDs generated using the Latin Hypercube method. From that library, an ensemble ML-based model employing the ridge regression, support vector regression, XGBoost, and Random Forest models was trained to estimate the underlying CDs using the corresponding reflectance spectrum as the input. Predictive capability of the ensemble model was tested using GAA samples virtually fabricated using a commercial fabrication simulation software, with CDs differing from those available in the training dataset. ML-model predictions of CDs based on the reflectance spectra simulated for those GAA transistors showed high predictive capability of the newly proposed method for GAA CD-inference. Furthermore, a formal Gage Repeatability and Reproducibility (Gage R&R) study was simulated for scatterometry-based GAA CD metrology relying on the newly proposed ML-based approach to CD inference. Depending on the nanosheet CD, Gage R&R values ranging between 1% and 14% were observed. Given that reflectance spectra acquisition and ML-based CD inference can be accomplished within 1 h for a 300 mm diameter wafer, these results indicate the feasibility of utilizing the concept of ML-based inference of CDs of GAA transistor arrays to enable their accurate, reliable, and fast scatterometry-based metrology.
Förderung
- NIST
Zitieren
APA
Wang, J., & Djurdjanovic, D. (2026). Machine-learning based inference of critical dimensions of gate-all-around transistors from reflectance spectroscopy measurements. Machine Learning: Engineering, 2(2), 025003. https://doi.org/10.1088/3049-4761/ae87a4
Vancouver
Wang J, Djurdjanovic D. Machine-learning based inference of critical dimensions of gate-all-around transistors from reflectance spectroscopy measurements. Machine Learning: Engineering. 2026;2(2):025003. doi:10.1088/3049-4761/ae87a4
BibTeX
@article{Wang2026,
author = {Wang, James and Djurdjanovic, Dragan},
title = {Machine-learning based inference of critical dimensions of gate-all-around transistors from reflectance spectroscopy measurements},
journal = {Machine Learning: Engineering},
year = {2026},
volume = {2},
number = {2},
pages = {025003},
doi = {10.1088/3049-4761/ae87a4}
}
EUVIMED