EUVIMEDEUVIMED entdecken
Crossref

Accurate SiC Epitaxial Thickness Measurement via Hybrid Deep Learning and Physical Interference Model for Power Device Applications

Minghui Li, Haojin Wang, Zekuo Zhang

Journal of Nanoelectronics and Optoelectronics · 01.12.2025 · 20 (12) · 1260-1266

DOI: 10.1166/jno.2025.3844 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Nanoelectronics and Optoelectronics, erschienen am 01.12.2025. Verfasst von Minghui Li und 2 weiteren Autor:innen. Die Arbeit nennt 14 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Silicon carbide has become indispensable for high-voltage power electronics, with device performance heavily dependent on epitaxial layer thickness control. Fourier transform infrared reflectance spectroscopy serves as the primary non-destructive technique for thickness measurement, yet its accuracy suffers from refractive index dispersion, strong phonon absorption near the reststrahlen band, and multi-beam interference effects that conventional peak analysis cannot adequately address. These physical complications introduce systematic errors that directly impact breakdown voltage uniformity and on-resistance characteristics in SiC devices. This work introduces a hybrid approach combining a physically interpretable peak-based interference model with a deep learning residual correction framework. The peak-based method establishes an initial thickness estimate using fringe spacing while avoiding complex nonlinear fitting procedures, and the neural network subsequently compensates for systematic deviations arising from dispersion and absorption phenomena. Simulation studies employing a multi-beam interference model with 4H-SiC optical properties demonstrate that the hybrid method achieves mean absolute errors below 0.1 μ m across the 5 to 40 μ m thickness range, representing approximately two orders of magnitude improvement over conventional techniques. The approach maintains this accuracy under significant spectral noise and across the entire thickness range relevant to modern power device fabrication. By preserving physical interpretability while leveraging data-driven correction, this framework offers a computationally efficient solution suitable for real-time in-line metrology in SiC manufacturing environments.

Zitieren

APA

Li, M., Wang, H., & Zhang, Z. (2025). Accurate SiC Epitaxial Thickness Measurement via Hybrid Deep Learning and Physical Interference Model for Power Device Applications. Journal of Nanoelectronics and Optoelectronics, 20(12), 1260-1266. https://doi.org/10.1166/jno.2025.3844

Vancouver

Li M, Wang H, Zhang Z. Accurate SiC Epitaxial Thickness Measurement via Hybrid Deep Learning and Physical Interference Model for Power Device Applications. Journal of Nanoelectronics and Optoelectronics. 2025;20(12):1260-1266. doi:10.1166/jno.2025.3844

BibTeX
@article{Li2025,
  author = {Li, Minghui and Wang, Haojin and Zhang, Zekuo},
  title = {Accurate SiC Epitaxial Thickness Measurement via Hybrid Deep Learning and Physical Interference Model for Power Device Applications},
  journal = {Journal of Nanoelectronics and Optoelectronics},
  year = {2025},
  volume = {20},
  number = {12},
  pages = {1260--1266},
  doi = {10.1166/jno.2025.3844}
}
Kostenlos starten →