EUVIMEDEUVIMED entdecken
Crossref

Measurement of silicon carbide epilayer thickness using physics-informed neural networks and hybrid intelligent systems

Changyuan Li, Dengkui Li

Intelligence & Robotics · 17.08.2026 · 6 (3) · 479-503

DOI: 10.20517/ir.2026.23 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Intelligence & Robotics, online erschienen am 17.08.2026. Verfasst von Changyuan Li und 1 weiteren Autor:innen. Die Arbeit nennt 33 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The thickness of silicon carbide (SiC) epilayers critically determines device performance; however, conventional optical methods suffer from low efficiency and limited robustness to noise. We propose an Uncertainty-Aware Hybrid Intelligent System (UAHIS) that integrates a physical benchmark model based on multi-beam interference with an attention-enhanced physics-informed neural network (AE-PINN). Trained on 50,000 real spectra, the AE-PINN serves as a rapid prediction engine, achieving a root mean square error (RMSE) of 1.194 μm for the test set. The Monte Carlo Dropout enables the UAHIS to quantify prediction uncertainty and adaptively switch between neural network prediction and physical model verification. The optimal uncertainty thresholds were 0.391 μm (global), 0.397 μm (SiC), and 0.252 μm (Si). Using real experimental data, the UAHIS achieved a 75.0% AE-PINN utilization rate and a mean absolute error of 0.12 μm for the SiC samples. Under various noise conditions, the average prediction error of the AE-PINN was 23.4% of that of a standalone physical model. An uncertainty-guided incremental learning mechanism further enables the continuous self-optimization of the model. This study offers a unified framework for intelligent optical thin-film thickness measurement, balancing accuracy and adaptability.

Zitieren

APA

Li, C., & Li, D. (2026). Measurement of silicon carbide epilayer thickness using physics-informed neural networks and hybrid intelligent systems. Intelligence & Robotics, 6(3), 479-503. https://doi.org/10.20517/ir.2026.23

Vancouver

Li C, Li D. Measurement of silicon carbide epilayer thickness using physics-informed neural networks and hybrid intelligent systems. Intelligence & Robotics. 2026;6(3):479-503. doi:10.20517/ir.2026.23

BibTeX
@article{Li2026,
  author = {Li, Changyuan and Li, Dengkui},
  title = {Measurement of silicon carbide epilayer thickness using physics-informed neural networks and hybrid intelligent systems},
  journal = {Intelligence & Robotics},
  year = {2026},
  volume = {6},
  number = {3},
  pages = {479--503},
  doi = {10.20517/ir.2026.23}
}
Kostenlos starten →