EUVIMEDEUVIMED entdecken
Crossref

Physics-Guided Latent Diffusion for Inverse Design of Plasmonic Nanostructure Photonic Chips

Ziwen Su

Journal of Nanoelectronics and Optoelectronics · 01.06.2026 · 21 (06) · 548

DOI: 10.1166/jno.2026.3910 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Nanoelectronics and Optoelectronics, online erschienen am 01.06.2026. Verfasst von Ziwen Su.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Inverse design turns a target optical response into a physical nanostructure, but plasmonic components on photonic chips remain difficult to design because their spectra depend nonlinearly on geometry and because many distinct geometries can produce a similar response. Iterative electromagnetic optimization reaches high accuracy yet is slow, while one-shot deep networks are fast but tend to collapse the one-to-many mapping and generalize poorly across device families. This paper presents PlasmoDiff, a physics-guided latent diffusion framework that generates plasmonic nanostructure layouts directly from a target transmission spectrum. Geometries are compressed by a variational autoencoder into a compact latent space, and a conditional denoising diffusion model is trained to sample latents that match the target. During sampling, a differentiable electromagnetic surrogate is added as a guidance term, so that each denoising step is nudged toward structures whose predicted spectrum agrees with the target. On a dataset of 24,000 finite-difference time-domain (FDTD) simulated structures spanning four device families, PlasmoDiff reaches a spectral fidelity of R 2 = 0.972 and a spectral mean squared error of 0.024, close to the accuracy of adjoint topology optimization while running about three orders of magnitude faster and returning a diverse set of valid candidates rather than a single point solution. A case study on an on-chip metal-insulator-metal refractive-index sensor shows a designed sensitivity near 1180 nm per refractive-index unit and a robustness window of about 10 nm in fabrication gap deviation. The framework offers a practical route to fast, diverse, and manufacturable plasmonic chip design.

Zitieren

APA

Su, Z. (2026). Physics-Guided Latent Diffusion for Inverse Design of Plasmonic Nanostructure Photonic Chips. Journal of Nanoelectronics and Optoelectronics, 21(06), 548. https://doi.org/10.1166/jno.2026.3910

Vancouver

Su Z. Physics-Guided Latent Diffusion for Inverse Design of Plasmonic Nanostructure Photonic Chips. Journal of Nanoelectronics and Optoelectronics. 2026;21(06):548. doi:10.1166/jno.2026.3910

BibTeX
@article{Su2026,
  author = {Su, Ziwen},
  title = {Physics-Guided Latent Diffusion for Inverse Design of Plasmonic Nanostructure Photonic Chips},
  journal = {Journal of Nanoelectronics and Optoelectronics},
  year = {2026},
  volume = {21},
  number = {06},
  pages = {548},
  doi = {10.1166/jno.2026.3910}
}
Kostenlos starten →