EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Process-Variation-Aware Physics-Informed Neural Network for Statistical Electromagnetic Analysis of Advanced VLSI Interconnects

Elena Novak, Anwar Siddiq

International Journal of Computational and Biological Sciences · 2026

DOI: 10.66238/ijcbs95 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Computational and Biological Sciences, 2026. Verfasst von Elena Novak und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Advanced VLSI interconnects are increasingly dominated by electromagnetic parasitics whose statistical spread depends on line width, spacing, dielectric height, material permittivity, and conductor resistivity. Conventional variation-aware extraction requires repeated field solves and therefore becomes expensive when design exploration or yield screening needs thousands of samples. This paper proposes a process-variation-aware physics-informed neural network (PV-PINN) that maps spatial coordinates and process variables directly to the electrostatic potential field of a coupled interconnect cross-section. The model combines finite-difference calibration snapshots with a Laplace-equation residual, then propagates manufacturing variations through the trained surrogate to estimate capacitance, peak electric field, and an RC delay proxy. The study intentionally uses a transparent synthetic benchmark generated by an included deterministic finite-difference solver rather than unverified measurement data. On 16 held-out geometries, the PV-PINN achieved a mean potential-field RMSE of 0.0231, a mean capacitance error of 9.54%, and a mean delay-proxy error of 9.54%. In a 60-sample Monte Carlo study, the PV-PINN reproduced the FDM capacitance mean within 2.1% and reduced mean per-geometry evaluation time from 31.12 ms to 2.46 ms on CPU, corresponding to a 12.7x speedup. The results support the feasibility of using physics-informed surrogates as fast statistical electromagnetic engines for early-stage interconnect design, while clearly indicating that full-wave, nonuniform-dielectric, and silicon-measurement validation remain future extensions.

Zitieren

APA

Novak, E., & Siddiq, A. (2026). Process-Variation-Aware Physics-Informed Neural Network for Statistical Electromagnetic Analysis of Advanced VLSI Interconnects. International Journal of Computational and Biological Sciences. https://doi.org/10.66238/ijcbs95

Vancouver

Novak E, Siddiq A. Process-Variation-Aware Physics-Informed Neural Network for Statistical Electromagnetic Analysis of Advanced VLSI Interconnects. International Journal of Computational and Biological Sciences. 2026. doi:10.66238/ijcbs95

BibTeX
@article{Novak2026,
  author = {Novak, Elena and Siddiq, Anwar},
  title = {Process-Variation-Aware Physics-Informed Neural Network for Statistical Electromagnetic Analysis of Advanced VLSI Interconnects},
  journal = {International Journal of Computational and Biological Sciences},
  year = {2026},
  doi = {10.66238/ijcbs95}
}
Kostenlos starten →