EUVIMEDEUVIMED entdecken
Crossref

Multi‐Task Deep Learning‐Assisted Global Optimization for Physical Parameter Extraction of PIN Diode Models

Chengjie Li, Yaogen Li, Yanwei Ye, Ruxin Zheng, Xingshen Dong, Zhongyuan Zhou

International Journal of Numerical Modelling: Electronic Networks, Devices and Fields · 2026-11 · 39 (6)

DOI: 10.1002/jnm.70228 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Numerical Modelling: Electronic Networks, Devices and Fields, online erschienen am 17.11.2026. Verfasst von Chengjie Li und 5 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate extraction of physical parameters is essential for high‐frequency simulation and reliability‐oriented design of semiconductor devices. However, conventional physics‐based global optimization suffers from parameter degeneracy and non‐uniqueness, while repeated simulations impose high computational cost. To address these limitations, we propose an efficient physical‐parameter extraction method for PIN diodes. A multi‐task deep neural network (MTDL) is developed as a surrogate model that jointly captures direct‐current characteristics and microwave transient responses. A quantum genetic algorithm (QGA) is then used for global optimization to extract three intrinsic parameters governing the high‐frequency macroscopic response: effective cross‐sectional area ( A ), intrinsic‐region thickness ( X i ), and intrinsic‐region doping concentration ( N i ). Cross‐validation against steady‐state DC and L‐band microwave measurements shows excellent agreement between simulated and measured response curves, with normalized root mean square error (NRMSE) below 2%. The extracted model also reproduces dynamic responses associated with space‐charge dielectric effects and displacement current. Simulated dielectric and breakdown characteristics further agree with reference handbook data, supporting the validity of the proposed method. The framework reduces the computational burden of conventional physics‐based extraction while providing an efficient and practical approach for semiconductor device modeling and parameter identification.

Zitieren

APA

Li, C., Li, Y., Ye, Y., Zheng, R., Dong, X., & Zhou, Z. (2026). Multi‐Task Deep Learning‐Assisted Global Optimization for Physical Parameter Extraction of PIN Diode Models. International Journal of Numerical Modelling: Electronic Networks, Devices and Fields, 39(6). https://doi.org/10.1002/jnm.70228

Vancouver

Li C, Li Y, Ye Y, Zheng R, Dong X, Zhou Z. Multi‐Task Deep Learning‐Assisted Global Optimization for Physical Parameter Extraction of PIN Diode Models. International Journal of Numerical Modelling: Electronic Networks, Devices and Fields. 2026;39(6). doi:10.1002/jnm.70228

BibTeX
@article{Li2026,
  author = {Li, Chengjie and Li, Yaogen and Ye, Yanwei and Zheng, Ruxin and Dong, Xingshen and Zhou, Zhongyuan},
  title = {Multi‐Task Deep Learning‐Assisted Global Optimization for Physical Parameter Extraction of PIN Diode Models},
  journal = {International Journal of Numerical Modelling: Electronic Networks, Devices and Fields},
  year = {2026},
  volume = {39},
  number = {6},
  doi = {10.1002/jnm.70228}
}
Kostenlos starten →