Wang D, Boroujeni PF, Koch M, Breun S, Weigel R, Franchi N (2026)
Publication Language: English
Publication Type: Conference contribution, Original article
Publication year: 2026
Publisher: IEEE
Event location: Karlsruhe Institute of Technology, Germany
URI: https://ieeexplore.ieee.org/abstract/document/11516451
DOI: 10.1109/GeMiC71240.2026.11516451
This work presents an AI-assisted and scalable 0 - 300 GHz generative surrogate model for synthesizing full-band six-port S-parameters of millimeter-wave on-chip transformers (OTs). The method constructs a rational backbone using shared-pole vector fitting (VF), while a multi-head multilayer perceptron (MLP) predicts the residues and affine terms for 21 unique channels. Through differentiable rational reconstruction, the model produces a SPICE-compatible broadband model in rational form (poles, residues, affine terms), from which continuous S-parameters and standard Touchstone.s6p files on arbitrary frequency grids can be synthesized on demand. A set of model-agnostic evaluation framework is introduced, enabling consistent benchmarking across different OT models. Within this framework, the relative deviation of the self-resonant frequency is only 0.1%, the worst-case inductance deviation is 0.4%, and the worst-case Q-factor deviation is 8.4%, demonstrating accurate capture of high-frequency effects.
APA:
Wang, D., Boroujeni, P.F., Koch, M., Breun, S., Weigel, R., & Franchi, N. (2026). Generative Surrogate Model for Continuous mm-Wave Six-Port Transformer S-Parameters. In Proceedings of the 2026 German Microwave Conference (GeMic). Karlsruhe Institute of Technology, Germany: IEEE.
MLA:
Wang, Dingan, et al. "Generative Surrogate Model for Continuous mm-Wave Six-Port Transformer S-Parameters." Proceedings of the 2026 German Microwave Conference (GeMic), Karlsruhe Institute of Technology, Germany IEEE, 2026.
BibTeX: Download