Summary
A research paper (arXiv:2611.21594) presents a machine-learning-assisted method for extracting BSIM4 compact-model parameters for SkyWater 130 nm (SKY130) CMOS transistors operating at cryogenic temperatures (77 K). The approach uses a neural surrogate model of NGSpice that maps compact-model parameters and device geometry to complete sets of output and transfer characteristics. Parameter extraction is formulated as a bounded least-squares search: a bank of NGSpice simulations provides initial candidates, finite differences of the surrogate supply the local Jacobian, and NGSpice re-simulates the resulting model cards before final selection. On 18 measured geometries, the authors compare a search over 7 card parameters against a search over 43 parameters selected by NGSpice sensitivity analysis. Relative to NGSpice re-simulation of a previously published card, the two searches reduce the mean in-sample relative root-mean-square error (RRMS) by 17% and 26%, respectively. The 43-parameter search achieves lower overall error, while the 7-parameter search yields better average accuracy on high-bias curves. This work supports cryogenic circuit design by enabling compact models that reproduce measured low-temperature behavior in standard circuit simulators.
Paper Overview
- Field: Machine Learning
- Authors: Aaron Wang, Olivia Gzamouranis, Vaibhav Lohia, Hershel Fei Xun Weiner, Olivia Seidel, Richard Cavanaugh, Giuseppe Di Guglielmo, Nhan Tran, Davide Braga
- Published: 2026-11-16
- arXiv: 2611.21594
Summary
Cryogenic circuit design requires compact transistor models that reproduce measured low-temperature behavior and run in standard circuit simulators. This work fits BSIM4 parameters to measured I-V curves of SkyWater 130 nm (SKY130) transistors at 77 K using a neural surrogate of NGSpice.
The surrogate maps compact-model parameters and device geometry to the complete set of output and transfer characteristics. Extraction is a bounded least-squares search:
- A bank of NGSpice simulations supplies initial candidates
- Finite differences of the surrogate supply the local Jacobian
- NGSpice re-simulates the resulting cards before selection
On 18 measured geometries, the authors compare a search over 7 card parameters with a search over 43 parameters selected by NGSpice sensitivity. Relative to NGSpice re-simulation of a previously published card, the two searches reduce the mean in-sample relative RMS error (RRMS) by 17% and 26%, respectively. The 43-parameter search achieves lower overall error, while the 7-parameter search yields a lower mean error on high-bias curves.
---
*Auto-collected on 2026-11-18*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178637207