Abstract
High-speed signal integrity (SI) modeling usually involves modeling non-linear subsystems. Traditional approaches like response surface modeling (RSM), artificial neural networks (ANN) and support vector machine (SVM) suffer from two major issues. Either they don't scale well when the number of dimensions or variables increase in the non-linear model, or they are sensitive to the initial training data used to create the models. For example: RSM approach requires large design of experiments (DoE) when the number of variables increase. With ANN, the designer has to spend a lot of time tuning or optimizing the hyperparameters. In this paper, a new approach based on sparsity constrained regression (SCR) is proposed. This approach can scale even when the non-linear problem has large number of variables, and it only has one hyper-parameter to tune making it easy to train and test. A test case based on printed circuit board (PCB) stack-up modeling is used in this paper to test the efficiency of SCR approach. The results are compared against traditional RSM and ANN approaches.
Recommended Citation
J. Zhuo et al., "Sparsity Constrained Regression for High-Speed Signal Integrity Modeling," IEEE Electrical Design of Advanced Packaging and Systems Symposium, article no. 9011671, Institute of Electrical and Electronics Engineers, Dec 2019.
The definitive version is available at https://doi.org/10.1109/EDAPS47854.2019.9011671
Department(s)
Electrical and Computer Engineering
Keywords and Phrases
artificial neural networks (ANN); design of experiments (DoE); response surface model (RSM); Sparsity constrained regression (SCR)
International Standard Book Number (ISBN)
978-172812432-2
International Standard Serial Number (ISSN)
2151-1233; 2151-1225
Document Type
Article - Conference proceedings
Document Version
Citation
File Type
text
Language(s)
English
Rights
© 2024 Institute of Electrical and Electronics Engineers, All rights reserved.
Publication Date
01 Dec 2019
Comments
National Science Foundation, Grant None