TY - JOUR
T1 - Laguerre-Volterra Feed-Forward Neural Network for Modeling PAM-4 High-Speed Links
AU - Wang, Xinying
AU - Nguyen, Thong
AU - Schutt-Aine, Jose E.
N1 - Manuscript received August 25, 2020; revised October 31, 2020; accepted November 5, 2020. Date of publication November 20, 2020; date of current version December 23, 2020. This work was supported in part by the National Science Foundation under Grant CNS 16-24810, in part by the U.S Army Small Business Innovation Research (SBIR) Program Office, and in part by the U.S. Army Research Office under Contract W911NF-16-C-0125. Recommended for publication by Associate Editor S. Grivet-Talocia upon evaluation of reviewers’ comments. (Corresponding author: Xinying Wang.) The authors are with the Department of Electrical and Computer Engineering, University of Illinois Urbana–Champaign, Champaign, IL 61820 USA (e-mail: [email protected]; [email protected]; [email protected]).
PY - 2020/12
Y1 - 2020/12
N2 - In this article, we present a PAM-4 IBIS-AMI model derived from machine learning for time-domain simulation. More specifically, we report a Laguerre-Volterra-expanded feed-forward neural network (LVFFN) approach with one hidden layer and ten neurons to model the 28-Gb/s PAM-4 high-speed link buffer. The proposed LVFFN model reduces the model size and improves the computational efficiency dramatically compared with the Volterra series model and other transitional artificial neural network models. The LVFFN model is implemented in IBIS-AMI, an industrial standard, and is simulated in existing software platforms for eye-diagram analysis. This work has two innovations: 1) we propose a method that dramatically reduces the neural network model complexity through a Laguerre-Volterra expansion when modeling weakly nonlinear systems with memory and 2) we implement an LVFFN model into IBIS-AMI to enhance the model interoperability and transportability.
AB - In this article, we present a PAM-4 IBIS-AMI model derived from machine learning for time-domain simulation. More specifically, we report a Laguerre-Volterra-expanded feed-forward neural network (LVFFN) approach with one hidden layer and ten neurons to model the 28-Gb/s PAM-4 high-speed link buffer. The proposed LVFFN model reduces the model size and improves the computational efficiency dramatically compared with the Volterra series model and other transitional artificial neural network models. The LVFFN model is implemented in IBIS-AMI, an industrial standard, and is simulated in existing software platforms for eye-diagram analysis. This work has two innovations: 1) we propose a method that dramatically reduces the neural network model complexity through a Laguerre-Volterra expansion when modeling weakly nonlinear systems with memory and 2) we implement an LVFFN model into IBIS-AMI to enhance the model interoperability and transportability.
KW - Behavior modeling
KW - PAM-4
KW - Volterra series
KW - eye diagram
KW - neural network
UR - https://www.scopus.com/pages/publications/85096855032
UR - https://www.scopus.com/inward/citedby.url?scp=85096855032&partnerID=8YFLogxK
U2 - 10.1109/TCPMT.2020.3039486
DO - 10.1109/TCPMT.2020.3039486
M3 - Article
AN - SCOPUS:85096855032
SN - 2156-3950
VL - 10
SP - 2061
EP - 2071
JO - IEEE Transactions on Components, Packaging and Manufacturing Technology
JF - IEEE Transactions on Components, Packaging and Manufacturing Technology
IS - 12
M1 - 9265412
ER -