Harvesting Design Knowledge From the Internet: High-Dimensional Performance Tradeoff Modeling for Large-Scale Analog Circuits

Published

Journal Article

© 2015 IEEE. Efficiently optimizing large-scale, complex analog systems requires to know the performance tradeoffs for various analog circuit blocks. In this paper, we propose a radically new approach for analog performance tradeoff modeling. Our key idea is to broadly search the rich design knowledge from the Internet, and then mathematically encode the knowledge as high-dimensional performance tradeoff curves that are referred to as Pareto fronts in the literature. Toward this goal, several novel numerical algorithms, such as sparse regression and semi-infinite programming, are developed in order to construct the high-dimensional Pareto front model while guaranteeing its monotonicity. Our numerical examples demonstrate that the proposed modeling technique can accurately capture the high-dimensional Pareto fronts for large-scale analog systems (e.g., analog-to-digital converter) while most traditional methods are limited to low-dimensional Pareto front modeling of small circuit blocks without considering layout parasitics and manufacturing nonidealities.

Full Text

Duke Authors

Cited Authors

  • Tao, J; Liao, C; Zeng, X; Li, X

Published Date

  • January 1, 2016

Published In

Volume / Issue

  • 35 / 1

Start / End Page

  • 23 - 36

International Standard Serial Number (ISSN)

  • 0278-0070

Digital Object Identifier (DOI)

  • 10.1109/TCAD.2015.2449240

Citation Source

  • Scopus