Skip to main content
eScholarship
Open Access Publications from the University of California

Energy Optimization for Two-Dimensional NoCs Using Genetic Algorithms

  • Author(s): Marafie, Zahraa A M R H
  • Advisor(s): Bagherzadeh, Nader
  • et al.
Abstract

The steadfast development of the computers world kept marching along with Moore’s predictions in the last two decades. Concurring to Moore’s prediction, more transistors result in gaining greater speed. This great speed comes with the trade-off of producing high heat. This prediction has eventually reached to a wall that cannot be crossed unless new technologies are discovered because the heat issues became uncontrollable. One of the greatest discoveries to get over this wall is the NoC infrastructure, which was presented by Benini in 2002. This technology defines a practical solution to improve the energy efficiency and performance.

The inspiration for this work came from Ogras’s paper: "It's A Small World After All", where performance is enhanced for the application-specific NoC-based SoC by adding extra long-range links to two-dimensional mesh topologies. The main focus in this work is to improve the energy efficiency for a general purpose NoC-based SoC by finding the best possible extra links to add to a two-dimensional mesh topology via genetic algorithms. In the genetic algorithm, extra links are added randomly to form the different solutions for this NP-Hard problem. Comparing the energy consumption results of the new NoC design to the regular mesh topology, an improvement of 19% in energy per throughput is obtained. Ultimately, it was found that the more and the longer the links, the higher energy efficiency is achieved.

Main Content
Current View