Abstract

Previously, research and design of Network-on-Chip (NoC) paradigms where mainly focused on improving the performance of the interconnection networks. With emerging wide range of low-power applications and energy constrained high-performance applications, it is highly desirable to have NoCs that are highly energy efficient without incurring performance penalty. In the design of high-performance massive multi-core chips, power and heat have become dominant constrains. Increased power consumption can raise chip temperature, which in turn can decrease chip reliability and performance and increase cooling costs.

It was proven that Small-world Wireless Network-on-Chip (SWNoC) architecture which replaces multi-hop wire-line path in a NoC by high-bandwidth single hop long range wireless links, reduces the overall energy dissipation when compared to wire-line mesh-based NoC architecture. However, the overall energy dissipation of the wireless NoC is still dominated by wire-line links and switches (buffers).

Dynamic Voltage Scaling is an efficient technique for significant power savings in microprocessors. It has been proposed and deployed in modern microprocessors by exploiting the variance in processor utilization. On a Network-on-Chip paradigm, it is more likely that the wire-line links and buffers are not always fully utilized even for different applications. Hence, by exploiting these characteristics of the links and buffers over different traffic, DVFS technique can be incorporated on these switches and wire-line links for huge power savings.

In this thesis, a history based DVFS mechanism is proposed. This mechanism uses the past utilization of the wire-line links & buffers to predict the future traffic and accordingly tune the voltage and frequency for the links and buffers dynamically for each time window. This mechanism dynamically minimizes the power consumption while substantially maintaining a high performance over the system. Performance analysis on these DVFS enabled Wireless NoC shows that, the overall energy dissipation is improved by around 40% when compared Small-world Wireless NoCs.

Library of Congress Subject Headings

Networks on a chip--Energy conservation

Publication Date

7-2015

Document Type

Thesis

Student Type

Graduate

Degree Name

Computer Engineering (MS)

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Amlan Ganguly

Advisor/Committee Member

Dhireesha Kudithipudi

Advisor/Committee Member

Reza Azarderakhsh

Comments

Physical copy available from RIT's Wallace Library at TK5105.546 .I78 2015

Campus

RIT – Main Campus

Plan Codes

CMPE-MS

Share

COinS