Abstract

Network-on-Chips (NoCs) have emerged as a communication infrastructure for the multi-core System-on-Chips (SoCs). Despite its advantages, due to the multi-hop communication over the metal interconnects, traditional Mesh based NoC architectures are not scalable in terms of performance and energy consumption. Folded architectures such as Torus and Folded Torus were proposed to improve the performance of NoCs while retaining the regular tile-based structure for ease of manufacturing. Ultra-low-latency and low-power express channels between communicating cores have also been proposed to improve the performance of conventional NoCs. However, the performance gain of these approaches is limited due to metal/dielectric based interconnection.

Many emerging interconnect technologies such as 3D integration, photonic, Radio Frequency (RF), and wireless interconnects have been envisioned to alleviate the issues of a metal/dielectric interconnect system. However, photonic and RF interconnects need the additional physically overlaid optical waveguides or micro-strip transmission lines to enable data transmission across the NoC. Several on-chip antennas have shown to improve energy efficiency and bandwidth of on-chip data communications. However, the date rates of the mm-wave wireless channels are limited by the state-of-the-art power-efficient transceiver design. Recent research has brought to light novel graphene based antennas operating at THz frequencies. Due to the higher operating frequencies compared to mm-wave transceivers, the data rate that can be supported by these antennas are significantly higher. Higher operating frequencies imply that graphene based antennas are just hundred micrometers in size compared to dimensions in the range of a millimeter of mm-wave antennas. Such reduced dimensions are suitable for integration of several such transceivers in a single NoC for relatively low overheads.

In this work, to exploit the benefits of a regular NoC structure in conjunction with emerging Graphene-based wireless interconnect. We propose a toroidal folding based NoC architecture. The novelty of this folding based approach is that we are using low power, high bandwidth, single hop direct point to point wireless links instead of multihop communication that happens through metallic wires. We also propose a novel phased based communication protocol through which multiple wireless links can be made active at a time without having any interference among the transceiver. This offers huge gain in terms of performance as compared to token based mechanism where only a single wireless link can be made active at a time. We also propose to extend Graphene-based wireless links to enable energy-efficient, phase-based chip-to-chip communication to create a seamless, wireless interconnection fabric for multichip systems as well. Through cycle-accurate system-level simulations, we demonstrate that such designs with torus like folding based on THz links instead of global wires along with the proposed phase based multichip systems. We provide estimates that they are able to provide significant gains (about 3 to 4 times better in terms of achievable bandwidth, packet latency and average packet energy when compared to wired system) in performance and energy efficiency in data transfer in a NoC as well as multichip system. Thus, realization of these kind of interconnection framework that could support high data rate links in Tera-bits-per-second that will alleviate the capacity limitations of current interconnection framework.

Library of Congress Subject Headings

Networks on a chip; Multichip modules (Microelectronics); Interconnects (Integrated circuit technology); Wireless communication systems

Publication Date

11-2017

Document Type

Thesis

Student Type

Graduate

Degree Name

Computer Engineering (MS)

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Amlan Ganguly

Advisor/Committee Member

Sonia Lopez Alarcon

Advisor/Committee Member

Andres Kwasinski

Campus

RIT – Main Campus

Plan Codes

CMPE-MS

Share

COinS