Hassan, F. (2011) Design and modelling of variability tolerant on-chip communication structures for future high performance system on chip designs. PhD thesis, University of Glasgow.
This is the latest version of this eprint.
PDF
Download (2MB) |
Abstract
The incessant technology scaling has enabled the integration of functionally complex System-on-Chip (SoC) designs with a large number of heterogeneous systems on a single chip. The processing elements on these chips are integrated through on-chip communication structures which provide the infrastructure necessary for the exchange of data and control signals, while meeting the strenuous physical and design constraints. The use of vast amounts of on chip communications will be central to future designs where variability is an inherent characteristic. For this reason, in this thesis we investigate the performance and variability tolerance of typical on-chip communication structures. Understanding of the relationship between variability and communication is paramount for the designers; i.e. to devise new methods and techniques for designing performance and power efficient communication circuits in the forefront of challenges presented by deep sub-micron (DSM) technologies.
The initial part of this work investigates the impact of device variability due to Random Dopant Fluctuations (RDF) on the timing characteristics of basic communication elements. The characterization data so obtained can be used to estimate the performance and failure probability of simple links through the methodology proposed in this work. For the Statistical Static Timing Analysis (SSTA) of larger circuits, a method for accurate estimation of the probability density functions of different circuit parameters is proposed. Moreover, its significance on pipelined circuits is highlighted. Power and area are one of the most important design metrics for any integrated circuit (IC) design. This thesis emphasises the consideration of communication reliability while optimizing for power and area. A methodology has been proposed for the simultaneous optimization of performance, area, power and delay variability for a repeater inserted interconnect. Similarly for multi-bit parallel links, bandwidth driven optimizations have also been performed. Power and area efficient semi-serial links, less vulnerable to delay variations than the corresponding fully parallel links are introduced. Furthermore, due to technology scaling, the coupling noise between the link lines has become an important issue. With ever decreasing supply voltages, and the corresponding reduction in noise margins, severe challenges are introduced for performing timing verification in the presence of variability. For this reason an accurate model for crosstalk noise in an interconnection as a function of time and skew is introduced in this work. This model can be used for the identification of skew condition that gives maximum delay noise, and also for efficient design verification.
Item Type: | Thesis (PhD) |
---|---|
Qualification Level: | Doctoral |
Keywords: | Variability tolerant, on-chip communication, Communication Structures, Network-on-Chip Links |
Subjects: | T Technology > TK Electrical engineering. Electronics Nuclear engineering |
Colleges/Schools: | College of Science and Engineering > School of Engineering |
Supervisor's Name: | Rodriguez-Salazar, Dr. Fernando |
Date of Award: | 2011 |
Depositing User: | Mr. F. Hassan |
Unique ID: | glathesis:2011-2742 |
Copyright: | Copyright of this thesis is held by the author. |
Date Deposited: | 28 Jan 2020 16:11 |
Last Modified: | 28 Jan 2020 16:11 |
URI: | https://theses.gla.ac.uk/id/eprint/2742 |
Available Versions of this Item
-
Design and modelling of variability tolerant
on-chip communication structures for
future high performance system on chip
designs. (deposited 05 Jul 2011)
- Design and modelling of variability tolerant on-chip communication structures for future high performance system on chip designs. (deposited 28 Jan 2020 16:11) [Currently Displayed]
Actions (login required)
View Item |
Downloads
Downloads per month over past year