Abstract
dc:descriptionThe importance of System-on-Chip (SoC) interconnect technology is growing with each generation of new SoC devices. Networks-on-chip (NoC) provided the needed disruptive interconnect technology to scale SoC interconnects from simple buses to fullyfl edged interconnection networks that behave like mini internets inside the chip. NoCs are now part of all SoCs and are being used in a variety of market segments, ranging from multimedia and telecom to automotive and medical devices. There are several signs, however, that the current NoC architectures are faced with major challenges in satisfying stringent performance and power requirements. To this end, we propose three complementary techniques that allow for the design of low-power NoC architectures and the development of a methodology for automatically verifying the NoC’s peak power consumption. Two of the techniques tackle the reduction of the power consumed by NoC buffers. In the first approach, a low-cost virtual-channel based shared buffer is proposed that drops the buffering requirements close to the absolute minimum, without sacrificing network performance. The proposed buffers can be used on the links, as a distributed elastic buffering architecture, or at the inputs and the outputs of NoC routers. The second approach employs multi-bit register composition to reduce the number of register cells of the design, thus significantly reducing the overall clock-tree complexity and power. Multi-bit register composition follows a balanced restructuring approach, where the reduction of the number of registers does not degrade timing slack, wire length or routing utilization. The introduced low-cost buffers, after being enhanced with self-testability hardware components, have been employed in the design of a scalable distributed NoC architecture that employs virtual channels. The proposed architecture removes the burden of the tight placement of NoC components and allows them to be physically spread throughout the chip, irrespective of the network topology. This feature can be used either for reducing power consumption, or achieving higher clock rates. The novel micro-architectures and the clock-tree complexity reduction methodology are complemented by a fully-automated methodology that produces appropriate traffic and data patterns that cause peak power consumption within the NoC. In this way, one can have a realistic estimate of a design’s peak power consumption, which directly impacts other salient system attributes, such as performance, implementation costs, battery life, and reliability.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2018
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Seitanidis, Ioannis
- Σεϊτανίδης, Ιωάννης
Subjects
dc:subject × 11- Δίκτυα σε ολοκληρωμένα κυκλώματα
- Ολοκληρωμένα κυκλώματα
- Συστήματα σε ολοκληρωμένα κυκλώματα
- Χαμηλή κατανάλωση ισχύος
- Networks-on-Chip
- Low power
- System-on-chip
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Engineering and Technology
- Electrical Engineering, Electronic Engineering, Information Engineering
Rights
dc:rights- Statement dc:rights
-
- BY
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/45974