Title
A low-power wireless-assisted multiple network-on-chip.
Abstract
Multiple network-on-chip (Multi-NoC) architectures are supposed to distribute the network traffic categorically among disjoint sub-networks. The main objective is significant energy reduction through power-gating of unused sub-networks. However, the packets are delayed due to sleep/wake cycles, which directly influences the overall performance of the system. In addition, the communication infrastructure of the Multi-NoC should be selected carefully to avoid performance degradation. Our solution to address these issues is using wireless links, which is used to relax the timing restrictions on sleep/wake cycles to save more power without losing performance. To realize wireless communications, we adopt two types of on-chip wireless technology that operate at different frequency bands, namely terahertz (THz) and millimeter-wave (mmW). To evaluate the merits of the proposed architecture that employs these wireless technologies, we used both real application benchmarks (PARSEC and SPLASH-2) and synthetic traffics on a many-core processor. For THz technology, the proposed architecture results in nearly 51% and 10% power reduction compared to traditional single network-on-chip (Single-NoC) and a power-gated 4-subnets Multi-NoC, respectively. The corresponding results for mmW technology show 46% and 6% power reduction. Also, the proposed architecture for THz and mmW technologies results in 10% and 7% latency reduction compared to Multi-NoC, respectively. The performance metrics of the proposed architecture is comparable to Single-NoC architecture, which demonstrates the effectiveness of our proposal.
Year
DOI
Venue
2018
10.1016/j.micpro.2018.09.001
Microprocessors and Microsystems
Keywords
Field
DocType
Multiple network-on-chip,Wireless network-on-chip,Hybrid network-on-chip,Low power,Energy proportionality,Power-gating
Architecture,Disjoint sets,Parsec,Wireless,Computer science,Latency (engineering),Network packet,Network on a chip,Real-time computing,Computer hardware,Radio spectrum
Journal
Volume
ISSN
Citations 
63
0141-9331
1
PageRank 
References 
Authors
0.35
40
2
Name
Order
Citations
PageRank
Mohammad Baharloo110.69
Ahmad Khonsari221042.43