2008 Asia and South Pacific Design Automation Conference 2008
DOI: 10.1109/aspdac.2008.4484015
|View full text |Cite
|
Sign up to set email alerts
|

Run-time power gating of on-chip routers using look-ahead routing

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
2
1
1

Citation Types

1
58
0

Year Published

2014
2014
2024
2024

Publication Types

Select...
6
2
1

Relationship

0
9

Authors

Journals

citations
Cited by 54 publications
(59 citation statements)
references
References 7 publications
1
58
0
Order By: Relevance
“…As opposed to this, our method pursues a minimal size hardware implementation for the power management, with fast response to changing communication load since the complete logic is implemented in hardware. In addition the centralized approach in [2] does not scale well, in contrast to our approach .Matsutani et al present run-time power gating for network routers in several publications [13], [14]. But besides a not negligible resource overhead, additional computation overhead is necessary inside the router to produce the information required to generate a wake-up request signal.…”
Section: Related Workmentioning
confidence: 93%
“…As opposed to this, our method pursues a minimal size hardware implementation for the power management, with fast response to changing communication load since the complete logic is implemented in hardware. In addition the centralized approach in [2] does not scale well, in contrast to our approach .Matsutani et al present run-time power gating for network routers in several publications [13], [14]. But besides a not negligible resource overhead, additional computation overhead is necessary inside the router to produce the information required to generate a wake-up request signal.…”
Section: Related Workmentioning
confidence: 93%
“…In contrast, powergating with lookahead [10,11], leads to 9-11% application slowdown if implemented at a router level, while causing as much as 20% performance loss if fine-grained power-gating is applied. Finally, Panthre leakage power savings can be as much as 36.9% on average, under the fair assumption that 10-16 nodes of a 64-node CMP are communication-idle at any given time.…”
Section: Introductionmentioning
confidence: 99%
“…However, it further worsens the problem of accumulated wakeup latencies, as it puts components to sleep more aggressively. Early wakeup with lookahead routing was proposed to compensate for wakeup latency [10]. However, for a typical 2-stage pipeline router, lookahead can only hide a small fraction of the wakeup latency, which is typically many cycles.…”
Section: Introductionmentioning
confidence: 99%
“…Therefore, certain mechanism should be there for maintaining an optimum wake up latency. Various approaches have been adopted like Look ahead method [10], where the output port of a flit is calculated in advance to reduce wake up latency. Another technique called Look ahead with ever-on may be used where the first hop is always on and no wake up signal is required for that.…”
Section: Wake Up Latencymentioning
confidence: 99%