workaddresses theproblem ofapplication mapping in networks-on-chip (NoCs). Itexplores theimportance of characterizing network traffic toeffectively predict NoC energy consumption. Traffic isseenas an important factor affecting the problem ofmapping applications intoNoCshaving as goalto minimize thetotal dynamic energyconsumption ofa complex system-on-a-chip (SoC). Experiments showedthatfailing to consider thebittransitions influence ontraffic inevitably leads to an energyestimation error.Thiserrorisproportional tothe amountofbittransitions intransmitted packets. Inapplications that present a large numberofpackets exchange, theerroris propagated, significantly affecting themapping results. This paper proposesa high-level application modelthat captures thetraffic effect andusesittodescribe thebehavior ofapplications. Inorder toevaluate thequality oftheproposed model, a setofembedded systems weredescribed using both, apreviously proposed model (that doesnotcapture thetraffic effect), andthemodelproposed here. Comparing theresulting mappings, those derived from the proposed modelshowedimprovements inenergysavings with regard totheother modelforallexperiments.
[1]
Luca Benini,et al.
Packetization and routing analysis of on-chip multiprocessor networks
,
2004,
J. Syst. Archit..
[2]
William J. Dally,et al.
Route packets, not wires: on-chip inteconnection networks
,
2001,
DAC '01.
[3]
Sharad Malik,et al.
A technology-aware and energy-oriented topology exploration for on-chip networks
,
2005,
Design, Automation and Test in Europe.