Docstoc

Fisher

Document Sample
Fisher Powered By Docstoc
					   Thermal-Aware Reliability for
       Real-Time Systems
  (inspired by title of concurrent talk in Workshop on Low
Power System on Chip -- “Reliability/Wear out-Aware Design”
                        by Mircea Stan)



                 Nathan Fisher
              Wayne State University
Reliability Model:
Electromigration (EM)
“[EM] is the aging process in metal
  interconnects caused by the exchange of
  momentum between electrons”
     -Z. Lu, J. Lach, M. Stan, and K. Skadron, IEEE Micro 2005


• Effect:
  – Atoms aggregation;
  – Voids form (open-circuit failure!)
Reliability Model:
Electromigration (EM)
• Mathematical Model of Void Growth Rate:

                        exp  Q / kT ( s, t ) 
r ( s, t )  I ( s, t )
                              kT ( s, t )
Where
 s(t) – speed function;
 I(s,t) – current density (e.g., speed-scaled:
                                     I(t) = σs2 (t) )
 kT(s,t) – thermal energy (e.g., Fourier’s Law :
                                     dT/dt = aP(t) – bT(t) ).
 Q – activation energy.
Lifetime Analysis:
Single Processor (Interconnect)
• L = lifetime of processor.
• Objective: Determine a speed schedule s that
  maximizes L s.t.
         L

          r(s, t)dt  F
         0
 and schedules real-time jobs {Ji = (ai, wi, di)}i
where
   F – Void Threshold
   ai – Arrival Time
   wi – Work Requirement
   di -- Deadline
             Issues/Questions
• How is reliability-aware different than
  temperature-aware?
  – Potential Answer: It might be better to run at a
    higher temp for short periods of time.
• Feasibility Algorithms:
  – Exact?
  – Approximation algorithms w/ ratio c that
    guarantee a lifetime Lopt/c?
• Online Algorithms

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:3
posted:8/13/2012
language:
pages:5