Energy efficiency of adaptive computing systems

Viktor Prasanna
University of Southern California
ceng.usc.edu/~prasanna

Recently energy dissipation has become an important performance 
metric. We outline an algorithm level design methodology for 
optimising the energy performance of configurable computing 
fabrics and introduce "malleable algorithms" as an approach to 
develop energy efficient soft IP(Intellectual Property) cores 
for Field Programmable Gate Arrays(FPGAs). We consider several 
kernels in adaptive signal processing in Software Defined Radio 
(SDR) and demonstrate energy and time efficient algorithms and 
implementations for these on configurable logic. We demonstrate 
improved performance using algorithmic optimizations. The 
performance of FPGAs is also compared against state of the art 
embedded processors, general purpose processors, and DSPs.