Dynamic power reduction of FPGA-based reconfigurable computers using precomputation

  • Authors:
  • Chi-Chiu Tsang;Hayden Kwok-Hay So

  • Affiliations:
  • The University of Hong Kong, Hong Kong;The University of Hong Kong, Hong Kong

  • Venue:
  • ACM SIGARCH Computer Architecture News
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper examines the effectiveness of employing precomputation techniques to reduce power consumption of field configurable computing systems. Multiplier is modified with precomputation techniques and are implemented using commercial off-the-shelf FPGAs. Precomputation techniques reduce dynamic power consumption of a module by eliminating unnecessary signal switching activities in inactive portions of the modules. Experiments have shown that up to 52% of logic and signal power consumption can be reduced in multiplier module. Furthermore, when compared to ASIC implementations, FPGA implementations of precomputation modules have the advantage of lower area overhead as most of them can be implemented using originally unoccupied related FPGA resources. Finally, it was found that the effectiveness of precomputation depends heavily on the input data statistics. It is expected that compilers for future reconfigurable computers may take full advantage of such power saving techniques by optimizing the architecture according to data input statistics.