Fixed Point Neuro-Inspired DSP Using Spiking Neurons in a GALS Architecture

Mesquida, Thomas;Valentian, Alexandre;Bol, David;Beigne, Edith
(2017) 2017 IEEE International Symposium on Asynchronous Circuits and Systems (ASYNC 2017) — Location: San Diego (USA) (21.May.2017)

Files

FixedPointNeuro-InspiredDSPUsingSpikingNeuronsinaGALSArchitecture.pdf
  • Restricted Access
  • Adobe PDF
  • 208.23 KB
Request a copy

Details

Authors
  • Mesquida, ThomasUniversité Grenoble Alpes, CEA, LETI/France
    Author
  • Valentian, AlexandreUCLouvain
    Author
  • Bol, Davidorcid-logoUCLouvain
    Author
  • Beigne, EdithUCLouvain
    Author
Abstract
In this paper, we expose architectural solutions for a neuromorphic signal processing chip for the Internet of Things. We use Spiking Neural Networks with set topology and weight to implement arithmetic and signal processing functions using the spikes to carry information. The obtained framework of operators is capable of flexible Fixed Point operations. We show in this paper that our Globally Asynchronous Locally Synchronous architecture using Dynamic Voltage and Frequency Scaling is very well suited to lower the overall energy of the proposed architecture and to bring more flexibility.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Mesquida, T., Valentian, A., Bol, D., & Beigne, E. (2017). Fixed Point Neuro-Inspired DSP Using Spiking Neurons in a GALS Architecture. 2017 IEEE International Symposium on Asynchronous Circuits and Systems (ASYNC 2017), San Diego (USA). https://hdl.handle.net/2078.5/252584