Exploring Spiking Neural Network on Coarse-Grain Reconfigurable Architectures

Abstract

<p> Today, recongurable architectures are becoming increas-<br /> ingly popular as the candidate platforms for neural net-<br /> works. Existing works, that map neural networks on re-<br /> congurable architectures, only address either FPGAs or<br /> Networks-on-chip, without any reference to the Coarse-Grain<br /> Recongurable Architectures (CGRAs). In this paper we<br /> investigate the overheads imposed by implementing spiking<br /> neural networks on a Coarse Grained Recongurable Ar-<br /> chitecture (CGRAs). Experimental results (using point to<br /> point connectivity) reveal that up to 1000 neurons can be<br /> connected, with an average response time of 4.4 msec.</p>