Performance study of spike visual processing on a Supercomputer: AER-based spike convolution processing on a 320-core cluster

Human brain is composed by millions of parallel neurons that process the visual information in a continuous way, spike by spike, from the information received through the retina. In this processing there are no frames in the video like it occurs in digital video. One frame every 40 ms implies information loss in between two frames. Address-Event-Representation (AER) is mechanism for multiplexing in time the stream of spikes generated by a set of neurons implemented on a chip, FPGA or microprocessor. These AER visual processing systems have been simulated in software. In this paper we present a performance study of the CRS supercomputer of the University of Cadiz (UCA) for AER-based convolution processing.