Monitoring implementation for spiking neural networks architecture on Zynq-7000 all programmable SoCs

Abstract:

This paper describes a Spiking Neural Networks (SNN) architecture mapped on a Zynq-7000 family of programmable system-on-chip (SoC) devices. The novel spike monitoring system based on AER Protocol, uses AXI4 Stream interface for high-speed streaming data communication between the PL and PS. Since the spike transmission is carried out during the distribution phase, we use an interruption to notify the PS that new spike data are being stored into the DDRAM in order to be retransmitted to a PC for monitoring purposes through Ethernet Protocol. The frame transmitted to the PS includes time-stamp, ID chip identifier, spikes packages followed by 16-bit CRC code for error detections. The solution is implemented using VHDL language to take advantage of the FPGA parallelism, developing its own dedicated IP modules for data validation and creating each frame. As result, we obtain an optimal and …

Año de publicación:

2021

Keywords:

Fuente:

google