Real-Time Monitoring Tool for SNN Hardware Architecture

Abstract

Spiking Neural Networks (SNN) are characterized by their brain-inspired biological computing paradigm. Large-scale hardware platforms are reported, where computational cost, connectivity, number of neurons and synapses, speed, configurability, and monitoring restriction, are some of the main concerns. Analog approaches are limited by their low flexibility and the amount of time and resources spent on prototype development design and implementation. On the other hand, the digital SNN platform based on System on Chip (SoC) offers the advantage of the Field-programmable Gate Array (FPGA) technology, along with a powerful Advanced RISC Machine (ARM) processor in the same chip, that can be used for peripheral control and high-bandwidth direct memory access. This paper presents a monitoring tool developed in Python that receives spike data from a large-scale SNN architecture called Hardware Emulator of Evolvable Neural System for Spiking Neural Network (HEENS) in order to on-line display in a dynamic raster plot in real-time. It is also possible to create a plain text file (.txt) with the entire spike activity with the aim to be analyzed offline. Overall, the monitoring tool and the HEENS functionalities working together show great potential for an end-user to bring up a neural application and monitor its evolution introducing a low delay, since a FIFO is used to temporarily store the incoming spikes to give the processor time to transmit data to the PC through Ethernet bus, without affecting the neural network execution.