Compact associative memory for AER spike decoding in FPGA-based evolvable SNN emulation

Abstract

A spike decoding scheme for Address Event Representation (AER)-based transmission in Spiking Neural Network (SNN) emulators is introduced. The proposed scheme is a modified associative memory based on an efficient use of BRAM, supporting connectivity upgrade in real-time for hardware implementations of evolutionary networks. After analysing the different options and selecting the most efficient one, a prototype example based on FPGA is provided together with a novel hashing technique to demonstrate a compact on-chip solution for implementing inter-chip connectivity in SNN.