Multi-objective Spiking Neural Network Hardware Mapping Based on Immune Genetic Algorithm

Abstract

For the Spiking Neuron Network (SNN) systems, the hardware implementation has unique advantages in terms of performance, energy, and scalability. The Networks-on-Chip (NoC) interconnection strategy has been widely used in hardware SNNs as it provides excellent interconnection mechanism for interneuronal communications. However, the mapping between the SNN models and NoC hardware systems remains a research challenge. In this paper, a multi-objective immune genetic algorithm is proposed for the mapping of SNN hardware system, which is based on the Immune Algorithm (IA) and Genetic Algorithm (GA). It can optimize the SNN hardware systems by reducing the energy consumption and communication delays. In the experiments, the spiking astrocyte neuron network model and the Star-Subnet-Based-3D Mesh (3D-SSBM) NoC hardware system are used for testing. Results demonstrate that the proposed algorithm provides an effective mapping solution for hardware SNNs with low energy consumption and communication delay.