Efficient modelling of spiking neural networks on a scalable chip multiprocessor

摘要: We propose a system based on the Izhikevich model running scalable chip multiprocessor - SpiNNaker for large-scale spiking neural network simulation. The design takes into account requirements processing, storage, and communication which are essential to efficient modelling of networks. To gain speedup processing as well saving storage space, is implemented in 16-bit fixed-point arithmetic. An approach using two scaling factors developed, making precision comparable original. With scheme, all firing patterns by original can be reproduced with much faster execution speed. reduce overhead, rather than sending synaptic weights communicating, we only send out event packets indicate neuron firings while holding memory post-synaptic neurons, so-called event-driven algorithm. handled efficiently multicast supported machine. also describe level simulation schemes above. has been functionally verified experimental results included. analysis performance whole presented at end paper.