Advanced dedicated hardware for simulating biologically inspired neural networks

In this paper, we present a digital system for simulating very large scale spiking neural networks (VLSNN). This system makes it possible to simulate a large neural network with features such as configurable connections, synaptic short term plasticity, long term plasticity with simultaneous impact of two variables, i.e. Hebbian learning rules of miscellaneous synoptic or dendritic parameters. We describe the data preparation for a neural network with a large amount of various parameters and the system components, with the efficiently data caching of the activity data in the tag-module and the topology unit. The hardware modules are described in VHDL and the system is implemented on a Virtex4-FPGA