A real-time hybrid neuron network for highly parallel cognitive systems

Abstract

For comprehensive understanding of how neurons communicate with each other, new tools need to be developed that can accurately reproduce and mimic the behaviour of such neurons in real-time. The proposed design in this thesis models an Inferior Olivary Nucleus network on an FPGA device, with a maximised amount of simulated neurons for the given FPGA family type. This has been achieved by the usage of a highly pipelined hybrid neuron network, which executes optimally scheduled floating-point operations that, together with open source IP, has resulted in a cost-effective solutions, capable of simulating responses faster or on par with their biological counterparts.

Digital Neuron Cells for Highly Parallel Cognitive Systems

“…The router fan-out in this case is 2, and can be changed according to the requirements of the implementation. The same holds true for the number of PhyCs in any cluster [15] [4], for the Hodgkin-Huxley model, implemented in the adjusted network . .…”

Energy-efficient multipath ring network for heterogeneous clustered neuronal arrays

“…Previous effort has been made to reduce throughput and resource costs by grouping neuron simulating hardware, memory blocks and control circuitry into units referred to as clusters [11,12,13]. The reuse of hardware for computational intensive mathematical operations such as exponentiation has successfully resulted in a reduction of resource consumption without increasing overall latency.…”

“…This thesis is based on work done in [11,12,13] where a hardware implementation of the extended Hodgkin Huxley model for the Xilinx Virtex 7 FPGA is presented with the overall system achieving real-time operation with a 50µs simulation step. Data locality is exploited, where neurons are connected with decreasing probability as the distance between them increases.…”

“…Apart from the implementation presented in section 2.3, several approaches have been suggested for simulating SNN with various degrees of biological realism on multiple platforms such as CPU and/or GPU, analog-digital hybrids, ASIC and FPGA. Among them it is worth mentioning: [11]. The box on the left is the axon and soma calculation.…”

“…• Configuration data is sent during setup or between simulation steps. It consists [11]. They represent ion concentrations and currents along with axon and soma potentials.…”