SNAVA—A real-time multi-FPGA multi-model spiking neural network simulation architecture

Abstract

Spiking Neural Networks (SNN) for Versatile Applications (SNAVA) simulation platform is a scalable and programmable parallel architecture that supports real-time, large-scale, multi-model SNN computation. This parallel architecture is implemented in modern Field-Programmable Gate Arrays (FPGAs) devices to provide high performance execution and flexibility to support large-scale SNN models. Flexibility is defined in terms of programmability, which allows easy synapse and neuron implementation. This has been achieved by using a special-purpose Processing Elements (PEs) for computing SNNs, and analyzing and customizing the instruction set according to the processing needs to achieve maximum performance with minimum resources. The parallel architecture is interfaced with customized Graphical User Interfaces (GUIs) to configure the SNN's connectivity, to compile the neuron-synapse model and to monitor SNN's activity. Our contribution intends to provide a tool that allows to prototype SNNs faster than on CPU/GPU architectures but significantly cheaper than fabricating a customized neuromorphic chip. This could be potentially valuable to the computational neuroscience and neuromorphic engineering communities.

flexHH: A Flexible Hardware Library for Hodgkin-Huxley-Based Neural Simulations

“…On the FPGA front, there are a few simulator proposals that are quite notable, as well. SNAVA, by Sripad et al [11], is a multi-FPGA SNN simulator focused on largescale neuron simulations. The simulator supports a variety of models using 16-bit fixed-point arithmetic operations.…”

“…To update the gate variables (y j ) without custom ion gates, the derivative dy j / dt from (9) is used. For this equation, the calculations of both α i and β i are done using (11). For the implementation of (11), the amount of divisions and exponentiations is minimized, as Algorithm 2 shows, to reduce the hardware usage of this function.…”

“…If custom ion gates are supported, (10) can also be used, depending on the model needs. The transition rates α j and β j or the target value inf j and the time constant τ j are calculated either via(11) (which mirrors the functions of gate equations in NeuroML) if custom ion gates are not VOLUME 4, 2016…”

LEGION-based image segmentation by means of spiking neural networks using normalized synaptic weights implemented on a compact scalable neuromorphic architecture

“…Since one of our objectives is to create a prototype to be used in vision processing systems for mobile robots in which real-time processing is demanded and area resources are limited, we optimize an existing multi-model SNN architecture called SNAVA [21] to simulate the proposed SNN-LEGION configuration at high processing speeds by requiring low area consumption. It should be mentioned that the existing SNAVA architecture was designed as a generic multimodel SNN platform to simulate large-scale SNN networks paying a penalty in terms of area consumption and speed processing.…”

Portable exploitation of parallel and heterogeneous HPC architectures in neural simulation using SkePU

“…Neural simulation, especially biophysically-detailed simulation, is a HPC application type presenting a very high computational load, due to both the numerical intensity of the models and the large space and time scale of the simulations. This immense computational load is evidently efficiently handled on HPC infrastructure; the required memory is provided by thousands of nodes, working in a single cluster [1], and computational performance has been greatly boosted by manycore [5], GPU [3] and reconfigurable [12] accelerators.…”