Skip to Main content Skip to Navigation
Conference papers

Spiking Neural Network Hardware implementation : from single neurons to networks

Abstract : Spiking Neural Networks (SNN) are used in computational neurosciences to simulate information processing in the brain. SNN implement biologically plausible models of neural networks, from the detailed neuron physiology to network adaptation and plasticity rules. Various implementation solutions exist for SNN, including software, hardware or mixed systems. In this presentation, we will focus on hardware-based SNN and propose a review of the most recently developed platforms as well as pioneer platforms. These hardware-based platforms often use analog cores to emulate the neuron-level behavior while other computation or configuration tasks are distributed on digital hardware or software units. First we will introduce the principle of hardware implementation, from the primitive computational elements (transistors, Boolean operators) to the description of ASIC (Application Specific Integrated Circuit) design flow. Typical performance criteria will be pointed out to compare analog/digital hardware designs, supported by a detailed example of a real analog solver (naturally continuous- and real-time). Before reviewing the hardware SNN solutions, we will present an overview of standard computational models, from the cellular or sub-cellular level, to the network level. At the neuron-level, models can be classified in two main categories: conductance-based or threshold-type, with several levels of abstraction in each category. At the network level, connectivity between neurons can be managed in different ways and plasticity rules can be also implemented with more or less details and parameters. This brief overview will highlight the trade-off between biological plausibility and computational cost. It finally shows how the models choice is closely related to the SNN application field. The reviewed hardware-based SNN will be compared in terms of: network size, neuron model accuracy, ability to real-time operation, multi-scale configurability and ability to hybrid living-artificial connection. A few systems will be described in a more detailed fashion, within their specific application context. Finally, some projects which are currently in progress using hardware SNN will be exposed. Their goal is to explore novel paradigms for information processing and overcome the related technological limitations.
Document type :
Conference papers
Complete list of metadatas
Contributor : Noëlle Lewis <>
Submitted on : Friday, May 2, 2014 - 10:50:04 AM
Last modification on : Monday, March 30, 2020 - 2:20:03 PM


  • HAL Id : hal-00986320, version 1


Noëlle Lewis, Yannick Bornat, Sylvie Renaud. Spiking Neural Network Hardware implementation : from single neurons to networks. in NeuroComp'09, Conférence de Neurosciences Computationnelles, 2009, Bordeaux, France. ⟨hal-00986320⟩



Record views