Implementation of a Minimal Recurrent Spiking Neural Network in a Solid-State Device - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue Physical Review Applied Année : 2021

Implementation of a Minimal Recurrent Spiking Neural Network in a Solid-State Device

Résumé

We study the minimal recurrent spiking neural network of a single neuron with an autaptic synapse. We implement the neural system in the solid state with a recently introduced ultracompact neuron (UCN) model, which is based on the memristive properties of a thyristor. The UCN is supplemented by a self-synaptic, autaptic, connection, where we control the feedback. Both excitatory and inhibitory cases are considered. We explore the systematic behavior as a function of autaptic intensity and feedback time delay. We realize a tunable dynamic memory, showing graded persistent activity, where short excitatory and inhibitory pulses allow the firing rate to be controlled. We finally reproduce recent experimentally observed behavior of a biological autapse measured in vivo, finding excellent qualitative agreement. Our work opens the way into the field of solid-state neuroscience, with the UCN as an accessible platform to implement and experimentally study the dynamic behavior of spiking neural networks.
Fichier principal
Vignette du fichier
Stoliar_autapse_PhysRevApp_2021.pdf (2.19 Mo) Télécharger le fichier
Origine Fichiers éditeurs autorisés sur une archive ouverte

Dates et versions

hal-03451369 , version 1 (26-11-2021)

Identifiants

Citer

Pablo Stoliar, Olivier Schneegans, Marcelo J Rozenberg. Implementation of a Minimal Recurrent Spiking Neural Network in a Solid-State Device. Physical Review Applied, 2021, 16, ⟨10.1103/physrevapplied.16.034030⟩. ⟨hal-03451369⟩
32 Consultations
107 Téléchargements

Altmetric

Partager

Gmail Mastodon Facebook X LinkedIn More