An exponential-decay synapse integrated circuit for bio-inspired neural networks

Abstract

Biologically-realistic artificial neurons and synapses are implemented on analog specific integrated circuits, to offer real-time computation. We present here specific circuits where synaptic interactions are modeled using exponential-decay synapses, which are digitally controlled. The design method is detailed, and experimental measurements on fabricated circuits are presented to illustrate the synaptic mechanisms. We finally propose to exploit the synapses to build a mixed analog-digital neural simulator, where an analog computing core is digitally controlled to simulate synaptic plasticity or learning algorithms.