An artificial dendrite using active channels

Abstract

Since their introduction, neural networks have become an accepted object of research in various disciplines. Most of these neural networks are implemented using digital hardware consisting of computers or dedicated processors.

Analogue implementations of artificial neurons, the elementary processing units, could be smaller than their digital counterparts, thus enabling more complex networks on a single chip. Conventional methods of learning cannot be used directly in these networks, due to practical limitations regarding on-chip interconnections. In order to achieve such a complexity, it is necessary to refine the neural networks.

This article proposes an artificial dendrite, one of the most important parts of the neuron. The artificial dendrite uses principles found in biology like active propagation and shaping of action potentials using active channels. A brief introduction in neurophysiology is given in order to explain the underlying mechanisms. The model is simulated in SPICE using models of conventional analogue electronic devices.