A current mode analogue circuit for implementing artificial neural networks

Analogue circuits provide an efficient way of implementing artificial neural networks. They carry out the required mathematical operations with low circuit complexity and are naturally asynchronous. A current mode circuit is described which emulates the important characteristics of neuron operation. The synapse essential to network operation may be configured as excitatory or inhibitory with a single input and the synaptic strength is proportional to the magnitude of T(i,j) with respect to Vref. An appropriate BIMOS implementation has been suggested which allows networks of arbitrary complexity to be realised with the minimum of ancillary components and would therefore enable rapid confirmation of theoretical results