Signalling techniques and their effect on neural network implementation sizes

A series of models are developed which predict the silicon area consumed by a neural network. These models predict the area consumed by different part s of a neural network and the effect of the use of different signalling typ es. The relative size of neural networks that use these different:signallin g types may thus be assessed. The silicon area consumed by neural networks implemented with local weights and single line inputs is shown to be orders of magnitude smaller than other possible neural network implementations. T he use of single line transmission is shown to be the next most effective m ethod. Differential or parallel digital data transmission techniques are sh own to be the least satisfactory options with respect to silicon area consu mption. In addition the use of rectangular synapse cells is shown to reduce the interconnect area consumed, while asymmetrical signalling techniques a re shown to be advantageous. (C) 2001 Elsevier Science Inc. All rights rese rved.