Quantization noise improvement in a hybrid distributed-neuron ANN architecture

This brief explores a useful self-scaling property of a hybrid (analog-digi tal) artificial neural network architecture based on distributed neurons. I n conventional sigmoidal neural networks with lumped neurons, the effect of weight quantization errors becomes more noticeable at the output as the ne twork becomes larger. However, it is shown here based on a stochastic model that the inherent self-sealing property of a distributed-neuron architectu re controls the output quantization noise (error) to signal ratio as the nu mber of inputs to an Adaline increases. This property contributes to a robu st hybrid VLSI architecture consisting of digital synaptic weights and anal og distributed neurons.