Arrangement of data cells and neural network system utilizing such an arrangement

Claim 1

A layered neural network system comprising: a) resolving means for determining, for each layer, states of output neurons i based on input neurons j which are interconnected by way of synapses, b) means for storing synaptic coefficients C ij associated with these synapses, c) means for training and for updating the synaptic coefficients C ij ; and d) means for storing the neuron states, wherein the improvement comprises that the means for storing the synaptic coefficients comprises at least one word-organized memory arrangement comprising: a plurality of data cells arranged in rows and columns to form a matrix, a single output bus comprising a plurality of lines for supplying a data word corresponding to a plurality of the bits stored in a row or column of the matrix; at least a plurality of first and a plurality of second switches, each of said first switches being connected in series with a different one of said second switches between one of said cells and one of said lines of said output bus; means for addressing said matrix in a first mode such that the stored bits appear as a word in a first bit arrangement on said single output bus, means for addressing said matrix in a second mode such that, with the same cell contents in the matrix, the stored bits appear as a word in a second bit arrangement on said same single output bus, and said addressing means being connected to such matrix such that, in either addressing mode, a same selection signal when applied to the matrix will activate the same switches thereby addressing the same word stored in the matrix.