Ferroelectric thin-film memory for electrically programmable IC neural networks

Abstract

The application of ferroelectric thin-film capacitors as connection element “synapses” in artificial neural network (ANN) integrated circuits (IC's) is described. A review of basic neural network structure, with an emphasis on electrical implementation in IC form is given. A binary (on/off efficacy) connection scheme of very high density is then described and analyzed. A continuous-valued synapse (analog memory) where the efficacy is controlled by the polarization of a ferroelectric capacitor element is then described. This analog synapse utilizes non-destructive readout and requires very infrequent refresh.