PULSE-DENSITY MODULATION TECHNIQUE IN VLSI IMPLEMENTATIONS OF NEURAL NETWORK ALGORITHMS

New implementations of fully connected neural network architecture are explored. We will present some efficient implementations based on the pulse-density modulation technique. These VLSI circuits are fully programmable and can thus be used in many applications. The architecture is implemented by using two different approaches: analog implementation with switched-capacitor structures and fully digital implementation. These approaches are also compared from the VLSI point of view. The advantages of the designs are the simple structures of the artificial synapse and neuron, and thus small area, modularity, and expandability. Thus, they can be used for various other neural network algorithms. © 1990 IEEE