A new VLSI architecture for perceptron network

A new architecture for multilayer perceptron network called the sliding feeder perceptron neural network is proposed. The sliding feeder is a shift register that can shift the input activities one by one to the neurons for processing. Compared to the conventional perceptron neutral network, this new architecture requires less multipliers. Only N multipliers are needed instead of N² multipliers for the conventional perceptron neural network. The connection between the neurons are also very simple. There are no overlapping connection between the neurons. This implies fewer layers of metal for fabrication. Furthermore, this new architecture is based on a modular design. Hence, the network can be extended easily. Only the micro-code need to be changed. Hardware simulations has been done on a 3-layer network and the feedforward and back propagation processes are proven to be functional