A general purpose neuro-image processor architecture
Gunhee Han, E. Sánchez‐Sinencio · 2002
A general purpose neuro-image processor architecture based on the Multiplexing Neuron Array (MNA) is proposed. It is a pseudo parallel matrix-vector multiplier. The proposed architecture can process not only vector signals but two dimensional signals as well, so various types of neural networks or image processing paradigm can be implemented without any hardware modification. The major advantage is an unlimited expansion capability due to simple routings between processing elements. The architecture can be implemented with any discrete-time system, i.e., digital system, Switched-Capacitor, Switched-Current. For illustration, a multilayer perceptron network and recurrent network are implemented with fabricated Switched-Capacitor CMOS chips.