A fully-differential CMOS implementation of Oja's learning rule in a dual-synapse neuron for extracting principal components for face recognition

A fully-differential, CMOS implementation of a self-organizing, dual-synapse neuron with on-chip learning for real-time facial feature extraction is presented. The adaptation of the network follows Oja's learning rule and the synaptic weight vector is shown to adapt to the principal component vector of the set of two-dimensional input vectors.