An efficient and compact integration of CMOS image sensors and cellular neural network (CNN) for intelligent processing

Abstract

By using the neuron-bipolar junction transistor (BJT) as phototransistor and single-transistor neuron, the cellular neural network (CNN) can be compactly integrated with CMOS image sensors so that the optical images can be input to the CNN directly for neural image processing. With the neuron-BJT, realized by the parasitic pnp BJT in n-well CMOS technology, the optical-input CNN with symmetric templates can be implemented in very small chip area. The cell area can be as small as 20 /spl mu/m/spl times/24 /spl mu/m. The simulation results have confirmed the correct function of the proposed optical-input compact CNN.