New digital architecture of CNN for pattern recognition

This paper deals with design of the novel architecture of the CNN (cellular neural network) for image processing. The main parameters were speed and area of the chip. Digital CNN network contains a lot of gates, which occupies a lot of area on chip. The most area on chip occupy multipliers, therefore we proposed CNN network, where multiplication of the input with weight is realized by block AND, while multiplied signal is special distributed in the time. Designed network we compared with standard CNN network which consists of parallel signed multipliers.