Image sensor integrated with convolutional neural network computation circuit

Abstract

An image sensor integrated with a convolutional neural network computation circuit is provided. The image sensor includes: a pixel array including pixels divided into pixel groups, wherein each pixel converts a light signal into a PWM signal; a convolution computation circuit controlling a turn-on time of a corresponding weighted current according to the first PWM signal of each pixel, and accumulating the weighted currents into an integrated current; a comparison circuit converting the integrated current into a second PWM signal and comparing it with that of an adjacent pixel group to output a larger one; and a classification circuit quantizing the second PWM signal to a quantization value according to a weight of a node in a fully-connected layer corresponding to each pixel group, accumulating the quantization values of all pixel groups into a feature value, and comparing the feature value with a feature threshold to obtain a classification result.