A neural network image sensor array architecture based on two-dimensional materials

Abstract

The invention discloses a neural network image sensor array structure based on two-dimensional materials. A pixel array circuit with the functions of photosensitive, photoelectric conversion, convolution calculation and data readout in the pixel; the basic unit is formed in an array arrangement in row and column directions, and the pixel array circuit is connected to the digital-to-analog conversion module through the convolution unit, and each pixel of the image is connected to the digital-to-analog conversion module. The data is input into the basic unit for convolution operation, and then output from the digital-to-analog conversion module; one basic unit controls one pixel of the image, and the weight value of the convolution kernel is used to control the back gate voltage value of one element in the basic unit, thereby controlling the photosensitivity ; Switch the charge and discharge of the conduction control capacitor through the PMOS transistor of the column, and then control the capacitor voltage of each group of components to realize the pixel convolution operation control. The invention can complete the first layer operation of the convolutional neural network at the same time of photoreception, and convert it into a digital signal for output, thereby reducing the energy consumption of subsequent calculations, and has the characteristics of high dynamic range, high frame rate, low power consumption and the like.

Description

technical field [0001] The invention discloses a novel two-dimensional material WSe based on 2 The neural network image sensor architecture. A pixel array circuit that realizes photosensitive, photoelectric conversion, convolution calculation and data readout in pixels and can be used to realize array convolution operation, especially relates to an image sensor capable of completing the first layer convolution operation of convolutional neural network. Convolution operation architecture. Background technique [0002] Modern artificial intelligence originated from the electronic neural network experiment in the 1950s, and in the following 70 years, it has had a huge impact on human society. Convolutional Neural Network (CNN), as one of the important algorithms in deep learning, has been widely used in image processing, target recognition, target tracking and other directions due to its superior performance in the field of machine vision. [0003] At the same time, the Inte...

AI Technical Summary

Problems solved by technology

However, these studies still have not fully utilized the characteristics of the image sensor itself. As an interface for photoelectric conversion, the pixel array itself has the function of analog storage and has the conditions to complete simple calculations.

Images