Convolutional neural network processor, implementation method, electronic equipment and storage medium

Abstract

The invention provides a convolutional neural network processor, an implementation method, electronic equipment and a storage medium, and the convolutional neural network processor is connected with a main processor through a bus protocol, and comprises an instruction module which is used for receiving and analyzing an instruction issued by the main processor; the content of the instruction at least comprises to-be-processed image related data, convolutional neural network related data and feature map related data; the buffer module is used for storing related data of convolutional neural network operation; the calculation module is used for realizing convolutional neural network operation; the control module is used for controlling the data to be written in and read out of the buffer module according to a corresponding instruction; and performing corresponding convolutional neural network operation on the corresponding to-be-processed image according to the related content of the execution instruction of the instruction control calculation module. According to the method and the device, the corresponding convolutional neural network operation can be performed according to the instruction sent by the main processor, the acceleration operation of the convolutional neural network operation can be realized, and the method and the device have universality and good flexibility.

Description

technical field [0001] The application belongs to the technical field of convolutional neural network, and specifically relates to a convolutional neural network processor, an implementation method, electronic equipment and a storage medium. Background technique [0002] Convolutional Neural Networks (CNN) is a classic and widely used deep learning neural network structure. The characteristics of local connection, weight sharing and pooling operation of convolutional neural network can effectively reduce the complexity of the network, reduce the number of training parameters, make the model invariant to translation, distortion and scaling to a certain extent, and It has strong robustness and fault tolerance, and is also easy to train and optimize the network structure. [0003] Since the calculation process of the convolutional neural network is relatively complicated and the data processed is large, most of the current convolutional neural network models have the problem o...

AI Technical Summary

Problems solved by technology

However, the GPU can only do instruction pipeline, not data pipeline, and the power consumption is too high

However, ASICs can only support specific convolutional neural network operations, and the development cycle is long. In the current era of rapid changes in neural network algorithms, the algorithms are often outdated when they are released.

Images