Convolutional neural network reasoning accelerator and acceleration method

Abstract

The invention discloses a convolutional neural network reasoning accelerator and an acceleration method, and belongs to the field of hardware implementation of an artificial intelligence algorithm. Aiming at the problems of high power consumption, poor configurability, low calculation precision and the like in the prior art, the invention provides the convolutional neural network reasoning accelerator and an acceleration method. The accelerator comprises a main control module and an address generation module. The system comprises an SRAM storage module, a data input module, a calculation engine module and a result output module, wherein parallel computing units in the computing engine module are arranged independently; on the basis of limited computing resources and storage resources, theperformance advantage of high parallelism is achieved, the parallel computing units have fixed-point interception and turn-off functions, the parallelism and the computing precision of the acceleratorare improved, and the use power consumption is reduced.

Description

technical field [0001] The invention relates to the field of hardware implementation of artificial intelligence algorithms, and more specifically, relates to a convolutional neural network reasoning accelerator and an acceleration method. Background technique [0002] Convolutional neural network is a deep feed-forward artificial neural network and one of the representative algorithms of deep learning. It has been successfully applied in computer vision, natural language processing and other fields. In a convolutional neural network, the convolutional layer can account for more than 80% of the total computing load and computing time of the entire network. Therefore, the acceleration of the convolutional layer is the key to improving the performance of the entire CNN network. In the calculation of the convolution layer, there are the following parameters: convolution kernel size (K), zero padding method (Pa), convolution step size (St), number of input image channels (Ch), nu...

AI Technical Summary

Problems solved by technology

[0007] Aiming at the problems of slow calculation speed, high power consumption, poor configurability and fixed-point accuracy existing in the prior art, the present invention provides a convolutional neural network reasoning accelerator and an acceleration method, which can increase the calculation speed, reduce power consumption, improve The configurability of the device ensures the accuracy of the calculation results

Images