Neural network acceleration device

Abstract

The invention relates to the technical field of computers, and provides a neural network acceleration device which is used for bearing the operation of a preset neural network. The device comprises aconvolution calculation unit, an input feature map unit and a weight unit which are connected with the input end of the convolution calculation unit, and an output feature map unit connected with theoutput end of the convolution calculation unit, wherein the convolution calculation unit is determined based on the structure of the preset neural network. The neural network acceleration device provided by the invention can be designed for different structures of the neural network, the calculation efficiency of the preset neural network is improved, and meanwhile, the calculation of the preset neural network is completed in the device without the participation of a central processing unit, so that the memory bandwidth requirement and the hardware resource consumption are reduced.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to a neural network acceleration device. Background technique [0002] Neural network algorithms have been widely used in the field of artificial intelligence, especially in subdivided fields such as speech recognition, image processing, and automatic driving. The convolutional neural network algorithm requires a lot of calculations in model training and application, but limited by the characteristics of the algorithm and the calculation itself, traditional computing chips can no longer meet the computing needs, and the neural network needs to be accelerated through accelerators. [0003] In the prior art, there are neural network accelerators designed for GPU (Graphics Processing Unit, graphics processing unit) / DPU (Deeplearning Processing Unit, deep learning processor) / NPU (Neural Network Processing Unit, neural network processor), but , this type of neural network ac...

AI Technical Summary

Problems solved by technology

The convolutional neural network algorithm requires a lot of calculations in model training and application, but limited by the characteristics of the algorithm and calculation itself, traditional computing chips can no longer meet the computing needs, and the neural network needs to be accelerated by accelerators

[0003] In the prior art, there are neural network accelerators designed for GPU (Graphics Processing Unit, graphics processing unit) / DPU (Deeplearning Processing Unit, deep learning processor) / NPU (Neural Network Processing Unit, neural network processor), but , this type of neural network accelerator adopts a general-purpose design, which has low computational efficiency, large memory bandwidth requirements, and consumes a lot of hardware resources.

Images