A convolutional neural network accelerator based on calculation optimization of an FPGA

Abstract

The invention discloses a convolutional neural network accelerator based on calculation optimization of an FPGA. The convolutional neural network accelerator comprises an AXI4 bus interface, a data cache region, a pre-fetched data region, a result cache region, a state controller and a PE array. The data cache region is used for caching feature map data, convolution kernel data and index values read from an external memory DDR through an AXI4 bus interface; The pre-fetched data area is used for pre-fetching feature map data needing to be input into the PE array in parallel from the feature mapsub-cache area; The result cache region is used for caching a calculation result of each row of PE; The state controller is used for controlling the working state of the accelerator to realize conversion between the working states; And the PE array is used for reading the data in the pre-fetched data area and the convolution kernel sub-cache area to carry out convolution operation. The accelerator utilizes the characteristics of parameter sparsity, repeated weight data and an activation function Relu to end redundant calculation in advance, so that the calculation amount is reduced, and the energy consumption is reduced by reducing the access memory frequency.

Description

technical field [0001] The invention belongs to the field of electronic information and deep learning, in particular to a computing-optimized convolutional neural network accelerator hardware structure based on FPGA (Filed Programmable Gate Array). Background technique [0002] In recent years, the use of deep neural networks has grown rapidly and has had a significant impact on the world's economic and social activities. Deep convolutional neural network technology has received widespread attention in many machine learning fields, including speech recognition, natural language processing, and intelligent image processing. Especially in the field of image recognition, deep convolutional neural networks have achieved some remarkable results. In these domains, deep convolutional neural networks are able to achieve superhuman accuracy. The excellence of deep convolutional neural networks stems from its ability to extract high-level features from raw data after performing stati...

AI Technical Summary

Problems solved by technology

[0006] Based on the foregoing analysis, there is a problem of excessive redundant calculations in convolution calculations in the prior art, and this case arises from this

Images