Alexnet forward network accelerator based on FPGA (Field Programmable Gate Array)

Abstract

The invention discloses an Alexnet forward network accelerator based on an FPGA. The Alexnet forward network accelerator comprises a processor, an external memory, a buffer, a controller, an AX14-Litebus module and an AX14 bus module, wherein the AX14-Lite bus module and the AX14 bus module are respectively connected with the processor; the external memory, the buffer and the processor are respectively connected with the AX14 bus module; the external memory is used for storing accelerator operation or calculation data; the buffer is used for storing an intermediate result calculated by the accelerator; the controller is used for controlling starting, calculation and time measurement of the accelerator; the AX14-Lite bus module is used for transmitting an instruction; the AX14 bus module is used for data transmission; the processor is used for realizing acceleration. According to the invention, redundant data generated by the Alexnet accelerator during operation is reduced, the contentof convolution calculation and memory access operation is reduced, and the consumption of computer hardware resources is also reduced.

Description

technical field [0001] The invention relates to the field of neural network accelerators, in particular to an FPGA-based Alexnet forward network accelerator. Background technique [0002] The Alexnet network model was proposed by Alex and won the first prize in the ImageNet2012 image classification competition. It is a classic model of CNN in image classification applications. The first five layers of Alexnet are convolutional layers, and the last three layers are fully connected layers. The model receives 3 images with dimensions (224, 224) as input, and after a series of convolution calculations and pooling operations, it finally outputs a 1000-dimensional vector, representing a classifier containing 1000 classification labels; volume The accumulation layer is composed of a certain number of feature maps, and its function is to extract the features of the input image. There are a large number of convolutional calculations in the convolutional layer, accounting for more t...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the invention of the present invention is: for the above-mentioned existing problems, provide a kind of Alexnet forward network accelerator based on FPGA; The present inventio

Images