FPGA-based neural network acceleration method and accelerator

Abstract

The invention belongs to the technical field of neural networks, and provides an FPGA-based convolutional neural network accelerator, which considers the requirements of the neural network acceleratoron performance and universality and has a wide application scene. Therefore, according to the accelerator, the neural network accelerator based on the FPGA comprises a convolution operation module, apooling module, a direct memory access DMA module, an instruction control module, an address control module, an internal random access memory RAM module and an instruction RAM module. Wherein the convolution operation module is used for carrying out operation on a convolution layer in the convolutional neural network. The method is mainly applied to design and manufacturing of neural network chips.

Description

technical field [0001] The invention belongs to the technical field of neural networks, in particular to a neural network accelerator architecture based on a field programmable gate array (FPGA) Background technique [0002] With the rapid development of artificial intelligence, Convolutional Neutral Network (CNN) has attracted more and more attention, and it plays an important role in many fields such as image processing. Compared with traditional algorithms, the computational complexity of CNN is much higher, and the general-purpose CPU can no longer meet the computational requirements. At present, the main solution is to use GPU for CNN calculation. Although GPU has natural advantages in parallel computing, it is There are great deficiencies in terms of cost and power consumption, which cannot meet the acceleration of CNN in some low-specific scenarios. FPGA has powerful parallel processing capability, flexible configurable features and ultra-low power consumption, which...

AI Technical Summary

Problems solved by technology

Compared with traditional algorithms, the computational complexity of CNN is much higher, and the general-purpose CPU can no longer meet the computational requirements. At present, the main solution is to use GPU for CNN calculation. Although GPU has natural advantages in parallel computing, it is There are great deficiencies in cost and power consumption, which cannot satisfy CNN acceleration in some low-specific scenarios

Images