A FPGA heterogeneous acceleration method, device and system

Abstract

The invention discloses an FPGA heterogeneous acceleration method, device and system; in this solution, interfaces facing different machine learning platforms are provided on the host end, thereby obtaining network description files and data of different networks, and through the network description files Different execution units in the underlying library of the FPGA can be configured to realize data calculation; through this acceleration method, the FPGA heterogeneous acceleration system can be used for new networks without secondary development, avoiding heterogeneous FPGA Accelerate system and source code recompilation, thus saving human resources and time resources.

Description

technical field [0001] The present invention relates to the technical field of heterogeneous acceleration, and more specifically, to an FPGA heterogeneous acceleration method, device and system. Background technique [0002] Convolutional Neural Network (CNN) is an efficient recognition method that has been developed in recent years and has attracted widespread attention. Hubel and Wiesel found that its unique network structure can effectively reduce the complexity of the feedback neural network when studying the neurons used for local sensitivity and direction selection in the cat's cerebral cortex, and then proposed a convolutional neural network, which includes a convolutional layer , activation layer, pooling layer, fully connected layer. Now, CNN has become one of the research hotspots in many scientific fields, especially in the field of pattern classification, because the network avoids the complex preprocessing of images and can directly input original images, so it...

AI Technical Summary

Problems solved by technology

[0003] At present, although the GPU is widely used for heterogeneous acceleration of CNN, the power consumption of the GPU is relatively large.

More and more studies have found that FPGA can play a huge role in the field of high-performance computing. The speed of processing convolution problems is completely comparable to that of GPU. Although there are already FPGAs used for CNN acceleration, most of them are for a specific Accelerate the network, such as alexnet, resnet50, resnet101, etc., if you change to a new network, you need to make major changes, secondary development and debugging, and a lot of manpower and time resources will be wasted

Images