Unlock instant, AI-driven research and patent intelligence for your innovation.

A multi-bank row and column interleaved read and write method for convolutional neural network data storage

A convolutional neural network and data storage technology, applied in the field of data storage mechanism, can solve problems such as data organization and arrangement of intermediate results that are difficult to achieve calculations, and achieve the effects of data arrangement and organization rules, efficient and effective storage, and reduced control

Active Publication Date: 2021-08-13
XI AN JIAOTONG UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, the intermediate result data cache of a single bank used by Eyeriss and DNA is difficult to achieve the purpose of regular organization and arrangement of the intermediate result data of the operation

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A multi-bank row and column interleaved read and write method for convolutional neural network data storage
  • A multi-bank row and column interleaved read and write method for convolutional neural network data storage
  • A multi-bank row and column interleaved read and write method for convolutional neural network data storage

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0027] The present invention will be further described in detail below in conjunction with specific embodiments, which are explanations of the present invention rather than limitations.

[0028] The CNN algorithm involves a large number of calculations. The optimization and acceleration of the CNN algorithm is realized on the chip. By increasing the operation array to improve the parallelism of the operation, the CNN algorithm can be executed in parallel. Taking AlexNet as an example, its architecture is as follows figure 1 As shown, when the convolutional neural network is operating, the layers are connected, and the output of the previous convolutional layer is the input data of the next convolutional layer. However, hardware resources are limited for on-chip systems. The CNN algorithm cannot be fully mapped to the hardware architecture. The output results generated by the previous layer need to be stored, and wait for the hardware resources to be free before serving as the ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The present invention is a multi-bank row-column interleaving reading and writing method for convolutional neural network data storage, and a multi-bank storage structure is designed, and the multi-bank memory adjusts the data storage path according to different operation modes, and each array subset is connected with the convolution kernel One-to-one correspondence, the intermediate result data of different channels generated by each array subset are directly divided into banks and written into the multi-bank memory. In the process of reading the intermediate result data, the method of reading out bank by bank is adopted. The result data are all intermediate result data belonging to the same channel, and can be called directly when the next data call is made. The invention improves the efficiency of data scheduling.

Description

technical field [0001] The invention relates to a data storage mechanism for accelerating a convolutional neural network based on a dynamic reconfigurable array, in particular to a multi-bank row-column interleaved read-write method for convolutional neural network data storage. Background technique [0002] Artificial intelligence is one of the most popular computer sciences at present. As the main way to realize artificial intelligence, deep learning has also achieved profound development. Convolution Neural Network (CNN) is one of the most studied and widely used network structures of artificial neural network structures. It has become one of the research hotspots in many scientific fields, especially in the field of pattern classification. Since CNN avoids the The complex preprocessing of the image can directly input the original image, so it has been more widely used. In recent years, convolutional neural networks have made great achievements in the field of computer v...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06N3/04G06F12/0802
CPCG06F12/0802G06N3/045
Inventor 杨晨张海波王逸洲王小力耿莉
Owner XI AN JIAOTONG UNIV