Device used for realizing a neural network processor with variable calculation precision

A neural network and processor technology, applied in the field of dedicated circuit computing devices, can solve the problem of no significant improvement in energy consumption and computing speed, and achieve the effects of reducing on-chip storage, reducing energy consumption, and ensuring computing accuracy

Active Publication Date: 2019-02-12
INST OF COMPUTING TECH CHINESE ACAD OF SCI
View PDF7 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, the programs of this type of technology still run on general-purpose processors, which can reduce the required storage capacity, but there is no significant improvement in energy consumption and computing speed

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
  • Device used for realizing a neural network processor with variable calculation precision
  • Device used for realizing a neural network processor with variable calculation precision
  • Device used for realizing a neural network processor with variable calculation precision

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0048] After research, the inventor found that the accuracy requirements of each network layer of the neural network are not the same. If the accuracy of the original data is used for storage and calculation without distinction for each network layer, it is a waste of computing resources and energy consumption. . Even if the existing technology quantifies the accuracy required by each network layer through software as mentioned in the background art, if the hardware structure of the processor is not changed, the hardware energy consumption and computing speed cannot be significantly improved. Therefore, the inventor believes that in order to solve the problems in the prior art, it is necessary to provide a dedicated neural network processor with variable calculation precision and provide a corresponding hardware structure to realize the adjustment of the precision of the neural network data.

[0049] Based on the particularity of the neural network algorithm, it has a certain ...

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 invention provides a data compression device for a neural network processor, a supporting data decompression device, and a neural network processor with variable precision. The data compression apparatus includes a rounding unit, a first comparator, a second comparator, and a first multiplexer for converting m bits of data to be compressed into n bits of compression results O, n (m. Through the data compression device, the neural network processor can adopt different data precision to participate in calculation at different layers of the neural network, and on the premise of ensuring the calculation precision, the on-chip storage amount is reduced, the data transmission energy loss is reduced, and the device is particularly suitable for the applications with limited on-chip resources.

Description

technical field [0001] The invention relates to a special-purpose circuit operation device, in particular to a neural network special-purpose processor with variable calculation precision. Background technique [0002] Neural network is one of the perception models with the highest level of development in the field of artificial intelligence. It simulates the neural connection structure of the human brain by establishing a model, and describes the data characteristics hierarchically through multiple transformation stages, providing image, video, audio, etc. The application of big data processing tasks has brought breakthrough progress. Since the neural network model is sufficiently robust to environmental noise and signal integrity issues, partial loss of calculation data will not have a catastrophic effect on the calculation results. [0003] Based on the above reasons, a dedicated processor for performing neural network operations, that is, a neural network processor, can...

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 Applications(China)
IPC IPC(8): G06N3/063
CPCG06N3/063
Inventor 韩银和许浩博王颖
Owner INST OF COMPUTING TECH CHINESE ACAD OF SCI
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap