Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

A computing device and method

A computing device and computing unit technology, applied in the field of artificial intelligence, can solve the problems such as the inability to further improve the computing speed, the flexibility is not high, the calculation amount is reduced, and the effect of reducing power consumption, high flexibility, and reducing energy consumption is achieved.

Active Publication Date: 2021-06-29
SHANGHAI CAMBRICON INFORMATION TECH CO LTD
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, this structure is not flexible enough to further increase the computing speed and reduce the amount of calculation

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 computing device and method
  • A computing device and method
  • A computing device and method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0080] Please refer to the attached Figure 4 As shown, in Embodiment 1, the arithmetic unit of the arithmetic module includes n (n is a positive integer) multipliers and an addition tree with n inputs. The operation control unit controls the operation module as a whole, and can send control signals to each multiplier. In addition to receiving data signals, each adder and multiplier can also receive control signals, that is, in addition to receiving data to be calculated, it can also receive control signals and send control signals to the next level.

[0081] The specific operation process of the low-power neural network operation device using the operation module of the first embodiment is as follows: first, the storage control unit controls the storage module to read out neuron data and synapse data. If the synaptic data is stored in a sparse coding method, the neuron data and the synaptic index value need to be transmitted to the data processing module, and the neuron data...

Embodiment 2

[0084] Please refer to the attached Figure 5 As shown, in Embodiment 2, the arithmetic module includes n (n is a positive integer) arithmetic units, and each arithmetic unit includes a multiplier, an adder, and a temporary cache, and each arithmetic control unit is also split into for each operation group. The synaptic data is transmitted to each computing unit in the form of broadcast, and the neurons are directly transmitted from the storage part.

[0085] The specific calculation process of the low-power neural network computing device using the computing module of Embodiment 2 is as follows: First, the temporary cache in each computing unit is initially set to 0. The storage control unit controls the storage module to read out the neuron data. If the synapse data is stored in a sparse coding manner, the index value of the synapse and the neuron data are transmitted to the data processing module, and the neuron data is stored according to the index. The value is compress...

Embodiment 3

[0087] Please refer to the attached Figure 6 As shown, in Embodiment 3, the computing module includes n computing units. Each arithmetic unit includes a multiplier and an adder and a selector. The operation control unit controls each operation unit separately, and the partial sums are transferred between the operation units.

[0088] The specific operation process of the low-power neural network computing device using the computing module of Embodiment 3 is as follows: First, the storage control unit reads the synaptic value from the storage module and sends it to the data processing module. The data processing module compresses the synapses by setting the compression threshold, that is, only selects the synaptic values ​​whose absolute value is not less than the given compression threshold, transmits them to each computing unit respectively, and keeps the synaptic values ​​unchanged. Then read the neuron value from the storage module, after the corresponding compression by...

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 disclosure provides a computing device, including: a computing module including one or more computing units; and a control module including a computing control unit, configured to control the closing of the computing unit through a judging condition. The present disclosure also provides an operation method. The disclosed low-power computing device and method have high flexibility and can be combined with software upgrading methods to further increase the computing speed, reduce the amount of computing, and reduce the computing power consumption of the accelerator.

Description

technical field [0001] The present disclosure relates to the technical field of artificial intelligence, and in particular to a computing device and method. Background technique [0002] Deep neural network is the basis of many artificial intelligence applications at present. It has achieved breakthrough applications in many aspects such as speech recognition, image processing, data analysis, advertising recommendation system, and automobile automatic driving, making deep neural network applied in daily life. every aspect. However, the huge amount of calculation of deep neural network has been restricting its faster development and wider application. When considering the use of accelerator design to accelerate the operation of deep neural networks, the huge amount of calculation will inevitably bring a lot of energy consumption, which also restricts the further wide application of accelerators. [0003] In terms of hardware, the existing commonly used accelerator architect...

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): G06N3/063G06N3/04G06F7/575
CPCG06N3/063
Inventor 不公告发明人
Owner SHANGHAI CAMBRICON INFORMATION TECH CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com