CN110147880A - A kind of Neural Network Data processing structure, method, system and relevant apparatus - Google Patents
A kind of Neural Network Data processing structure, method, system and relevant apparatus Download PDFInfo
- Publication number
- CN110147880A CN110147880A CN201910429293.8A CN201910429293A CN110147880A CN 110147880 A CN110147880 A CN 110147880A CN 201910429293 A CN201910429293 A CN 201910429293A CN 110147880 A CN110147880 A CN 110147880A
- Authority
- CN
- China
- Prior art keywords
- neural network
- network data
- array
- data
- data processing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 77
- 238000012545 processing Methods 0.000 title claims abstract description 34
- 238000000034 method Methods 0.000 title claims abstract description 22
- 238000003860 storage Methods 0.000 claims abstract description 55
- 238000004364 calculation method Methods 0.000 claims abstract description 19
- 241001269238 Data Species 0.000 claims abstract description 6
- 238000003672 processing method Methods 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 7
- 238000004422 calculation algorithm Methods 0.000 abstract description 6
- 238000013527 convolutional neural network Methods 0.000 description 13
- 238000004891 communication Methods 0.000 description 8
- 230000005236 sound signal Effects 0.000 description 4
- 210000004218 nerve net Anatomy 0.000 description 3
- 230000006870 function Effects 0.000 description 2
- 230000014759 maintenance of location Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- KLDZYURQCUYZBL-UHFFFAOYSA-N 2-[3-[(2-hydroxyphenyl)methylideneamino]propyliminomethyl]phenol Chemical compound OC1=CC=CC=C1C=NCCCN=CC1=CC=CC=C1O KLDZYURQCUYZBL-UHFFFAOYSA-N 0.000 description 1
- 239000011248 coating agent Substances 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 201000001098 delayed sleep phase syndrome Diseases 0.000 description 1
- 208000033921 delayed sleep phase type circadian rhythm sleep disease Diseases 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000013486 operation strategy Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000000547 structure data Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Neurology (AREA)
- Image Analysis (AREA)
Abstract
This application discloses a kind of Neural Network Data processing structure, method, system and a kind of electronic equipment and computer readable storage mediums, the structure includes: and to pass through the three-dimensional memory array of multiple output channel parallel output target datas for being pre-stored the Neural Network Data；It is connected with the three-dimensional memory array, for carrying out the computing array of convolutional calculation to the target data.Neural Network Data processing structure provided by the present application, three-dimensional memory array therein is pre-stored Neural Network Data, when computing array needs to carry out convolution algorithm to Neural Network Data, multiple output channel parallel output datas can be passed through, a large amount of calculating data can be provided in a short time for computing array, the degree of parallelism for improving Neural Network Data processing, accelerates the speed of convolutional calculation.
Description
Technical field
This application involves nerual network technique fields, more specifically to a kind of Neural Network Data processing structure, side
Method, system and a kind of electronic equipment and a kind of computer readable storage medium.
Background technique
Research in terms of deep learning at present is mainly using CNN as research object.And the difference due to handling scene, to CNN
Performance requirement it is also not identical, to develop multiple network structure.But CNN it is basic composition be it is fixed, it is respectively defeated
Enter layer, convolutional layer, active coating, pond layer and full articulamentum.Wherein calculation amount the best part is convolutional layer, main function
Exactly complete the convolution algorithm between image (feature) and neuron (filter).
For different CNN neural network structures, the data length of processing is different.For the same CNN
Neural network, data length handled by every layer are also in variation.In CNN neural network, input data amount=input figure
Image width degree × input picture height × input picture number of active lanes.Output data quantity=output picture traverse × output picture altitude
× output image channel number.Convolutional calculation total degree=output picture traverse × output picture altitude × input picture port number
Mesh × output image channel number.As it can be seen that for common CNN network structure, the data volume of input and output is very big.Example
It can achieve 512 input channels such as one layer of ResNet50,512 output channels, multiplied by the size of image, data
Byte quantity can achieve million grades.The rate of convolutional calculation is to measure the important indicator of CNN network performance, and this requires short
A large amount of calculating data can be provided in the time for computing array.
Therefore, how can provide a large amount of data that calculate for computing array in a short time is that those skilled in the art need
Technical problems to be solved.
Summary of the invention
The application's is designed to provide a kind of Neural Network Data processing structure, method, system and a kind of electronic equipment
With a kind of computer readable storage medium, a large amount of calculating data can be provided for computing array in a short time.
To achieve the above object, this application provides a kind of Neural Network Data processing structures, comprising:
It is deposited for being pre-stored the Neural Network Data, and by the three-dimensional of multiple output channel parallel output target datas
Store up array；
It is connected with the three-dimensional memory array, for carrying out the computing array of convolutional calculation to the target data.
Wherein, every row of the three-dimensional memory array corresponds to that row write enters to enable and row reads enabled, the equal respective column of each column
It is enabled that enabled and column reading is written, the every row row's of correspondence write-in enables and row's reading is enabled.
Wherein, the three-dimensional memory array is specially the three-dimensional memory array of ping-pong structure.
To achieve the above object, this application provides a kind of Neural Network Data processing methods, comprising:
It determines pending data, and determines that the pending data corresponding all storages in three-dimensional memory array are single
Member；
By the data in each storage unit by multiple output channel parallel outputs to computing array, so as to described
Computing array carries out convolutional calculation to the pending data.
Wherein, the data by each storage unit are by multiple output channel parallel outputs to calculating battle array
Column, comprising:
It controls each storage unit row of the row and reads the row that enabled, column column read enabled and place row
It reads and enables the data in each storage unit passing through multiple output channel parallel outputs to computing array.
Wherein, the data by each storage unit are by multiple output channel parallel outputs to calculating battle array
Column, comprising:
It determines configuration parameter, and the data in each storage unit is passed through by multiple outputs according to the configuration parameter
Channel parallel is exported to computing array；Wherein, the configuration parameter includes at least the number of the output channel and each described
Relationship between output channel.
Wherein, further includes:
The input width of three-dimensional memory array is determined according to the data-interface width of external memory space；
Neural Network Data in the external memory space is stored according to the input width to the three-dimensional storage
In array.
To achieve the above object, this application provides a kind of Neural Network Data processing systems, comprising:
Determining module for determining pending data, and determines that the pending data is corresponding in three-dimensional memory array
All storage units；
Output module, for the data in each storage unit to be passed through multiple output channel parallel outputs to calculating
Array, so that the computing array carries out convolutional calculation to the pending data.
To achieve the above object, this application provides a kind of electronic equipment, comprising:
Memory, for storing computer program；
Processor is realized when for executing the computer program such as the step of above-mentioned Neural Network Data processing method.
To achieve the above object, this application provides a kind of computer readable storage medium, the computer-readable storages
It is stored with computer program on medium, realizes that above-mentioned Neural Network Data such as is handled when the computer program is executed by processor
The step of method.
By above scheme it is found that a kind of Neural Network Data processing structure provided by the present application, comprising: for being pre-stored
The Neural Network Data, and pass through the three-dimensional memory array of multiple output channel parallel output target datas；With the three-dimensional
Storage array is connected, for carrying out the computing array of convolutional calculation to the target data.
Neural Network Data processing structure provided by the present application, three-dimensional memory array therein are pre-stored neural network number
According to, when computing array need to Neural Network Data carry out convolution algorithm when, multiple output channel parallel output numbers can be passed through
According to, a large amount of calculating data can be provided in a short time for computing array, improve the degree of parallelism of Neural Network Data processing,
Accelerate the speed of convolutional calculation.Disclosed herein as well is a kind of Neural Network Data processing system and a kind of electronic equipment and one
Kind computer readable storage medium, is equally able to achieve above-mentioned technical effect.
It should be understood that the above general description and the following detailed description are merely exemplary, this can not be limited
Application.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of application for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.Attached drawing is and to constitute specification for providing further understanding of the disclosure
A part, be used to explain the disclosure together with following specific embodiment, but do not constitute the limitation to the disclosure.Attached
In figure:
Fig. 1 is a kind of structure chart of Neural Network Data processing structure shown according to an exemplary embodiment；
Fig. 2 is the structure chart of another Neural Network Data processing structure shown according to an exemplary embodiment；
Fig. 3 is a kind of flow chart of Neural Network Data processing method shown according to an exemplary embodiment；
Fig. 4 is the flow chart of another Neural Network Data processing method shown according to an exemplary embodiment；
Fig. 5 is a kind of structure chart of Neural Network Data processing system shown according to an exemplary embodiment；
Fig. 6 is the structure chart according to a kind of electronic equipment shown in an exemplary embodiment.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of embodiments of the present application, instead of all the embodiments.It is based on
Embodiment in the application, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall in the protection scope of this application.
The embodiment of the present application discloses a kind of Neural Network Data processing structure, can mention in a short time for computing array
For largely calculating data.
Referring to Fig. 1, a kind of structure chart of Neural Network Data processing structure shown according to an exemplary embodiment is such as schemed
Shown in 1, comprising:
It is deposited for being pre-stored the Neural Network Data, and by the three-dimensional of multiple output channel parallel output target datas
Store up array 100；
It is connected with the three-dimensional memory array 100, for carrying out the computing array of convolutional calculation to the target data
200。
In the present embodiment, three-dimensional memory array 100 includes multiple storage units, for being pre-stored Neural Network Data,
The width of input data can be according to such as DDR (Chinese name: Double Data Rate, full name in English: Double Data
Rate) the data-interface width of the external memory spaces such as random access memory is controlled by configuration parameter.Three-dimensional storage
Array 100 can read the data in any storage unit under the control of configuration parameter, without considering the mode of data storage
And the bit wide of data port.The data of output can according to need any combination, extremely by multiple output channel parallel outputs
Computing array 200, the number of active lanes of output, data length, the correlation between beat and each output channel can pass through
Configuration parameter is controlled.Under different configuration parameter control, the data way of output of multiple combinations is realized.Be conducive to CNN
(Chinese name: convolutional neural networks, full name in English: Convolutional Neural Networks) dynamic adjustment structure, expands
The concrete function for having opened up CNN enriches the implementation of CNN.
It should be noted that three-dimensional memory array 100 can run on FPGA (Chinese name: field programmable gate
Array, full name in English: Field Programmable Gate Array) accelerate on board, due to the standard component inside FPGA
Speed and hardware resource cost it is controllable, the Neural Network Data of needs can be provided for computing array, further increased
The computational efficiency of computing array.
Preferably, every row of the three-dimensional memory array corresponds to that row write enters to enable and row reads enabled, and each column is corresponding
Column write-in is enabled and column read it is enabled, the corresponding row's write-in of every row is enabled and row read it is enabled.In specific implementation, when some is deposited
When the row that storage unit row reading of the row is enabled, column column reading is enabled and place is arranged reads enabled be all turned on, from
Data are read in the storage unit, pass through multiple output channel parallel outputs to computing array.Equally when some storage unit institute
When the row write being expert at enters that enabled, column column write-in is enabled and row's write-in of place row is enabled is all turned on, to the storage list
Member write-in data.
On the basis of above-mentioned three-dimensional memory array, in order to further improve the handling capacity of data, as shown in Fig. 2, can
To take ping-pong operation strategy, i.e. the three-dimensional memory array three-dimensional memory array that is specially ping-pong structure.For ping-pong structure
Data input sequence can also be controlled according to configuration parameter.
Neural Network Data processing structure provided by the embodiments of the present application, three-dimensional memory array therein are pre-stored nerve net
Network data can be defeated parallel by multiple output channels when computing array needs to carry out convolution algorithm to Neural Network Data
Data out, can be provided in a short time for computing array it is a large amount of calculate data, improve Neural Network Data processing and
Row degree accelerates the speed of convolutional calculation.
The embodiment of the present application discloses a kind of the embodiment of the present application and discloses a kind of Neural Network Data processing method, specifically
:
Referring to Fig. 3, a kind of flow chart of Neural Network Data processing method shown according to an exemplary embodiment is such as schemed
Shown in 3, comprising:
S101: determining pending data, and determines that the pending data is corresponding in three-dimensional memory array and all deposit
Storage unit；
The executing subject of the present embodiment can for Neural Network Data processing processor, purpose be in a short time be meter
It calculates array and a large amount of Neural Network Data is provided.In this step, it is first determined the Neural Network Data that computing array needs is (i.e.
Pending data) corresponding all storage units in three-dimensional memory array, i.e., these storage units are in three-dimensional memory array
Coordinate position.
S102: by the data in each storage unit by multiple output channel parallel outputs to computing array, with
Toilet states computing array and carries out convolutional calculation to the pending data.
In this step, the coordinate position of the storage unit determined according to previous step, by the data in each storage unit
Pass through multiple output channel parallel outputs to computing array.The number of active lanes of output, data length, beat and each output are logical
Correlation between road can be controlled by configuration parameter.That is this step can include determining that configuration parameter, and according to
Data in each storage unit are passed through multiple output channel parallel outputs to computing array by the configuration parameter；Its
In, the configuration parameter includes at least relationship between the number and each output channel of the output channel.
In specific implementation, enabled, column column reading is read by controlling each storage unit row of the row
It takes the row of enabled and place row to read to enable the data in each storage unit passing through multiple output channel parallel outputs
To computing array.
Neural Network Data processing method provided by the embodiments of the present application, three-dimensional memory array therein are pre-stored nerve net
Network data can be defeated parallel by multiple output channels when computing array needs to carry out convolution algorithm to Neural Network Data
Data out, can be provided in a short time for computing array it is a large amount of calculate data, improve Neural Network Data processing and
Row degree accelerates the speed of convolutional calculation.
The process of Neural Network Data write-in three-dimensional memory array is introduced below, specific:
Referring to fig. 4, the flow chart of a kind of Neural Network Data processing method shown according to an exemplary embodiment is such as schemed
Shown in 4, comprising:
S201: the input width of three-dimensional memory array is determined according to the data-interface width of external memory space；
In this step, the input width of data can be according to the data-interface width of external memory space by configuration parameter
To be controlled.
S202: the Neural Network Data in the external memory space is stored according to the input width to the three-dimensional
In storage array.
In this step, three-dimensional memory array is written into according to the input width that previous step determines in Neural Network Data
In.It can be written herein according to the sequence of storage unit in three-dimensional memory array, or Neural Network Data is specified
Specific storage unit, herein without specifically limiting.
A kind of Neural Network Data processing system provided by the embodiments of the present application is introduced below, described below one
Kind of Neural Network Data processing system can be cross-referenced with a kind of above-described Neural Network Data processing method.
Referring to Fig. 5, a kind of structure chart of Neural Network Data processing system shown according to an exemplary embodiment is such as schemed
Shown in 5, comprising:
Determining module 501 for determining pending data, and determines that the pending data is right in three-dimensional memory array
All storage units answered；
Output module 502, for by the data in each storage unit by multiple output channel parallel outputs extremely
Computing array, so that the computing array carries out convolutional calculation to the pending data.
Neural Network Data processing system provided by the embodiments of the present application, three-dimensional memory array therein are pre-stored nerve net
Network data can be defeated parallel by multiple output channels when computing array needs to carry out convolution algorithm to Neural Network Data
Data out, can be provided in a short time for computing array it is a large amount of calculate data, improve Neural Network Data processing and
Row degree accelerates the speed of convolutional calculation.
On the basis of the above embodiments, the output module 502 is specially to control often as a preferred implementation manner,
Row's reading that a storage unit row reading of the row enables, the column reading of column is enabled and place is arranged is enabled will be each
The module that data in the storage unit pass through multiple output channel parallel outputs to computing array.
On the basis of the above embodiments, the output module 502 is specially that determination is matched as a preferred implementation manner,
Set parameter, and according to the configuration parameter by the data in each storage unit by multiple output channel parallel outputs extremely
The module of computing array；Wherein, the configuration parameter includes at least the number and each output channel of the output channel
Between relationship.
On the basis of the above embodiments, as a preferred implementation manner, further include:
Input width module is determined, for determining three-dimensional memory array according to the data-interface width of external memory space
Input width；
Memory module, for by the Neural Network Data in the external memory space according to the input width store to
In the three-dimensional memory array.
About the system in above-described embodiment, wherein modules execute the concrete mode of operation in related this method
Embodiment in be described in detail, no detailed explanation will be given here.
Present invention also provides a kind of electronic equipment, referring to Fig. 6, a kind of electronic equipment 600 provided by the embodiments of the present application
Structure chart, as shown in fig. 6, may include processor 11 and memory 12.The electronic equipment 600 can also include multimedia group
Part 13, one or more of input/output (I/O) interface 14 and communication component 15.
Wherein, processor 11 is used to control the integrated operation of the electronic equipment 600, to complete above-mentioned Neural Network Data
All or part of the steps in processing method.Memory 12 is for storing various types of data to support in the electronic equipment
600 operation, these data for example may include any application or method for operating on the electronic equipment 600
Instruction and the relevant data of application program, such as contact data, the message of transmitting-receiving, picture, audio, video etc..This is deposited
Reservoir 12 can be realized by any kind of volatibility or non-volatile memory device or their combination, such as static random
It accesses memory (Static Random Access Memory, abbreviation SRAM), electrically erasable programmable read-only memory
(Electrically Erasable Programmable Read-Only Memory, abbreviation EEPROM), erasable programmable
Read-only memory (Erasable Programmable Read-Only Memory, abbreviation EPROM), programmable read only memory
(Programmable Read-Only Memory, abbreviation PROM), and read-only memory (Read-Only Memory, referred to as
ROM), magnetic memory, flash memory, disk or CD.Multimedia component 13 may include screen and audio component.Wherein shield
Curtain for example can be touch screen, and audio component is used for output and/or input audio signal.For example, audio component may include one
A microphone, microphone is for receiving external audio signal.The received audio signal can be further stored in memory
It 12 or is sent by communication component 15.Audio component further includes at least one loudspeaker, is used for output audio signal.I/O interface
14 provide interface between processor 11 and other interface modules, other above-mentioned interface modules can be keyboard, mouse, button
Deng.These buttons can be virtual push button or entity button.Communication component 15 for the electronic equipment 600 and other equipment it
Between carry out wired or wireless communication.Wireless communication, such as Wi-Fi, bluetooth, near-field communication (Near Field
Communication, abbreviation NFC), 2G, 3G or 4G or they one or more of combination, therefore corresponding communication
Component 15 may include: Wi-Fi module, bluetooth module, NFC module.
In one exemplary embodiment, electronic equipment 600 can be by one or more application specific integrated circuit
(Application Specific Integrated Circuit, abbreviation ASIC), digital signal processor (Digital
Signal Processor, abbreviation DSP), digital signal processing appts (Digital Signal Processing Device,
Abbreviation DSPD), programmable logic device (Programmable Logic Device, abbreviation PLD), field programmable gate array
(Field Programmable Gate Array, abbreviation FPGA), controller, microcontroller, microprocessor or other electronics member
Part is realized, for executing above-mentioned Neural Network Data processing method.
In a further exemplary embodiment, a kind of computer readable storage medium including program instruction is additionally provided, it should
The step of above-mentioned Neural Network Data processing method is realized when program instruction is executed by processor.For example, this computer-readable is deposited
Storage media can be the above-mentioned memory 12 including program instruction, and above procedure instruction can be by the processor 11 of electronic equipment 600
It executes to complete above-mentioned Neural Network Data processing method.
Each embodiment is described in a progressive manner in specification, the highlights of each of the examples are with other realities
The difference of example is applied, the same or similar parts in each embodiment may refer to each other.For system disclosed in embodiment
Speech, since it is corresponded to the methods disclosed in the examples, so being described relatively simple, related place is referring to method part illustration
?.It should be pointed out that for those skilled in the art, under the premise of not departing from the application principle, also
Can to the application, some improvement and modification can also be carried out, these improvement and modification also fall into the protection scope of the claim of this application
It is interior.
It should also be noted that, in the present specification, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning
Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that
A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or
The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged
Except there is also other identical elements in the process, method, article or apparatus that includes the element.
Claims (10)
1. a kind of Neural Network Data processing structure characterized by comprising
For being pre-stored the Neural Network Data, and the three-dimensional storage battle array for passing through multiple output channel parallel output target datas
Column；
It is connected with the three-dimensional memory array, for carrying out the computing array of convolutional calculation to the target data.
2. Neural Network Data processing structure according to claim 1, which is characterized in that every row of the three-dimensional memory array
It corresponds to row write to enter to enable and go to read to enable, the equal respective column write-in of each column, which enables and arrange to read, to be enabled, and every row corresponding arrange is write
It is enabled to enter to enable and arrange reading.
3. Neural Network Data processing structure according to claim 1 or claim 2, which is characterized in that the three-dimensional memory array tool
Body is the three-dimensional memory array of ping-pong structure.
4. a kind of Neural Network Data processing method characterized by comprising
It determines pending data, and determines pending data corresponding all storage units in three-dimensional memory array；
By the data in each storage unit by multiple output channel parallel outputs to computing array, so as to the calculating
Array carries out convolutional calculation to the pending data.
5. Neural Network Data processing method according to claim 4, which is characterized in that described by each storage unit
In data pass through multiple output channel parallel outputs to computing array, comprising:
The row that the row reading of the row of each storage unit is enabled, column column reading is enabled and place is arranged is controlled to read
It is enabled that data in each storage unit are passed through into multiple output channel parallel outputs to computing array.
6. Neural Network Data processing method according to claim 4, which is characterized in that described by each storage unit
In data pass through multiple output channel parallel outputs to computing array, comprising:
It determines configuration parameter, and the data in each storage unit is passed through by multiple output channels according to the configuration parameter
Parallel output is to computing array；Wherein, the configuration parameter includes at least the number and each output of the output channel
Relationship between channel.
7. the Neural Network Data processing method according to any one of claim 4 to 6, which is characterized in that further include:
The input width of three-dimensional memory array is determined according to the data-interface width of external memory space；
Neural Network Data in the external memory space is stored according to the input width to the three-dimensional memory array
In.
8. a kind of Neural Network Data processing system characterized by comprising
Determining module for determining pending data, and determines pending data corresponding institute in three-dimensional memory array
There is storage unit；
Output module, for by the data in each storage unit by multiple output channel parallel outputs to calculating battle array
Column, so that the computing array carries out convolutional calculation to the pending data.
9. a kind of electronic equipment characterized by comprising
Memory, for storing computer program；
Processor is realized as described in any one of claim 4 to 7 when for executing the computer program at Neural Network Data
The step of reason method.
10. a kind of computer readable storage medium, which is characterized in that be stored with computer on the computer readable storage medium
Program realizes that Neural Network Data is handled as described in any one of claim 4 to 7 when the computer program is executed by processor
The step of method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910429293.8A CN110147880A (en) | 2019-05-22 | 2019-05-22 | A kind of Neural Network Data processing structure, method, system and relevant apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910429293.8A CN110147880A (en) | 2019-05-22 | 2019-05-22 | A kind of Neural Network Data processing structure, method, system and relevant apparatus |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110147880A true CN110147880A (en) | 2019-08-20 |
Family
ID=67592728
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910429293.8A Pending CN110147880A (en) | 2019-05-22 | 2019-05-22 | A kind of Neural Network Data processing structure, method, system and relevant apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110147880A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111091188A (en) * | 2019-12-16 | 2020-05-01 | 腾讯科技（深圳）有限公司 | Forward computing method and device for neural network and computer readable storage medium |
| CN111506518A (en) * | 2020-04-13 | 2020-08-07 | 湘潭大学 | Data storage control method and device |
| CN111857999A (en) * | 2020-07-10 | 2020-10-30 | 苏州浪潮智能科技有限公司 | Data scheduling method, device and equipment and computer readable storage medium |
| CN112016522A (en) * | 2020-09-25 | 2020-12-01 | 苏州浪潮智能科技有限公司 | Video data processing method, system and related components |
| CN112395247A (en) * | 2020-11-18 | 2021-02-23 | 北京灵汐科技有限公司 | Data processing method and storage and calculation integrated chip |
| CN112596684A (en) * | 2021-03-08 | 2021-04-02 | 成都启英泰伦科技有限公司 | Data storage method for voice deep neural network operation |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102339846A (en) * | 2010-07-19 | 2012-02-01 | 旺宏电子股份有限公司 | Semiconductor memory element possessing transistor with adjustable grid resistance value |
| CN103533378A (en) * | 2013-10-09 | 2014-01-22 | 天津大学 | Three-dimensional integer DCT (Discrete Cosine Transform) transformation system on basis of FPGA (Field Programmable Gate Array) and transformation method thereof |
| CN103578552A (en) * | 2012-08-10 | 2014-02-12 | 三星电子株式会社 | Nonvolatile memory device and operating method with variable memory cell state definitions |
| CN103681679A (en) * | 2012-08-30 | 2014-03-26 | 成都海存艾匹科技有限公司 | Three-dimensional offset-printed memory |
| US20160092122A1 (en) * | 2014-09-30 | 2016-03-31 | Sandisk Technologies Inc. | Method and apparatus for wear-levelling non-volatile memory |
| CN106898371A (en) * | 2017-02-24 | 2017-06-27 | 中国科学院上海微系统与信息技术研究所 | Three-dimensional storage reading circuit and its wordline and bit-line voltage collocation method |
| CN107220704A (en) * | 2016-03-21 | 2017-09-29 | 杭州海存信息技术有限公司 | Integrated neural network processor containing three-dimensional memory array |
| CN107316014A (en) * | 2016-03-07 | 2017-11-03 | 杭州海存信息技术有限公司 | Have the memory of image identification function concurrently |
| CN108596331A (en) * | 2018-04-16 | 2018-09-28 | 浙江大学 | A kind of optimization method of cell neural network hardware structure |
| CN109446996A (en) * | 2018-10-31 | 2019-03-08 | 北京智慧眼科技股份有限公司 | Facial recognition data processing unit and processing method based on FPGA |
-
2019
- 2019-05-22 CN CN201910429293.8A patent/CN110147880A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102339846A (en) * | 2010-07-19 | 2012-02-01 | 旺宏电子股份有限公司 | Semiconductor memory element possessing transistor with adjustable grid resistance value |
| CN103578552A (en) * | 2012-08-10 | 2014-02-12 | 三星电子株式会社 | Nonvolatile memory device and operating method with variable memory cell state definitions |
| CN103681679A (en) * | 2012-08-30 | 2014-03-26 | 成都海存艾匹科技有限公司 | Three-dimensional offset-printed memory |
| CN103533378A (en) * | 2013-10-09 | 2014-01-22 | 天津大学 | Three-dimensional integer DCT (Discrete Cosine Transform) transformation system on basis of FPGA (Field Programmable Gate Array) and transformation method thereof |
| US20160092122A1 (en) * | 2014-09-30 | 2016-03-31 | Sandisk Technologies Inc. | Method and apparatus for wear-levelling non-volatile memory |
| CN107316014A (en) * | 2016-03-07 | 2017-11-03 | 杭州海存信息技术有限公司 | Have the memory of image identification function concurrently |
| CN107220704A (en) * | 2016-03-21 | 2017-09-29 | 杭州海存信息技术有限公司 | Integrated neural network processor containing three-dimensional memory array |
| CN106898371A (en) * | 2017-02-24 | 2017-06-27 | 中国科学院上海微系统与信息技术研究所 | Three-dimensional storage reading circuit and its wordline and bit-line voltage collocation method |
| CN108596331A (en) * | 2018-04-16 | 2018-09-28 | 浙江大学 | A kind of optimization method of cell neural network hardware structure |
| CN109446996A (en) * | 2018-10-31 | 2019-03-08 | 北京智慧眼科技股份有限公司 | Facial recognition data processing unit and processing method based on FPGA |
Non-Patent Citations (1)
| Title |
|---|
| 罗玉平等: "《高清晰度电视3D-DCT变换硬件核解决方案》", 《图形与图像》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111091188A (en) * | 2019-12-16 | 2020-05-01 | 腾讯科技（深圳）有限公司 | Forward computing method and device for neural network and computer readable storage medium |
| CN111091188B (en) * | 2019-12-16 | 2022-03-25 | 腾讯科技（深圳）有限公司 | Forward computing method and device for neural network and computer readable storage medium |
| CN111506518A (en) * | 2020-04-13 | 2020-08-07 | 湘潭大学 | Data storage control method and device |
| CN111857999A (en) * | 2020-07-10 | 2020-10-30 | 苏州浪潮智能科技有限公司 | Data scheduling method, device and equipment and computer readable storage medium |
| CN111857999B (en) * | 2020-07-10 | 2023-01-10 | 苏州浪潮智能科技有限公司 | Data scheduling method, device and equipment and computer readable storage medium |
| CN112016522A (en) * | 2020-09-25 | 2020-12-01 | 苏州浪潮智能科技有限公司 | Video data processing method, system and related components |
| CN112016522B (en) * | 2020-09-25 | 2022-06-07 | 苏州浪潮智能科技有限公司 | Video data processing method, system and related components |
| CN112395247A (en) * | 2020-11-18 | 2021-02-23 | 北京灵汐科技有限公司 | Data processing method and storage and calculation integrated chip |
| CN112395247B (en) * | 2020-11-18 | 2024-05-03 | 北京灵汐科技有限公司 | Data processing method and memory and calculation integrated chip |
| CN112596684A (en) * | 2021-03-08 | 2021-04-02 | 成都启英泰伦科技有限公司 | Data storage method for voice deep neural network operation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110147880A (en) | A kind of Neural Network Data processing structure, method, system and relevant apparatus | |
| US20130262774A1 (en) | Method and apparatus to manage object based tier | |
| TWI554883B (en) | Systems and methods for segmenting data structures in a memory system | |
| KR20210045509A (en) | Accessing data in multi-dimensional tensors using adders | |
| Van Houdt | Performance of garbage collection algorithms for flash-based solid state drives with hot/cold data | |
| CN106201923B (en) | Method for reading and writing data and device | |
| US9335947B2 (en) | Inter-processor memory | |
| CN108885596A (en) | Data processing method, equipment, dma controller and computer readable storage medium | |
| CN105573917B (en) | Techniques for selecting an amount of reserved space | |
| CN110688256B (en) | Metadata power-on recovery method and device, electronic equipment and storage medium | |
| CN109754359A (en) | A kind of method and system that the pondization applied to convolutional neural networks is handled | |
| CN107832151A (en) | A kind of cpu resource distribution method, device and equipment | |
| CN112906865B (en) | Neural network architecture searching method and device, electronic equipment and storage medium | |
| CN109472361A (en) | Neural network optimization | |
| CN109074335A (en) | Data processing method, equipment, dma controller and computer readable storage medium | |
| CN105528183B (en) | A kind of method and storage equipment of storing data | |
| CN103714010B (en) | Storage device write-in method and storage device | |
| US20090055616A1 (en) | Maintaining reserved free space for segmented logical volumes | |
| CN112396072B (en) | Image classification acceleration method and device based on ASIC (application specific integrated circuit) and VGG16 | |
| CN109902821A (en) | A kind of data processing method, device and associated component | |
| CN104951243B (en) | Storage extended method and device in virtual storage system | |
| US8458405B2 (en) | Cache bank modeling with variable access and busy times | |
| CN114444673A (en) | Artificial neural network memory system based on artificial neural network data locality | |
| US20120079188A1 (en) | Method and apparatus to allocate area to virtual volume based on object access type | |
| CN106155910A (en) | A kind of methods, devices and systems realizing internal storage access |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190820 |