CN209388304U - Can software definition deposit the integrated chip of calculation and electronic equipment - Google Patents
Can software definition deposit the integrated chip of calculation and electronic equipment Download PDFInfo
- Publication number
- CN209388304U CN209388304U CN201920246699.8U CN201920246699U CN209388304U CN 209388304 U CN209388304 U CN 209388304U CN 201920246699 U CN201920246699 U CN 201920246699U CN 209388304 U CN209388304 U CN 209388304U
- Authority
- CN
- China
- Prior art keywords
- module
- flash memory
- configuration information
- programmable
- memory process
- 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.)
- Active
Links
- 238000004364 calculation method Methods 0.000 title claims abstract description 46
- 230000015654 memory Effects 0.000 claims abstract description 123
- 238000000034 method Methods 0.000 claims abstract description 106
- 230000008569 process Effects 0.000 claims abstract description 103
- 239000011159 matrix material Substances 0.000 claims abstract description 27
- 238000004088 simulation Methods 0.000 claims abstract description 21
- 238000006243 chemical reaction Methods 0.000 claims description 32
- 230000032258 transport Effects 0.000 claims description 16
- 230000001276 controlling effect Effects 0.000 claims description 5
- 239000000758 substrate Substances 0.000 claims description 4
- 230000001105 regulatory effect Effects 0.000 claims description 3
- 238000000465 moulding Methods 0.000 claims 1
- 230000002093 peripheral effect Effects 0.000 abstract description 5
- 239000004065 semiconductor Substances 0.000 description 20
- 238000013528 artificial neural network Methods 0.000 description 18
- 210000004027 cell Anatomy 0.000 description 11
- 210000002569 neuron Anatomy 0.000 description 8
- 238000000151 deposition Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 4
- 210000003205 muscle Anatomy 0.000 description 4
- 150000001875 compounds Chemical class 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 239000004020 conductor Substances 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 238000002347 injection Methods 0.000 description 1
- 239000007924 injection Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000005641 tunneling Effects 0.000 description 1
Landscapes
- Logic Circuits (AREA)
Abstract
The utility model provide it is a kind of can software definition deposit the integrated chip of calculation, this can software definition deposit that calculate the flash memory process array of integrated chip include for executing the multiple flash memory process subarrays of different simulation vector-matrix multiplication operations respectively, programmable computing module includes multiple programmable arithmetic elements for realizing different arithmetical operations respectively, control module is combined configuration to each module in the integrated chip of calculation is deposited according to the configuration information and finite state machine information of practical application, realize the dynamic configuration of circuit structure in chip, enable the chip to the circuit structure according to actual task flexible modulation chip, and ADC, DAC, register, the peripheral circuits such as programmable arithmetic element can be realized multiplexing, and then reduce circuit area, it adapts to integrated, the needs of miniaturization, and effectively reduce chip at This.
Description
Technical field
The utility model relates to semiconductor integrated circuit field, more particularly to it is a kind of can software definition deposit the integrated core of calculation
Piece.
Background technique
Flash memory is a kind of nonvolatile memory, realizes depositing for data by regulating and controlling the threshold voltage of flash memory transistor
Storage.According to the difference of flash memory transistor and array structure, flash memory is broadly divided into NOR- type flash memory and NAND- type flash memory. NAND-
As unit of page and block, capacity is big, at low cost for the read-write of type flash memory, is widely used in extensive free-standing memory;NOR-
Type flash memory supports the arbitrary access of data, and compared with NAND- type flash memory, density is lower, capacity is smaller, higher cost, mainly answers
For in-line memory.
In recent years, in order to solve traditional von Neumann counting system structure bottleneck, interior calculating (Computing-In- is deposited
Memory, CIM) chip architecture obtains the extensive concern of people, basic thought be directly using memory progress logic calculation,
To reduce the volume of transmitted data and transmission range between memory and processor, performance is improved while reducing power consumption.
Existing deposit calculates the one customized of integrated chip architecture, and circuit structure is fixed, and cannot be appointed according to practical
Business carries out flexible modulation, and circuit module can not achieve shared, cause circuit area big, does not adapt to integrated, miniaturization
It needs.
Utility model content
In view of this, the present invention provides it is a kind of can software definition deposit the integrated chip of calculation and equipment, by using
Multiple flash memory process subarrays, multiple programmable arithmetic elements and control module cooperation, according to practical application request pair
The circuit structure of chip carries out dynamic configuration, can carry out flexible modulation according to actual task, and ADC, DAC, register, can compile
The peripheral circuits such as journey arithmetic operation unit can be realized multiplexing, and then reduce circuit area, adapt to the need of integrated miniaturization
It wants.
To achieve the goals above, the utility model adopts the following technical solution:
In a first aspect, provide it is a kind of can software definition deposit the integrated chip of calculation, comprising: flash memory process array programmable is calculated
Art computing module and the control module being connect with the flash memory process array and the programmable computing module,
The flash memory process array includes for executing the different simulation multiple flash memory process of vector-matrix multiplication operation respectively
Subarray;
The programmable computing module includes multiple programmable operations for realizing different arithmetical operations respectively
Unit;
The control module according to configuration information to multiple flash memory process subarrays and multiple programmable arithmetic elements into
Row combination configuration, realizes the dynamic configuration of circuit structure in chip.
Further, can software definition deposit the integrated chip of calculation further include:
Input interface module, for receiving outer input data;
Input register heap connects the input interface module, for storing the outer input data or pending data;
D/A converter module, input terminal connect the input register heap, and output end connects the flash memory process array, and being used for will
The outer input data or pending data are converted to analog signal and transport to the flash memory process array, the flash memory process array pair
The analog signal carries out simulation vector-matrix multiplication operation and exports operation result;
Analog-to-digital conversion module, input terminal connect the flash memory process array, and output end connects the programmable computing module,
It, should for the simulation vector-matrix multiplication operation result to be converted to digital signal and transports to the programmable computing module
Programmable computing module carries out arithmetical operation to the digital signal and exports arithmetic operation results;
Output register heap connects the programmable computing module and the input register heap, for keeping in the arithmetic
Operation result, and the arithmetic operation results are exported or transport to the input register heap as the pending data;
Output interface module connects the output register heap, receives the output data of the output register heap, and this is defeated
Data export outward out;
Wherein, which connects the input interface module, the input register heap, the D/A converter module, the sudden strain of a muscle
Deposit processing array, the analog-to-digital conversion module, the output register heap, the programmable computing module and the output interface mould
Block, for carrying out dynamic configuration to foregoing circuit module according to practical application request.
Further, the output end of the input register heap is also connected with the programmable computing module.
Further, multiple programmable arithmetic element serial connections, each programmable arithmetic element are equal
It include: demultplexer, arithmetical operation subelement and multiple selector;
The input terminal of the demultplexer connects upper programmable arithmetic element or the analog-to-digital conversion module, wherein one
A output end connects the arithmetical operation subelement, and another output and the output end of the arithmetical operation subelement pass through the multichannel
Selector connects next programmable arithmetic element or output register heap, control terminal connect the control module.
Further, can software definition deposit the integrated chip of calculation further include: programmed circuit connect with the control module, this
Programmed circuit connects source electrode, grid and/or the substrate of each flash cell in the flash memory process array, for regulating and controlling flash memory
The threshold voltage of unit;
Wherein, which includes: the voltage generation circuit and use for generating program voltage or erasing voltage
In the voltage control circuit that the program voltage is loaded onto selected flash cell.
Further, can software definition deposit the integrated chip of calculation further include:
Ranks decoder connects the flash memory process array and the control module, under the control of the control module
The flash memory process array procession is decoded.
Further, which carries out dynamic configuration to each circuit module connected to it according to configuration information, should
Configuration information includes: the configuration information of flash memory process subarray, the configuration information of programmable arithmetic element, digital-to-analogue conversion mould
The configuration information of block, the configuration information of analog-to-digital conversion module, input interface module configuration information, output interface module configuration
The configuration information of information, the configuration information of input register heap and output register heap, this connects according to configuration information pair with it
Each circuit module connect carries out dynamic configuration
The flash memory process array is divided into multiple flash memory process submatrixs according to the configuration information of the flash memory process subarray
Column, and control the working sequence of multiple flash memory process subarrays;
The corresponding multichannel of each programmable arithmetic element is controlled according to the configuration information of the programmable arithmetic element
The working condition of distributor and multiple selector makes multiple programmable arithmetic elements realize any combination operation;
The D/A converting circuit open and-shut mode for participating in actual task is controlled according to the configuration information of the D/A converter module;
The analog to digital conversion circuit open and-shut mode for participating in actual task is controlled according to the configuration information of the analog-to-digital conversion module;
The input interface circuit open and-shut mode for participating in actual task is controlled according to the configuration information of the input interface module;
The output interface circuit open and-shut mode for participating in actual task is controlled according to the configuration information of the output interface module;
Input register data source to be stored is controlled according to the configuration information of the input register heap to connect in input
The input data or output register heap pending data of mouth mold block;
The output register heap is controlled by the data output or conduct in it according to the configuration information of the output register heap
Pending data transports to the input register heap.
Second aspect provides a kind of electronic equipment, including it is above-mentioned can software definition deposit the integrated chip of calculation.
It is provided by the utility model can software definition deposit calculation integrated chip, electronic equipment, this can software definition deposit calculation
The flash memory process array of integrated chip includes for executing the different simulation multiple flash memory process of vector-matrix multiplication operation respectively
Subarray, programmable computing module include multiple programmable operation lists for realizing different arithmetical operations respectively
Member, control module are combined with finite state machine information to the integrated each module of chip of calculation is deposited according to the configuration information of practical application
Configuration is realized the dynamic configuration of circuit structure in chip, is enabled the chip to according to circuit knot in actual task flexible modulation chip
Structure, and the peripheral circuits such as ADC, DAC, register, programmable arithmetic element can be realized multiplexing, and then reduce circuit face
Product, adapts to the needs of integrated miniaturization, and effectively reduces chip cost.
For the above and other objects, features and advantages of the utility model can be clearer and more comprehensible, preferable reality is cited below particularly
Example is applied, and cooperates institute's accompanying drawings, is described in detail below.
Detailed description of the invention
In order to illustrate the embodiment of the utility model or the technical proposal in the existing technology more clearly, below will be to embodiment
Or attached drawing needed to be used in the description of the prior art is briefly described, it should be apparent that, the accompanying drawings in the following description is only
It is some embodiments of the utility model, for those of ordinary skill in the art, in the premise not made the creative labor
Under, it is also possible to obtain other drawings based on these drawings.
Fig. 1 be the utility model embodiment can software definition deposit the structure chart one for calculating integrated chip;
Fig. 2 be the utility model embodiment can software definition deposit the structure chart two for calculating integrated chip;
Fig. 3 be the utility model embodiment can software definition deposit programmable arithmetic element 30 in the integrated chip of calculation
Structure chart;
Fig. 4 be the utility model embodiment can depositing for software definition calculate programmable operation subelement in integrated chip
Structure chart;
Fig. 5 be the utility model embodiment can the programmable computing module in the integrated chip of calculation of depositing of software definition realize
The schematic diagram of compound operation;
Fig. 6 be the utility model embodiment can software definition deposit the structure chart for calculating flash memory process subarray in integrated chip
One;
Fig. 7 be the utility model embodiment can software definition deposit the structure chart for calculating flash memory process subarray in integrated chip
Two;
Fig. 8 be the utility model embodiment can software definition deposit the structure chart for calculating flash memory process subarray in integrated chip
Three;
Fig. 9 be the utility model embodiment can software definition deposit the structure chart three for calculating integrated chip;
Specific embodiment
The following will be combined with the drawings in the embodiments of the present invention, carries out the technical scheme in the embodiment of the utility model
Clearly and completely describe, it is clear that the described embodiments are only a part of the embodiments of the utility model, rather than whole
Embodiment.Based on the embodiments of the present invention, those of ordinary skill in the art are without making creative work
Every other embodiment obtained, fall within the protection scope of the utility model.
Existing deposit calculates the one customized of integrated chip architecture, and circuit structure is fixed, and cannot be appointed according to practical
Business carries out flexible modulation, and circuit module can not achieve shared, cause circuit area big.
To solve the above problem in the prior art, the utility model embodiment provide it is a kind of it is provided by the utility model can
Software definition deposits calculation integrated chip, electronic equipment, this can depositing for software definition calculate the flash memory process array of integrated chip and include
For executing the different multiple flash memory process subarrays of simulation vector-matrix multiplication operation, programmable operation mould respectively
Block includes multiple programmable arithmetic elements for realizing different arithmetical operations respectively, and control module is according to practical application
Configuration information and finite state machine information are combined configuration to the integrated each module of chip of calculation is deposited, and realize circuit structure in chip
Dynamic configuration is enabled the chip to according to circuit structure in actual task flexible modulation chip, and ADC, DAC, register, can be compiled
The peripheral circuits such as journey arithmetic operation unit can be realized multiplexing, and then reduce circuit area, adapt to the need of integrated miniaturization
It wants.
Fig. 1 be the utility model embodiment can software definition deposit the structure chart one for calculating integrated chip.As shown in Figure 1, should
Can software definition deposit the integrated chip of calculation include: flash memory process array 20, programmable computing module 30 and with the flash memory
The control module 10 of array 20 and the programmable computing module 30 connection is handled,
The flash memory process array 20 includes for being executed at the different simulation multiple flash memories of vector-matrix multiplication operation respectively
Manage subarray (not shown in figure 1).
Wherein, multiple flash memory process subarrays can be the identical flash memory process subarray of structure, can also be according to actually answering
It is difference, such as the line number and column of each flash memory process subarray by the structure setting of each flash memory process subarray with demand
Number can be configured according to practical application request, the utility model embodiment to this with no restriction.
The programmable computing module 30 includes for realizing that multiple programmables of different arithmetical operations are transported respectively
Calculate unit (not shown in figure 1).
Programmable arithmetic element uses hardware realization, for executing specific arithmetical operation.
Wherein, arithmetical operation includes: multiplying, add operation, subtraction, division arithmetic, shift operation, activation letter
It counts, be maximized, being minimized, being averaged, the combination in one or more of pond etc..
The control module 10 posts the input interface module in chip, input according to configuration information and finite state machine information
Storage heap, D/A converter module, flash memory process array, analog-to-digital conversion module, output register heap, programmable computing module
And output interface module is combined configuration, realizes the dynamic configuration of circuit structure in chip.
Wherein it is possible to, by compilation tool, obtain configuration information and finite state machine information according to practical application request.
Wherein, configuration information is usually state that is static, such as specifying the modules of participation task, each unit
Configure size;Configuration information is generally held in memory, is waited and is scheduled before task runs.And finite state machine information is logical
It is often dynamic, in task run, when control actual task is run timing and state.
Specifically, the control module 10 is according to configuration information to multiple flash memory process subarray and multiple programmables
Arithmetic element is combined configuration, selects the flash memory process subarray devoted oneself to work and programmable arithmetic element, and to sudden strain of a muscle
The combination matching method for depositing processing subarray and programmable arithmetic element is controlled to realize certain operations.
It is understood that certain may be implemented in each programmable arithmetic element in multiple programmable arithmetic elements
One or more of arithmetical operations, multiple programmable arithmetic elements can go out a variety of compound operations with permutation and combination, with multiple sudden strains of a muscle
Processing subarray cooperation is deposited, can be realized multiple combinations configuration, and then realize complicated calculation function.
As can be seen from the above description, it is provided by the embodiment of the utility model can software definition deposit the integrated chip of calculation, at flash memory
Managing array includes that can compile for executing the different multiple flash memory process subarrays of simulation vector-matrix multiplication operation respectively
Journey arithmetical operation module includes multiple programmable arithmetic elements for realizing different arithmetical operations respectively, control module
Configuration is combined to multiple flash memory process subarrays and multiple programmable arithmetic elements according to configuration information, realizes chip
Framework dynamic configuration, can not only be according to actual task flexible modulation chip architecture, moreover it is possible to realize the calculation function of Various Complex,
And the peripheral circuits such as ADC, DAC, register, programmable arithmetic element can be realized multiplexing, and then reduce circuit area,
Adapt to the needs of integrated miniaturization.
In an alternative embodiment, referring to fig. 2, this can the integrated chip of calculation of depositing of software definition can also include: defeated
Incoming interface module 40, input register heap 50, D/A converter module 60, analog-to-digital conversion module 70, output register heap 80 and
Output interface module 90.
The input terminal of input interface module 40 connects external equipment, for receiving the input data from external equipment (i.e.
Need the data of operation).
The input terminal of input register heap 50 connects the output end of the input interface module 40, for keeping in the input data
An or pending data.
The input terminal of D/A converter module 60 connects the output end of the input register heap 50, and output end connects at the flash memory
The input terminal for managing array 20, for that will be converted from the outer input data that the input register heap 50 exports or pending data
For analog signal and the flash memory process array 20 is transported to, which carries out simulation vector-square to the analog signal
Battle array multiplying simultaneously exports simulation vector-matrix multiplication operation result.
70 input terminal of analog-to-digital conversion module connects the flash memory process array 20, and output end connects the programmable operation mould
Block 30, for the simulation vector-matrix multiplication operation result to be converted to digital signal and transports to the programmable operation mould
Block 30, the programmable computing module 30 carry out arithmetical operation to the digital signal and export arithmetic operation results.
The input terminal of output register heap 80 connects the programmable computing module 30, and output end connects input deposit
Device heap 50 for keeping in the arithmetic operation results, and the arithmetic operation results is exported or transport to this as the pending data
Input register heap 50.
The output end of the input terminal connection output register heap 80 of output interface module 90, receives the output register heap
80 output data, and the output data is exported to external equipment.
Wherein, which connects the input interface module 40, the input register heap 50, the D/A converter module
60, the flash memory process array 20, the analog-to-digital conversion module 70, the output register heap 80, the programmable computing module 30
And the output interface module 90, for carrying out dynamic configuration to foregoing circuit module according to configuration information.
Wherein, which carries out dynamic configuration to each circuit module connected to it according to configuration information, this is matched
Confidence breath includes: flash memory process subarray 201~20nConfiguration information, programmable arithmetic element 301~30nMatch confidence
Breath, the configuration information of D/A converter module 60, the configuration information of analog-to-digital conversion module 70, input interface module 40 match confidence
The configuration of breath, the configuration information of output interface module 90, the configuration information of input register heap 50 and output register heap 80
Information etc., it may include following content that this, which carries out dynamic configuration to each circuit module connected to it according to the configuration information:
According to the flash memory process subarray 201~20nConfiguration information the flash memory process array 20 is divided into multiple flash memories
Handle subarray 201~20n, and control multiple flash memory process subarrays 201~20nWorking sequence.
According to the programmable arithmetic element 301~30nConfiguration information to control each programmable arithmetic element corresponding
Selector working condition, make multiple programmable arithmetic elements realize any combination operation participate in work.
The D/A converting circuit opening and closing shape for participating in actual task is controlled according to the configuration information of the D/A converter module 60
State;
The analog to digital conversion circuit opening and closing shape for participating in actual task is controlled according to the configuration information of the analog-to-digital conversion module 70
State;
The input interface circuit opening and closing shape for participating in actual task is controlled according to the configuration information of the input interface module 40
State;
The output interface circuit opening and closing shape for participating in actual task is controlled according to the configuration information of the output interface module 90
State;
Input register data source to be stored is controlled in defeated according to the configuration information of the input register heap 50
The input data or output register heap pending data of incoming interface module;
According to the configuration information of the output register heap 80 control the output register heap 80 by it data output or
The input register heap 50 is transported to as pending data.
Specifically, the input terminal of the input register heap 50 passes through the connection of multiple selector (MUX) the 110 input interface mould
The output end of the output end of block 40 and the output register heap 80 receives the outside from the input interface module 40 with selectivity
Input data or pending data from the output register heap 80.The control module 10 connects the multiple selector (MUX)
100, which is controlled according to the configuration information, and then controlling the input register heap 50 is that the reception outside is defeated
Enter data or the pending data.
The D/A converter module 60 passes through the multiple flash memory process submatrix of demultplexer (DEMUX) 120 selective connection
Column (201~20n).Control module 10 connects the demultplexer 120, to control demultplexer Q according to configuration information, into
And which flash memory process subarray is selected to participate in work.
Multiple flash memory process subarray (201~20n) output end by a multiple selector 130 connect the modulus turn
Change the mold block 70.The control module 10 connects the multiple selector 130, controls the multiple selector 130 according to configuration information, in turn
The output of which flash memory process subarray is selected to connect the input terminal of the analog-to-digital conversion module 70, i.e., by above-mentioned participation work
The output of flash memory process subarray is connected to the input terminal of the analog-to-digital conversion module 70.
The input terminal of the programmable computing module 30 connects the demultplexer 110 by multiple selector 140
The output end of output end and the analog-to-digital conversion module 70.
Multiple programmable arithmetic elements 30 of the programmable computing module 301~30nSerial connection, each
The programmable arithmetic element includes: demultplexer 30a, arithmetical operation subelement 30b and multiple selector 30c,
Referring to Fig. 3.
The input terminal of demultplexer 30a connects upper programmable arithmetic element or the analog-to-digital conversion module 70,
One of output end connects arithmetical operation subelement 30b, the output end of arithmetical operation subelement 30b and the multichannel point
Orchestration 30a another output passes through the next programmable arithmetic element of a multiple selector 30c connection or output deposit
Device heap 80, in addition, the control terminal of demultplexer 30a and multiple selector 30c are all connected with the control module 20.
Specifically, the first programmable arithmetic element 301In the input terminal of demultplexer connect the analog-to-digital conversion
The output end of module 70, one of output end connect the first programmable arithmetic element 301In arithmetical operation it is single
The input terminal of member, another output connects second by a multiple selector with the output end of the arithmetical operation subelement can
Program arithmetic operation unit 302Input terminal, which connects the control module with the control terminal of the multiple selector
20。
Second programmable arithmetic element 302In demultplexer input terminal connect first programmable fortune
Calculate unit 301Output end, one of output end connects the second programmable arithmetic element 302In arithmetical operation son
The input terminal of unit, another output and the output end of the arithmetical operation subelement connect third by a multiple selector
Programmable arithmetic element 303Input terminal, which connects the control mould with the control terminal of the multiple selector
Block 20.The rest may be inferred, until the n-th programmable arithmetic element 30n, the n-th programmable arithmetic element 30nIn multichannel
The input terminal of distributor connects the (n-1)th programmable arithmetic element 30n-1Output end, one of output end connect this n-th
Programmable arithmetic element 30nIn arithmetical operation subelement input terminal, another output and arithmetical operation are single
The output end of member connects the input terminal of output register heap 80 by a multiple selector, and the demultplexer and the multichannel are selected
The control terminal for selecting device connects the control module 20.
Control module 20 connects the demultplexer and the multiple selector, root in each programmable arithmetic element
The demultplexer and the multiple selector in each programmable arithmetic element are controlled according to configuration information, to select this can
Whether the arithmetical operation subelement in programming arithmetic operation unit participates in operation, realizes multiple programmable arithmetic elements with this
Permutation and combination configuration, realize different complex calculations, flexible configuration arithmetical operation function.
In an alternative embodiment, each programmable operation subelement may each comprise multiple be arranged side by side
Arithmetic unit, such as multiplier, adder, subtracter, divider, shift unit, activation primitive device, be maximized operation
Device one or more of is minimized arithmetic unit, is averaged arithmetic unit, Chi Huaqi, in parallel between each arithmetic unit, defeated
Enter the output end that end is respectively connected to corresponding demultplexer, output end is respectively connected to the input of corresponding multiple selector
End, referring to fig. 4.
The process that the programmable computing module executes compound operation is as shown in Figure 5.
The output end of the output register heap 80 passes through 150 selective connection of the demultplexer output interface module 90
Input terminal or the input register heap 50 input terminal.The control module 20 connects the demultplexer 150, according to confidence
Breath controls the working condition of the demultplexer 150, and the output register heap 80 output result is transported to output interface with selection
Module 90 or input register heap 50, when the output register heap 80 output result is transported to input register heap 50 by selection
When, it is meant that the calculation process of a new round will be carried out to the output result.
In an alternative embodiment, the output end of the input register heap 50 can also pass through a demultplexer
The input terminal of the 110 selective connection D/A converter module 50 or the input terminal of the programmable computing module 30, the control
Module 10 connect the demultplexer 110, the working condition of the demultplexer 110 is controlled according to configuration information, with select by
The output end of the input register heap 50 is connected to the input terminal or the programmable operation mould of the D/A converter module 50
The input terminal of block 30, wherein when the output end of the input register heap 50 is connected to the input terminal of the D/A converter module 50,
Mean that the output to the input register heap 50 carries out simulation vector-matrix multiplication operation and arithmetical operation;When the input is posted
When the output end of storage heap 50 is connected to the input terminal of the input terminal of the programmable computing module 30, it is meant that the input
The output of register file 50 carries out certain arithmetical operation, further increases the flexibility of chip architecture with this.
In an alternative embodiment, each flash memory process subarray is all made of using source electrode coupling, drain electrode summation
Topological structure, referring to Fig. 6, multiple programmable semiconductor devices (also referred to as flash cell) including array arrangement.
Wherein, the source electrode of all programmable semiconductor devices of each column is connected to same analog voltage input, more
Column programmable semiconductor device is correspondingly connected with multiple analog voltage inputs;The leakage of all programmable semiconductor devices of each column
Pole is connected to the same analog current output end, and multiple row programmable semiconductor device is correspondingly connected with multiple analog current outputs
End;The grid of all programmable semiconductor devices of every a line is connected to same bias voltage input, multirow programmable half
Conductor device is correspondingly connected with multiple bias voltage inputs;Wherein, the threshold voltage of each programmable semiconductor device
It adjusts.
In another alternative embodiment, each flash memory process subarray includes the multiple programmable of array arrangement
Semiconductor devices;The grid of all programmable semiconductor devices of every a line is connected to same analog voltage input, multirow
Programmable semiconductor device is correspondingly connected with multiple analog voltage inputs;The drain electrode of all programmable semiconductor devices of each column
It is connected to same first end, multiple row programmable semiconductor device is correspondingly connected with multiple first ends;The all programmable of each column
The source electrode of semiconductor devices is connected to same second end, and multiple row programmable semiconductor device is correspondingly connected with multiple second ends;Often
The threshold voltage of a programmable semiconductor device is adjustable;Wherein, which is bias voltage input, the second end
For analog current output end, the topological structure of grid coupling, source electrode summation is realized, referring to Fig. 7;Alternatively, the first end is simulation
Current output terminal, the second end are bias voltage input, the topological structure of grid coupling, drain electrode summation are realized, referring to Fig. 8.
Specifically, which can will each be compiled by the threshold voltage of adjusting programmable semiconductor device
Journey semiconductor devices regards a variable equivalent simulation weight as, analog matrix data is equivalent to, to programmable semiconductor device
Array applies analog voltage, realizes matrix multiplication operation function.
In an alternative embodiment, this can software definition to deposit the integrated chip of calculation can also include: programmed circuit 22.
The programmed circuit 22 connects source electrode, grid and/or the substrate of each flash cell in flash memory process array, is used for
Regulate and control the threshold voltage of flash cell.
Wherein, the programmed circuit include: voltage generation circuit for generating program voltage or erasing voltage and
For the program voltage to be loaded onto the voltage control circuit of selected flash cell.
Specifically, programmed circuit utilizes thermoelectron injection effect, according to flash cell threshold voltage demand data, to flash memory
The source electrode of unit applies high voltage, channel electrons is accelerated to high speed, to increase the threshold voltage of flash cell.
Also, programmed circuit utilizes tunneling effect, according to flash cell threshold voltage demand data, to the grid of flash cell
Pole or substrate apply high voltage, to reduce the threshold voltage of flash cell.
In addition, control module 10 connects the programmed circuit, for controlling the programmed circuit according to configuration information, to the sudden strain of a muscle
The weight stored in processing array 20 is deposited to be adjusted.
In an alternative embodiment, this can software definition to deposit the integrated chip of calculation can also include: ranks decoder.
The ranks decoder connects the flash memory process array 20 and the control module 10, in the control module 10
Control under to 20 procession of flash memory process array decode.
In an alternative embodiment, programmable semiconductor device can be realized using floating transistor.
Wherein, which includes: that NOR type flash memory process array and NAND-type flash memory handle array, certainly, this
Utility model is not limited.
Based on the above content, the application provide it is a kind of using the utility model embodiment can software definition deposit the integrated chip of calculation
Realize the scene of neural network computing, with illustrate this can software definition deposit and calculate the workflow of integrated chip.
The neural network carries out operation for realizing to data P, which includes R layers of neuron, every layer of neuron
It is main to realize vector-matrix multiplication operation, and be attached by certain arithmetical operation (because of the application weight between each layer neuron
Point be can software definition deposit the integrated chip of calculation and only describe its fortune so do not carry out going deep into description to neural network computing herein
Calculate framework, with exemplary illustration can software definition deposit and calculate the workflow of integrated chip, be not limitation of the utility model).
For the neural network computing, this can software definition deposit calculate integrated chip workflow it is as follows:
Control module 10 obtains configuration information and finite state machine information, and the configuration information and finite state machine information include
The configuration information in R period and finite state machine information, R period (for example roll up corresponding to the operation of neural network R layers of neuron
Product, pond etc.), the operation of corresponding one layer of neuron of each period.The configuration information in each period includes: flash memory process submatrix
The configuration information of column, the configuration information of programmable arithmetic element, the configuration information of output register heap, input register heap
Configuration information etc..Flash memory process array 20 is divided into R flash memory process subarray according to the configuration information by control module 10,
Each flash memory process subarray corresponds to a cycle, i.e., each flash memory process subarray realizes one layer of neural network of operation, so
Control module 10 controls the working sequence of each circuit module according to finite state machine information afterwards.
Input interface module 40 receives data P;
Control module 10 controls the input register heap 50 according to the configuration information and finite state machine information of period 1
Multiple selector (DEMUX) A of front end, is connected to input interface module 40 with the input register heap 50, controls at the flash memory
Demultplexer (MUX) Q for managing 20 front end of array, so that the flash memory of the D/A converter module 60 and corresponding neural network first layer
Handle subarray 1 be connected to, control the multiple selector B of 20 rear end of flash memory process array, the flash memory process subarray 1 made with
Analog-to-digital conversion module 70 be connected to, control programmable computing module 30 each programmable arithmetic element selector and
Alternative selector realizes the arithmetical operation 1 for corresponding to neural network first layer, and transports to the input register heap in data P
The demultplexer W of 80 output end of output register heap and the multiple selector of 50 front end of input register heap are controlled after 50
(DEMUX) A makes the input terminal of the input register heap 50 be connected to the output end of the output register heap 80, realizes the period 1
Operation architecture configuration;
Data P transports to D/A converter module 60 after keeping in by the input register heap 50, after being converted to analog signal
Flash memory process subarray 1 is transported to, which carries out simulation vector-matrix multiplication operation 1 to the analog signal
(such as matrix multiplication operation), simulation vector-matrix multiplication operation result 1 switch to digital letter by the analog-to-digital conversion module 70
Number, arithmetic operation results 1 are obtained after programmable computing module 30, are transported to the input through output register heap is after 80s and are posted
Storage heap 50 so far completes the operation of first layer neural network;
Automatic trigger control module 10 at this time, control module 10 are believed according to the configuration information and finite state machine of second round
Breath controls demultplexer (MUX) Q of 20 front end of flash memory process array so that the D/A converter module 60 with it is corresponding neural
The flash memory process subarray 2 of the network second layer is connected to, and controls the multiple selector B of 20 rear end of flash memory process array, and what is made should
Flash memory process subarray 2 is connected to analog-to-digital conversion module 70, each programmable fortune of control programmable computing module 30
The selector of unit is calculated, the arithmetical operation 2 for corresponding to the neural network second layer is realized, realizes that the operation framework of second round is matched
It sets.
The arithmetic operation results 1 of the first layer neural network transport to digital-to-analogue conversion after keeping in by the input register heap 50
Module 60 transports to flash memory process subarray 2 after being converted to analog signal, which carries out the analog signal
It simulates vector-matrix multiplication operation 2 (such as matrix multiplication operation), simulation vector-matrix multiplication operation result passes through the modulus
Conversion module 70 switchs to digital signal, and arithmetic operation results 2 are obtained after programmable computing module 30, deposits through output
Device heap is after 80s to transport to the input register heap 50, so far completes the operation of second layer neural network, and so on, until last
Layer neural network, wherein when carrying out the configuration of the last layer neural network, the multichannel distribution of control 80 output end of output register heap
Device W makes the output end of the output register heap 80 connect the input terminal of output interface module 90, so that entire neural network
Operation result exported by the output interface module 90 to external equipment.
It will be appreciated by persons skilled in the art that when a certain layer neural network only needs arithmetical operation without simulating
When vector-matrix multiplication operation, only the multichannel point that the defeated register file 50 exports need to be controlled in 10 configuration circuit of control module
Orchestration E, so that the output end of the input register heap 50 is connected to the input terminal of the arithmetical operation module 30, other configurations
Process repeats no more.
Through the above technical solution it is known that it is provided by the embodiment of the utility model can software definition deposit the integrated core of calculation
Piece cooperates multiple flash memory process subarrays and multiple programmable arithmetic elements by control module, can be according to actually answering
Flexible combination is carried out to chip architecture with demand, can be realized complicated processor active task, is suitable for speech processes, image procossing, machine
The various application occasions such as device processing, artificial intelligence (AI), and the periphery such as ADC, DAC, register, programmable arithmetic element electricity
Road can be realized multiplexing, and then reduce circuit area, adapt to it is integrated, miniaturization needs, and effectively reduce chip at
This.
Fig. 9 be the utility model embodiment can software definition deposit the structure chart three for calculating integrated chip.As shown in figure 9, in Fig. 2
It is shown can be on the basis of software definition deposits the integrated chip of calculation, the input terminal of the input register heap 50 passes through multiple selector
(DEMUX) 100 the output end of the input interface module 40 and the output end of the output register heap 80 are connected, with selective reception
Outer input data from the input interface module 40 or the pending data from the output register heap 80.The control
Module 10 connects the multiple selector (DEMUX) 100.
The D/A converter module 60 passes through the multiple flash memory process subarray of demultplexer (MUX) 120 selective connection
(201~20n).Control module 10 connects demultplexer Q.
Multiple flash memory process subarray (201~20n) output end by a multiple selector 130 connect the modulus turn
Change the mold block 70.The control module 10 connects multiple selector B.
The input terminal of the programmable computing module 30 connects the demultplexer 110 by multiple selector 140
The output end of output end and the analog-to-digital conversion module 70.
Multiple programmable arithmetic elements 30 of the programmable computing module 301~30nSerial connection, each
The programmable arithmetic element includes selector 30a and arithmetical operation subelement 30b.
The input terminal of selector 30a connects upper programmable arithmetic element or the analog-to-digital conversion module 70, wherein
One output end connects arithmetical operation subelement 30b, the output of another output and arithmetical operation subelement 30b
End connects next programmable arithmetic element or output register heap 80 by an alternative selector, and control terminal connection should
Control module 20.
The output end of the output register heap 80 passes through 150 selective connection of the demultplexer output interface module 90
Input terminal or the input register heap 50 input terminal.The control module 20 connects demultplexer W, according to configuration information
The working condition of demultplexer W is controlled, the output register heap 80 output result is transported to by output interface module with selection
90 or input register heap 50, when the output register heap 80 output result is transported to input register heap 50 by selection, meaning
Taste will to the output result carry out a new round calculation process.
The output end of the input register heap 50 passes through 110 selective connection of the demultplexer D/A converter module
The input terminal of 50 input terminal or the programmable computing module 30, the control module 10 connect demultplexer E, according to
Configuration information controls the working condition of demultplexer E, and the output end of the input register heap 50 is connected to this with selection
The input terminal of the input terminal of D/A converter module 50 or the programmable computing module 30, wherein when the input register
When the output end of heap 50 is connected to the input terminal of the D/A converter module 50, it is meant that the output of the input register heap 50 into
Row simulation vector-matrix multiplication operation and arithmetical operation;When the output end of the input register heap 50 is connected to the programmable calculation
When the input terminal of the input terminal of art computing module 30, it is meant that carry out certain arithmetic to the output of the input register heap 50 and transport
It calculates, the flexibility of chip architecture is further increased with this.
It will be appreciated by persons skilled in the art that when a certain layer neural network only needs arithmetical operation without simulating
When vector-matrix multiplication operation, only the multichannel point that the defeated register file 50 exports need to be controlled in 10 configuration circuit of control module
Orchestration E, so that the output end of the input register heap 50 is connected to the input terminal of the arithmetical operation module 30, other configurations
Process repeats no more.
In addition, it will be appreciated by persons skilled in the art that according to practical application request generate configuration information when, Ke Yigen
It is realized according to pre-set instruction-framework mapping table.
It is worth noting that when generating configuration information according to practical application request, it is to be understood that the flash memory process for needing to put into
The scale of the quantity of subarray and each flash memory process subarray, at this point, flash memory process can be obtained according to practical application request
The division of array instructs, and is then instructed according to the division flash memory process array being divided into multiple flash memory process subarrays, right
Answer a variety of matrix multiplication operation scales.
It will be appreciated by persons skilled in the art that using the utility model embodiment can software definition deposit the integrated chip of calculation
When, when carrying out multiple cycle operations, the period corresponding flash memory process subarray can be programmed in each period,
Each flash memory process subarray can be uniformly programmed according to programming instruction before carrying out each cycle operation.
The utility model embodiment also provides a kind of electronic equipment, and neural network algorithm can be performed, which includes
Multilayer neuron, every layer of neuron carry out corresponding operation according to the output result of one layer of neuron thereon, which includes
It is above-mentioned can software definition deposit the integrated chip of calculation.
Wherein, which for example can be personal computer, laptop computer, cellular phone, camera phone, intelligence
Can phone, personal digital assistant, media player, navigation equipment, electronic mail equipment, game console, tablet computer, can
The combination of any equipment in wearable device or these equipment.
Specific embodiment is applied in the utility model to be expounded the principles of the present invention and embodiment, with
The explanation of upper embodiment is merely used to help understand the method and its core concept of the utility model;Meanwhile for this field
Those skilled in the art, based on the idea of the present invention, there will be changes in the specific implementation manner and application range, comprehensive
Upper described, the content of the present specification should not be construed as a limitation of the present invention.
Claims (8)
1. one kind can software definition deposit the integrated chip of calculation characterized by comprising flash memory process array, programmable fortune
The control module calculating module and being connect with the flash memory process array and the programmable computing module,
The flash memory process array includes for executing different multiple flash memory process of simulation vector-matrix multiplication operation respectively
Array;
The programmable computing module includes multiple programmable operation lists for realizing different arithmetical operations respectively
Member;
The control module carries out multiple flash memory process subarrays and multiple programmable arithmetic elements according to configuration information
Combination configuration, realizes the dynamic configuration of circuit structure in chip.
2. it is according to claim 1 can software definition deposit the integrated chip of calculation, which is characterized in that further include:
Input interface module, for receiving outer input data;
Input register heap connects the input interface module, for storing the outer input data or pending data;
D/A converter module, input terminal connect the input register heap, and output end connects the flash memory process array, and being used for will
The outer input data or pending data are converted to analog signal and transport to the flash memory process array, the flash memory process
Array carries out simulation vector-matrix multiplication operation to the analog signal and exports operation result;
Analog-to-digital conversion module, input terminal connect the flash memory process array, and output end connects the programmable computing module,
For the simulation vector-matrix multiplication operation result to be converted to digital signal and transports to the programmable operation mould
Block, the programmable computing module carry out arithmetical operation to the digital signal and export arithmetic operation results;
Output register heap connects the programmable computing module and the input register heap, for keeping in the calculation
Art operation result, and the arithmetic operation results are exported or transport to the input register heap as the pending data;
Output interface module connects the output register heap, receives the output data of the output register heap, and by the output
Data export outward;
Wherein, the control module connect the input interface module, the input register heap, the D/A converter module,
The flash memory process array, the analog-to-digital conversion module, the output register heap, the programmable computing module and
The output interface module, for carrying out dynamic configuration to foregoing circuit module according to practical application request.
3. it is according to claim 2 can software definition deposit the integrated chip of calculation, which is characterized in that the input register heap
Output end be also connected with the programmable computing module.
4. it is according to claim 2 can software definition deposit the integrated chip of calculation, which is characterized in that multiple programmable calculations
The serial connection of art arithmetic element, each programmable arithmetic element includes: demultplexer, arithmetical operation subelement
And multiple selector;
The input terminal of the demultplexer connects upper programmable arithmetic element or the analog-to-digital conversion module, wherein one
A output end connects the arithmetical operation subelement, and another output and the output end of the arithmetical operation subelement pass through institute
It states multiple selector and connects next programmable arithmetic element or output register heap, control terminal connects the control module.
5. it is according to claim 4 can software definition deposit the integrated chip of calculation, which is characterized in that further include: with the control
The programmed circuit of molding block connection, the programmed circuit connect the source of each flash cell in the flash memory process subarray
Pole, grid and/or substrate, for regulating and controlling the threshold voltage of flash cell;
Wherein, the programmed circuit includes: voltage generation circuit for generating program voltage or erasing voltage and is used for
The program voltage is loaded onto the voltage control circuit of selected flash cell.
6. it is according to claim 1 or 2 can software definition deposit the integrated chip of calculation, which is characterized in that further include:
Ranks decoder connects the flash memory process array and the control module, for the control in the control module
Under to the flash memory process array procession decode.
7. it is according to claim 4 can software definition deposit the integrated chip of calculation, which is characterized in that the control module according to
Configuration information carries out dynamic configuration to each circuit module connected to it, and the configuration information includes: flash memory process subarray
Configuration information, the configuration information of programmable arithmetic element, the configuration information of D/A converter module, analog-to-digital conversion module are matched
Confidence breath, the configuration information of input interface module, the configuration information of output interface module, input register heap configuration information with
And the configuration information of output register heap, it is described that dynamic configuration packet is carried out to each circuit module connected to it according to configuration information
It includes:
The flash memory process array is divided into multiple flash memory process submatrixs according to the configuration information of the flash memory process subarray
Column, and control the working sequence of multiple flash memory process subarrays;
The corresponding multichannel point of each programmable arithmetic element is controlled according to the configuration information of the programmable arithmetic element
The working condition of orchestration and multiple selector makes multiple programmable arithmetic elements realize any combination operation;
The D/A converting circuit open and-shut mode for participating in actual task is controlled according to the configuration information of the D/A converter module;
The analog to digital conversion circuit open and-shut mode for participating in actual task is controlled according to the configuration information of the analog-to-digital conversion module;
The input interface circuit open and-shut mode for participating in actual task is controlled according to the configuration information of the input interface module;
The output interface circuit open and-shut mode for participating in actual task is controlled according to the configuration information of the output interface module;
Input register data source to be stored is controlled in input interface according to the configuration information of the input register heap
The input data or output register heap pending data of module;
According to the configuration information of the output register heap control the output register heap by it data output or as to
Processing data transport to the input register heap.
8. a kind of electronic equipment, which is characterized in that including as described in any one of claims 1 to 7 can software definition deposit calculation
Integrated chip.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201920246699.8U CN209388304U (en) | 2019-02-26 | 2019-02-26 | Can software definition deposit the integrated chip of calculation and electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201920246699.8U CN209388304U (en) | 2019-02-26 | 2019-02-26 | Can software definition deposit the integrated chip of calculation and electronic equipment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN209388304U true CN209388304U (en) | 2019-09-13 |
Family
ID=67854445
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201920246699.8U Active CN209388304U (en) | 2019-02-26 | 2019-02-26 | Can software definition deposit the integrated chip of calculation and electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN209388304U (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111128279A (en) * | 2020-02-25 | 2020-05-08 | 杭州知存智能科技有限公司 | Memory computing chip based on NAND Flash and control method thereof |
CN112599132A (en) * | 2019-09-16 | 2021-04-02 | 北京知存科技有限公司 | Voice processing device and method based on storage and calculation integrated chip and electronic equipment |
CN116414456A (en) * | 2023-01-19 | 2023-07-11 | 杭州知存智能科技有限公司 | Weighted fusion conversion component in memory chip, memory circuit and cooperative computing method |
-
2019
- 2019-02-26 CN CN201920246699.8U patent/CN209388304U/en active Active
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112599132A (en) * | 2019-09-16 | 2021-04-02 | 北京知存科技有限公司 | Voice processing device and method based on storage and calculation integrated chip and electronic equipment |
CN111128279A (en) * | 2020-02-25 | 2020-05-08 | 杭州知存智能科技有限公司 | Memory computing chip based on NAND Flash and control method thereof |
CN116414456A (en) * | 2023-01-19 | 2023-07-11 | 杭州知存智能科技有限公司 | Weighted fusion conversion component in memory chip, memory circuit and cooperative computing method |
CN116414456B (en) * | 2023-01-19 | 2024-01-19 | 杭州知存智能科技有限公司 | Weighted fusion conversion component in memory chip, memory circuit and cooperative computing method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN209388304U (en) | Can software definition deposit the integrated chip of calculation and electronic equipment | |
CN106844294B (en) | Convolution algorithm chip and communication equipment | |
Park et al. | A 65k-neuron 73-Mevents/s 22-pJ/event asynchronous micro-pipelined integrate-and-fire array transceiver | |
CN108090560A (en) | The design method of LSTM recurrent neural network hardware accelerators based on FPGA | |
US20200192971A1 (en) | Nand block architecture for in-memory multiply-and-accumulate operations | |
US20160196488A1 (en) | Neural network computing device, system and method | |
CN111611197B (en) | Operation control method and device of software-definable storage and calculation integrated chip | |
CN111611195A (en) | Software-definable storage and calculation integrated chip and software definition method thereof | |
KR20130090147A (en) | Neural network computing apparatus and system, and method thereof | |
WO2023045114A1 (en) | Storage and computation integrated chip and data processing method | |
CN108170640B (en) | Neural network operation device and operation method using same | |
CN108763163A (en) | Simulate vector-matrix multiplication operation circuit | |
CN109086249A (en) | Simulate vector-matrix multiplication operation circuit | |
WO2020258360A1 (en) | Compute-in-memory chip, and memory unit array structure | |
TWI634489B (en) | Multi-layer artificial neural network | |
CN109146070A (en) | A kind of peripheral circuit and system of neural network training of the support based on RRAM | |
CN209182823U (en) | A kind of numerical model analysis deposits the integrated chip of calculation and the arithmetic unit for neural network | |
CN209766043U (en) | Storage and calculation integrated chip and storage unit array structure | |
CN110163359A (en) | A kind of computing device and method | |
CN111767994B (en) | Neuron computing device | |
JP7150998B2 (en) | Superconducting neuromorphic core | |
US11797830B2 (en) | Flexible accelerator for sparse tensors in convolutional neural networks | |
CN109670581B (en) | Computing device and board card | |
CN109086871A (en) | Training method, device, electronic equipment and the computer-readable medium of neural network | |
CN209388707U (en) | One kind depositing the integrated chip of calculation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CP03 | Change of name, title or address |
Address after: Room 213-175, 2nd Floor, Building 1, No. 180 Kecheng Street, Qiaosi Street, Linping District, Hangzhou City, Zhejiang Province, 311100 Patentee after: Hangzhou Zhicun Computing Technology Co.,Ltd. Country or region after: China Address before: 1416, shining building, No. 35, Xueyuan Road, Haidian District, Beijing 100083 Patentee before: BEIJING WITINMEM TECHNOLOGY Co.,Ltd. Country or region before: China |
|
CP03 | Change of name, title or address |