CN109784482A - Class nerve computing system and its current estimation method - Google Patents
Class nerve computing system and its current estimation method Download PDFInfo
- Publication number
- CN109784482A CN109784482A CN201711105185.2A CN201711105185A CN109784482A CN 109784482 A CN109784482 A CN 109784482A CN 201711105185 A CN201711105185 A CN 201711105185A CN 109784482 A CN109784482 A CN 109784482A
- Authority
- CN
- China
- Prior art keywords
- terminal
- sensing
- voltage value
- line
- circuit
- 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.)
- Granted
Links
Landscapes
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
- Image Analysis (AREA)
Abstract
One type nerve computing system, including cynapse cell array, switching circuit, sensing circuit and processing circuit.Cynapse cell array includes a plurality of alignment, a plurality of line and multiple cynapse units.Cynapse unit is located at the infall of alignment and line.Switching circuit couples cynapse cell array, line is electrically connected to first terminal or second terminal.Sensing circuit couples cynapse cell array, to sense voltage value and current value on line.Processing circuit couples switching circuit and sensing circuit, and be configured and to: pass through switching circuit, a specific line in a plurality of line is electrically connected to first terminal, passes through sensing circuit, a first voltage value is obtained from specific line when being electrically connected to first terminal, passes through switching circuit, specific line is electrically connected to second terminal, passes through sensing circuit, second voltage value is obtained, according to the voltage difference between the first voltage value and second voltage value from specific line when being electrically connected to second terminal, estimates product item and sensing current value.
Description
Technical field
The present invention relates generally to a type nerve computing system, and real based on hardware array structure institute more particularly to one kind
Existing class nerve computing system.
Background technique
Recently, the class nerve computing device realized using hardware array structure is suggested.Compared to utilization processor (example
Such as CPU) device of Lai Zhihang class nerve calculation, class nerve computing device has the advantages that low-power consumption.
Class nerve computing device generally includes multiple cynapse units (synapse).Each cynapse unit corresponds to a weight
Value.When an input vector is applied to class nerve computing device, this input vector will be right with one or more associated cynapse unit institutes
The weight vectors that the weighted value answered is constituted are multiplied, and product item and (sum of product) sensing electricity are formed in output channel
Stream.This product item and the one product item of size reflection and result for sensing electric current.
However, product item and sensing electric current in output channel may become quite big as cynapse element number increases, make
Must consume energy raising.
Summary of the invention
The present invention relates generally to a kind of class nerve computing system realized based on hardware array structure.It is real according to the present invention
Example is applied, the output channel of cynapse cell array is switchably coupled to first terminal or second terminal.Output channel is being connected to
The first voltage value can be presented when first terminal, and second voltage value is presented when being connected to second terminal.Product item and sensing electric current
Value can be estimated out according to the difference between the first voltage value and second voltage value.It may be straight compared in conventional method
Connect and the product item and high current that flow through output channel measured to carry out operation, according to the present invention, be connected to first terminal or
The only possible confined electric current of conducting size of the output channel of second terminal, or even it is not turned on electric current, therefore can effectively reduce energy consumption.
According to an aspect of the invention, it is proposed that a type nerve computing device.Class nerve computing system includes cynapse unit
Array, switching circuit, sensing circuit and processing circuit.Cynapse cell array includes a plurality of alignment, a plurality of line and multiple
Cynapse unit.Cynapse unit is located at the infall of alignment and line.Switching circuit couples cynapse cell array, and to by line
It is connected to first terminal or second terminal.Sensing circuit couples cynapse cell array, to sense voltage value on line and
Current value.Processing circuit couples switching circuit and sensing circuit, be configured and to: will be in a plurality of line by switching circuit
A specific line be electrically connected to first terminal, by sensing circuit from be electrically connected to first terminal when specific line take
A first voltage value is obtained, specific line is electrically connected to by second terminal by switching circuit, is connected by sensing circuit from electrical
Specific line when being connected to second terminal obtains second voltage value, according to the voltage between the first voltage value and second voltage value
Difference estimates product item and sensing current value.
According to another aspect of the invention, it is proposed that the current estimation method of a type nerve computing device.Class nerve calculates
System includes cynapse cell array, switching circuit, sensing circuit and processing circuit, cynapse cell array include a plurality of alignment,
A plurality of line and multiple cynapse units positioned at the infall of alignment and line.The current estimation method includes: to pass through switching
A specific line in a plurality of line is electrically connected to first terminal, is electrically connected to first certainly by sensing circuit eventually by circuit
Specific line when end obtains the first voltage value, specific line is electrically connected to second terminal by switching circuit, passes through sense
Slowdown monitoring circuit obtains second voltage value from specific line when being electrically connected to second terminal, passes through processing circuit according to first voltage
One product item of voltage difference estimation and sensing current value between value and second voltage value.
More preferably understand to have to above-mentioned and other aspect of the invention, special embodiment below, and cooperates appended attached
Detailed description are as follows for figure:
Detailed description of the invention
Fig. 1 is the block diagram for the class nerve computing system that an embodiment according to the present invention is schematically painted.
Fig. 2 is schematically painted the circuit structure diagram of cynapse cell array and switching circuit.
Fig. 3 is the flow chart of the current estimation method of class nerve computing system depicted in an embodiment according to the present invention.
[symbol description]
102: cynapse cell array
104: switching circuit
106: sensing circuit
108: processing circuit
201,203,205: alignment
202,204,206: line
210: cynapse unit
WU: resistive element
SU: selector
V1、V2、V3: input voltage
I1、I2、I3: sensing electric current
T1: first terminal
T2: second terminal
SW: switch element
302,304,306,308,310: step
Specific embodiment
Fig. 1 is the block diagram for the class nerve computing system that an embodiment according to the present invention is schematically painted.Class nerve
Computing system includes cynapse cell array 102, switching circuit 104, sensing circuit 106 and processing circuit 108.Switching circuit
104 and sensing circuit 106 couple cynapse cell array 102.Processing circuit 108 couples switching circuit 104 and sensing circuit
106。
Input vector can be formed by weighing vector progress phase with by one or more cynapse units by cynapse cell array 102
Multiplied by carrying out long-pending item and operation.Switching circuit 104 is controlled by processing circuit 108, to by each defeated of cynapse cell array 102
Channel is connected to first terminal or second terminal out.Sensing circuit 106 can sense the voltage value and current value of output channel.Cause
This, the first electricity that the output channel that sensing circuit 106 can obtain cynapse cell array 102 is presented when being connected to first terminal
Pressure value and the second voltage value presented when being connected to second terminal.Processing circuit 108 can according to the first voltage value and
Voltage difference between second voltage value estimates long-pending item and senses the size of current value.
Sensing circuit 106 is for example including sensing amplifier (sensing amplifier).Processing circuit 108 can for example with
Microprocessor, microcontroller, chip and/or circuit board are realized.
Fig. 2 is schematically painted the circuit structure diagram of cynapse cell array 102 and switching circuit 104.Although being drawn in Fig. 2
Show 3 × 3 cynapse units, but should be noted that cynapse cell array 102 may include any number of cynapse unit and combination.
As shown in Fig. 2, cynapse cell array 102 includes a plurality of alignment 201,203,205, a plurality of line 202,204,206
And multiple cynapse units 210 positioned at alignment 201,203,205 and the infall of line 202,204,206.
Alignment 201,203,205 is the input channel as cynapse cell array 102, to receive input voltage respectively
V1、V2、V3.Input voltage V1、V2、V3Input vector [the V provided system is provided1、V2、V3].Cynapse unit 210 can respond
Received from the input voltage V on alignment 201,203,2051、V2、V3, the output channel as array line 202,204,
Sensing electric current I is respectively formed on 2061、I2、I3。
Cynapse unit 210 can be any weight elements suitable for class nerve computing device, such as by resistive element WU
And " 1S1R " circuit structure that selector SU (such as transistor) series winding is formed.
One end of each line 202,204,206 can be connected to first terminal by switch element SW by switching circuit 104
T1 or second terminal T2.The all ungrounded end first terminal T1 and second terminal T2.It can different from traditional class nerve computing device
Product item and big sensing electric current can be formed on the line of ground connection, when line 202,204,206 is connected to first terminal T1 or second
Terminal T2, the sensing electric current I on line 202,204,2061、I2、I3To be constrained to preset low current (it is aobvious be less than product item and
Electric current) or even no current generation.
In an example, first terminal T1 is a suspension joint node, and second terminal T2 is a current limiting element, such as electric current
Mirror, transistor or other current sources that can provide fixation/limitation electric current.
Although should be noted that switching circuit 104 it should be noted that being painted 3 groups of first terminal T1 and second terminal T2 in Fig. 2
It may include any number of first terminal T1/ second terminal T2 and combination.For example, a plurality of line can share identical first
Terminal T1 and/or second terminal T2.
Sensing circuit 106 couples line 202,204,206.Sensing circuit 106 can detect line 202,204,206 first
State (being connected to the state of first terminal T1) or the second state (i.e. the state that one end of line is connected to second terminal T2)
When the current value, the voltage value that are presented, and detecting result is supplied to processing circuit 108 and estimates product item and current value.
As an example it is assumed that first terminal T1 is a suspension joint node, second terminal T2 is a current limiting element, processing electricity
The line (such as line 202) that road 108 can be intended to read first with switching circuit 104 is set in first state, and utilizes sensing
Circuit 106 obtains the first voltage value on the line.Sensing circuit 106 then can be set in the second state by processing circuit 108,
With the sensing current value for obtaining second voltage value on the line and the line is connected.In this way, processing circuit 108
It can be according to the first voltage value (Va) and second voltage value (Vb) between voltage difference and sensing current value (Is) between product,
Estimate long-pending item and sensing current value (Isp).For example, product item and sensing current value IspIt can be expressed as follows:
By taking line 202 as an example, in order to estimate the product item and current value of line 202, processing circuit 108 can first pass through switching electricity
Road 104 obtains line by 202 suspension joint of line (line 202 is namely connected to first terminal T1), and by sensing circuit 106
The first voltage value on 202, such as 0.5V.
Then, in response to line 202 is switched the second terminal T2 for being connected to and realizing with 50 μ A current sources, processing circuit
108 second voltage values that will be obtained by sensing circuit 106 on line 202, such as 0.4V.
After obtaining the first voltage value and second voltage value, processing circuit 108 can estimate line according to (formula 0)
202 product item and sensing current value is as follows:
To help to understand illustrate why (formula 0) can be used for estimating long-pending item and current value below.
Firstly, know to be grounded when one end of a certain line, the electric current on the line that is, product item and sensing electric current,
It can be expressed as follows:
Wherein gout,iIndicate the weighted value for being coupled to the cynapse unit of i-th alignment and the line to be read, ViTable
Show the input voltage value for being applied to i-th alignment, Iout|groundIt indicates to be formed by sensing current value when line ground connection (also
It is product item and sensing current value (I to be estimatedsp))。
In order to reduce the electric current generated on line when long-pending item and operation, switching circuit 104 can respond processing circuit 108,
The line is connected to suspension joint node (i.e. first terminal T1 in this example).When the line is suspension joint, which will not electric conduction
Flow (the sensing electric current I namely on lineout|floatingFor 0), and equilibrium voltage value V is presentedout|floating(i.e. the in this example
One voltage value Va).Therefore, (formula 1) can be rewritten as follows:
Wherein
Switching circuit 104 can more respond processing circuit 108, which is connected to current limiting element (in this instance i.e.
Second terminal T2).It at this time will one sensing electric current I of conducting on the lines, and there is voltage value Vout|Iout=Is(in this instance i.e. the
Two voltage value Vb) it is as follows:
Wherein the value of α is between 0 to 1.
According to (formula 1), (formula 2) and (formula 3), can be obtained:
It can be seen that, (formula 4) and (formula 0) mathematical notation having the same.
In other examples, first terminal T1 and second terminal T2 are two current limit members of corresponding different current values
Part.At this point, when a line is on the line to sense electric current using first terminal T1 as terminal and have the first electricity conducting first
Pressure value.When the line is on the line to sense electric current using second terminal T2 as terminal and have second voltage conducting second
Value.By simply modifying the derivation process of (formula 1) to (formula 4), processing circuit 108 is using the first sensing current value, first
Voltage value, the second sensing current value and second voltage value estimate the product item and current value of the corresponding line.
Fig. 3 is the flow chart of the current estimation method of class nerve computing system depicted in an embodiment according to the present invention.
In step 302, the specific line that switching circuit 104 is intended to read is electrically connected to first terminal T1.
In step 304, for processing circuit 108 by sensing circuit 106, this when being electrically connected to first terminal T1 certainly is specific
Line obtains the first voltage value.
In step 306, which is electrically connected to second terminal T2 by switching circuit 104.
In step 308, for processing circuit 108 by sensing circuit 106, this when being electrically connected to second terminal T2 certainly is specific
Line obtains second voltage value.
In step 310, processing circuit 108 is according to the voltage difference between the first voltage value and second voltage value, estimation
One product item and sensing current value.
In one embodiment, too small and be not easy interpretation in order to solve voltage difference between the first voltage value and second voltage value
The problem of, the voltage difference can be converted to time domain (time domain), according to conversion by specially designed detection technology
As a result long-pending item and sensing current value are estimated.For example, can plan line under first state/second state to capacitor charging,
To obtain the voltage difference between the first voltage value and second voltage value according to the charge and discharge time of capacitor, so estimate product item and
Sense current value.
In conclusion the present invention relates generally to a kind of class nerve computing system realized based on hardware array structure.Root
According to the embodiment of the present invention, the output channel of cynapse cell array is switchably coupled to first terminal or second terminal.Output is logical
The first voltage value can be presented when being connected to first terminal for road, and second voltage value is presented when being connected to second terminal.Product item
And sensing current value can be estimated out according to the difference between the first voltage value and second voltage value.Compared to tradition side
Directly the product item and high current that flow through output channel may be measured to carry out operation in method, according to the present invention, are connected to
The confined electric current of the only possible conducting size of the output channel of first terminal or second terminal, or even it is not turned on electric current, therefore can have
Effect reduces energy consumption.
Although the present invention has been disclosed by way of example above, it is not intended to limit the present invention..The neck of technology belonging to the present invention
Those of ordinary skill in domain, without departing from the spirit and scope of the present invention, when can be used for a variety of modifications and variations.Therefore, originally
The protection scope of invention, which is worked as, is subject to what claim was defined.
Claims (10)
1. a type nerve computing system, comprising:
One cynapse cell array, comprising:
A plurality of alignment；
A plurality of line；And
Multiple cynapse units, positioned at the infall of these alignments and these lines；
One switching circuit couples the cynapse cell array, the respectively line is electrically connected to a first terminal or one second
Terminal；
One sensing circuit couples the cynapse cell array, to sense voltage value and current value on these lines；And
One processing circuit couples the switching circuit and the sensing circuit, and be configured and to:
By the switching circuit, the specific line in these lines is electrically connected to the first terminal；
By the sensing circuit, specific line when being electrically connected to the first terminal certainly obtains a first voltage value；
By the switching circuit, which is electrically connected to the second terminal；
By the sensing circuit, specific line when being electrically connected to the second terminal certainly obtains a second voltage value；And
According to the voltage difference between the first voltage value and the second voltage value, one product item of estimation and sensing current value.
2. class nerve computing system as described in claim 1, wherein the first terminal is a suspension joint node, which is
One current limiting element.
3. class nerve computing system as claimed in claim 2, wherein when the specific line is electrically connected to the second terminal, it should
Processing circuit also to obtain the sensing current value on the specific line from the sensing circuit, and according to the voltage difference and
Product between the sensing current value estimates the product item and sensing current value.
4. class nerve computing system as claimed in claim 3, wherein the product item and sensing current value (Isp) are as follows:
Wherein IsFor the sensing current value, VaFor the first voltage value, VbFor the second voltage value.
5. class nerve computing system as claimed in claim 2, wherein the current limiting element is a current mirror or a transistor.
6. the current estimation method of a type nerve computing system, such neural computing system includes a cynapse cell array, one
Switching circuit, a sensing circuit and a processing circuit, the cynapse cell array include a plurality of alignment, a plurality of line and are located at
Multiple cynapse units of the infall of these alignments and these lines, the current estimation method include:
By the switching circuit, the specific line in these lines is electrically connected to the first terminal；
By the sensing circuit, specific line when being electrically connected to the first terminal certainly obtains a first voltage value；
By the switching circuit, which is electrically connected to the second terminal；
By the sensing circuit, specific line when being electrically connected to the second terminal certainly obtains a second voltage value；And
Circuit through this process, according to the voltage difference between the first voltage value and the second voltage value, one product of estimation
Item and sensing current value.
7. current estimation method as claimed in claim 6, wherein the first terminal is a suspension joint node, which is one
Current limiting element.
8. current estimation method as claimed in claim 7, further includes:
When the specific line is electrically connected to the second terminal, it is electric to obtain the sensing on the specific line using the sensing circuit
Flow valuve；And
Using the processing circuit, according to the product between the voltage difference and the sensing current value, the product item and sensing are estimated
Current value.
9. current estimation method as claimed in claim 8, wherein the product item and sensing current value (Isp) are as follows:
Wherein IsFor the sensing current value, VaFor the first voltage value, VbFor the second voltage value.
10. current estimation method as claimed in claim 7, wherein the current limiting element is a current mirror or a transistor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711105185.2A CN109784482B (en) | 2017-11-10 | 2017-11-10 | Neural computing system and current estimation method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711105185.2A CN109784482B (en) | 2017-11-10 | 2017-11-10 | Neural computing system and current estimation method thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109784482A true CN109784482A (en) | 2019-05-21 |
| CN109784482B CN109784482B (en) | 2023-01-13 |
Family
ID=66485500
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711105185.2A Active CN109784482B (en) | 2017-11-10 | 2017-11-10 | Neural computing system and current estimation method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109784482B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112085149A (en) * | 2019-06-14 | 2020-12-15 | 旺宏电子股份有限公司 | Neural computing device |
| WO2022121644A1 (en) * | 2020-12-09 | 2022-06-16 | International Business Machines Corporation | Bias scheme for single-device synaptic element |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4963769A (en) * | 1989-05-08 | 1990-10-16 | Cypress Semiconductor | Circuit for selective power-down of unused circuitry |
| US20160343433A1 (en) * | 2014-02-28 | 2016-11-24 | Hewlett Packard Enterprise Development Lp | Sensing circuit for resistive memory array |
| CN106470023A (en) * | 2015-08-18 | 2017-03-01 | 华为技术有限公司 | Neurn simulation circuit |
| US20170308788A1 (en) * | 2011-09-16 | 2017-10-26 | International Business Machines Corporation | Event-driven universal neural network circuit |
-
2017
- 2017-11-10 CN CN201711105185.2A patent/CN109784482B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4963769A (en) * | 1989-05-08 | 1990-10-16 | Cypress Semiconductor | Circuit for selective power-down of unused circuitry |
| US20170308788A1 (en) * | 2011-09-16 | 2017-10-26 | International Business Machines Corporation | Event-driven universal neural network circuit |
| US20160343433A1 (en) * | 2014-02-28 | 2016-11-24 | Hewlett Packard Enterprise Development Lp | Sensing circuit for resistive memory array |
| CN106470023A (en) * | 2015-08-18 | 2017-03-01 | 华为技术有限公司 | Neurn simulation circuit |
Non-Patent Citations (3)
| Title |
|---|
| Y. HORIO等: "Neuron-synapse IC chip-set for large-scale chaotic neural networks", 《IEEE TRANSACTIONS ON NEURAL NETWORKS》 * |
| 孔晓荣等: "对人工智能中的类神经网络方面的研究与探讨", 《内蒙古科技与经济》 * |
| 王蓓: "神经元电路设计实现的研究", 《中国优秀硕士学位论文全文数据库信息科技辑（月刊）》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112085149A (en) * | 2019-06-14 | 2020-12-15 | 旺宏电子股份有限公司 | Neural computing device |
| WO2022121644A1 (en) * | 2020-12-09 | 2022-06-16 | International Business Machines Corporation | Bias scheme for single-device synaptic element |
| US11741353B2 (en) | 2020-12-09 | 2023-08-29 | International Business Machines Corporation | Bias scheme for single-device synaptic element |
| GB2616393A (en) * | 2020-12-09 | 2023-09-06 | Ibm | Bias scheme for single-device synaptic element |
| GB2616393B (en) * | 2020-12-09 | 2024-05-08 | Ibm | Bias scheme for single-device synaptic element |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109784482B (en) | 2023-01-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US6700358B2 (en) | Automatic power factor correction system | |
| JP2017208819A (en) | Maintain power signature controller at power interface of poe or podl system | |
| CN104169843B (en) | Sparse sensing on interactive screens | |
| TW469411B (en) | Unevenness detecting sensor, fingerprint ID system and personal ID system | |
| CN109784482A (en) | Class nerve computing system and its current estimation method | |
| CN107110900A (en) | Insulation resistance measuring device and method | |
| CN207780205U (en) | A kind of high-voltage DC power supply ground insulation impedance detection circuit | |
| TW201044746A (en) | Control circuit and electronic apparatus comprising the same | |
| CN106645893A (en) | Charge pump and voltage detection circuit thereof | |
| CN110031705A (en) | Electrochemical capacitor life monitoring method, device, electronic equipment and storage medium | |
| CN209417155U (en) | A kind of exchange insulation resistance monitoring device | |
| CN103403640B (en) | Adapter and voltage setting of adapter | |
| CN107340924A (en) | Touch-control circuit, contact panel, display panel and display device | |
| TWI504905B (en) | Non-intrusive load monitoring method and device | |
| CN106033242A (en) | An external power supply unit and a system connection detecting unit for the same | |
| TWI647627B (en) | Neuromorphic computing system and current estimation method using the same | |
| US11062197B2 (en) | Neuromorphic computing system and current estimation method using the same | |
| EP2368303A1 (en) | Islanding detection system | |
| CN115167703A (en) | Touch driving device, touch device and touch driving method | |
| CN105333808B (en) | Displacement detecting circuit, eddy current displacement sensor and its correcting circuit and method | |
| JP2019518405A (en) | Determining the feed mode for DC powered devices | |
| CN207586404U (en) | Abnormal detection circuit and Switching Power Supply | |
| JP6465607B2 (en) | Power measuring apparatus and voltage system discrimination method | |
| CN103006205A (en) | Device and method for electrocardiosignal baseline recognition | |
| CN109031058B (en) | Insulation detection device and insulation detection circuit |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |