CN110263926A - Impulsive neural networks and its system and operation method based on photoelectricity computing unit - Google Patents
Impulsive neural networks and its system and operation method based on photoelectricity computing unit Download PDFInfo
- Publication number
- CN110263926A CN110263926A CN201910415826.7A CN201910415826A CN110263926A CN 110263926 A CN110263926 A CN 110263926A CN 201910415826 A CN201910415826 A CN 201910415826A CN 110263926 A CN110263926 A CN 110263926A
- Authority
- CN
- China
- Prior art keywords
- pulse
- computing unit
- read
- out area
- neural networks
- 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
- 230000005622 photoelectricity Effects 0.000 title claims abstract description 50
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 41
- 238000000034 method Methods 0.000 title claims abstract description 18
- 210000005036 nerve Anatomy 0.000 claims abstract description 33
- 210000002569 neuron Anatomy 0.000 claims abstract description 19
- 238000004020 luminiscence type Methods 0.000 claims abstract description 13
- 230000000694 effects Effects 0.000 claims abstract description 11
- 230000000946 synaptic effect Effects 0.000 claims abstract description 10
- 239000000203 mixture Substances 0.000 claims abstract description 5
- 230000000737 periodic effect Effects 0.000 claims abstract description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 34
- 230000005284 excitation Effects 0.000 claims description 27
- 238000013507 mapping Methods 0.000 claims description 15
- 230000005540 biological transmission Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 210000001367 artery Anatomy 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 8
- 210000003462 vein Anatomy 0.000 claims description 8
- 230000010354 integration Effects 0.000 claims description 4
- 230000008335 axon cargo transport Effects 0.000 claims description 3
- 230000008054 signal transmission Effects 0.000 claims description 3
- 230000003760 hair shine Effects 0.000 claims 1
- 239000000758 substrate Substances 0.000 description 22
- 239000004065 semiconductor Substances 0.000 description 12
- 210000004556 brain Anatomy 0.000 description 8
- 230000008878 coupling Effects 0.000 description 7
- 238000010168 coupling process Methods 0.000 description 7
- 238000005859 coupling reaction Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 238000002955 isolation Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 4
- 238000005468 ion implantation Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- VYPSYNLAJGMNEJ-UHFFFAOYSA-N Silicium dioxide Chemical compound O=[Si]=O VYPSYNLAJGMNEJ-UHFFFAOYSA-N 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000000151 deposition Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000005684 electric field Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005530 etching Methods 0.000 description 1
- 238000005194 fractionation Methods 0.000 description 1
- 230000009916 joint effect Effects 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000012856 packing Methods 0.000 description 1
- 238000005036 potential barrier Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000000377 silicon dioxide Substances 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/067—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means
- G06N3/0675—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using optical means using electro-optical, acousto-optical or opto-electronic means
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention discloses a kind of impulsive neural networks based on photoelectricity computing unit and its systems and operation method.It includes synaptic web, dendron, aixs cylinder, pulse receiver and neuron that the nerve of impulsive neural networks, which calculates nuclear structure, the array of multiple photoelectricity computing unit periodic arrangement compositions is as synaptic web, each photoelectricity computing unit includes luminescence unit and computing unit, and the light that luminescence unit issues is incident in computing unit；Each computing unit includes carrier control zone, coupled zone and photo-generated carrier collecting region and read-out area；The read-out area of each column count unit is sequentially connected in array, as dendron；The read-out area output end of dendron is connected with neuron；The carrier control zone of every a line computing unit is sequentially connected in array, as aixs cylinder；Carrier control zone in aixs cylinder is connected with pulse receiver.Neural network of the invention is not necessarily to access chip external memory repeatedly in actual operation, has achieved the effect that reduce power consumption.
Description
Technical field
The present invention relates to a kind of impulsive neural networks based on photoelectricity computing unit and its system and methods for using thems, belong to meter
Calculation field and photodetection field.
Background technique
Traditional computer takes von Neumann framework mostly, however, because von Neumann framework storage unit and operation
Unit it is discrete, result in and produce great energy consumption in data transmission, and influence arithmetic speed.Photoelectricity calculates single
Member can combine to carry out the calculating device of operation for one kind with operation independent or with current electronic computation technology, can be real
Now high-precision to deposit-calculate a body function, individual devices can store the optical signal of light input end and save for a long time after disconnected light,
And individual devices may be implemented and complete multiplying.These characteristics make photoelectricity computing unit be highly suitable for forming class brain
Impulsive neural networks.
Existing nerual network technique uses the deep neural network of CNN RNN class mostly, flat by general-purpose computations
Platform, such as CPU, GPU complete operation.CNN RNN class deep neural network weighting parameter is big, and computation complexity is high, needs big
Caching reads weight and completes multiple operation outside amount ground access piece, has high requirement to the calculating power of conventional computing system.And by
It is cached in outside access piece repeatedly, a large amount of energy losses cause its calculating efficiency low, be not able to satisfy and increasingly increase in data transmission procedure
Long application demand.
Summary of the invention
In order to improve the calculated performance of system, the present invention provides a kind of impulsive neural networks based on photoelectricity computing unit,
And the impulsive neural networks operation method and the application system that is made of the neural network.
The technical solution adopted by the invention is as follows:
Nuclear structure is calculated based on the impulsive neural networks of photoelectricity computing unit, including nerve, nerve calculates nuclear structure and includes
Synaptic web, dendron, aixs cylinder, pulse receiver and neuron；The array of multiple photoelectricity computing unit periodic arrangement compositions is made
For synaptic web, each photoelectricity computing unit includes luminescence unit and computing unit, and the light that luminescence unit issues is incident on calculating
In unit；Each computing unit includes carrier control zone, coupled zone and photo-generated carrier collecting region and read-out area；It is described
Carrier control zone is for controlling and modulating the carrier in photoproduction carrier collection area and read-out area；The photo-generated carrier is received
Collecting region in Ji Qu and read-out area is used to absorb the photon of luminescence unit transmitting and collects the photo-generated carrier of generation；The load
The read-out area flowed in sub- control zone or photo-generated carrier collecting region and read-out area is connect with electric signal, and read-out area is for exporting quilt
Carrier after the photo-generated carrier and electric signal effect；The coupled zone connection collecting region and read-out area；In array
The read-out area of each column count unit is sequentially connected, as dendron；The read-out area output end of dendron is connected with neuron；Array
In the carrier control zone of every a line computing unit be sequentially connected, as aixs cylinder；Carrier control zone in aixs cylinder connects with pulse
Device is received to be connected.
Further, impulsive neural networks include that multiple nerves calculate nuclear structure, pulse processor, address arbiter sum number
According to coffret, multiple nerves calculate nuclear structure and arrange in web form, and it is secondary that each nerve calculating nuclear structure connects an address
Cut out device；The pulse processor, the pulse signal of coffret input and each nerve calculate nuclear structure for receiving data
The pulse signal that transmits of aixs cylinder, and coding carried out according to each pulse signal of different calculation and check and by the pulse after coding
Signal passes to address arbiter；The address arbiter, for receiving the pulse signal of pulse processor transmission, and according to arteries and veins
Rush address arbiter of the encoded information of signal by pulse signal transmission to pulse receiver or neighborhood calculation core；The data
Coffret is converted into the arteries and veins suitable for pulse processor for receiving external pulse signal, and by external pulse signal
It rushes signal form and is transferred to pulse processor.
A kind of operation method of the impulsive neural networks based on photoelectricity computing unit of the present invention, specific steps include: luminous
Unit is calculated in the weighted value input synaptic web of nuclear structure and is saved under the driving of driver, by nerve；Pulse receiver
It receives from external pulse excitation, caches pulse excitation, and by the arteries and veins of caching after receiving impulsive synchronization control signal
Impulse is encouraged through axonal transport to the carrier control zone of every a line computing unit；The pulse that carrier control zone receives is swashed
It encourages and carries out multiplying with the weighted value stored in synaptic web, the calculated result of each computing unit is connected by dendron, from
And the electric current of the read-out area output end of each column count unit is converged, and exports to neuron；After neuron will converge
The current signal calculating current conversion that is AD converted into digital signal, and constantly the pulse excitation of input is generated after number
Word signal results add up, while accumulated result being brought into the neuron models function being made of Digital Logic, sentence
It is disconnected whether to generate output pulse signal.
A kind of application system of the impulsive neural networks based on photoelectricity computing unit of the present invention, including the pulse nerve net
Network and: Algorithm mapping module, for decomposing using algorithm for impulsive neural networks processing will to be needed；Weight write-in
Module, for weight needed for the algorithm after decomposing to be written in computing unit according to mapping relations by luminescence unit；Arteries and veins
Generation module is rushed, for generating corresponding pulse excitation according to the excited data in algorithm actual operation, and inputs to pulse
Receiver；As a result conversion module, the calculated result pulse signal that coffret transmits for receiving data, and according to corresponding calculation
Calculated result pulse signal is carried out integration processing by method model, is converted into the dominant calculated result of algorithm.
The present invention makes full use of depositing for photoelectricity computing unit to calculate integrated characteristic, by the weighting parameter in impulsive neural networks
By light be input in array save to simulate the connection weight in human brain, using the characteristic of the achievable multiplying of single device come
The operation for simulating neuron in human brain, to constitute the novel pulse neural network of a type brain.It, will be to reality in application process
Existing algorithm model is mapped on impulsive neural networks by algorithm decomposition, then data stimuli is converted by Algorithm mapping relationship
It at pulse excitation, is input in network and carries out operation, the acceleration to neural network may be implemented.Since photoelectricity deposits the integrated device of calculation
Collection storage and operation one, neural network of the invention are not necessarily to access chip external memory repeatedly in actual operation, can be very big
Power consumption is reduced, operation efficiency is improved, relatively existing deep neural network has very big advantage.
Detailed description of the invention
Fig. 1 is the multi-functional-area block diagram of computing unit.
Fig. 2 is the structural schematic diagram of photoelectricity computing array.
Fig. 3 is 1 computing unit structure (a) sectional view of embodiment and (b) perspective view.
Fig. 4 is 2 computing unit structure (a) sectional view of embodiment and (b) perspective view.
Fig. 5 is 3 computing unit (a) structural schematic diagram of embodiment and the multi-functional-area (b) schematic diagram.
Fig. 6 is that the nerve based on photoelectricity computing unit of 4*4 calculates the schematic diagram of nuclear structure.
Fig. 7 is the schematic diagram for the 4 nuclear network structures that the nerve based on photoelectricity computing unit calculates core.
Fig. 8 is the schematic diagram of the application flow of the impulsive neural networks based on photoelectricity computing unit.
In figure: 1- light emitting array, 2- computing array.
Specific embodiment
Computing unit in photoelectricity computing unit of the present invention is the multi-functional-area structure for including three zones area, such as Fig. 1 institute
Show, three zones area are as follows: carrier control zone, coupled zone, photo-generated carrier collecting region and read-out area, concrete function is respectively such as
Under:
Carrier control zone: it is responsible for controlling and modulates the carrier in photoelectricity computing unit, and as computing unit
Electrical input mouth inputs one of operand as electric input quantity；Or the carrier in computing unit is only controlled and modulates,
Electric input quantity is inputted by other regions.
Coupled zone: being responsible for connection photo-generated carrier collecting region and read-out area, so that the photo-generated carrier that photon incidence generates
The carrier in photoelectricity computing unit is acted on, operation relation is formed.
Photo-generated carrier collecting region and read-out area: wherein collecting region is responsible for absorbing incident photon and collects the photoproduction of generation
Carrier, and the light input port as computing unit input one of operand as light input quantity；Read-out area can be with
As the electrical input mouth of computing unit, one of operand is inputted as electric input quantity, and as the defeated of computing unit
Exit port, output is by the carrier after light input quantity and electric input quantity effect as unit output quantity；Or pass through other regions
Electric input quantity is inputted, read-out area is only used as the output port of computing unit, and output is by after light input quantity and electric input quantity effect
Carrier, as unit output quantity.
The light that luminescence unit issues is collected as incident computing unit photo-generated carrier and the photon of read-out area, participates in fortune
It calculates.Photoelectricity computing array includes light emitting array 1 and computing array 2, and structure is as shown in Figure 2.Light emitting array 1 is by multiple luminescence units
Periodic arrangement composition, computing array 2 are made of multiple computing unit periodic arrangements.
Embodiment 1
As shown in figure 3, the computing unit of the present embodiment includes: as the control grid of carrier control zone, as coupling
The Charged Couple floor in area, and as the P type substrate of photo-generated carrier collecting region and read-out area, left side is divided into P type substrate and is received
Ji Qu and right side read-out area, wherein including shallow-trench isolation in the read-out area of right side, by the N-type source and N-type of ion implanting formation
Drain terminal.Shallow-trench isolation is located at the centre at semiconductor substrate middle part, collecting region and read-out area, and shallow-trench isolation is by etching and being packed into
Silica is formed, with the electric signal for collecting region and read-out area to be isolated.N-type source is located in read-out area and is situated between by near-bottom
The side of matter layer is adulterated by ion implantation and is formed.N-type drain terminal is located in semiconductor substrate close to underlying dielectric layer and N
The opposite other side of type source is equally doped method by ion implantation and is formed.It should be understood that left side mentioned in this article,
Right side, top and lower section, which are only represented, is changing change with observation visual angle by the relative position under view as shown in the figure
Change, and is not understood to the limitation to specific structure.
Apply the pulse that a voltage range is negative pressure on the substrate of collecting region, or applies a voltage on the control gate
Range is the pulse of positive pressure, so that generating the depletion layer collected for photoelectron in collecting region substrate, and passes through right side read-out area
Read the photoelectron quantity collected, the input quantity as light input end.When reading, applies a positive voltage on the control gate, make N
Conducting channel is formed between type source and collecting region N-type drain terminal, then by applying a biasing arteries and veins between N-type source and N-type drain terminal
Voltage is rushed, so that the electronics in conducting channel accelerates to be formed the electric current between source and drain.The load of electric current is formed between source and drain in channel
Stream is controlled the photoelectron quantity collective effect that gate voltage, source-drain voltage and collecting region are collected, as by light input quantity
Electronics with after electric input quantity collective effect, is exported in the form of electric current, and wherein control-grid voltage, source-drain voltage can be with
As the electric input quantity of device, photoelectron quantity is then the light input quantity of device.
The Charged Couple layer of coupled zone makes depletion region in collecting region substrate start to collect for connecting collecting region and read-out area
After photoelectron, the photoelectron quantity that collecting region substrate surface gesture just will receive collection influences；By the connection of Charged Couple layer,
So that read-out area semiconductor substrate surface gesture is influenced by collecting region semiconductor substrate surface gesture, and then between influence read-out area source and drain
Size of current, to read the photoelectron quantity of collecting region collection by judging electric current between read-out area source and drain；
The control gate of carrier control zone, to apply a pulse voltage on it, so that being read in P-type semiconductor substrate
It generates in area for exciting photoelectronic depletion region out, while can also be used as electrical input, input a wherein bit arithmetic amount.
In addition, there is the underlying dielectric layer for isolation between P-type semiconductor substrate and Charged Couple layer；Charged Couple layer
Also there is the top layer dielectric layer for isolation between control gate.
Embodiment 2
As shown in figure 4, the computing unit of the present embodiment includes: as the control grid of carrier control zone, as coupling
The Charged Couple floor in area, and as the P-type semiconductor substrate of photo-generated carrier collecting region and read-out area, wherein in P type substrate
Include the N-type source formed by ion implanting and drain terminal.P-type semiconductor substrate can undertake work that is photosensitive and reading simultaneously
Make.N-type source is located at the side in read-out area close to underlying dielectric layer, is adulterated and is formed by ion implantation.N-type drain terminal position
It is same to be carried out by ion implantation close to the underlying dielectric layer other side opposite with the N-type source in semiconductor substrate
Doping method is formed.
When photosensitive, apply the pulse that a voltage range is negative pressure on P-type semiconductor substrate, while as carrier
Apply the pulse that a voltage range is positive pressure on the control grid of control zone, is received so that being generated in P type substrate for photoelectron
The depletion layer of collection is generated and is accelerated under the electric field action in the electronics in depletion region between control grid and P type substrate both ends,
And sufficiently high energy is obtained reaching, the underlying dielectric layer potential barrier across P type substrate and Charged Couple layer, into charge
Coupling layer is simultaneously stored in this, the amount of charge in Charged Couple layer, when will affect threshold value when device is opened, and then influencing to read
Source and drain between size of current；When reading, apply a pulse voltage on the control gate, makes to be formed between N-type source and N-type drain terminal and lead
Electric channel, then by applying a pulse voltage between N-type source and N-type drain terminal, so that the electronics in conducting channel accelerates shape
At the electric current between source and drain.Electric current between source and drain is controlled in gate pulse voltage, source-drain voltage and Charged Couple layer and deposits
The electron amount collective effect of storage, as by the electronics after light input quantity and electric input quantity collective effect, in the form of electric current into
Row output, wherein control-grid voltage, source-drain voltage can be used as the electric input quantity of device, the photoelectricity stored in Charged Couple layer
Subnumber amount is then the light input quantity of device.
The Charged Couple layer of coupled zone enters photoelectron therein for storing, and device threshold size when changing reading,
And then electric current between read-out area source and drain is influenced, thus by judge between read-out area source and drain electric current come generation when reading photosensitive and entering
Photoelectron quantity in Charged Couple layer.
The control gate of carrier control zone, to apply a pulse voltage on it, so that being read in P-type semiconductor substrate
It generates in area for exciting photoelectronic depletion region out, while can also be used as electrical input, input a wherein bit arithmetic amount.
In addition, there are one layer of underlying dielectric layers for isolation between P-type semiconductor substrate and Charged Couple layer；Charge coupling
It closes and also there is one layer of top layer dielectric layer for isolation between layer and control gate.
Embodiment 3
As shown in figure 5, the computing unit of the present embodiment includes: two pole of photoelectricity collected as photo-generated carrier with read-out area
Pipe and readout tube, wherein photodiode is formed by ion doping, is responsible for photosensitive.The area N of photodiode passes through as coupling
The photoelectron coupling lead for closing area is connected on the control gate of readout tube and the source of reset transistor, and the drain terminal of readout tube is applying one just
Voltage pulse, the driving voltage as read current；Before exposure, reset transistor is opened, and reset transistor drain terminal voltage is applied to photoelectricity two
In pole pipe, the photodiode as collecting region is made to be in reverse-biased, generates depletion layer；When exposure, reset transistor shutdown, photoelectricity
Diode is electrically isolating, and photoelectron is generated behind photon incidence photodiode depletion region, and accumulate in the diode, two poles
The area N of pipe with electrically by as coupled zone photoelectron couple lead connect with the area N readout tube control gate potential opening
Begin decline, and then the electron concentration in influence readout tube channel.Readout tube is responsible for reading, and drain terminal applies a positive pulse voltage,
Source is connected with addressing pipe drain terminal, when reading, is opened addressing pipe, is generated circuit current in readout tube, size of current is resetted
Pipe drain terminal voltage, readout tube drain terminal voltage and incident light subnumber joint effect, the electronics in readout tube channel, input as by light
Electronics after amount and electric input quantity collective effect, exports in the form of electric current, wherein reset transistor drain terminal voltage, readout tube drain terminal electricity
Pressure can be used as the electric input quantity of device, and electric incident light subnumber is then the light input quantity of device.
The photoelectron coupling lead of coupled zone is used to be connected to the light of collecting region in photo-generated carrier collection and read-out area
Electric diode and readout tube as read-out area, the area photodiode N potential is applied on readout tube control gate.
As the reset transistor of carrier control zone, a positive voltage is inputted by its drain terminal and acts on photodiode, when
When reset transistor is opened, positive voltage can be acted on the photodiode, and photodiode is made to generate depletion region and photosensitive, while
It can be used as electrical input, input a wherein bit arithmetic amount.
In addition, addressing pipe is used to control output of the entire arithmetic unit as the output electric current of output quantity, it can be in photoelectricity
Ranks addressing uses when computing unit forms array.
Embodiment 4
The present embodiment is calculated using the photoelectricity of multiple luminescence units and the computing unit structure composition 4*4 of embodiment 1,2 or 3
Array forms nerve and calculates nuclear structure, to simulate the nerve nucleus of human brain.
As shown in Fig. 6, the square V in figure represents photoelectricity computing unit, and photoelectricity computing array simulates the nerve in human brain
Synaptic web receives the value information of optical signal transmission and storage, completes when receiving pulse signal after storage corresponding
Operation.The grid of all computing units of every a line of photoelectricity computing array is connected, and forms the aixs cylinder in human brain nerve nucleus, aixs cylinder
The pulse excitation signal in the pulse receiver reception neural network in core is calculated by nerve.Each column of photoelectricity computing array
The source of all computing units is connected, and completes the convergence of electric current, forms the dendron in human brain nerve nucleus, dendron will be after convergence
Electric current passes to neuron.After neuron receives the electric current after dendron convergence, operation is carried out according to corresponding neural model,
Judge whether that generating pulse excitation passes to network according to operation result.
Wherein neuron models can be described in following formula:
Vj(t)=Vj(t-1)+∑ _ (i=0) ^3Ai(t)wi,j–λj
In formula, Ai(t) pulse excitation that pulse receiver receives in network when indicating some, if having received arteries and veins
Impulse is encouraged, AiIt (t) is then 1, pulse signal is by axonal transport to all cynapse points (i.e. photoelectricity calculating list of a line every in Fig. 6
Member).wi,jIndicate the weight of each photoelectricity computing unit storage in 4*4 array.If the point has weight and is not 0, then it represents that should
For point in a network there is connection relationship, which can pass through dendron biography according to corresponding weight generation electric current after receiving pulse
It passs.The summation between i=0~3 in formula is corresponding with the dendron in Fig. 6, and the electric current of all cynapse points of each column passes through tree
Superposition is completed in prominent convergence.Vj(t-1) it is the intermediate result of previous moment storage, is deposited by the digital logic portion in neuron
Storage, and with it is current when calculated result be added.λjIndicate the threshold value of each neuron, it is related to network model.Neuron
The calculated result and threshold value comparison at middle current time are gone out and current value is clear if it exceeds threshold value then sends pulse signal
Zero；If calculated result is less than threshold value, continue to keep, receives the calculating intermediate result of subsequent time.
Embodiment 5
The present embodiment proposes the multi-core network structure based on photoelectricity computing unit, the mind including multiple embodiments 1
It is computed nuclear structure, pulse processor, address arbiter and data transmission interface, each nerve calculates nuclear structure and connects a ground
Location moderator.Wherein, data transmission interface receives external input pulse and sends the internal pulse signal for needing to spread out of, with pulse
Processor is connected；Pulse processor is connected with the address arbiter that each nerve calculates core, and address arbiter and each nerve are counted
The pulse receiver for calculating core is connected.The input terminal of pulse processor receives the pulse signal and net that data transmission interface receives
Each nerve calculates the pulse signal that core transmits in network, and checks each pulse signal according to different disposal and encoded, and will compile
Pulse signal after code passes to the pulse address arbiter that nerve calculates core, so that pulse excitation signal be passed in a network
It broadcasts.The input terminal of address arbiter receives the pulse signal that pulse processor is sent, and transmits in network calculating core and receiving
When pulse signal, whether address arbiter, to the encoded information of pulse, judges the pulse by the calculating core according to pulse processor
Processing: if assessing calculations by the calculating, by the pulse be transferred to above-mentioned photoelectricity calculate array group at nerve calculating core arteries and veins
Receiver is rushed, the pulse signal is calculated into core to other of specific direction according to encoded information if not from calculating core processing
Address arbiter is sent.Data transmission interface carries out interface protocol conversion to the pulse signal that outside is transmitted, and changes into suitable for more
The pulse signal form of pulse processor and pulse processor is transferred in nuclear network；Meanwhile coffret can be by this multicore
Calculated result pulse in network carries out packing coding according to the agreement of interface, and is broadcast to other multi-core networks outward.
As shown in Fig. 7, shared in figure 4 photoelectricity as described in Example 4 calculate array groups at calculating core, calculate core
It is arranged according to web form, i.e., each nerve calculates connecting up and down with 4 other nerve calculating nuclear phases for core.Each
The pulse receiver for calculating core is connected with the network of on piece, for receiving the pulse excitation for being transferred to the calculating core in network；Often
A neuron for calculating core is equally connected with neural network, for the calculated result pulse of this core to be transmitted in network.
It include address arbiter in the impulsive neural networks of 4 cores, address arbiter uses the strategy of 2D-mesh, each
The corresponding nerve of address arbiter calculates core, for managing the input and output of the calculating core；Each address arbitration for calculating core
Device receives the pulse excitation transmitted in network, judges the encoded information of pulse excitation, will if the excitation passes to this calculating core
Pulse excitation unpacks, and excitation is passed to the pulse receiver of this calculating core；If the excitation is not handled by this calculating core, basis
Information in address code propagates the pulse excitation according to certain direction of transfer, can transmit to four direction up and down
The address arbiter of other cores is completed pulse excitation and is propagated.
Embodiment 6
The present embodiment proposes the application system and process of a kind of impulsive neural networks based on photoelectricity computing unit.
Application system includes Algorithm mapping module, weight writing module, pulse generation module and result conversion module.Its
In, what Algorithm mapping module was used to that impulsive neural networks will to be needed to handle decomposes using algorithm, makes the calculating process of algorithm
It can be mapped on impulsive neural networks；Weight writing module will be calculated according to the algorithm after the fractionation of the algorithm above mapping block
Weight needed for method is written in photoelectricity computing array according to mapping relations；Pulse generation module is according in algorithm actual operation
Excited data generates corresponding pulse excitation, and pulse excitation meets the operation requirement of impulsive neural networks, can be in pulse mind
It is propagated in network and participates in operation；As a result conversion module is used to collect the calculated result pulse letter that impulsive neural networks transmit
Number, and calculated result pulse signal is carried out by integration processing according to corresponding algorithm model, it is converted into the dominant calculating knot of algorithm
Fruit.
As shown in Fig. 8, the application flow of the impulsive neural networks based on photoelectricity computing unit is as described below:
The first step, determines application scenarios, such as the application scenarios that image recognition, speech recognition are different；
Second step, algorithm picks.According to determining applicating category and usage scenario, suitable algorithm and training test are selected
Data set.Such as the application of recognition of face class, CNN class algorithm is just chosen；
Third step, pulse network training.Network structure is built according to the algorithm type of selection, and according to training test training
Practice network model, gets a desired effect；
4th step, Algorithm mapping.Trained Algorithm mapping to existing photoelectricity is calculated into battle array by Algorithm mapping module
In the impulsive neural networks of column, it converts the operation process of algorithm to the pulse of the impulsive neural networks support of photoelectricity computing array
Form.Analog simulation is carried out to the network after mapping simultaneously, it is ensured that the network after mapping can normally complete work；
5th step, hardware bottom layer are realized.According to the mapping relations of algorithm, pulse network model is carried out by main control part
It recompilates, by information such as the corresponding weights of algorithm by weight writing module, is controlled with luminescence unit and weighted data is written
It is saved into actual pulse network；
6th step after multi-core network completes weight input, receives the external excited data transmitted, and pass through the pulse
External drive signal is converted to the manageable impulse form of multi-core network, and is transmitted in multi-core network by generation module
Pulse receiver, and then calculate core by multiple nerves in multi-core network and calculated, complete corresponding function.
7th step, after the excitation of input is carried out calculation processing by multi-core network, by result conversion module, by multicore net
The calculating impulse response of network carries out integration processing according to algorithm model, is converted into dominant calculated result and in the application system
In be shown.
Claims (4)
1. based on the impulsive neural networks of photoelectricity computing unit, including nerve calculates nuclear structure, it includes prominent that nerve, which calculates nuclear structure,
Net-fault network, dendron, aixs cylinder, pulse receiver and neuron, which is characterized in that
For the array of multiple photoelectricity computing unit periodic arrangement compositions as synaptic web, each photoelectricity computing unit includes shining
Unit and computing unit, the light that luminescence unit issues are incident in computing unit；Each computing unit include carrier control zone,
Coupled zone and photo-generated carrier collecting region and read-out area；The carrier control zone is for controlling and modulating photo-generated carrier
Carrier in collecting region and read-out area；Collecting region in the photo-generated carrier collecting region and read-out area is for absorbing the list that shines
The photon of member transmitting and the photo-generated carrier for collecting generation；The carrier control zone or photo-generated carrier collecting region and reading
Read-out area in area is connect with electric signal, and read-out area is used to export by the load after the photo-generated carrier and electric signal effect
Stream；The coupled zone connection collecting region and read-out area；
The read-out area of each column count unit is sequentially connected in array, as dendron；The read-out area output end and neuron of dendron
It is connected；
The carrier control zone of every a line computing unit is sequentially connected in array, as aixs cylinder；Carrier control zone in aixs cylinder
It is connected with pulse receiver.
2. the impulsive neural networks according to claim 1 based on photoelectricity computing unit, which is characterized in that including multiple institutes
It states nerve and calculates nuclear structure, pulse processor, address arbiter and data transmission interface, multiple nerves calculate nuclear structure in netted
Form arrangement, each nerve calculate nuclear structure and connect an address arbiter；
The pulse processor, the pulse signal of coffret input and each nerve calculate nuclear structure for receiving data
The pulse signal that aixs cylinder transmits, and coding is carried out according to each pulse signal of different calculation and check and believes the pulse after coding
Number pass to address arbiter；
The address arbiter, for receiving the pulse signal of pulse processor transmission, and according to the encoded information of pulse signal
By pulse signal transmission to the address arbiter of pulse receiver or neighborhood calculation core；
The data transmission interface for receiving external pulse signal, and external pulse signal is converted into be suitable for arteries and veins
It rushes the pulse signal form of processor and is transferred to pulse processor.
3. the operation method of the impulsive neural networks as described in claim 1 based on photoelectricity computing unit, which is characterized in that specific
Step includes:
Luminescence unit is calculated in the weighted value input synaptic web of nuclear structure and is saved under the driving of driver, by nerve；
Pulse receiver is received from external pulse excitation, caches pulse excitation, and receiving impulsive synchronization control signal
The pulse excitation of caching is passed through into axonal transport to the carrier control zone of every a line computing unit later；
The weighted value stored in pulse excitation that carrier control zone receives and synaptic web is subjected to multiplying, Mei Geji
The calculated result for calculating unit is connected by dendron, to converge to the electric current of the read-out area output end of each column count unit
It is poly-, and export to neuron；
Current signal after convergence is AD converted into digital signal by neuron, and constantly the pulse excitation of input is generated
Digital signal result after calculating current conversion adds up, while accumulated result being brought into the mind being made of Digital Logic
In meta-model function, judged whether to generate output pulse signal.
4. the application system of the impulsive neural networks as claimed in claim 2 based on photoelectricity computing unit, which is characterized in that including
The impulsive neural networks and:
Algorithm mapping module, for decomposing using algorithm for the impulsive neural networks processing will to be needed；
Weight writing module, based on the weight needed for by the algorithm after decomposing is written to by luminescence unit according to mapping relations
It calculates in unit；
Pulse generation module for generating corresponding pulse excitation according to the excited data in algorithm actual operation, and inputs
To pulse receiver；
As a result conversion module, the calculated result pulse signal that coffret transmits for receiving data, and according to corresponding algorithm
Calculated result pulse signal is carried out integration processing by model, is converted into the dominant calculated result of algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910415826.7A CN110263926B (en) | 2019-05-18 | 2019-05-18 | Pulse neural network based on photoelectric computing unit, system and operation method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910415826.7A CN110263926B (en) | 2019-05-18 | 2019-05-18 | Pulse neural network based on photoelectric computing unit, system and operation method thereof |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110263926A true CN110263926A (en) | 2019-09-20 |
| CN110263926B CN110263926B (en) | 2023-03-24 |
Family
ID=67913384
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910415826.7A Active CN110263926B (en) | 2019-05-18 | 2019-05-18 | Pulse neural network based on photoelectric computing unit, system and operation method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110263926B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111309670A (en) * | 2020-02-18 | 2020-06-19 | 电子科技大学 | CMOS (complementary Metal oxide semiconductor) inductive storage and calculation integrated circuit structure |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140114893A1 (en) * | 2011-05-31 | 2014-04-24 | International Business Machines Corporation | Low-power event-driven neural computing architecture in neural networks |
| CN105095967A (en) * | 2015-07-16 | 2015-11-25 | 清华大学 | Multi-mode neural morphological network core |
| CN107333040A (en) * | 2017-07-13 | 2017-11-07 | 中国科学院半导体研究所 | Bionical visual imaging and processing unit |
-
2019
- 2019-05-18 CN CN201910415826.7A patent/CN110263926B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140114893A1 (en) * | 2011-05-31 | 2014-04-24 | International Business Machines Corporation | Low-power event-driven neural computing architecture in neural networks |
| CN105095967A (en) * | 2015-07-16 | 2015-11-25 | 清华大学 | Multi-mode neural morphological network core |
| CN107333040A (en) * | 2017-07-13 | 2017-11-07 | 中国科学院半导体研究所 | Bionical visual imaging and processing unit |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111309670A (en) * | 2020-02-18 | 2020-06-19 | 电子科技大学 | CMOS (complementary Metal oxide semiconductor) inductive storage and calculation integrated circuit structure |
| CN111309670B (en) * | 2020-02-18 | 2023-05-05 | 电子科技大学 | CMOS sense-memory-calculation integrated circuit structure |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110263926B (en) | 2023-03-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TWI774147B (en) | Pulse convolutional neural network algorithm and related integrated circuits and method of manufacture thereof, computing devices and storage media | |
| Xiong et al. | Parameter extraction of solar photovoltaic models using an improved whale optimization algorithm | |
| Meng et al. | An intelligent hybrid wavelet-adversarial deep model for accurate prediction of solar power generation | |
| CN108256636A (en) | A kind of convolutional neural networks algorithm design implementation method based on Heterogeneous Computing | |
| CN105354646B (en) | Power load forecasting method for hybrid particle swarm optimization and extreme learning machine | |
| CN109410575A (en) | A kind of road network trend prediction method based on capsule network and the long Memory Neural Networks in short-term of nested type | |
| CN109784483A (en) | Computation accelerator in binaryzation convolutional neural networks memory based on FD-SOI technique | |
| CN106126863B (en) | Photovoltaic cell parameter identification method based on artificial fish-swarm and the algorithm that leapfrogs | |
| CN109460819B (en) | Method and device for simulating optical synapse of organism | |
| CN105046325A (en) | Circuit simulating biological neural network based on similar MOS luminescent devices | |
| CN105654136A (en) | Deep learning based automatic target identification method for large-scale remote sensing images | |
| CN110263295A (en) | A kind of operation optimization method of the matrix-vector multiplier based on photoelectricity computing array | |
| CN110009102A (en) | A kind of accelerated method of the depth residual error network based on photoelectricity computing array | |
| CN110263926A (en) | Impulsive neural networks and its system and operation method based on photoelectricity computing unit | |
| Mirmahaleh et al. | DNN pruning and mapping on NoC-Based communication infrastructure | |
| Yang et al. | Research on subway pedestrian detection algorithms based on SSD model | |
| Bakır | Comparative performance analysis of metaheuristic search algorithms in parameter extraction for various solar cell models | |
| CN109993283A (en) | The accelerated method of depth convolution production confrontation network based on photoelectricity computing array | |
| CN110244817A (en) | A kind of Solving Partial Differential Equations device and its method based on photoelectricity computing array | |
| Ge et al. | A virtual data collection model of distributed PVs considering spatio-temporal coupling and affine optimization reference | |
| Junior et al. | Filter-free color pixel sensor using gated PIN photodiodes and machine learning techniques | |
| CN111667064B (en) | Hybrid neural network based on photoelectric computing unit and operation method thereof | |
| CN116110232A (en) | Traffic flow prediction method based on hierarchical dynamic residual map convolution network | |
| CN109033603A (en) | Secondary system of intelligent substation emulation mode based on source stream path chain | |
| CN106952215A (en) | A kind of image pyramid characteristic extracting circuit, device and method |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240409 Address after: 210000 two, B unit 300, Zhihui Road, Kirin science and Technology Innovation Park, Jiangning District, Nanjing, Jiangsu. Patentee after: NANJING JIXIANG SENSING IMAGING TECHNOLOGY RESEARCH INSTITUTE Co.,Ltd. Country or region after: China Address before: 9/F, Building B, No. 100, Tianjiao Road, Qilin Hi-tech Industrial Development Zone, Jiangning District, Nanjing, Jiangsu, 210000 Patentee before: Nanjing Weixin Photoelectric System Co.,Ltd. Country or region before: China |