CN113867474A - Optical computing device, computing method and computing system - Google Patents
Optical computing device, computing method and computing system Download PDFInfo
- Publication number
- CN113867474A CN113867474A CN202010623939.9A CN202010623939A CN113867474A CN 113867474 A CN113867474 A CN 113867474A CN 202010623939 A CN202010623939 A CN 202010623939A CN 113867474 A CN113867474 A CN 113867474A
- Authority
- CN
- China
- Prior art keywords
- signals
- feedback
- matrix
- feedback signals
- unit
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06E—OPTICAL COMPUTING DEVICES; COMPUTING DEVICES USING OTHER RADIATIONS WITH SIMILAR PROPERTIES
- G06E3/00—Devices not provided for in group G06E1/00, e.g. for processing analogue or hybrid data
- G06E3/008—Matrix or vector computation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Optical Communication System (AREA)
- Optical Modulation, Optical Deflection, Nonlinear Optics, Optical Demodulation, Optical Logic Elements (AREA)
Abstract
A light computing device includes a management unit, a first Italic unit, and a second Italic unit. The management unit is connected with the first Yixin unit and the second Yixin unit. The first and second Isci units receive a first set of signals. The first IshCi unit generates a first set of feedback signals based on the first set of signals and the first problem sub-matrix. The second IshCi unit generates a second group of feedback signals according to the first group of signals and a second problem sub-matrix, wherein the first problem sub-matrix and the second problem sub-matrix are sub-matrices of the problem matrix respectively, and the problem matrix indicates first data to be calculated. The management unit receives a first plurality of sets of feedback signals including a first set of feedback signals and a second set of feedback signals, and generates a first target feedback signal according to the first plurality of sets of feedback signals. The optical computing equipment provided by the application can increase the operation efficiency by adopting a mode of parallel operation of a plurality of Italian units.
Description
Technical Field
The present application relates to the field of information technology, and in particular, to an optical computing device, an optical computing method, and an optical computing system.
Background
The Esin model describes a complex system comprising a large number of spin nodes, each spin node has a spin state of +1 and-1, in the system, the spin nodes have interaction, the interaction between the spin nodes can change the spin states of the spin nodes, and based on the interaction between the spin nodes, the Esin model can gradually realize an annealing process, namely the Hamming metric of the system is gradually reduced until convergence. The combinatorial optimization problem can be changed into an Esinc model through conversion, and parameters in the combinatorial optimization problem are represented by utilizing the interaction between the spin nodes.
The light Exin machine utilizes physical phenomenon simulation to solve the Exin model. The optical Eschen machine utilizes interference between multiple optical signals to simulate the interaction between spin nodes. The phase of the optical signal can represent the spin state of the spin node in the Esin model, and the interference effect between the optical signals can realize the change of the phase of the optical signal, so that the change of the spin state of the spin node in the Esin model is simulated.
When the number of spin nodes in the Itanium model is increased, the number of optical signals in the Itanium machine needs to be increased when the Itanium machine is used for simulating and solving the Itanium model, but the number of optical signals which can be introduced at one time by the Itanium machine constructed in an on-chip integration mode is limited at present, and the Itanium model with a large number of spin nodes cannot be simulated and solved.
Disclosure of Invention
The application provides an optical computing device, a computing method and a computing system, which are used for enabling an optical Esino machine to solve an Esino model with a large number of spinning nodes.
In a first aspect, the present application provides a light computing device comprising a management unit, a first inching unit and a second inching unit. The management unit is connected with the first Yixin unit and the second Yixin unit. A plurality of isooctyl units (e.g., comprising m units, m being a positive integer not less than 2) can be included in the optical computing device, and only the first and second unit included in the plurality of unit are taken as an example for illustration.
The first and second incarnation units may receive the first set of signals simultaneously. The first IshCi unit generates a first set of feedback signals based on the first set of signals and the first problem sub-matrix. The second IshCi unit generates a second group of feedback signals according to the first group of signals and a second problem sub-matrix, wherein the first problem sub-matrix and the second problem sub-matrix are different sub-matrices of the problem matrix respectively, and the problem matrix is used for indicating first data to be calculated.
Each Itanium unit can output a group of feedback signals; the management unit may receive a first plurality of sets of feedback signals output by each of the plurality of yixing units, where the first plurality of sets of feedback signals include a first set of feedback signals and a second set of feedback signals, and generate a first target feedback signal according to the first plurality of sets of feedback signals. The first target feedback signal is to instruct the light computing device to perform a first intermediate result of the incarnation calculation on the first data.
In the optical computing equipment provided by the application, the first Italic unit and the second Italic unit can run in parallel, a group of feedback signals are generated based on the received first group of signals and the configured problem submatrix, when a plurality of Italic units are operated in parallel, more optical signals can be received, the simulation solution of the Italic model with more spin nodes is supported, and meanwhile, the operation efficiency can be increased.
In one possible design, the problem matrix may include a third problem sub-matrix and a fourth problem sub-matrix in addition to the first problem sub-matrix and the second problem sub-matrix. The third problem sub-matrix and the fourth problem sub-matrix are different from the first problem sub-matrix and the second problem sub-matrix.
The first and second Isci units receive the first set of signals simultaneously. The first IshCi unit may obtain a third set of feedback signals according to the first set of signals and the third problem sub-matrix; the second Ishig generates a fourth group of feedback signals according to the first group of signals and the fourth problem submatrix; wherein the third problem sub-matrix and the fourth problem sub-matrix are different sub-matrices of the problem matrix.
The management unit receives a second plurality of groups of feedback signals including a third group of feedback signals and a fourth group of feedback signals, and generates a second target feedback signal according to the second plurality of groups of feedback signals. The second target feedback signal is to instruct the light computing device to perform a second intermediate result of the inching calculation on the first data.
In the optical computing device provided by the present application, the problem matrix is decomposed into a plurality of groups of problem sub-matrices, and for any group of problem sub-matrices, a plurality of problem sub-matrices (e.g., a first problem sub-matrix and a second problem sub-matrix, a third problem sub-matrix and a fourth problem sub-matrix) may be respectively configured in a plurality of yixing units in advance, and then the plurality of yixing units generate a plurality of groups of feedback signals in parallel based on the received first group of signals and the configured problem sub-matrices. For the correlation operation of a problem matrix, the method can be decomposed into a plurality of operation processes. For example, the calculation of the plurality of problem sub-matrices including the first problem sub-matrix and the second problem sub-matrix by the plurality of iton units is one operation, and the calculation of the plurality of problem sub-matrices including the third problem sub-matrix and the fourth problem sub-matrix by the plurality of iton units is another operation. Correspondingly, multiple parallel operations can be realized by multiple IshIn units based on other problem sub-matrixes. According to the method, the time-sharing multiplexing of the Itanium unit is realized in the process of operating a problem matrix. The optical computing device performs the plurality of operations to complete a round of operation. According to the mode, in the process of carrying out Yixin calculation on data to be calculated, not only a plurality of Yixin units are adopted for carrying out parallel calculation, but also the plurality of Yixin units can be subjected to related calculation in a time-sharing calculation mode (one calculation process is completed in one time period, and the other calculation process is completed in the other time period), so that the utilization rate of the Yixin units can be improved, the number of spinning nodes in an Yixin model for simulation solution is effectively increased, and the calculation efficiency of optical calculation equipment is increased.
In a possible design, after a round of operation is performed, the optical computing device may further continue to perform a plurality of operation processes (i.e. perform the next round of operation process) based on the target feedback signal generated in the round of operation process, and the following description will be given by taking a first operation process in the plurality of operation processes that are continued later as an example: the management unit may send the first target feedback signal to the first itu unit and send the second target feedback signal to the second itu unit.
The first yixin unit may obtain a fifth set of feedback signals according to the first set of signals, the first target feedback signal, and the first problem submatrix; the second Italic unit may generate a sixth set of feedback signals from the first set of signals, the second target feedback signal, and the second problem sub-matrix.
The management unit may receive a third plurality of sets of feedback signals including a fifth set of feedback signals and a sixth set of feedback signals, and generate a third target feedback signal from the third plurality of sets of feedback signals. The third target feedback signal is to instruct the light computing device to perform a third intermediate result of the inching calculation on the first data.
In the optical computing device provided by the application, the optical computing device completes one round of operation process every time the optical computing device completes multiple operation processes (such as m operation processes), and the optical computing device can perform multiple rounds of operation processes to realize iteration of the multiple rounds of operation processes, so that the Itanium model can be accurately solved.
In one possible design, the management unit may decompose the problem matrix, determine a plurality of sets of problem sub-matrices from the problem matrix, where each set of problem sub-matrices includes a plurality of problem sub-matrices, such as a first problem sub-matrix and a second problem sub-matrix, or includes a third problem sub-matrix and a fourth problem sub-matrix, and the management unit may configure the plurality of problem sub-matrices in a set in a plurality of yixin units, respectively.
In the optical computing device provided by the application, the management unit can realize the decomposition of the problem matrix, so that the optical computing device can simulate and solve the Esino model with more nodes.
In one possible design, taking the first isooctyl unit as an example, the structures of the first isooctyl unit and the second isooctyl unit are explained: the first IshCi unit comprises a spin signal generation module and a feedback calculation module. The spin signal generation module can generate a set of spin signals and the feedback computation module can generate a set of feedback signals from the set of spin signals.
When the first IshC unit generates the first set of feedback signals, the spin signal generation module may obtain a first set of spin signals according to the first set of signals; the feedback computation module may generate a first set of feedback signals from the first set of spin signals and the first problem submatrix.
When the first IshC unit generates the third set of feedback signals, the spin signal generation module may obtain a second set of spin signals according to the first set of signals; the feedback calculation module may generate a third set of feedback signals based on the second set of spin signals and the third problem submatrix.
When the first IshC unit generates the fifth set of feedback signals, the spin signal generation module may obtain a third set of spin signals according to the first set of signals and the first target feedback signal; the feedback calculation module may generate a fifth set of feedback signals based on the third set of spin signals and the first problem sub-matrix.
In the light calculation device provided by the application, the spin signal generation module and the feedback calculation module in the first incarnation unit can cooperate to generate a set of spin signals and a set of feedback signals, so that the generation process of the feedback signals can be simplified.
In one possible design, the spin signal generating module is capable of generating spin signals, and the structure of the spin signal generating module is not limited herein, for example, the spin signal generating module includes a phase modulator array and an intensity modulator array. The spin signal generation module may intensity modulate the first set of signals before phase modulating. The intensity modulator array carries out intensity modulation on the first group of signals to obtain a first group of modulation signals; then, the phase modulator array performs phase modulation on the first group of modulation signals to obtain a first group of spin signals. The spin signal generation module may also perform phase modulation on the first set of signals before performing intensity modulation. The phase modulator array performs phase modulation on the first group of signals to obtain a second group of modulation signals; the intensity modulator array then intensity modulates the second set of modulation signals to obtain a first set of spin signals.
In the light computing device provided by the application, the spin signal generation module comprises a phase modulator array and an intensity modulator array, and the phase and the intensity of the signal can be adjusted in a targeted mode.
In one possible design, the intensity modulator array includes a plurality of intensity modulators, which may be mach-zehnder interferometers, or electro-absorption modulators, to suit different application scenarios.
In one possible design, the feedback calculation module includes a plurality of mach-zehnder interferometers so that the problem sub-matrix (e.g., the first problem sub-matrix or the third problem sub-matrix) may be better loaded.
In one possible design, the management unit may calculate a hamiltonian according to the obtained multiple target feedback signals after the optical computing device completes each round of the operation process, and the hamiltonian is used to represent the system energy corresponding to the first data. When the hamilton is no longer reduced, i.e., the hamilton converges, the optical computing device stops the operation, and the management unit may obtain the calculation result of the first data from the plurality of sets of spin signals generated in the respective itu units. When the Hamiltonian does not converge, the optical computing device continues to operate, after which the Hamiltonian converges.
In a second aspect, the present application provides a calculation method, and beneficial effects may refer to related descriptions of the first aspect, which are not described herein again. The method is performed by an optical computing device comprising a management unit, a first inching unit and a second inching unit, the management unit connecting the first inching unit and the second inching unit.
The first Italic unit may obtain a first set of feedback signals according to the first set of signals and the first problem sub-matrix.
The second Italic unit may generate a second set of feedback signals according to the first set of signals and a second problem submatrix, where the first problem submatrix and the second problem submatrix are different submatrices of the problem matrix, and the problem matrix is used to indicate the first data to be calculated.
The management unit receives a first plurality of sets of feedback signals including a first set of feedback signals and a second set of feedback signals, and generates a first target feedback signal according to the first plurality of sets of feedback signals.
In one possible design, the first itu unit may obtain a third set of feedback signals based on the first set of signals and the third problem sub-matrix;
the second IshCi unit may generate a fourth set of feedback signals according to the first set of signals and the fourth problem sub-matrix; wherein the third problem sub-matrix and the fourth problem sub-matrix are different sub-matrices of the problem matrix;
the management unit may receive a second plurality of sets of feedback signals including a third set of feedback signals and a fourth set of feedback signals, and generate a second target feedback signal from the second plurality of sets of feedback signals.
In one possible design, the management unit may send a first target feedback signal to the first inching unit and a second target feedback signal to the second inching unit.
The first Italic unit may obtain a fifth set of feedback signals according to the first set of signals, the first target feedback signal, and the first problem submatrix.
The second Italic unit may generate a sixth set of feedback signals from the first set of signals, the second target feedback signal, and the second problem sub-matrix.
The management unit may receive a third plurality of sets of feedback signals including a fifth set of feedback signals and a sixth set of feedback signals, and generate a third target feedback signal from the third plurality of sets of feedback signals.
In a third aspect, the present application provides a computing system that may include an optical computing device as in the first aspect or any one of the possible implementations of the first aspect, and a laser for transmitting a first set of signals to the optical computing device.
In a fourth aspect, the present application further provides a computer program product, which includes program code, where the program code includes instructions to be executed by a computer to implement the computing method in the second aspect or any one implementation manner of the second aspect.
In a fifth aspect, the present application further provides a computer-readable storage medium for storing program code, where the program code includes instructions to be executed by a computer to implement the computing method in any one of the implementations of the foregoing second aspect or second aspect.
Drawings
FIG. 1 is a schematic diagram of a light computing device according to the present application;
FIG. 2 is an exploded view of a problem sub-matrix provided herein;
FIG. 3 is a schematic structural diagram of a spin signal generating module provided herein;
fig. 4 is a schematic structural diagram of a feedback calculation module provided in the present application;
FIG. 5 is a schematic diagram of an operation process provided herein;
FIG. 6 is a schematic diagram of a round of computation provided herein;
FIG. 7 is a schematic diagram of a computing system provided herein;
fig. 8 is a schematic diagram of a calculation method provided in the present application.
Detailed Description
As shown in fig. 1, for a light computing device 10 provided in an embodiment of the present application, the light computing device 10 includes a management unit 100, m ircin units 200, where m is an integer not less than 2. It should be noted that, in the embodiment of the present invention, the optical computing device 10 may be in the form of a circuit or a chip, and a specific implementation form of the optical computing device 10 is not limited herein. As shown in fig. 1, the management unit 100 can decompose the problem matrix, and can decompose the problem matrix into m sets of problem sub-matrices (in the embodiment of the present application, the problem sub-matrices are sub-matrices of the problem matrix), where each set of problem sub-matrices includes m problem sub-matrices, and for any set of problem sub-matrices, the management unit 100 can configure the m problem sub-matrices of the set of problem sub-matrices in m ircin units 200, respectively. Each of the inct cells 200 is configured with one of the set of question sub-matrices.
For any of the m Italic units 200, a set of feedback signals may be generated from the received first set of signals and the configured problem sub-matrix. Specifically, the yixin unit 200 may generate a set of spin signals according to the first set of signals (and the target feedback signal), and then may also generate a set of feedback signals using the configured problem submatrix and the set of spin signals.
The management unit 100 may also receive a set of feedback signals output in each of the plurality of converter units 200. The management unit 100 may generate a target feedback signal from the acquired m sets of feedback signals. When the management unit 100 generates m target feedback signals, the m target feedback signals may be respectively transmitted to the m yixin units 200. Each of the itu units 200 receives one of the m target feedback signals.
Specifically, inside any of the Isci units 200, each Isci unit 200 includes a spin signal generation module 210 and a feedback calculation module 220.
When the management unit 100 configures m problem sub-matrices of a set of problem sub-matrices in m inching units 200, respectively, the m problem sub-matrices of the set of problem sub-matrices may be configured on the feedback calculation module 220 in the m inching units 200, respectively. Each feedback calculation module 220 configures one of the set of problem sub-matrices.
When the management unit 100 sends the m target feedback signals to the m inching units 200, respectively, the m target feedback signals may be sent to the spin signal generating modules 210 in each inching unit 200, respectively, and each spin signal generating module 210 receives one target feedback signal of the m target feedback signals.
The spin signal generation module 210 is configured to generate spin signals, and the spin signal generation module 210 may generate a set of spin signals according to the received first set of signals, and the set of spin signals may be transmitted to the feedback calculation module 220 in the inching unit 200.
It should be noted that, when the spin signal generation module 210 receives the target feedback signal, the spin signal generation module 210 may generate a set of spin signals according to the received first set of signals and the target feedback signal when generating a set of spin signals.
The feedback computation module 220 in the IshX unit 200 may generate a set of feedback signals based on the received set of spin signals and the configured problem submatrix.
In the embodiment of the present application, the problem sub-matrices simultaneously configured on the m feedback calculation modules 220 are a set of problem sub-matrices after the problem matrix decomposition. For convenience of description, the process that the management unit 100 configures a group of problem sub-matrices for the m feedback calculation modules 220 and the m irxin units 200 operate to generate m groups of feedback signals is called a one-time operation process. The management unit 100 configures m sets of problem sub-matrices (i.e., configures a problem matrix) for the m feedback calculation modules 220, and the m ircin units 200 perform m operation processes in the process of generating m × m sets of feedback signals, where the m operation processes are referred to as a round of operation processes. In each calculation process, the management unit 100 receives a set of feedback signals (m sets of feedback signals) output by each of the yixin units 200, and may generate one target feedback signal according to the m sets of feedback signals, and when the optical calculation device 10 performs m calculation processes, the management unit 100 may generate m target feedback signals in total, and the management unit 100 may transmit the m target feedback signals to the m yixin units 200 (the spin signal generation module 210). Then, the optical computing device 10 may further perform m operation processes, that is, perform one round of operation process, so that m target feedback signals generated in each round of operation process are sent to m ircin units 200 (spin signal generating module 210) to perform the next round of operation process, thereby implementing iteration of multiple rounds of operation processes.
The following describes the iteration of one operation process, and multiple operation processes.
(1) A one-time operation process
The management unit 100 may decompose the problem matrix first, and the way in which the management unit 100 decomposes the problem matrix is not limited herein, and any way that the problem matrix can be decomposed into m sets of problem sub-matrices is suitable for the embodiments of the present application. The embodiment of the present application does not limit the structure of the management unit 100, for example, the management unit 100 may be a Field Programmable Gate Array (FPGA) or a Digital Signal Processing (DSP).
The problem matrix is used for indicating first data to be calculated, the first data can also be understood as an Eschen model converted from a combinatorial optimization problem, the problem matrix can be obtained by performing data simulation extraction on the combinatorial optimization problem, each element in the matrix represents interaction between different spin nodes in the Eschen model, and the problem matrix can be a symmetric matrix generally.
As shown in fig. 2, for a problem matrix decomposition diagram provided in the embodiment of the present application, the management unit 100 may decompose the problem matrix into m × m problem sub-matrices Jn*nEach problem submatrix Jn*nThe problem sub-matrix is divided into a group of problem sub-matrices. In fig. 2, each column of problem sub-matrices is grouped into a group of problem sub-matrices in units of columns. The management unit 100 may sequentially allocate the set of problem sub-matrices to the m ircin units 200 (feedback calculation module 220).
In one operation, each of the incarnation units 200 can simultaneously receive a first group of signals, where the first group of signals includes at least one signal, and in the embodiment of the present application, the first group of signals includes n signals as an example for illustration. The embodiment of the present application does not limit the type of the signal in the first group of signals, and may be an optical signal or an electrical signal. Each of the ircin units 200 generates a set of feedback signals based on the received first set of signals and the configured problem sub-matrix.
Taking the example that the management unit 100 configures the first set of problem sub-matrices for the multiple yixin units 200, a description will be given of a single operation process involving a first yixin unit and a second yixin unit in the multiple yixin units 200, where the management unit 100 configures a first problem sub-matrix in the first set of problem sub-matrices in the first yixin unit, and the management unit 100 configures a second problem sub-matrix in the first set of problem sub-matrices in the second yixin unit.
The first Italic unit may obtain a first set of feedback signals according to the first set of signals and the first problem submatrix; the second Italic unit may generate a second set of feedback signals based on the first set of signals and the second problem sub-matrix.
The management unit 100 receives a plurality of sets of feedback signals (for convenience of distinction, the plurality of sets of feedback signals are referred to as a first plurality of sets of feedback signals) output by the plurality of yixin units 200, wherein the first plurality of sets of feedback signals include a first set of feedback signals and a second set of feedback signals, and the management unit 100 generates a first target feedback signal according to the first plurality of sets of feedback signals.
In the optical computing device shown in fig. 1, each of the inching units 200 outputs a set of feedback signals, and the management unit 100 may receive a set of feedback signals output by m inching units 200, and may receive m sets of feedback signals, where the m sets of feedback signals are first sets of feedback signals, and may generate a first target feedback signal according to the m sets of feedback signals, where the first target feedback signal is a first intermediate result of the optical computing device performing an inching calculation on first data.
For any of the Isci units 200, the spin signal generation module 210 in the Isci unit 200 can adjust the phase and intensity of each of the first set of signals, outputting a set of spin signals. The set of spin signals includes n spin signals.
In the embodiment of the present application, the way in which the spin signal generation module 210 adjusts the phase and intensity of each signal in the first group of signals is not limited, and the way in which the spin signal generation module 210 adjusts the phase and intensity of each signal in the first group of signals is also different for different types of signals, and the following description will be given by taking the signals in the first group of signals as optical signals (such as optical pulses) as an example, and the way in which the spin signal generation module 210 adjusts the phase and intensity of each signal in the first group of signals is described:
referring to fig. 3, a schematic structural diagram of a spin signal generation module 210 provided in an embodiment of the present application is shown, where the spin signal generation module 210 includes an intensity modulator array 211 and a phase modulator array 212.
The intensity modulator array 211 includes a plurality of intensity modulators capable of intensity modulating the received signal. The phase modulator array 212 includes a plurality of phase modulators capable of phase modulating the received signal.
The intensity modulator may be a mach-zehnder interferometer (MZI), an electro-absorption modulator (EAM), a Semiconductor Optical Amplifier (SOA), or a Variable Optical Attenuator (VOA), among others. The phase modulator may be a waveguide.
The embodiment of the present application does not limit the sequence of intensity modulation and phase modulation on the first group of signals, and may perform intensity modulation first and then perform phase modulation (the intensity modulator array 211 receives the first group of signals first, performs intensity modulation on the first group of signals, and outputs a group of modulation signals, and the phase modulator array 212 performs phase modulation on a group of modulation signals output by the intensity modulator array 211, and outputs a group of spin signals); or, the phase modulation may be performed first, and then the intensity modulation may be performed (the phase modulator array 212 receives the first group of signals first, performs the phase modulation on the first group of signals, and the intensity modulator array 211 performs the intensity modulation on the group of signals output by the phase modulator array 212, and outputs a group of spin signals).
If the signals in the first set of signals are electrical signals, the spin signal generating module 210 may be an FPGA or a DSP, so as to modulate the intensity of the electrical signals.
The structure of the spin signal generating module 210 in the above description is only an example, and the embodiment of the present application is not limited to the structure of the spin signal generating module 210, and any module capable of generating a set of spin signals according to the first set of signals is applicable to the embodiment of the present application.
After the spin signal generation module 210 outputs a set of spin signals. The feedback computation module 220 receives the set of spin signals and generates a set of feedback signals based on the configured problem submatrix and the set of spin signals.
In the embodiment of the present application, the way in which the feedback calculation module 220 generates a set of feedback signals based on the problem submatrix and a set of spin signals is not limited, and the way in which the feedback calculation module 220 generates a set of feedback signals based on the problem submatrix and a set of spin signals is also different for different types of signals, and the following description will be given by taking signals in a set of spin signals as optical signals, and the way in which the feedback calculation module 220 generates a set of feedback signals based on the problem submatrix and a set of spin signals is described.
Referring to fig. 4, which is a schematic structural diagram of a feedback calculation module 220 provided in an embodiment of the present application, the feedback calculation module 220 includes an interaction matrix 221 and a photodetector array 222.
The interaction matrix 221 includes a plurality of Mach-Zehnder interferometers (MZIs), the number of MZIs and the problem sub-matrix Jn*nThe number of elements in the interaction matrix 221 includes n in the embodiment of the present application2Mach Zehnder interferometer using n2Problem sub-matrix J constructed by Mach Zehnder interferometern*n. The phase modulator cell 213 on each mach-zender interferometer may be implemented using thin film lithium niobate, and the thin film lithium niobate and the transmission waveguide are integrated by a hetero-integration technique. J realized by using electro-optical characteristic of thin-film lithium niobaten*nThe problem submatrix J can be realized by loading and utilizing a transmission waveguiden*nTime-sharing refresh. n is2Problem sub-matrix J constructed by Mach Zehnder interferometern*nThen, the problem submatrix J may be constructedn*nDecomposed into a unitary matrix U, a diagonal matrix sigma and a transposed unitary matrix U, and after a set of spin signals (which may be considered as column vectors) is transmitted through the interaction matrix 221, a multiplication of the matrix vectors is completed and a set of feedback optical signals is generated. The group of feedback optical signals includes n feedback optical signals.
The photodetector array 222 includes n photodetectors, and the photodetectors may convert optical signals into electrical signals, and a group of feedback optical signals passes through the photodetector array 222 to implement photoelectric conversion, and output a group of feedback signals, where the group of feedback signals includes n feedback signals.
Referring to fig. 5, taking the example that the management unit 100 configures the first set of problem submatrices for the plurality of inct units 200, a set of feedback signals generated by the feedback calculation module 220 in the first inct unit and the feedback calculation module 220 in the second inct unit will be described, where the first set of spin signals output by the spin signal generation module 210 in the first inct unit is { σ }1，σ2The first problem sub-matrix loaded by the feedback calculation module 220 in the first Italic unit is
The first set of feedback signals is output as
The second group of spin signals output by the spin signal generation module 210 in the second IshCi cell is { σ3，σ4A second problem sub-matrix loaded by the feedback calculation module 220 in the second Italic unit is
The second set of feedback signals is output as
using JN*NIs a characteristic of a symmetric matrix (J)ij＝Jji) The above formula can be converted as follows:
the management unit 100 acquires m sets of feedback signals, sums the m sets of feedback signals, and generates a first target feedback signal.
Still taking the first set of feedback signals and the second set of feedback signals shown in fig. 5 as an example for explanation, the management unit 100 is right
And
summing to generate a first target feedback signal
Where N is the total number m × N of spin signals included in the m sets of spin signals output by the m spin signal generation modules 210, in the above example, it can be considered that m and N are both 2, and N is 4.
Based on the foregoing reasoning, any target feedback signal is a signal sequence including n signals, and the jth signal f in the target feedback signaljThe following were used:
wherein N is the total number m × N, σ of spin signals included in the m sets of spin signals output by the m spin signal generation modules 210iThe ith spin signal of all spin signals is included in the m groups of spin signals.
From the above, one objective feedback signal is the result of the operation of the spin signal and a set of problem submatrices in the problem matrix. In the embodiment of the present application, the first target feedback signal is an operation result of the spin signals generated by the m ircin units and the first set of problem submatrices during the operation.
(2) A round of operation process
In one operation process, only m problem sub-matrices in the problem matrix are loaded in the m feedback calculation modules 220, where the time for the optical calculation apparatus 10 to perform one operation process is t, and the time for the optical calculation apparatus 10 to perform the first operation process is t 1. m feedback calculation modules 220 load m × m problem sub-matrices in the problem matrix, and the required time is m × t.
The following describes a calculation procedure in the remaining m-1 calculation procedures in a round of calculation procedure, with the aforementioned calculation procedure as the first calculation procedure:
the management unit 100 respectively configures m problem sub-matrixes in another group of problem sub-matrixes in m Italic units 200, and each Italic unit 200 is configured with one problem sub-matrix;
during this operation, each of the i-cin units 200 receives the first set of signals at the same time, and each of the i-cin units 200 generates a set of feedback signals according to the received first set of signals and the configured problem submatrix.
Taking the example that the management unit 100 configures the second group of problem sub-matrices for the multiple ecin units 200, a description will be given of a primary operation process in which a first ecin unit and a second ecin unit of the multiple ecin units 200 participate, where the management unit 100 configures a third problem sub-matrix of the second group of problem sub-matrices in the first ecin unit, and the management unit 100 configures a fourth problem sub-matrix of the second group of problem sub-matrices in the second ecin unit. The second set of problem sub-matrices are different from the first set of problem sub-matrices, i.e. the second problem sub-matrix and the first problem sub-matrix, and the third problem sub-matrix and the fourth problem sub-matrix are different sub-matrices in the problem matrix.
The first IshCi unit may obtain a third set of feedback signals according to the first set of signals and the third problem sub-matrix; the second itu unit may generate a fourth set of feedback signals based on the first set of signals and the fourth problem sub-matrix.
The management unit 100 receives a plurality of sets of feedback signals (for convenience of distinction, the plurality of sets of feedback signals are referred to as a second plurality of sets of feedback signals) output by the plurality of yixin units 200, wherein the second plurality of sets of feedback signals include a third set of feedback signals and a fourth set of feedback signals, and the management unit 100 generates a second target feedback signal according to the second plurality of sets of feedback signals.
In the optical computing device shown in fig. 1, each of the inching units 200 outputs a set of feedback signals, and the management unit 100 may receive a set of feedback signals output by m inching units 200, and may receive m sets of feedback signals in total, where the m sets of feedback signals are second sets of feedback signals, and may generate a second target feedback signal according to the m sets of feedback signals, and the second target feedback signal is a second intermediate result of the optical computing device performing an inching calculation on the first data.
In the embodiment of the present application, the second target feedback signal is an operation result of the spin signals generated by the m ircin units and the second set of problem submatrices in the operation process.
For any of the Isci units 200, the spin signal generation module 210 in the Isci unit 200 performs intensity modulation and phase modulation on the first set of signals, and outputs a set of spin signals. The manner of the set of spin signals output by the spin signal generating module 210 can be referred to the foregoing description, and is not described herein again.
Each feedback calculation module 220 generates a set of feedback signals based on the configured problem sub-matrix and a set of spin signals; the way in which the feedback calculation module 220 generates a set of feedback signals based on the problem submatrix and a set of spin signals can be referred to the foregoing description, and is not described herein again.
Taking as an example that the m feedback calculation modules 220 load a column of problem sub-matrices in the problem matrix every time the optical calculation apparatus 10 performs an operation process, a round of calculation process of the optical calculation apparatus 10 will be described.
As shown in fig. 6, in m calculation processes of one round of calculation process of the light calculation apparatus 10, m problem sub-matrices loaded in the feedback calculation module 220 in each calculation process are fed back.
From FIG. 6, during time t1, light calculationIn the first operation process of the device 10, the m feedback calculation modules 220 load the first column of problem submatrixes in the problem matrix, and the m feedback calculation modules 220 output m sets of feedback signals { f }11，f21，...fm1In which f11A set of feedback signals, f, output by the first feedback calculation module 220 during time t121A set of feedback signals, f, output by the second feedback calculation module 220 during time t1m1A set of feedback signals output by the mth feedback calculation module 220 at time t 1; the management unit 100 feeds back the m sets of feedback signals { f }11，f21，...fm1The sum is output as a target feedback signal f1. Target feedback signal f1Is a signal column comprising n signals. { f11，f21，…fm1}。
During the second operation of the light computing apparatus 10 at time t2, the m feedback computing modules 220 load the problem submatrix of the 2 nd column in the problem matrix, and the m feedback computing modules 220 output m sets of feedback signals { f } f12，f22，...fm2In which f12A set of feedback signals, f, output by the first feedback calculation module 220 during time t222A set of feedback signals, f, output by the second feedback calculation module 220 during time t2m2A set of feedback signals output by the mth feedback calculation module 220 at time t 2; the management unit 100 feeds back the m sets of feedback signals { f }12，f22，...fm2The sum is output as a target feedback signal f2. Target feedback signal f2Is a signal column comprising n signals.
During the operation of the light computing device 10 for the mth time within tm, the m feedback computing modules 220 load the mth column of problem submatrix in the problem matrix, and the m feedback computing modules 220 output m sets of feedback signals { f }1m，f2m，...fmmIn which f1mA set of feedback signals, f, output by the first feedback calculation module 220 during time tm2mA set of feedback signals, f, output by the second feedback calculation module 220 during time tmmmCalculating a modulus for the mth feedbackBlock 220 outputs a set of feedback signals at time tm. The management unit 100 feeds back the signals { f) to the m groups1m，f2m，...fmmThe sum is output as a target feedback signal fm. Target feedback signal fmIs a signal column comprising n signals.
After the m-th operation process of the optical computing device 10 is finished, the optical computing device 10 completes a round of operation process in which the optical computing device 10 cumulatively generates m target feedback signals { f }1，f2，...fm}。
(3) Multiple rounds of operational process iteration
The light computing device 10 completes a round of computation, and the management unit 100 generates m target feedback signals { f }1，f2，...fmFor the next round of calculation, the management unit 100 may send m target feedback signals { f }1，f2，...fmSent to m isooctane units 200, respectively. An ecin unit 200 receives a target feedback signal.
Specifically, the management unit 100 may feed back m target feedback signals { f }1，f2，...fmSending the m isooctane units 200 to the m isooctane units 200 in sequence according to the arrangement sequence of the m isooctane units 200. For example, the management unit 100 may target the feedback signal f1Sent to the first yixin unit 200, the management unit 100 may send a target feedback signal f2Sent to a second exine unit 200. The management unit 100 may feed back the target feedback signal fmTo the mth yixin unit 200.
When sending a target feedback signal to an inching unit 200, the management unit 100 may send the target feedback signal to the spin signal generation module 210 of the inching unit 200.
The management unit 100 feeds back the target feedback signal f1For illustration, the target feedback signal f is sent to a spin signal generating module 2101Comprising n signals, the management unit 100 loads the n signals on n intensity modulators of the intensity modulator array 211 in the spin signal generation module 210, respectivelyOne of the n signals is loaded on one intensity modulator.
The management unit 100 feeds back m target feedback signals { f1，f2，...fmAfter being sent to the m ecin units 200 (spin signal generating modules 210), respectively, the optical computing device 10 may perform m operation processes, i.e., one operation process.
The manner in which the optical computing device 10 can perform a round of operation process can be referred to the foregoing description, and it should be noted that, since each spin signal generation module 210 loads a target feedback signal, each spin signal generation module 210 can perform intensity modulation and phase modulation on the first group of signals based on the target feedback signal when performing intensity modulation and phase modulation on the first group of signals in one operation process.
When the spin signal generating module 210 performs intensity modulation and phase modulation on the first group of signals based on the target feedback signal, if the signals in the first group of signals are optical signals, the spin signal generating module 210 may perform intensity modulation and phase modulation on the optical signals in the first group of signals according to the amplitude and phase of the target feedback signal. If the signals in the first set of signals are electrical signals, the spin signal generation module 210 may modulate the intensity of the electrical signals in the first set of signals according to the amplitude of the target feedback signal. Each of the itu units 200 generates a set of feedback signals from the received first set of signals, the target feedback signal and the configured problem sub-matrix.
Taking the example that the management unit 100 configures the first set of problem sub-matrices for the multiple yixin units 200, a description will be given of a single operation process involving a first yixin unit and a second yixin unit in the multiple yixin units 200, where the management unit 100 configures a first problem sub-matrix in the first set of problem sub-matrices in the first yixin unit, and the management unit 100 configures a second problem sub-matrix in the first set of problem sub-matrices in the second yixin unit. The management unit 100 sends the first target feedback signal to the first yixin unit, and the management unit 100 sends the second target feedback signal to the second yixin unit.
The first yixin unit may obtain a fifth set of feedback signals according to the first set of signals, the first target feedback signal, and the first problem submatrix; the second Italic unit may generate a sixth set of feedback signals from the first set of signals, the second target feedback signal, and the second problem sub-matrix.
The management unit 100 receives a plurality of sets of feedback signals (for convenience of distinction, the plurality of sets of feedback signals are referred to as a third plurality of sets of feedback signals) output by the plurality of yixin units 200, wherein the third plurality of sets of feedback signals include a fifth set of feedback signals and a sixth set of feedback signals, and the management unit 100 generates a third target feedback signal according to the third plurality of sets of feedback signals.
For any of the Isci units 200, the spin signal generation module 210 in Isci unit 200 outputs a set of spin signals based on a set of signals and the target feedback signal.
The feedback calculation module 220 in the Itanium unit 200 generates a set of feedback signals based on the problem submatrix and the set of spin signals;
the management unit 100 receives a group of feedback signals output by the feedback calculation modules 220 in m of the ecin units 200, and may sum the m groups of feedback signals to generate a third target feedback signal.
In the optical computing device shown in fig. 1, each of the inching units 200 outputs a set of feedback signals, and the management unit 100 may receive a set of feedback signals output by m inching units 200, and may receive m sets of feedback signals in total, where the m sets of feedback signals are third sets of feedback signals, and may generate a third target feedback signal according to the m sets of feedback signals, where the third target feedback signal is a third intermediate result of the optical computing device performing an inching calculation on the first data.
In the embodiment of the present application, the third target feedback signal is a result of the operation between the spin signal generated by the m ircin units and the first set of problem submatrices during the operation. The spin signal is generated based on the first target feedback signal.
After each round of operation process is finished, the management unit 100 may calculate the hami metric H and the hami metric H by using the m target feedback signals determined in the round
Wherein σiAnd σjIs the spin signal in the set of spin signals output by the spin signal generation module 210 during the round of operation.
The hami metric H indicates the system energy corresponding to the first data, and when the system energy corresponding to the first data is no longer reduced, that is, the system energy of the izod model is no longer reduced, that is, the hami metric H converges and is no longer reduced, the izod model solved by the optical computing device 10 completes convergence, and the optical computing device 10 may stop operating. At this time, the set of spin signals output by each spin signal generation module 210 is the optimal solution. When the system energy corresponding to the first data is still reduced, that is, the system energy of the ixing model is reduced, that is, the hami metric H is not converged and is continuously reduced, the optical computing device 10 still needs to continue to perform the operation, that is, to continue to perform a round of operation until the hami metric H is converged.
As shown in fig. 7, a computing system provided for the embodiment of the present application includes an optical computing device 10 and a laser 20, where the laser 20 is capable of outputting a first set of signals, so that the optical computing device 10 receives the first set of signals to perform an operation.
In order to make the description of the scheme clearer, the workflow of the optical computing device provided in the embodiment of the present invention will be generally described below by taking the computing method shown in fig. 8 as an example in conjunction with the previous embodiment. As shown in fig. 8, during operation: the management unit 100 decomposes the problem matrix and determines a first set of problem sub-matrices from the problem matrix. The management unit 100 configures a first question sub-matrix of the first set of question sub-matrices in a first yixin unit. A second problem sub-matrix of the first set of problem sub-matrices is configured in a second incarnation cell.
The first IshC unit receives the first set of signals (step 1), and generates a first set of feedback signals based on the received first set of signals and the first problem submatrix (step 2).
The first IshC unit receives the first set of signals (step 3), and generates a first set of feedback signals based on the received first set of signals and the second problem submatrix (step 4).
The management unit 100 receives a first plurality of sets of feedback signals (including a first set of feedback signals and a second set of feedback signals) (step 5), and generates a first target feedback signal based on the first plurality of sets of feedback signals (step 6).
The above process is an execution mode of a single operation process in the optical computing device, and the optical computing device can continue to execute m-1 operation processes, wherein the single operation process is as follows:
the management unit 100 arranges a third problem sub-matrix of the second group of problem sub-matrices in the first yixing unit, and the management unit 100 arranges a fourth problem sub-matrix of the second group of problem sub-matrices in the second yixing unit.
The first IshCi unit may obtain a third set of feedback signals according to the first set of signals and the third problem sub-matrix; the second itu unit may generate a fourth set of feedback signals based on the first set of signals and the fourth problem sub-matrix.
The management unit 100 receives a second plurality of sets of feedback signals, wherein the second plurality of sets of feedback signals includes a third set of feedback signals and a third set of feedback signals, and the management unit 100 generates a second target feedback signal according to the second plurality of sets of feedback signals.
In the foregoing description, the m-time operation process of the optical computing device is a round of operation process performed by the optical computing device, the optical computing device may perform multiple rounds of operation processes in a similar manner, the foregoing process is taken as a first round of operation process, and the following first operation process in the next round of operation process performed by the optical computing device is taken as an example to describe any operation process of the optical computing device in the next round of operation process:
the management unit 100 sends the first target feedback signal to the first yixin unit, and the management unit 100 sends the second target feedback signal to the second yixin unit.
The first yixin unit may obtain a fifth set of feedback signals according to the first set of signals, the first target feedback signal, and the first problem submatrix; the second Italic unit may generate a sixth set of feedback signals from the first set of signals, the second target feedback signal, and the second problem sub-matrix.
The management unit 100 receives a plurality of sets of feedback signals (for convenience of distinction, the plurality of sets of feedback signals are referred to as a third plurality of sets of feedback signals) output by the plurality of yixin units 200, wherein the third plurality of sets of feedback signals include a fifth set of feedback signals and a sixth set of feedback signals, and the management unit 100 generates a third target feedback signal according to the third plurality of sets of feedback signals.
The optical computing equipment provided by the embodiment of the invention decomposes the process of solving the Itanium model into one or more rounds of operation processes, each round of operation process comprises m operation processes (time-sharing computation), and each operation process can be realized by utilizing a parallel operation mode of m Itanium units (parallel computation). The method can support the solution of the Esin model with a large number of spinning nodes, and improve the operation efficiency.
In addition, because the optical computing device provided by the embodiment of the invention has a simple structure and can be realized on a chip, the whole computing process is realized in an optical signal or electric signal mode, the signal transmission speed is high, and the computing speed is greatly improved, the computing device improved by the embodiment of the invention can be applied to a neural network system, for example, can be used for realizing feedback control in the neural network system.
It should be noted that the examples provided in this application are only illustrative. It will be apparent to those skilled in the art that, for convenience and brevity of description, the description of the various embodiments has been focused on, and for parts of one embodiment that are not described in detail, reference may be made to the description of other embodiments. The features disclosed in the embodiments of the invention, in the claims and in the drawings may be present independently or in combination. Features described in hardware in embodiments of the invention may be implemented by software and vice versa. And are not limited herein.
Claims (14)
1. A light computing device, comprising:
the first Yixinu unit is used for obtaining a first group of feedback signals according to the first group of signals and the first problem submatrix;
the second Yixinu unit is used for generating a second group of feedback signals according to the first group of signals and a second problem sub-matrix, wherein the first problem sub-matrix and the second problem sub-matrix are respectively different sub-matrices of the problem matrix, and the problem matrix is used for indicating first data to be calculated;
and the management unit is connected with the first and second IshC units and used for receiving a first plurality of groups of feedback signals comprising the first and second groups of feedback signals and generating a first target feedback signal according to the first plurality of groups of feedback signals, wherein the first target feedback signal is used for indicating the optical computing equipment to execute a first intermediate result of IshC calculation on the first data.
2. The apparatus of claim 1,
the first Yixinu unit is further configured to obtain a third set of feedback signals according to the first set of signals and a third problem submatrix;
the second IshC unit is further used for generating a fourth group of feedback signals according to the first group of signals and a fourth problem submatrix; wherein the third problem sub-matrix and the fourth problem sub-matrix are different sub-matrices of the problem matrix;
the management unit is further configured to receive a second plurality of sets of feedback signals including the third set of feedback signals and the fourth set of feedback signals, and generate a second target feedback signal according to the second plurality of sets of feedback signals.
3. The apparatus of claim 1 or 2,
the management unit is further configured to send the first target feedback signal to the first yixin unit, and send the second target feedback signal to the second yixin unit;
the first IshCi unit is further configured to obtain a fifth set of feedback signals according to the first set of signals, the first target feedback signal, and the first problem submatrix;
the second Italic unit is configured to generate a sixth set of feedback signals according to the first set of signals, the second target feedback signals, and the second problem submatrix;
the management unit is configured to receive a third plurality of sets of feedback signals including the fifth set of feedback signals and the sixth set of feedback signals, and generate a third target feedback signal according to the third plurality of sets of feedback signals.
4. The light computing device of any of claims 1-3, wherein the first Ill cell comprises:
a spin signal generating module for obtaining a first set of spin signals from the first set of signals;
and the feedback calculation module is used for generating a first group of feedback signals according to the first group of spin signals and the first problem submatrix.
5. The light computing device of claim 4, wherein the spin signal generation module is to intensity modulate and phase modulate the first set of signals to obtain the first set of spin signals.
6. The device of claim 4 or 5, wherein the spin signal generation module comprises:
the intensity modulator array is used for carrying out intensity modulation on the first group of signals to obtain a first group of modulation signals;
and the phase modulator array is used for carrying out phase modulation on the first group of modulation signals to obtain the first group of spin signals.
7. The apparatus of claim 6, wherein the array of intensity modulators comprises a plurality of Mach-Zehnder interferometers (MZIs), or electroabsorption modulators (EAMs).
8. The device of claim 6, wherein the phase modulator array comprises a plurality of waveguides.
9. The apparatus of claim 4, wherein the feedback calculation module comprises a plurality of Mach-Zehnder interferometers (MZIs).
10. The apparatus of any of claims 3-9, wherein the management unit is further to:
calculating a Hamiltonian according to the third target feedback signal, wherein the Hamiltonian is used for representing system energy corresponding to the first data;
and when the Hamiltonian convergence is determined, obtaining a calculation result of the first data according to the third group of spin signals.
11. A calculation method is characterized in that the method is executed by an optical calculation device, the optical calculation device comprises a management unit, a first Itanium unit and a second Itanium unit, and the management unit is connected with the first Itanium unit and the second Itanium unit;
the first IshCi unit obtains a first group of feedback signals according to the first group of signals and the first problem submatrix;
the second IshCi unit generates a second group of feedback signals according to the first group of signals and a second problem sub-matrix, wherein the first problem sub-matrix and the second problem sub-matrix are different sub-matrices of a problem matrix respectively, and the problem matrix is used for indicating first data to be calculated;
the management unit receives a first plurality of sets of feedback signals including the first set of feedback signals and the second set of feedback signals, and generates a first target feedback signal according to the first plurality of sets of feedback signals.
12. The method of claim 11, further comprising:
the first IshCi unit obtains a third group of feedback signals according to the first group of signals and a third problem sub-matrix;
the second IshCi unit generates a fourth group of feedback signals according to the first group of signals and a fourth problem sub-matrix; wherein the third problem sub-matrix and the fourth problem sub-matrix are different sub-matrices of the problem matrix;
the management unit receives a second plurality of sets of feedback signals including the third set of feedback signals and the fourth set of feedback signals, and generates a second target feedback signal according to the second plurality of sets of feedback signals.
13. The method of claim 11 or 12, further comprising:
the management unit sends the first target feedback signal to the first Ishige unit and sends the second target feedback signal to the second Ishige unit;
the first Yixin unit obtains a fifth group of feedback signals according to the first group of signals, the first target feedback signals and the first problem submatrix;
the second IshCi unit generates a sixth set of feedback signals according to the first set of signals, the second target feedback signals and the second problem sub-matrix;
the management unit receives a third plurality of sets of feedback signals including the fifth set of feedback signals and the sixth set of feedback signals, and generates a third target feedback signal according to the third plurality of sets of feedback signals.
14. A computing system comprising an optical computing device according to any of claims 1 to 10 and a laser for transmitting the first set of signals to the optical computing device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010623939.9A CN113867474A (en) | 2020-06-30 | 2020-06-30 | Optical computing device, computing method and computing system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010623939.9A CN113867474A (en) | 2020-06-30 | 2020-06-30 | Optical computing device, computing method and computing system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113867474A true CN113867474A (en) | 2021-12-31 |
Family
ID=78980990
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010623939.9A Pending CN113867474A (en) | 2020-06-30 | 2020-06-30 | Optical computing device, computing method and computing system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113867474A (en) |
-
2020
- 2020-06-30 CN CN202010623939.9A patent/CN113867474A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Larger et al. | High-speed photonic reservoir computing using a time-delay-based architecture: Million words per second classification | |
| Nahmias et al. | Photonic multiply-accumulate operations for neural networks | |
| CN109784486B (en) | Optical neural network processor and training method thereof | |
| US11398871B2 (en) | Systems and methods for analog computing using a linear photonic processor | |
| US11373089B2 (en) | Serialized electro-optic neural network using optical weights encoding | |
| TWI767877B (en) | Optoelectronic processing system | |
| CN109784485B (en) | Optical neural network processor and calculation method thereof | |
| TWI806042B (en) | Optoelectronic processing apparatus, system and method | |
| WO2020092899A1 (en) | Matrix multiplication using optical processing | |
| Shiflett et al. | Albireo: Energy-efficient acceleration of convolutional neural networks via silicon photonics | |
| US20200110432A1 (en) | Reservoir computing operations using multi-mode photonic integrated circuits | |
| CN111753977B (en) | Optical neural network convolution layer chip, convolution calculation method and electronic equipment | |
| CN111723337B (en) | Photon tensor core integrated circuit architecture for neural network training and neural network training method thereof | |
| CN109376855A (en) | A kind of smooth neuronal structure and the Processing with Neural Network system comprising the structure | |
| CN115169542B (en) | Two-dimensional photon convolution acceleration system and device for convolution neural network | |
| CN113570051A (en) | Photoelectric processing system | |
| WO2023109065A1 (en) | Method and system for implementing recurrent neural network, and electronic device and storage medium | |
| CN114970836A (en) | Reservoir neural network implementation method and system, electronic equipment and storage medium | |
| CN113325917A (en) | Light computing device, system and computing method | |
| CN113867474A (en) | Optical computing device, computing method and computing system | |
| CN113705774A (en) | Optical circuit construction method, optical circuit, optical signal processing method and device | |
| EP4040258A1 (en) | Optical vector multiplier | |
| De Marinis et al. | Analysis of Integration Technologies for High-Speed Analog Neuromorphic Photonics | |
| JPH05233859A (en) | Optical neural network system | |
| WO2024012335A1 (en) | Computing system and method for processing photonic computing result |
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 |