WO2021047614A1 - 一种光计算设备以及计算方法 - Google Patents

一种光计算设备以及计算方法 Download PDF

Info

Publication number
WO2021047614A1
WO2021047614A1 PCT/CN2020/114614 CN2020114614W WO2021047614A1 WO 2021047614 A1 WO2021047614 A1 WO 2021047614A1 CN 2020114614 W CN2020114614 W CN 2020114614W WO 2021047614 A1 WO2021047614 A1 WO 2021047614A1
Authority
WO
WIPO (PCT)
Prior art keywords
signals
spin
optical
feedback
array
Prior art date
Application number
PCT/CN2020/114614
Other languages
English (en)
French (fr)
Inventor
云志强
张翔
董晓文
Original Assignee
华为技术有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to EP20863261.2A priority Critical patent/EP4027250A4/en
Priority to JP2022515890A priority patent/JP7485758B2/ja
Publication of WO2021047614A1 publication Critical patent/WO2021047614A1/zh
Priority to US17/691,945 priority patent/US20220197328A1/en

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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F15/00—Digital computers in general; Data processing equipment in general
    • G06F15/76—Architectures of general purpose stored program computers
    • G06F15/78—Architectures of general purpose stored program computers comprising a single central processing unit
    • G06F15/7807—System on chip, i.e. computer system on a single chip; System in package, i.e. computer system on one or more chips in a single package
    • G06F15/7817—Specially adapted for signal processing, e.g. Harvard architectures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks

Definitions

  • This application relates to the field of information technology, in particular to an optical computing device and a computing method.
  • the combinatorial optimization problems involved in many fields such as drug design, transportation physics planning, network resource allocation are non-deterministic polynomial hard (NP-hard) problems, because the solution time varies with the number of variables in the problem. The index increases and it is difficult to solve in a reasonable time.
  • NP-hard polynomial hard
  • the Ising model describes the random process that represents the phase transition of N nodes in the magnetic structure.
  • One of the nodes represents a magnetic dipole, and the magnetic dipole has a spin state of +1 or -1.
  • a node can be coupled with many other nodes to produce a phase change of matter (such as a change in spin state).
  • the optical Ising machine simulates and solves the Ising model by using multiple optical signals (or light pulses) to form the physical phenomenon of a grid point network.
  • the phase of the optical signal (or optical pulse) is used to represent the spin state of the node in the Ising model, and the optical signal or optical pulses are coupled and interfered with each other to change the phase of the optical signal (or optical pulse).
  • the optical signal in the implementation scheme of the optical Ising machine usually runs in series. As the number of nodes in the Ising model increases, more optical signals need to be introduced. The increase in the number of optical signals will also increase the number of optical signals in the optical signal. The transmission time in the Ising machine greatly increases the computing time, which limits the computing efficiency of the Ising machine.
  • the present application provides an optical computing device and a computing method, which are used to provide an optical Issing machine with higher operating efficiency and improve the computing speed.
  • this application provides an optical computing device, which includes a first spin array, an optical feedback network, and a second spin array, and the optical feedback network is connected to the first spin array and the second spin array, respectively .
  • the first spin array may receive a first set of signals, and generate a first set of spin signals according to the first set of signals, the first set of spin signals includes N spin signals, and N is not less than 2. Integer.
  • the first spin array may send the first set of spin signals to the optical feedback network.
  • the optical feedback network After receiving the first set of spin signals, the optical feedback network may generate a first set of feedback signals according to the first set of spin signals and the set first data, where the first set of feedback signals Including N feedback signals.
  • the optical feedback network may send the first set of feedback signals to the second spin array.
  • the second spin array may also receive the first set of signals, and the second spin array may be generated according to the first set of feedback signals and the first set of signals The second set of spin signals, wherein the second set of spin signals includes N spin signals.
  • a set of spin signals generated by the first spin array or the second spin array each time can instruct the optical computing device to process an intermediate calculation result in the process of processing the optical signal.
  • the group of spin signals can instruct the optical computing device to process the final calculation result of the optical signal at this time.
  • the first spin array and the second spin array receive the same set of signals, and the first spin array can send feedback signals to the second spin array through the optical feedback network, So that the second spin array can also generate a set of spin signals.
  • the first spin array and the second spin array process the signals in a set of signals in parallel. Therefore, while increasing the number of signals, it does not increase the computing time of the optical computing device, which can effectively improve the optical computing The computing efficiency of the equipment.
  • the second spin array may send the second set of spin signals to the optical feedback network, and the optical feedback network may receive the second set of spin signals, or A second set of feedback signals are generated according to the second set of spin signals and the first data, where the second set of feedback signals includes N feedback signals.
  • the first spin array may also receive the second set of signals.
  • the first spin array may also generate a third set of spin signals according to the second set of feedback signals and the second set of signals, where the third set of spin signals includes N spin signals.
  • the first spin array and the second spin array are coupled to establish a connection through the optical feedback network. Adopting this connection method simplifies the structure of the optical computing device.
  • the first spin array and the second spin array can receive the same set of signals, and each spin array can process the set of signals in parallel.
  • feedback signals can be sent to each other through an optical feedback network.
  • the signal transmission process between the first spin array and the second spin array through the optical feedback network is simpler and more efficient, which makes the calculation efficiency of the optical computing device more efficient. It can be improved, and the first spin array and the second spin array can send feedback signals to each other through the optical feedback network to realize the self generated by the first spin array and the second spin array.
  • the phase of the spin signal is interlocked.
  • the optical computing device since the first spin array and the second spin array are coupled through the optical feedback network to establish a connection, this way makes the feedback signal the first spin array and the second spin array
  • the two-spin array injects feedback signals into each other through the optical feedback network, and multiple feedback signals are injected at the same time, without increasing the length of the fiber, which can effectively reduce the signal transmission time, reduce the calculation time of the optical computing device, and help improve the computing efficiency of the optical computing device .
  • the first spin array includes N optical parametric oscillators, and each optical parametric oscillator in the first spin array can receive one signal of the first group of signals.
  • the N optical parametric oscillators included in the first spin array can process multiple signals in the first group of signals in parallel, and each optical parametric oscillator can generate A spin signal, the first spin array processes the first set of signals in parallel, which can effectively shorten the calculation time of the optical computing device, and can effectively improve the signal processing efficiency of the first spin array shown, thereby ensuring The calculation process of the optical computing device can be performed efficiently.
  • the second spin array includes N optical parametric oscillators, and each optical parametric oscillator in the second spin array can receive one signal of the first group of signals, It is also possible to receive a feedback signal in the first group of feedback signals, and a signal received by each optical parametric oscillator and a feedback signal interact in the optical parametric oscillator.
  • N optical parametric oscillators in the second spin array and the N optical parametric oscillators in the first spin array.
  • the signals received by the corresponding optical parametric oscillators in the first spin array are the same.
  • the N optical parametric oscillators included in the second spin array can process N feedback signals and multiple signals in a group of signals in parallel, and each optical parametric oscillator The second spin array can generate a spin signal.
  • the manner in which the second spin array processes multiple signals in a group of signals in parallel can improve the signal processing efficiency of the second spin array, and can ensure that the calculation process of the optical computing device can be
  • the corresponding relationship between the optical parametric oscillators in the second spin array and the first spin array can ensure that each optical parametric oscillator of the first spin array and the second spin array
  • the optical parametric oscillator corresponding to the array receives the same optical pulse.
  • the optical feedback network includes the first signal processing module, the problem loading module, and the second signal processing module.
  • the problem loading module is respectively connected to the first signal processing module and the second signal processing module.
  • the first signal processing module may receive the first set of spin signals, and the first signal processing module may convert the first set of spin signals into a first optical signal matrix, the first optical signal matrix Including N*N light pulses.
  • the first signal processing module may transmit the first optical signal matrix to the question loading module, and the question loading module may generate a first feedback signal matrix according to the first optical signal matrix and the first data ,
  • the first feedback signal matrix includes N*N optical signals.
  • the second signal processing module may convert the first feedback signal matrix into the first group of feedback signals.
  • optical computing device when signal processing is performed inside the optical feedback network, photoelectric conversion is not required, and a set of spin signals can be converted into a set of feedback signals, which can effectively shorten the calculation time and thereby Improve the computational efficiency of optical computing equipment.
  • the second signal processing module may also receive the second set of spin signals, and convert the second set of spin signals into a second optical signal matrix, wherein the first The second optical signal matrix includes N*N optical pulses, and the second signal processing module can send the second optical signal matrix to the question loading module; then, the question loading module can be based on the second The optical signal matrix and the first data generate a second feedback signal matrix, wherein the second feedback signal matrix includes N*N optical signals; the question loading module may send the second feedback signal matrix to The first signal processing unit and the first signal processing module may convert the second feedback signal matrix into the second group of feedback signals.
  • optical computing device when signal processing is performed inside the optical feedback network, photoelectric conversion is not required, and a set of spin signals can be converted into a set of feedback signals, which can effectively shorten the calculation time and thereby Improve the computational efficiency of optical computing equipment.
  • the optical feedback network may include a plurality of cascaded Mach-Zehnder interference units, and each Mach-Zehnder interference unit includes Mach-Zehnder interferometers and optical switches arranged at intervals.
  • the Mach-Zehnder interference unit has a small size and can form an optical computing device with a compact structure, which can enable the optical computing device to be implemented on a chip, ensuring the stability of the system.
  • the optical computing device further includes a detector array, and the detector array can be connected to any spin array.
  • the third group of spin arrays can be detected.
  • the detector array can conveniently detect the phase of a group of spin signals generated by the first spin array to determine the final output result.
  • the set of signals (such as the first set of signals and the second set of signals) received by the first spin array and the second spin array may include multiple optical pulses or multiple electrical pulses.
  • a signal, an optical pulse or an electrical signal can be referred to as a signal in a group of signals.
  • the number of optical pulses or electrical signals may be N.
  • the first group of signals includes a first group of light pulses
  • the second group of signals includes a second group of light pulses, wherein the amplitude of the second group of light pulses is greater than that of the first group of light pulses Amplitude.
  • the first spin array and the second spin array can directly process the light pulses or perform photoelectric conversion, and then process the converted light pulses, which can be applied to different The application scenarios of the company have expanded the scope of application.
  • the present application provides a calculation method.
  • the method is executed by an optical computing device, the optical computing device includes a first spin array, a second spin array, and an optical feedback network connected to the first spin array and the second spin array, respectively.
  • the method includes: the first spin array first receives a first set of signals, and generates a first set of spin signals according to the first set of signals, and the first set of spin signals includes N spin signals, where N is An integer not less than 2; the first spin array sends the first set of spin signals to the optical feedback network.
  • the optical feedback network After receiving the first set of spin arrays, the optical feedback network generates a first set of feedback signals according to the first set of spin signals and the set first data, where the first set of feedback signals includes N A feedback signal.
  • the optical feedback network sends the first set of feedback signals to the second spin array.
  • the second spin array receives the first set of signals and the first set of feedback signals, and can generate a second set of spin signals according to the first set of feedback signals and the first set of signals.
  • the second set of spin signals includes N spin signals.
  • the second spin array may send the second set of spin signals to the optical feedback network, and the optical feedback network receives the first spin signal sent by the second spin array.
  • a second set of feedback signals may be generated according to the second set of spin signals and the first data, where the second set of feedback signals includes N feedback signals.
  • the optical feedback network may send the second set of feedback signals to the first spin array.
  • the first spin array can also receive a second set of signals
  • the first spin array generates a third set of spin signals according to the received second set of feedback signals and the second set of signals, and the third set of spin signals includes N spin signals.
  • the optical computing device further includes a detector array capable of detecting the third set of spin signals; determining each spin signal in the third set of spin signals ⁇ The phase.
  • the detector array may obtain the calculation result of the first data according to the third set of spin signals.
  • each optical parametric oscillator in the first spin array can receive one signal in the first set of signals.
  • the second spin array includes N optical parametric oscillators, and when the second spin array receives the first set of signals and the first set of feedback signals, the Each optical parametric oscillator in the second spin array can receive one signal in the first set of signals and one feedback signal in the first set of feedback signals, wherein
  • the N optical parametric oscillators have a one-to-one correspondence with the N optical parametric oscillators in the first spin array, and the second spin array has a corresponding relationship with the optical parametric oscillators in the first spin array
  • the received light pulses are the same.
  • the present application provides an optical computing chip.
  • the optical computing chip may include the optical computing device as described in the first aspect or any one of the possible implementations of the first aspect.
  • Figure 1 is a schematic structural diagram of an optical computing device provided by this application.
  • Figure 2 is a flow chart of signal transmission in an optical computing device provided by this application.
  • FIG. 3 is a flow chart of signal transmission in another optical computing device provided by this application.
  • FIG. 4 is a schematic diagram of light pulse groups received by two spin arrays provided by this application.
  • FIG. 5 is a schematic structural diagram of a second spin array provided by this application.
  • FIG. 6 is a schematic structural diagram of an optical parametric oscillation cavity provided by this application.
  • FIG. 7 is a schematic structural diagram of an optical feedback network provided by this application.
  • 8A is a flowchart of processing spin signal group E by the first signal processing module on the XY plane provided by this application;
  • FIG. 9 is a schematic diagram of an optical signal matrix generated by the first signal processing module provided by this application.
  • FIG. 10 is a flowchart of signal processing by a problem loading module provided by this application.
  • FIG. 11 is a schematic diagram of a feedback signal matrix provided by this application.
  • 12A is a flowchart of processing the feedback signal matrix by the second signal processing module on the XY plane provided by this application;
  • 12B is a flow chart of processing a column of signals in the feedback signal matrix provided by the first signal processing module of the YZ plane provided by this application;
  • FIG. 13 is a schematic structural diagram of an optical computing device provided by this application.
  • FIG. 14 is a schematic diagram of the structure of an MZIU provided by this application.
  • 15A is a schematic diagram of a signal transmission path in an MZIU provided by this application.
  • 15B is a schematic diagram of a signal transmission path in another MZIU provided by this application.
  • FIG. 16 is a schematic structural diagram of an optical computing device provided by this application.
  • FIG. 17 is a schematic diagram of a calculation method provided by this application.
  • This application provides a calculation method and optical computing equipment to provide an optical Ising machine with higher operating efficiency.
  • the optical computing device 10 includes two spin arrays and an optical feedback network 300.
  • the two spin arrays use a first spin array 100 and a second spin array respectively.
  • the difference between the two spin arrays 200 is that the optical feedback network 300 is connected to the first spin array 100 and the second spin array 200 respectively.
  • the spin array (the first spin array 100 or the second spin array 200) can receive a set of signals (and multiple feedback signals fed back by the optical feedback network 300), according to the received set of signals ( And multiple feedback signals) to generate multiple spin signals.
  • the embodiment of the present application does not limit the signal types included in a group of signals received by the spin array, and the group of signals may include multiple optical pulses or multiple electrical signals (such as electrical pulses).
  • the number of optical signals and electrical signals can be N.
  • the spin array can convert the received set of electrical signals into N optical pulses.
  • the spin array can convert the received electrical signals into N optical pulses.
  • the light pulses (and N feedback signals) are converted into N spin signals.
  • the signal received by the spin array is taken as an example of a group of optical pulses, and the group of optical pulses includes N optical pulses.
  • the optical feedback network 300 can receive multiple spin signals generated by a spin array, and based on preset first data, process the multiple spin signals generated by the first spin array 100 to generate multiple feedback signals , The multiple feedback signals generated are fed back to another spin array.
  • a group of signals received by the first spin array 100 is an example of a group of light pulses including N light pulses.
  • the case of (or the second spin array 200) including N electrical signals is similar to the case of a group of optical pulses received by the first spin array 100 (or the second spin array 200), except that the first spin array 100
  • the first spin array 100 (or the second spin array 200) needs to first convert the N electrical signals into N optical pulses .
  • the subsequent processing of the N optical pulses is the same as the processing of the N optical pulses when the first spin array 100 (or the second spin array 200) receives a group of optical pulses.
  • the signal transmission process of the first spin array 100, the optical feedback network 300, and the second spin array 200 will be described below.
  • the signal between the first spin array 100, the optical feedback network 300, and the second spin array 200 According to the direction of signal transmission, the transmission can include two processes: the signal transmission of the spin signal generated by the first spin array 100 through the optical feedback network 300 to the second spin array 200 (process 1), and the second spin array
  • the spin signal generated by 200 passes through the optical feedback network 300 to the signal transmission of the first spin array 100 (process 2):
  • the first process is a signal transmission process in which the spin signal generated by the first spin array 100 passes through the optical feedback network 300 to the second spin array 200.
  • the first spin array 100 can receive a group of light pulses (for convenience of description, the group of light pulses received by the first spin array 100 is represented by light pulse group A), and the light pulse group A includes Multiple identical light pulses are A 1 , A 2 , A 3 ,...A N.
  • the first spin array 100 can generate a group of spin signals according to the optical pulse group A (for convenience of description, the group of spin signals is represented by a spin signal group E), and multiple spin signals in the spin signal group E, respectively, E 1, E 2, E 3 ?? E N, then, the group E spin signals input to the optical feedback network 300.
  • the group of spin signals is represented by a spin signal group E
  • multiple spin signals in the spin signal group E respectively, E 1, E 2, E 3 — E N, then, the group E spin signals input to the optical feedback network 300.
  • the optical feedback network 300 uses the preset first data to process the spin signal group E to generate a group of feedback signals (for convenience of description, the group of feedback signals is represented by the feedback signal group fA) , the feedback signal comprising a plurality of feedback signals fA group, respectively fA 1, fA 2, ising, fA N, the feedback signal input value of the second spin group fA array 200.
  • the spin signal generated by the second spin array 200 passes through the optical feedback network 300 to reach the signal transmission process of the first spin array 100.
  • the second spin array 200 can also receive the same group of light pulses as the light pulse group A.
  • the group of optical pulses that are the same as the optical pulse group A received by the rotating array 200 is represented by the optical pulse group B.
  • the optical pulse group B includes a plurality of optical pulses, namely B 1 , B 2 , B 3 , ... B N .
  • B 1 and A 1 are the same light pulses
  • B 2 and A 2 are the same light pulses
  • B , 3 and A 3 are the same light pulses
  • B , N and A N are the same light pulses.
  • the second spin array 200 generates a group of spin signals according to the feedback signal group fA and the optical pulse group B (for the convenience of description, the group of spin signals is represented by a spin signal group F), and there are multiple spin signals in the spin signal group F.
  • the spin signals are respectively F 1 , F 2 , F 3, ... F N , and then the spin signal group F is input to the optical feedback network 300.
  • the optical feedback network 300 uses the preset first data to process the spin signal group F to generate a group of feedback signals (for convenience of description, the group of feedback signals is represented by the feedback signal group fB), and the feedback signal group fB includes a plurality of The feedback signals are fB 1 , fB 2 , ..., fB N , and the feedback signal group fB is input to the first spin array 100.
  • the first spin matrix and the second spin matrix can send feedback signals to each other through the optical feedback network 300, which can achieve the phase of the spin signals generated by the first spin array and the second spin array. The effect of interlocking.
  • the first spin array 100 and the second spin array 200 process the optical pulses in a group of optical pulses in parallel, which can effectively reduce the computing time of the optical computing device 10. If the number of optical pulses is increased, it will not Increasing the computing time of the optical computing device effectively guarantees the computing efficiency of the optical computing device.
  • the first spin array 100 After the first spin array 100 receives the light pulse group A, it will continue to receive other groups of light pulses. Taking a group of light pulses received later as the light pulse group A'as an example, the first spin array 100 is receiving light pulses. After the pulse group A', a group of spin signals is generated according to the feedback signal group fB and the optical pulse group A'(for the convenience of explanation, the group of spin signals is represented by the spin signal group P), and the spin signal group P is more There are two spin signals, respectively P 1 , P 2 , P 3 ... P N.
  • the first spin array 100 inputs the spin signal group P to the optical feedback network 300, and the optical feedback network 300 can generate a group of feedback signals according to the spin signal group P and preset first data (for convenience of description, the The group feedback signal is represented by a feedback signal group fA′), the feedback signal group fA′ includes a plurality of feedback signals, and the feedback signal group fA′ is input to the second spin array 200.
  • This process is similar to process one, except that the first spin array 100 generates a set of spin signals based on the subsequent received optical pulse set and multiple feedback signals received by the optical feedback network 300.
  • the second spin array 200 group can also receive the same group of pulsed light as the optical pulse group A'(which can be represented by the optical pulse group B'); using the optical pulse group B'and the feedback signal group fA' input by the optical feedback network 300 Generate a set of spin signals (for convenience, this set of spin signals is represented by a spin signal group Q), and multiple spin signals in the spin signal group Q are Q 1 , Q 2 , Q 3 ?? Q N , the first spin array 100 inputs the spin signal group Q to the optical feedback network 300, and the optical feedback network 300 can generate a set of feedback signals according to the spin signal group and preset first data (for the convenience of description ,
  • the group of feedback signals is represented by a feedback signal group fB′), the feedback signal group fB′ includes a plurality of feedback signals, and the feedback signal group fB′ is input to the first spin array 100.
  • This process is similar to process two, except that the signals (light pulses and feedback signals) received by the second spin array 200 are different.
  • the first spin array 100 and the second spin array 200 in the optical computing device will continue to receive the same optical pulse group subsequently, and perform a signal transmission process similar to the above process. In other words, each time the first spin array 100 and the second spin array 200 in the optical computing device receive the same set of light pulses, a signal transmission process similar to the above process is performed once.
  • a set of spin signals generated by the first spin array 100 or the second spin array 200 during signal processing similar to process one and process two indicate an intermediate calculation result of the optical computing device 10 until the first spin
  • a group of optical pulses received by the first spin array 100 and the second spin array 200 are the same (e.g. The light pulse group A is the same as the light pulse group B, and the light pulse group A'is the same as the light pulse group B').
  • each group of optical pulses received by the first spin array 100 and the second spin array 200 can be emitted by the same pump source. After the pump source emits a group of optical pulses, it can pass through the beam splitter. , Divide the group of light pulses into two groups of the same light pulses, and send them to the first spin array 100 and the second spin array 200, respectively.
  • each group of optical pulses received by the first spin array 100 and the second spin array 200 may be emitted by the same pump source; the first spin array 100 and the second spin array 200.
  • the pump source can send two sets of the same optical pulse at a set time interval, one of which is sent to the first spin array 100, and the other is The light pulse is sent to the second spin array 200.
  • the pump source can generate a group of light pulses every time T/2, and generate the same two groups of light pulses during the time T, and the generated light pulses in different times T can be different.
  • each group of optical pulses received by the first spin array 100 and the second spin array 200 can also be emitted by two different pump sources, but the two different pump sources emit two groups.
  • the light pulse is the same.
  • a spin array (take the second spin array 200 as an example) needs to generate a set of spin signals based on receiving a set of optical pulses and a set of feedback signals received from the optical feedback network 300, where the second spin The set of feedback signals received by the array 200 is generated by the optical feedback network 300 according to the first spin array 100 using the same set of optical pulses (and another set of feedback signals). That is, the time when the second spin array 200 receives the group of light pulses should be later than the time when the first spin array 100 receives the same group of light pulses, so that the second spin array 200 can be When the group of optical pulses is received, the group of feedback signals can be obtained from the optical feedback network 300.
  • the time for the set of optical pulses to reach the two sets of spin arrays is different, and there is a time difference.
  • the time difference can enable the second spin array 200 to receive the group of optical pulses and a group of feedback signals at the same time.
  • the time difference is determined based on the time from when the first spin array 100 receives the group of optical pulses to the time the optical feedback network 300 feeds back a group of feedback signals to the second spin array 200. It is related to the time when the corresponding feedback signal is generated after processing.
  • each group of optical pulses received by the first spin array 100 and the second spin array 200 can be emitted by the same pump source as an example, if the period of each group of optical pulses emitted by the pump source is T, that is to say, the pump The pump source generates a group of light pulses every T. After a group of light pulses are split from the pump source, the time difference between the group of light pulses reaching the first spin array 100 and the second spin array 200 is T/ 2.
  • FIG. 4 it is a schematic diagram of the optical pulse groups received by two spin arrays.
  • the first spin array 100 uses the spin signals generated by the optical pulse group A to generate feedback signal group E through optical feedback signal processing, and inject Into the second spin array 200.
  • the second spin array 200 uses the received optical pulse group B and the feedback signal group E to generate a new spin signal, and the new spin signal is processed by the optical feedback signal to generate the feedback signal group F, which is injected into the first spin array 100 , Cyclically stack in this way.
  • the pump source can adjust a set of light pulses generated each time. For example, the pump source can increase the amplitude of each light pulse in the group of light pulses to generate a group of light pulse groups different from the light pulse group generated before T, and input the generated light pulse groups into the first self Spin array 100 and second spin array 200. Therefore, the first spin array 100 and the second spin array 200 and the optical feedback network 300 in the optical computing device can change a set of spin signals generated by the second spin array 200 according to the set of optical pulses generated by the pump source.
  • the first spin array 100 and the second spin array 200 continue to receive the same set of optical pulses, perform the above-mentioned process, and send feedback signals to each other through the optical feedback network 300. Since the first spin array 100 and the second spin array 200 are coupled to establish a connection through the optical feedback network 300 in the embodiment of the present application, the processing of the optical pulses in each group of optical pulses is parallel, increasing the number of optical pulses, It will not increase the computing time of the optical computing device, and can effectively improve the computing efficiency of the optical computing device.
  • first spin array 100 and the second spin array 200 are similar, and only the second spin array 200 is taken as an example for description here.
  • the second spin array 200 includes a plurality of parallel optical parametric oscillating cavities, and the number of optical parametric oscillating cavities is the same as the number of optical pulses included in a group of received optical pulses.
  • Each optical parametric oscillation cavity is connected to an optical feedback network 300, and receives a feedback signal in a group of feedback signal groups from the optical feedback network 300.
  • the optical parametric oscillating cavity includes, but is not limited to, an optically pumped optical parametric oscillating cavity and an electrically pumped optical parametric oscillating cavity.
  • the second spin array 200 includes N optical parametric oscillators (optical parametric oscillators), and each optical parametric oscillator can receive one optical pulse in the optical pulse group B. In this way, the N optical parametric oscillators can receive optical pulses. N light pulses in group B. As a result, the second spin array 200 can process the N optical pulses of the optical pulse group B in parallel, which can better improve the signal processing efficiency.
  • an optical parametric oscillator (also called an optical parametric oscillator) is a parametric oscillator that can oscillate based on the frequency of an optical signal (or optical pulse).
  • the optical parametric oscillating cavity can generate a new optical signal (or optical pulse) through the nonlinear optical interaction of the optical signal (or optical pulse) input into the optical parametric oscillating cavity.
  • the second spin array 200 when the second spin array 200 receives the optical pulse group B, it will also receive the feedback signal group fA from the optical feedback network 300.
  • the optical feedback network 300 may input multiple feedback signals in the feedback signal group fA into each optical parametric oscillation cavity in the second spin array 200, and each optical parametric oscillation cavity receives a feedback signal. That is, the signal input to one optical parametric oscillation cavity in the second spin array 200 includes one optical pulse in the optical pulse group B and one feedback signal in the feedback signal group fA.
  • the transmission path length of the optical signal in the optical feedback network 300, the transmission path of the optical pulse emitted by the pump source, or the pump source can be adjusted by adjusting the transmission path length of the optical signal in the optical feedback network 300.
  • the period of the light pulse is generated, so that the light pulse and the feedback signal can simultaneously enter the optical parametric oscillation cavity within a set time period.
  • FIG. 6 it is a schematic diagram of the structure of the optical parametric oscillation cavity, which includes two Bragg reflection areas and one parametric oscillation area, and the Bragg reflection areas are located at both ends of the parametric oscillation area.
  • the two Bragg reflection regions form a resonant cavity.
  • the feedback signal enters the optical parametric oscillation cavity, it travels back and forth between the two Bragg reflection regions to oscillate; and the optical pulse is coupled and input to the parametric oscillation region to produce nonlinearity with the feedback signal. After that, the optical pulse is coupled and filtered out from the parametric oscillation area. After filtering out, the optical signal left in the optical parametric oscillation cavity is the spin signal.
  • the optical signal output by the optical parametric oscillator is referred to as a spin signal.
  • Each optical parametric oscillation cavity in the second spin array 200 can generate a spin signal, and the second spin array 200 can output a group of N spin signals (respectively Q 1 , Q 2 , Q 3 ... ...Q N ) of the spin signal group Q.
  • optical parametric oscillation cavity included in the first spin array 100 and the action process of the signal in each optical parametric oscillation cavity are compared with the optical parametric oscillation cavity included in the second spin array 200 and the signal in each optical parametric oscillation cavity. The effect is the same, so I won't repeat them here.
  • a group of signals received by the second spin array 200 includes N optical pulses as an example for description here, for the case where a group of signals received by the second spin array 200 includes N electrical signals Similar to the case where the set of signals received by the second spin array 200 includes N optical pulses, the difference is that when the set of signals received by the second spin array 200 includes N electrical signals, the second spin array 200 needs to first The N electrical signals are converted into N light pulses.
  • the processing process for the N light pulses is the same as the processing method for the N light pulses when the set of signals received by the second spin array 200 includes N light pulses.
  • each optical parametric oscillator receives an electrical signal in the group of electrical signals, and converts the electrical signal into an optical pulse for processing. After that, each optical parametric oscillator is converted according to the converted electrical signal. The optical pulse and the received feedback signal generate a spin signal. Each optical parametric oscillator generates a spin signal according to the converted optical pulse and the received feedback signal. When a group of signals includes N optical pulses, each optical parametric oscillator generates a spin signal according to a received optical pulse and a feedback signal in the same way.
  • the optical feedback network 300 may be an all-optical feedback network, or may not be an all-optical feedback network. All-optical feedback means that the process of generating feedback signals is realized by optical signals.
  • Non-full optical feedback means that the feedback signal can be generated in the form of a circuit or photoelectric combined feedback. For example, in the case of non-full optical feedback, multiple optical pulses can be converted into electrical signals first, and the electrical signals can be loaded with feedback information through the action of a field programmable gate array (FPGA). The electrical signal with feedback information modulates the new optical signal (such as changing the phase or intensity of the optical signal), and the modulated optical signal is the feedback signal.
  • the implementation of the optical feedback network 300 is not limited in the embodiment of the present invention.
  • the optical feedback network 300 is an all-optical feedback network
  • the signal processing of the optical feedback network 300 does not require photoelectric conversion to generate a set of feedback signals (in fact, the feedback signals are also optical signals). And send the group of feedback signals to the first spin array 100 or the second spin array 200. Since the all-optical feedback network omits photoelectric conversion, the calculation time can be effectively shortened, and the signal processing efficiency of the optical feedback network 300 can be improved.
  • the optical feedback network 300 can process a set of spin signals input from any spin array based on preset first data, where the preset first data is related to the NP-hard problem to be solved. Relevant is obtained after NP-hard is mathematically abstracted. For different structures of the optical feedback network 300, the representation form of the first data is also different. In the embodiment of the present application, two types of optical feedback networks 300 are provided, which will be introduced separately below.
  • the optical feedback network 300 includes two signal processing modules (respectively a first signal processing module and a second signal processing module) and a problem loading module.
  • the first signal processing module receives each group of spin signals from the first spin array 100
  • the second signal processing module receives each group of spin signals from the second spin array 200.
  • the signal processing module can convert a one-dimensional set of spin signals into a two-dimensional optical signal matrix, and it can also convert a two-dimensional feedback signal matrix into a set of feedback signals. That is to say, the signal processing module can increase the signal dimension for subsequent follow-up
  • the problem loading module can also reduce the signal dimension to facilitate the input of a set of feedback signals into the spin array.
  • the signal matrix (such as the optical signal matrix and the feedback signal matrix) is a matrix composed of N*N optical signals.
  • the question loading module is provided with first data, and the first data is applied to the two-dimensional optical signal matrix to generate a feedback signal matrix.
  • the manner in which the signal processing module converts a one-dimensional set of spin signals into a two-dimensional optical signal matrix will be described below.
  • the first signal processing module can shape and collimate each spin signal in the spin signal group E, so that each spin signal in the spin signal group is transmitted in parallel without crosstalk.
  • the signal shaping and collimation can be realized by a micromirror array, or by other optical devices, which is not limited in the embodiment of the present application.
  • the first information processing module splits each parallel spin signal on the transmission plane of the spin signal.
  • Each spin signal is divided into N identical spin signals, and the N spin signals are split to form N columns of spin signals.
  • Spin signals, each column of spin signals includes N spin signals, and the N spin signals in each column are the same, forming an N*N optical signal matrix.
  • the light splitting of the spin signal may be realized by a cylindrical mirror, or may be realized by other optical devices, which is not limited in the embodiment of the present application.
  • FIG. 8A is a flow chart of processing the spin signal group E by the first signal processing module on an XY plane.
  • Each spin signal (E 1 , E 2 , E 3 ... E N ) in the spin signal group E is collimated by the micromirror array to form parallel N spin signals; after passing through the cylindrical mirror, N is formed Column spin signal.
  • the plane formed by each column of spin signals is perpendicular to the XY plane and parallel to the YZ plane.
  • FIG. 8B it is a flow chart of processing a spin signal by the first signal processing module on the YX plane.
  • One spin signal (take E 1 as an example) in the spin signal group E is collimated by the micromirror array; after passing through the cylindrical mirror, N spin signals E 1 are formed .
  • Fig. 9 shows the optical signal matrix generated by the first signal processing module.
  • the size of the optical signal matrix is N*N, and the optical signal in each column is the same as a spin signal in the spin signal group E.
  • the first spin array 100 may be located on the focal plane of the cylindrical mirror and the micromirror array in the first signal processing module.
  • the signal processing process performed by the problem loading module will be described below by taking the problem loading module receiving the optical signal matrix as shown in FIG. 9 to generate a feedback signal matrix as an example.
  • the first data set in the problem loading module can be abstracted as a two-dimensional action matrix.
  • the two-dimensional action matrix can be realized by a spatial light modulator.
  • the spatial light modulator modulates the optical phase or phase of the optical signal. strength.
  • an element in the two-dimensional action matrix indicates the degree of modulation of the optical phase or intensity of a signal in the signal matrix by the spatial light modulator.
  • the two-dimensional action matrix can also be realized by a digital micro-mirror array, which modulates the light intensity of the optical signal through the digital micro-mirror array.
  • an element in the two-dimensional action matrix indicates the degree of optical phase or intensity modulation of a signal in the signal matrix by the digital micromirror array.
  • the two-dimensional action matrix is a symmetric matrix, and the elements on the diagonal are not zero.
  • the signal processing process of the problem loading module can be abstracted as a matrix operation, and the feedback signal matrix is generated using the optical signal matrix generated by the first signal processing module and the two-dimensional matrix.
  • An optical signal of the optical signal matrix interacts with an element at a corresponding position in the two-dimensional matrix to generate a signal at a corresponding position in the feedback signal matrix.
  • the optical signal E 1 in the first row of the optical signal matrix interacts with the element a 1.1 z in the first row and the first column of the two-dimensional matrix to generate a signal a 1.1 E 1 at the corresponding position in the feedback signal matrix.
  • Figure 11 shows the feedback signal matrix received by the second signal processing module from the optical feedback network 300.
  • the size of the feedback signal matrix is N*N.
  • An optical signal in the matrix can be regarded as a spin in the spin signal group E. The product of a signal and an element in a two-dimensional matrix.
  • the signal combining and signal splitting processes are reversed, and the optical signal combining can be realized by a cylindrical mirror or by other optical devices, which is not limited in the embodiment of the present application.
  • the second signal processing module can shape and collimate each of the N feedback signals, so that each feedback signal in the feedback signal group fA is transmitted in parallel without crosstalk, and the signal can be shaped and collimated by the micromirror array
  • the implementation may also be implemented by other optical devices, which is not limited in the embodiment of the present application.
  • FIG. 12A is a flow chart of processing the feedback signal matrix by the second signal processing module on an XY plane.
  • the N columns of feedback signals form N feedback signals after passing through the cylindrical mirror; the N feedback signals are collimated by the micromirror array to form parallel N feedback signals.
  • FIG. 12B it is a flowchart of processing a column of signals in the feedback signal matrix by the second signal processing module on the YX plane.
  • the column of feedback signals are combined after passing through the cylindrical mirror to form a feedback signal, and then collimated by the micro-mirror array.
  • the second spin array 200 may be located on the focal plane of the cylindrical mirror and the micro-mirror array in the second signal processing module.
  • the second signal processing module forms N feedback signals, and transmits the N feedback signals to each optical parametric oscillation cavity in the second spin array 200, and each optical parametric oscillation cavity receives a feedback signal.
  • the feedback signal received by each optical parametric oscillation cavity in the second spin array 200 is generated after the spin signal output by the corresponding optical parametric oscillation cavity in the first spin array 100 is processed by the optical feedback network 300.
  • the spin signal output by the optical parametric oscillator in the first spin array 100 is E 1
  • the spin signal output by the optical parametric oscillator in the first spin array 100 is E 2
  • the corresponding optical parametric oscillator in the second spin array 200 receives from the optical feedback network 300
  • FIG. 13 is a schematic structural diagram of another optical computing device provided by an embodiment of the present invention.
  • the first spin array 100 and the second spin array 200 each include four optical parametric oscillating cavities as an example, a schematic structural diagram of an optical feedback network 300, and the optical feedback network 300 includes multiple stages.
  • the Mach-Zehnder Interferometer Unit (MZIU) is composed of a connected Mach-Zehnder Interferometer Unit, each MZIU can receive two optical signals, and each MZIU includes multiple Mach-Zehnder Interferometers (Mach-Zehnder Interferometer , MZI) and optical switches.
  • the small size of the Mach-Zehnder interference unit makes the structure of the optical computing device more compact and can be implemented on-chip, ensuring the stability of the system.
  • the MZI is used to realize the mutual interference between the two input optical signals, and the corresponding output signal is controlled by the MZI through the phase parameter of the MZI.
  • the optical switch in each MZIU can adjust the optical signal from different directions (that is, the transmission direction is the optical signal from the first spin array 100 to the second spin array 200 and from the second spin array 200 to the first spin array 200). 100 of the optical signal) transmission path.
  • the phase parameters of the MZI included in the MZIU can be different.
  • the optical signals from different directions pass through the same number of MZIs, and the MZI with different phase parameters passes through
  • the order is the same.
  • the transmission direction of the extended optical signal in the MZIU is set in a bilaterally symmetrical manner, and the phase parameters of the two MZIs in the bilaterally symmetrical position are the same.
  • the optical switch is identified by SW in Fig. 14, and the phase parameters of the MZI on the left and right sides of the MZIU are the same.
  • the phase parameter of MZI-1 and MZI-3 is ⁇ i
  • the phase parameter of MZI-2 is ⁇ i .
  • the optical signal received from the first spin array 100 and the optical signal received from the second spin array 200 have different transmission paths, but the number of MZIs passed through is the same, and the order of the MZIs passed through with different phase parameters the same.
  • FIG. 15A it is a transmission path in the MZIU of an optical signal whose transmission direction is from the first spin array 100 to the second spin array 200. Pass through MZI-2 with a phase parameter of ⁇ i and MZI-3 with a phase parameter of ⁇ i in turn.
  • FIG. 15B it is the transmission path in the MZIU of the optical signal whose transmission direction is from the first spin array 100 to the second spin array 200. Pass through MZI-2 with a phase parameter of ⁇ i and MZI-1 with a phase parameter of ⁇ i in turn.
  • the embodiment of the present application does not limit the cascading manner of the multiple MZIUs included in the optical feedback network 300. It is only necessary to ensure that multiple MZIUs can form a UDU + structure, where U is a unitary matrix, U + is a transposed matrix of U, and D is a diagonal matrix.
  • the first data in this optical feedback network 300 is the phase parameter of the MZI in the cascaded multiple MZIUs.
  • the optical computing device provided by the embodiment of the present application further includes a detector array 400.
  • the detector array 400 can be connected to any spin array to detect a set of spin signals output by the spin array.
  • the phase of each spin signal may randomly collapse to 0 or ⁇ .
  • the phase of the spin signal collapses to 0 or ⁇ it is considered that the set of spin signals currently output corresponds to the first data, and the final calculation result can be determined according to the set of spin signals.
  • the connection between the detector array 400 and the first spin array 100 is taken as an example.
  • the detector array 400 may include N detectors, and each detector is connected to an optical parametric oscillator in the first spin array 100.
  • the optical parametric oscillator connected to each detector is different.
  • Each detector can detect the phase of the spin signal generated by the connected optical parametric oscillator.
  • the phase of the spin signal collapses to 0 or ⁇ , and the final calculation result is determined according to the set of spin signals.
  • FIG. 16 is another optical computing device provided by an embodiment of the present invention. It should be noted that the following description of the workflow is also applicable to all optical computing devices described above. As shown in FIG. 16 and FIG.
  • the first spin array 100 can receive a first group of signals including N signals (such as the optical pulse group A in the embodiment of the present application), and based on the first The group signal generates a first group of spin signals including N spin signals (such as the spin signal group E in the embodiment of the present application), where N is a positive integer greater than 2.
  • the optical feedback network 300 may receive a first set of spin signals, and generate a first set of feedback signals including N feedback signals according to the first set of spin signals and the set first data (such as the feedback signal in the embodiment of the present application).
  • Signal group fA It can be understood that the signal transmission process from the first spin array 100 to the optical feedback network 300 described here is similar to the process one in the foregoing embodiment.
  • the second spin array 200 can receive the first set of feedback signals (such as the feedback signal group fA in the embodiment of the present application) and the first set of signals (such as the feedback signal group fA in the embodiment of the present application).
  • the second spin array 200 can generate a second set of spin signals including N spin signals according to the first set of feedback signals and the first set of signals (such as the spin signal set F in the embodiment of the present application).
  • the optical feedback network 300 receives the second set of spin signals, and generates a second set of feedback signals including N feedback signals (such as the feedback signal set fB in the embodiment of the present application) according to the second set of spin signals and the first data.
  • the signal transmission process from the second spin array 200 to the optical feedback network 300 is similar to the process two in the foregoing embodiment.
  • the first spin array 100 may also receive a second group of feedback signals and a second group of signals including N signals (such as the optical pulse group A'in the embodiment of the present application), and according to the second group of feedback signals and the first group of signals
  • the two sets of signals generate a third set of spin signals including N spin signals (such as the spin signal set P in the embodiment of the present application).
  • the optical feedback network 300 receives the third set of spin signals, generates a new set of feedback signals according to the received third set of spin signals, and sends the newly generated set of feedback signals to the second spin array 200.
  • the second spin array 200 generates a set of spin signals according to the received feedback signal and the second set of signals, and transmits the set of spin signals to the optical feedback network 300.
  • the optical feedback network 300 receives the set of spin signals, generates a set of feedback signals according to the spin signals received from the second spin array, and feeds back the set of feedback signals to the first spin array 100. This cycle is repeated until the detector array 400 detects that the phase of each spin signal output by the first spin array 100 collapses to 0 or ⁇ .
  • the optical computing device since the first spin array and the second spin array process the n optical pulses in a group of optical pulses in parallel, the number of optical pulses is increased while increasing the number of optical pulses. , Will not increase the computing time of the optical computing device, and can effectively improve the computing efficiency of the optical computing device.
  • the optical computing device improved by the embodiment of the present invention has a simple structure, and the interaction is realized through mutual injection coupling of the dual-spin array, which greatly simplifies the architecture of the optical Ising machine, and may realize the integration of the all-optical Ising machine on the chip.
  • the optical feedback network adopts an all-optical feedback network, no photoelectric conversion is required, and parallel search of spin configuration is realized, which greatly shortens the calculation time, improves the stability of the system, and improves the calculation efficiency of the optical computing device. . Moreover, in the calculation process, the injection signal of photoelectric detection is not required, which improves the signal transmission time.
  • first spin array and the second spin array in the embodiment of the present invention may not be limited to spin arrays that can generate light pulses, and dual-spin arrays that can be mapped in other ways may also be used.
  • Rotational arrays such as laser arrays composed of lasers, polarized sub-arrays composed of polarons, etc.
  • the optical computing device provided by the embodiment of the present invention has a simple structure and can be implemented on a chip, and the entire calculation process is implemented by means of optical signals, the signal transmission speed is fast, and the calculation speed is greatly improved. Therefore, the implementation of the present invention
  • the improved optical computing device can be applied to a neural network system, for example, can be used to realize feedback control in a neural network system.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Hardware Design (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Signal Processing (AREA)
  • Microelectronics & Electronic Packaging (AREA)
  • Probability & Statistics with Applications (AREA)
  • Neurology (AREA)
  • Optical Modulation, Optical Deflection, Nonlinear Optics, Optical Demodulation, Optical Logic Elements (AREA)

Abstract

一种光计算设备以及计算方法，用以提供一种运行效率较高的光伊辛机。所述光计算设备包括第一自旋阵列、光反馈网络以及第二自旋阵列，光反馈网络分别与第一自旋阵列和第二自旋阵列连接。第一自旋阵列可以接收包括N个光脉冲或N个电信号的第一组信号，产生包括N个自旋信号的第一组自旋信号。光反馈网络可以接收第一组自旋信号；以及根据第一组自旋信号以及设置的第一数据生成包括N个反馈信号的第一组反馈信号。第二自旋阵列接收第一组反馈信号以及第一组信号；根据第一组反馈信号以及第一组光脉冲产生包括N个自旋信号的第二组自旋信号。第一自旋阵列和第二自旋阵列可以并行处理对多个信号，可以提高光运算设备的计算效率。

Description

一种光计算设备以及计算方法 技术领域
本申请涉及信息技术领域，尤其涉及一种光计算设备以及计算方法。
背景技术
药物设计、交通物理规划、网络资源分配等许多领域中涉及的组合优化问题属于非确定性多项式时间困难(non-deterministic polynomial hard，NP-hard)问题，因其求解时间随问题中变量的数目呈指数增加而很难在合理的时间内进行求解。这些组合优化问题可以先转化为伊辛模型，借由光伊辛机对该伊辛模型进行模拟求解。
伊辛模型描述了表示磁结构中的N个节点物质相变的随机过程，其中的一个节点表征一个磁偶极子，且该磁偶极子具有+1或-1的自旋状态，每个节点与许多其他节点之间可以耦合作用，产生物质相变(如自旋状态发生改变)。
光伊辛机利用多个光信号(或光脉冲)构成格点网络的物理现象模拟求解伊辛模型。光伊辛机中利用光信号(或光脉冲)的相位表示伊辛模型中的节点的自旋状态，光信号或光脉冲间相互耦合干涉，改变光信号(或光脉冲)的相位。
目前光伊辛机中实现方案中光信号通常是串行运行，随着伊辛模型中的节点数量增加，需要引入较多的光信号，光信号数量的增加，也会增大光信号在光伊辛机中的传输时间，导致运算时间大大增加，限制了光伊辛机的运算效率。
发明内容
本申请提供一种光计算设备以及计算方法，用以提供一种运行效率较高的光伊辛机，提高计算速度。
第一方面，本申请提供了一种光计算设备，该设备包括第一自旋阵列、光反馈网络以及第二自旋阵列，光反馈网络分别与第一自旋阵列和第二自旋阵列连接。所述第一自旋阵列可以接收第一组信号，并根据所述第一组信号产生第一组自旋信号，所述第一组自旋信号包括N个自旋信号，N为不小于2的整数。所述第一自旋阵列可以将所述第一组自旋信号发送给所述光反馈网络。所述光反馈网络在接收到所述第一组自旋信号后，可以根据所述第一组自旋信号以及设置的第一数据生成第一组反馈信号，其中，所述第一组反馈信号包括N个反馈信号。所述光反馈网络可以将所述第一组反馈信号发送给所述第二自旋阵列。所述第二自旋阵列除了接收到第一组反馈信号，还可以接收所述第一组信号，所述第二自旋阵列可以根据所述第一组反馈信号以及所述第一组信号产生第二组自旋信号，其中，所述第二组自旋信号包括N个自旋信号。
第一自旋阵列或第二自旋阵列每次产生的一组自旋信号能够指示光计算设备处理光信号过程中的一个中间计算结果。当第一自旋阵列或第二自旋阵列产生的一组自旋信号的相位达到预设值时，此时该组自旋信号可以指示光计算设备处理光信号过程的最终计算结果。
在本申请提供的光计算设备中，所述第一自旋阵列与第二自旋阵列接收相同的一组信号，第一自旋阵列可以通过光反馈网络向第二自旋阵发送反馈信号，以使得第二自旋阵列也可以产生一组自旋信号。第一自旋阵列和第二自旋阵列对一组信号中的信号的处理过程是并行的，因此，在增加信号的数量的同时，不会增加光计算设备的运算时间，可以有效提高光运算设 备的计算效率。
在一种可能的设计中，所述第二自旋阵列可以将所述第二组自旋信号发送给所述光反馈网络，所述光反馈网络接收所述第二组自旋信号，还可以根据所述第二组自旋信号以及所述第一数据生成第二组反馈信号，其中，所述第二组反馈信号包括N个反馈信号。所述第一自旋阵列在接收到第二组反馈信号时，还可以接收第二组信号。所述第一自旋阵列还可以根据所述第二组反馈信号以及所述第二组信号产生第三组自旋信号，其中，所述第三组自旋信号包括N个自旋信号。
本申请提供的光计算设备中，所述第一自旋阵列与所述第二自旋阵列通过所述光反馈网络耦合建立连接。采用这种连接方式，使得光计算设备的结构得到简化，所述第一自旋阵列与所述第二自旋阵列能够接收相同的一组信号，每个自旋阵列可以并行处理该组信号，并且能够通过光反馈网络互相发送反馈信号，所述第一自旋阵列与所述第二自旋阵列之间通过所述光反馈网络的信号传输过程更加简单、高效，使得光计算设备的计算效率得以提高，并且，所述第一自旋阵列和所述第二自旋阵列通过所述光反馈网络互相发送反馈信号能够实现所述第一自旋阵列和所述第二自旋阵列产生的自旋信号的相位互锁。此外，本申请提供的光计算设备中，由于所述第一自旋阵列与所述第二自旋阵列通过所述光反馈网络耦合建立连接，这种方式使得反馈信号第一自旋阵列和第二自旋阵列通过光反馈网络相互注入反馈信号，多个反馈信号同时注入，无需增加光纤长度，可以有效减少信号传递时间，减少了光计算设备的计算时间，有利于提高光计算设备的运算效率。
在一种可能的设计中，所述第一自旋阵列包括N个光参量振荡器，所述第一自旋阵列中的每个光参量振荡器可以接收所述第一组信号的一个信号。
在本申请提供的光计算设备中，所述第一自旋阵列包括的N个光参量振荡器，能够并行的对第一组信号中的多个信号进行处理，每个光参量振荡器能够产生一个自旋信号，所述第一自旋阵列并行处理第一组信号的方式，能够有效缩短光计算设备的运算时间，可以有效提高所示第一自旋阵列的信号处理效率，进而，可以保证光计算设备的计算过程能够高效进行。
在一种可能的设计中，所述第二自旋阵列包括N个光参量振荡器，所述第二自旋阵列中的每个光参量振荡器可以接收所述第一组信号的一个信号，还可以接收所述第一组反馈信号中的一个反馈信号，每个光参量振荡器接收的一个信号和一个反馈信号在该光参量振荡器中进行相互作用。所述第二自旋阵列中的N个光参量振荡器与所述第一自旋阵列中的N个光参量振荡器之间存在一一对应的关系，所述第二自旋阵列和所述第一自旋阵列中具有对应关系的光参量振荡器接收的信号相同。
在本申请提供的光计算设备中，所述第二自旋阵列包括的N个光参量振荡器能够并行的对N个反馈信号和一组信号中的多个信号进行处理，每个光参量振荡器能够产生一个自旋信号，所述第二自旋阵列并行处理一组信号中的多个信号的方式可以提高所述第二自旋阵列的信号处理效率，可以保证光运算设备的计算过程能够高效进行；且所述第二自旋阵列和所述第一自旋阵列中光参量振荡器的对应关系，能够保证所述第一自旋阵列每个光参量振荡器和所述第二自旋阵列对应的光参量振荡器接收到相同的光脉冲。
在一种可能的设计中，所述光反馈网络包括所述第一信号处理模块、所述问题加载模块以及所述第二信号处理模块。所述问题加载模块分别连接所述第一信号处理模块和所述第二信号处理模块。
所述第一信号处理模块可以接收所述第一组自旋信号，所述第一信号处理模块可以将所述第一组自旋信号转换为第一光信号矩阵，所述第一光信号矩阵中包括N*N个光脉冲。所述 第一信号处理模块可以将所述第一光信号矩阵传输至所述问题加载模块，所述问题加载模块可以根据所述第一光信号矩阵和所述第一数据生成第一反馈信号矩阵，所述第一反馈信号矩阵中包括N*N个光信号。之后，所述第二信号处理模块可以将所述第一反馈信号矩阵转换为所述第一组反馈信号。
在本申请提供的光计算设备中，所述光反馈网络内部在进行信号处理时，不需要进行光电转换，可以将一组自旋信号转换为一组反馈信号，能够有效的缩短计算时间，进而提高光计算设备的计算效率。
在一种可能的设计中，所述第二信号处理模块还可以接收所述第二组自旋信号，并将所述第二组自旋信号转换为第二光信号矩阵，其中，所述第二光信号矩阵中包括N*N个光脉冲，所述第二信号处理模块可以将所述第二光信号矩阵发送给所述问题加载模块；之后，所述问题加载模块可以根据所述第二光信号矩阵和所述第一数据生成第二反馈信号矩阵，其中，所述第二反馈信号矩阵中包括N*N个光信号；所述问题加载模块可以将所述第二反馈信号矩阵发送给所述第一信号处理单元，所述第一信号处理模块可以将所述第二反馈信号矩阵转换为所述第二组反馈信号。
在本申请提供的光计算设备中，所述光反馈网络内部在进行信号处理时，不需要进行光电转换，可以将一组自旋信号转换为一组反馈信号，能够有效的缩短计算时间，进而提高光计算设备的计算效率。
在一种可能的设计中，所述光反馈网络可以包括多个级联的马赫-曾德尔干涉单元，每个马赫-曾德尔干涉单元包括间隔设置的马赫-曾德尔干涉仪和光开关。
在本申请提供的光计算设备中，马赫-曾德尔干涉单元尺寸较小，能够形成具备紧凑结构的光计算设备，能够使得该光计算设备可以在片上实现，保证了系统的稳定性。
在一种可能的设计中，所述光计算设备还包括探测器阵列，所述探测器阵列可以连接任一自旋阵列，当连接到第一自旋阵列时，可以探测所述第三组自旋信号；当所述第三组自旋信号的相位为预设值时，根据所述第三组自旋信号获得所述第一数据的计算结果。
在本申请提供的光计算设备中，所述探测器阵列可以较为方便的通过探测第一自旋阵列产生的一组自旋信号的相位，确定最终的输出结果。
在一种可能的实现方式中，第一自旋阵列和第二自旋阵列接收的一组信号(如第一组信号和第二组信号)可以包括多个光脉冲，也可以是多个电信号，其中一个光脉冲或一个电信号可以称为一组信号中一个信号。示例性的，光脉冲或电信号的数目可以是N，当第一自旋阵列和第二自旋阵列接收的一组信号包括多个电信号时，第一自旋阵列和第二自旋阵列可以将该多个电信号转换为一组光脉冲，该组光脉冲包括N个光脉冲。
示例性的，所述第一组信号包括第一组光脉冲，所述第二组信号包括第二组光脉冲，其中，所述第二组光脉冲的幅值大于所述第一组光脉冲幅值。
在本申请提供的光计算设备中，第一自旋阵列和第二自旋阵列可以对光脉冲进行直接处理，也可以进行光电转换，之后再对转换后的光脉冲进行处理，能够适用于不同的应用场景，扩展了应用范围。
第二方面，本申请提供了一种计算方法，有益效果可以参见第一方面的相关描述，此处不再赘述。该方法由光计算设备执行，所述光计算设备包括第一自旋阵列、第二自旋阵列以及分别与所述第一自旋阵列和所述第二自旋阵列连接的光反馈网络，该方法包括：所述第一自旋阵列先接收第一组信号，并根据所述第一组信号产生第一组自旋信号，所述第一组自旋信号包括N个自旋信号，N为不小于2的整数；所述第一自旋阵列将所述第一组自旋信号发 送给所述光反馈网络。
所述光反馈网络接收到所述第一组自旋阵列后，根据所述第一组自旋信号以及设置的第一数据生成第一组反馈信号，其中，所述第一组反馈信号包括N个反馈信号。所述光反馈网络将所述第一组反馈信号发送给所述第二自旋阵列。
所述第二自旋阵列接收所述第一组信号以及所述第一组反馈信号，可以根据所述第一组反馈信号以及所述第一组信号产生第二组自旋信号，其中，所述第二组自旋信号包括N个自旋信号。
在一种可能的设计中，所述第二自旋阵列可以将所述第二组自旋信号发送给所述光反馈网络，所述光反馈网络接收到所述第二自旋阵列发送的第二组自旋信号后，可以根据所述第二组自旋信号以及所述第一数据生成第二组反馈信号，其中，所述第二组反馈信号包括N个反馈信号。所述光反馈网络可以将所述第二组反馈信号发送给所述第一自旋阵列。
所述第一自旋阵列还可以接收第二组信号；
所述第一自旋阵列根据接收到的所述第二组反馈信号以及所述第二组信号产生第三组自旋信号，所述第三组自旋信号包括N个自旋信号。
在一种可能的设计中，所述光计算设备还包括探测器阵列，所述探测器阵列能探测所述第三组自旋信号；确定所述第三组自旋信号中每个自旋信号的相位。
当所述第三组自旋信号的相位为预设值时，所述探测器阵列可以根据所述第三组自旋信号获得所述第一数据的计算结果。
在一种可能的设计中，所述第一自旋阵列接收第一组信号时，所述第一自旋阵列中的每个光参量振荡器能接收所述第一组信号中的一个信号。
在一种可能的设计中，所述第二自旋阵列包括N个光参量振荡器，所述第二自旋阵列在接收所述第一组信号以及所述第一组反馈信号时，所述第二自旋阵列中的每个光参量振荡器可以接收所述第一组信号中的一个信号以及所述第一组反馈信号中的一个反馈信号，其中，所述第二自旋阵列中的N个光参量振荡器与所述第一自旋阵列中的N个光参量振荡器一一对应，所述第二自旋阵列和所述第一自旋阵列中具有对应关系的光参量振荡器接收的光脉冲相同。
第三方面，本申请提供了一种光计算芯片，所述光计算芯片可以包括如第一方面或第一方面的任意一种可能的实现方式中所述的光计算设备。
附图说明
图1为本申请提供的一种光计算设备的结构示意图；
图2为本申请提供的一种光计算设备中信号传输的流程图；
图3为本申请提供的另一种光计算设备中信号传输的流程图；
图4为本申请提供的两个自旋阵列接收到的光脉冲组的示意图；
图5为本申请提供的一种第二自旋阵列的结构示意图；
图6为本申请提供的一种光参量振荡腔的结构示意图；
图7为本申请提供的一种光反馈网络的结构示意图；
图8A为本申请提供的XY平面第一信号处理模块对自旋信号组E的处理流程图；
图8B为本申请提供的YZ平面第一信号处理模块对自旋信号E 1的处理流程图；
图9为本申请提供的第一信号处理模块产生的光信号矩阵示意图；
图10为本申请提供的一种问题加载模块对信号处理的流程图；
图11为本申请提供的一种反馈信号矩阵的示意图；
图12A为本申请提供的XY平面第二信号处理模块对反馈信号矩阵的处理流程图；
图12B为本申请提供的YZ平面第一信号处理模块对反馈信号矩阵中的一列信号的处理流程图；
图13为本申请提供的一种光计算设备的结构示意图；
图14为本申请提供的一种MZIU的结构示意图；
图15A为本申请提供的一种MZIU中信号的传输路径示意图；
图15B为本申请提供的另一种MZIU中信号的传输路径示意图；
图16为本申请提供的一种光计算设备的结构示意图；
图17为本申请提供的一种计算方法示意图。
具体实施方式
本申请提供了一种计算方法以及光计算设备，用以提供一种运行效率较高的光伊辛机。
如图1所示，为本申请实施例提供的一种光计算设备，光计算设备10其中包括两个自旋阵列和光反馈网络300，两个自旋阵列分别用第一自旋阵列100和第二自旋阵列200区别，光反馈网络300分别与第一自旋阵列100和第二自旋阵列200连接。
本申请实施例中自旋阵列(第一自旋阵列100或第二自旋阵列200)能够接收一组信号(以及光反馈网络300反馈的多个反馈信号)，根据接收到的一组信号(和多个反馈信号)产生多个自旋信号。
本申请实施例并不限定自旋阵列接收的一组信号中包括的信号类型，该组信号可以包括多个光脉冲，也可以包括多个电信号(如电脉冲)。例如光信号和电信号的数量可以为N。当接收到一组电信号(包括N个电信号)时，自旋阵列能够将接收到的一组电信号转换为N个光脉冲，当转换为N个光脉冲后，自旋阵列可以根据N个光脉冲(和N个反馈信号)转换为N个自旋信号。在本申请实施例中仅是以自旋阵列接收的信号为一组光脉冲为例进行说，该组光脉冲包括N个光脉冲。
光反馈网络300能够接收一个自旋阵列产生的多个自旋信号，并基于预先设置的第一数据，对第一自旋阵列100产生的多个自旋信号进行处理，产出多个反馈信号，将产生的多个反馈信号反馈给另一个自旋阵列。
在本申请实施例中以第一自旋阵列100(或第二自旋阵列200)接收的一组信号为包括N个光脉冲的一组光脉冲为例进行说明，对于第一自旋阵列100(或第二自旋阵列200)包括N个电信号的情况与第一自旋阵列100(或第二自旋阵列200)接收的一组光脉冲的情况类似，区别在于第一自旋阵列100(或第二自旋阵列200)接收的一组信号包括N个电信号时，第一自旋阵列100(或第二自旋阵列200)需要先将该N个电信号转换为N个光脉冲。对于后续处理该N个光脉冲的方式与第一自旋阵列100(或第二自旋阵列200)接收一组光脉冲时，处理N个光脉冲的方式相同。
下面对第一自旋阵列100、光反馈网络300以及第二自旋阵列200的信号传输过程进行说明，第一自旋阵列100、光反馈网络300以及第二自旋阵列200之间的信号传输按照信号传输的方向，可以包括两个过程：第一自旋阵列100产生的自旋信号经过光反馈网络300到达第二自旋阵列200的信号传输(过程一)，以及第二自旋阵列200产生的自旋信号经过光反馈网络300到达第一自旋阵列100的信号传输(过程二)：
过程一、第一自旋阵列100产生的自旋信号经过光反馈网络300到达第二自旋阵列200 的信号传输过程。
如图2所示，第一自旋阵列100可以接收一组光脉冲(为方便说明，第一自旋阵列100接收的该组光脉冲用光脉冲组A表示)，该光脉冲组A中包括多个相同的光脉冲，分别为A 1、A 2、A 3、……A N。
第一自旋阵列100可以根据光脉冲组A产生一组自旋信号(为方便说明，该组自旋信号用自旋信号组E表示)，该自旋信号组E中多个自旋信号，分别为E 1、E 2、E 3……E N，之后，将该自旋信号组E输入至光反馈网络300。
光反馈网络300接收到自旋信号组E后，利用预设的第一数据对自旋信号组E进行处理，产生一组反馈信号(为方便说明，该组反馈信号用反馈信号组fA表示)，反馈信号组fA包括多个反馈信号，分别为fA 1、fA 2、……、fA N，将反馈信号组fA输入值第二自旋阵列200。
过程二、第二自旋阵列200产生的自旋信号经过光反馈网络300到达第一自旋阵列100的信号传输过程。
如图3所示，第二自旋阵列200除了接收反馈信号组fA外，还可以接收与光脉冲组A相同的一组光脉冲，为了区分接收光脉冲组的不同自旋阵列，第二自旋阵列200接收到的与光脉冲组A相同的一组光脉冲用光脉冲组B表示，该光脉冲组B包括多个光脉冲，分别为B 1、B 2、B 3、……B N。B 1与A 1为相同的光脉冲，B 2与A 2为相同的光脉冲，B ,3与A 3为相同的光脉冲，B ,N与A N为相同的光脉冲。
第二自旋阵列200根据反馈信号组fA和光脉冲组B，产生一组自旋信号(为方便说明，该组自旋信号用自旋信号组F表示)，该自旋信号组F中多个自旋信号，分别为F 1、F 2、F 3……F N，之后，将该自旋信号组F输入至光反馈网络300。
光反馈网络300利用预设的第一数据对该自旋信号组F进行处理，产生一组反馈信号(为方便说明，该组反馈信号用反馈信号组fB表示)，反馈信号组fB包括多个反馈信号，分别为fB 1、fB 2、……、fB N，将反馈信号组fB输入值第一自旋阵列100。
上述过程一和过程二中，第一自旋矩阵和第二自旋矩阵通过光反馈网络300可以互相发送反馈信号，可以达到第一自旋阵列和第二自旋阵列产生的自旋信号的相位互锁的效果。
第一自旋阵列100和第二自旋阵列200对一组光脉冲中的光脉冲的处理过程是并行的，可以有效减少光计算设备10的运算时间，若增加光脉冲的数量，也不会增加光计算设备的运算时间，有效的保证了光运算设备的计算效率。
第一自旋阵列100在接收了光脉冲组A之后，还会继续接收其他组光脉冲，以之后接收一组光脉冲为光脉冲组A’为例，第一自旋阵列100在接收到光脉冲组A’后，根据反馈信号组fB和光脉冲组A’，产生一组自旋信号(为方便说明，该组自旋信号用自旋信号组P表示)，该自旋信号组P中多个自旋信号，分别为P 1、P 2、P 3……P N。第一自旋阵列100将自旋信号组P输入至光反馈网络300，该光反馈网络300可以根据该自旋信号组P与预设的第一数据产生一组反馈信号(为方便说明，该组反馈信号用反馈信号组fA’表示)，反馈信号组fA’包括多个反馈信号，将反馈信号组fA’输入至第二自旋阵列200。该过程与过程一类似，区别在于，第一自旋阵列100是根据后续接收到的光脉冲组以及光反馈网络300接收的多个反馈信号产生一组自旋信号的。
第二自旋阵列200组还可以接收与光脉冲组A’相同的一组脉冲光(可以用光脉冲组B’表示)；利用光脉冲组B’和光反馈网络300输入的反馈信号组fA’产生一组自旋信号(为方便说明，该组自旋信号用自旋信号组Q表示)，该自旋信号组Q中多个自旋信号，分别为Q 1、Q 2、Q 3……Q N，第一自旋阵列100将自旋信号组Q输入至光反馈网络300，该光反馈网络300 可以根据该自旋信号组与预设的第一数据产生一组反馈信号(为方便说明，该组反馈信号用反馈信号组fB’表示)，反馈信号组fB’包括多个反馈信号，将反馈信号组fB’输入至第一自旋阵列100。该过程与过程二类似，区别在于，第二自旋阵列200接收的信号(光脉冲以及反馈信号)是不同的。
光计算设备中的第一自旋阵列100和第二自旋阵列200后续还会持续接收相同的光脉冲组，执行与上述过程相似的信号传输过程。也就是说，光计算设备中的第一自旋阵列100和第二自旋阵列200每接收到一组相同的光脉冲，就会执行一次与上述过程相似的信号传输过程。
第一自旋阵列100或第二自旋阵列200在类似过程一和过程二的信号处理过程中产生的一组自旋信号指示的是光计算设备10的一次中间计算结果，直至第一自旋阵列100或第二自旋阵列200产生的一组自旋信号中的N个自旋信号的相位坍缩到0或者π，则该组自旋信号则可以指示最终的计算结果。
应需理解的是，在光计算设备内部循环一次类似过程一和过程二的信号传输过程时，第一自旋阵列100和第二自旋阵列200接收到的一组光脉冲是相同的(如光脉冲组A与光脉冲组B相同、光脉冲组A’与光脉冲组B’相同)。在一种实现方式中，第一自旋阵列100和第二自旋阵列200接收的每组光脉冲组可以由同一个泵浦源发出，该泵浦源发出一组光脉冲后，可以通过分光，将该组光脉冲分为两组相同的光脉冲，分别发送给第一自旋阵列100和第二自旋阵列200。在另一种实现方式中，第一自旋阵列100和第二自旋阵列200接收的每组光脉冲组可以由同一个泵浦源发出的；第一自旋阵列100和第二自旋阵列200每次循环一次过程一和过程二的信号传输过程时，该泵浦源可以间隔设定时间发送两组相同的光脉冲，其中一组光脉冲发送给第一自旋阵列100，另一组光脉冲发送给第二自旋阵列200。例如该泵浦源可以每隔时间T/2产生一组光脉冲，在时间T内产生相同两组光脉冲，不同时间T内的产生光脉冲可以不同。
在另一种实现方式中，第一自旋阵列100和第二自旋阵列200接收的每组光脉冲组也可以由两个不同泵浦源发出，但这两个不同泵浦源发出两组光脉冲是相同的。
另外，一个自旋阵列(以第二自旋阵列200为例)需要根据接收一组光脉冲以及从光反馈网络300接收的一组反馈信号产生一组自旋信号，其中，该第二自旋阵列200接收的该组反馈信号是光反馈网络300根据第一自旋阵列100利用相同的一组光脉冲(以及另一组反馈信号)产生的。也就是说，第二自旋阵列200接收到该组光脉冲的时间，应该晚于所述第一自旋阵列100接收相同的该组光脉冲的时间，这样，第二自旋阵列200才能在接收到该组光脉冲时，可以从光反馈网络300中获取该组反馈信号。
也就是说，虽然两组自旋阵列接收的是相同的一组光脉冲，该组光脉冲到达两组自旋阵列的时间却不同，存在时间差值。该时间差值可以使得第二自旋阵列200能够同时接收到该组光脉冲和一组反馈信号。该时间差值是根据从第一自旋阵列100接收到该组光脉冲到光反馈网络300向第二自旋阵列200反馈一组反馈信号的时间确定的，与一组光脉冲在光计算设备中经过处理生成对应的反馈信号的时间有关。
以第一自旋阵列100和第二自旋阵列200接收的每组光脉冲组可以由同一个泵浦源发出为例，若泵浦源发出每组光脉冲的周期为T，也就是说泵浦源每隔T产生一组光脉冲，可以设置对泵浦源发出一组光脉冲分光后，该组光脉冲到达第一自旋阵列100与第二自旋阵列200的时间差值为T/2。
如图4所示，为两个自旋阵列接收到的光脉冲组的示意图，第一自旋阵列100利用光脉冲组A产生的自旋信号，经过光反馈信号处理生成反馈信号组E，注入到第二自旋阵列200中。
第二自旋阵列200利用接收的光脉冲组B与反馈信号组E产生新的自旋信号，新的自旋信号经过光反馈信号处理生成反馈信号组F，注入到第一自旋阵列100中，如此进行循环叠加。
而泵浦源可以对每次产生的一组光脉冲进行调整。例如泵浦源可以增大该组光脉冲中每个光脉冲的幅值，以产生与T之前产生的光脉冲组不同的一组光脉冲组，并将产生的光脉冲组分别输入第一自旋阵列100和第二自旋阵列200。从而，光计算设备中第一自旋阵列100和第二自旋阵列200以及光反馈网络300，能够根据泵浦源产生的光脉冲组改变第二自旋阵列200产生的一组自旋信号。
第一自旋阵列100与第二自旋阵列200继续接收相同的一组光脉冲，执行上述过程，通过光反馈网络300互相发送反馈信号。由于本申请实施例中第一自旋阵列100与第二自旋阵列200通过光反馈网络300耦合建立连接，对每组光脉冲中的光脉冲的处理过程是并行的，增加光脉冲的数量，也不会增加光计算设备的运算时间，可以有效提高光运算设备的计算效率。
前述说明中，对光计算设备中的信号传输过程进行了说明，下面对光计算设备中各个组成部分对信号的处理方式进行说明：
(一)、第一自旋阵列100以及第二自旋阵列200。
第一自旋阵列100与第二自旋阵列200的工作原理类似，此处仅以第二自旋阵列200为例进行说明。
如图5所示，第二自旋阵列200包括多个并行的光参量振荡腔，光参量振荡腔的数量与接收的一组光脉冲中包括的光脉冲数量相同。每个光参量振荡腔与光反馈网络300连接，从光反馈网络300接收一组反馈信号组中的一个反馈信号。其中，光参量振荡腔包括但不限于光泵浦的光参量振荡腔、电泵浦的光参量振荡腔。
以第二自旋阵列200接收的一组光脉冲为光脉冲组B，光脉冲组B中的光脉冲分别为B 1、B 2、B 3、……B N，光脉冲数量等于N为例，第二自旋阵列200包括N个光参量振荡腔(optical parametric oscillator)，每个光参量振荡腔可以接收光脉冲组B中的一个光脉冲，这样，N个光参量振荡腔可以接收光脉冲组B中的N个光脉冲。使得第二自旋阵列200对光脉冲组B的N个光脉冲可以并行处理，能够较好的提升信号处理效率。
其中，光参量振荡腔(也称为光参量振荡器)是一种能够基于光信号(或光脉冲)频率进行振荡的参量振荡器。光参量振荡腔可以将输入到光参量振荡腔的光信号(或光脉冲)通过非线性的光学相互作用产生新的光信号(或光脉冲)。
在本申请实施例中，第二自旋阵列200在接收到光脉冲组B时，还会接收来自光反馈网络300的反馈信号组fA。光反馈网络300可以将反馈信号组fA中的多个反馈信号分别输入到第二自旋阵列200中的每一个光参量振荡腔中，每个光参量振荡腔接收一个反馈信号。也就是说，输入到第二自旋阵列200中的一个光参量振荡腔的信号包括光脉冲组B中的一个光脉冲、以及反馈信号组fA中的一个反馈信号。
需理解的是，由于光参量振荡腔内一个光脉冲、以及反馈信号组fA中的一个反馈信号需要相互作用，因此，需要光参量振荡腔接收光脉冲时间和接收反馈信号的时间有重叠。为了 保证该光脉冲和该反馈信号能够给在光参量振荡腔中相互作用，可以通过调整光反馈网络300中光信号的传输的路径长度、泵浦源发出的光脉冲的传输路径或泵浦源产生光脉冲的周期，从而使得该光脉冲和该反馈信号能在设定的时间段内同时进入光参量振荡腔。
如图6所示，为光参量振荡腔的结构示意图，其中包括两个布拉格反射区和一个参量振荡区域，布拉格反射区位于参量振荡区域两端。两个布拉格反射区形成谐振腔，反馈信号在进入到光参量振荡腔后，在两个布拉格反射区之间来回传输，进行振荡；而光脉冲耦合输入至参量振荡区域，与反馈信号产生非线性作用，之后光脉冲再从参量振荡区域耦合滤出，滤出后，光参量振荡腔中遗留的光信号即为自旋信号。在本申请实施例中，由于光参考振荡腔输出的光信号的相位可以对应一种自旋状态，故而将光参量振荡器输出的光信号称为自旋信号。
第二自旋阵列200中的每个光参量振荡腔均可以产生一个自旋信号，第二自旋阵列200可以输出一组包括N个自旋信号(分别为Q 1、Q 2、Q 3……Q N)的自旋信号组Q。
第一自旋阵列100中包括的光参量振荡腔以及每个光参量振荡腔内信号的作用过程，与第二自旋阵列200中包括的光参量振荡腔以及每个光参量振荡腔内信号的作用相同，此处不再赘述。
需要说明的是，第二自旋阵列200中的光参量振荡腔与第一自旋阵列100中的光参量振荡腔之间存在一一对应的关系，当第二自旋阵列200和第一自旋阵列100接收相同的一组光脉冲时，第二自旋阵列200和第一自旋阵列100中两个对应的光参量振荡腔所接收的光脉冲是相同的。
应需理解的是，此处仍以第二自旋阵列200接收的一组信号包括N个光脉冲为例进行说明，对于第二自旋阵列200接收的一组信号包括N个电信号的情况与第二自旋阵列200接收的一组信号包括N个光脉冲的情况类似，区别在于第二自旋阵列200接收的一组信号包括N个电信号时，第二自旋阵列200需要先将该N个电信号转换为N个光脉冲。对于该N个光脉冲的处理过程与第二自旋阵列200接收的一组信号包括N个光脉冲时，处理N个光脉冲的方式相同。具体到第二自旋阵列200内部，每个光参量振荡器接收该组电信号中的一个电信号，并将该电信号转换为一个光脉冲进行处理，之后每个光参量振荡器根据转换后的光脉冲以及接收的一个反馈信号产生一个自旋信号，每个光参量振荡器根据转换后的光脉冲以及接收的一个反馈信号产生一个自旋信号的方式与以第二自旋阵列200接收的一组信号包括N个光脉冲时，每个光参量振荡器根据接收的一个光脉冲以及一个反馈信号产生一个自旋信号的方式相同。
(二)、光反馈网络300。
在本申请实施例中，光反馈网络300可以为全光反馈网络，也可以不是全光反馈网络。全光反馈是指产生反馈信号的过程都是通过光信号来实现。非全光反馈是指可以通过电路形式或光电结合反馈的方式来产生反馈信号。例如，在采用非全光反馈的情况下，可以先将多个光脉冲转换为电信号，将电信号经过现场可编程门阵列(fieldprogrammable gate array，FPGA)的作用加载反馈信息，之后，根据加载了反馈信息的电信号对新的光信号进行调制(如改变光信号的相位或强度)，调制后的光信号即为反馈信号。在本发明实施例中并不对光反馈网络300的实现方式进行限制。可以理解的是，当光反馈网络300为全光反馈网络时，光反馈网络300的信号处理过程中不需要借助光电转换，就可以产生一组反馈信号(实质上反馈信号也为光信号)，并将该组反馈信号发送给第一自旋阵列100或第二自旋阵列200。由于全光反 馈网络省略了光电转换，可以有效的缩短计算时间，提高光反馈网络300对信号的处理效率。
在本申请实施例中，光反馈网络300可以基于预设的第一数据对任一自旋阵列输入的一组自旋信号进行处理，其中预设的第一数据与待解决的NP-hard问题有关，是将NP-hard进行数学抽象后获得的。对于不同的光反馈网络300的结构，第一数据的表现形式也不同。在本申请实施例中提供了两种光反馈网络300，下面分别进行介绍。
第一种、如图7所示，光反馈网络300中包括两个信号处理模块(分别为第一信号处理模块和第二信号处理模块)以及问题加载模块。其中第一信号处理模块接收来自第一自旋阵列100的每组自旋信号，第二信号处理模块接收来自第二自旋阵列200的每组自旋信号。
信号处理模块可以将一维的一组自旋信号转换为二维光信号矩阵，还可以将二维反馈信号矩阵转换为一组反馈信号，也就是说，信号处理模块可以增加信号维度，便于后续问题加载模块进行处理，还可以缩小信号维度，便于将一组反馈信号输入到自旋阵列中。
需要说明的是，在本申请实施例中涉及到信号矩阵(如光信号矩阵以及反馈信号矩阵均)为N*N的光信号构成的矩阵。
问题加载模块中设置有第一数据，该第一数据作用在二维光信号矩阵中，可以产生反馈信号矩阵。
下面以第一信号处理模块接收自旋信号组E为例，对信号处理模块将一维的一组自旋信号转换为二维光信号矩阵的方式进行说明。
第一信号处理模块在接收到自旋信号组E后，可以对自旋信号组E中各个自旋信号进行整形准直，使得自旋信号组中的各个自旋信号平行传输，不会发生串扰，信号的整形准直可以由微镜阵列实现，也可以由其他光学器件实现，本申请实施例并不限定。
第一信息处理模块在自旋信号的传输平面上，对各个平行的自旋信号进行分光，每个自旋信号分为N个相同的自旋信号，N个自旋信号经过分光形成N列自旋信号，每列自旋信号包括N个自旋信号，每列的N个自旋信号相同，形成N*N的光信号矩阵。自旋信号的分光可以由柱面镜实现，也可以由其他光学器件实现，本申请实施例并不限定。
下面结合图示进行说，如图8A为在一个XY平面上，第一信号处理模块对自旋信号组E的处理流程图。自旋信号组E中的各个自旋信号(E 1、E 2、E 3……E N)经过微镜阵列进行准直，形成平行的N个自旋信号；在经过柱面镜后形成N列自旋信号。每列自旋信号构成的平面垂直于XY平面，平行于YZ平面。
如图8B所示，为YX平面上第一信号处理模块对一个自旋信号的处理流程图。
自旋信号组E中的一个自旋信号(以E 1为例)经过微镜阵列进行准直；在经过柱面镜后形成N个自旋信号E 1。
如图9为第一信号处理模块产生的光信号矩阵，该光信号矩阵大小为N*N，每列的光信号与自旋信号组E中的一个自旋信号相同。
为了能够更好的实现对自旋信号的准直以及分光，第一自旋阵列100可以位于第一信号处理模块中柱面镜以及微镜阵列的焦平面上。
下面以问题加载模块接收到如图9所示的光信号矩阵产生反馈信号矩阵为例，对问题加载模块进行的信号处理过程进行说明。
在该种结构中问题加载模块中设定的第一数据可以抽象为一个二维的作用矩阵，该二维作用矩阵可以由空间光调制器实现，通过空间光调制器调制光信号的光相位或强度。也就是说，二维作用矩阵中的一个元素指示空间光调制器对信号矩阵中一个信号的光相位或强度调 制程度。
该二维作用矩阵也可以由数字微镜阵列实现，通过数字微镜阵列调制光信号的光强度。也就是说，二维作用矩阵中的一个元素指示数字微镜阵列对信号矩阵中一个信号的光相位或强度调制程度。
该二维作用矩阵为对称矩阵，且对角线上的元素不为零。
如图10所示，问题加载模块对信号的处理过程可以抽象为矩阵运算，利用第一信号处理模块产生的光信号矩阵与该二维矩阵生成反馈信号矩阵。
光信号矩阵的一个光信号与二维矩阵对应位置的元素作用后生成反馈信号矩阵中的对应位置的信号。例如，光信号矩阵中第一行光信号E 1与二维矩阵中第一行第一列的元素a 1.1z作用，产生反馈信号矩阵中的对应位置的信号a 1.1E 1。
下面以第二信号处理模块输出反馈信号组fA为例，对信号处理模块将二维反馈信号矩阵转换为一组反馈信号的方式进行说明。
如图11为第二信号处理模块从光反馈网络300接收的反馈信号矩阵，该反馈信号矩阵大小为N*N，矩阵中的一个光信号可以看做是自旋信号组E中的一个自旋信号与二维矩阵中的一个元素的乘积。
第二信号处理模块在接收到反馈信号矩阵后，可以将该反馈信号矩阵中每列信号合束为一个反馈信号，对于N*N列的反馈信号矩阵，可经过合束后，形成N个反馈信号，反馈信号矩阵中一列的信号合并后可以生成一个反馈信号，也就是说，反馈信号fA 1＝∑ 1＜i＜Na i.1E 1，反馈信号fA 2＝∑ 1＜i＜Na i.2E 2，反馈信号fA 3＝∑ 1＜i＜Na i.3E 3，反馈信号fA N＝∑ 1＜i＜Na i.NE N。信号合束和信号的分光过程互逆，光信号合束的可以由柱面镜实现，也可以由其他光学器件实现，本申请实施例并不限定。
之后，第二信号处理模块可以对N个反馈信号中各个反馈信号进行整形准直，使得反馈信号组fA中的各个反馈信号平行传输，不会发生串扰，信号的整形准直可以由微镜阵列实现，也可以由其他光学器件实现，本申请实施例并不限定。
下面结合图示进行说明，如图12A为在一个XY平面上，第二信号处理模块对反馈信号矩阵的处理流程图。反馈信号矩阵中N列反馈信号在经过柱面镜后形成N个反馈信号；N个反馈信号经过微镜阵列进行准直，形成平行的N个反馈信号。
如图12B所示，为YX平面上第二信号处理模块对反馈信号矩阵中的一列信号的处理流程图。该列反馈信号在经过柱面镜后进行合束，形成一个反馈信号，之后再经过微镜阵列进行准直。
为了能够更好的实现对反馈信号的准直以及合术，第二自旋阵列200可以位于第二信号处理模块中柱面镜以及微镜阵列的焦平面上。
第二信号处理模块形成N个反馈信号，将N个反馈信号分别传输至第二自旋阵列200中的每个光参量振荡腔中，每个光参量振荡腔接收一个反馈信号。
第二自旋阵列200中的每个光参量振荡腔接收的反馈信号是第一自旋阵列100中的对应的光参量振荡腔输出的自旋信号经过光反馈网络300处理后的产生的。例如第一自旋阵列100中光参量振荡腔输出的自旋信号为E 1，第二自旋阵列200中对应的光参量振荡腔从光反馈网络300接收的反馈信号为fA 1＝∑ 1＜i＜Na i.1E 1；第一自旋阵列100中光参量振荡腔输出的自旋信号为E 2，第二自旋阵列200中对应的光参量振荡腔从光反馈网络300接收的反馈信号为fA 1＝∑ 1＜i＜Na i.1E 2。
图13为本发明实施例提供的另一种光计算设备的结构示意图。如图13所示，为第一自旋阵列100和第二自旋阵列200均包括四个光参量振荡腔为例，一种光反馈网络300的结构示意图，光反馈网络300中包括多个级联的马赫-曾德尔干涉单元(Mach-Zehnder Interferometer Unit，MZIU)构成，每个MZIU可以接收两路光信号，每个MZIU中包括多个间隔设置的马赫-曾德尔干涉仪(Mach-Zehnder Interferometer，MZI)和光开关。
马赫-曾德尔干涉单元尺寸较小，使得该光计算设备的结构更加紧凑，能够在片上实现，保证了系统的稳定性。
本申请实施例中，MZI用于实现输入的两路光信号之间的相互干涉作用，对应的输出信号由MZI通过MZI的相位参数控制。每个MZIU中光开关可以调整来自不同方向的光信号(即传输方向为从第一自旋阵列100到第二自旋阵列200的光信号和从第二自旋阵列200到第一自旋阵列100的光信号)的传输路径。
MZIU中包括的MZI的相位参数可以不同，通过控制光开关与不同相位参数的MZI的在MZIU的位置，使得来自不同方向的光信号经过的相同数量的MZI，且经过的不同相位参数的MZI的顺序相同。示例性的，MZIU中延光信号的传输方向，采用左右对称的方式设置MZI，处于左右对称位置的两个的MZI的相位参数相同。
下面介绍一种本申请实施例提供的MZIU的结构，如图14所述示例，该MZIU中包括三个MZI(分别为MZI-1，MZI-2，MZI-3)和四个光开关(optical switch)，在图14中以SW标识光开关，MZIU中左右两侧的MZI的相位参数相同。MZI-1和MZI-3的相位参数为φ i，MZI-2的相位参数为θ i。
通过控制光开关可以使得从第一自旋阵列100接收的光信号和从第二自旋阵列200接收的光信号传输路径不同，但经过的MZI的数量相同、经过的不同相位参数的MZI的顺序相同。
示例性的，如图15A所示，为传输方向为从第一自旋阵列100到第二自旋阵列200的光信号在MZIU中的传输路径。依次经过相位参数为θ i的MZI-2和相位参数为φ i的MZI-3。
如图15B所示，为传输方向为从第一自旋阵列100到第二自旋阵列200的光信号在MZIU中的传输路径。依次经过相位参数为θ i的MZI-2和相位参数为φ i的MZI-1。
本申请实施例并不限定光反馈网络300中包括的多个MZIU的级联方式。只需保证多个MZIU能够形成UDU +的结构即可，其中U为酉矩阵(Unitary Matrix),U +为U的转置矩阵，D为对角矩阵。
在这种光反馈网络300中第一数据为级联的多个MZIU中MZI的相位参数。
如图16所示，本申请实施例提供的光计算设备中还包括探测器阵列400，探测器阵列400可以与任一自旋阵列连接，对自旋阵列输出一组自旋信号进行探测，确定该组自旋信号中包括的各个自旋信号的相位，当自旋阵列接收的光脉冲的光功率达到设定值时，各个自旋信号的相位可能会随机坍缩到0或者π，当各个自旋信号的相位坍缩到0或者π时，则认为当前输出的该组自旋信号与第一数据相对应，可以根据该组自旋信号确定最终的计算结果。
图16中以探测器阵列400与第一自旋阵列100连接为例，探测器阵列400中可以包括N个探测器，每个探测器连接第一自旋阵列100中的一个光参量振荡器，每个探测器连接的光参量振荡器不同，每个探测器可以对连接的光参量振荡器产生的自旋信号的相位进行检测，当探测器阵列400探测到第一自旋阵列100输出的各个自旋信号的相位坍缩到0或者π，根据该组自旋信号确定最终的计算结果。
为了使方案描述更加清楚，下面将结合前面的实施例，以如图16所示的光计算设备和图 17所示的计算方法为例，对本发明实施例中提供的光计算设备的工作流程进行概括介绍。图16为本发明实施例提供的又一种光计算设备。需要说明的是，下面的工作流程的介绍也同样适用于前面描述的所有光计算设备。如图16和图17所示，在工作过程中，第一自旋阵列100可以接收包括有N个信号的第一组信号(如本申请实施例中的光脉冲组A)，并基于第一组信号产生包括N个自旋信号的第一组自旋信号(如本申请实施例中的自旋信号组E)，其中，N为大于2的正整数。光反馈网络300可以接收第一组自旋信号，并根据所述第一组自旋信号以及设置的第一数据生成包括N个反馈信号的第一组反馈信号(如本申请实施例中的反馈信号组fA)。可以理解的是，这里所述的第一自旋阵列100到光反馈网络300的信号传输过程类似于前述实施例中的过程一。在光反馈网络300生成第一组反馈信号后，第二自旋阵列200可以接收第一组反馈信号(如本申请实施例中的反馈信号组fA)以及第一组信号(如本申请实施例中的光脉冲组B)。并且，第二自旋阵列200可以根据第一组反馈信号以及第一组信号产生包括N个自旋信号的第二组自旋信号(如本申请实施例中的自旋信号组F)。光反馈网络300接收第二组自旋信号，并根据第二组自旋信号以及第一数据生成包括N个反馈信号的第二组反馈信号(如本申请实施例中的反馈信号组fB)。其中，第二自旋阵列200到光反馈网络300的信号传输过程类似于前述实施例中的过程二。
可以理解的是，实际应用中，还可以多次执行前述的过程一和过程二。例如，第一自旋阵列100还可以接收第二组反馈信号以及包括N个信号的第二组信号(如本申请实施例中的光脉冲组A’)，并根据第二组反馈信号以及第二组信号产生包括N个自旋信号的第三组自旋信号(如本申请实施例中的自旋信号组P)。之后，光反馈网络300接收第三组自旋信号，并根据接收的第三组自旋信号产生新的一组反馈信号，并向第二自旋阵列200发送新产生的一组反馈信号。第二自旋阵列200再根据接收的反馈信号和第二组信号生一组自旋信号，并将该组自旋信号传输至光反馈网络300。光反馈网络300接收到该组自旋信号，又根据从第二自旋阵列接收的自旋信号产生一组反馈信号，将该组反馈信号反馈给第一自旋阵列100。如此循环往复，直至探测器阵列400探测到第一自旋阵列100输出的各个自旋信号的相位坍缩到0或者π。
本发明实施例提供的光计算设备，由于第一自旋阵列和第二自旋阵列对一组光脉冲中的n个光脉冲的处理过程是并行的，因此，在增加光脉冲的数量的同时，不会增加光计算设备的运算时间，可以有效提高光运算设备的计算效率。并且，本发明实施例提高的光计算设备结构简单，通过双自旋阵列相互注入耦合实现相互作用，极大的简化了光伊辛机的架构，可能实现在片上集成全光伊辛机。并且，由于光反馈网络采用的是全光反馈网络，无需进行光电转换，实现自旋组态的并行搜索，极大的缩短了计算时间，提升了系统的稳定性，提高光计算设备的计算效率。并且，在计算过程中，不需要采用光电探测的注入信号，提升了信号传输时间。
可理解的是，实际应用中，本发明实施例中的第一自旋阵列和第二自旋阵列也可以不局限于能产生光脉冲的自旋阵列，还可以采用通过其他方式映射的双自旋阵列，比如激光器构成的激光器阵列、极化子构成的极化子阵列等。
此外，由于本发明实施例提供的光计算设备结构简单，能够在片上实现，并且，整个计算过程通过光信号的方式实现，信号传输速度快，计算速度也得到很大提升，因此，本发明实施例提高的光计算设备可以应用于神经网络系统中，例如，可以用于实现神经网络系统中的反馈控制。
需要说明的是，本申请所提供的实施例仅仅是示意性的。所属领域的技术人员可以清楚 的了解到，为了描述的方便和简洁，在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其他实施例的相关描述。在本发明实施例、权利要求以及附图中揭示的特征可以独立存在也可以组合存在。在本发明实施例中以硬件形式描述的特征可以通过软件来执行，反之亦然。在此不做限定。

Claims (14)

  1. 一种光计算设备，其特征在于，包括：
    第一自旋阵列，用于接收第一组信号，并根据所述第一组信号产生第一组自旋信号，其中，所述第一组自旋信号包括N个自旋信号，N为不小于2的整数；
    光反馈网络，连接所述第一自旋阵列并用于：
    接收所述第一组自旋信号；
    根据所述第一组自旋信号以及设置的第一数据生成第一组反馈信号，所述第一组反馈信号包括N个反馈信号；
    第二自旋阵列，连接所述光反馈网络，用于：
    接收所述第一组反馈信号以及所述第一组信号；
    根据所述第一组反馈信号以及所述第一组信号产生第二组自旋信号，所述第二组自旋信号包括N个自旋信号。
  2. 如权利要求1所述的设备，其特征在于，所述光反馈网络还用于：
    接收所述第二组自旋信号；
    根据所述第二组自旋信号以及所述第一数据生成第二组反馈信号，所述第二组反馈信号包括N个反馈信号；
    所述第一自旋阵列，还用于：
    接收所述第二组反馈信号以及第二组信号，其中，所述第二组光脉冲包括N个光脉冲，所述第二组光脉冲的幅值大于所述第一组光脉冲幅值；
    根据所述第二组反馈信号以及所述第二组信号产生第三组自旋信号，所述第三组自旋信号包括N个自旋信号。
  3. 如权利要求1或2所述的设备，其特征在于，所述第一自旋阵列包括N个光参量振荡器，其中，所述第一自旋阵列中的每个光参量振荡器用于接收所述第一组信号的一个信号。
  4. 如权利要求1或2所述的设备，其特征在于，所述第二自旋阵列包括N个光参量振荡器，所述第二自旋阵列中的每个光参量振荡器用于接收所述第一组信号的一个信号以及所述第一组反馈信号中的一个反馈信号，其中，所述第二自旋阵列中的N个光参量振荡器与所述第一自旋阵列中的N个光参量振荡器一一对应，所述第二自旋阵列和所述第一自旋阵列中具有对应关系的光参量振荡器接收的信号相同。
  5. 如权利要求1或2所述的设备，其特征在于，所述光反馈网络包括：
    第一信号处理模块，用于接收所述第一组自旋信号，将所述第一组自旋信号转换为第一光信号矩阵，所述第一光信号矩阵中包括N*N个光脉冲；
    问题加载模块，连接所述第一信号处理模块，用于根据所述第一光信号矩阵和所述第一数据生成第一反馈信号矩阵，所述第一反馈信号矩阵中包括N*N个光信号；
    第二信号处理模块，连接所述问题加载模块，用于将所述第一反馈信号矩阵转换为所述第一组反馈信号。
  6. 如权利要求5所述的设备，其特征在于：
    所述第二信号处理模块还用于接收所述第二组自旋信号，将所述第二组自旋信号转换为第二光信号矩阵，所述第二光信号矩阵中包括N*N个光脉冲；
    所述问题加载模块，还用于根据所述第二光信号矩阵和所述第一数据生成第二反馈信号 矩阵，所述第二反馈信号矩阵中包括N*N个光信号；
    所述第一信号处理模块，还用于将所述第二反馈信号矩阵转换为所述第二组反馈信号。
  7. 如权利要求1或2所述的设备，其特征在于，所述光反馈网络包括多个级联的马赫-曾德尔MZ干涉单元，每个MZ干涉单元包括间隔设置的马赫-曾德尔干涉仪和光开关。
  8. 如权利要求1-7任意一项所述的设备，其特征在于，所述光计算设备还包括：
    探测器阵列，连接所述第一自旋阵列并用于：
    探测所述第三组自旋信号；
    当所述第三组自旋信号的相位为预设值时，根据所述第三组自旋信号获得所述第一数据的计算结果。
  9. 如权利要求2所述的设备，其特征在于，所述第一组信号包括第一组光脉冲，所述第二组信号包括第二组光脉冲，其中，所述第二组光脉冲的幅值大于所述第一组光脉冲的幅值。
  10. 一种计算方法，其特征在于，所述方法由光计算设备执行，所述光计算设备包括第一自旋阵列、第二自旋阵列以及分别与所述第一自旋阵列和所述第二自旋阵列连接的光反馈网络，所述方法包括：
    所述第一自旋阵列接收第一组信号，并根据所述第一组信号产生第一组自旋信号，其中，所述第一组自旋信号包括N个自旋信号，N为不小于2的整数；
    所述光反馈网络根据所述第一组自旋信号以及设置的第一数据生成第一组反馈信号，所述第一组反馈信号包括N个反馈信号；
    所述第二自旋阵列接收所述第一组信号以及所述第一组反馈信号，并根据所述第一组反馈信号以及所述第一组信号产生第二组自旋信号，其中，所述第二组自旋信号包括N个自旋信号。
  11. 根据权利要求10所述的方法，其特征在于，还包括：
    所述光反馈网络接收所述第二自旋阵列发送的第二组自旋信号；
    所述光反馈网络根据所述第二组自旋信号以及所述第一数据生成第二组反馈信号，所述第二组反馈信号包括N个反馈信号；
    所述第一自旋阵列接收第二组信号；
    所述第一自旋阵列根据所述第二组反馈信号以及所述第二组信号产生第三组自旋信号，所述第三组自旋信号包括N个自旋信号。
  12. 如权利要求11所述的方法，其特征在于，所述光计算设备还包括探测器阵列，该方法还包括：
    所述探测器阵列探测所述第三组自旋信号；
    当所述第三组自旋信号的相位为预设值时，根据所述第三组自旋信号获得所述第一数据的计算结果。
  13. 如权利要求10-12任意一项所述的方法，其特征在于，所述第一自旋阵列接收第一组信号包括：
    所述第一自旋阵列中的每个光参量振荡器用于接收所述第一组信号的一个信号，其中，所述第一自旋阵列包括N个光参量振荡器。
  14. 如权利要求13所述的方法，其特征在于，所述第二自旋阵列包括N个光参量振荡器，所述第二自旋阵列接收所述第一组信号以及所述第一组反馈信号包括：
    所述第二自旋阵列中的每个光参量振荡器用于接收所述第一组信号的一个信号以及所述第一组反馈信号中的一个反馈信号，其中，所述第二自旋阵列中的N个光参量振荡器与所述 第一自旋阵列中的N个光参量振荡器一一对应，所述第二自旋阵列和所述第一自旋阵列中具有对应关系的光参量振荡器接收的信号相同。
PCT/CN2020/114614 2019-09-11 2020-09-10 一种光计算设备以及计算方法 WO2021047614A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
EP20863261.2A EP4027250A4 (en) 2019-09-11 2020-09-10 OPTICAL COMPUTING DEVICE AND COMPUTING METHOD
JP2022515890A JP7485758B2 (ja) 2019-09-11 2020-09-10 光計算装置及び計算方法
US17/691,945 US20220197328A1 (en) 2019-09-11 2022-03-10 Optical computing device and computing method

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910860799.4 2019-09-11
CN201910860799.4A CN112486898B (zh) 2019-09-11 2019-09-11 一种光计算设备以及计算方法

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/691,945 Continuation US20220197328A1 (en) 2019-09-11 2022-03-10 Optical computing device and computing method

Publications (1)

Publication Number Publication Date
WO2021047614A1 true WO2021047614A1 (zh) 2021-03-18

Family

ID=74866587

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/114614 WO2021047614A1 (zh) 2019-09-11 2020-09-10 一种光计算设备以及计算方法

Country Status (5)

Country Link
US (1) US20220197328A1 (zh)
EP (1) EP4027250A4 (zh)
JP (1) JP7485758B2 (zh)
CN (1) CN112486898B (zh)
WO (1) WO2021047614A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019014345A1 (en) * 2017-07-11 2019-01-17 Massachusetts Institute Of Technology OPTICAL ISING MACHINES AND OPTICAL CONVOLUTIVE NEURAL NETWORKS
WO2019078354A1 (ja) * 2017-10-19 2019-04-25 日本電信電話株式会社 イジングモデルの計算装置
WO2019078355A1 (ja) * 2017-10-19 2019-04-25 日本電信電話株式会社 ポッツモデルの計算装置
CN109814319A (zh) * 2019-03-06 2019-05-28 电子科技大学 一种基于横自旋结构光场的xnor逻辑门

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2009015690A1 (en) * 2007-07-31 2009-02-05 Telefonaktiebolaget Lm Ericsson (Publ) Optical circuit for comparing two n-bit binary words
JP6143325B2 (ja) * 2013-01-11 2017-06-07 大学共同利用機関法人情報・システム研究機構 イジングモデルの量子計算装置及びイジングモデルの量子計算方法
JP2014174301A (ja) * 2013-03-08 2014-09-22 Nippon Telegr & Teleph Corp <Ntt> 光パルス列発生器及び発生方法
JP6029072B2 (ja) * 2014-02-28 2016-11-24 日本電信電話株式会社 光パラメトリック発振器とそれを用いたランダム信号発生装置及びイジングモデル計算装置
WO2015156126A1 (ja) * 2014-04-11 2015-10-15 大学共同利用機関法人情報・システム研究機構 イジングモデルの量子計算装置、イジングモデルの量子並列計算装置及びイジングモデルの量子計算方法
WO2017044077A1 (en) 2015-09-08 2017-03-16 Hewlett Packard Enterprise Development Lp Apparatus for solving ising problems
US9870273B2 (en) * 2016-06-13 2018-01-16 1Qb Information Technologies Inc. Methods and systems for quantum ready and quantum enabled computations
JP6980185B2 (ja) * 2017-03-06 2021-12-15 日本電信電話株式会社 イジングモデルの計算装置
JP6533544B2 (ja) * 2017-03-06 2019-06-19 日本電信電話株式会社 イジングモデルの計算装置
JP6818320B2 (ja) 2017-03-06 2021-01-20 日本電信電話株式会社 イジングモデルの計算装置
JP6684259B2 (ja) * 2017-10-19 2020-04-22 日本電信電話株式会社 ポッツモデルの計算装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2019014345A1 (en) * 2017-07-11 2019-01-17 Massachusetts Institute Of Technology OPTICAL ISING MACHINES AND OPTICAL CONVOLUTIVE NEURAL NETWORKS
WO2019078354A1 (ja) * 2017-10-19 2019-04-25 日本電信電話株式会社 イジングモデルの計算装置
WO2019078355A1 (ja) * 2017-10-19 2019-04-25 日本電信電話株式会社 ポッツモデルの計算装置
CN109814319A (zh) * 2019-03-06 2019-05-28 电子科技大学 一种基于横自旋结构光场的xnor逻辑门

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4027250A4 *

Also Published As

Publication number Publication date
JP7485758B2 (ja) 2024-05-16
CN112486898A (zh) 2021-03-12
US20220197328A1 (en) 2022-06-23
EP4027250A1 (en) 2022-07-13
CN112486898B (zh) 2023-02-10
EP4027250A4 (en) 2022-10-26
JP2022548567A (ja) 2022-11-21

Similar Documents

Publication Publication Date Title
Babazadeh et al. High-dimensional single-photon quantum gates: concepts and experiments
WO2021047620A1 (zh) 一种光计算设备以及光信号处理方法
Hurtado et al. Controllable spiking patterns in long-wavelength vertical cavity surface emitting lasers for neuromorphic photonics systems
Armstrong et al. Programmable multimode quantum networks
JP6300049B2 (ja) 光パラメトリック発振器のネットワークを使用する計算
Schreiber et al. Photons walking the line: a quantum walk with adjustable coin operations
JP2022533053A (ja) ガウスボソンサンプリングの装置及び方法
JP6029072B2 (ja) 光パラメトリック発振器とそれを用いたランダム信号発生装置及びイジングモデル計算装置
JP2022526208A (ja) 偏光変調軌道角運動量のための軌道角運動量生成装置および方法
Bian et al. Experimental implementation of a quantum walk on a circle with single photons
JP2023109931A (ja) スケーラブルでプログラム可能なコヒーレント波形発生器
WO2021104535A1 (zh) 一种光计算设备以及光信号处理方法
Luo et al. Topological photonic orbital-angular-momentum switch
Cen et al. Microwave photonic ising machine
Alata et al. Phase noise robustness of a coherent spatially parallel optical reservoir
WO2021047614A1 (zh) 一种光计算设备以及计算方法
Consoli et al. Networks of random lasers: current perspective and future challenges
WO2022136146A1 (en) Optical computing and reconfiguring with spatiotemporal nonlinearities in waveguides
Zhu et al. Photonic discrete-time quantum walks
Kumar et al. Cylindrical vector beams by tailoring single polarization component
Rigas et al. Generation of continuous-variable cluster states of cylindrically polarized modes
Antonio et al. Networks of random lasers: current perspective and future challenges
JPH01179910A (ja) パルス多重光学系
Akel Abrahao Frontiers of quantum optics: photonics tolls, computational complexity, quantum metrology, and quantum correlations
Sinha Generalized Laser Architecture for High-Powered Applications

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20863261

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2022515890

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2020863261

Country of ref document: EP

Effective date: 20220404