Apparatus and method for Gaussian boson sampling

A photonic platform for Gaussian boson sampling addresses the lack of fault-tolerant hardware by using a light source, beam splitters, and detectors to perform unitary transformations, enhancing computational speed and enabling applications in graph problems and vibronic spectrum modeling.

JP7590346B2Active Publication Date: 2024-11-26XANADU QUANTUM TECHNOLOGIES HOLDINGS ULC
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
JP2021566944
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2019-05-22
Filing Date
2020-05-20
Publication Date
2024-11-26
Estimated Expiration
2040-05-20

AI Technical Summary

Technical Problem

Current fault-tolerant hardware is lacking for executing general-purpose quantum computing algorithms on large-scale problems, particularly in Gaussian boson sampling.

Method used

A photonic platform apparatus for Gaussian boson sampling, comprising a light source, a network of reconfigurable beam splitters, and photon counting detectors, controlled by a controller to manage squeeze factor, unitary transform angle, and phase, to perform unitary transformations and measure output photons.

Benefits of technology

Enables efficient and flexible implementation of Gaussian boson sampling, facilitating faster computational speeds and practical applications in graph problems, vibronic spectrum modeling, and other complex computations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007590346000009
    Figure 0007590346000009
  • Figure 0007590346000010
    Figure 0007590346000010
  • Figure 0007590346000011
    Figure 0007590346000011
Patent Text Reader

Abstract

The apparatus includes a light source that provides a plurality of input optical modes in squeezed states. The apparatus also includes a network of interconnected reconfigurable beam splitters (RBSs) configured to perform unitary transformation of the plurality of input optical modes to generate a plurality of output optical modes. An array of photon-counting detectors is in optical communication with the network of interconnected RBSs and configured to measure the number of photons in each of the plurality of output optical modes after unitary transformation. The apparatus also includes a controller operably coupled to the light source and the network of interconnected RBSs. The controller is configured to control at least one of a squeeze factor of the squeezed-state light, an angle of the unitary transformation, or a phase of the unitary transformation.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[1001] This application claims the benefit of and priority to U.S. Provisional Patent Application No. 62 / 851,312, entitled "Apparatus and Methods for Gaussian Boson Sampling," filed May 22, 2019, the entire contents of which are incorporated herein by reference. Field

[0002]

[1002] One or more embodiments relate to Gaussian Boson sampling. [Background technology]

[0003] background

[1003] General-purpose quantum computing algorithms often promise much faster computational speeds than their classical counterparts, but to date there is no known hardware capable of running these algorithms on large-scale problems, let alone in a fault-tolerant manner. Summary of the Invention [Means for solving the problem]

[0004] overview

[1004] Some embodiments described herein relate generally to Gaussian boson sampling, and in particular to performing Gaussian boson sampling on a photonic platform. In some embodiments, an apparatus includes a light source configured to provide a plurality of input optical modes of squeezed state light. The apparatus also includes a network of interconnected reconfigurable beam splitters (RBS) in optical communication with the light source. The network of interconnected RBSs is configured to perform unitary transformation of the plurality of input optical modes to generate a plurality of output optical modes. An array of photon counting detectors is in optical communication with the network of interconnected RBSs and configured to measure the number of photons in each of the plurality of output optical modes after the unitary transformation. The apparatus also includes a controller operably coupled to the light source and the network of interconnected RBSs. The controller is configured to control at least one of a squeeze factor of the squeezed state light, an angle of the unitary transformation, or a phase of the unitary transformation.

[0005] BRIEF DESCRIPTION OF THE DRAWINGS

[1005] The drawings are primarily for illustrative purposes and are not intended to limit the scope of the subject matter described herein. The drawings are not necessarily to scale, and in some cases, various aspects of the subject matter disclosed herein may be shown exaggerated or enlarged in the drawings to facilitate understanding of different features. In the drawings, like reference characters generally refer to like features (e.g., functionally similar and / or structurally similar elements). [Brief description of the drawings]

[0006] [Figure 1]

[1006] A schematic of an apparatus for Gaussian Boson Sampling (GBS) according to one embodiment is shown. [Diagram 2]

[1007] A schematic of a GBS system using spatial coding according to one embodiment is shown. [Diagram 3]

[1008] A schematic of a quantum processing unit (QPU) implemented in a photonic integrated circuit performing GBS, according to one embodiment, is shown. [Figure 4A]

[1009] A schematic of a power splitting module that can be used in the QPU shown in Figure 3, according to one embodiment. [Figure 4B]

[1009] A schematic of a power splitting module that can be used in the QPU shown in Figure 3, according to one embodiment. [Diagram 5]

[1010] A schematic of a displacement module that can be used to generate displacement squeezed light for GBS, according to one embodiment. [Figure 6]

[1011] A block diagram illustrating a method of operating a GBS device according to one embodiment is shown. [Figure 7]

[1012] A schematic of a GBS system that provides user access to quantum hardware configuration, according to one embodiment. [Figure 8]

[1013] A block diagram illustrating a method for solving graph problems using a GBS device according to one embodiment is shown. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0007] Detailed Description

[1014] In light of the problem of realizing universal quantum computing, alternative non-universal quantum computing models are currently being actively researched. One such model is Gaussian Boson Sampling (GBS). A GBS system prepares input squeezed vacuum or squeezed coherent states of a set of quantum harmonic oscillators (e.g., light field modes) in M ​​modes, inputs them into an N-mode linear optical interferometer, and detects the number of photons in each of the N output modes of the interferometer. In other words, the GBS system samples from the probability distribution of outputs based on the number of photons.

[0008]

[1015] 1 shows a schematic of a GBS apparatus 100 according to one embodiment. The apparatus 100 includes an input state generator 110 that prepares light in a desired input state (optical mode), followed by a linear interferometer 120 that performs a unitary transformation on the input state of light received from the input state generator 110. A detector 130 is utilized to measure the output of the linear interferometer 120 (e.g., the output optical mode, also referred to as the transformed optical mode). The combination of the input state generator 110 and the linear interferometer 120 is also referred to as a quantum processing unit (QPU) 150, and the components within the dashed line are also referred to as quantum hardware (QH) 160.

[0009]

[1016] In addition to the QH 160, the apparatus 100 also includes an input controller 115 that controls the operation of the input state generator 110, an interferometer controller 125 that controls the operation of the linear interferometer 120, and a detector controller 135 that controls the operation of the detector 130. Furthermore, a master controller 140 is utilized to coordinate and / or control the operation of the three controllers 115, 125, and 135. The controllers 115, 125, and 135 may include one or more field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), graphics processing units (GPUs), and / or central processing units (CPUs). The use of FPGAs, ASICs, GPUs, and / or CPUs may increase operational speed, reliability, and flexibility as well as lower monetary costs compared to customized integrated circuits.

[0010]

[1017] In some embodiments, the controllers 115, 125, and 135 may include small computers that execute control logic and run drivers that operate the various subsystem hardware elements. The master controller 140 may include a computer that coordinates the control logic of the sub-controllers 115, 125, and 135. For example, the master controller 140 may be configured to receive instructions from a user and then send the received instructions to one or more specific sub-controllers.

[0011]

[1018] In some embodiments, the input state generator 110 is configured to generate the input states as an optical beam or optical pulses (e.g., squeezed light). In these embodiments, the input state generator 110 is also referred to as the light source 110. For example, the input state generator 110 may convert laser light to create a total number M of squeezed and / or displaced squeezed states of light via nonlinear optical elements, which may be

number

number

number

[0012]

[1019] The linear interferometer 120 is configured to perform a unitary operation on up to N input modes, thus

number

number

number

[0013]

[1020] The detector 130 in the apparatus 100 has photon counting capability, i.e., the detector 130 is configured to measure the number of photons at each output mode of the linear interferometer 120. Without being bound by any particular theory or mode of operation, each output mode of the linear interferometer 120 may be represented as a projected value measurement element.

number

[0014]

[1021] The modes i generated by the input state generator 110 (i.e., state encoding) can correspond to any suitable set of quantum harmonic oscillator modes. Herein, optical field modes are used as an illustrative example. In general, optical field modes can be characterized by frequency / time, polarization, and spatial distribution. Therefore, various encoding schemes can be utilized to prepare the appropriate optical field modes of the apparatus 100 to perform GBS.

[0015]

[1022] In some embodiments, the input state generator 110 can be configured to generate the optical modes using spatial encoding. In these embodiments, the different optical modes are characterized by transverse field distributions that do not overlap in space. For example, the different optical modes can include separate light beams that can propagate in free space or separate optical waveguides.

[0016]

[1023] In some embodiments, the apparatus 100 can include multiple waveguides that propagate optical modes with spatial encoding. For example, the linear interferometer 120 can include M input ports that receive M input optical modes, each of which can be coupled to an input waveguide that propagates a distinct input optical mode. In addition, the output modes of the linear interferometer 120 can also be transmitted to the detector 130 via the waveguides. In this case, the detector 130 can include an array of photon-counting detectors, each of which is coupled to a corresponding output port of the linear interferometer 120 to measure the number of photons in the output optical mode sent from the output port.

[0017]

[1024] In some embodiments, the input state generator 110 can be configured to generate the optical modes using time encoding. In these embodiments, the different optical modes are characterized by longitudinal distributions that do not overlap in space. For example, each optical mode can include a clearly defined optical pulse in one longitudinal path.

[0018]

[1025] In some embodiments, the input state generator 110 can be configured to generate the optical modes using frequency encoding. In these embodiments, the different optical modes are characterized by non-overlapping frequency distributions in frequency, i.e., each optical mode has a distinct frequency or wavelength. In some embodiments, the input state generator 110 can utilize more than one encoding scheme, i.e., use a hybrid scheme to prepare the input optical modes.

[0019]

[1026] 2 shows a schematic of a GBS system 200 using spatial coding, according to one embodiment. System 200 includes a server 210 that networks the system to a user terminal 201. User terminal 201 may include any user hardware that provides access to system 200. Access may be achieved locally or by an Internet connection. System 200 also includes a master unit 240 that controls system 200, receives user commands, and / or provides status information and computation results to the user.

[0020]

[1027] In system 200, QPU 250 is used to perform GBS. QPU 250 can be substantially similar to QPU 150 shown in FIG. 1 and described above (e.g., including an input state generator and a linear interferometer). In some embodiments, QPU 250 can be configured as a photonic chip (see more detail below with reference to FIGS. 3 and 4). QPU driver 252 is operatively coupled to QPU 250 to configure, program, and operate the elements of QPU 250. For example, QPU driver 252 can include appropriate voltage / current drivers for on-chip electrical elements (e.g., phase shifters) included in QPU 250. In addition, lock loop controller 254 is utilized for any active feedback system associated with QPU 250, such as a lock loop of a resonant device. QPU driver 252 and / or lock loop controller 254 can include one or more FPGAs, ASICs, GPUs, and / or CPUs.

[0021]

[1028] QPU 250 is also operably coupled to QPU environmental controller 255, which is configured to control the mechanical, thermal, and / or optical environment of QPU 250. For example, QPU environmental controller 255 can be configured to ensure that QPU 250 is operating within a stable environment (e.g., low levels of temperature fluctuations, etc.). QPU environmental controller 255 can include one or more FPGAs, ASICs, GPUs, and / or CPUs. One or more monitor photodiodes 260 are operably coupled to QPU 250 and QPU environmental controller 255 to assess the stability of the QPU elements. Monitor photodiode 260 can be configured to measure operating parameters (e.g., optical output) of QPU 250 and provide the measurements to QPU environmental controller 255 so that QPU environmental controller 255 can generate appropriate control signals.

[0022]

[1029] System 200 also includes a pump source 220 that provides a pump light beam to an input state generator in QPU 250 to generate appropriate input states. A pump controller 225 is operatively coupled to pump source 220 to control the operation of pump source 220. Pump controller 225 may include one or more FPGAs, ASICs, GPUs, and / or CPUs. In some embodiments, pump source 220 may include a laser.

[0023]

[1030] The output of the QPU 250 is monitored by a detector array 230, which includes a set of photon-counting detectors, such as photon-number-resolving transition edge sensors, superconducting nanowire photon-counting detectors, and / or avalanche photodiodes. The detector array 230 is controlled by a detection controller 235. For example, the detection controller 235 can adjust the bias voltage of the detector array 230 to improve detection efficiency. The detection controller 235 can include one or more FPGAs, ASICs, GPUs, and / or CPUs. In some embodiments, an optional cryogenic controller 234 can be utilized to control a cryogenic system (not shown in FIG. 2 ) that supports the operation of the detector array 230. A data acquisition system (DAQ) 232 is operably coupled to the detector array 230 to translate the output signal of the detector array 230 into a series of integers representing a photon number readout, which can then be reported to a user (e.g., via the user terminal 201).

[0024]

[1031] 3 shows a schematic of a quantum processing unit (QPU) 300 implemented in a photonic integrated circuit performing GBS, according to one embodiment. The QPU 300 includes an input state generator 310 and an interferometer 320 fabricated on a substrate 305 (also referred to as a platform 305). The substrate may include one or more of silicon, silicon oxide (e.g., silicon dioxide), silicon nitride, etc. The input state generator 310 includes: Power division The interferometer 320 includes a pulse distribution module (PDM) 312, a squeezer 314, a filter 316, and an optional displacement module 318 (which generates the displacement squeezer light). The interferometer 320 includes an input port 322, a network of RBS 325 that performs a unitary transformation, and an output port 328.

[0025]

[1032] QPU 300 also includes a QPU driver 340 operably coupled to PDM 312, squeezer 314, filter 316, displacement module 318, and RBS network 325 to control their operation. In addition, each of the above elements has a corresponding monitor (e.g., photodetector and electronics) that monitors the optical status output of that element. More specifically, PDM 312 is coupled to PDM monitor 313, squeezer 314 is coupled to squeezer monitor 315, filter 316 is coupled to filter monitor 317, displacement module 318 is coupled to displacement monitor 319, and RBS network 325 is coupled to interferometer monitor 324. Monitors 313, 315, 317, 319, and 324 are all also operably coupled to QPU driver 340. In FIG. 3, thin lines (e.g., between each monitor and its corresponding element) indicate optical signals, and thick lines with arrows (e.g., between QPU driver 340 and components within input state generator 310 and interferometer 320) indicate electrical signals.

[0026]

[1033] The platform 305 can be selected for low optical loss and high index contrast for high confinement of optical modes. The resulting building blocks (e.g., waveguides) can then be used to build deep optical circuits or circuit elements (e.g., PDM 312, squeezer 314, displacement module 318, or connections between these elements) that allow optical modes to propagate long distances. In some embodiments, the platform 305 includes a silicon-on-insulator wafer that supports the optical waveguide structure and photonic integration components.

[0027]

[1034] In addition to good passive performance, the waveguide material may also possess high second- or third-order nonlinear optical response to induce efficient parametric fluorescence, which can be exploited by the input state generator 310 to generate the squeezed light. In some embodiments, the waveguide material may include silicon nitride, silicon nanowires, lithium niobate, and aluminum nitride, among others.

[0028]

[1035] In some embodiments, the QPU 300 can be fabricated using a fabrication platform that includes multiple waveguiding layers, providing enhanced design flexibility. For example, the nonlinear optical components (e.g., squeezer 314) can be fabricated on a silicon nitride waveguide layer, and the interferometer 320 can be implemented on a silicon layer. In these embodiments, interlayer couplers can be utilized to enable the transfer of light between the different layers.

[0029]

[1036] In some embodiments, it may be advantageous to have tunability and programmability in some active components of QPU 300, such as interferometer 320. Tunability and programmability may be achieved via electrical input from QPU driver 340. In some embodiments, thermo-optical, electro-optical phase shifters, or any suitable type of phase shifter, may be integrated with the optical waveguide structure to achieve tunability and programmability.

[0030]

[1037] In the QPU 300, an input state generator 310 is configured to create squeezer light and / or displacement squeezer state light that inputs into a linear optical interferometer 320. As an input, the state generator 310 drives a nonlinear optical process that transforms coherent laser light to generate the squeeze and perform the coherent displacement. In general, the squeezer and displacer (e.g., 314 and 318) are excited by mutually coherent beams, and a PDM 312 is utilized to take one or more mutually coherent laser input beams and generate multiple phase-locked, mutually coherent output laser beams to drive the squeezer and displacer. The output of the PDM 312 is then directed to the squeezer 314 and the displacement module 318.

[0031]

[1038] In some embodiments, the displacement module 318 can be optional, i.e., the input optical mode to the interferometer 320 does not include displacement. Power An array of laser beams may be provided and configured to excite the squeezer 314. In some embodiments, the input optical mode to the interferometer 320 includes a displacement squeezer light.

[0032]

[1039] In these embodiments, the PDM 312 is configured to generate a low Power Typically, the wavelength of the pump light beam that drives the nonlinear process to create the squeezed light (i.e., pump wavelength) is different from the wavelength of the resulting squeezed light (i.e., squeezed light wavelength). Thus, the wavelength of the displacement beam (i.e., displacement wavelength) is also different from the pump wavelength. In addition, in many mode systems (i.e., large M systems, where M is the number of optical modes), highly optical Power Depending on the demand, one channel can supply the entire pump light required. Power It can be difficult to receive.

[0033]

[1040] In some embodiments, the QPU 300 may include a beam splitter that splits the pump beam into multiple separate pump beams (e.g., M pump beams) that can be monitored to provide information about the relative phase between a set of split input pump beams. The phase information is then used by the QPU driver 340 to feedback the associated input pumps and lock their relative phase. As used herein, a beam splitter (BS) refers to ... from two input modes via, for example, a directional coupler or a multimode interferometer. Power This refers to a component that transfers the input to two output modes.

[0034]

[1041] In some embodiments, the beam splitter used is an adjustable PowerThe optical waveguide may include a reconfigurable beam splitter (RBS) with a splitting ratio. For example, the RBS may include a Mach-Zehnder interferometer formed from two subsequent beam splitters, with a tunable phase shifter (TPS) in one intermediate arm providing controllable transmission between the four ports. As used herein, a TPS is an element placed adjacent to or within the path of a waveguide that provides a controllable optical phase delay. The TPS may be operated thermo-optically or electro-optically (or any other suitable mechanism). In general, the TPS may be tuned by an input voltage applied to electrical contacts on the photonic chip that includes the TPS.

[0035]

[1042] In some embodiments, phase information can be obtained by tapping off a small portion of each input pump beam using an on-chip beam splitter (e.g., a directional coupler) as it enters the PDM 312. Each tapped portion can be interfered with a neighboring tapped portion via another beam splitter, whose resulting optical output is provided to a photodetector. Information from the signals incident on these photodetectors can be used by the QPU driver 340 for feedback and locking the relative pump input phases. In some embodiments, feedback is activated before the pump beams enter the QPU 300. In some embodiments, feedback is activated by a TPS unit integrated into the PDM 312 via an electrical signal from the QPU driver 340.

[0036]

[1043] It is also advantageous that the displacement beam (labeled "Displacement Input" in FIG. 3) is phase stabilized with respect to the pump beam. Thus, the PDM 312 can include a monitor to check this relative phase. Because the displacement wavelength is usually different from the pump wavelength, interference of the displacement beam with the pump beam may not be effective for extracting relative phase information. To address this issue, a separate control beam carried in the same optical channel as the pump and displacement beams can be used to probe the relative phase of the displacement and pump beams. The wavelength of the control beam is different from any other wavelengths used (including the wavelength of the pump beam, the wavelength of the displacement beam, and the wavelength of the resulting squeezed light). In addition, the difference between the wavelength of the control beam and the wavelength of the squeezed light can be large enough to facilitate high extinction filtering of the control beam before detection of the quantum light by a photon-counting detector. The phase information is extracted by interfering the control beams in different channels with each other and measuring the amplitude of the resulting control beam after interference.

[0037]

[1044] Following this phase-locking layer of the PDM 312, the input pump beam is split to generate balanced output pump beams (e.g., M output pump beams) that drive the squeezer 314. In some embodiments, the splitting can be achieved by a set of beam splitter trees. In some embodiments, the splitting can be achieved by an interferometer network. Before the pump beam is directed to the squeezer 314, the optical Power It may also be advantageous to independently control (eg, using RBS).

[0038]

[1045] Depending on the nature of the squeezer 314, the output pump light beam (i.e., the light beam driving the squeezer 314) can be continuous, pulsed, monochromatic, or a combination of one or more continuous inputs and one or more pulsed inputs of different wavelengths. For example, the squeezer 314 may generate squeezed light using a strong continuous drive beam and a weak pulsed beam of another wavelength. Details of this scheme for generating squeezed light can be found in U.S. Patent No. 10,649,307, filed May 12, 2020, entitled "INTEGRATED DEVICES FOR SQUEEZED LIGHT GENERATION," which is incorporated herein in its entirety. In this case, the PDM 312 allows the appropriate routing of the bichromatic inputs based on the operation of the squeezer 314. If the two separate wavelength components of the pump beam are carried in separate channels prior to the QPU, a similar phase stabilization mechanism can be incorporated to provide information about the necessary relative phase.

[0039]

[1046] In some embodiments, the PDM 312 is configured to operate with a bichromatic pump split into multiple inputs and separate displacement and control beams at each input. In these embodiments, the PDM 312 operates as follows: First, the PDM 312 receives in separate channels: (i) one or more pump inputs (continuous or pulsed) at a first wavelength, (ii) one or more pump inputs (continuous or pulsed) at a second wavelength, and (iii) one input at a third wavelength in a mode suitable for coherent displacement. A control beam that probes the phase displacement co-propagates with any input channel whose propagating beams have a wavelength substantially different from the others.

[0040]

[1047] The PDM 312 then obtains information about the relative phase of the inputs. This information can be provided to an external driver (e.g., 340) via a photodetector monitor (e.g., 313). The external driver then modulates a phase shifter (e.g., within the PDM 312 or before the PDM 312) to actuate a phase shift to correct for phase drift between the various inputs and generate a phase-locked pump beam. The PDM 312 then adjusts the tunable Power and directs them to the squeezer 314. The PDM 312 also directs the phase-locked displacement beam to the displacement module 318.

[0041]

[1048] 4A illustrates a schematic of a PDM 400 that can be used in the QPU 300 shown in FIG. 3, according to one embodiment. The PDM 400 can arbitrarily split an input signal 401 into 2N output channels (i.e., P1 to P2 as shown in FIG. 4A). 2N ). The PDM 400 includes 2N RBS units 410(1)-410(2N), each RBS unit 410 including an input port for receiving an input signal and two output ports for sending two output signals. In each RBS 410, the first output port sends the first output signal to an integrated delay line (IDL) unit, which can delay the phase of the first output signal by a desired amount. These delays can synchronize the array of outputs of the PDM 400. The second output port (e.g., 410(N)) of each RBS sends the second output signal to an input port of the next RBS (e.g., 410(N+1)), i.e., the second output signal serves as the input signal for the next RBS. The last RBS (i.e., RBS 2N ), the second output signal can be used as a test channel for calibration.

[0042]

[1049] The transmittance of each RBS (i.e., T1, T2, . . . , T 2N ) can be controlled by a corresponding phase shifter in each RBS (e.g., a TPS located in one arm of the RBS). These RBS units are then controlled to produce the desired output, which can be determined by the configuration of the PDM 400. Power , N。 For example, the odd output channels can carry pump light to drive the squeezer (e.g., 314 in FIG. 3), and the even output channels can carry light to perform the displacement (e.g., 318 in FIG. 3). That is, the (2j-1)th output, for j=1, 2, 3,..., N, corresponds to a channel carrying pump light for the squeezer element of the jth squeezer-displacer, while the (2j)th output corresponds to the displacer input of the jth squeezer-displacer.

[0043]

[1050] Output of a given sequence Power The RBS settings for generating the part can be determined as follows. Power P i Consider that the output channel i is in the i-th channel. Then, the input to the PDM is Power P in (ignoring losses) is P in =Σ i P i (1) can be written as:

[0044]

[1051] The total output from the i-th RBS Power Part α of i is α i =P i / P0 and therefore the output channel of the first RBS Power The transmittance is T1 = α1, and the output channel of the i-th RBS (i>1) Power The transmittance is

number

[0045]

[1052] In some cases, these ratios (α i ) can be adjusted to account for propagation losses. For example, the ratio (α i The adjustment of the ratio (α) can be determined during calibration of the PDM 400. i) may be performed during operation of the PDM 400 in response to changes in environmental parameters such as temperature.

[0046]

[1053] For pulsed operation, an IDL unit can be placed before the output port of each channel to provide an appropriate time delay that balances the optical path length of each channel. In this way, one input pulse can appear as a set of time-synchronized output pulses.

[0047]

[1054] The PDM 400 described herein can be understood as a subset of a directed binary tree, with each RBS corresponding to a node of the tree and the intervening waveguides corresponding to the branches connecting the nodes.

[0048]

[1055] Alternatively, the PDM can be realized by a full binary tree. In this case, the PDM can include a cascade of RBS layers. FIG. 4B shows a schematic of a PDM 420 based on a full binary tree structure. The first layer includes RBS 430(1) and the second layer includes RBS 430(2) and 430(3). Two layers are shown in FIG. 4B for illustrative purposes. Each layer of RBS can be split into two channels until the desired number of output modes is realized. This approach can use fewer RBS units (approximately half when N is large), but can have a greater depth in the horizontal dimension. Depending on the footprint constraints, the PDM can be replaced by such a full tree system.

[0049]

[1056] Both of the above approaches to constructing the PDMs 400 and 420 use a global reconfiguration for each desired power setting. In general, each RBS reconfigures the power in any channel. Power This approach adds complexity to the control system, but all inputs Power These techniques have the distinct advantage of being energy efficient because Power This is also called the efficiency method.

[0050]

[1057] Alternatively, another approach to PDM could use a fixed division ratio for each RBS, and then an attenuator controls the output, i.e., an equal division approach. This approach may have a simpler control system, but requires less power for the input Power Some of this may be lost during operation (eg, in an attenuator).

[0051]

[1058] Several factors can influence the decision about which method to use, such as large squeeze levels, a large number of modes in a realistic simulation (e.g., 5-10 modes), and large outputs from the PDM. Power In a simulation involving Power The efficiency approach is advantageous. In another example, if simplification of the control system is desired, an equal split approach followed by variable damping may be used.

[0052]

[1059] 3, the squeezer 314 is configured to generate squeezed light using tunable squeezing. In some embodiments, the output of the squeezer 314 includes a frequency-degenerate single-mode squeezed state. In some embodiments, the output of the squeezer 314 includes a two-mode or multi-mode squeezed state by utilizing multiple frequency, polarization, or time bins.

[0053]

[1060] Various approaches can be used to generate squeezed light by the squeezer 314. Typically, the squeezer 314 uses parametric interactions in a second- or third-order nonlinear optical medium to generate the squeezed light, i.e., spontaneous parametric down-conversion or spontaneous four-wave mixing. In some embodiments, the squeezer 314 employs a dual-pump scheme to generate squeezed light via spontaneous four-wave mixing in a resonant structure, such as a microring resonator. In some embodiments, the squeezer 314 uses a single-pump scheme to generate squeezed light via spontaneous four-wave mixing in a resonant structure. In some embodiments, spontaneous four-wave mixing can be performed in long waveguide segments (e.g., using either a dual-pump or single-pump scheme).

[0054]

[1061] In some embodiments, the squeezer 314 can generate the squeezed light using spontaneous parametric down-conversion in a resonant structure (e.g., a microring resonator or a system of coupled microring resonators). In some embodiments, spontaneous parametric down-conversion can be performed in a phase-matched waveguide segment or a quasi-phase-matched periodically poled waveguide segment.

[0055]

[1062] In some embodiments, in each of the squeezed light generation approaches described herein, auxiliary structures can be included in the squeezer 314 to enhance the attributes of the generated squeezed light. For example, it is advantageous for a resonant structure (e.g., a microring) to have a resonance that matches the generated squeezed light have a high escape efficiency (i.e., the efficiency with which the generated squeezed light couples out of the resonant structure). This can be achieved, for example, by strongly over-coupling the corresponding resonant mode with the resonant structure.

[0056]

[1063] The squeeze factor of the output of the squeezer 314 can be adjusted by the QPU driver 340 (e.g., manually by a user or automatically according to a particular protocol). Poweris a variable amount of optical pump Power to each squeezer 314. In some embodiments, for a resonant squeezer, the detuning of the pump beam away from a resonant mode in the resonator can be adjusted to change the squeeze factor. In some embodiments, a signal from the QPU driver 340 is used to activate an on-chip tunable phase shifter, which in turn performs the adjustment.

[0057]

[1064] In some embodiments, the squeezers 314 also include monitoring and active feedback for stabilization (especially in the case of resonant squeezer structures). In these embodiments, optical signals providing information about the status of each squeezer 314 are provided to a set of monitoring photodetectors (e.g., 315) in communication with the QPU driver 340. Electrical feedback signals are then conveyed from the QPU driver 340 to the chip (i.e., components fabricated on the substrate 305) to actuate stabilization changes, thereby implementing a feedback loop. In some embodiments, the monitoring optical signals can be obtained directly from the squeezer 314. This approach can be used, for example, if the squeezer 314 includes a resonant system in which pump light is conveyed to and extracted from the squeezer 314 in a separate optical channel from the generated squeezed light. In some implementations, the optical signals can be obtained from an output port of the filter 316, as described below.

[0058]

[1065] In some embodiments, the optical pump of the squeezer 314 PowerSome or all of the pump light is carried out through the same optical channel as the squeezed light. In these embodiments, it can be useful to divert as much of the pump light as possible (e.g., about 99% or more) prior to the interferometer 320. Such a diversion has at least two advantages. First, the diversion can provide optical information about the squeezer status via the post-squeezer pump light for monitoring and locking. Second, the diversion can reduce or eliminate noise generated after the squeezer 314 from unwanted linear effects driven by the pump light, such as spontaneous Raman scattering or four-wave mixing.

[0059]

[1066] In QPU 300, diversion of the pump light to squeezer 314 is accomplished by filter 316. Various types of filtering mechanisms can be used to construct filter 316. In some embodiments, filter 316 includes one or more drop-add ring filters configured to separate the pump light beam while transmitting the squeezed light. In some embodiments, filter 316 includes an asymmetric Mach-Zehnder interferometer (AMZI). In some embodiments, filter 316 includes a lattice filter (also called a cascaded AMZI). In some embodiments, filter 316 includes a Bragg grating structure. In some embodiments, filter 316 includes a coupled ring structure. In some embodiments, a TPS unit can be used to apply bias for filter 316.

[0060]

[1067] FIG. 5 shows a schematic of a displacement module 500 that can be used to generate displacement squeezed light of a GBS, according to one embodiment. The displacement module 500 can also be used as the displacement module 318 in the QPU 300 shown in FIG. 3. The displacement module 500 includes an array of RBSs 530(1)-530(N). Each RBS 530(j) receives a first input beam from a corresponding squeezer 510(j) and a second input beam (i.e., displacement beam) 520(j), where j=1, 2,...,N. The squeezers 510(1)-510(N) can collectively form the squeezer 314 of FIG. 3.

[0061]

[1068] As shown in Figure 5, the displacement can be achieved by mixing a displacement light beam provided by a PDM (e.g., PDM 312 in Figure 3) with squeezed light provided by squeezer 314 (e.g., after filter 316). The mixing can be performed with RBSs biased to transmit the squeezed light close to 100%. This can avoid adding significant losses to the squeezed light path while allowing variable displacement according to the transmission of each RBS set by the QPU driver. The phase of the displacement beam 520 can be controlled by a TPS element on the input waveguide segment that propagates the displacement beam 520 prior to the RBS 530.

[0062]

[1069] Each RBS 530 has two output ports: one port is configured to send the displacement squeezed light towards an interferometer (e.g., 320 in FIG. 3), and the other port can be configured to provide a signal to monitor a photodiode for status monitoring, feedback, and stabilization.

[0063]

[1070] 3, the interferometer 320 may be implemented by a network of beam splitters and phase shifters (e.g., RBS). The interferometer 320 may include N input ports 322 and N output ports 328, thereby performing a general transformation U(N) for N optical modes. In some embodiments, the N * An interferometer 320 can be constructed by interconnecting (N-1) / 2 RBSs and N phase shifters. * The (N-1) / 2 RBSs can perform the SU(2) transform, and the N phase shifters can perform the U(1) transform. Building the interferometer 320 using RBSs allows for low loss and a small physical footprint.

[0064]

[1071] The QPU 300 also includes a detector (not shown in FIG. 3). In some embodiments, the detector may be fabricated on the same platform 305. In some embodiments, the detector may be fabricated on a separate platform and may be detachable from the platform 305 (e.g., via fiber coupling).

[0065]

[1072] The detectors in QPU 300 are configured to perform a photon count readout of each separate optical mode provided by interferometer 320. In other words, the detectors in QPU 300 have a photon number resolution. In some embodiments, the detectors may include an array of sensors, each coupled to an output port 328 of interferometer 320 and generating data corresponding to the number of photons detected at a corresponding output port 328.

[0066]

[1073] In some embodiments, the detector comprises a transition edge sensor (TES) detector. TES detectors can provide accurate photon number resolution with negligible dark counts and high quantum efficiency, which can be advantageous for GBS applications. TES detectors usually operate at low temperatures, e.g., in the mK range. Thus, the TES detector can be coupled to the interferometer 320 via optical fiber (instead of fabricating the TES detector on the platform 305). Data acquisition of the TES detector can be achieved by digitizing the voltage pulses coming from the detector and performing appropriate computational analysis to resolve the number of photons in each pulse and channel. This data acquisition can be triggered by appropriate timing pulses provided by the QPU control system or the input pump subsystem.

[0067]

[1074] In some embodiments, the detector may include any other suitable type of photon counting detector, for example, a single-photon counting detector, such as a superconducting nanowire single-photon counting detector or a single-photon sensitive avalanche photodiode. In some such embodiments, a suitable multiplexing scheme (e.g., spatial, temporal, or modal multiplexing) may be used to ensure that the probability of more than one photon being incident on any single detector element is low. Such a multiplexing strategy may involve some multiple L of detectors, where L is a multiple of N and is large enough to suppress multi-photon counting events per detector. To ensure that the detector detects only the desired photons, the optical fiber connected to the detector may include multiple stages of filtering.

[0068]

[1075] The GBS device described herein above with reference to Figures 1-5 can be configured such that the output samples (i.e., statistics of the number of photons in each optical mode at the output) contain useful information. By varying the input squeeze and / or N-mode linear interferometer settings, a wide range of problems can be encoded into the GBS. These problems can be graph-based or combinatorial optimization based and have several practical applications including social network optimization and chemistry. The output samples from a suitably programmed GBS device can then be used to provide potential solutions to the encoding problem. These samples may be processed by a classical computer, including as part of a classical-quantum hybrid algorithm. The results can further lead to a reconfiguration / update of the GBS device as part of a continuous algorithm.

[0069]

[1076] FIG. 6 shows a block diagram illustrating a method 600 of operating a GBS device, according to one embodiment. The method 100 includes incorporating an application 610 into a GBS system 620 (e.g., any GBS device shown and / or described with reference to FIGS. 1-5). For example, the squeeze factor and / or phase settings of an interferometer in the GBS system can be configured to solve a specific problem (e.g., a graph or vibronic spectrum, as described below). The output 630 of the GBS system 620 is then sent to a classical computer 640 (or simply a processor) for post-processing (e.g., analysis of the photon distribution in each optical mode) to search for a solution 650 of the specific problem incorporated into the GBS system 620. Additionally, processed information from the classical computer 640 can be sent to the application 610 to improve the implementation (e.g., adjustment of the input optical mode or phase settings of the interferometer). Example applications that can be incorporated into the GBS system 620 include identifying highly connected influencers in social networks, improving the efficiency of solar cells, or reducing the time it takes for customers to place and receive their online orders.

[0070]

[1077] The GBS devices described herein may also enable user interaction such as state generation and setting parameters of linear optical interferometer components, etc. To this end, the hardware running the GBS-based application may be programmable, i.e., squeeze parameters, coherent displacement settings (if applicable), and interferometer matrix may be specified and operated accordingly.

[0071]

[1078] FIG. 7 shows a schematic of a GBS system 700 that provides user access to the configuration of quantum hardware, according to one embodiment. In the GBS system 700, the hardware elements responsible for input state preparation, interferometric conversion, and output state detection are hereafter referred to as quantum hardware (QH) 750 (e.g., similar to QH 160 in FIG. 1). Parameterization of the QH can be accomplished using a set of classical computers and associated communication interfaces. More specifically, a master unit 730 is dedicated to running, monitoring, and receiving results from the QH 750 via a set of control systems 740. Clients can access the master unit 730 (and thus the QH 750) via a user interface 710. In some embodiments, the user interface 710 is locally connected to the master unit 730. In some embodiments, clients can access the master unit 730 via the Internet (i.e., the user interface 710 includes a network interface). In either case, instructions can be passed to the hardware via an application program interface (API). In the Internet access model (also called the cloud access model), clients can communicate with the server 720 via a secure web link, which in turn communicates with the master unit 730.

[0072]

[1079] As described herein, the GBS devices and systems shown in Figures 1-7 can be used to solve a wide range of problems. Two examples: graph problems and vibronic spectra are described in more detail below for illustrative purposes.

[0073]

[1080] Solving problems with graphs has both theoretical and practical interest due to its applications in computer vision, social networks, and finance. For example, graph problem algorithms can help identify highly connected influencers in social networks, improve the efficiency of solar cells, and reduce the time it takes for customers to place and receive online orders. Graph problems can be specified through the adjacency matrix of the graph, which can be encoded by varying the input squeeze and N-mode linear interferometer settings. In this encoding, each vertex of the graph corresponds to a mode of the device / system in the GBS. More details can be found, for example, in Kamil Bradler, Pierre-Luc Dallaire-Demers, Patrick Rebentrost, Daiqin Su, and Christian Weedbrook, Gaussian boson sampling for perfect matchings of arbitrary graphs, Phys. Rev. A. vol. 98, page 032310, Sep 2018, which is incorporated herein by reference in its entirety.

[0074]

[1081] FIG. 8 illustrates a block diagram showing a method 800 of solving a graph problem using a GBS device described herein, according to one embodiment. In the method 800, a graph problem 810 is solved using algorithms and heuristics 820 to arrive at a solution 830, and a GBS system 850 can be used to improve the solution. More specifically, the graph problem 810 can first be encoded (at 840) into a GBS system. Next, samples are taken from the GBS system 850 (e.g., any of the GBS systems shown in and described with reference to FIGS. 1-7) and processed in the algorithms and heuristics 820 to help solve the graph problem. In some embodiments, the processing includes iteratively updating the GBS device based on its output. The GBS model of computation can be adapted within the algorithm to find dense subgraphs and compute graph matching, graph isomorphism, and graph similarity, which are described in more detail below.

[0075]

[1082] Identifying highly connected subgraphs is a relevant problem in bioinformatics, drug design, data mining, finance, and community detection in social networks. Unfortunately, the computational resources required to find these subgraphs grow rapidly with the size of the graph. GBS can be utilized to randomly generate highly connected subgraphs with high probability. This feature allows GBS to improve existing conventional algorithms that rely on random searches over the space of possible subgraphs.

[0076]

[1083] Classical algorithms often proceed with a mixture of global exploration of the problem space and local searches for solution candidates that exploit local structure. Both the exploration and local searches can include probabilistic elements, e.g., choosing a random subgraph as a starting point or a local search to build larger cliques. With the goal of finding improved solution candidates, these probabilistic elements are often biased towards the selection of highly connected subgraphs. An appropriately encoded GBS system can then be used as a probability source for selecting highly connected subgraphs.

[0077]

[1084] The graph isomorphism problem is the decision problem as to whether two isospectral graphs (i.e., graphs with the same eigenspectrum) are isomorphic, i.e., related to each other, by a simple relabeling of the vertices. The main open question is whether there exists a polynomial-time algorithm that can determine whether two graphs are isomorphic or not. In practice, graph isomorphism likely belongs to the class of NP intermediate computation problems. A further generalized problem concerns how two non-isomorphic graphs (including graphs with different numbers of vertices) are related to each other, i.e., the graph similarity problem. Graph isomorphism, but mainly graph similarity, provides solutions to problems as different as computer vision, detecting anomalous activity (e.g., fraudulent withdrawals) in financial trading networks, and identifying attributes of molecular families without the need for synthesis (e.g., in medicine, genetics, or biochemistry).

[0078]

[1085] GBS can be utilized to yield a complete set of graph invariants, and two graphs are isomorphic if and only if these graph invariants are equal. To this end, we first describe the link between graph isomorphism and GBS. It can be understood by considering a cyclic permutation of photon events from GBS. One useful concept is the orbit representative, which serves as a seed event that describes all permutation-equivalent events that form an orbit. For example, an orbit (0,0,1,2) describes possible events (0,1,2,0), (2,1,0,0), etc. These events can be associated with output samples (also called click patterns) from a GBS device, and each orbit has a corresponding GBS probability. The orbit probabilities can be found and compared for two distinct graphs. If the orbit probabilities differ, it is sufficient to conclude that the two graphs are not isomorphic.

[0079]

[1086] Implementation of this scheme can proceed by encoding the two graphs on the GBS device as described above and evaluating the probability of a given trajectory by taking multiple samples from the GBS device. In some embodiments, the graphs can be encoded on the GBS device sequentially, i.e., evaluating the trajectory probability of one graph and then evaluating the trajectory probability of another graph. In some embodiments, the graphs can be encoded on the GBS device alternately, i.e., taking a sample from one graph and then a sample from the other graph, successively updating the trajectory probability. If experimental uncertainty is also considered, enough samples can be taken such that the trajectory probabilities do not overlap. This can be useful to conclude that the two graphs are non-isomorphic.

[0080]

[1087] To explain the application of GBS to graph similarity problems, it is also useful to link GBS trajectories with two concepts in graph theory: k-matchings and matching polynomials. k-matchings in a simple (i.e., undirected and loop-free) graph with 2M vertices is the number of k-independent (sparsely related) edges in the graph. By definition, 1≦k≦M. The k-matchings are the coefficients of the graph matching polynomial. Computing the matching polynomial can be a difficult computational problem that becomes intractable for conventional computers when M is large enough. In fact, there are no known efficient conventional algorithms to even approximate the matching polynomial, except for certain special classes of graphs.

[0081]

[1088] A GBS device can be used to estimate a matching polynomial through the use of a GBS polynomial, which is of the same order as the matching polynomial and whose coefficients are related to, but not identical to, the k-matchings of the graph. Some of the terms (also called coefficients) of the GBS polynomial can be efficiently inferred using a GBS device, and furthermore, the GBS polynomial is more suitable for graph isomorphism or similarity problems than using information that can be obtained from the matching polynomial of the graph.

[0082]

[1089] GBS polynomials can be extracted from the GBS device output as follows: The output click pattern of a GBS device is called collision-free if at most one photon is detected in each output mode. When a graph is encoded with a GBS device, it can be shown that the probability of observing a collision-free trajectory with 2k photons is equal to the k-th coefficient of the GBS polynomial. An important aspect of sampling trajectories with 2k photons is that it is tractable for some k. This is because the number of collision-free trajectories simply grows polynomially with the encoded graph size 2M. Since the trajectory probability is a graph invariant, it provides an efficient way for GBS to determine whether two graphs are isomorphic or not.

[0083]

[1090] The extension to graph similarity is given by considering the full range of possible trajectories given a constant total number of photons. This takes the problem out of the collisionless regime and therefore requires a corresponding generalization of the GBS polynomial insight. Higher order probability distributions contain combinations of different trajectory probabilities. These coarse-grained probability distributions are the coefficients of another GBS polynomial corresponding to the extended graph obtained by tensorizing the original graph with a complete graph with self-loops.

[0084]

[1091] There are three aspects to the above process. First, the coarser the probability distribution, the easier it is to sample from it using the GBS device. Second, the tensorization procedure can amplify the differences between two graphs. It is the comparison of the extended GBS polynomial coefficients that makes it suitable for graph similarity problems. Taking coarser trajectories with a larger number of photons allows us to obtain more precise information about the differences between two graphs. Finally, these insights lead to a comparison of the matching polynomial (as if we had access to it) with the GBS polynomial, and we find that in all cases the GBS polynomial reveals the dissimilarity of the two graphs with fewer detected photons. This last point suggests that the GBS polynomial can be more powerful than the matching polynomial for this type of problem, and this has consequences for the experimental implementation of the algorithm.

[0085]

[1092] A GBS scheme that may be more suitable for graph similarity problems involves an initial uniformly displaced squeezed state that enters a linear interferometer to encode the graph. The regimes considered can be both collision-containing or collision-free. A comparison of the detected click patterns with their coarse-grained versions leads to a measure of the similarity of the encoded graphs. In general, by evaluating the trajectory distribution through GBS, with or without displacement, the user can associate the encoded graph with a feature vector filled with the trajectory probabilities. This association can be understood as an embedding of the graph into the real space using GBS, which can be used to formulate a graph kernel. The graph kernel and feature vector can be used within machine learning such as classification and recursion. This concept can be extended to a form of graph classification or clustering using methods such as support vector machines and k-means clustering.

[0086]

[1093] Based on the above description, a GBS-based solution to a graph problem (e.g., a graph isomorphism or similarity problem) can be performed as follows: First, the graph is made GBS-encodable by converting it into a Gaussian covariance matrix. The Gaussian covariance matrix is ​​used to find the squeeze parameters as well as the unitary angle and phase values ​​of the linear interferometer in the GBS device. The vacuum displacement values ​​optimized for the displacement-based version of the setup are found. Next, the statistics of the click patterns are collected from the detectors in the GBS device. Then, classical post-processing gives coarse-grained statistics. The trajectory probabilities are estimated in the collision or collisionless regime. The trajectory probabilities are interpreted as the coefficients of the GBS polynomial of the graph or its tensor expansion.

[0087]

[1094] Then, in a graph isomorphism problem, the difference in the estimated GBS polynomial coefficients is used to conclude that the two graphs are not isomorphic. In a graph similarity problem, the associated estimated GBS polynomial coefficients are associated with feature vectors and processed in a machine learning layer to find the similarity of the two graphs. For example, the feature vectors are processed by a classical computer running a machine learning algorithm such as a neural network or kernel machine, among others. Finally, the decision problem or similarity measure can be reported to a user.

[0088]

[1095] Efficient modeling of molecular vibronic spectra is useful for understanding and engineering chemical reactions for both academic research and industrial purposes. Accurate prediction of the structure of molecular vibronic spectra can be a computationally intensive task, and typically involves prohibitively large computational resources as the size and complexity of the target molecules increase. Therefore, strategies to efficiently perform these calculations represent an important step forward for chemical simulations.

[0089]

[1096] The GBS system enables such a strategy by exploiting the link between the dynamics of vibronic modes and the dynamics of quantum optics. For a molecule with n vibrational modes, the GBS strategy involves the provision of n displaced squeezed states, an n×n linear optical interferometer, and photon-number-resolved detection at the output mode. In all the above cases, the associated GBS algorithms scale better with problem size in computational resources and / or number of computational steps than any known classical algorithm.

[0090]

[1097] In a GBS-based approach to solving vibronic spectrum problems (e.g., the system shown in FIG. 7), a client can first communicate with a server over a secure web link using a suitable graphical or command line user interface or custom application program interface (API). The information sent by the client to the server describes the molecular structures to be simulated. These molecular structures are then interpreted by an interpreter to determine the physical parameters (e.g., optical Power , beam splitter ratios, and phases) to a control system that includes auxiliary drivers that communicate directly with the quantum hardware. The quantum hardware includes optical chips and photon-counting detectors. Output from these detectors is fed to software running on a decoder that parses the photon-counting distribution into a usable representation of the vibronic spectrum. This information is passed to a server, which sends the results of the calculations back to the client.

[0091]

[1098] In some embodiments, the client access model shown in FIG. 7 can use a molecular coding scheme described in Huh et al., Boson sampling for molecular vibronic spectra, Nature Photonics 9, 615 (2015), which is incorporated herein by reference in its entirety. The interpreter uses this coding scheme to convert the vibronic transitions to be simulated into a set of parameters including the beam splitter ratio and phase of the interferometer, the squeeze factor, and the displacement to be performed by the squeezer / displacer. In the system described herein, these parameters can be controlled on-chip by the phase at each module. Thus, the interpreter ultimately tells the control system what phase to set at each module. The control system converts this into a set of voltages to be applied to the phase shifters.

[0092]

[1099] In the case of decoding, the photon counts measured by the detector can precisely form a probability distribution that constitutes a set of Franck-Condon (FC) factors that can be used to reconstruct the vibronic spectrum. The detector can therefore take this photon count distribution, compile it into a set of FC factors, and report the results to the client via the server.

[0093]

[1100] In addition to solving graph problems and modeling vibronic spectra, the GBS system described herein can also be used as a point process. Without being bound by any particular theory or mode of operation, a point process is a statistical model that generates and interprets random point patterns that may appear in nature, technology, and human affairs. Matrix function point processes are a class of point processes in which the probability of observing a particular point pattern depends on matrix functions such as determinants and permanents. Determinant point processes can be applied to a variety of problems in machine learning, including text summarization, human pose estimation, and news threading. However, point processes that depend on permanents or Hafnians have not been widely used due to the computational difficulties associated with computing these matrix functions on classical computers.

[0094]

[1101] A GBS machine can be viewed as a point process that generates random point patterns in the form of photons in a particular optical mode. The probability of a pattern generated by the GBS is proportional to the Hafnian of the matrix that specifies the GBS configuration. Thus, a GBS machine can be programmed to generate point patterns that inherit typical patterns of Hafnian and permanent point processes. As a result, GBS eliminates the difficulties associated with the application of Hafnian and permanent point processes on classical computers. In particular, two examples of GBS point processes are provided below.

[0095]

[1102] One feature of the GBS point process is that there is a higher probability of point patterns being locally clustered. This inherent clustering property of GBS can be used for portfolio optimization and machine learning tasks. One aspect of portfolio optimization is diversifying assets to avoid having highly correlated stocks in one portfolio. The GBS machine can be programmed with a stock covariance matrix that is derived based on the time correlation of stock properties such as market prices. This GBS point process is capable of detecting highly correlated stocks, and thus helps provide a well diversified and stable portfolio. The GBS point process can also be combined with classical algorithms to increase the efficiency of classical machine learning clustering methods. In particular, the GBS point process can be used to provide more realistic cluster centers used to initiate classical clustering algorithms such as k-means.

[0096]

[1103] Although various embodiments have been described and illustrated herein, a wide variety of other means and / or structures for performing the functions and / or obtaining one or more of the results and / or advantages described herein, and each of such variations and / or modifications, are possible. More generally, it is intended that all parameters, dimensions, materials, and configurations described herein are examples, and that the actual parameters, dimensions, materials, and / or configurations will depend on the particular application or applications for which the disclosure is used. It is understood that the above embodiments are presented merely as examples, and that other embodiments may be practiced other than those specifically described and claimed. An embodiment of the present disclosure relates to each individual feature, system, article, material, kit, and / or method described herein. In addition, any combination of two or more such features, systems, articles, materials, kits, and / or methods is included within the inventive scope of the present disclosure, if such features, systems, articles, materials, kits, and / or methods are not mutually inconsistent.

[0097]

[1104] Also, various concepts may be implemented as one or more methods, examples of which are provided. The acts performed as part of a method may be ordered in any suitable manner. Thus, even if an example embodiment shows acts as sequential, embodiments may be constructed in which acts are performed in a different order than that shown, which may include performing some acts simultaneously.

[0098]

[1105] All definitions defined and used herein are understood to take precedence over dictionary definitions, definitions in documents incorporated by reference, and / or ordinary meanings of the defined terms.

[0099]

[1106] As used herein, a "module" may be, for example, any assembly and / or set of operably coupled electrical components associated with the performance of a particular function, and may include, for example, memory, processors, electrical traces, optical connectors, software (stored in and executed by hardware), etc.

[0100]

[1107] The indefinite articles "a" and "an," as used in the specification and claims, unless clearly indicated to the contrary, should be understood to mean "at least one."

[0101]

[1108] The phrase "and / or," as used herein and in the claims, should be understood to mean "either one or both" of the elements so joined, i.e., elements that may be conjunctive or disjunctive. Multiple elements listed with "and / or" should be understood in the same manner, i.e., "one or more" of the elements so joined. Other elements other than the elements specifically identified by the "and / or" clause may optionally be present, whether related to the specifically identified elements or not. Thus, as a non-limiting example, a reference to "A and / or B," when used in conjunction with an open-ended term such as "comprising," can refer in one embodiment to only A (optionally including elements other than B); in another embodiment to only B (optionally including elements other than A); in yet another embodiment to both A and B (optionally including other elements), etc.

[0102]

[1109] As used herein and in the claims, "or" should be understood to have the same meaning as "and / or" as defined above. For example, when separating items in a list, "or" or "and / or" shall be construed to be inclusive, i.e., including at least one of the elements or list of elements, but including two or more and optionally additional unlisted items. When used only with terms clearly indicating the contrary, such as "only one of" or "exactly one of," or in the claims, "consisting of" refers to the inclusion of exactly one element of the elements or list of elements. In general, the term "or," as used herein, shall be construed to indicate exclusive alternatives (i.e., "one or the other, but not both") only when used in conjunction with terms of exclusivity, such as "any one of," "one of," "only one of," or "exactly one of." "Consisting essentially of," when used in the claims, shall have its ordinary meaning as used in the field of patent law.

[0103]

[1110] As used herein and in the claims, the phrase "at least one" in reference to a list of one or more elements should be understood to mean at least one element selected from any one or more elements in the list of elements, but does not necessarily include at least one of each and every element specifically listed in the list of elements, and does not exclude any combination of elements in the list of elements. This definition also allows for elements other than the elements specifically identified in the list of elements to which the phrase "at least one" refers, whether related to the specifically identified elements or not, to be optionally present. Thus, as a non-limiting example, "at least one of A and B" (or, equivalently, "at least one of A or B" or, equivalently, "at least one of A and / or B") can refer, in one embodiment, to at least one A, optionally including more than one, and no B (optionally including elements other than B); in another embodiment, to at least one B, optionally including more than one, and no A (optionally including elements other than A); in yet another embodiment, to at least one A, optionally including more than one, and at least one B, optionally including more than one (optionally including other elements), etc.

[0104]

[1111] In the claims, as in the specification above, all transitional phrases such as "comprises," "includes," "have," "having," "including," "involving," "holding," "comprising," and the like, shall be understood to be open-ended, i.e., to mean including but not limited to. Only the transitional phrases "consisting of" and "consisting essentially of" shall be closed or semi-closed transitional phrases, respectively, as set forth in Section 2111.03 of the United States Patent Office Guidelines for Patent Examining Procedures.

Claims

1. a light source configured to provide a plurality of input optical modes with squeezed state light; a network of interconnected reconfigurable beam splitters (RBS) in optical communication with the light source, the network of interconnected RBSs configured to perform unitary transformation of the plurality of input optical modes to generate a plurality of output optical modes; an array of photon-counting detectors in optical communication with the network of interconnected RBSs, the array of photon-counting detectors configured to measure a number of photons in each of the plurality of output optical modes after the unitary transformation; a controller operatively coupled to the light source and the network of interconnected RBSs, the controller being configured to perform at least one of adjusting an angle of the unitary transformation or adjusting a phase of the unitary transformation to represent an adjacency matrix of a graph, each input optical mode in the plurality of input optical modes representing a vertex of the graph; An apparatus comprising:

2. The apparatus of claim 1 , wherein the light source is configured to provide a plurality of optical modes in a displaced squeezed state.

3. The light source is a power division module (PDM) that receives an input optical beam and divides the input optical beam into 2N power division module (PDM) beams, where N is a positive integer; a squeezer-displacer module (SDM) in optical communication with the PDM to receive the 2N PDM beams and generate the plurality of input optical modes having N squeezed optical beams; The apparatus of claim 1 , comprising:

4. 2. The apparatus of claim 1, wherein the network of interconnected RBSs includes an array of input ports that direct the plurality of input optical modes into the network of interconnected RBSs and an array of output ports that send the plurality of output optical modes to the array of photon counting detectors.

5. The apparatus of claim 1 , wherein the array of photon counting detectors comprises an array of superconducting transition edge detectors.

6. The apparatus of claim 1 , further comprising a substrate comprising at least one of silicon, silicon oxide, or silicon nitride, wherein the light source and the network of interconnected RBSs are fabricated in the substrate.

7. The apparatus of claim 1 , wherein the light source is configured to provide the plurality of input optical modes that represent an adjacency matrix of a graph, each input optical mode in the plurality of input optical modes representing a vertex of the graph.

8. The light source is configured to provide the plurality of input optical modes that represent an adjacency matrix of a graph, each input optical mode in the plurality of input optical modes representing a vertex of the graph, and the apparatus further comprises:

2. The apparatus of claim 1 , further comprising a processor operatively coupled to the array of photon-counting detectors and configured to find a dense subgraph in the graph based on a number of photons in each output optical mode of the plurality of output optical modes after the unitary transformation.

9. The light source is configured to provide the plurality of input optical modes that represent an adjacency matrix of a graph, each input optical mode in the plurality of input optical modes representing a vertex of the graph, and the apparatus further comprises:

2. The apparatus of claim 1, further comprising a processor operatively coupled to the array of photon-counting detectors and configured to calculate a graph invariant that determines a graph isomorphism and a GBS polynomial for the graph isomorphism based on a number of photons in each output optical mode of the plurality of output optical modes after the unitary transformation.

10. The light source is configured to provide the plurality of input optical modes that represent an adjacency matrix of a graph, each input optical mode in the plurality of input optical modes representing a vertex of the graph, and the apparatus further comprises:

2. The apparatus of claim 1, further comprising a processor operatively coupled to the array of photon-counting detectors and configured to evaluate graph isomorphism based on a number of photons in each output optical mode of the plurality of output optical modes after the unitary transformation.

11. a user interface operatively coupled to the light source and the network of interconnected RBSs and configured to receive information about molecular structures provided by a user; an interpreter operatively coupled to the user interface to generate control signals based on the information about the molecular structure, the controller being configured to cause the light source to generate the plurality of input optical modes in response to the control signals; The apparatus of claim 1 further comprising:

12. 10. The apparatus of claim 1, further comprising a decoding unit operatively coupled to the array of photon-counting detectors and configured to convert an array of detection signals acquired by the array of photon-counting detectors into a molecular vibronic spectrum representation.

Citation Information

Patent Citations

  • Methods, systems, and apparatus for programmable quantum photonic processing

    US20160245639A1