Light quantum computing chip structure oriented to quantum neural network

Abstract

The invention discloses a light quantum chip which is combined with high-dimensional coding, is provided with two data coding layers and two trainable entanglement layers, is compatible with a data recoding technology, and can be used for variable component subtasks such as a quantum neural network. The chip structure comprises a cascaded Mach-Zehnder interferometer binary tree array, d groups of photon pair sources, d groups of controlled unitary gate modules, a multi-quantum bit control Z gate, four d-dimensional adjustable linear networks and two state tomography linear networks. The cascaded MZI binary tree array and the controlled unitary group jointly form a first coding layer and a parameter-containing entanglement layer, and the adjustable optical linear network and the multi-quantum bit control Z gate jointly form a second coding layer and a parameter-containing entanglement layer. The two state chromatography linear networks can further improve the computing power of the chip in a mode of changing a measurement basis or executing quantum state chromatography. According to the method, the expression ability of the light quantum neural network is improved, and the development of a light quantum computing chip and the application of the light quantum computing chip in actual tasks are facilitated.

Description

Technical Field

[0001] This invention relates to the field of quantum information technology, and more specifically, to an optical quantum computing chip structure for quantum neural networks. Background Technology

[0002] Quantum computers, leveraging properties such as superposition and entanglement, exhibit quantum advantages over classical computers in specific tasks. For example, Shor's algorithm can crack RSA keys in polynomial time, while classical computers require exponentially longer times. However, current quantum computers are in the era of noisy, medium-scale quantum computing, with a limited number of qubits and noisy quantum gates, making it impossible to run algorithms like Shor's algorithm, which requires a fault-tolerant quantum computing environment.

[0003] Quantum Neural Networks (QNNs) combine quantum computing and classical neural networks, possessing higher expressive power and potentially faster operating speeds, and effectively utilizing currently limited quantum resources to solve practical problems. A typical QNN consists of an encoding layer, parameterized quantum circuits (PQCs), and a measurement layer. The encoding layer encodes classical data into quantum states; the PQC has multiple training modules composed of parametric quantum gates, whose parameters can be trained to approximate the objective function; the measurement layer reads the results through quantum measurements and optimizes the PQC parameters using a loss function. To improve the expressive power of QNNs, on the one hand, entanglement gates need to be introduced into the PQC training modules, and the number of training module layers needs to be increased; on the other hand, data recoding techniques can be used (Physical Review A, 2021, 103(3): 032430), that is, adding an additional encoding layer to each training module of the PQC, which can introduce nonlinearity into the QNN.

[0004] QNNs can be implemented using quantum computing technologies such as superconductivity, ion traps, and optics. Among these, integrated optics is one of the most promising platforms for quantum computing, and photonic quantum chips based on it offer advantages such as small size, programmability, and room-temperature operation. Photonic quantum chips have been applied to supervised and unsupervised quantum machine learning tasks, but current photonic quantum chips can only implement a single training layer, limiting the expressive power of QNNs and thus restricting their application scope.

[0005] Therefore, a new solution is needed to address this problem. Summary of the Invention

[0006] To enhance the expressive power of photonic quantum neural networks (QNNs), this invention proposes a photonic quantum chip that combines high-dimensional encoding with two encoding layers and two trainable entangled layers, compatible with data recoding techniques. Furthermore, the chip is equipped with two state tomography layers to further improve its computational capabilities.

[0007] The above-mentioned technical objective of this invention is achieved through the following technical solution: a photonic quantum computing chip structure for quantum neural networks, comprising a cascaded MZI binary tree array, d groups of photon pair sources, d groups of controlled unitary gates, a multi-qubit controlled Z-gate, four d-dimensional tunable linear networks, and two d-dimensional state tomography linear networks, wherein, The cascaded MZI binary tree array is connected to the photon pair source, and the cascaded MZI binary tree array adjusts the pump power of the photon pair source to control the photon generation rate of each photon pair source. Each set of photon source sources can generate identical photon pairs that are anti-focused and input them into the controlled unitary gate group in a specific order; The controlled unitary gate group performs a unitary transformation operation controlled by multiple qubits; The multi-qubit control Z-gate enables photon-to-photon interaction to achieve entanglement; The d-dimensional tunable linear network transforms the quantum state of a single photon. The d-dimensional tunable linear network includes linear network one, linear network two, linear network three, and linear network four. The photon pair includes photon one and photon two. Linear network one and linear network three transform the state of photon one, and linear network two and linear network four transform the state of photon two. Linear network one and linear network two serve as intermediate layers connecting the controlled unitary gate group and the multi-qubit control Z-gate. Linear network three and linear network four are connected after the multi-qubit control Z-gate. The d-dimensional state tomography linear network transforms or performs quantum state tomography on the measurement basis of the quantum states of photon one and photon two.

[0008] Furthermore, the waveguide path of each photon in the photon pair is d-dimensional, satisfying... , in It is an integer greater than or equal to 1, and each waveguide corresponds to 1 quantum ground state.

[0009] Furthermore, the cascaded MZI binary tree array, combined with a pump photon pair source, generates entangled states with tunable coefficients: The coefficients satisfy .

[0010] Furthermore, the output port of the pumped photon pair source is connected to the controlled unitary gate group in a specific order, such that the input quantum state is: .

[0011] Furthermore, the controlled unitary group comprises d CU gate modules, each of which performs a multi-qubit controlled unitary gate operation on the quantum states represented by two adjacent waveguides, such that the quantum states are: .

[0012] Furthermore, the cascaded MZI binary tree array and the controlled unitary group together constitute the first coding layer and the first parametric entanglement layer; As a globally entangled state, by changing the phase shifter parameters, one qubit can be extracted from qubit 1 and one from qubit n to 2n-1. The gate encodes 2n+2 classic data.

[0013] Furthermore, the complete linear network one and linear network two perform arbitrary unitary transformations on photon one and photon two respectively, extracting one unitary transformation from qubit 1 and one unitary transformation from qubit n to 2n-1. The gate serves as the second coding layer.

[0014] Furthermore, the multi-qubit control Z-gate comprises 2d-1 MZIs with a splitting ratio of 1:2, and the input and output of the d-th MZI are connected to the d-1 path of photon 1 and the 0 path of photon 2, respectively. When both photons enter, and the two photons are then selected to output from two ports, the quantum interference introduces one phase into the quantum state. .

[0015] Furthermore, the multi-qubit control Z-gate has adjustable linear networks before and after it, which together realize the second parametric entanglement layer.

[0016] In summary, the present invention has the following characteristics: 1) The pump light is tunably distributed to each photon pair source by a cascaded MZI binary tree array to change the photon pair generation rate.

[0017] 2) The CU gate group and the photon pair source are connected in a specific waveguide order to meet the initial state requirements for executing a multi-qubit controlled unitary gate, and each CU gate module is a simple 2D case.

[0018] 3) By combining high-dimensional coding and quantum interference, a multi-qubit control Z-gate between two photons was realized.

[0019] The beneficial effects of this invention are: This invention provides an optical quantum computing chip architecture with two encoding layers and a global entanglement layer, which can be extended to higher dimensions. Compared with optical quantum chip structures that only have photon pairs pre-entangled at the source, the chip of this invention has stronger expressive power, computational power, and scalability, and can be used in complex quantum machine learning tasks and variable quantum algorithms. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the present invention, in which each solid line represents one optical waveguide; Figure 2 This is a schematic diagram of some of the optical structures used in this invention; Figure 2 In the diagram, (a) represents the structure of a cascaded MZI binary tree, which consists of d-1 MZIs; Figure 2 (b) is a schematic diagram of the state tomography network, which contains 2(d-1) phase shifters; Figure 2 (c) in the diagram represents the complete tunable linear optical network structure. Figure 3 This is a schematic diagram of the remaining optical structure used in this invention; Figure 3 (a) in the diagram is the specific structure of the photon pair source, which can generate and separate identical photon pairs into two paths; Figure 3 (b) in the diagram represents the specific structure of the CU gate; Figure 3 (c) in the diagram represents the specific optical structure of the multi-qubit controlled Z-gate; Figure 4 The quantum circuitry corresponding to the optical quantum chip was demonstrated. Figure 4 (a) in the middle is Figure 1 The corresponding overall quantum circuit; Figure 4 (b) in the diagram represents the quantum circuit corresponding to the cascaded MZI binary tree array; Figure 4 (c) in the figure represents the correspondence between the basic parametric quantum gate and the specific optical structure. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] Example 1: The quantum chip structure of the present invention is as follows: Figure 1 As shown, each solid line represents an optical waveguide. The chip comprises one cascaded MZI binary tree structure, d sets of photon pair sources, d sets of CU gates, one multi-qubit controlled Z-gate, four d×d tunable linear optical networks, and two state tomography optical networks. The structure of the d-dimensional cascaded MZI binary tree is shown below. Figure 2 As shown in (a), it consists of d-1 cascaded tunable MZIs used to adjust the pump power of the photon pair source. The photon pair source is used to generate identical, path-separated photon pairs, and the structure of each photon pair source is as follows. Figure 3 As shown in (a), it includes two 50:50 MMIs and two nonlinear single-photon sources, with the phase shifter set to... This ensures deterministic separation of photon pairs generated by the single-photon source at the output, and the upper output waveguide is equipped with a phase shifter. CU gates are used to perform controlled unitary operations on multiple qubits; the structure of each CU gate group is as follows: Figure 3As shown in (b), the input waveguide is extended into 2D by the MZI, and two MZIs perform a complete 2D unitary transformation. Then, paths 2 and 3 are swapped using cross-waveguides, and the photon path information is erased by two 50:50 MMIs. The output only selects the results from paths 1 and 3. The structure of the multi-qubit controlled Z-gate is as follows: Figure 3 As shown in (c), the beam splitting ratio of each MZI is adjusted to 1:2 (corresponding to a phase shifter phase of approximately 1.231), resulting in a two-photon transmission probability of 1 / 9. When the input quantum state is That is, when two photons simultaneously enter the middle MZI, quantum interference occurs, and at this time the quantum state introduces an additional ion. Phase shift, quantum state becomes The remaining input quantum states remain unchanged. The d×d tunable linear optical network, implemented using an MZI array based on the Clement architecture, can perform complete unitary transformations on single-photon states. Figure 3 (c) shows an example of a 4x4 MZI array. The quantum state tomography network is also implemented using an MZI array, with the structure shown below. Figure 3 As shown in (b), quantum states can be measured on any basis. The chip has a total of 2 output terminals. root waveguide, satisfying , in The integers are greater than or equal to 1, and each waveguide corresponds to one quantum ground state. The output single photons will be detected by a single-photon detector, and only the result of selecting photon 1 to be output from the first d waveguides and photon 2 to be output from the last d waveguides will be considered.

[0023] The quantum circuits corresponding to the chip, such as Figure 4 As shown in (a), each solid line represents 1 qubit, and the first n qubits ( ) is implemented by photon 1, and the next n qubits ( The quantum qubit is realized by photon 2, and the ground state of each photon corresponds one-to-one with the output waveguide mode of the photon in the chip. In the quantum circuit, the order of action of the quantum gates is... Figure 1 The order of the components in the chip shown is consistent. The correspondence between quantum gates and the optical components of the quantum chip is as follows: Figure 4 As shown in (c) in the figure.

[0024] like Figure 1As shown, the optical quantum computing chip for quantum machine learning of the present invention includes the design and connection of various optical quantum devices. The chip is fabricated based on mature CMOS technology, using an on-chip laser as the pump source, and generating photon pairs using an on-chip spontaneous four-wave mixing single-photon source. MZI and phase shifters are used to control the transformation of the optical quantum states, and a single-photon detector is used to detect the output single photons. The phase of the phase shifter is an adjustable parameter controlled by a current source. To run the quantum machine learning task, after the output of the optical quantum chip is detected, a suitable loss function is designed according to the specific application (the loss function is related to the phase shift parameter of the optical quantum chip). The detection results are processed using a classical computer, and the chip parameters are optimized through gradient or gradient-free optimization methods to reduce the loss function. After multiple rounds of iterative training, the quantum chip outputs the solution to the target task.

[0025] When the chip is working, two pump beams at different frequencies and Coupled to the chip, the pump power is distributed to d groups of photon pair sources by a cascaded MZI binary tree array. The photon pair sources generate frequencies of [frequency value missing] through a spontaneous four-wave mixing effect. Identical photon pairs are separated into two paths using quantum interference. Since the single-photon generation rate of a spontaneous four-wave mixing single-photon source is related to the pump power (in the case of weak pumping, the photon rate is proportional to the square of the pump power), a d-dimensional two-photon entangled state with adjustable coefficients can be generated by adjusting the binary tree array and the phase shifter at the output port of the photon pair source. satisfy ,in Representing photon 1 and photon 2 in the... A pair of photons is generated. To facilitate the correspondence between the optical circuitry and the quantum circuitry of the quantum chip, photon 1 can be routed to the first n waveguides and photon 2 to the last n waveguides via cross-waveguides, thereby converting the quantum state into path encoding and rewriting the quantum state into binary form. in, for The decimal representation. At this point, each photon encodes one n-qubit state, and each quantum ground state corresponds to one waveguide mode, such as... Figure 1 and Figure 4 As shown in (a) above. Next, the output port of the photon pair source is connected to the controlled unitary module in a specific order, and the quantum state evolves before entering the controlled unitary module. In the context of quantum circuitry, this quantum state can be achieved by... Figure 4 (a) Transformation acts on Achieve, and The specific structure is as follows Figure 4 As shown in (b) of the diagram. The controlled unitary module controls qubit-0 ( ) and qubit-n ( ) Performs a 2D unitary operation with multi-qubit control, such as Figure 4 As shown in (a), the quantum state evolves into in and All represent the general 2D unitary transform, implemented by an MZI with two additional phase shifters, such as Figure 4 As shown in (c) of the reference. Figure 4 (b) and Figure 4 In (a), because able to and Arbitrary quantum states are generated on the ground (Physical Review A, 2001, 64(1): 014303), and each quantum ground state controls the generation of arbitrary quantum states ... and Execute arbitrary quantum gates, therefore it is possible to execute Previously, one qubit was extracted from these 2n+2 qubits. Door. Therefore, angular encoding can be used to encode classic data into... During the rotation of the door, a total of 2n+2 classic data can be encoded, thus realizing the first encoding layer. Furthermore, because... The controlled unitary module performs entanglement operations on all qubits, thus serving as the first globally entangled layer. The training parameters of this layer are related to the phase shifter phases of the cascaded MZI binary tree array, the photon pair source, and the controlled unitary module. Let the data vector to be encoded be... , The parameter vectors of the controlled unitary (CU) module are respectively and Then the quantum state can be described as Next, photon 1 and photon 2 each pass through a complete d-dimensional linear network. and (remember for The parameters are ), and then through multi-qubit control of the Z-gate ( Global entanglement is achieved, and finally, the network passes through a d-dimensional linear network. and (remember for The parameters are Since the coding layer does not involve entanglement between photons, it can be modified... The parameters are used to separate the second coding layer. Meanwhile, Together, they achieved the second entanglement layer, such as Figure 4 As shown in (a) above. At this point, the quantum state can be written as... according to and It can be seen that the optical quantum chip has successfully implemented two coding layers and two parametric, trainable entangled layers, and therefore can be used for quantum machine learning tasks.

[0026] In addition, the chip is finally connected to two state tomography modules. and It can perform measurement basis transformation or quantum state tomography on two photon states respectively, further meeting the needs of different tasks.

[0027] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A photonic quantum computing chip structure capable of performing variable quantum tasks such as quantum neural networks, characterized in that: It includes one cascaded MZI binary tree array, d groups of photon pair sources, d groups of controlled unitary gates, one multi-qubit controlled Z-gate, four d-dimensional tunable linear networks, and two d-dimensional state tomography linear networks, among which, The cascaded MZI binary tree array is connected to the photon pair source, and the cascaded MZI binary tree array adjusts the pump power of the photon pair source to control the photon generation rate of each photon pair source. Each set of photon source sources can generate identical photon pairs that are anti-focused and input them into the controlled unitary gate group in a specific order; The controlled unitary gate group performs a unitary transformation operation controlled by multiple qubits; The multi-qubit control Z-gate enables photon-to-photon interaction to achieve entanglement; The d-dimensional tunable linear network transforms the quantum state of a single photon. The d-dimensional tunable linear network includes linear network one, linear network two, linear network three, and linear network four. The photon pair includes photon one and photon two. Linear networks one and three transform the state of photon one, and linear networks two and four transform the state of photon two. Linear networks one and two serve as intermediate layers connecting the controlled unitary gate group and the multi-qubit control Z-gate. Linear networks three and four are connected after the multi-qubit control Z-gate. The d-dimensional state tomography linear network transforms or performs quantum state tomography on the measurement basis of the quantum states of photon one and photon two.

2. The optical quantum computing chip structure for quantum neural networks according to claim 1, characterized in that: The waveguide path of each photon in the photon pair is d-dimensional, satisfying... , in It is an integer greater than or equal to 1, and each waveguide corresponds to 1 quantum ground state.

3. The optical quantum computing chip structure for quantum neural networks according to claim 1, characterized in that: The cascaded MZI binary tree array, combined with a pump photon pair source, generates entangled states with tunable coefficients: The coefficients satisfy .

4. The optical quantum computing chip structure for quantum neural networks according to claim 3, characterized in that: The output port of the pumped photon pair source is connected to a controlled unitary gate group in a specific order, such that the input quantum state is: 。 5. The optical quantum computing chip structure for quantum neural networks according to claim 4, characterized in that: The controlled unitary group comprises d CU gate modules. Each CU gate module performs a multi-qubit controlled unitary gate operation on the quantum states represented by two adjacent waveguides, such that the quantum states are: 。 6. The optical quantum computing chip structure for quantum neural networks according to claim 5, characterized in that: The cascaded MZI binary tree array and the controlled unitary group together constitute the first coding layer and the first parametric entanglement layer; As a globally entangled state, by changing the phase shifter parameters, one qubit can be extracted from qubit 1 and one from qubit n to 2n-1. The gate encodes 2n+2 classic data.

7. The optical quantum computing chip structure for quantum neural networks according to claim 1, characterized in that: The complete linear networks one and two perform arbitrary unitary transformations on photon one and photon two respectively, extracting one unitary transformation from qubit 1 and qubits n to 2n-1. The gate serves as the second coding layer.

8. The optical quantum computing chip structure for quantum neural networks according to claim 1, characterized in that: The multi-qubit control Z-gate comprises 2d-1 MZIs with a splitting ratio of 1:2, and the input and output of the d-th MZI are connected to the d-1 path of photon 1 and the 0 path of photon 2, respectively. When both photons enter, and the two photons are then selected to output from two different ports, the quantum interference introduces a phase into the quantum state. .

9. The optical quantum computing chip structure for quantum neural networks according to claim 1, characterized in that: The multi-qubit control Z-gate has adjustable linear networks before and after it, which together realize the second parametric entanglement layer.

Images