Superconducting quantum bit control method and system
By introducing a dynamic noise perception and adaptive pulse optimization mechanism, and utilizing a hybrid neural network model and quantum state tomography, the problems of noise environment adaptability and real-time feedback in superconducting quantum bit manipulation were solved, achieving more efficient and stable manipulation results.
Patent Information
- Application Number
- CN202510994254.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-18
- Publication Date
- 2025-11-28
AI Technical Summary
Existing microwave pulse manipulation methods cannot adapt to dynamic noise environments and lack real-time feedback mechanisms, resulting in low stability and efficiency of superconducting quantum bit manipulation and sensitivity to non-ideal pulse responses.
A dynamic noise perception and adaptive impulse optimization mechanism is introduced. Noise environment data is collected and processed in real time through a hybrid neural network model to generate optimal impulse parameters, and closed-loop optimization is performed using quantum state tomography.
It significantly improves the stability and efficiency of superconducting quantum bit manipulation, breaks through the limitations of traditional static pulse shaping methods, and has good engineering feasibility and practical value.
Smart Images
Figure CN121031807A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of quantum computing technology, and more specifically to methods and systems for manipulating superconducting qubits. Background Technology
[0002] A superconducting quantum bit (qubit) is a qubit built using superconducting circuits and is the fundamental information unit for quantum computing. Unlike classical bits in traditional computers, which can only exist in a 0 or 1 state, superconducting qubits can exist in a superposition of 0 and 1 simultaneously. This quantum superposition property gives them an exponential computational advantage when dealing with certain specific problems. From a physical perspective, superconducting qubits utilize the special quantum properties exhibited by superconducting materials at extremely low temperatures. Their core principle is based on the Josephson effect of superconductors, where superconducting quantum pairs can tunnel through an insulating layer to form a superconducting current when two superconductors are separated by an extremely thin insulating layer. Superconducting qubits typically employ a Josephson junction structure, which consists of two superconductors and an insulator sandwiched in between. When the qubit is in its ground state, the current in the Josephson junction is zero; when the qubit is in an excited state, the current is not zero. By precisely controlling this superconducting current, superposition and entanglement states of the qubit can be achieved, thus enabling quantum computing.
[0003] Manipulation of superconducting qubits is a crucial step in realizing quantum computing. Microwave pulse manipulation is currently the most commonly used method for manipulating superconducting qubits. By applying microwave pulses of specific frequency and intensity, the energy levels of the superconducting qubit can be changed, thereby achieving quantum state transitions. Specifically, the transition frequency between the ground state and excited state of a qubit is fixed. When the applied microwave frequency matches this transition frequency, it induces a transition from the ground state to the excited state, and vice versa. This manipulation method is similar to tuning in classical radio, where a specific state is excited by adjusting the frequency. In two-qubit systems, microwave pulses can be used to implement quantum gate operations, such as CNOT gates. By precisely controlling the microwave field strength and phase between two qubits, entanglement and interaction between them can be achieved.
[0004] However, existing microwave pulse manipulation typically uses Gaussian pulses or Blackman pulses of fixed shape for single-qubit gate operations. These methods often suffer from the following problems:
[0005] Static pulse design cannot adapt to dynamic noise environments: In actual operation, thermal noise, electromagnetic interference and crosstalk around the quantum chip will cause the quantum bit frequency to drift, affecting the gate operation accuracy.
[0006] Lack of real-time feedback mechanism: traditional pulse shaping relies on offline calibration and cannot dynamically adjust according to the current environmental state.
[0007] Sensitive to non-ideal pulse response: due to transmission line loss and mismatch, the actual pulse shape reaching the quantum bit may deviate from the designed value.
[0008] Therefore, how to improve the stability and efficiency of superconducting quantum bit manipulation and break through the limitations of traditional static pulse shaping method is a problem that technicians in the field need to solve. SUMMARY
[0009] Therefore, the present application provides a superconducting quantum bit manipulation method and system, which significantly improves the stability and efficiency of superconducting quantum bit manipulation by introducing a dynamic noise perception and adaptive pulse optimization mechanism.
[0010] In order to achieve the above purpose, the present application adopts the following technical solutions:
[0011] Firstly, the present application discloses a superconducting quantum bit manipulation method, comprising the following steps:
[0012] Collecting noise environment data of a superconducting quantum chip;
[0013] Inputting the collected noise environment data into a trained hybrid neural network model to predict the optimal pulse parameters of the current micro-pulse for manipulating the superconducting quantum bit;
[0014] Generating a microwave pulse based on the current micro-pulse optimal pulse parameters and sending it to the quantum bit control port through an IQ modulator for quantum gate operation;
[0015] Real-time evaluation of quantum gate operation results using quantum state tomography method, and optimization of the hybrid neural network model based on the quantum gate operation results;
[0016] Using the optimized hybrid neural network model to predict new noise environment data to form a closed-loop control cycle.
[0017] Further, in the step of collecting noise environment data of a superconducting quantum chip, the noise environment data includes electromagnetic interference spectrum data around the quantum chip, thermal noise time series data and environmental scalar data, and the environmental scalar data includes temperature data and magnetic field data.
[0018] Further, the collected noise environment data is input into the trained hybrid neural network model, which includes an input layer, a feature processing layer, a feature fusion layer, and a multi-task output layer connected in sequence; wherein the feature processing layer includes a CNN convolution module for electromagnetic interference spectrum data, an LSTM module for processing thermal noise time series data, and a fully connected layer module for processing environmental scalar data.
[0019] Further, the feature fusion layer in the hybrid neural network model performs the following operations:
[0020] The spatial attention module is used to process the spectral features output by the CNN convolution module to obtain weighted spatial features;
[0021] The time attention module is used to process the noise time series features output by the LSTM module to obtain weighted time series features;
[0022] The environmental feature projection module is used to sequentially normalize, expand features, and nonlinearly transform the environmental scalar features output by the fully connected layer module to obtain high-order environmental features;
[0023] The spatio-temporal cross-attention fusion module is used to fuse the weighted spatial features, weighted time series features, and high-order environmental features to obtain spatio-temporal fusion features.
[0024] Further, the multi-task output layer in the hybrid neural network model performs the following operations:
[0025] The spatio-temporal fusion features output by the spatio-temporal cross-attention fusion module are received, and corresponding pulse parameters are output through different task branches, including amplitude modulation parameters, phase compensation parameters, and pulse rise / fall time parameters.
[0026] Further, the quantum state tomography method is used to evaluate the quantum gate operation results in real time, and the hybrid neural network model is optimized based on the quantum gate operation results, including the following steps:
[0027] After the quantum gate operation is completed, measurements under multiple different measurement bases are performed on the qubits, including X base, Y base, and Z base;
[0028] The measurement results are counted multiple times, and the actual quantum state density matrix is reconstructed using the maximum likelihood estimation method;
[0029] The reconstructed quantum state density matrix is compared with the target ideal state density matrix, the fidelity of the quantum gate operation is calculated, and is used as error information;
[0030] A composite loss function is constructed based on the error information and the difference between the predicted pulse parameters and the real pulse parameters, and the hybrid neural network model is optimized using the composite loss function.
[0031] Further, in the above method, the trained hybrid neural network model is trained using a transfer strategy.
[0032] Secondly, the application also discloses a superconducting quantum bit control system, comprising,
[0033] A noise acquisition subsystem is configured to measure and acquire electromagnetic interference spectrum data, thermal noise time series data and environmental scalar data around the quantum chip, wherein the environmental scalar data includes temperature data and magnetic field data.
[0034] A hybrid neural network model is configured to input the acquired noise environment data into the trained hybrid neural network model to predict and output the optimal pulse parameters of the current micro-pulse for controlling the superconducting quantum bit.
[0035] A quantum gate operation subsystem is configured to generate a microwave pulse based on the current optimal pulse parameters of the micro-pulse, and send the microwave pulse into a quantum bit control port through an IQ modulator to perform quantum gate operation.
[0036] A hybrid neural network model optimization subsystem is configured to evaluate the quantum gate operation result in real time using a quantum state tomography method, and optimize the hybrid neural network model based on the quantum gate operation result.
[0037] A closed-loop optimization subsystem is configured to predict new noise environment data using the optimized hybrid neural network model to form a closed-loop control cycle.
[0038] Preferably, the hybrid neural network model comprises an input layer, a feature processing layer, a feature fusion layer and a multi-task output layer connected in sequence; wherein the feature processing layer comprises a CNN convolution module for processing electromagnetic interference spectrum data, an LSTM module for processing thermal noise time series data, and a fully connected layer module for processing environmental scalar data.
[0039] Preferably, the feature fusion layer performs the following operations:
[0040] The spatial attention module is used to process the spectral features output by the CNN convolution module to obtain weighted spatial features;
[0041] The time attention module is used to process the noise time series features output by the LSTM module to obtain weighted time series features;
[0042] The environmental feature projection module is used to sequentially normalize, expand and nonlinearly transform the environmental scalar features output by the fully connected layer module to obtain high-order environmental features.
[0043] The spatial-temporal fusion module is used for fusing the weighted spatial features, the weighted time sequence features and the high-order environment features to obtain a spatial-temporal fusion feature.
[0044] Compared with the prior art, the superconducting quantum bit control method and system provided by the present application has the following beneficial effects:
[0045] The present application discloses a superconducting quantum bit control method and system based on adaptive pulse shaping with dynamic noise suppression (APS-DNS), which significantly improves the stability and efficiency of superconducting quantum bit control by introducing a dynamic noise perception and adaptive pulse optimization mechanism, breaking through the limitations of traditional static pulse shaping methods, and having good engineering implementation and practical value.
[0046] On the other hand, in order to improve the prediction accuracy of pulse parameters, the present application uses a hybrid neural network model to predict the noise environment data of the quantum chip to improve the accuracy of the predicted pulse parameters.
[0047] In addition, the present application uses the superconducting quantum bit control result and the output result of the hybrid neural network model as the loss function of the neural network model, and realizes the optimization feedback of the hybrid neural network model based on the loss function. BRIEF DESCRIPTION OF DRAWINGS
[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of the provided drawings.
[0049] Figure 1 The present application provides a superconducting quantum bit control method overall flowchart.
[0050] Figure 2 The present application provides a hybrid neural network model framework diagram. DETAILED DESCRIPTION
[0051] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.
[0052] Embodiment 1
[0053] Embodiment 1 of the present application discloses a method for operating a superconducting quantum bit, which is implemented based on adaptive pulse shaping with dynamic noise suppression (APS-DNS), as shown in the operation method Figure 1 The operation method comprises the following steps:
[0054] Collecting noise environment data of a superconducting quantum chip;
[0055] Inputting the collected noise environment data into a trained hybrid neural network model to predict and output optimal pulse parameters of a micro pulse for operating the superconducting quantum bit;
[0056] Generating a microwave pulse based on the optimal pulse parameters of the micro pulse and sending the microwave pulse into a quantum bit control port through an IQ modulator to perform quantum gate operation;
[0057] Real-time evaluating quantum gate operation results by using a quantum state tomography method, and optimizing the hybrid neural network model based on the quantum gate operation results;
[0058] Using the optimized hybrid neural network model to predict new noise environment data to form a closed-loop control cycle.
[0059] The operation method of the present application will be explained in detail.
[0060] In a specific embodiment, collecting noise environment data of a superconducting quantum chip specifically comprises collecting electromagnetic interference spectrum data, thermal noise time sequence data and environmental scalar data around the quantum chip, wherein the environmental scalar data includes temperature data and magnetic field data. It should be noted that in addition to electromagnetic interference, thermal noise and other noises, temperature, magnetic field and other environmental scalar data also have a great impact on quantum operation. Therefore, the data collection step also includes environmental scalar data.
[0061] To realize pulse control, corresponding pulse parameters need to be obtained based on the noise environment data collected in the above steps, and in the present application, this process is realized by a designed hybrid neural network model. The hybrid neural network model inputs noise environment data, and processes spatial features, time sequence features and scalar data through different branches, and obtains a fused feature vector. Figure 2 The hybrid neural network model in the present application comprises an input layer, a feature processing layer, a feature fusion layer and a multi-task output layer connected in sequence; wherein the feature processing layer comprises a CNN convolution module for electromagnetic interference spectrum data, an LSTM module for processing thermal noise time sequence data, and a fully connected layer module for processing environmental scalar data. Specifically, the feature fusion layer performs the following operations:
[0062] The spatial attention module is used to process the spectral features output by the CNN convolution module to obtain weighted spatial features; the time attention module is used to process the noise time sequence features output by the LSTM module to obtain weighted time sequence features; the environmental feature projection module is used to sequentially normalize, expand and nonlinearly transform the environmental scalar features output by the fully connected layer module to obtain high-order environmental features; finally, the spatio-temporal cross-attention fusion module is used to fuse the weighted spatial features, the weighted time sequence features and the high-order environmental features to obtain spatio-temporal fusion features.
[0063] According to the obtained spatio-temporal fusion features, the multi-task output layer performs the following operations:
[0064] The spatio-temporal fusion features output by the spatio-temporal cross-attention fusion module are received, and corresponding pulse parameters are output through different task branches, the pulse parameters including amplitude modulation parameters, phase compensation parameters and pulse rise / fall time parameters.
[0065] In one specific embodiment, the noise environment data input by the input layer includes electromagnetic interference spectrum data, thermal noise time sequence data and environmental scalar data, and corresponding feature vectors are processed through different branches, wherein the 1D-CNN spectrum processing branch extracts spectral features of the electromagnetic interference spectrum data through multiple convolution layers, such as the first convolution layer for detecting local frequency correlation, such as harmonic components, and the second convolution layer for capturing wideband spectral features, finally obtaining a 32-dimensional spatial spectral feature F s ; the long short-term memory network LSTM branch is used to process the time sequence noise data, and also obtains a 32-dimensional noise time sequence feature F t ; the fully connected layer branch is used to process the magnetic field and temperature environmental scalar data, and finally obtains the environmental features.
[0066] The spatial attention module calculates the spatial spectral features F simportance weights of different frequency components, and finally obtain the weighted spatial feature A s The time attention module calculates the weight of the periodic interference feature in the noise time sequence feature through the attention mechanism, and finally obtains the weighted time sequence feature A t The environment feature projection module sequentially performs normalization, feature expansion and nonlinear transformation on the environment scalar feature output by the full connection layer module to obtain a 16-dimensional high-order environment feature F env The process is specifically represented as follows:
[0067] F env = ReLU(W2·ReLU(W1·E+b1)+b2)
[0068] Wherein, ReLU represents an activation function, W1 and W2 represent learnable weight matrices, b1 and b2 are bias vectors corresponding to network layers, wherein the dimension of b1 is 8, and the dimension of b2 is 16, and finally a 16-dimensional high-order feature F env is obtained.
[0069] The spatio-temporal cross-attention fusion module fuses the weighted spatial feature, the weighted time sequence feature and the high-order environment feature to obtain a spatio-temporal fusion feature, and the mathematical representation of the specific operation process is as follows:
[0070] F fusion = σ(W g F all )⊙A s +(1-σ(W g F all ))⊙A t +f env (F env )
[0071] Wherein, W g represents a learnable gating weight matrix, F all represents a feature set composed of spatial spectral features and noise time sequence features, σ represents a weight coefficient, F env represents an environment scalar feature, and f env () represents a projection transformation function.
[0072] Finally, the multi-task output layer of the hybrid neural network model receives the spatio-temporal fusion feature output by the spatio-temporal cross-attention fusion module, and outputs corresponding pulse parameters through different task branches, including amplitude modulation parameters, phase compensation parameters, and pulse rise / fall time parameters.
[0073] The pulse parameters are used as the current best pulse parameters to generate the current microwave pulse. In one specific embodiment, the pulse parameters can be input into an arbitrary waveform generator (AWG) to generate the microwave pulse, and the microwave pulse is sent to the quantum bit control port of the quantum chip through an IQ modulator for quantum gate operation. The IQ modulator is an electronic device used for signal modulation, and its core function is to decompose the baseband signal into two components, in-phase (I) and quadrature (Q), and modulate the carrier signal using the two components, thereby achieving efficient and stable signal transmission.
[0074] After the quantum gate operation is completed, an online error feedback mechanism is enabled synchronously. In the mechanism, the quantum gate operation result is evaluated in real time using quantum state tomography (QST), and the hybrid neural network model is optimized based on the quantum gate operation result. The optimization includes the following steps:
[0075] After the quantum gate operation is completed, measurements are performed on the quantum bits in multiple different measurement bases, including X base, Y base, and Z base.
[0076] The measurement results are counted, and the actual quantum state density matrix is reconstructed using the maximum likelihood estimation method.
[0077] The reconstructed quantum state density matrix is compared with the target ideal state density matrix, the fidelity of the quantum gate operation is calculated, and the fidelity is used as error information.
[0078] Based on the error information and the difference between the predicted pulse parameters and the real pulse parameters, a composite loss function is constructed, and the hybrid neural network model is optimized using the composite loss function. The specific loss function of the hybrid neural network model is as follows:
[0079] L = a1L 保真度误差信息 + a2L 预测差值 .
[0080] In the formula, a1 and a2 represent the corresponding weight coefficients, L 保真度误差信息 represents the loss function composed of error information, and L 预测差值 represents the loss function composed of the difference between the predicted pulse parameters and the real pulse parameters. The two loss functions together constitute the final loss function to optimize the training process of the hybrid neural network model.
[0081] Then, the optimized hybrid neural network model is used to predict new noise environment data, forming a closed-loop control cycle, thereby completing the superconducting quantum bit control method based on dynamic noise suppression and adaptive pulse shaping technology.
[0082] In the present application, quantum state tomography (QST) is used, and the steps of the QST technique are further described below:
[0083] Select measurement bases: Select three mutually orthogonal measurement bases, such as Z base (|0>, |1>), X base and Y base
[0084] Perform measurements and record results: For each measurement base, repeat the measurement multiple times, and record the results of each measurement (such as the number of times |0> is measured and the number of times |1> is measured under the Z base).
[0085] Calculate probability distribution: Calculate the measurement probability under each base according to the measurement results. For example, under the Z base, p0 represents the probability of measuring |0>, and p1 represents the probability of measuring |1>.
[0086] Reconstruct the density matrix: Using the probability values obtained above, combined with appropriate mathematical tools (such as MLE), the density matrix form of the quantum state is reconstructed.
[0087] In the quantum state tomography method, fidelity (Fidelity) is used to measure the similarity between an actual quantum state and an ideal target state. The fidelity calculation formula is as follows:
[0088]
[0089] Where F represents the calculated fidelity value, ranging from 0 to 1, 1 indicating that the two quantum states are completely identical, and 0 meaning that they are completely different. ρ ideal represents the density matrix corresponding to the ideal quantum state, and ρ actual represents the density matrix of the reconstructed actual quantum state.
[0090] Further, in order to optimize the technical solutions, in the above method, the trained hybrid neural network model is trained using a transfer strategy.
[0091] The present application
[0092] Embodiment 2
[0093] Embodiment 2 based on the control method of embodiment 1 discloses a superconducting quantum bit control system, comprising,
[0094] A noise acquisition subsystem is used to measure and acquire electromagnetic interference spectrum data, thermal noise time sequence data, and environmental scalar data around the quantum chip, wherein the environmental scalar data includes temperature data and magnetic field data.
[0095] The hybrid neural network model is used for inputting the collected noise environment data into the trained hybrid neural network model to predict the optimal pulse parameters of the current micro-pulse for operating the superconducting quantum bit.
[0096] The quantum gate operation subsystem is used for generating a microwave pulse based on the current micro-pulse optimal pulse parameters and sending the microwave pulse into a quantum bit control port through an IQ modulator for quantum gate operation.
[0097] The hybrid neural network model optimization subsystem is used for real-time evaluation of the quantum gate operation result by using a quantum state tomography method and optimization of the hybrid neural network model based on the quantum gate operation result.
[0098] The closed-loop optimization subsystem is used for prediction of new noise environment data by using the optimized hybrid neural network model to form a closed-loop control cycle.
[0099] According to the method in Reference Example 1, the hybrid neural network model in the superconducting quantum bit operating system comprises an input layer, a feature processing layer, a feature fusion layer and a multi-task output layer connected in sequence; wherein the feature processing layer comprises a CNN convolution module for electromagnetic interference spectrum data, an LSTM module for processing thermal noise time series data, and a fully connected layer module for processing environmental scalar data.
[0100] In order to improve the pulse parameters, the feature fusion layer in the hybrid neural network model performs the following operations:
[0101] The spatial attention module is used to process the spectral features output by the CNN convolution module to obtain weighted spatial features;
[0102] The time attention module is used to process the noise time series features output by the LSTM module to obtain weighted time series features;
[0103] The environmental feature projection module is used to sequentially normalize, expand and nonlinearly transform the environmental scalar features output by the fully connected layer module to obtain high-order environmental features;
[0104] The space-time cross-attention fusion module is used to fuse the weighted spatial features, the weighted time series features and the high-order environmental features to obtain space-time fusion features.
[0105] The disclosed operation method and system of adaptive pulse shaping with dynamic noise suppression (APS-DNS) can be widely applied in scenarios of improving the operation fidelity of single / double quantum bit gates in superconducting quantum computers, and can enhance the robustness of quantum circuits in complex electromagnetic environments, and provide scalable operation solutions for future large-scale integrated quantum chips.
[0106] The various embodiments are described in the specification with progressive progression, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be mutually referred to. For the device disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.
[0107] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for manipulating superconducting qubits, characterized in that, Includes the following steps: Collect noise environment data for superconducting quantum chips; The collected noise environment data is input into the trained hybrid neural network model to predict and output the optimal pulse parameters for the micropulse that is currently manipulating the superconducting quantum bits. Microwave pulses are generated based on the current optimal pulse parameters and then sent to the quantum bit control port via an IQ modulator for quantum gate operation. The quantum state tomography method is used to evaluate the results of quantum gate operations in real time, and the hybrid neural network model is optimized based on the results of quantum gate operations. The optimized hybrid neural network model is used to predict new noise environment data, forming a closed-loop control.
2. The method for manipulating superconducting qubits according to claim 1, characterized in that, In the step of collecting noise environment data of superconducting quantum chips, the noise environment data includes electromagnetic interference spectrum data, thermal noise time series data, and environmental scalar data around the quantum chip. The environmental scalar data includes temperature data and magnetic field data.
3. The method for controlling a superconducting quantum bit according to claim 1, characterized in that, In the step of inputting the collected noise environment data into the trained hybrid neural network model, the hybrid neural network model includes an input layer, a feature processing layer, a feature fusion layer, and a multi-task output layer connected in sequence; wherein the feature processing layer includes a CNN convolutional module for electromagnetic interference spectrum data, an LSTM module for processing thermal noise time series data, and a fully connected layer module for processing environmental scalar data.
4. The method for controlling a superconducting quantum bit according to claim 3, characterized in that, The feature fusion layer performs the following operations: The spatial attention module is used to process the spectral features output by the CNN convolutional module to obtain weighted spatial features; The time attention module is used to process the noisy temporal features output by the LSTM module to obtain weighted temporal features; The environmental scalar features output by the fully connected layer module are normalized, feature expanded, and nonlinearly transformed sequentially using the environmental feature projection module to obtain higher-order environmental features. The weighted spatial features, weighted temporal features, and higher-order environmental features are fused using a spatiotemporal cross-attention fusion module to obtain spatiotemporal fusion features.
5. The method for controlling a superconducting quantum bit according to claim 4, characterized in that, The multi-task output layer performs the following operations: The spatiotemporal fusion features output by the spatiotemporal cross-attention fusion module are received, and corresponding pulse parameters are output through different task branches. The pulse parameters include amplitude modulation parameters, phase compensation parameters, and pulse rise / fall time parameters.
6. The method for controlling a superconducting quantum bit according to claim 1, characterized in that, The quantum state tomography method is used to evaluate the results of quantum gate operations in real time, and the hybrid neural network model is optimized based on the results of quantum gate operations. The specific steps include: After the quantum gate operation is completed, measurements are performed on the qubit under multiple different measurement bases, including X-basis, Y-basis and Z-basis; The results of multiple measurements were statistically analyzed, and the actual quantum state density matrix was reconstructed using the maximum likelihood estimation method. The reconstructed quantum state density matrix is compared with the target ideal state density matrix to calculate the fidelity of quantum gate operations, which is then used as error information. Based on the error information and the difference between the predicted pulse parameters and the actual pulse parameters, a composite loss function is constructed, and the composite loss function is used to optimize the hybrid neural network model.
7. The method for controlling a superconducting quantum bit according to claim 1, characterized in that, The trained hybrid neural network model is trained using a transfer learning strategy.
8. A manipulation system for superconducting qubits, characterized in that, include, The noise acquisition subsystem is used to measure and acquire electromagnetic interference spectrum data, thermal noise time series data, and environmental scalar data around the quantum chip. The environmental scalar data includes temperature data and magnetic field data. A hybrid neural network model is used to input the collected noise environment data into the trained hybrid neural network model and predict the optimal pulse parameters for the micropulse that is currently manipulating the superconducting quantum bits. The quantum gate operation subsystem is used to generate microwave pulses based on the current optimal pulse parameters of the micropulse and send them to the quantum bit control port for quantum gate operation via an IQ modulator. A hybrid neural network model optimization subsystem is used to evaluate the results of quantum gate operations in real time using quantum state tomography and optimize the hybrid neural network model based on the results of quantum gate operations. The closed-loop optimization subsystem is used to predict new noise environment data using the optimized hybrid neural network model, forming a closed-loop cyclic control.
9. A manipulation system for superconducting qubits according to claim 8, characterized in that, The hybrid neural network model includes an input layer, a feature processing layer, a feature fusion layer, and a multi-task output layer connected in sequence; wherein the feature processing layer includes a CNN convolutional module for electromagnetic interference spectrum data, an LSTM module for processing thermal noise time series data, and a fully connected layer module for processing environmental scalar data.
10. A method for controlling superconducting qubits according to claim 9, characterized in that, The feature fusion layer performs the following operations: The spatial attention module is used to process the spectral features output by the CNN convolutional module to obtain weighted spatial features; The time attention module is used to process the noisy temporal features output by the LSTM module to obtain weighted temporal features; The environmental scalar features output by the fully connected layer module are normalized, feature expanded, and nonlinearly transformed sequentially using the environmental feature projection module to obtain higher-order environmental features. The weighted spatial features, weighted temporal features, and higher-order environmental features are fused using a spatiotemporal cross-attention fusion module to obtain spatiotemporal fusion features.