Optical computing error adjustment method, system, medium, and program product

Abstract

The application discloses an optical computing error adjustment method, system, medium and program product, and relates to the technical field of optical computing, and comprises the following steps: determining a unified correction deviation value corresponding to phase noise of an interferometer array; taking the minimum value of the square sum of the difference between the measured light intensity value and the ideal light intensity value of a photoelectric detector array as the target, and generating a target function based on the correction deviation direction variable; solving the target function to obtain the optimal value of the correction deviation direction variable of each phase shifter; and sending the optimal value of the correction deviation direction variable of each phase shifter and the unified correction deviation value to a control device, so that the control device adjusts each phase shifter simultaneously. In this way, through the determined unified correction deviation value and the optimal value of the correction deviation direction variable of each phase shifter, collective adjustment of all phase shifters can be realized, additional overhead caused by individual calibration can be avoided, the complexity of optical computing noise mitigation is reduced, the noise mitigation process is accelerated, and the output accuracy of the optical computing device is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical computing, in particular to an optical computing error adjustment method, system, medium and program product. BACKGROUND

[0002] With the semiconductor process approaching the physical limit, Moore's law gradually fails, and the traditional electrical computer faces many limitations: energy consumption increases exponentially with frequency, copper interconnection bandwidth bottleneck will cause data congestion, and actual computing power is difficult to meet the standard. Optical computing has become an important direction to support the continuous evolution of computing power, and the interferometer array is the core component of the linear operation of the optical computing device. However, the optical computing device usually needs to frequently calibrate each phase shifter in the interferometer array to lock the working point during operation, which increases the system delay and affects the output accuracy of the optical computing device as the array depth increases. SUMMARY

[0003] The present application provides an optical computing error adjustment method, system, medium and program product, which can realize collective adjustment of all phase shifters instead of individual calibration of each phase shifter, speed up the noise relief process, and improve the output accuracy of the optical computing device.

[0004] The present application provides an optical computing error adjustment method, comprising:

[0005] determining a unified correction bias value corresponding to the phase noise of the interferometer array in the optical computing device; the interferometer array comprises at least two phase shifters;

[0006] minimizing the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device as the target, and generating a target function based on the correction bias direction variable;

[0007] solving the target function to obtain the optimal value of the correction bias direction variable of each phase shifter;

[0008] sending the optimal value of the correction bias direction variable of each phase shifter and the unified correction bias value to the control device, so that the control device adjusts each phase shifter according to the optimal value of the correction bias direction variable of each phase shifter and the unified correction bias value.

[0009] The present application also provides an optical computing error adjustment system, comprising: an optical computing device, a control device and an external computing device;

[0010] The optical computing device comprises a light source, an interferometer array and a photodetector array; the interferometer array comprises at least two phase shifters;

[0011] The external computing device is configured to determine a unified correction bias value corresponding to the phase noise of the interferometer array, generate a target function based on a correction bias direction variable by minimizing the sum of squares of differences between the measured light intensity values and the ideal light intensity values of the photodetector array in the optical computing device, and solve the target function to obtain optimal values of the correction bias direction variable of each phase shifter.

[0012] The control device is configured to simultaneously adjust each phase shifter according to the optimal values of the correction bias direction variable of each phase shifter and the unified correction bias value.

[0013] The present application also provides a computer-readable storage medium having a computer program stored therein, wherein the computer program, when executed by a processor, implements the steps of any of the above optical computing error adjustment methods.

[0014] The present application also provides a computer program product comprising a computer program, which, when executed by a processor, implements the steps of any of the above optical computing error adjustment methods.

[0015] According to the present application, a unified correction bias value corresponding to the phase noise of the interferometer array in the optical computing device is first determined; the interferometer array comprises at least two phase shifters; then a target function based on a correction bias direction variable is generated by minimizing the sum of squares of differences between the measured light intensity values and the ideal light intensity values of the photodetector array in the optical computing device; thereafter, the target function is solved to obtain optimal values of the correction bias direction variable of each phase shifter; finally, the optimal values of the correction bias direction variable of each phase shifter and the unified correction bias value are sent to the control device, so that the control device simultaneously adjusts each phase shifter according to the optimal values of the correction bias direction variable of each phase shifter and the unified correction bias value. In this way, by using the determined unified correction bias value and the optimal values of the correction bias direction variable of each phase shifter, collective adjustment of all phase shifters can be realized, so that the measured light intensity values of the photodetector array approach the ideal light intensity values, effectively improving the output accuracy of the optical computing device, avoiding additional expenses caused by individual calibration of phase shifters, and greatly reducing the overall expenses during the operation of the entire interferometer array. At the same time, the collective adjustment mode does not require individual adjustment of each phase shifter, greatly simplifying the operation process of optical computing noise mitigation, reducing the complexity of optical computing noise mitigation, speeding up the noise mitigation process, reducing the time required for calibration, and can be widely applied in integrated optical networks, providing strong support for efficient and low-cost operation of integrated optical networks.

[0016] In addition, the present application also provides a corresponding optical calculation error adjustment system, a computer readable storage medium and a program product for the optical calculation error adjustment method, which have the same or corresponding technical features and effects as the optical calculation error adjustment method mentioned above. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0018] Figure 1 The flow chart of the optical calculation error adjustment method provided for the embodiments of the present application;

[0019] Figure 2 The structural schematic diagram of a single Mach-Zehnder interferometer provided for the embodiments of the present application;

[0020] Figure 3 The structural schematic diagram of a Mach-Zehnder interferometer array provided for the embodiments of the present application. DETAILED DESCRIPTION

[0021] As the semiconductor process approaches the physical limit, the tunneling effect makes the transistor miniaturization come to an end, and Moore's law has become a certainty. More seriously, the energy crisis of traditional electrical computers is increasingly highlighted, and the joule heat caused by resistance increases exponentially with the increase of frequency, and heat dissipation has become a shackles to restrict the development of computing power, which is difficult to continue in the context of sustainable development. At the same time, the bandwidth bottleneck of copper interconnection seriously restricts data flow, and the "data congestion" of traditional computers makes the actual computing power much lower than the theoretical peak value. The demand for computing power (especially parallel matrix operation) in the field of artificial intelligence model training, climate simulation, real-time big data processing, etc. is exploding at a speed far beyond the supply capacity of traditional computers, and the traditional architecture has been unable to support the computing power demand of the future digital society. Optical computing is considered to be able to solve the problems faced by electronic computing system to a certain extent. Optical computing refers to using the physical properties of light to perform a series of computing tasks, and its core breakthrough is to use light to replace electrons for information transmission and processing: light propagates at a speed of 300,000 kilometers per second in vacuum, breaking free from the constraints of resistance and capacitance, and theoretically realizing terahertz-level bandwidth and femtosecond-level delay, completely breaking through the data transmission bottleneck restricting computing power. More importantly, photon transmission almost does not produce joule heat, and the energy efficiency of optical computing is several orders of magnitude higher than that of electrical computing in certain computing tasks, which is expected to reduce the energy consumption of data centers by more than ten times, and solve the energy consumption problem from the root. Light also has a natural parallel advantage: different wavelengths of light can be transmitted independently in the same channel (wavelength division multiplexing), making the transmission capacity of a single optical fiber far exceed that of copper cable by tens of thousands of times. This feature has a revolutionary value in the field of artificial intelligence: light can directly complete matrix multiplication in the optical domain through physical processes such as interference and diffraction, and matrix operation is the core of neural networks.

[0022] In optical computing architecture, Mach Zehnder interference (MZI) array is the core component to realize linear operation, and its performance is extremely sensitive to noise. The existence of noise fundamentally destroys the accuracy, reliability and potential advantages of optical computing. Noise directly disturbs the interference process on which Mach Zehnder interferometer works. For example, environmental temperature fluctuations will cause slight changes in the refractive index of waveguide materials, causing phase drift that is difficult to track and compensate in real time; slight mechanical stress or vibration will cause random fluctuations in the optical path difference. These phase noises superimposed on the phase noise of the light source itself (limited line width of the laser) make the light field state that needs to be precisely constructive or destructive interference become blurred. The inherent non-uniformity of the manufacturing process (such as etching depth deviation) will also introduce static beam splitting ratio error. The direct consequence is the catastrophic decline of computing accuracy. When optical neural network performs large-scale matrix multiplication, every small phase or amplitude error will be amplified by multi-layer structure, resulting in a huge deviation of the final output result from the theoretical value, which seriously affects the accuracy of image recognition, signal processing and other applications. In addition, noise forces the system to bear a huge additional burden. In order to resist dynamic drift, complex online calibration must be performed frequently to relock the working point, which significantly increases the system delay and power consumption. Maintaining an acceptable signal-to-noise ratio even requires increasing the input optical power or increasing the error correction overhead, which weakens the inherent low energy consumption advantage of optical computing. Noise also limits the size of non-crosstalk units that can be used simultaneously, restricting the implementation of large-scale parallel computing. Essentially, noise erodes the precision interference foundation on which optical computing relies, and if it cannot be effectively suppressed, its theoretical speed and energy efficiency advantage will be difficult to translate into reliable performance in practical applications. Although the influence of noise on a single phase shifter is not large, as the depth of the Mach Zehnder interferometer array increases, the influence of noise gradually increases, which will affect the accuracy of the final output result. In order to solve the above problems, the present application provides an optical computing error adjustment method.

[0023] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0024] It should be noted that in the description of the present application, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or further includes elements inherent to such a process, method, article or device. The terms "first", "second" and the like in the present application are used to distinguish similar objects, not to describe a specific order or sequence.

[0025] In order to enable those skilled in the art to better understand the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments.

[0026] In combination with the specific application environment architecture or specific hardware architecture on which the light computing error adjustment method is executed, the specific application environment architecture or specific hardware architecture is described here.

[0027] The embodiments of the present application provide a light computing error adjustment method, and the method is described in detail in combination with the execution flow of the light computing error adjustment method. Figure 1 The flowchart of the light computing error adjustment method provided by the embodiments of the present application is shown in Figure 1 The method comprises the following steps.

[0028] S101, determining a uniform correction bias value corresponding to phase noise of an interferometer array in an optical computing device; the interferometer array comprises at least two phase shifters.

[0029] Before step S101 is executed, initialization setting can be performed first. The optical computing device is started, a test light signal is input through a control device, a simple vector (such as a full 1 vector) can be loaded through a spatial light modulator, and it is ensured that all phase shifters are in a preset state (phase , is set to a known default value) of the control device; an array of photodetectors is connected, the light intensity of the output port is measured, and the result is returned to the control device. The control device can control the optical computing device to emit a preset light signal, control the selection of the phase shifter and the corresponding phase setting. The execution subject of the light computing error adjustment method provided by the present application can be an external computing device, and the external computing device is responsible for the calculation task. The above-mentioned interferometer array can be a Mach-Zehnder interferometer array, or other interferometer arrays, which are not limited here.

[0030] It should be noted that the core idea of the present application is not to perfectly correct the error (i.e. noise) of each phase shifter, but to reduce the error range of the phase shifter, thereby reducing the overall error degree of the interferometer array. Since the training task of the optical neural network can tolerate a certain degree of error itself, the optical calculation error adjustment method provided by the present application can increase the noise robustness of the optical neural network based on the interferometer array.

[0031] In the present application, a kind of phase drift noise is mainly considered, which is caused by the length deviation during waveguide manufacturing, so that light will accumulate additional phase when propagating in the waveguide, which can be represented as a slight change in the phase And Due to the imperfection of the phase shifter during the manufacturing process and other factors, the phase cannot be accurately adjusted to Or There will be a certain background error.

[0032] When performing step S101, the phase noise of the interferometer array, the external computing device determines the corresponding uniform correction deviation value The uniform correction deviation value refers to a global compensation phase value set to alleviate the phase noise faced by all phase shifters in the interferometer array, and can also be understood as a uniform correction deviation size variable. The uniform correction deviation can be replaced by a global control instead of modifying each phase shifter.

[0033] S102, the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device is minimized, and a target function based on the correction deviation direction variable is generated.

[0034] In implementation, the present application compensates for the phase noise of the phase shifter in the optical computing device, and constructs a target function: minimizing the sum of the squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array as the core target, and taking the correction deviation direction variable as the optimization core variable. The correction deviation direction variable can be , which is related to the adjustment direction of the phase noise compensation; +1 represents the direction in which the actual phase deviates from the ideal value due to the phase noise of the current phase shifter, which needs to be adjusted by positive phase compensation; -1 represents the opposite case, which needs to be adjusted by reverse phase compensation. The present application converts the noise relief problem of the interferometer array into a global optimization task, and finally finds the optimal phase correction scheme that can make the measured light intensity accurately fit the ideal value by solving the minimum value of the target function through the mapping of the variable and the light intensity deviation.

[0035] S103, solve the target function to get the optimal value of the correction deviation direction variable of each phase shifter.

[0036] In the implementation, the present application determines the deviation adjustment direction of each phase shifter, such as phase compensation or subtraction, by solving the objective function.

[0037] S104, the optimal value of the correction deviation direction variable of each phase shifter and the unified correction deviation value are sent to the control device, so that the control device adjusts each phase shifter according to the optimal value of the correction deviation direction variable of each phase shifter and the unified correction deviation value.

[0038] It should be noted that the present application synchronously transmits the optimal value of the correction deviation direction variable of each phase shifter obtained in step S103 and the unified correction deviation value determined in step S101 to the control device. In other words, the present application provides two parameters, the optimal value of the correction deviation direction variable and the unified correction deviation value, for each phase shifter, and the control device adjusts each phase shifter in combination with these two parameters, that is, adds the product of the optimal value of the correction deviation direction variable and the unified correction deviation value to the phase of all phase shifters for collective adjustment, so that the phase state of the interferometer array is restored from the noisy state to the ideal state, the measured light intensity value of the photodetector array approaches the ideal light intensity value, and the accuracy of subsequent optical calculation is ensured.

[0039] In the above optical calculation error adjustment method provided by the embodiment of the present application, the unified correction deviation value and the optimal value of the correction deviation direction variable of each phase shifter can be used to realize the collective adjustment of all phase shifters, so that the measured light intensity value of the photodetector array approaches the ideal light intensity value, the output accuracy of the optical calculation device is effectively improved, and the additional cost caused by the separate calibration of the phase shifter is avoided, thereby greatly reducing the overall cost during the operation of the entire interferometer array. At the same time, the collective adjustment mode does not need to debug each phase shifter one by one, greatly simplifies the operation process of optical calculation noise relief, reduces the complexity of optical calculation noise relief, speeds up the process of noise relief, reduces the time required for calibration, and can be widely applied to integrated optical networks to provide strong support for efficient and low-cost operation of integrated optical networks.

[0040] In practical applications, the present application focuses on optical calculation based on interference, the core device of which is an interferometer array. The calculation directly realizes linear algebra operation at the physical layer by using the interference principle of light, and the core is to construct a unitary matrix transformation network through programmable phase modulation. The architecture is integrated on a photonic chip in a triangular z-grid or rectangular grid topology by multiple interferometer units. The input light field is transformed by the array and output , wherein is the target operation matrix. Figure 2 is the structure schematic diagram of a single Mach-Zehnder interferometer provided by the embodiment of the present application. As shown in Figure 2As shown, a single Mach-Zehnder interferometer consists of two 50:50 beam splitters and two parallel waveguide arms, with one arm integrating two phase shifters (introducing a controllable phase difference). and Its transmission characteristics are described by a unitary matrix:

[0041] The first phase modulation matrix is:

[0042] ;

[0043] The transmission matrix of the first beam splitter is:

[0044] ;

[0045] The second phase modulation matrix is:

[0046] ;

[0047] The transmission matrix of the second beam splitter is:

[0048] ;

[0049] The unitary transmission matrix of a single Mach-Zehnder interferometer is:

[0050] ;

[0051] By adjusting It can achieve any Unitary transformation.

[0052] To construct large-scale matrix operations, Mach-Zehnder interferometer elements are cascaded in a specific topology (such as a triangular mesh) to form a Mach-Zehnder interferometer array. Figure 3 This is a schematic diagram of the structure of a Mach-Zehnder interferometer array provided in an embodiment of the present invention. Figure 3 As shown, a single Mach-Zehnder interferometer is simplified to the portion shown in the dashed line.

[0053] for 3D input vector Its light field representation injection After entering the port, the signal undergoes a series of interferometric transformations within the array. The transmission matrix of the entire network is the product of the unitary matrices of each Mach-Zehnder interferometer element:

[0054] ;

[0055] in It is important to note that this network can implement arbitrary unitary transformations, because according to the Clems decomposition theorem, any... unitary matrix It can be decomposed into:

[0056] ;

[0057] here This is the diagonal phase compensation matrix. It is a specific Mach-Zehnder interferometer sequence. This can be understood as the coordinates of a Mach-Zehnder interferometer. Through precise configuration of all... This allows for the programming implementation of the target unitary matrix. Output light field This is the result of matrix multiplication.

[0058] In optical computation based on Mach-Zehnder interferometer arrays, the input vector The data loading process can be completed by injecting the optical signal into the input port via a spatial light modulator. speed( The light travels through the array (at the effective refractive index of the waveguide), undergoing continuous interference transformations, with a transmission time on the order of picoseconds (ps). Finally, the light intensity is measured at the output end via a photodetector array to obtain the result.

[0059] Furthermore, in a specific implementation, in the optical computing error adjustment method provided in the embodiments of the present invention, step S101, which determines the uniform correction deviation value corresponding to the phase noise of the interferometer array in the optical computing device, may specifically include: after the control device randomly selects a set number of phase shifters and adjusts the phase of the phase shifters, obtaining the light intensity change output by the photodetector array; obtaining the average noise amplitude based on the fitting curve corresponding to the light intensity change output by the photodetector array, and using the average noise amplitude as the uniform correction deviation value corresponding to the phase noise of the interferometer array in the optical computing device.

[0060] In practice, this invention can use a control device to randomly select a small number (e.g., at least three) of phase shifters for phase offset testing to calibrate their phase error. The specific method is as follows: Use the control device to fine-tune the phase of one phase shifter (e.g., increase the phase by a small amount). The method uses a photodetector array to measure changes in output light intensity. The average noise amplitude is estimated by fitting the change curve on a computing device and set as a uniform correction deviation value. This allows for the random selection and adjustment of phase shifters by the control device, combined with the fitted curve of the photodetector array's output light intensity change, to obtain the average noise amplitude. This eliminates the need for complex point-by-point detection across the entire array, making the operation simple and efficient. Using this as a uniform correction deviation value allows for the rapid capture of the overall characteristics of the array's phase noise, laying the foundation for subsequent accurate noise correction and mitigation of noise interference with the output. This helps improve the stability and output accuracy of the interferometer array while reducing resource consumption and time costs in the noise detection process.

[0061] Furthermore, in a specific implementation, in the above-mentioned optical calculation error adjustment method provided in the embodiments of the present invention, step S102, which generates an objective function based on the correction deviation direction variable, may specifically include: after obtaining the measured light intensity value and the ideal light intensity value of the photodetector array, generating the objective function by calculating the sum of squares of the differences between the measured light intensity value and the ideal light intensity value of the photodetector array.

[0062] In practice, a single interferometer contains two phase shifters, with the first phase shifter having a phase of... And the noise is The phase of the second phase shifter is And the noise is The transmission matrix of the noisy interferometer is as shown above.

[0063] Assume the input of the interferometer is Therefore, its output under the first-order approximation should be:

[0064] ;

[0065] Note that the matrix above contains... The first term enclosed in parentheses represents the output of the interferometer under ideal conditions. If... The coefficient matrix above is denoted as The output of a noisy interferometer under ideal conditions is denoted as Therefore, under the first-order approximation, the output of the noisy interferometer should be:

[0066] ;

[0067] The above equation expresses the output error as a linear function of the noise in the first order approximation, makes the correction process independent of the specific phase value, means that the output in the presence of noise is the output in the ideal case plus some linear noise terms, at the same time, these linear noise can be mitigated by applying a small phase offset on the phase shifter. These linear noise terms are independent of the specific value of the phase at this time, which means that after a one-time calibration, no matter what value the phase is set to, the calibration result can continue to be used, reducing the subsequent calculation overhead.

[0068] The above analysis only considers the transformation of an ideal light input into a single interferometer. But in actual conditions, it is often the light with noise after the interferometer again through the next interferometer, at this time, the transformation of the noisy input in the noisy interferometer needs to be considered. Assuming that the noisy input of a certain noisy interferometer is , wherein and represent the deviation of this light from the noise-free case, which can be expressed as a linear combination of the noise amount of the phase shifter through which the light passes in the first order approximation. After passing through the noisy interferometer, its output in the first order approximation can be expressed as:

[0069] ;

[0070] Similar to the above analysis, in the above expression, the first term is the output in the ideal case. At the same time, it is noted that and represent the noise terms of the noisy phase shifter through which the light passes, at this time, the output can be rewritten as:

[0071] ;

[0072] , wherein represents the noise of the phase shifter passed through, is the corresponding noise coefficient vector, represents the phase shifter.

[0073] In this case, this noise can be mitigated by introducing a small offset in each phase shifter, in this case, after the same analysis, it can be obtained that the output after correction is:

[0074] ;

[0075] , wherein is the correction deviation direction variable, which can only take , represents the correction direction, and is the unified correction deviation value, which represents the size of the correction error. Because it is assumed that the noise in the interferometer array is The size is close and the direction is random, so most of the errors can be corrected in this way. Among them The value can be roughly determined by pre-calibrating several phase shifters, and The specific value needs to be determined by judging the noisy output . This can reduce the parameter complexity, accurately match the deviation direction of each phase shifter, and without complex calculation of the independent compensation amount of each phase shifter, only by judging the noisy output Determine , can efficiently approximate to offset most of the noise, so that the output Close to the ideal value , while ensuring the noise mitigation effect, greatly simplifies the phase calibration process of the interferometer array in the optical computing device.

[0076] If there are interferometers in total, there are phase shifters, by comparing the difference between the noisy output after correction and the ideal output, we can get:

[0077] ;

[0078] Here, the constant term is ignored. In this case, the problem of mitigating the noise of the phase shifter in the interferometer array becomes finding a set of Distribution, so that the difference between And The difference is determined by comparing the modulus sum of the difference between the output complex amplitude of all ports and the ideal complex amplitude. The ideal output complex amplitude can be obtained by calculating the matrix multiplication between the ideal interferometer array transmission matrix and the ideal input, but the problem is that the noisy output complex amplitude of the interferometer needs to be measured. This complex amplitude measurement is difficult to perform in practical applications, and in experiments, the commonly used method is to measure the output light intensity, and then sum the squares of the light intensity differences between each output port and the ideal value.

[0079] First, consider the output port, then its noisy output complex amplitude can be represented as:

[0080] ;

[0081] Among them, is the component of the noise coefficient vector . Then the light intensity of the port output can be represented as:

[0082] ;

[0083] in, Let represent the real part. Then the sum of squares of the light intensity differences between all ports and the ideal output can be written as:

[0084] ;

[0085] in, Represent the objective function; This represents the measured light intensity value of the photodetector array; Indicates the ideal light intensity value; , They represent the first Phase shifter, the first A phase shifter.

[0086] If the two bodies The preceding coefficient is denoted as , to monomer The preceding coefficient is denoted as Then the above formula can be written as:

[0087] ;

[0088] This means that it can be adjusted This is done to minimize the difference between the noisy output and the ideal output, thereby mitigating the noise.

[0089] As can be seen from the above, the present invention can be adjusted That is, the noise mitigation direction of each phase shifter, measuring the output light intensity of each port, and calculating the sum of squares of the difference from the ideal output light intensity.

[0090] Based on this, in the specific implementation of the above-mentioned optical computing error adjustment method provided in the embodiments of the present invention, step S103 solves the objective function to obtain the optimal value of the correction deviation direction variable of each phase shifter. Specifically, it may include: mapping the objective function to the Ising model; solving the ground state of the Ising model to obtain the optimal value of the correction deviation direction variable of each phase shifter.

[0091] In implementation, the application can accurately locate the variable combination that minimizes the sum of squares of differences between the measured light intensity and the ideal light intensity by mapping the objective function (i.e. the problem of minimizing the output error) into an Ising model based on the correction deviation direction variables, and then solving the ground state of the Ising model to obtain the optimal values of the correction deviation direction variables of each phase shifter. In this way, the optimization objective function is directly generated by the physical response of the optical system, without the need for separate measurement of the noise offset of each phase shifter, significantly reducing the calibration complexity, providing a reliable basis for subsequent accurate adjustment of the phase shifter, further relieving the phase noise interference, improving the output precision of the optical computing device, and also ensuring the efficiency and stability of the noise mitigation process.

[0092] Further, in the above-described optical computing error adjustment method provided by the embodiments of the application, the constructed objective function can be mapped into an Ising model using the following formula:

[0093] ;

[0094] wherein, is the objective function; is the correction deviation direction variable of the noise of the i th phase shifter; is the correction deviation direction variable of the noise of the j th phase shifter; is a quadratic term coefficient matrix element, reflecting the coupling relationship between and , determined by the noise coefficient and the ideal output; is a linear term coefficient, reflecting the linear influence of , determined by the inherent noise, the noise coefficient and the ideal output.

[0095] In implementation, the application can clearly quantify the coupling relationship between the correction deviation direction variables of each phase shifter and the linear influence of a single variable by mapping the objective function into an Ising model using the above formula, and convert the complex noise mitigation optimization problem into a structured physical model solving problem. This mapping not only preserves the core optimization logic of the objective function, but also provides an adaptive framework for efficient solving through the characteristics of the Ising model, which helps to accurately find the variable combination that minimizes the sum of squares of differences between the light intensity and the ideal light intensity, and provides guidance for subsequent phase shifter adjustment, thereby improving the accuracy and efficiency of noise correction.

[0096] Furthermore, in specific implementation, in the above steps, solving the ground state of the Ising model to obtain the optimal values ​​of the correction bias direction variables for each phase shifter can specifically include: using a heuristic algorithm to iteratively update and solve the ground state of the Ising model; wherein, starting from the initial state, the correction bias direction variables and auxiliary variables are updated step by step; each time it is updated, the next correction bias direction variable is calculated based on the current correction bias direction variables and auxiliary variables, and then the auxiliary variables are updated in combination with the time evolution function that changes with the number of steps; during the update process, when the correction bias direction variables reach the positive and negative limit values, they remain unchanged, and the corresponding auxiliary variables are set to zero; when all correction bias direction variables reach the positive and negative limit values, the combination of the values ​​of all correction bias direction variables is used as the ground state solution of the Ising model.

[0097] In implementation, a heuristic algorithm is used to iteratively solve the ground state of the Ising model. By progressively updating the correction bias direction variables and auxiliary variables, and combining this with a time evolution function that changes with the number of steps to optimize the update logic, variables that reach positive or negative limits are locked and their corresponding auxiliary variables are reset to zero. This approach can efficiently converge to a stable solution, simultaneously optimizing the correction direction of all phase shifters. This method avoids complex computational redundancy and accurately identifies the variable combination that minimizes the objective function, quickly obtaining the optimal values ​​of the correction bias direction variables for each phase shifter. This provides a reliable basis for subsequent phase shifter adjustments, helps improve the efficiency and accuracy of interferometer array noise mitigation, and further ensures output accuracy.

[0098] Furthermore, in specific implementations, in the optical computing error adjustment method provided in the embodiments of the present invention, the iterative update solution method of the heuristic algorithm can be as follows:

[0099] ;

[0100] ;

[0101] in, For time interval variables, Let be the time evolution function at step t+1. For the corresponding time evolution coefficient, Let be the direction variable for correcting the noise of the i-th phase shifter at step t. For step t, and Related auxiliary variables, Let be the direction variable for correcting the noise of the i-th phase shifter at step t+1. When at step t+1 and Related auxiliary variables, The constant coefficients, Let $\frac{t+1}{t}$ be the direction variable for correcting the noise of the $j$-th phase shifter.

[0102] In implementation, the above heuristic algorithm can select a simulation bifurcation algorithm. When solving by the iteration formula of the above heuristic algorithm, the correction deviation direction variable and the auxiliary variable can be updated and corrected in order through the time interval variable, the time evolution function dynamically changing with the step number, and the constant coefficient: first updating the correction deviation direction variable based on the current auxiliary variable, and then optimizing the auxiliary variable in combination with the variable coupling relationship and the linear influence. This structured iteration logic not only guarantees the accuracy of variable updating, but also efficiently promotes the solving process and quickly converges to the stable value of each correction deviation direction variable, providing reliable support for obtaining the ground state of the Ising model and determining the optimal adjustment direction of the phase shifter, thereby improving the efficiency of interferometer array noise mitigation.

[0103] Further, in specific implementation, in the above steps, the next correction deviation direction variable is calculated according to the current correction deviation direction variable and the auxiliary variable, and the auxiliary variable is updated in combination with the time evolution function changing with the step number, which can specifically include: at the t+1 step, the next correction deviation direction variable is calculated according to the current correction deviation direction variable and the auxiliary variable and the auxiliary variable , the new auxiliary variable is obtained by combining the time evolution function . .

[0104] Correspondingly, when the correction deviation direction variable reaches the positive and negative limit value and no longer changes, the corresponding auxiliary variable is set to zero, which can specifically include: when reaches +1 or -1, it stays on and does not change, and the corresponding is set to 0.

[0105] In implementation, at the t+1 step, updating the variables according to the above rules can not only rely on the state of the current correction deviation direction variable and the auxiliary variable, but also realize the ordered iteration of the variables in combination with the time evolution function, guaranteeing the coherence and accuracy of the updating logic; and by locking the correction deviation direction variable reaching and zeroing the corresponding auxiliary variable, it can avoid the interference of the stable variables by subsequent iterations and accelerate the convergence to the stable state. In this way, the Ising model ground state solving can be efficiently promoted, the optimal correction direction of the phase shifter can be quickly determined, the foundation for accurate mitigation of phase noise of the interferometer array and improvement of output precision is laid, while redundant calculation is reduced and the noise mitigation efficiency is improved.

[0106] Furthermore, in specific implementation, in the above steps, when all correction bias direction variables reach their positive and negative limit values, the combination of all correction bias direction variable values ​​is used as the ground state solution of the Ising model. Specifically, this may include: when all correction bias direction variables reach their positive and negative limit values ​​and there is no numerical change in multiple consecutive iterations, the correction bias direction variables are binarized to obtain a sequence containing +1 and -1, and the sequence is used as the ground state solution of the Ising model; the ground state solution minimizes the objective function.

[0107] In practice, the solution method of this invention is a heuristic algorithm that utilizes the current... At that time, the stable point of the system is the point that minimizes the objective function. Then, when each... After evolution ceases, by keeping it in Binarize it to obtain a set of values. of The sequence is used to adjust each phase shifter to mitigate noise in the interferometer array. By waiting for all correction deviation direction variables to reach their positive and negative limits and remain unchanged for several consecutive steps, and then binarizing them to obtain the ground state solution containing +1 and -1, the stability and reliability of the solution are ensured, avoiding optimization deviations caused by incomplete convergence. Furthermore, binarization precisely identifies the variable combination that minimizes the objective function, providing a basis for determining the optimal correction direction of the phase shifter. This improves output accuracy while ensuring the effectiveness and efficiency of the noise mitigation process.

[0108] Furthermore, in a specific implementation, in the optical computing error adjustment method provided in the embodiments of the present invention, the quadratic coefficient matrix elements of the Ising model are calculated using the following formula. and coefficient of the first term :

[0109] ;

[0110] ;

[0111] in, Noise figure vector The kth component, Noise figure vector The kth component, for The complex conjugate, To achieve the ideal output complex amplitude, To standardize the correction deviation values, Let be the noise of the j-th phase shifter.

[0112] In practice, the quadratic coefficient matrix elements of the Ising model are calculated using the above formula. and the coefficient of the first term , can accurately integrate key parameters such as ideal output complex amplitude, noise coefficient vector, unified correction deviation value and phase shifter noise into the model, and clearly quantify the coupling relationship between each correction deviation direction variable and the linear influence of a single variable. This accurate parameter calculation method provides a basis for the construction of the Ising model that fits the actual noise scene, ensuring that the model can accurately map the objective function, and then helping to efficiently solve the ground state, finding the optimal correction direction for the phase shifter, and effectively improving the accuracy of interferometer array noise mitigation.

[0113] Further, in the above optical calculation error adjustment method provided by the embodiments of the present application, before solving the objective function, the method can further include: obtaining a correction deviation direction variable randomly assigned to each phase shifter by the control device; the value of the correction deviation direction variable is a positive limit value or a negative limit value; wherein the positive limit value indicates an increase in the unified correction deviation value, and the negative limit value indicates a decrease in the unified correction deviation value; and calculating an initial value of the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array using the correction deviation direction variable randomly assigned by the control device.

[0114] In implementation, the present application can quickly provide an initial reference benchmark for subsequent optimization and solution by randomly assigning a correction deviation direction variable (such as +1 or -1, corresponding to an increase or decrease in the unified correction deviation value) to each phase shifter by the control device, and calculating the initial value of the light intensity difference sum of squares accordingly, avoiding the solution blind area without an initial state, and also relying on the randomly assigned discrete variable to preliminarily cover the possible correction direction range, laying a foundation for subsequent accurate iterative optimization, while simplifying the initial value calculation process, reducing the time-consuming of the early preparation, and helping to improve the overall efficiency of interferometer array noise mitigation.

[0115] Further, in the above optical calculation error adjustment method provided by the embodiments of the present application, before generating the objective function based on the correction deviation direction variable, the method can further include: obtaining a test optical signal input by the control device and a transmission matrix corresponding to the interferometer array under ideal conditions; operating the test optical signal with the transmission matrix under ideal conditions to obtain a signal matrix at the output end of the interferometer array under ideal conditions through corresponding calculation between matrices; processing the signal corresponding to each output port in the obtained signal matrix to obtain an ideal light intensity value.

[0116] In implementation, the ideal output signal matrix is obtained by obtaining the test optical signal and the ideal transmission matrix and operating them The ideal light intensity values of each port are obtained by reprocessing the matrix, a reference for noise mitigation can be accurately constructed, the ideal light intensity values are highly matched with the ideal working state of the interferometer array, reliable basis is provided for subsequent calculation of the square sum of the difference between the measured light intensity and the ideal light intensity, and the ideal light intensity values are systematically and accurately obtained through matrix operation and port signal processing, the error of manual setting of the reference is avoided, and an accurate foundation is laid for subsequent positioning of phase noise interference and determination of the optimal correction direction of the phase shifter.

[0117] It should be noted that the light calculation error adjustment method provided by the present application can first simplify the noise of the phase shifter. Assuming that the noise on each phase shifter is similar in size and small, the influence of the noise on the phase can be described by the deviation from the ideal value. Under the above assumption, under the consideration of only the first-order approximation, the difference between the output complex amplitude and the complex amplitude in the ideal case is a linear function. Under this condition, the phase or can be set to or , that is, the adjustment of is introduced to the phase, wherein characterizes the direction of calibration, represents the actual calibration size, which can be obtained by calibration of a separate phase shifter. Under this phase adjustment, by comparing the difference between the output light intensity of the interferometer array under the noisy condition and the ideal light intensity, the problem of whether to perform positive or negative calibration on each phase shifter can be equivalent to an Ising model, and the ground state of the Ising model can be obtained by solving the ground state of the Ising model through some heuristic algorithms, such as the simulated bifurcation algorithm. It needs to be emphasized that the present application is to execute the heuristic algorithm such as the simulated bifurcation algorithm to solve the ground state of the Ising model on the actual interferometer array, and in the process of executing the ground state solving algorithm, the adjustment direction of each phase shifter is simultaneously synchronized, and therefore it is called a collective adjustment method.

[0118] The present application uses the control device to set the current correction deviation direction variable to all phase shifters (that is, the phase adjustment of each phase shifter is or ). The control device inputs a test light signal, and the photodetector array measures the actual light intensity of the output port . The error is calculated on the external computing device. The the value of the step of repeating the calculation of the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device, and updating the value of the step of repeating the calculation of the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device, and updating the value of the step of repeating the calculation of the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device, and updating the value of the step of repeating the calculation of the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device, and updating the value of the step of repeating the calculation of the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device, and updating the value of the step of repeating the calculation of the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device, and updating the value of the step of repeating the calculation of the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device, and updating

[0119] In addition, it should be noted that the present application converts noise mitigation into a global optimization problem by collectively synchronously adjusting the phase direction of all phase shifters, uses a heuristic algorithm to solve the optimal solution at one time, changes the calibration operation from point-by-point debugging to batch processing, avoids independent operation of 2M phase shifters (M is the number of interferometers), reduces the dependence on high-precision calibration equipment, and significantly saves hardware resources and energy consumption. Moreover, the present application minimizes the output light intensity error by globally optimizing the phase shifter correction direction, approximates the ideal value under first-order approximation, not only reduces the error rate of optical neural network tasks, but also improves the robustness and reliability of the system in complex scenes due to long-term effectiveness of single calibration. The present application does not need to modify the physical structure of the interferometer array, and only needs to embed an optimization algorithm in the control system to realize noise mitigation. This feature enables the present application to be directly applied to integrated optical computing devices and seamlessly compatible with traditional optical computing architectures. The method provides high-precision and low-energy computing power support for artificial intelligence training, real-time big data processing and other scenarios, accelerates the transformation process of optical computing technology from the laboratory to industrial applications, and breaks through the physical bottleneck of traditional electronic computing.

[0120] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be realized by software and a general hardware platform as required, of course, it can also be realized by hardware, but in many cases the former is a better embodiment.

[0121] Embodiments of the present application also provide an optical computing error adjustment device. Based on the angle of the functional module, the device comprises:

[0122] a correction bias value determination module for determining a unified correction bias value corresponding to the phase noise of the interferometer array in the optical computing device; the interferometer array comprises at least two phase shifters;

[0123] a target function generation module for generating a target function based on the correction bias direction variable, with the minimization of the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device as the target;

[0124] a target function solving module configured to solve the target function to obtain optimal values of the correction deviation directional variables of the phase shifters;

[0125] a correction parameter sending module configured to send the optimal values of the correction deviation directional variables of the phase shifters and the unified correction deviation value to the control device, so that the control device adjusts the phase shifters simultaneously according to the optimal values of the correction deviation directional variables of the phase shifters and the unified correction deviation value to obtain the interferometer array after noise mitigation.

[0126] In the optical computing error adjustment device provided by the embodiment of the present application, the collective adjustment of all the phase shifters can be realized by the determined unified correction deviation value and the optimal values of the correction deviation directional variables of the phase shifters, so that the measured light intensity value of the photodetector array is close to the ideal light intensity value, the output accuracy of the optical computing device is effectively improved, the additional cost caused by the individual calibration of the phase shifters is avoided, and the overall cost in the running process of the entire interferometer array is greatly reduced. At the same time, the collective adjustment mode does not need to debug each phase shifter one by one, greatly simplifies the operation process of optical computing noise mitigation, reduces the complexity of optical computing noise mitigation, speeds up the process of noise mitigation, reduces the time required for calibration, and can be widely applied to integrated optical networks to provide strong support for efficient and low-cost operation of integrated optical networks.

[0127] Since the embodiments of the optical computing error adjustment device part correspond to the embodiments of the optical computing error adjustment method part, the description of the features in the corresponding embodiments of the optical computing error adjustment device can refer to the related description of the corresponding embodiments of the optical computing error adjustment method, which will not be repeated here. And has the same beneficial effects as the above-mentioned optical computing error adjustment method.

[0128] Further, in specific implementation, in the optical computing error adjustment device provided by the embodiment of the present application, the correction deviation value determination module can be specifically configured to obtain the change in light intensity output by the photodetector array after the control device randomly selects a set number of phase shifters and adjusts the phases of the phase shifters; obtain the average noise amplitude from the fitting curve corresponding to the change in light intensity output by the photodetector array, and take the average noise amplitude as the unified correction deviation value corresponding to the phase noise of the interferometer array in the optical computing device.

[0129] Further, in specific implementation, in the optical computing error adjustment device provided by the embodiment of the present application, the target function solving module can be specifically configured to map the target function to an Ising model; and solve the ground state of the Ising model to obtain the optimal values of the correction deviation directional variables of the phase shifters.

[0130] Embodiments of the present invention also provide an optical computing error adjustment system, including an optical computing device, a control device, and an external computing device. The optical computing device includes a light source, an interferometer array, and a photodetector array; the interferometer array includes at least two phase shifters. The external computing device is used to determine the unified correction deviation value corresponding to the phase noise of the interferometer array in the optical computing device; generate an objective function based on the correction deviation direction variable with the objective of minimizing the sum of squares of the differences between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device; solve the objective function to obtain the optimal value of the correction deviation direction variable for each phase shifter; and send the optimal value of the correction deviation direction variable for each phase shifter to the control device. The control device is used to simultaneously adjust each phase shifter according to the optimal value of the correction deviation direction variable for each phase shifter and the unified correction deviation value.

[0131] The aforementioned optical computing device can specifically consist of a light source, an interferometer array composed of multiple interferometers, waveguides connecting the interferometers, and a photodetector array. Each interferometer includes two noisy phase shifters, two beam splitters, and a corresponding waveguide. The optical computing device includes both calibration and computation modes. The control device controls the optical computing device to emit a preset light signal and controls the selection of the phase shifters and their corresponding phase settings. The external computing device is responsible for the computational tasks, including fitting the relationship between the output light intensity and the phase deviation to obtain a uniform correction value for the deviation. Calculate the magnitude and direction of the correction deviation. The updated value is then output to the control device.

[0132] Embodiments of the present invention also provide a computer-readable storage medium storing a computer program configured to execute the steps in any of the above-described embodiments of the optical computing error adjustment method when running.

[0133] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0134] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the optical computing error adjustment method.

[0135] The embodiment of the present application further provides another computer program product comprising a nonvolatile computer readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above optical computing error adjustment method embodiments.

[0136] Those skilled in the art will further appreciate that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or both, and that the interchangeability of hardware and software methods is contemplated. Accordingly, the examples described herein are presented for purposes of illustration and not limitation. The steps of the examples are not necessarily limited to the order described, and the order of steps can be varied.

[0137] The above provides a detailed introduction to the optical computing error adjustment method, system, medium and program product provided by the present application. The principles and implementation modes of the present application are described herein by applying specific examples, and the above description of the examples is only applicable to help understand the method of the present application and its core idea. It should be pointed out that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and modifications can be made to the present application, and these improvements and modifications also fall within the protection scope of the present application.

Claims

1. A method of optical computing error adjustment, the method comprising: The method comprises the following steps: After the control device randomly selects a set number of phase shifters and adjusts the phases of the phase shifters, the light intensity change output by the photodetector array is obtained; According to the fitting curve corresponding to the light intensity change output by the photodetector array, the average noise amplitude is obtained, and the average noise amplitude is taken as a unified correction bias value corresponding to the phase noise of the interferometer array in the optical computing device; The interferometer array comprises at least two phase shifters; A target function based on the correction bias direction variable is generated by minimizing the sum of squares of the difference between the measured light intensity value and the ideal light intensity value of the photodetector array in the optical computing device; The target function is mapped into an Ising model by using the following formula: ； wherein, is the objective function; is the correction bias direction variable of the i-th phase shifter noise; is the correction bias direction variable of the j-th phase shifter noise; is the quadratic term coefficient matrix element, reflecting the coupling relationship between and ; is the linear term coefficient, reflecting the linear influence of ; The quadratic coefficient matrix elements of the Ising model are calculated using the following formula and the linear coefficient : ； ； wherein is the kth component of the noise coefficient vector is the kth component of the noise coefficient vector is the kth component of the noise coefficient vector is the kth component of the noise coefficient vector is the complex conjugate of is the complex conjugate of is the ideal output complex amplitude is the uniform correction bias value is the noise of the jth phase shifter; The ground state of the Ising model is solved to obtain the optimal value of the correction bias direction variable of each phase shifter; The optimal value of the correction bias direction variable of each phase shifter and the unified correction bias value are sent to the control device, so that the control device adjusts each phase shifter according to the optimal value of the correction bias direction variable of each phase shifter and the unified correction bias value.

2. The method of claim 1, wherein, Solving the ground state of the Ising model to obtain the optimal value of the correction bias direction variable of each phase shifter comprises: The ground state of the Ising model is iteratively updated by using a heuristic algorithm; Wherein, starting from the initial state, the correction bias direction variable and the auxiliary variable are updated step by step; each time the correction bias direction variable is updated, the next correction bias direction variable is calculated according to the current correction bias direction variable and the auxiliary variable, and the auxiliary variable is updated in combination with the time evolution function varying with the step number; during the updating process, when the correction bias direction variable reaches the positive and negative limit values, it remains unchanged, and the corresponding auxiliary variable is set to zero; when all the correction bias direction variables reach the positive and negative limit values, the combination of the values of all the correction bias direction variables is taken as the ground state solution of the Ising model.

3. The method of claim 2, wherein, The iterative updating solving method of the heuristic algorithm is: ； ； wherein, is a time interval variable, is a time evolution function of the t+1 step, is a corresponding time evolution coefficient, is a correction deviation direction variable of the i-th phase shifter noise at the t step, is an auxiliary variable related to at the t step, is a correction deviation direction variable of the i-th phase shifter noise at the t+1 step, is an auxiliary variable related to at the t+1 step, is a constant coefficient, is a correction deviation direction variable of the j-th phase shifter noise at the t+1 step.

4. The method of claim 3, wherein, According to the current correction bias direction variable and the auxiliary variable, the next correction bias direction variable is calculated, and the auxiliary variable is updated in combination with the time evolution function varying with the step number, comprising: At the t+1 step, according to the current and auxiliary variable , calculate , and combine the time evolution function to obtain the new auxiliary variable ; When the correction bias direction variable reaches the positive and negative limit values, it remains unchanged, and the corresponding auxiliary variable is set to zero, comprising: When reaches +1 or -1, it is kept unchanged and the corresponding is set to 0.

5. The method of claim 2, wherein, When all the correction bias direction variables reach the positive and negative limit values, the combination of the values of all the correction bias direction variables is taken as the ground state solution of the Ising model, comprising: When all the correction bias direction variables reach the positive and negative limit values and there is no any numerical change in continuous multiple iterations, the correction bias direction variable is binary processed to obtain a sequence containing +1 and -1, and the sequence is taken as the ground state solution of the Ising model; the ground state solution makes the target function reach the minimum value.

6. The method of claim 1, wherein, Before solving the target function, the method further comprises the following steps: The correction bias direction variable randomly allocated by the control device for each phase shifter is obtained; the value of the correction bias direction variable is a positive limit value or a negative limit value; wherein the positive limit value indicates that the unified correction bias value is increased, and the negative limit value indicates that the unified correction bias value is decreased; An initial value of a sum of squares of differences between measured light intensity values of the photodetector array and ideal light intensity values is calculated using a randomly assigned correction bias direction variable of the control device.

7. The method of claim 1, wherein, Before generating the objective function based on the correction bias direction variable, further comprising: Obtaining a test light signal input by the control device and a transmission matrix corresponding to the interferometer array in an ideal case; Operating the test light signal with the transmission matrix in the ideal case to obtain a signal matrix of the output end of the interferometer array in the ideal case through corresponding calculation between matrices; Processing the signal corresponding to each output port in the obtained signal matrix to obtain the ideal light intensity value.

8. The method of claim 1, wherein, Generating the objective function based on the correction bias direction variable, comprising: After obtaining the measured light intensity values of the photodetector array and the ideal light intensity values, a sum of squares of differences between the measured light intensity values of the photodetector array and the ideal light intensity values is calculated to generate the objective function.

9. An optical computing error adjustment system, comprising: Comprising: An optical computing device, a control device and an external computing device; The optical computing device comprises a light source, an interferometer array and a photodetector array; The interferometer array comprises at least two phase shifters; The external computing device is configured to obtain light intensity changes output by the photodetector array after the control device randomly selects a set number of phase shifters and adjusts the phases of the phase shifters; According to a fitting curve corresponding to the light intensity changes output by the photodetector array, an average noise amplitude is obtained, and the average noise amplitude is taken as a unified correction bias value corresponding to phase noise of the interferometer array in the optical computing device; A target function based on the correction bias direction variable is generated by minimizing a sum of squares of differences between measured light intensity values of the photodetector array and ideal light intensity values in the optical computing device, the target function is mapped to an Ising model, and the target function is mapped to the Ising model by using the following formula: ； wherein, is the objective function; is the correction bias direction variable of the i-th phase shifter noise; is the correction bias direction variable of the j-th phase shifter noise; is the quadratic term coefficient matrix element, reflecting the coupling relationship between and is the linear term coefficient, reflecting the linear influence of ;​ The quadratic coefficient matrix elements of the Ising model are calculated using the following formula and the linear coefficient : ； ； wherein is the kth component of the noise coefficient vector is the kth component of the noise coefficient vector is the kth component of the noise coefficient vector is the kth component of the noise coefficient vector is the complex conjugate of is the complex conjugate of is the ideal output complex amplitude is the uniform correction bias value is the noise of the jth phase shifter; The ground state of the Ising model is solved to obtain optimal values of the correction bias direction variable of each phase shifter, and the optimal values of the correction bias direction variable of each phase shifter are sent to the control device; The control device is configured to adjust each phase shifter according to the optimal values of the correction bias direction variable of each phase shifter and the unified correction bias value.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, wherein the computer program is executed by a processor to implement the steps of the optical computing error adjustment method according to any one of claims 1 to 8.

11. A computer program product comprising a computer program, characterized in that, The computer program is executed by a processor to implement the steps of the optical computing error adjustment method according to any one of claims 1 to 8.

Images