A method of electromagnetic structure yield analysis and related apparatus

Abstract

This application provides a method and related equipment for analyzing the yield rate of electromagnetic structures, belonging to the field of electromagnetic structure design and manufacturing technology. The method includes: acquiring multiple disturbance samples of the electromagnetic structure to be analyzed under the influence of processing errors to form an evaluation sample set; training a ranking surrogate model to predict the relative performance ranking among samples rather than absolute performance values; using the model to rank all samples to obtain an ordered sample set; employing a binary search strategy to iteratively locate a sliding window, performing full-wave simulation on the samples within the window, and determining the boundary position between good and bad products based on the simulation results; when the boundary is within the window, accurately determining the boundary position and calculating the yield rate based on the simulation results of all samples within the window. This application significantly reduces the amount of simulation computation by introducing a ranking learning and sliding window-binary search collaborative strategy, breaking through the accuracy bottleneck of existing surrogate models, and achieving efficient and accurate electromagnetic structure yield rate analysis.

Description

Technical Field

[0001] This application relates to the field of electromagnetic structure design and manufacturing technology, and in particular to a method and related equipment for analyzing the yield rate of electromagnetic structures. Background Technology

[0002] In the electromagnetic device design process, the optimized solution that meets the design requirements must ultimately be put into manufacturing. However, due to limitations in manufacturing process precision, the actual product cannot accurately reproduce the design parameters. These unavoidable parameter deviations can cause product performance to deviate from the target, or even lead to failure, affecting the feasibility of productizing the designed structure. Therefore, before putting the structural design into actual manufacturing, it is necessary to analyze the yield rate of the structure under manufacturing errors.

[0003] Traditional yield analysis uses the Monte Carlo method, which requires sampling local perturbations in the evaluated structure and feeding these samples into simulation software to assess its performance. The yield rate is then evaluated based on these performance values. However, this method requires a large number of samples and simulations, consuming significant computational resources and time. Existing methods typically use surrogate models to replace expensive simulation calculations, but surrogate models cannot meet the high-precision prediction requirements of local performance under small perturbations, making it difficult to accurately calculate the yield rate and limiting the accuracy of the yield rate evaluation.

[0004] Therefore, there is an urgent need in this field for a new method for analyzing the yield rate of electromagnetic structures, which can break through the bottleneck of analytical accuracy under limited computational budget and provide a low-cost, high-precision robust index analysis scheme. Summary of the Invention

[0005] The main objective of this application is to propose an electromagnetic structure yield analysis method, electronic device, storage medium, and program product, aiming to solve the technical problems of high computational resource consumption and long processing time in the traditional Monte Carlo method, as well as the insufficient analysis accuracy of the existing surrogate model method, and to achieve efficient and accurate yield evaluation.

[0006] To achieve the above objectives, one aspect of this application proposes a method for analyzing the yield rate of electromagnetic structures, the method comprising: S1: Obtain multiple disturbance samples of the electromagnetic structure to be analyzed under the influence of processing errors to form an evaluation sample set; S2: Train a ranking proxy model, which is used to predict the relative performance ranking of any two samples in the evaluation sample set, rather than predicting their absolute performance values; S3: Use the ranking proxy model to rank all samples in the evaluation sample set according to their performance to obtain an ordered sample set; S4: Using a binary search strategy, a sliding window is iteratively located in the ordered sample set. In each iteration, full-wave simulation is performed on all samples in the current sliding window to obtain their true performance values, and the boundary between good and bad products is determined based on the simulation results. S5: When the boundary is located within the current sliding window, determine the precise position of the boundary in the ordered sample set based on the simulation results of all samples within the window, and calculate the yield rate of the electromagnetic structure based on the precise position.

[0007] In some embodiments, training the ranking agent model in step S2 further includes: Construct an initial training dataset, which includes a subset of samples drawn from the evaluation sample set and their corresponding simulated real performance values; Using the initial training dataset, the ranking proxy model is trained by jointly optimizing the objective loss function, which includes a first loss component for measuring the accuracy of performance value prediction and a second loss component for measuring the consistency of ranking among samples.

[0008] In some embodiments, the second loss component is a ranking loss function used to minimize the difference between the sample performance ranking predicted by the model and the sample ranking obtained based on the true performance values ​​during training.

[0009] In some embodiments, step S4, which employs a binary search strategy to iteratively locate the sliding window, further includes: S41: Initialize the left and right boundaries of the binary search in the ordered sample set; S42: Calculate the middle position based on the current left and right boundaries, and determine the sliding window with the middle position as the center; S43: Perform full-wave simulation on all samples within the sliding window, and determine their good product status based on the simulation results; S44: If all samples in the sliding window are good or all are bad, then update the left or right boundary according to the judgment result and return to step S42 for the next iteration; if both good and bad samples exist in the sliding window, then determine that the boundary has been located and proceed to step S5.

[0010] In some embodiments, determining the precise location of the boundary and calculating the yield rate in step S5 further includes: When it is determined that the boundary is within the current sliding window, the range of the sliding window is expanded, and full-wave simulation is performed on the samples that have not yet been simulated within the expanded window to obtain the true performance values ​​of all samples. Based on the actual performance values ​​of all samples within the expanded window, the dividing line between good and bad products is precisely determined in the ordered sample set. The yield rate of the electromagnetic structure is calculated based on the boundary subscript: Yield rate = (Total number of evaluation samples - Boundary subscript) / Total number of evaluation samples.

[0011] In some embodiments, the electromagnetic structure has multiple performance targets, and the method further includes: In step S3, the multi-objective performance values ​​of each sample are combined into a single comprehensive performance value according to a preset scalarization rule, and the evaluation sample set is sorted based on the comprehensive performance value.

[0012] In some embodiments, in each iteration of step S4, after performing full-wave simulation on the samples within the current sliding window, the method further includes: Using the samples obtained in this iteration and their actual performance simulation values, the current ranking proxy model is fine-tuned and updated online to improve the model's ability to predict the ranking of samples in key areas in subsequent iterations.

[0013] In some embodiments, when performing online fine-tuning and updating of the ranking agent model, the loss function used is consistent with the target loss function used when training the initial ranking agent model, including the first loss component and the second loss component.

[0014] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0015] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0016] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0017] Compared with the prior art, this application has the following beneficial effects: 1) Introducing a ranking learning mechanism to overcome the accuracy bottleneck of surrogate models: This application innovatively applies a ranking neural network model to the yield analysis of electromagnetic structures. By predicting the relative performance ranking among samples rather than the absolute performance values, the learning difficulty of the model is significantly reduced, thereby achieving more reliable yield estimation under limited simulation samples. The ranking model is not sensitive to fluctuations in absolute performance values, but focuses more on the relative relationships between samples. Therefore, it can maintain stable analytical performance even when simulation errors or noise exist.

[0018] 2) Efficient collaboration between sliding window and binary search significantly reduces simulation computation: This application combines sliding window and binary search strategies, simulating only a small number of samples near the performance ranking boundary. Compared with the traditional Monte Carlo method, this reduces the number of simulations by more than 70%, greatly saving computational resources and time costs. The window mechanism ensures that only samples most likely to contain the good product boundary are simulated each time, while the binary search quickly narrows down the boundary range. The combination of the two greatly improves search efficiency while maintaining accuracy.

[0019] 3) Supports continuous online learning, forming a virtuous cycle of analysis and learning: During the analysis process, this application continuously uses new simulation data to fine-tune the surrogate model, gradually improving the model's predictive ability in key sample regions, forming a virtuous cycle of "analysis-learning-reanalysis". Even with a small initial number of simulation samples, the analysis accuracy can still be gradually improved through the iterative update mechanism, lowering the data acquisition threshold.

[0020] 4) Achieving a dynamic balance between computational resources and accuracy: Users can flexibly control the analysis accuracy and computational cost by adjusting parameters such as window size and iteration count to adapt to the needs of different application scenarios. Expanding the window strategy further reduces the impact of surrogate model sorting errors on the final yield assessment, improving analysis accuracy.

[0021] 5) Excellent scalability and generalization ability: This method is not only applicable to electromagnetic structures, but can also be widely used in yield analysis of various structures such as metasurfaces, photonic devices, and mechanical bearings, demonstrating strong cross-domain adaptability. The Transformer-based surrogate model can effectively capture the complex relationships between high-dimensional parameters, making it suitable for modern electromagnetic structures with a large number of design variables.

[0022] 6) Supports multi-objective performance evaluation: Through preset scalarization rules or weighting methods, multi-objective performance can be integrated into a single ranking criterion, which is suitable for yield analysis scenarios under complex constraints.

[0023] 7) Compatible with existing simulation tools and workflows: The method can be directly integrated into mainstream electromagnetic simulation platforms such as CST and HFSS without changing the existing design process, which facilitates engineering application and promotion. Attached Figure Description

[0024] Figure 1 This is a flowchart of the electromagnetic structure yield analysis method provided in the embodiments of this application.

[0025] Figure 2 This is a schematic diagram illustrating the key steps of the electromagnetic structure yield analysis method in the embodiments of this application.

[0026] Figure 3 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0029] In the electromagnetic device design process, after designers obtain the nominal design parameters that meet the design specifications through optimization algorithms, the design needs to be put into actual production. However, limited by the current processing technology precision, the manufactured products cannot completely and accurately reproduce the design parameters, and there will always be a certain degree of random processing errors. These unavoidable parameter deviations can cause the actual performance of the product to deviate from the design target, and in severe cases, even lead to product failure, directly affecting the product feasibility and mass production yield of the designed structure. Therefore, accurate analysis of the yield of electromagnetic structures under the influence of processing errors during the design phase is of great significance for guiding design improvements and assessing mass production risks.

[0030] Traditional yield analysis methods primarily employ the Monte Carlo method. This method first samples numerous random disturbances to the nominal design parameters based on a pre-defined processing error distribution model, obtaining tens of thousands of disturbance samples. Then, each disturbance sample is individually fed into electromagnetic simulation software (such as CST STUDIO or HFSS) for full-wave simulation, calculating the true performance value of each sample. Finally, the proportion of samples meeting the design specifications is statistically analyzed, yielding the yield rate. While the Monte Carlo method is simple in principle and accurate in results, its fatal flaw lies in its enormous computational resource consumption and extremely long analysis cycle. For a complex electromagnetic structure, a single full-wave simulation may take several minutes or even hours, while the Monte Carlo method requires thousands or even tens of thousands of simulations, which is often unsustainable in engineering practice.

[0031] To reduce computational costs, existing technologies have proposed yield rate evaluation methods based on surrogate models. The core idea of ​​this method is to first train a surrogate model (such as a neural network or Gaussian process regression) using a small number of simulation samples to replace time-consuming full-wave simulations; then, this surrogate model is used to quickly predict the performance values ​​of all perturbation samples, thereby calculating the yield rate. However, such methods face severe accuracy bottlenecks in practical applications. Manufacturing errors typically manifest as small perturbations near the nominal design parameters. The impact of these perturbations on electromagnetic performance is often highly nonlinear, and the performance differences between perturbation samples are extremely subtle. While surrogate models perform well in predicting macroscopic trends, their ability to predict local performance under small perturbations is severely insufficient, leading to large prediction errors for the absolute performance values ​​of individual samples. Since yield rate statistics depend on the accurate judgment of whether the performance of each sample meets the standards, the prediction errors of the surrogate model directly cause the yield rate calculation results to deviate significantly from the true value, making it difficult to meet engineering accuracy requirements.

[0032] In view of this, this application provides a method, electronic device, storage medium, and program product for electromagnetic structure yield analysis based on a ranking neural network model and a sliding window. This scheme obtains the relative performance ranking of locally perturbed sampled samples through a ranking surrogate model. Combining binary search and sliding window techniques, high-precision yield analysis can be completed by simulating only a small number of samples. An iterative analysis process and continuous learning framework are adopted to gradually improve the prediction accuracy and ranking reliability of the ranking surrogate model for important sample regions. This application significantly reduces the dependence on computing resources, overcomes the accuracy bottleneck of existing surrogate model analysis methods, and is applicable to yield analysis of various structures such as electromagnetic structures and metasurfaces, possessing technical advantages of high efficiency, accuracy, and wide adaptability.

[0033] The electromagnetic structure yield analysis method provided in this application relates to the field of electromagnetic structure design and manufacturing technology. This method can be applied to terminals, servers, or software running on either terminal or server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the electromagnetic structure yield analysis method, but is not limited to the above forms.

[0034] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0035] like Figure 1 As shown, this embodiment provides a method for analyzing the yield rate of electromagnetic structures. The method specifically includes the following steps: Step S101: Obtain multiple disturbance samples of the electromagnetic structure to be analyzed under the influence of processing errors to form an evaluation sample set; Step S102: Train the ranking proxy model. The ranking proxy model is used to predict the relative ranking of the performance of any two samples in the evaluation sample set, rather than predicting their absolute performance values. Step S103: Use the ranking proxy model to rank all samples in the evaluation sample set according to their performance to obtain an ordered sample set; Step S104: Using a binary search strategy, a sliding window is iteratively located in the ordered sample set. In each iteration, full-wave simulation is performed on all samples in the current sliding window to obtain their true performance values, and the boundary between good and bad products is determined based on the simulation results. Step S105: When the boundary is within the current sliding window, determine the precise position of the boundary in the ordered sample set based on the simulation results of all samples within the window, and calculate the yield rate of the electromagnetic structure based on the precise position.

[0036] Below, in conjunction with Figure 2 The solutions of the embodiments of this application will be described in detail and explained with specific application examples.

[0037] This embodiment proposes a method for analyzing the yield rate of electromagnetic structures based on a sorting neural network model and a sliding window. The specific process consists of the following four steps: Step S1: Problem definition and processing error modeling.

[0038] S1-1: First, the electromagnetic structure needs to be modeled as a floating-point vector of length n. Each floating-point number This corresponds to a specific structural parameter value. Based on a pre-parameterized electromagnetic structure model, each n-dimensional floating-point vector... This allows for a specific electromagnetic structure design.

[0039] S1-2: Define the objective function used to quantify the performance of the electromagnetic structure. Common objectives include gain and bandwidth. Using simulation software (such as CST STUDIO, HFSS, etc.), we can simulate each structure and calculate its corresponding performance curves. .in The parameter vector represents the structure, and m represents different curve types, such as... Electromagnetic structures possess various performance curves, such as gain curves. However, during the design process, only certain performance characteristics within specific frequency bands are considered, such as the maximum gain in the 10-15 GHz band. Therefore, further calculations of the corresponding target values ​​for the structure are necessary. Because electromagnetic scenarios involve multiple objectives, we define each objective as having a performance target value. A sample is considered good only if all objectives meet their performance targets. For simplicity, this good-product judgment logic is defined here as an indicator function. When electromagnetic structure When it is a good product Conversely .

[0040] S1-3: To quantify the processing error for a given electromagnetic structure To assess the performance impact, the random error model is defined as follows:

[0041] Where the covariance matrix It is a diagonal matrix, representing the independence of the perturbations of each parameter; The value representing the processing error intensity of the i-th parameter is provided by the user to reflect the actual manufacturing precision. Therefore, the actual structural parameter vector after perturbation is: In the defined random error model Internal collection The perturbation samples constitute the error evaluation sample set. ,in Then the electromagnetic structure The formula for calculating the yield rate is:

[0042] S2: Training the ranking proxy model.

[0043] S2-1: To reduce analysis costs and accelerate the analysis speed, this embodiment uses a Transformer model as a surrogate model to replace the time-consuming full-wave simulation. The input of the surrogate model is a vector of structural parameters. The output is its corresponding multi-objective performance value. To construct the initial model, this embodiment uses the model generated in steps S1-3. Randomly selected Each sample is fed into the simulation to obtain its corresponding simulation performance value, forming the initial training dataset. ,use Train the surrogate model. Since the surrogate model will be used in subsequent steps to rank the performance of the entire evaluation set, an additional ranking loss (HR) is introduced during training, in addition to the standard mean squared error (MSE) loss, to improve the model's ranking ability. The loss function is defined as follows:

[0044] in, and These are weighting coefficients used to balance the target loss and the ranking loss; It is the number of multiple objectives; Let be the model's predicted value for the i-th target; The ranking loss is used to improve the accuracy of the model's prediction of the relative magnitude of the performance values ​​of two samples. It is calculated by comparing the predicted ranking with the actual ranking. Based on this step, we obtain the initial surrogate model. , These are the parameters corresponding to the initial proxy model.

[0045] S3: Determine the performance ranking of samples and the positioning sliding window; fine-tune the samples and online surrogate model within the simulation window.

[0046] S3-1: This embodiment uses a surrogate model to measure the error evaluation sample set. The performance values ​​of all samples are used to obtain an ordered sample set, which is then sorted in ascending order of performance values. (For multiple objectives, the values ​​can be mapped to a comprehensive performance value according to a preset scalarization rule and then sorted.) Assumption The order of the samples in the dataset is consistent with the order of the true performance values, and there must be boundary subscripts in this order. , satisfying the location The samples on the left are all defective. ),lie in The samples on the right are all good quality. Then its yield rate To quickly find the boundary index This invention uses a binary search method, initially setting the left subscript. subscript Subscript To further improve search efficiency, this invention uses the intermediate index after determining the intermediate index. Establish a size of at the center point A sliding window, by traversing the window sample set The sample performance values ​​are based on the search boundary index. .

[0047] S3-2: Due to the window sample set obtained in step S3-1 The vast majority of sample performance values ​​are obtained through the surrogate model. The predicted values ​​have a certain error compared to the actual performance values. To compensate for this error, this embodiment uses a windowed sample set. All samples are fed into the simulation to obtain their true performance values, thereby improving the accuracy of yield analysis values.

[0048] S3-3: At the same time, in order to enable the surrogate model to gradually improve its performance on the evaluation sample set... To improve the prediction accuracy and ranking reliability, this embodiment employs an online model update method during the analysis process, thereby enhancing the accuracy of the indicators and the efficiency of the analysis. Therefore, in this step, the window sample set obtained in step S3-2 will be utilized. The new simulation data in the current agent model Fine-tuning is performed, and the training loss is made consistent with the loss defined in step S2, resulting in a new surrogate model. .

[0049] S4: Determine whether to reposition the sliding window based on the simulation performance values ​​of the samples within the window, or provide the yield rate value.

[0050] S4-1: The window sample set obtained in step S3-2 After obtaining the simulation performance values ​​of all samples, a good product indicator function can be used. Judgment Sample Set For information on good quality products, please refer to [link / reference]. Figure 2 In practical problems, the following three situations may occur: a) All samples in the text are good: This indicates that the boundary subscripts are all good. Not in window sample set In the middle, and On the left side of the sliding window (in the area with smaller performance values), move the lower right index... Update to middle subscript ,Right now Return to step S3 to reposition the sliding window. b) All samples in the text are defective: This indicates that the boundary subscripts are all defective. Not in window sample set In the middle, and On the right side of the sliding window (the area with larger performance values), move the left index... Update to middle subscript ,Right now Return to step S3 to reposition the sliding window. c) The samples included both good and bad products: this indicates the boundary subscripts In window sample set In the middle, the boundary subscript has been located. Once the approximate location is found, the binary search process can be exited, and step S4-2 can be initiated.

[0051] S4-2: To further reduce the impact of surrogate model prediction errors on the accuracy of yield rate analysis, after case c in step S4-1 occurs, the size of the sliding window will be increased to... Obtain an expanded window sample set This will change the original sliding window. Expand to Then... After inputting unsimulated samples into the simulation and obtaining the corresponding true performance values, the boundary indices are located. And calculate the yield rate:

[0052] This step reduces yield rate assessment errors caused by surrogate model sorting errors and improves analysis accuracy.

[0053] In summary, existing methods for analyzing the yield rate of electromagnetic structures have significant drawbacks: traditional Monte Carlo evaluation methods require each sample of local perturbations to be individually fed into simulation software to calculate the true performance values ​​and statistically analyze robustness indicators, resulting in high computational resource consumption and long processing times; existing surrogate model-based methods aim to accurately predict the absolute performance values ​​of samples, but surrogate models struggle to meet the high-precision prediction requirements of local performance under small perturbations, thus limiting the accuracy of indicator evaluation. In contrast, this application does not require surrogate models to accurately predict absolute performance data, but rather obtains the relative performance ranking between samples through surrogate models. High-precision yield rate analysis can be completed by simulating only a small number of samples. This method significantly reduces the dependence on computational resources and overcomes the accuracy bottleneck of existing surrogate models, achieving efficient and accurate analysis within a limited budget.

[0054] In summary, this application has the following advantages and beneficial effects compared to the prior art: 1) Innovative introduction of ranking learning mechanism: This application innovatively applies the ranking neural network model to the yield analysis of electromagnetic structures. By predicting the relative performance ranking among samples rather than the absolute performance value, the difficulty of model prediction is greatly reduced, and the feasibility and reliability of yield analysis are improved.

[0055] 2) Overcoming the accuracy bottleneck of surrogate models: By using the ranking loss function (HR Loss) to guide the model to learn the order relationship between samples, the requirement for high-precision fitting of absolute performance values ​​is avoided, thereby achieving more reliable yield estimation under limited simulation samples.

[0056] 3) Enhanced robustness to noise and uncertainty: The ranking model is not sensitive to fluctuations in absolute performance values ​​and focuses more on the relative relationships between samples. Therefore, it can maintain stable analytical performance even when simulation errors or noise exist.

[0057] 4) Significantly reduce simulation computation: By combining the sliding window and binary search strategies, simulations are performed only on a small number of samples near the performance ranking boundary. Compared with the traditional Monte Carlo method, the number of simulations can be reduced by more than 70%, which greatly saves computing resources and time costs.

[0058] 5) Supports online continuous learning and adaptive model updates: During the analysis process, new simulation data is continuously used to fine-tune the surrogate model, gradually improving the model's predictive ability in key sample areas, forming a virtuous cycle of "analysis-learning-reanalysis".

[0059] 6) Reduce dependence on the amount of initial training data: Even with a small number of initial simulation samples, the analysis accuracy can still be gradually improved through the iterative update mechanism, which lowers the threshold for data acquisition.

[0060] 7) Sliding window and binary search work together efficiently: The window mechanism ensures that only the sample most likely to contain the good product boundary is simulated each time, and the binary search quickly narrows down the boundary range. The combination of the two greatly improves the search efficiency while ensuring accuracy.

[0061] 8) Achieve dynamic balance between computing resources and accuracy: Users can flexibly control the analysis accuracy and computing cost by adjusting parameters such as window size and number of iterations to adapt to the needs of different application scenarios.

[0062] 9) Possesses good scalability and generalization ability: This method is not only applicable to electromagnetic structures, but can also be widely used in yield analysis of various structures such as metasurfaces, photonic devices, and mechanical bearings, and has strong cross-domain adaptability.

[0063] 10) Supports multi-objective performance evaluation: Through preset scalarization rules or weighting methods, multi-objective performance can be integrated into a single ranking criterion, which is suitable for yield analysis scenarios under complex constraints.

[0064] 11) Provide clear interpretability and decision support: By visualizing the sorting results and the position of the sliding window, users can intuitively understand the distribution of good product boundaries, providing a basis for actual production and processing.

[0065] 12) Compatible with existing simulation tools and workflows: The method can be directly integrated into mainstream electromagnetic simulation platforms such as CST and HFSS without changing the existing design process, which facilitates engineering application and promotion.

[0066] 13) Supports high-dimensional parameter space analysis: The surrogate model based on Transformer can effectively capture the complex relationships between high-dimensional parameters and is suitable for modern electromagnetic structures with a large number of design variables.

[0067] 14) It has good convergence and stability: In internal experiments, it was verified that the method can quickly converge to the true yield rate in most cases, and the results have good repeatability, making it suitable for engineering practice.

[0068] This application not only realizes a paradigm shift from "absolute performance prediction" to "relative ranking learning" at the technical level, but also provides an efficient, accurate, and scalable yield analysis solution at the engineering level, demonstrating significant technological advancement and broad industrial application value.

[0069] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0070] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0071] Please see Figure 3 , Figure 3 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 301 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 302 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 302 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 302 and is called and executed by the processor 301 using the methods described above in the embodiments of this application. Input / output interface 303 is used to implement information input and output; The communication interface 304 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 305 transmits information between various components of the device (e.g., processor 301, memory 302, input / output interface 303, and communication interface 304); The processor 301, memory 302, input / output interface 303, and communication interface 304 are connected to each other within the device via bus 305.

[0072] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0073] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0074] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0075] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0076] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented in the embodiments of this program product are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments. The executable computer program code or "code" used to perform the various embodiments can be written in high-level programming languages ​​such as C, C++, Python, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (e.g., Transact-SQL), Perl, or in various other programming languages.

[0077] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0078] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0079] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0080] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0081] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0082] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0083] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0084] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0085] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0086] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0087] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for analyzing the yield rate of electromagnetic structures, characterized in that, The method includes the following steps: S1: Obtain multiple disturbance samples of the electromagnetic structure to be analyzed under the influence of processing errors to form an evaluation sample set; S2: Train a ranking proxy model, which is used to predict the relative performance ranking of any two samples in the evaluation sample set, rather than predicting their absolute performance values; S3: Use the ranking proxy model to rank all samples in the evaluation sample set according to their performance to obtain an ordered sample set; S4: Using a binary search strategy, a sliding window is iteratively located in the ordered sample set. In each iteration, full-wave simulation is performed on all samples in the current sliding window to obtain their true performance values, and the boundary between good and bad products is determined based on the simulation results. S5: When the boundary is located within the current sliding window, determine the precise position of the boundary in the ordered sample set based on the simulation results of all samples within the window, and calculate the yield rate of the electromagnetic structure based on the precise position.

2. The method according to claim 1, characterized in that, The training of the ranking agent model in step S2 further includes: Construct an initial training dataset, which includes a subset of samples drawn from the evaluation sample set and their corresponding simulated real performance values; Using the initial training dataset, the ranking proxy model is trained by jointly optimizing the objective loss function, which includes a first loss component for measuring the accuracy of performance value prediction and a second loss component for measuring the consistency of ranking among samples.

3. The method according to claim 2, characterized in that, The second loss component is a ranking loss function, which is used to minimize the difference between the sample performance ranking predicted by the model and the sample ranking obtained based on the true performance values ​​during training.

4. The method according to claim 1, characterized in that, Step S4, which uses a binary search strategy to iteratively locate the sliding window, further includes: S41: Initialize the left and right boundaries of the binary search in the ordered sample set; S42: Calculate the middle position based on the current left and right boundaries, and determine the sliding window with the middle position as the center; S43: Perform full-wave simulation on all samples within the sliding window, and determine their good product status based on the simulation results; S44: If all samples in the sliding window are good or all are bad, then update the left or right boundary according to the judgment result and return to step S42 for the next iteration; if both good and bad samples exist in the sliding window, then determine that the boundary has been located and proceed to step S5.

5. The method according to claim 4, characterized in that, Determining the precise location of the boundary and calculating the yield rate in step S5 further includes: When it is determined that the boundary is within the current sliding window, the range of the sliding window is expanded, and full-wave simulation is performed on the samples that have not yet been simulated within the expanded window to obtain the true performance values ​​of all samples. Based on the actual performance values ​​of all samples within the expanded window, the dividing line between good and bad products is precisely determined in the ordered sample set. The yield rate of the electromagnetic structure is calculated based on the boundary subscript: Yield rate = (Total number of evaluation samples - Boundary subscript) / Total number of evaluation samples.

6. The method according to claim 1, characterized in that, The electromagnetic structure has multiple performance objectives, and the method further includes: In step S3, the multi-objective performance values ​​of each sample are combined into a single comprehensive performance value according to a preset scalarization rule, and the evaluation sample set is sorted based on the comprehensive performance value.

7. The method according to claim 1, characterized in that, In each iteration of step S4, after performing full-wave simulation on the samples within the current sliding window, the method further includes: Using the samples obtained in this iteration and their actual performance simulation values, the current ranking proxy model is fine-tuned and updated online to improve the model's ability to predict the ranking of samples in key areas in subsequent iterations.

8. The method according to claim 7, characterized in that, When performing online fine-tuning and updating of the ranking agent model, the loss function used is consistent with the target loss function used when training the initial ranking agent model, including the first loss component and the second loss component.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 8.

Images