A method and system for suppressing electromagnetic interference of a chip

Abstract

The present invention provides an electromagnetic interference suppression method and system for a chip, which relates to the technical field of intelligent chips, and comprises the following steps: obtaining layout data of a target chip, extracting structural parameters and device parameters based on the layout data through an image segmentation algorithm, and constructing a multi-physical field coupling simulation model; determining the optimal excitation source corresponding to the target chip through a machine learning algorithm, adopting a time-domain finite element method, calculating and obtaining transient multi-physical field distribution through an adaptive time step control strategy, extracting multi-scale distribution features in the time-frequency domain, constructing a spectrum fingerprint library of electromagnetic interference using a deep convolutional neural network, and generating an electromagnetic interference identification result; determining the optimal suppression circuit topology based on the electromagnetic interference identification result through a suppression topology search algorithm, determining the optimal suppression device parameters through a multi-objective optimization algorithm combined with electromagnetic compatibility index constraints, and generating an electromagnetic interference suppression strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of smart chips, and in particular to a method and system for suppressing electromagnetic interference of chips. Background Art

[0002] With the rapid development of electronic technology and the high integration of integrated circuits, the functions of chips are becoming increasingly complex and the operating frequency is constantly increasing, resulting in increasingly prominent electromagnetic interference problems within and between chips. Electromagnetic interference not only affects the normal operation of chips, reduces the reliability and stability of the system, but may also lead to information leakage and security risks.

[0003] Although the existing electromagnetic interference suppression methods have alleviated the electromagnetic interference problem of chips to a certain extent, they still have many limitations and are difficult to meet the increasingly stringent electromagnetic compatibility requirements of modern electronic systems. Traditional methods mainly rely on manual experience and repeated experiments, lack intelligence and adaptive capabilities, and are difficult to cope with complex and changing electromagnetic environments. At the same time, most of the existing methods are targeted at specific types of electromagnetic interference and lack a general and systematic suppression solution. Therefore, how to effectively suppress the electromagnetic interference of chips has become one of the key issues that need to be solved in the field of electronic design.

[0004] In summary, there is an urgent need for an intelligent and adaptive electromagnetic interference suppression method that can automatically generate the optimal suppression strategy according to the actual working environment and electromagnetic interference characteristics of the chip, and effectively suppress various types of electromagnetic interference. At the same time, the new suppression method should be universal and scalable, meet the increasingly complex electromagnetic compatibility design requirements, improve the reliability and stability of the chip, and ensure the normal operation of the electronic system. The present invention can solve the problems in the prior art. Summary of the invention

[0005] The embodiments of the present invention provide a method and system for suppressing electromagnetic interference of a chip, which can solve the problems in the prior art.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provided is a method for suppressing electromagnetic interference for a chip, comprising:

[0008] Acquire layout data of a target chip whose electromagnetic interference is to be suppressed, extract structural parameters of each functional module in the target chip and device parameters of corresponding devices through an image segmentation algorithm based on the layout data, and construct a multi-physics field coupling simulation model including the overall structure of the target chip and each device model based on the structural parameters and the device parameters;

[0009] Based on the multi-physics field coupling simulation model, the optimal excitation source corresponding to the target chip is determined through a machine learning algorithm. Based on the optimal excitation source, a time-domain finite element method is used to dynamically adjust the time step of the simulation calculation through an adaptive time step control strategy to calculate and obtain the transient multi-physics field distribution of the target chip. By post-processing the transient multi-physics field distribution, the multi-scale distribution characteristics of the electromagnetic interference in the time and frequency domains are extracted. The multi-scale distribution characteristics in the time and frequency domains are learned and classified using a deep convolutional neural network to construct a spectrum fingerprint library of electromagnetic interference, and generate an electromagnetic interference identification result including the electromagnetic interference type and interference severity.

[0010] Based on the electromagnetic interference identification result, the optimal suppression circuit topology is determined through the suppression topology search algorithm, and the optimal suppression device parameters are determined through the multi-objective optimization algorithm combined with the electromagnetic compatibility index constraints. Based on the optimal suppression circuit topology and the optimal suppression device parameters, an electromagnetic interference suppression strategy is generated.

[0011] In an optional embodiment,

[0012] Acquiring layout data of a target chip for suppressing electromagnetic interference, extracting structural parameters of each functional module in the target chip and device parameters of corresponding devices through an image segmentation algorithm based on the layout data, and constructing a multi-physics field coupling simulation model including the overall structure of the target chip and each device model based on the structural parameters and the device parameters, including:

[0013] Obtain the layout data file of the target chip;

[0014] Converting the layout data file into a layout image, encoding the layout image into a quantum state, extracting multi-scale texture features and topological structure features of the layout image in the corresponding space of the quantum state based on quantum Fourier transform and quantum phase estimation algorithm, determining the layout features, constructing a similarity measurement model based on quantum gate circuits, calculating the similarity of different layout images in the feature space through entanglement measurement between quantum states, retrieving a layout similar to the layout image in a preset layout database using a quantum search algorithm, obtaining candidate areas for layout segmentation and parameter extraction, and extracting geometric structure parameters and device parameters of each functional module in the candidate areas;

[0015] Based on the geometric structure parameters, a three-dimensional geometric model of the target chip is constructed, and based on the device parameters, a device model is constructed, and the corresponding device model is placed in the three-dimensional geometric model to establish a multi-physical field coupling simulation model.

[0016] In an optional embodiment,

[0017] The multi-physics field coupling simulation model also includes:

[0018] The multi-physics field coupling simulation model is discretized by using an adaptive meshing method;

[0019] Adaptively adjust the grid density and size according to the electromagnetic field intensity distribution and device position distribution in different areas of the target chip;

[0020] In combination with multi-layer mesh nesting technology, the multi-physics field coupling simulation model is meshed at different scales to determine the overall electromagnetic interference characteristics and local electromagnetic distribution;

[0021] The grid division is based on the coupling relationship between different physical fields to determine a unified grid division scheme.

[0022] In an optional embodiment,

[0023] Based on the multi-physics field coupling simulation model, the optimal excitation source corresponding to the target chip is determined by a machine learning algorithm, including:

[0024] Based on the multi-physics field coupling simulation model, a group of partial differential equations describing the coupling effects of the electromagnetic field, thermal field and stress field inside the chip is constructed, and the group of partial differential equations is discretized using a preset finite element method to obtain a group of multi-physics field coupling algebraic equations;

[0025] The multi-physics field coupling algebraic equation group is converted into an optimization model with the excitation source as the design variable, and based on the machine learning algorithm, the excitation source parameter combination that makes the multi-physics field distribution optimal is obtained through iterative search to obtain the optimal excitation source;

[0026] The optimal excitation source is input into an excitation source signal generation module to synthesize an excitation signal sequence matching the actual scene, and the distribution characteristics corresponding to the optimal excitation source are maintained in combination with a preset data enhancement method to expand the excitation source samples.

[0027] In an optional embodiment,

[0028] Based on the optimal excitation source, the time domain finite element method is used to dynamically adjust the time step of the simulation calculation through an adaptive time step control strategy to calculate the transient multi-physical field distribution of the target chip, including:

[0029] Based on the multi-physics field coupled algebraic equations, a time domain term is introduced to construct a time domain finite element model, and a time stepping algorithm is used to calculate the multi-physics field distribution of the internal nodes of the target chip at each time step;

[0030] Based on the severity of changes in multi-physics fields, combined with an adaptive time step control strategy, the time step of the time domain solution is dynamically adjusted through error estimation and iterative correction;

[0031] Based on the hybrid parallel mode, the time-domain finite element model is divided into multiple sub-regions, in which the communication synchronization between sub-regions is realized through the message passing interface, and the multi-threaded parallel execution mechanism is used to achieve multi-threaded parallelism within the sub-region.

[0032] In an optional embodiment,

[0033] By post-processing the transient multi-physical field distribution, the time-frequency domain multi-scale distribution features of electromagnetic interference are extracted, and the time-frequency domain multi-scale distribution features are learned and classified using a deep convolutional neural network to build an electromagnetic interference spectrum fingerprint library. The electromagnetic interference identification results containing the electromagnetic interference type and interference severity are generated, including:

[0034] Perform short-time Fourier transform on transient multi-physical field distribution to obtain time-frequency representation of multi-physical fields, construct time-frequency spectrum, extract local texture features at different time and frequency scales from the time-frequency spectrum based on preset multi-scale convolution kernel, and determine multi-scale distribution characteristics of electromagnetic interference in time-frequency domain;

[0035] Construct a deep convolutional neural network model with multi-scale distribution features in the time-frequency domain as input, optimize the network parameters of the deep convolutional neural network model through end-to-end supervised learning, and combine the attention mechanism to determine the sensitivity of the deep convolutional neural network model to each time-frequency domain to classify the electromagnetic interference type;

[0036] Based on the trained deep convolutional neural network model, a preset set of electromagnetic interference samples is input, the time-frequency domain features corresponding to each electromagnetic interference sample are extracted, and mapped to the corresponding electromagnetic interference type, the corresponding relationship between the electromagnetic interference type and the time-frequency domain features is determined, and a spectrum fingerprint library is constructed;

[0037] Combined with the interference severity assessment module, the severity assessment features are extracted, including interference energy, interference frequency distribution and interference duration, and combined with the electromagnetic interference type to generate electromagnetic interference identification results.

[0038] In an optional embodiment,

[0039] Based on the electromagnetic interference identification results, the optimal suppression circuit topology is determined through the suppression topology search algorithm, including:

[0040] The suppression topology search algorithm is constructed based on the fruit fly optimization algorithm;

[0041] Performing quantum encoding on the topology of the inhibition circuit, mapping each device in the inhibition circuit to a quantum bit, wherein the state of the device corresponds to the probability amplitude of the quantum bit, and multiple quantum bits form a quantum chromosome, corresponding to a candidate solution of the inhibition circuit topology;

[0042] A set of quantum chromosomes is randomly generated, each quantum chromosome corresponds to a fruit fly individual, and the initial fruit fly population is determined;

[0043] Performing a taste search on the initial fruit fly population, observing the quantum bits according to the quantum chromosomes of the individual fruit flies, determining the inhibitory circuit topology determination solution corresponding to the inhibitory circuit topology candidate solution from the superposition state of the quantum bits, calculating the fitness corresponding to the inhibitory circuit topology determination solution, determining the initial optimal inhibitory circuit topology in the search space, and updating the probability amplitude of all quantum bits in the corresponding quantum chromosome;

[0044] Taking the initial optimal inhibitory circuit topology as the visual search center point, generating a new set of quantum chromosomes near the visual search center point according to a preset neighbor threshold, determining the visual search sample points, calculating the fitness corresponding to the visual search sample points, selecting the visual search sample points corresponding to the maximum fitness, determining the advanced optimal inhibitory circuit topology, and using it as the new visual search center point, while updating the probability amplitude of the quantum bit;

[0045] Based on the quantum revolving door operation, the probability amplitude of the quantum bit is adjusted and updated, and the rotation angle of the quantum revolving door is adaptively determined within a preset revolving door angle range, and the rotation angle is inversely proportional to the fitness;

[0046] Based on the advanced optimal inhibitory circuit topology and combined with random perturbations, a new generation of fruit fly populations was generated;

[0047] The iteration is repeated until the preset number of iterations is reached, and the advanced optimal inhibitory circuit topology obtained by the last taste search and visual search is used as the final optimal inhibitory circuit topology.

[0048] According to a second aspect of the embodiments of the present invention,

[0049] An electromagnetic interference suppression system for a chip is provided, comprising:

[0050] The first unit is used to obtain layout data of a target chip whose electromagnetic interference is to be suppressed, extract structural parameters of each functional module in the target chip and device parameters of corresponding devices through an image segmentation algorithm based on the layout data, and construct a multi-physics field coupling simulation model including the overall structure of the target chip and each device model based on the structural parameters and the device parameters;

[0051] The second unit is used to determine the optimal excitation source corresponding to the target chip based on the multi-physics field coupling simulation model through a machine learning algorithm, and based on the optimal excitation source, use a time-domain finite element method to dynamically adjust the time step of the simulation calculation through an adaptive time step control strategy to calculate and obtain the transient multi-physics field distribution of the target chip, extract the multi-scale distribution characteristics of the electromagnetic interference in the time and frequency domains by post-processing the transient multi-physics field distribution, use a deep convolutional neural network to learn and classify the multi-scale distribution characteristics in the time and frequency domains, build a spectrum fingerprint library of electromagnetic interference, and generate an electromagnetic interference identification result including the type of electromagnetic interference and the severity of interference;

[0052] The third unit is used to determine the optimal suppression circuit topology based on the electromagnetic interference identification result through the suppression topology search algorithm, determine the optimal suppression device parameters through the multi-objective optimization algorithm combined with the electromagnetic compatibility index constraints, and generate an electromagnetic interference suppression strategy based on the optimal suppression circuit topology and the optimal suppression device parameters.

[0053] According to a third aspect of the embodiments of the present invention,

[0054] An electronic device is provided, comprising:

[0055] processor;

[0056] a memory for storing processor-executable instructions;

[0057] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0058] A fourth aspect of the embodiments of the present invention is:

[0059] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0060] In an embodiment of the present invention, by acquiring the layout data of the target chip and using an image segmentation algorithm to extract the structural parameters and device parameters of each functional module, the physical layout and device information of the target chip can be accurately reconstructed. Based on these data, a multi-physical field coupling simulation model containing the overall structure of the target chip and each device model is constructed, which can fully simulate the behavior of the chip under different physical fields; the optimal excitation source of the target chip is identified and determined by a machine learning algorithm, so that the simulation is closer to the actual working conditions, and the authenticity and accuracy of the simulation results are improved; the spectrum fingerprint library of electromagnetic interference is constructed, and the characteristics of electromagnetic interference are effectively extracted and classified, so that the identification of electromagnetic interference is more accurate and comprehensive; the acquisition of electromagnetic interference identification results improves the accuracy and comprehensiveness of identification; the application of the suppression topology search algorithm effectively reduces the complexity of electromagnetic interference design. Through global search and iterative optimization, the circuit topology with the best suppression effect is found to ensure the practical feasibility of the suppression circuit; through a multi-objective optimization algorithm, the optimal suppression device parameters are determined to improve the suppression efficiency of the circuit and the reliability of the system, so that the suppression circuit can effectively cope with different electromagnetic interference situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A schematic diagram of a flow chart of an electromagnetic interference suppression method for a chip according to an embodiment of the present invention;

[0062] Figure 2 The present invention is a schematic diagram of the structure of an electromagnetic interference suppression system for a chip according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0065] Figure 1 FIG. 1 is a flow chart of a method for suppressing electromagnetic interference of a chip according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0066] S101. Acquire layout data of a target chip for suppressing electromagnetic interference, extract structural parameters of each functional module in the target chip and device parameters of corresponding devices through an image segmentation algorithm based on the layout data, and construct a multi-physics field coupling simulation model including the overall structure of the target chip and each device model based on the structural parameters and the device parameters;

[0067] The layout data specifically refers to the detailed information generated during the chip design process to represent the physical layout of the chip, including the geometry, position, connection mode and other information of each circuit component in the chip, which is an indispensable reference file in the chip manufacturing process. The structural parameters specifically refer to the geometry and topology information of each functional module in the target chip, including the position, size, shape and connection relationship between the modules. The structural parameters determine the physical layout of the chip and the interaction between the circuits. The device specifically refers to the physical components that implement specific functions on the chip, such as transistors, resistors, capacitors, etc. Each device has specific electrical characteristics and plays a key role in the circuit. The device parameters specifically refer to the physical and electrical characteristics of each device that constitutes the chip, such as type, material, operating voltage, power consumption, response speed, etc. These parameters directly affect the performance and function of the chip.

[0068] In this embodiment, by acquiring the layout data of the target chip and using the image segmentation algorithm to extract the structural parameters and device parameters of each functional module, the physical layout and device information of the target chip can be accurately reconstructed. Based on these data, a multi-physics field coupling simulation model including the overall structure of the target chip and each device model is constructed, which can fully simulate the behavior of the chip under different physical fields (such as electromagnetic field, thermal field, stress field), and help to deeply analyze the performance of the chip under actual working conditions, optimize the design, and formulate strategies for electromagnetic interference suppression, thereby improving the reliability and functional stability of the chip.

[0069] In an optional embodiment, obtaining layout data of a target chip to suppress electromagnetic interference, extracting structural parameters of each functional module in the target chip and device parameters of corresponding devices through an image segmentation algorithm based on the layout data, and constructing a multi-physics field coupling simulation model including the overall structure of the target chip and each device model based on the structural parameters and the device parameters, including:

[0070] Obtain the layout data file of the target chip;

[0071] Converting the layout data file into a layout image, encoding the layout image into a quantum state, extracting multi-scale texture features and topological structure features of the layout image in the corresponding space of the quantum state based on quantum Fourier transform and quantum phase estimation algorithm, determining the layout features, constructing a similarity measurement model based on quantum gate circuits, calculating the similarity of different layout images in the feature space through entanglement measurement between quantum states, retrieving a layout similar to the layout image in a preset layout database using a quantum search algorithm, obtaining candidate areas for layout segmentation and parameter extraction, and extracting geometric structure parameters and device parameters of each functional module in the candidate areas;

[0072] Based on the geometric structure parameters, a three-dimensional geometric model of the target chip is constructed, and based on the device parameters, a device model is constructed, and the corresponding device model is placed in the three-dimensional geometric model to establish a multi-physical field coupling simulation model.

[0073] Read the layout data file of the target chip from the chip design layout database. When reading the layout data file, you need to parse the file format and extract the geometric graphics, layer information, text annotations and other data content in the layout. For different layout data formats, you need to use corresponding parsers and reading libraries to complete data extraction. The extracted layout data is generally stored in the form of vector graphics, which contains the shape, coordinates, size and other geometric information of each layer in the layout.

[0074] Convert the extracted vector layout data into a rasterized layout image. Determine the pixel size of the layout image based on the layout size and the required resolution. Perform geometric transformation and sampling on the layout data to convert the vector graphics into a pixel grid. Rasterize the layout geometry into pixels using scan conversion algorithms, such as scan line algorithms and polygon filling algorithms. The generated layout image is usually a grayscale or binary image, and different layers can be represented by different grayscale values.

[0075] Encode the generated layout image into a quantum state. Determine the dimension of the quantum state and the number of quantum bits according to the pixel size of the layout image. Map the pixel values ​​of the layout image to the probability amplitude of the quantum state. For grayscale images, the pixel grayscale values ​​can be normalized to between 0 and 1 as the probability amplitude of the corresponding quantum state basis vector. For binary images, the pixel values ​​0 and 1 can be directly mapped to the basis vector of the quantum state. Construct the quantum state corresponding to the layout image through quantum gate circuits or tensor networks. The quantum state exists in the form of a vector in a high-dimensional Hilbert space, which can efficiently represent the pixel information of the layout image.

[0076] Based on quantum Fourier transform, the quantum state of the layout image is converted from the pixel domain to the frequency domain. Fast Fourier transform is implemented through quantum gate circuits to map the quantum state of the spatial domain to the quantum state of the frequency domain. In the frequency domain, different frequencies correspond to texture features of different scales and directions in the layout image. Based on the quantum phase estimation algorithm, the multi-scale texture features of the layout image in the frequency domain are extracted. By building a phase estimation circuit, the amplitude and phase of different frequency components are estimated. By using quantum parallelism, the eigenvalues ​​of multiple frequency components are obtained simultaneously. The extracted amplitude and phase are used as the multi-scale texture feature vector of the layout image. Based on the topological structure analysis of the quantum state, the topological structure features of the layout image are extracted. By measuring the probability distribution of the quantum state on different basis vectors, the topological relationship between different regions and graphic elements in the layout image is obtained. The entanglement property of the quantum state is used to analyze the connectivity and adjacency relationship between different regions. The topological structure features of the layout image are extracted, such as the number of connected components, the number of holes, the Euler number, etc. The extracted multi-scale texture features and topological structure features are combined to form the feature vector of the layout image. The representation capability of the feature vector is optimized through weighted fusion or feature selection, and a mapping relationship is established between the quantum state of the layout image and its feature vector to form a quantum code of the layout features.

[0077] According to the quantum encoding of layout features, a quantum gate circuit for similarity measurement is constructed. A set of parameterized quantum gates is designed to realize the transformation and mapping of feature space. By adjusting the parameters of the quantum gate, different similarity measurement requirements can be flexibly adapted. The parameters of the quantum gate circuit are learned through training and optimization. The parameters of the quantum gate circuit are trained by using known layout image pairs and similarity labels in a supervised learning or unsupervised learning manner. The quantum gate parameters are optimized by minimizing the similarity prediction error or maximizing the feature discrimination and other objective functions. The trained quantum gate circuit can realize the similarity measurement of layout images in the feature space. Using the trained quantum gate circuit, a similarity measurement model based on quantum states is constructed. Given the quantum states of two layout images, they are transformed and processed by quantum gate circuits to obtain quantum state outputs representing similarity. By measuring the probability amplitude or expected value of the output quantum state, the similarity value between the two layout images is obtained. Entanglement measurement between quantum states can efficiently calculate the similarity of layout images in the feature space.

[0078] Convert the layout images in the preset layout database into quantum states to build a layout quantum database. For each layout image, extract its features and encode them into corresponding quantum states. Store the quantum states of all layout images in the quantum memory to form a layout quantum database for search. Use the quantum search algorithm to retrieve layouts similar to the target layout image in the layout quantum database. Use the quantum state of the target layout image as the query state and compare the similarity with the quantum states in the layout quantum database. Calculate the similarity between the target layout image and each layout image in the database by measuring the entanglement between quantum states. Sort the layout images in the database according to the similarity value. Select several layout images with the highest similarity as the retrieval results. By measuring the quantum state with the largest similarity value, obtain the layout most similar to the target layout image. According to the needs, the top-k similar layouts can be selected as candidate results.

[0079] Perform layout segmentation on the retrieved candidate layout images. Use image segmentation algorithms to divide the layout images into different functional areas. According to the design rules and topological structure of the layout, extract the regional marks and boundary contours of each functional module. Extract the geometric structure parameters in the area of ​​each functional module. Through contour tracing and shape analysis, extract the geometric information of the module, such as size, position, and direction. Measure the length, width, area and other parameters of the module to determine its layout and arrangement in the layout. Extract the relative position relationship between modules. Extract the device parameters contained in each functional module. Identify the basic devices in the module through pattern matching and parameter estimation. Extract the key parameters such as device type, size, threshold voltage, etc. According to the connection topology of the device, extract the interconnection relationship and circuit structure between the devices. Organize and store the extracted geometric structure parameters and device parameters. Establish a hierarchical structured representation of the layout, including chip-level, module-level and device-level parameter information. Associate the parameters with the corresponding layout elements to form a complete layout feature description. These extracted parameters will be used for subsequent 3D modeling and simulation analysis.

[0080] Based on the extracted geometric structure parameters, the overall size and outline of the chip are determined, and the three-dimensional shell model of the chip is constructed to model the chip shell. Inside the chip, the three-dimensional geometric model of each functional module is constructed according to the hierarchical structure of the layout. The three-dimensional entity of each module is generated in the chip coordinate system using the extracted module size, position, direction and other parameters. Considering the interconnection relationship between modules, the three-dimensional wiring structure between modules is constructed. In each functional module, a detailed three-dimensional device model is constructed based on the device parameters. According to the type and size of the device, the corresponding three-dimensional geometric entity is generated. The device model is placed in the corresponding module area, and the three-dimensional connection between the devices is constructed according to the interconnection relationship. The constructed chip shell, functional module and device model are assembled and integrated. According to the layout and hierarchical structure of the layout, the three-dimensional models of each part are spatially arranged and aligned, and the generated three-dimensional geometric model can fully represent the physical structure and layout of the target chip.

[0081] Based on the extracted device parameters, the physical model of the device is constructed. According to the type and characteristics of the device, appropriate physical equations and constitutive relations are selected. For transistors, semiconductor device physical equations such as drift-diffusion equations and Poisson's equations are used to describe carrier transport and potential distribution. For passive devices such as resistors and capacitors, circuit theory and electromagnetic field equations are used to describe their electrical characteristics. Combine the constructed device physical model with the three-dimensional geometric model to establish a multi-physics field simulation model of the device. Apply the physical equations and constitutive relations to the three-dimensional geometric region of the device, and set appropriate boundary conditions and initial conditions. Consider the material properties and impurity distribution inside the device, discretize and solve the physical field of the device. Integrate the device model into the overall simulation model of the chip. Based on the three-dimensional geometric model of the chip, a chip-level multi-physics field coupling simulation environment is established. The interaction and coupling relationship between different physical fields such as electric field, magnetic field, and thermal field are taken into consideration. Use the finite element method numerical solution technology to model and simulate the multi-physics field problems of the chip. Set the boundary conditions and excitation signals of the simulation, and run the multi-physics field coupling simulation. Perform a comprehensive physical field analysis of the chip according to the actual working conditions. Obtain physical quantities such as the potential distribution, current density, and temperature gradient inside the chip to evaluate the performance indicators and reliability of the chip. The simulation results can be used to optimize chip design, verify functional correctness, and predict potential failure modes.

[0082] In this embodiment, by using image segmentation algorithms and quantum state coding technology, multi-scale texture features and topological structure features can be efficiently extracted from complex layout data, key structural information of chip functional modules can be captured, and accurate geometric and circuit parameters can be provided to ensure the accuracy of subsequent analysis; the similarity measurement model implemented by quantum gate circuits can quickly and accurately compare the similarities between different layouts, thereby improving the efficiency of layout retrieval. The quantum search algorithm further speeds up the speed of locating similar layouts in a large database, providing an efficient tool for design optimization and troubleshooting; the constructed three-dimensional geometric model and multi-physical field coupling simulation model can accurately simulate the working state of the chip under the combined action of multiple physical fields such as electromagnetic, thermal, and stress. This helps to comprehensively evaluate the performance of the chip, discover potential problems, optimize the design, and ensure the stability and reliability of the chip in actual use.

[0083] In an optional embodiment, the multi-physics field coupling simulation model further includes:

[0084] The multi-physics field coupling simulation model is discretized by using an adaptive meshing method;

[0085] Adaptively adjust the grid density and size according to the electromagnetic field intensity distribution and device position distribution in different areas of the target chip;

[0086] In combination with multi-layer mesh nesting technology, the multi-physics field coupling simulation model is meshed at different scales to determine the overall electromagnetic interference characteristics and local electromagnetic distribution;

[0087] The grid division is based on the coupling relationship between different physical fields to determine a unified grid division scheme.

[0088] Determine the initial parameters of meshing based on the geometric structure and physical characteristics of the multi-physics coupling simulation model. Set the type of mesh (such as tetrahedron, hexahedron, etc.), as well as the initial size and density of the mesh. Consider the characteristic scale of the model and the calculation accuracy requirements, and select appropriate mesh parameters.

[0089] Adaptive meshing algorithm is used to generate meshes for multi-physics coupling simulation models. The distribution and size of the mesh are adaptively adjusted according to the geometry and boundary conditions of the model. In key areas of the model, such as areas with drastic changes in electromagnetic field intensity and areas with densely distributed devices, the mesh density is appropriately increased to capture local physical field changes. In areas with small field intensity gradients or simple geometric structures, sparser meshes can be used to reduce the amount of calculation.

[0090] In the process of meshing, the strategy of adaptive mesh encryption and coarsening is introduced. The density of the mesh is dynamically adjusted according to the gradient of the electromagnetic field distribution and the complexity of the device layout. For areas with large field intensity gradient or dense devices, the mesh is encrypted and the local mesh size is refined. For areas with gentle field intensity changes or sparse devices, the mesh is coarsened to reduce the number of meshes. Through adaptive adjustment, the mesh distribution is matched with the physical field distribution and device layout, improving the calculation efficiency and accuracy.

[0091] The multi-layer mesh nesting technology is used to mesh the multi-physics field coupling simulation model at different scales. The multi-level mesh structure is introduced, and a coarser mesh is used on the overall model to capture the global electromagnetic interference characteristics. In local key areas, such as the functional modules and key components of the chip, more refined sub-grids are nested to accurately describe the local electromagnetic field distribution. Through the nesting and connection of multi-layer meshes, the coordinated solution of physical fields at different scales is achieved, taking into account both global characteristics and local details.

[0092] Consider the coupling relationship between multiple physical fields and coordinate and unify the meshing of different physical fields. In the case of coupling of multiple physical fields such as electromagnetic fields, thermal fields, and structural mechanics, it is necessary to ensure the consistency and continuity of the meshes of different physical fields at the interface. Through the matching and interpolation technology of the mesh, data transmission and coupling calculation between physical fields can be realized. A unified meshing scheme can reduce the interpolation error between physical fields and improve the accuracy and efficiency of multi-physical field coupling simulation.

[0093] In this embodiment, the adaptive meshing algorithm dynamically adjusts the mesh density according to the complexity of the physical field distribution, especially in areas where the electromagnetic field intensity changes dramatically or the devices are densely packed, and performs local mesh encryption to ensure that the physical field changes are accurately captured; by using sparse meshes in areas with small gradients or simple geometric structures, unnecessary calculations are reduced, and multi-layer mesh nesting technology is used to achieve global and local fine solutions, thereby taking into account both global calculation speed and local detail capture; in the simulation of multi-physical field coupling, a unified meshing and matching technology is used to ensure the consistency between different physical fields at the interface, reduce interpolation errors, and improve the reliability of coupled simulation results.

[0094] S102. Based on the multi-physics field coupling simulation model, the optimal excitation source corresponding to the target chip is determined through a machine learning algorithm. Based on the optimal excitation source, a time-domain finite element method is used to dynamically adjust the time step of the simulation calculation through an adaptive time step control strategy to calculate and obtain the transient multi-physics field distribution of the target chip. By post-processing the transient multi-physics field distribution, the multi-scale distribution characteristics of the electromagnetic interference in the time and frequency domains are extracted. The multi-scale distribution characteristics in the time and frequency domains are learned and classified using a deep convolutional neural network to construct a spectrum fingerprint library of electromagnetic interference, and generate an electromagnetic interference identification result including the electromagnetic interference type and interference severity.

[0095] The excitation source specifically refers to the input signal or physical quantity applied to the target chip to induce a response in the multi-physics field simulation. The excitation source can be a physical quantity such as voltage, current, magnetic field, temperature, etc., which is used to simulate the external stimulation of the chip under actual working conditions, thereby generating corresponding electromagnetic, thermal, mechanical and other multi-physics field responses.

[0096] In this embodiment, a machine learning algorithm is used to identify and determine the optimal excitation source of the target chip, so that the simulation is closer to the actual working conditions, and the authenticity and accuracy of the simulation results are improved; an adaptive time step control strategy is adopted to dynamically adjust the time step during the simulation process to improve the efficiency of the simulation calculation, especially when the transient multi-physical field changes drastically, the physical phenomena at the critical moment can be captured more accurately; the multi-scale distribution features in the time and frequency domains in the simulation results are learned and classified through a deep convolutional neural network, and a spectrum fingerprint library of electromagnetic interference is constructed, which effectively extracts and classifies the features of electromagnetic interference, making the identification of electromagnetic interference more accurate and comprehensive; the electromagnetic interference identification result finally generated includes the interference type and severity, providing a reliable reference for the anti-interference design and optimization of the chip.

[0097] In an optional embodiment, based on the multi-physics field coupling simulation model, determining the optimal excitation source corresponding to the target chip through a machine learning algorithm includes:

[0098] Based on the multi-physics field coupling simulation model, a group of partial differential equations describing the coupling effects of the electromagnetic field, thermal field and stress field inside the chip is constructed, and the group of partial differential equations is discretized using a preset finite element method to obtain a group of multi-physics field coupling algebraic equations;

[0099] The multi-physics field coupling algebraic equation group is converted into an optimization model with the excitation source as the design variable, and based on the machine learning algorithm, the excitation source parameter combination that makes the multi-physics field distribution optimal is obtained through iterative search to obtain the optimal excitation source;

[0100] The optimal excitation source is input into an excitation source signal generation module to synthesize an excitation signal sequence matching the actual scene, and the distribution characteristics corresponding to the optimal excitation source are maintained in combination with a preset data enhancement method to expand the excitation source samples.

[0101] Based on the multi-physics coupling simulation model, a mathematical model is established to describe the coupling effects of electromagnetic field, thermal field and stress field inside the chip. For the electromagnetic field, Maxwell's equations are used to describe the propagation and interaction of electric field and magnetic field. For the thermal field, Fourier heat conduction equation is used to characterize the temperature distribution and heat flow transmission inside the chip. For the stress field, elastic mechanics equation is used to describe the stress-strain relationship and deformation inside the chip. Considering the coupling effect between different physical fields, coupling terms are introduced into the partial differential equations. The coupling between electromagnetic field and thermal field is reflected by Joule heat and thermoelectric effect, and the coupling terms of temperature gradient and current density are introduced into the equation. The coupling between electromagnetic field and stress field is reflected by piezoelectric effect and magnetostrictive effect, and the coupling terms of strain tensor and electromagnetic field intensity are introduced into the equation. The coupling between thermal field and stress field is reflected by thermal expansion and contraction effect, and the strain term caused by temperature change is introduced into the equation. By introducing appropriate coupling terms into the partial differential equations, a complete mathematical model describing the coupling effect of multi-physics fields is established. The equations include the unknowns of electric field, magnetic field, temperature field and stress field, as well as the coupling relationship between them. The boundary conditions and initial conditions are determined according to the actual working environment and excitation conditions of the chip. Finally, a coupled partial differential equation group is obtained, which describes the distribution and evolution of the multi-physical fields inside the chip.

[0102] Select a suitable finite element method to spatially discretize the partial differential equations. Commonly used finite element methods, preferably the Galerkin method, select appropriate finite element formats and interpolation functions according to the characteristics of the equations and the solution requirements; convert the partial differential equations into discrete algebraic equations. On the basis of finite element discretization, the unknowns in the partial differential equations are expanded using finite element basis functions, and the equations are integrated on each finite element. By introducing test functions and weighted residual methods, a discretized algebraic equation system is obtained. The unknowns in the equation system are the physical values ​​at the finite element nodes, such as electric field intensity, temperature, displacement, etc.; in the discretization process, the influence of multi-physical field coupling effects on the equation system is considered. In the discretization of a single physical field, the coupling terms of other physical fields are retained to ensure that the coupling effects are reflected. For the coupling terms across physical fields, appropriate interpolation and projection are performed during discretization to ensure data transfer and consistency between physical fields. A multi-physical field coupled algebraic equation system is obtained. The coefficient matrix of the equation group contains the relationship between physical quantities after finite element discretization, and the right-hand side term contains the contribution of boundary conditions and excitation sources. By solving this algebraic equation group, we can obtain the numerical solution of the multi-physical fields inside the chip and describe the distribution of electromagnetic fields, thermal fields and stress fields at discrete points.

[0103] The multi-physics field coupling algebraic equations are transformed into an optimization model with the excitation source as the design variable. The terms related to the excitation source in the equations are extracted to construct the objective function and constraints. The objective function can be an expression that makes the multi-physics field distribution meet specific performance indicators, such as minimizing electromagnetic interference, optimizing the uniformity of heat distribution, etc. The constraints include the equilibrium relationship of the equations themselves, as well as the boundary restrictions of device parameters, physical field strength, etc. The machine learning algorithm is used to solve the optimization model. Through iterative search, the optimal excitation source parameter combination is found in the design variable space. In each iteration, the multi-physics field coupling algebraic equations are solved according to the current excitation source parameters to evaluate the satisfaction of the objective function value and the constraints. According to the evaluation results, the excitation source parameters are updated to generate new solution candidates. In the optimization search process, the machine learning algorithm continuously improves the excitation source parameter combination through intelligent exploration and utilization strategies. The algorithm conducts a global search in the solution space and combines local search methods to accelerate the convergence speed. Through iterative optimization, the optimal solution is gradually approached to obtain the excitation source parameter combination that makes the multi-physics field distribution optimal. The optimized optimal excitation source parameters are obtained. Under the optimal combination of excitation source parameters, the electromagnetic field, thermal field and stress field distribution inside the chip will meet the preset performance indicators, realizing the optimal design of multi-physics fields. The optimal excitation source can serve as an important reference for chip simulation and design, guiding the physical implementation and optimization of the chip.

[0104] The parameters of the optimal excitation source are input into the excitation source signal generation module. The excitation source signal generation module synthesizes an excitation signal sequence that matches the actual scene according to the excitation source type and parameters. For electromagnetic excitation, a voltage or current signal with a specific frequency, amplitude and phase can be generated. For thermal excitation, a temperature curve or heat flow pulse that varies with time can be generated. For mechanical stress excitation, a force or displacement load sequence of a specific form can be generated. Considering the characteristics of the actual scene, the excitation signal sequence is modulated and optimized. According to the working environment and application requirements of the chip, the noise, interference and distortion factors of the actual scene are introduced. Through signal processing techniques such as filtering, modulation, encoding, etc., the excitation signal sequence is made closer to the real environment. At the same time, the time domain and frequency domain characteristics of the signal are optimized to ensure that the excitation signal can effectively stimulate the multi-physical field response inside the chip. In order to enhance the diversity and robustness of the excitation source samples, the excitation signal sequence is expanded by using data enhancement methods. On the premise of maintaining the distribution characteristics of the optimal excitation source, random perturbations and transformations are introduced to generate multiple samples that are similar to but different from the original excitation signal. Data enhancement methods can include adding random noise, time shifting, amplitude scaling, frequency modulation, etc. Through data enhancement, the coverage of stimulus source samples is expanded, and the generalization ability of chip simulation and optimization is improved. A set of diverse stimulus signal sequences that match the actual scenario is obtained. The stimulus signal sequence is used as the stimulus input of the chip to simulate the multi-physical field stimulus in the real environment. By applying these optimized stimulus signals to the chip, the performance and reliability of the chip can be fully evaluated and the effectiveness of the design can be verified. The expanded stimulus source samples provide a more comprehensive and reliable test basis for chip simulation verification and optimization.

[0105] In this embodiment, by establishing a mathematical model and introducing coupling terms, the complex physical phenomena inside the chip are accurately simulated, providing a solid foundation for subsequent simulation and optimization; the multi-physical field coupling equation is converted into an optimization model using a machine learning algorithm, and the optimal excitation source parameter combination is successfully found to ensure that the distribution of each physical field inside the chip meets the preset performance indicators; based on the optimal excitation source parameters, an excitation signal sequence is generated, and the diversity and robustness of the samples are expanded through data enhancement methods, so that the simulation results are closer to reality and the reliability of the chip design is enhanced; through spatial discretization processing and solving a group of algebraic equations, the full embodiment of the multi-physical field coupling effect is ensured, and through the diversified processing of signal sequences, the adaptability and accuracy of the simulation model in different actual scenarios are improved.

[0106] In an optional embodiment, based on the optimal excitation source, a time-domain finite element method is used to dynamically adjust the time step of the simulation calculation through an adaptive time step control strategy, and the transient multi-physical field distribution of the target chip is calculated and obtained, including:

[0107] Based on the multi-physics field coupled algebraic equations, a time domain term is introduced to construct a time domain finite element model, and a time stepping algorithm is used to calculate the multi-physics field distribution of the internal nodes of the target chip at each time step;

[0108] Based on the severity of changes in multi-physics fields, combined with an adaptive time step control strategy, the time step of the time domain solution is dynamically adjusted through error estimation and iterative correction;

[0109] Based on the hybrid parallel mode, the time-domain finite element model is divided into multiple sub-regions, in which the communication synchronization between sub-regions is realized through the message passing interface, and the multi-threaded parallel execution mechanism is used to achieve multi-threaded parallelism within the sub-region.

[0110] On the basis of the multi-physics field coupled algebraic equations, the time derivative term is introduced to expand the static model into a time domain model; the time-domain finite element model is solved by the time-stepping algorithm, and the total simulation time is divided into multiple discrete time steps; within each time step, the space domain is discretized using the finite element method to obtain a discrete algebraic equation group; the algebraic equation group under the current time step is solved to obtain the electromagnetic field, thermal field and stress field distribution of the internal nodes of the chip; the discretization and solution process is repeated, and the time is gradually advanced until the preset total simulation time is reached.

[0111] According to the severity of the multi-physics field changes, the applicability of the current time step is evaluated; an error estimation mechanism is introduced to estimate the time discretization error by comparing the sub-solutions of different stages; based on the error estimation results, it is determined whether the time step needs to be adjusted. If the error exceeds the threshold, the step size is reduced; if the error is much smaller than the threshold, the step size is increased; through iterative correction, the time step is continuously updated until the error requirements are met; the adaptive time step control strategy is combined with the time-domain finite element solution, and the step size is dynamically adjusted within each time step according to the changes in the field distribution.

[0112] The time-domain finite element model is divided into multiple sub-regions, taking into account the balance of computing load and the minimization of communication overhead; the message passing interface (MPI) is used to achieve communication synchronization between sub-regions, and each sub-region is assigned to an independent process; in each sub-region, the open multi-threaded (OpenMP) parallel execution mechanism is used to assign the finite element calculation tasks of the sub-region to multiple threads; the boundary information is exchanged between sub-regions through MPI to achieve global field distribution synchronization; OpenMP is used to execute computing tasks in parallel with multiple threads in the sub-region to accelerate local calculations; the hybrid parallel mode is combined with the time-domain finite element solution to complete parallel calculations within each time step, thereby promoting the solution of the field distribution in the entire time domain.

[0113] In this embodiment, by expanding the static model into a time-domain model and performing finite element solution in each time step, the dynamic evolution process is successfully simulated to capture more detailed physical changes; by introducing an error estimation mechanism and an adaptive time step adjustment strategy, the simulation accuracy is guaranteed while optimizing the computational efficiency. When the field distribution changes dramatically, the step length is automatically shortened to improve the accuracy; when the change is gentle, the step length is extended to save computing resources; a hybrid parallel mode is adopted to divide the time-domain finite element solution task into multiple sub-regions, and multi-threaded parallel computing is implemented in each sub-region, while communication synchronization between regions is achieved through MPI to improve computing efficiency; by using MPI to allocate sub-region tasks globally and using OpenMP to execute computing tasks in parallel locally, the computing load is balanced and the communication overhead is minimized, which accelerates the overall simulation process.

[0114] In an optional embodiment, by post-processing the transient multi-physical field distribution, extracting the time-frequency domain multi-scale distribution features of electromagnetic interference, using a deep convolutional neural network to learn and classify the time-frequency domain multi-scale distribution features, constructing a spectrum fingerprint library of electromagnetic interference, and generating electromagnetic interference identification results including electromagnetic interference type and interference severity, including:

[0115] Perform short-time Fourier transform on transient multi-physical field distribution to obtain time-frequency representation of multi-physical fields, construct time-frequency spectrum, extract local texture features at different time and frequency scales from the time-frequency spectrum based on preset multi-scale convolution kernel, and determine multi-scale distribution characteristics of electromagnetic interference in time-frequency domain;

[0116] Construct a deep convolutional neural network model with multi-scale distribution features in the time-frequency domain as input, optimize the network parameters of the deep convolutional neural network model through end-to-end supervised learning, and combine the attention mechanism to determine the sensitivity of the deep convolutional neural network model to each time-frequency domain to classify the electromagnetic interference type;

[0117] Based on the trained deep convolutional neural network model, a preset set of electromagnetic interference samples is input, the time-frequency domain features corresponding to each electromagnetic interference sample are extracted, and mapped to the corresponding electromagnetic interference type, the corresponding relationship between the electromagnetic interference type and the time-frequency domain features is determined, and a spectrum fingerprint library is constructed;

[0118] Combined with the interference severity assessment module, the severity assessment features are extracted, including interference energy, interference frequency distribution and interference duration, and combined with the electromagnetic interference type to generate electromagnetic interference identification results.

[0119] The transient multi-physics field distribution inside the chip is sampled to obtain a discrete data sequence in the time domain; the time domain data sequence is subjected to a short-time Fourier transform to convert the time domain signal into a time-frequency domain representation. The sliding window method is used to divide overlapping time slices in the time domain, and each time slice is subjected to a Fourier transform to obtain the spectrum information within the time slice; the spectrum of different time slices is spliced ​​to construct a time-frequency spectrum of the multi-physics field. The horizontal axis of the time-frequency spectrum represents time, the vertical axis represents frequency, and the pixel value represents the signal strength at the corresponding time and frequency; a set of multi-scale convolution kernels are designed to cover different time and frequency scales. The size and shape of the convolution kernel are selected according to the typical characteristics of electromagnetic interference to capture local time-frequency domain texture information; the multi-scale convolution kernel is used to perform convolution operations on the time-frequency spectrum to extract local texture features at different time and frequency scales. The convolution operation locally scans the time-frequency spectrum through the sliding convolution kernel to obtain the feature response at each position; the convolution result is pooled to reduce the spatial dimension of the feature while retaining significant time-frequency domain features. The pooling operation can use maximum pooling or average pooling to extract significant responses in local areas; the extracted multi-scale features in the time-frequency domain are combined to form a time-frequency domain distribution feature vector of electromagnetic interference. This feature vector contains local texture information at different time and frequency scales, and describes the multi-scale distribution characteristics of electromagnetic interference in the time-frequency domain.

[0120] The architecture of the deep convolutional neural network is designed, including convolutional layers, pooling layers, and fully connected layers. The convolutional layer is responsible for extracting local features in the time-frequency domain, the pooling layer reduces the feature dimension, and the fully connected layer performs feature combination and classification decisions; the multi-scale distribution features in the time-frequency domain are used as the input of the network, and the feature information is transmitted and abstracted layer by layer through the forward propagation process. The convolutional layer extracts features of different scales and directions through convolution operations, the pooling layer reduces the spatial resolution of the features, and the fully connected layer flattens and combines the features; a classification layer is added at the end of the network, and the features are mapped to the probability distribution of different electromagnetic interference types through the softmax activation function. The output dimension of the classification layer is consistent with the number of electromagnetic interference types; an end-to-end supervised learning method is used to train the network using labeled electromagnetic interference samples. The weight parameters of the network are optimized by minimizing the cross entropy loss function between the predicted type and the true type; during the training process, an attention mechanism is introduced to dynamically adjust the network's attention to different time-frequency domain features through the attention layer. The attention mechanism can adaptively allocate network resources according to task requirements to improve classification performance; through multiple rounds of iterative training, network parameters are continuously updated until the preset performance indicators or convergence conditions are reached. After the training is completed, an optimized deep convolutional neural network model is obtained, which can classify the electromagnetic interference type based on the input time-frequency domain features.

[0121] Collect and organize a set of pre-set electromagnetic interference samples, covering different electromagnetic interference types and intensities. The sample collection should be representative and diverse to ensure the completeness of the spectrum fingerprint library; extract time and frequency domain features for each electromagnetic interference sample, and use the trained deep convolutional neural network model to map the time and frequency domain features of the sample to the corresponding electromagnetic interference type; establish a correspondence between the electromagnetic interference type and the time and frequency domain features, and associate the time and frequency domain features of each electromagnetic interference sample with its corresponding type label to form a spectrum fingerprint record; organize all spectrum fingerprint records into a structured database to build a spectrum fingerprint library. Each record in the spectrum fingerprint library contains the time and frequency domain features and corresponding type labels of the electromagnetic interference sample; index and optimize the spectrum fingerprint library to improve the efficiency of retrieval and matching. Appropriate data structures and algorithms, such as hash tables and trees, can be used to accelerate the query and comparison process of spectrum fingerprints.

[0122] The electromagnetic interference signal to be identified is obtained, and its time-frequency domain features are extracted to obtain multi-scale distribution features in the time-frequency domain; the extracted time-frequency domain features are input into the trained deep convolutional neural network model, and the prediction result of the electromagnetic interference type is obtained through the forward propagation process; the interference severity assessment module is used to extract the interference energy, interference frequency distribution and interference duration and other features to quantify the severity of the interference; the electromagnetic interference type and severity assessment results are combined to generate the final electromagnetic interference identification result, including the interference type, severity and other information; the identification result is compared with the records in the spectrum fingerprint library to find the spectrum fingerprint most similar to the interference signal to be identified, obtain the corresponding interference type and feature information, and further verify and refine the identification result; the electromagnetic interference identification report is output, including detailed information such as interference type, severity, spectrum characteristics, etc., to provide a basis for subsequent interference analysis and processing.

[0123] In this embodiment, the multi-scale distribution characteristics of electromagnetic interference in the time and frequency domain are captured, which effectively reveals the behavioral characteristics of electromagnetic interference at different times and frequencies; by classifying and identifying the multi-scale characteristics in the time and frequency domain, efficient identification of electromagnetic interference types is achieved; a spectrum fingerprint library containing electromagnetic interference types and their corresponding time and frequency domain characteristics is established to achieve rapid retrieval and matching of electromagnetic interference types, providing an important basis for subsequent interference analysis and processing; in the identification process, by extracting information such as the energy, frequency distribution and duration of electromagnetic interference, the severity of the interference is quantified, so that the identification results are not limited to the interference type, but also include the degree of influence of the interference, providing key data support for the anti-interference design and optimization of the chip; by introducing the attention mechanism, the network can adaptively adjust the degree of attention to different features, thereby improving the classification performance. At the same time, multiple rounds of iterative optimization are performed during the training process to ensure that the model achieves the best recognition effect.

[0124] S103. Based on the electromagnetic interference identification result, the optimal suppression circuit topology is determined through the suppression topology search algorithm, and the optimal suppression device parameters are determined through the multi-objective optimization algorithm combined with the electromagnetic compatibility index constraints. Based on the optimal suppression circuit topology and the optimal suppression device parameters, an electromagnetic interference suppression strategy is generated.

[0125] According to the electromagnetic interference identification results, the characteristic parameters of the interference source are obtained, including interference frequency, interference intensity, interference mode, etc. A candidate set of suppression circuit topology is established, including common suppression circuit structures such as filters, limiters, absorbers, etc. The candidate set should cover various types of suppression circuits to provide a wide range of choices. The evaluation indicators of the suppression circuit topology are defined, such as suppression effect, circuit complexity, cost, etc. The evaluation indicators are used to measure the advantages and disadvantages of different suppression circuit topologies. The suppression topology search algorithm, such as genetic algorithm, particle swarm optimization algorithm, etc., is used to search for the optimal suppression circuit topology in the candidate set. During the search process, the candidate topologies are scored and sorted according to the evaluation indicators. During the search process, the suppression circuit topology is continuously updated and improved through iterative optimization until the set termination conditions are met, such as reaching the maximum number of iterations or finding the optimal topology that meets the requirements. The optimal suppression circuit topology obtained by the search is output.

[0126] Based on the optimal suppression circuit topology, the key components involved in the suppression circuit, such as resistors, capacitors, inductors, etc., are determined; the parameter range of each key component, such as the resistance value range, the capacitance value range, etc., is defined to form a parameter search space; a multi-objective optimization model is established, and the objective function includes multiple aspects such as suppression effect, circuit performance, and cost. The objective function is used to evaluate the advantages and disadvantages of different parameter combinations; the electromagnetic compatibility index is used as a constraint condition of the optimization model, such as electromagnetic interference suppression rate, signal integrity and other indicators. The constraint condition ensures that the parameter optimization meets the electromagnetic compatibility requirements; a multi-objective optimization algorithm, such as the non-dominated sorting genetic algorithm (NSGA-II) and the multi-objective particle swarm optimization algorithm (MOPSO), is used to find the optimal device parameter combination in the parameter search space; in the optimization process, the device parameters are continuously updated and improved through iterative search and non-dominated sorting until the set termination conditions are met, such as reaching the maximum number of iterations or finding the optimal parameter combination that meets the constraint conditions; the optimal suppression device parameters obtained by the search are output, and this parameter combination performs best under multiple objective functions while meeting the electromagnetic compatibility index constraints, and can effectively suppress electromagnetic interference.

[0127] Build a complete electromagnetic interference suppression scheme by combining the optimal suppression circuit topology and the optimal suppression device parameters; simulate and verify the suppression scheme, use electromagnetic simulation software to model and analyze the suppression circuit, and evaluate its suppression effect and performance under different working conditions; optimize and improve the suppression scheme based on the simulation results, adjust the circuit topology and device parameters until the electromagnetic compatibility requirements and performance indicators are met; generate a detailed electromagnetic interference suppression strategy document, including the suppression circuit topology, device parameter values, wiring layout guidelines, installation and debugging steps, etc.; provide implementation guidance and precautions for the suppression strategy, such as device selection, circuit board layout, grounding treatment, etc., to ensure the correct implementation and effectiveness of the suppression scheme; test and verify the suppression strategy, deploy the suppression scheme in the actual circuit, and conduct electromagnetic compatibility testing to evaluate its suppression effect in the real environment; make necessary corrections and optimizations to the suppression strategy based on the test results and feedback, and continuously improve and improve the suppression scheme to adapt to different application scenarios and interference environments.

[0128] In this embodiment, the acquisition of electromagnetic interference identification results improves the accuracy and comprehensiveness of identification, laying a solid foundation for the subsequent design of suppression circuits; the application of the suppression topology search algorithm effectively reduces the complexity of electromagnetic interference design. Through global search and iterative optimization, the circuit topology with the best suppression effect is found to ensure the practical feasibility of the suppression circuit; through the multi-objective optimization algorithm, the optimal suppression device parameters are determined to improve the suppression efficiency of the circuit and the reliability of the system, so that the suppression circuit can effectively cope with different electromagnetic interference situations; the final generated suppression strategy integrates the optimal topology and device parameters to ensure that electromagnetic interference can be effectively suppressed. After simulation verification and actual testing, the effectiveness of the strategy has been confirmed.

[0129] In an optional embodiment, based on the electromagnetic interference identification result, determining the optimal suppression circuit topology through a suppression topology search algorithm includes:

[0130] The suppression topology search algorithm is constructed based on the fruit fly optimization algorithm;

[0131] Performing quantum encoding on the topology of the inhibition circuit, mapping each device in the inhibition circuit to a quantum bit, wherein the state of the device corresponds to the probability amplitude of the quantum bit, and multiple quantum bits form a quantum chromosome, corresponding to a candidate solution of the inhibition circuit topology;

[0132] A set of quantum chromosomes is randomly generated, each quantum chromosome corresponds to a fruit fly individual, and the initial fruit fly population is determined;

[0133] Performing a taste search on the initial fruit fly population, observing the quantum bits according to the quantum chromosomes of the individual fruit flies, determining the inhibitory circuit topology determination solution corresponding to the inhibitory circuit topology candidate solution from the superposition state of the quantum bits, calculating the fitness corresponding to the inhibitory circuit topology determination solution, determining the initial optimal inhibitory circuit topology in the search space, and updating the probability amplitude of all quantum bits in the corresponding quantum chromosome;

[0134] Taking the initial optimal inhibitory circuit topology as the visual search center point, generating a new set of quantum chromosomes near the visual search center point according to a preset neighbor threshold, determining the visual search sample points, calculating the fitness corresponding to the visual search sample points, selecting the visual search sample points corresponding to the maximum fitness, determining the advanced optimal inhibitory circuit topology, and using it as the new visual search center point, while updating the probability amplitude of the quantum bit;

[0135] Based on the quantum revolving door operation, the probability amplitude of the quantum bit is adjusted and updated, and the rotation angle of the quantum revolving door is adaptively determined within a preset revolving door angle range, and the rotation angle is inversely proportional to the fitness;

[0136] Based on the advanced optimal inhibitory circuit topology and combined with random perturbations, a new generation of fruit fly populations was generated;

[0137] The iteration is repeated until the preset number of iterations is reached, and the advanced optimal inhibitory circuit topology obtained by the last taste search and visual search is used as the final optimal inhibitory circuit topology.

[0138] The search problem of the inhibitory circuit topology is transformed into an optimization problem of the fruit fly optimization algorithm. The fitness of the inhibitory circuit topology is used as the optimization target to find the optimal inhibitory circuit topology. The inhibitory circuit topology is encoded by quantum coding. Each device in the inhibitory circuit is mapped to a quantum bit, and the state of the device corresponds to the probability amplitude of the quantum bit. Multiple quantum bits form a quantum chromosome, which represents a candidate solution of the inhibitory circuit topology. The encoding scheme of the quantum bit is designed to determine the dimension and representation of the quantum bit. The different states of the device, such as the switch state, parameter value, etc., are represented by the superposition state of the quantum bit. The fitness function is defined to evaluate the advantages and disadvantages of the candidate solutions of the inhibitory circuit topology. The fitness function comprehensively considers factors such as the inhibitory effect, circuit complexity, and cost, and assigns a fitness value to each candidate solution.

[0139] A set of quantum chromosomes is randomly generated, each corresponding to a fruit fly individual. Each quantum bit in the quantum chromosome is in a different state with a certain probability amplitude. The generated quantum chromosomes are checked for legitimacy to ensure that they can correspond to a valid inhibitory circuit topology. The illegal quantum chromosomes are corrected or regenerated. The generated quantum chromosomes are formed into an initial fruit fly population, and the size of the population is determined according to the complexity of the problem and the size of the search space. For each individual in the initial fruit fly population, the quantum bits are observed according to their quantum chromosomes to obtain a definite candidate solution for the inhibitory circuit topology. The generation of candidate solutions follows the probability distribution of quantum bits.

[0140] For each individual in the initial fruit fly population, according to its quantum chromosome, by observing the quantum bits, a definite solution of the inhibitory circuit topology is determined from the superposition state of the quantum bits. The definite solution of the inhibitory circuit topology is converted into the corresponding circuit structure and parameter settings to construct the actual inhibitory circuit. The fitness value of each definite solution of the inhibitory circuit topology is calculated to evaluate its inhibitory effect and performance. The fitness values ​​of all definite solutions are compared, and the individual with the largest fitness is selected to determine the initial optimal inhibitory circuit topology. According to the initial optimal inhibitory circuit topology, the probability amplitude of all quantum bits in the corresponding quantum chromosome is updated. The probability amplitude of the quantum bit state that leads to the optimal solution is increased, and the probability amplitude of other states is reduced.

[0141] The initial optimal inhibitory circuit topology is used as the center point of the visual search, and a set of new quantum chromosomes are generated within a certain range near it as sample points for the visual search. For each visual search sample point, the corresponding inhibitory circuit topology is determined by observing the quantum bits. The fitness value of each visual search sample point is calculated to evaluate its inhibitory effect and performance. The fitness values ​​of all visual search sample points are compared, and the sample point with the largest fitness is selected to determine the advanced optimal inhibitory circuit topology. The advanced optimal inhibitory circuit topology is used as the new visual search center point, and the probability amplitude of the quantum bits in the corresponding quantum chromosome is updated to increase the probability amplitude of the quantum bit state that leads to the optimal solution.

[0142] According to the fitness value, the rotation angle of the quantum revolving gate is determined. The rotation angle is inversely proportional to the fitness. The greater the fitness, the smaller the rotation angle; the smaller the fitness, the larger the rotation angle. Apply the quantum revolving gate operation to the quantum bit, and adjust the probability amplitude of the quantum bit according to the rotation angle. The larger the rotation angle, the greater the adjustment of the probability amplitude. Through the quantum revolving gate operation, the probability amplitude of the quantum bit state leading to the optimal solution is increased, and the probability amplitude of other states is reduced, guiding the search in a more optimal direction. The rotation angle of the quantum revolving gate is adaptively adjusted in each iteration. As the search proceeds, the rotation angle gradually decreases to achieve a refined search.

[0143] Based on the advanced optimal inhibitory circuit topology, a new set of quantum chromosomes is generated near it by random perturbation. Random perturbations include slight changes to the state of quantum bits, such as slight adjustments to the probability amplitude, to introduce new search directions and diversity. The generated new quantum chromosomes are checked for legitimacy to ensure that they can correspond to valid inhibitory circuit topologies. The newly generated quantum chromosomes are merged with individuals in the original population to form a new generation of fruit fly populations in preparation for the next round of search.

[0144] The taste search, visual search, and quantum revolving door operations are repeatedly performed to continuously update and improve the inhibitory circuit topology. In each iteration, the search strategy and parameters, such as the visual search range and the quantum revolving door angle, are dynamically adjusted according to the change in fitness value to adapt to the progress of the search. The iterative process continues until the preset number of iterations is reached or the convergence condition is met.

[0145] In the last iteration, the advanced optimal inhibitory circuit topology obtained by taste search and visual search is taken as the final optimal solution.

[0146] In this embodiment, by combining the fruit fly optimization algorithm with quantum coding, an efficient global search can be performed in a huge search space. The superposition state representation of quantum bits enables the algorithm to explore multiple possible inhibitory circuit topologies at the same time, improve the globality and diversity of the search process, reduce the possibility of falling into a local optimal solution, and automatically generate and screen the optimal circuit topology structure, greatly reducing the design complexity and manpower investment; by adaptively adjusting the rotation angle of the quantum revolving door and the visual search range and other parameters, the algorithm can dynamically adjust the search strategy according to the fitness change in the search process, realize the gradual advancement from rough search to fine search, and improve the accuracy and optimization effect of the final solution; the quantum revolving door operation continuously adjusts the probability amplitude of the quantum bit, so that the search process can gradually focus on the optimal solution, thereby accelerating the convergence speed of the algorithm and reducing the waste of computing resources; through multiple rounds of taste search, visual search and random perturbation, the inhibitory circuit topology finally obtained reaches the global optimum after multiple optimization iterations, so that the inhibitory circuit can effectively deal with complex electromagnetic interference problems in practical applications.

[0147] Figure 2 FIG. 1 is a schematic diagram of a structure of an electromagnetic interference suppression system for a chip according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0148] The first unit is used to obtain layout data of a target chip whose electromagnetic interference is to be suppressed, extract structural parameters of each functional module in the target chip and device parameters of corresponding devices through an image segmentation algorithm based on the layout data, and construct a multi-physics field coupling simulation model including the overall structure of the target chip and each device model based on the structural parameters and the device parameters;

[0149] The second unit is used to determine the optimal excitation source corresponding to the target chip based on the multi-physics field coupling simulation model through a machine learning algorithm, and based on the optimal excitation source, use a time-domain finite element method to dynamically adjust the time step of the simulation calculation through an adaptive time step control strategy to calculate and obtain the transient multi-physics field distribution of the target chip, extract the multi-scale distribution characteristics of the electromagnetic interference in the time and frequency domains by post-processing the transient multi-physics field distribution, use a deep convolutional neural network to learn and classify the multi-scale distribution characteristics in the time and frequency domains, build a spectrum fingerprint library of electromagnetic interference, and generate an electromagnetic interference identification result including the type of electromagnetic interference and the severity of interference;

[0150] The third unit is used to determine the optimal suppression circuit topology based on the electromagnetic interference identification result through the suppression topology search algorithm, determine the optimal suppression device parameters through the multi-objective optimization algorithm combined with the electromagnetic compatibility index constraints, and generate an electromagnetic interference suppression strategy based on the optimal suppression circuit topology and the optimal suppression device parameters.

[0151] According to a third aspect of the embodiments of the present invention,

[0152] An electronic device is provided, comprising:

[0153] processor;

[0154] a memory for storing processor-executable instructions;

[0155] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0156] A fourth aspect of the embodiments of the present invention is:

[0157] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0158] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for suppressing electromagnetic interference of a chip, characterized in that: include: Acquire layout data of a target chip whose electromagnetic interference is to be suppressed, extract structural parameters of each functional module in the target chip and device parameters of corresponding devices through an image segmentation algorithm based on the layout data, and construct a multi-physics field coupling simulation model including the overall structure of the target chip and each device model based on the structural parameters and the device parameters; Based on the multi-physics field coupling simulation model, the optimal excitation source corresponding to the target chip is determined through a machine learning algorithm. Based on the optimal excitation source, a time-domain finite element method is used to dynamically adjust the time step of the simulation calculation through an adaptive time step control strategy to calculate and obtain the transient multi-physics field distribution of the target chip. By post-processing the transient multi-physics field distribution, the multi-scale distribution characteristics of the electromagnetic interference in the time and frequency domains are extracted. The multi-scale distribution characteristics in the time and frequency domains are learned and classified using a deep convolutional neural network to construct a spectrum fingerprint library of electromagnetic interference, and generate an electromagnetic interference identification result including the electromagnetic interference type and interference severity. Based on the electromagnetic interference identification result, the optimal suppression circuit topology is determined by a suppression topology search algorithm, the optimal suppression device parameters are determined by a multi-objective optimization algorithm combined with electromagnetic compatibility index constraints, and an electromagnetic interference suppression strategy is generated based on the optimal suppression circuit topology and the optimal suppression device parameters; Acquiring layout data of a target chip for suppressing electromagnetic interference, extracting structural parameters of each functional module in the target chip and device parameters of corresponding devices through an image segmentation algorithm based on the layout data, and constructing a multi-physics field coupling simulation model including the overall structure of the target chip and each device model based on the structural parameters and the device parameters, including: Obtain the layout data file of the target chip; Converting the layout data file into a layout image, encoding the layout image into a quantum state, extracting multi-scale texture features and topological structure features of the layout image in the corresponding space of the quantum state based on quantum Fourier transform and quantum phase estimation algorithm, determining the layout features, constructing a similarity measurement model based on quantum gate circuits, calculating the similarity of different layout images in the feature space through entanglement measurement between quantum states, retrieving a layout similar to the layout image in a preset layout database using a quantum search algorithm, obtaining candidate areas for layout segmentation and parameter extraction, and extracting geometric structure parameters and device parameters of each functional module in the candidate areas; Based on the geometric structure parameters, a three-dimensional geometric model of the target chip is constructed, and based on the device parameters, a device model is constructed, and the corresponding device model is placed in the three-dimensional geometric model to establish a multi-physical field coupling simulation model.

2. The method according to claim 1, characterized in that: The multi-physics field coupling simulation model also includes: The multi-physics field coupling simulation model is discretized by using an adaptive meshing method; Adaptively adjust the grid density and size according to the electromagnetic field intensity distribution and device position distribution in different areas of the target chip; In combination with multi-layer mesh nesting technology, the multi-physics field coupling simulation model is meshed at different scales to determine the overall electromagnetic interference characteristics and local electromagnetic distribution; The grid division is based on the coupling relationship between different physical fields to determine a unified grid division scheme.

3. The method according to claim 1, characterized in that Based on the multi-physics field coupling simulation model, the optimal excitation source corresponding to the target chip is determined by a machine learning algorithm, including: Based on the multi-physics field coupling simulation model, a group of partial differential equations describing the coupling effects of the electromagnetic field, thermal field and stress field inside the chip is constructed, and the group of partial differential equations is discretized using a preset finite element method to obtain a group of multi-physics field coupling algebraic equations; The multi-physics field coupling algebraic equation group is converted into an optimization model with the excitation source as the design variable, and based on the machine learning algorithm, the excitation source parameter combination that makes the multi-physics field distribution optimal is obtained through iterative search to obtain the optimal excitation source; The optimal excitation source is input into an excitation source signal generation module to synthesize an excitation signal sequence matching the actual scene, and the distribution characteristics corresponding to the optimal excitation source are maintained in combination with a preset data enhancement method to expand the excitation source samples.

4. The method according to claim 3, characterized in that Based on the optimal excitation source, the time domain finite element method is used to dynamically adjust the time step of the simulation calculation through an adaptive time step control strategy to calculate the transient multi-physical field distribution of the target chip, including: Based on the multi-physics field coupled algebraic equations, a time domain term is introduced to construct a time domain finite element model, and a time stepping algorithm is used to calculate the multi-physics field distribution of the internal nodes of the target chip at each time step; Based on the severity of changes in multi-physics fields, combined with an adaptive time step control strategy, the time step of the time domain solution is dynamically adjusted through error estimation and iterative correction; Based on the hybrid parallel mode, the time-domain finite element model is divided into multiple sub-regions, in which the communication synchronization between sub-regions is realized through the message passing interface, and the multi-threaded parallel execution mechanism is used to achieve multi-threaded parallelism within the sub-region.

5. The method according to claim 4, characterized in that By post-processing the transient multi-physical field distribution, the multi-scale distribution features of electromagnetic interference in the time and frequency domain are extracted, and the multi-scale distribution features in the time and frequency domain are learned and classified using a deep convolutional neural network to build a spectrum fingerprint library of electromagnetic interference. The electromagnetic interference identification results containing the electromagnetic interference type and interference severity are generated, including: Perform short-time Fourier transform on transient multi-physical field distribution to obtain time-frequency representation of multi-physical fields, construct time-frequency spectrum, extract local texture features at different time and frequency scales from the time-frequency spectrum based on preset multi-scale convolution kernel, and determine multi-scale distribution characteristics of electromagnetic interference in time-frequency domain; Construct a deep convolutional neural network model with multi-scale distribution features in the time-frequency domain as input, optimize the network parameters of the deep convolutional neural network model through end-to-end supervised learning, and combine the attention mechanism to determine the sensitivity of the deep convolutional neural network model to each time-frequency domain to classify the electromagnetic interference type; Based on the trained deep convolutional neural network model, a preset set of electromagnetic interference samples is input, the time-frequency domain features corresponding to each electromagnetic interference sample are extracted, and mapped to the corresponding electromagnetic interference type, the corresponding relationship between the electromagnetic interference type and the time-frequency domain features is determined, and a spectrum fingerprint library is constructed; Combined with the interference severity assessment module, the severity assessment features are extracted, including interference energy, interference frequency distribution and interference duration, and combined with the electromagnetic interference type to generate electromagnetic interference identification results.

6. The method according to claim 1, characterized in that Based on the electromagnetic interference identification results, the optimal suppression circuit topology is determined through the suppression topology search algorithm, including: The suppression topology search algorithm is constructed based on the fruit fly optimization algorithm; Performing quantum encoding on the topology of the inhibition circuit, mapping each device in the inhibition circuit to a quantum bit, wherein the state of the device corresponds to the probability amplitude of the quantum bit, and multiple quantum bits form a quantum chromosome, corresponding to a candidate solution of the inhibition circuit topology; A set of quantum chromosomes is randomly generated, each quantum chromosome corresponds to a fruit fly individual, and the initial fruit fly population is determined; Performing a taste search on the initial fruit fly population, observing the quantum bits according to the quantum chromosomes of the individual fruit flies, determining the inhibitory circuit topology determination solution corresponding to the inhibitory circuit topology candidate solution from the superposition state of the quantum bits, calculating the fitness corresponding to the inhibitory circuit topology determination solution, determining the initial optimal inhibitory circuit topology in the search space, and updating the probability amplitude of all quantum bits in the corresponding quantum chromosome; Taking the initial optimal inhibitory circuit topology as the visual search center point, generating a new set of quantum chromosomes near the visual search center point according to a preset neighbor threshold, determining the visual search sample points, calculating the fitness corresponding to the visual search sample points, selecting the visual search sample points corresponding to the maximum fitness, determining the advanced optimal inhibitory circuit topology, and using it as the new visual search center point, while updating the probability amplitude of the quantum bit; Based on the quantum revolving door operation, the probability amplitude of the quantum bit is adjusted and updated, and the rotation angle of the quantum revolving door is adaptively determined within a preset revolving door angle range, and the rotation angle is inversely proportional to the fitness; Based on the advanced optimal inhibitory circuit topology and combined with random perturbations, a new generation of fruit fly populations was generated; The iteration is repeated until the preset number of iterations is reached, and the advanced optimal inhibitory circuit topology obtained by the last taste search and visual search is used as the final optimal inhibitory circuit topology.

7. An electromagnetic interference suppression system for a chip, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to obtain layout data of a target chip whose electromagnetic interference is to be suppressed, extract structural parameters of each functional module in the target chip and device parameters of corresponding devices through an image segmentation algorithm based on the layout data, and construct a multi-physics field coupling simulation model including the overall structure of the target chip and each device model based on the structural parameters and the device parameters; The second unit is used to determine the optimal excitation source corresponding to the target chip based on the multi-physics field coupling simulation model through a machine learning algorithm, and based on the optimal excitation source, use a time-domain finite element method to dynamically adjust the time step of the simulation calculation through an adaptive time step control strategy to calculate and obtain the transient multi-physics field distribution of the target chip, extract the multi-scale distribution characteristics of the electromagnetic interference in the time and frequency domains by post-processing the transient multi-physics field distribution, use a deep convolutional neural network to learn and classify the multi-scale distribution characteristics in the time and frequency domains, build a spectrum fingerprint library of electromagnetic interference, and generate an electromagnetic interference identification result including the type of electromagnetic interference and the severity of interference; The third unit is used to determine the optimal suppression circuit topology based on the electromagnetic interference identification result through the suppression topology search algorithm, determine the optimal suppression device parameters through the multi-objective optimization algorithm combined with the electromagnetic compatibility index constraints, and generate an electromagnetic interference suppression strategy based on the optimal suppression circuit topology and the optimal suppression device parameters.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images