Electromagnetic simulation grid density prediction method and system

Abstract

The invention provides an electromagnetic simulation grid density prediction method and system. Collecting geometric structure data of the radio frequency integrated circuit and a corresponding mesh generation result; extracting a geometric contour according to the geometric structure data, and then converting the geometric contour into a geometric binary image; predicting a grayscale image of the geometric binary image by using a deep learning model; and acquiring the electromagnetic simulation grid density according to the predicted grayscale image. According to the method, high-density grids are automatically distributed in a special mark area with violent electromagnetic field intensity change, so that the simulation precision is ensured; a low-density grid is adopted in an electromagnetic field distribution gentle parameterization area, and the number of redundant units is reduced; moreover, the topological information of the original geometric structure is reserved, and the local grid density demand is visually reflected through the gray gradient, so that the deep learning model can directly utilize the local receptive field characteristics of the deep learning network to efficiently capture the complex space correlation between the geometric contour and the grid density.

Description

Technical Field

[0001] The present invention belongs to the technical field of electromagnetic simulation of intelligent integrated circuits, and relates to an electromagnetic simulation grid density prediction method and a system thereof. Background Art

[0002] Electromagnetic simulation is a core technology in modern electronic design, widely used in fields such as intelligent integrated circuits, radio frequency integrated circuits, antenna design, and electromagnetic compatibility analysis. Its core goal is to simulate electromagnetic field distribution through numerical calculations, providing a basis for design optimization. Mesh generation is a critical step in the electromagnetic simulation process, directly impacting simulation accuracy, computational efficiency, and result reliability. The quality of the mesh determines the ability to accurately capture the electromagnetic characteristics of complex geometric structures. This is particularly true in critical areas such as boundary layers and microstructures, where mesh density is extremely sensitive.

[0003] Traditional mesh generation methods usually adopt a uniform mesh division strategy. Although uniform meshes are easy to implement, they have significant inherent defects: in critical areas with high geometric complexity or drastic changes in field strength (such as conductor edges and dielectric interfaces), uniform meshes may cause simulation accuracy to decrease due to insufficient resolution and fail to accurately capture local physical phenomena; in non-critical areas, overly refined meshes will introduce redundant calculations, significantly increasing computing time and resource consumption. In addition, traditional methods rely on manual experience to adjust local mesh density, resulting in a cumbersome process and low degree of automation, making it difficult to adapt to the needs of large-scale complex designs. Additional steps are required in the post-processing stage to correct the mesh distribution, further increasing the engineering burden.

[0004] In recent years, deep learning technology has provided new insights into mesh generation. Deep learning models can learn the mapping relationship between complex geometric features and mesh density from data, and in theory can automatically generate adaptive meshes. However, existing technologies face two major bottlenecks: First, high-quality annotated mesh datasets are scarce. Mesh data in the field of electromagnetic simulation requires a combination of physical and geometric characteristics. The annotation cost is high and difficult to obtain, resulting in insufficient model training data. Second, complex deep learning models (such as multi-layer convolutional networks) are prone to overfitting in small sample scenarios, with limited generalization capabilities, making it difficult to adapt to diverse geometric structures and electromagnetic scenarios.

[0005] Therefore, due to the contradiction between the accuracy and efficiency of uniform grids and the data and generalization bottlenecks of deep learning in grid generation, it has been impossible to form an efficient and high-precision adaptive grid generation strategy. Summary of the Invention

[0006] The purpose of the present invention is to address the deficiencies of the existing technology and provide an electromagnetic simulation grid density prediction method and system. By mapping the area information of the grid unit into a grayscale image and using a deep learning network to realize the prediction from geometric contour to grid density, the shortcomings of traditional grid generation methods in accuracy and efficiency are solved.

[0007] In a first aspect, the present invention provides a method for predicting electromagnetic simulation grid density, the method comprising:

[0008] Collect the geometric structure data of the RFIC and the corresponding mesh generation results;

[0009] Extracting geometric contours according to geometric structure data, and then converting the geometric contours into geometric binary images;

[0010] The grid cell area in the grid generation result is mapped into a grayscale image through normalization processing; wherein the grayscale image includes the device area covered by the circuit and the non-simulation area not covered by the circuit, and the device in the device area includes a special marking area and a parameterized area; the special marking area, the parameterized area, and the non-simulation area are distinguished by different grayscale values;

[0011] Construct a data set based on geometric binary images and grayscale images;

[0012] Construct a deep learning model and use the data set for training and testing; wherein the input of the deep learning model is a geometric binary image and the output is a corresponding grayscale image;

[0013] The trained and tested deep learning model is used to predict the corresponding grayscale image of the geometric binary image of the RF integrated circuit; the electromagnetic simulation grid density is obtained based on the predicted grayscale image.

[0014] Preferably, the other conductive structure is a functional unit in a radio frequency integrated circuit.

[0015] Preferably, the special marking area includes ports and vias; the parameterized area includes the dielectric layer and other conductive structure areas; wherein the dielectric layer and other conductive structure areas are filled with different grayscale values according to the area of the grid unit; the via area is filled with a grayscale value of 255, and the port area is filled with a grayscale value of 254; the non-simulation area not covered by the circuit is filled with a grayscale value of 0.

[0016] More preferably, the grayscale value filling of the dielectric layer and other conductive structure areas is as follows:

[0017] Preferably, the grayscale value range of the dielectric layer and other conductive structure areas in the device area is [1,200].

[0018] Preferably, the process of mapping the grid unit area in the grid subdivision result into a grayscale image through normalization processing includes:

[0019] First, the grid cell area is normalized, and then the original coordinates are scaled proportionally. The scaling factor is the smaller ratio of the target size to the original size. The grayscale value 0 area is symmetrically filled in the direction of the short side of the scaled area.

[0020] A grayscale image is generated based on the normalized value of each grid cell area and pixel mapping rules.

[0021] Preferably, the deep learning model is a ladder-type U-Net, whose input is a binary image of the geometric contour and whose output is a corresponding grayscale image reflecting the size of the grid unit.

[0022] Preferably, data enhancement is also performed on the dataset input into the deep learning model.

[0023] In a second aspect, the present invention provides an electromagnetic simulation grid density prediction system, comprising:

[0024] The data acquisition module is responsible for collecting the geometric structure data of the RFIC; extracting the geometric contour based on the geometric structure data, and then converting the geometric contour into a geometric binary image;

[0025] The density prediction module is responsible for using the trained and tested deep learning model to predict the corresponding grayscale image for the geometric binary image; the electromagnetic simulation grid density is obtained based on the predicted grayscale image.

[0026] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method described above.

[0027] In a fourth aspect, the present invention provides a computing device comprising a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the method described is implemented.

[0028] This paper proposes a method for predicting electromagnetic simulation mesh density that combines image processing and deep learning techniques. This method transforms the complex mesh generation problem into an image-to-image mapping task, significantly improving the automation level of mesh generation. Compared with existing technologies, this paper has at least the following advantages:

[0029] (1) Improving the model's ability to model the spatial relationship between geometric features and grid density: This method converts the grid cell size into a grayscale image and marks special marked areas (such as vias and ports) with specific grayscale values (254, 255), converting the grid density distribution into image data with clear physical meaning. This representation method not only retains the topological information of the original geometric structure but also intuitively reflects the local grid density requirements through grayscale gradients. This allows deep learning models to directly utilize the local receptive field characteristics of deep learning networks to efficiently capture the complex spatial relationship between geometric contours and grid density.

[0030] (2) Achieve coordinated optimization of computing efficiency and resource utilization: The present invention automatically allocates high-density grids in special marked areas where the electromagnetic field intensity changes dramatically to ensure simulation accuracy; and uses low-density grids in parameterized areas where the electromagnetic field distribution is smooth to reduce the number of redundant units.

[0031] (2) Efficiency: This invention establishes an end-to-end process of "geometry input - model prediction - mesh generation" through standardized grayscale image generation and reverse mapping mechanisms. The predicted grayscale image can be directly visualized, and engineers can quickly verify the rationality of the mesh distribution by adjusting the grayscale threshold, reducing the cost of manual corrections later. In addition, the invention can also accurately restore the mesh coordinates based on the predicted grayscale image, ensuring compatibility with simulation tools.

[0032] (3) Flexibility: By converting grid data into image format, the present invention increases the flexibility of model selection, enabling the use of existing powerful image classification and object detection models without having to rebuild a specific model for grid data. This flexibility enables the present invention to adapt to different application scenarios and needs.

[0033] (4) Resource Conservation: By predicting the grid cell size, this method avoids using too small grid cells in unimportant areas, effectively reducing computational effort and resource consumption. This not only improves computational efficiency but also reduces the demand for computing resources, making it suitable for large-scale simulation tasks.

[0034] (5) Reduced time and cost: This invention predicts the mesh cell size at an early stage, which can reduce the need for mesh modification and recalculation later, thus saving time and cost. By automating the mesh generation process, engineers can focus more on design and optimization rather than on tedious mesh processing.

[0035] (6) Strong adaptability: The technology of the present invention can adapt to different types of objects and complex geometric shapes, and has a wide range of application prospects. Whether in aerospace, automotive design or electromagnetic simulation of electronic equipment, it can play an important role. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Schematic diagram of the flow of the electromagnetic simulation grid density prediction method of the present invention;

[0037] Figure 2 Schematic diagram of the process of generating a grayscale image in the method of the present invention;

[0038] Figure 3 The following are comparison diagrams of examples of converting grid cells into grayscale images in the present invention, where (a) is the grid division result and (b) is the grayscale image;

[0039] Figure 4 This is the model architecture diagram of Staircase U-Net in the present invention;

[0040] Figure 5 is the training loss function of the Staircase U-Net model in the present invention. DETAILED DESCRIPTION

[0041] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0042] The terms "including," "having," and any variations thereof, as used in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to the process, method, product, or apparatus.

[0043] This embodiment provides a method for predicting the density of electromagnetic simulation grids, which converts the grid cell size into a grayscale image to intuitively display the density and distribution of the grid cells, thereby improving the model's understanding and processing capabilities of the spatial relationship of the data, and solving the problems of insufficient accuracy and low computational efficiency of traditional grid generation. It includes data preparation, grayscale image generation, model training, and using the trained model to generate grayscale images. In the data preparation stage, the grid division results of the two-dimensional object are collected, and the grid cell size is standardized. When generating a grayscale image, the standardized value is mapped to the image pixel according to the grid cell position, and the size is adjusted and normalized. During the model training process, the input and output are designed, and data enhancement is performed to improve the generalization ability of the model. Finally, after the two-dimensional object is input, the model outputs the corresponding grayscale image and performs reverse mapping according to the size of the original object. See Appendix Figure 1 , specifically including:

[0044] Step S1: Data acquisition

[0045] Collect the geometric structure data of RF integrated circuits (such as filters and power dividers) and the corresponding mesh generation results, see the attached Figure 3 (a)

[0046] For example, the geometric structure data of this embodiment includes: a set of coordinates of the metal layer polygon vertices {(x i ,y i )|i=1,2,...,N}, dielectric layer profile and thickness, via position and size, port position, etc.

[0047] For example, the mesh generation result of this embodiment includes: a mesh unit area set {(A j )|j=1,2,...,M}, each grid cell corresponds to the material attribute mark C j ∈{metal, dielectric};

[0048] Step S2: extracting a geometric outline according to the geometric structure data, and then converting the geometric outline into a geometric binary image;

[0049] For example, this embodiment can create a blank image W×H, initialize a completely black background (pixel value 0), convert the acquired geometric structure data into a geometric contour, fill the metal layer area with a pixel value of 200, and obtain a geometric binary image; the position information of special marking areas (such as vias and ports) is retained through coordinate mapping.

[0050] Step S3: Generate a grayscale image corresponding to the binary image. Figure 2

[0051] S3-1 normalizes the area of the grid cells in the grid generation results;

[0052] The calculation is performed using the following standardized formula:

[0053]

[0054] where s j Represents the normalized value of the j-th grid cell area, A j represents the area of the jth grid cell, A min Represents the minimum area of the grid cell, A max Indicates the maximum area of the grid cell; Q j represents the j-th grid cell.

[0055] This method collects the meshing results of representative two-dimensional objects and standardizes the size of each grid cell. By mapping the grid cell size to a grayscale value range of 1 to 200, using 254 and 255 as special markers, this data is used as input for subsequent model training. This process ensures data consistency and usability, providing high-quality training samples for deep learning models.

[0056] S3-2 generates a grayscale image, which intuitively reflects the grid requirements of different areas and facilitates the subsequent deep learning model processing. This embodiment uses coordinate information to construct an image to ensure that the generated grayscale image (see Appendix Figure 3 (b) can accurately reflect the geometric characteristics of the object.

[0057] This embodiment extracts the coordinate information of each grid cell in the grid; performs proportional scaling and filling; and maps the standardized area value of each grid cell to the image pixels to generate a grayscale image that can reflect the grid density of different areas. Specifically:

[0058] 3-2-1 Get the coordinates of all vertices in all grid cells and extract the maximum horizontal coordinate x max , maximum vertical coordinate y max , minimum horizontal coordinate x min , minimum ordinate y min , calculate the minimum bounding box of the geometric contour;

[0059] 3-2-2 Perform proportional scaling and filling:

[0060] Set the target length and width of the grayscale image to W×H and calculate the scaling factor s:

[0061]

[0062] Add padding area Δ on both sides of the short side of the minimum bounding box of the geometric outline J , the fill value is set to 0;

[0063] Fill area Δ J for:

[0064]

[0065] 3-2-3 Establish a normalized coordinate system:

[0066]

[0067] Where (x i 、y i ) represents the coordinates of the i-th vertex;

[0068] 3-2-4 pixel mapping rule:

[0069]

[0070] Where floor() represents the rounding down function; (x pix ,y pix ) means (x i 、y i )’s mapped pixel coordinates;

[0071] 3-2-5 Standardized value s according to the area of each grid unit j , and pixel mapping rules to generate grayscale images that can reflect the grid density in different areas.

[0072] The dielectric layer and other conductive structure areas of the device area are filled with gray values of [1,200] according to the pixel mapping rule; the via area is filled with gray values of [1,200] according to the via coordinates (v' x ,v' y ) is used as the center to generate a 5×5 pixel square filled with 255; port area: if the port is located at the boundary, it will be expanded outward by 3 pixels along the boundary and filled with 254; the non-simulation area not covered by the circuit is filled with a grayscale value of 0.

[0073] Step S4: Data enhancement and normalization

[0074] S4-1 data enhancement:

[0075] 4-1-1 Geometric transformation:

[0076] The binary image and the grayscale image are rotated synchronously. For example, the angle θ∈[-15°,15°] can be used. The newly generated area is filled with zero values. The grayscale image interpolation method is the nearest neighbor interpolation to ensure the consistency of the data space.

[0077] 4-1-2 Add noise:

[0078] Gaussian noise (σ = 0.0.1) is added only to the device dielectric layer and other conductive structures to avoid contamination of vias and port markings, thereby maintaining the physical rationality of the label while improving data diversity.

[0079] Through the above method, the scale of the training dataset can be effectively expanded, the overfitting of the model can be suppressed, and the generalization ability of unseen geometric structures can be enhanced.

[0080] S4-2 normalization processing:

[0081]

[0082] Among them I raw (x,y) represents the pixel value of the grayscale image, I raw (x,y)∈[1,200],I norm(x,y) represents the normalized pixel value.

[0083] Therefore, the device dielectric layer and other conductive structure mask mask = (0 norm <0.785), via mark mask via_mask=(I norm =1.0), port mark mask port_mask=(0.996≤I norm <1.0).

[0084] This embodiment uses image processing technology to perform data enhancement on grayscale images to improve the generalization ability of the model, including operations such as rotation and adding noise to increase the diversity of training data.

[0085] Step S5: Deep learning model construction: Build a deep learning model based on a convolutional neural network and train it using the dataset;

[0086] For example, the deep learning model based on convolutional neural network adopts the ladder structure U-Net model Staircase U-Net with Skip connection, see the attached Figure 4 , whose input is the contour image of the 2D object of RFIC, and whose output is a grayscale image that can reflect the grid density of different areas (i.e., the size of the grid unit).

[0087] Staircase U-Net model Staircase U-Net training loss function see the attached Figure 5 The optimizer uses Adam, and the parameters are set to β1 = 0.9, β2 = 0.999; the initial learning rate is 1×10^(-4), and it decays exponentially by 0.89 after every 650 steps.

[0088] This embodiment adopts a ladder-structured U-Net network, combined with deformable convolution and channel attention mechanisms, to enhance the model's ability to extract geometric features. Through multi-scale feature extraction and jump connection mechanisms, geometric features can be gradually abstracted in the encoding stage, and high-resolution grayscale images can be accurately reconstructed in the decoding stage, significantly improving the grid density prediction accuracy of tiny structures (such as the edge of the dielectric layer). The input of the model is a binary image of the geometric contour, and the output is the corresponding grayscale image, reflecting the size of the grid unit. Through data enhancement technology, the generalization ability of the model is improved to ensure that it can work effectively under different objects and complex shapes.

[0089] Step S6: Use the trained and tested deep learning model to predict the corresponding grayscale image of the geometric binary image of the radio frequency integrated circuit, and then obtain the size of the grid unit.

[0090] ​The grayscale image output by the deep learning model can be reverse-mapped to the original object's size to help recreate an intuitive mesh. This process ensures that the generated mesh maintains the same geometric characteristics as the original object, facilitating subsequent simulation and analysis.

[0091] For example, the reverse mapping is:

[0092] 1) Density inversion calculation:

[0093]

[0094] Among them I pred (x,y) is the pixel value on the predicted grayscale image output by the deep learning model. max AA min The maximum and minimum values of the grid area are set by yourself.

[0095] 2) Perform inverse transformation on all pixel values on the predicted grayscale image to obtain the coordinates (x orig ,y orig ):

[0096]

[0097] Where (x' pix , y' pix ) represents the pixel value on the predicted grayscale image, Δ w , Δ h The width of the single side padding.

[0098] 3) Adaptive mesh generation:

[0099] According to the predicted grayscale image, for all coordinates (x orig ,y orig ) Use a mesh generation tool (such as Gmsh) to generate an adaptive non-uniform mesh.

[0100] This embodiment also provides an electromagnetic simulation grid density prediction system, including:

[0101] The data acquisition module is responsible for collecting the geometric structure data of the RFIC; extracting the geometric contour based on the geometric structure data, and then converting the geometric contour into a geometric binary image;

[0102] The density prediction module is responsible for using the trained and tested deep learning model to predict the corresponding grayscale image for the geometric binary image; the electromagnetic simulation grid density is obtained based on the predicted grayscale image.

[0103] An embodiment of the present invention provides an electronic device. Specifically, the electronic device includes a memory and a processor. The memory stores executable code. When the processor executes the executable code, the method described in any one of the embodiments is implemented.

[0104] The memory may include high-speed random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage. The system network element communicates with at least one other network element via at least one communication interface (which may be wired or wireless), such as the Internet, a wide area network, a local area network, a metropolitan area network, or the like.

[0105] The bus may be an ISA bus, a PCI bus or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc.

[0106] The memory is used to store the program, and the processor executes the program after receiving the execution instruction. The method executed by the device for flow process definition disclosed in any of the aforementioned embodiments of the present invention can be applied to the processor or implemented by the processor.

[0107] The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method may be completed by hardware integrated logic circuits in the processor or by software instructions. The above processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The methods, steps, and logic block diagrams disclosed in the embodiments of the present invention may be implemented or executed. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in conjunction with the embodiments of the present invention may be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium well-known in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or the like. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0108] The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can be referred to the previous method embodiment and will not be repeated here.

[0109] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0110] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A method for predicting electromagnetic simulation grid density, characterized in that The method comprises: Collect the geometric structure data of the RFIC and the corresponding mesh generation results; Extracting geometric contours according to geometric structure data, and then converting the geometric contours into geometric binary images; The grid cell area in the grid generation result is mapped into a grayscale image through normalization processing; wherein the grayscale image includes the device area covered by the circuit and the non-simulation area not covered by the circuit, and the device in the device area includes a special marking area and a parameterized area; the special marking area, the parameterized area, and the non-simulation area are distinguished by different grayscale values; Construct a data set based on geometric binary images and grayscale images; Construct a deep learning model and use the data set for training and testing; wherein the input of the deep learning model is a geometric binary image and the output is a corresponding grayscale image; The trained and tested deep learning model is used to predict the corresponding grayscale image of the geometric binary image of the RF integrated circuit; the electromagnetic simulation grid density is obtained based on the predicted grayscale image.

2. The method according to claim 1, characterized in that The other conductive structures are functional units in a radio frequency integrated circuit.

3. The method according to claim 1, characterized in that The special marking area includes ports and vias; the parameterized area includes dielectric layers and other conductive structure areas; the dielectric layers and other conductive structure areas are filled with different grayscale values according to the area of the grid units, the via area is filled with a grayscale value of 255, the port area is filled with a grayscale value of 254, and the non-simulation area not covered by the circuit is filled with a grayscale value of 0.

4. The method according to claim 3, characterized in that The grayscale value range of the dielectric layer and other conductive structure areas in the device area is [1,200].

5. The method according to claim 1, characterized in that: The process of mapping the grid cell area in the grid generation result into a grayscale image through normalization processing includes: First, the grid cell area is normalized, and then the original coordinates are scaled proportionally. The scaling factor is the smaller ratio of the target size to the original size. The grayscale value 0 area is symmetrically filled in the direction of the short side of the scaled area. A grayscale image is generated based on the normalized value of each grid cell area and pixel mapping rules.

6. The method according to claim 1, characterized in that The deep learning model is a ladder-type U-Net, whose input is a binary image of the geometric contour and the output is a corresponding grayscale image reflecting the size of the grid unit.

7. The method according to claim 1, characterized in that: Data augmentation is also performed on the dataset input into the deep learning model.

8. An electromagnetic simulation grid density prediction system for implementing the method according to any one of claims 1 to 7, characterized in that include: Data acquisition module, responsible for collecting geometric structure data of RF integrated circuits; Extracting geometric contours according to geometric structure data, and then converting the geometric contours into geometric binary images; The density prediction module is responsible for using the trained and tested deep learning model to predict the corresponding grayscale image for the geometric binary image; the electromagnetic simulation grid density is obtained based on the predicted grayscale image.

9. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the method according to any one of claims 1 to 7.

10. A computing device comprising a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, the method according to any one of claims 1 to 7 is implemented.

Images