Lithography prediction model training method and related product
By enhancing the focus on contour regions during the training of the lithography prediction model, the problem of inaccurate contour region prediction in the optical proximity effect correction process of the lithography prediction model is solved, thereby improving the prediction accuracy of the model and the manufacturing yield of the chip.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-28
- Publication Date
- 2026-04-07
AI Technical Summary
Existing lithography prediction models are prone to blurring or loss of detail in local sensitive areas during optical proximity effect correction, resulting in inaccurate prediction of contour areas and affecting chip performance and manufacturing yield.
By acquiring the contour region of the benchmark image during the training process of the lithographic prediction model, and adding extra weights to the pixels within the contour region in the loss function, global and local loss functions are constructed to enhance attention to edge details and improve prediction accuracy.
This improves the prediction accuracy of the lithography prediction model for the contour region, reduces contour distortion, and enhances the circuit connectivity and manufacturing yield of the chip.
Smart Images
Figure CN121806366A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of integrated circuit technology, and in particular to a training method for a photolithography prediction model, a computer-readable storage medium, a computer program product, and a computer device. Background Technology
[0002] In the semiconductor EDA industry, layout image processing is a crucial step. A layout image typically contains complex geometric patterns, lines, and contours, which directly affect chip performance, reliability, and manufacturing yield. As semiconductor process nodes continue to shrink, the precision requirements for layout images are becoming increasingly stringent.
[0003] In related technologies, deep learning-based lithography prediction methods have been widely applied in areas such as layout optimization, lithography simulation, defect detection, and image enhancement. Taking optical proximity correction (OPC) as an example, the lithography prediction model needs to simulate and predict the OPC-optimized mask layout, output the predicted optical spatial image, and then correct the mask layout based on the predicted optical spatial image.
[0004] However, existing technologies are prone to problems such as blurring of sensitive areas or loss of detail. Summary of the Invention
[0005] One object of the present invention is to provide a training method, computer-readable storage medium, computer program product and computer device for a lithographic prediction model, so as to enhance the lithographic prediction model’s attention to contour regions in an image, thereby improving the prediction accuracy of contour regions and reducing contour distortion.
[0006] Specifically, according to one aspect of the present invention, the present invention provides a method for training a lithography prediction model, comprising: Acquire training samples, which include the initial design layout to be predicted and a benchmark image used to characterize the ideal prediction result; The contour region is determined in the reference image; A loss function is obtained, which is constructed to build a true value based on the pixel value of each pixel in the reference image, and the true value weight of the pixels within the contour region is greater than the true value weight of the pixels outside the contour region. The initial lithography prediction model is trained using the training samples and the loss function to obtain the target lithography prediction model.
[0007] Optionally, determining the contour region in the reference image includes: Edge detection is performed on the reference image to extract multiple closed contour lines; Obtain the contour point set for each contour line, and generate the contour region based on the contour point set.
[0008] Optionally, determining the contour region in the reference image includes: Edge detection is performed on the reference image to extract multiple closed contour lines; Obtain the contour point set for each contour line, and generate an initial contour region based on the contour point set; Each of the initial contour regions is expanded outward and / or shrunken inward to form the contour region.
[0009] Optionally, the step of expanding and / or shrinking each of the initial contour regions to form the contour region includes: Obtain the geometric features of the initial design layout; Based on the aforementioned geometric features, the extended range parameters are obtained; Based on the expansion range parameter, each of the initial contour regions is expanded outward and / or shrunken inward to form the contour region.
[0010] Optionally, the loss function includes: The global branch loss function is constructed to build a first true value based on the pixel values of each pixel in the reference image and to build a first predicted value based on the pixel values of each pixel in the predicted image of the initial lithography prediction model. The local branch loss function is constructed to build a second true value based on the pixel values of each pixel in the contour region of the reference image, and to build a second predicted value based on the pixel values of each pixel in the region corresponding to the contour region of the predicted image of the initial lithography prediction model.
[0011] Optionally, the first true value weight of the global branch loss function is less than or equal to the second true value weight of the local branch loss function.
[0012] Optionally, the global branch loss function is constructed based on the mean absolute error function.
[0013] Optionally, the local branch loss function is constructed based on the mean absolute error function.
[0014] Optionally, the initial design layout includes a mask design layout; The lithography prediction model is used to simulate and predict the input mask design layout, and outputs the corresponding optical spatial image, photoresist development image and / or wafer etching image.
[0015] According to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the training method for any of the above-described lithography prediction models.
[0016] According to another aspect of the present invention, a computer program product is also provided, comprising a computer program that, when executed by a processor, implements the steps of the training method for any of the above-described lithography prediction models.
[0017] According to another aspect of the present invention, a computer device is also provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the training method for any of the above-described lithography prediction models.
[0018] The training method of the lithography prediction model of the present invention obtains the contour region of the reference image and modifies the loss function by adding extra weights to the pixels in the contour region in the loss function. This enhances the attention of the lithography prediction model to edge details during the training process, thereby improving the prediction accuracy of the lithography prediction model for the contour region and reducing contour distortion.
[0019] The above and other objects, advantages and features of the present invention will become more apparent to those skilled in the art from the following detailed description of specific embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description
[0020] The following sections will describe some specific embodiments of the invention in detail by way of example and not limitation, with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar parts or portions. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings: Figure 1 This is a flowchart illustrating a training method for a lithography prediction model according to an embodiment of the present invention. Figure 2 This is a schematic flowchart illustrating the process of obtaining the contour region of a reference image using a training method according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of the contour region of a reference image obtained by a training method according to an embodiment of the present invention; Figure 4 This is a schematic flowchart illustrating the process of obtaining the contour region of a reference image using a training method according to another embodiment of the present invention. Figure 5 This is a schematic diagram of the structure of the contour region of a reference image obtained by a training method according to another embodiment of the present invention; Figure 6This is a schematic flowchart illustrating the process of obtaining the contour region of a reference image using a training method according to yet another embodiment of the present invention. Figure 7 This is a schematic diagram of a computer program product according to an embodiment of the present invention; Figure 8 This is a schematic diagram of a computer-readable storage medium according to an embodiment of the present invention; and Figure 9 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation
[0021] In the field of Electronic Design Automation (EDA), layout image processing is a crucial step. A layout image typically contains complex geometric patterns, lines, and contours, which directly affect chip performance, reliability, and manufacturing yield. As semiconductor process nodes continue to shrink (reaching the nanometer level), the precision requirements for layout images are becoming increasingly stringent. Image restoration techniques, especially deep learning-based image restoration methods, have been widely applied in areas such as layout optimization, lithography simulation, defect detection, and image enhancement. These methods aim to recover high-precision layout details from noisy, blurred, or low-resolution input images to support subsequent mask fabrication and chip production.
[0022] In related technologies, deep learning-based lithography prediction models mainly rely on convolutional neural networks (CNN), generative adversarial networks (GAN), or other deep learning network architectures. These network models are usually trained based on loss functions (such as L1 loss, L2 loss, or perceptual loss) built on pixel-level loss, which can globally optimize the image, but lacks specificity for local sensitive areas (such as contours), resulting in blurred edges or loss of details.
[0023] For example, in optical proximity correction (OPC), the contour regions of the optical spatial image (i.e., pattern edges and boundaries) are often the decisive factors. These regions account for a small proportion of the total pixels, but have a huge impact on the error. Whether the contour regions can be accurately reproduced directly affects the connectivity, parasitic effects, and manufacturing tolerances of the circuit. If the error in the contour regions is too large, it may lead to problems such as short circuits, open circuits, or performance deviations. Existing lithography prediction models often ignore the weight differences in contour regions, resulting in poor model performance in high-density pattern regions.
[0024] The purpose of the training method for the lithography prediction model in this embodiment is to enhance the lithography prediction model's attention to contour regions in the image, thereby improving the prediction accuracy of contour regions and reducing contour distortion.
[0025] Figure 1 This is a flowchart illustrating a training method for a lithography prediction model according to an embodiment of the present invention. The method generally includes: S100, Obtain training samples, which include the initial design layout to be predicted and a benchmark image used to characterize the ideal prediction result; S200, the contour region is determined in the reference image; S300, Obtain the loss function. The loss function is constructed to build the true value based on the pixel value of each pixel in the reference image, and the true value weight of the pixels within the contour area is greater than the true value weight of the pixels outside the contour area. S400 uses training samples and a loss function to train the initial lithography prediction model to obtain the target lithography prediction model.
[0026] In this embodiment, the lithography prediction model can be a network model based on algorithms such as convolutional neural networks (e.g., U-Net, ResNet, etc.) and generative adversarial networks (GANs). Specifically, the lithography prediction model can be an optical simulation prediction model, wherein the initial design layout can be the mask design layout before optical proximity correction, and the reference image (ground truth) can be a measured spatial image obtained by using a charge-coupled device (CCD) to acquire its optical spatial image. The lithography prediction model is used to simulate the input initial design layout and output the predicted optical spatial image.
[0027] The lithography prediction model can also be an exposure and development prediction model. The initial design layout can be a mask design layout, and the ground truth image can be an SEM image of the developed photoresist obtained using a scanning electron microscope (SEM). The lithography prediction model is used to simulate the input initial design layout and output the predicted photoresist development image.
[0028] The lithography prediction model can also be an etching prediction model. The initial design layout can be a mask design layout, and the ground truth image can be a SEM image obtained by scanning electron microscopy (SEM) of the etched wafer. The lithography prediction model is used to simulate the input initial design layout and output the predicted wafer etching image.
[0029] The training method of this embodiment will be explained below using the lithography prediction model as an example of the optical simulation prediction model.
[0030] First, training samples (including mask design layouts and measured spatial images) are input into the lithography prediction model, which then identifies contour regions in the measured spatial images. For example, the lithography prediction model can use edge detection algorithms (such as the Sobel operator, Canny operator, etc.) to identify contour regions in the measured spatial images.
[0031] In some embodiments, the lithography prediction model includes a separate contour region recognition module, which identifies contour regions in the input reference image. This contour region recognition module can be used only during the training phase; after training is complete, it can be removed when the lithography prediction model is actually used.
[0032] Next, the loss function for the lithography prediction model is constructed. Specifically, firstly, pixels within the contour region of the measured spatial image are labeled. When constructing the loss function, the weight of the true values of pixels within the contour region is increased, making the weight of the true values of pixels within the contour region greater than that of pixels outside the contour region. The loss function can be a single L1 loss function, L2 loss function, etc., or multiple L1 loss functions or multiple L2 loss functions can be used in combination, depending on the needs.
[0033] It is important to understand that the initial design layout, the reference image, and the predicted image output by the lithography prediction model are usually aligned with each other. After determining the contour region of the measured spatial image, the contour region can be easily mapped onto the predicted optical spatial image based on the coordinates. This allows the predicted pixels within the contour region in the predicted optical spatial image to be marked, and then the weighted loss between the predicted value and the true value within the contour region can be calculated using the loss function.
[0034] By adding extra weights to pixels within the contour region in the loss function, the lithography prediction model's attention to edge details can be enhanced, improving prediction accuracy of image contour regions and reducing contour distortion. Furthermore, it can accelerate model convergence, reduce the number of iterations during training, and improve training efficiency. Moreover, this approach only requires modification to the loss function, thus allowing seamless integration into existing deep learning-based lithography prediction models, such as seamless integration into the PyTorch framework, improving model development efficiency and reducing development costs.
[0035] Next, the initial lithographic prediction model is trained using training samples and a loss function. Taking a deep learning network model based on the U-Net architecture as an example, the initial lithographic prediction model can be configured with multi-layered (e.g., 5-layer) encoders, decoders, and the loss function obtained in the previous steps. Skip connections are also set between the encoders and decoders at corresponding layers. The encoder part consists of multiple convolutions and downsampling operations, used for feature extraction from the input image. The decoder consists of multiple convolution operations and upsampling operations, used to generate the predicted image. In the U-Net architecture, the corresponding features and the decoded features of the same scale are integrated through skip connections, thereby preserving the detailed information of the lower layers under different resolution conditions. During network training, some features can be directly passed through this connection. Therefore, the initial lithographic prediction model based on the U-Net architecture has a good effect on improving the detail representation of the contour region.
[0036] Specifically, the mask design layout and the measured spatial image are input into the initial lithography prediction model. The initial lithography prediction model performs feature extraction (such as line width features, spacing features, corner features, sub-resolution auxiliary graphic features, etc.) and simulation prediction on the mask design layout to obtain the preliminary predicted optical spatial image.
[0037] Next, the predicted values of pixels in the contour and non-contour regions of the initially predicted optical spatial image are obtained, and the true values of pixels in the contour and non-contour regions of the measured spatial image are obtained. A loss function is then used to calculate the overall loss between the initially predicted optical spatial image and the measured spatial image. Since the true value weight of pixels within the contour region is greater than that of pixels outside the contour region, the loss of pixel values within the contour region in the initially predicted optical spatial image and the measured spatial image is amplified.
[0038] Next, a preset loss threshold is obtained, and it is determined whether the overall loss of the loss function is greater than the preset loss threshold. If it is greater, the initial weights of the initial lithography prediction model are adjusted based on the result, and then prediction is performed again. This process is repeated multiple times until the overall loss of the loss function meets the requirement of the preset loss threshold, at which point training stops, and the trained target lithography prediction model is obtained.
[0039] In the training method of the lithography prediction model in this embodiment, the contour region of the reference image is obtained and the loss function is modified. Additional weights are added to the pixels in the contour region in the loss function, which enhances the attention of the lithography prediction model to edge details during the training process, thereby improving the prediction accuracy of the lithography prediction model for the contour region and reducing contour distortion.
[0040] In some embodiments of the training method for the lithography prediction model of the present invention, such as Figure 2 As shown, the contour region is determined in the reference image, including: S211, perform edge detection in the reference image and extract multiple closed contour lines; S213, obtain the contour point set of each contour line respectively, and generate the contour region based on the contour point set.
[0041] In this embodiment, the lithography prediction model can use edge detection algorithms (such as the Sobel operator, Canny operator, etc.) to perform edge detection on the reference image. Specifically, multiple edge-finding points can be placed in the reference image first. Then, using each edge-finding point as a starting point, a search is performed within a preset window to locate the point with the largest absolute gradient value, which is then used as a contour point. After determining the precise location of the contour point within the preset window, the coordinate information of that point is recorded. After contour points are found within the preset windows of all edge-finding points, the corresponding contour points are connected to form multiple closed contour lines.
[0042] Next, the pixels that make up the contour line are organized into a series of discrete but high-precision two-dimensional contour point sets, and the contour point sets are used as contour regions to obtain contour masks, thereby enabling the marking of pixels within the contour regions.
[0043] Please see Figure 3 , Figure 3 This is a schematic diagram illustrating the structure of the contour region obtained from the reference image. Each dashed box represents a pixel, the red closed line represents the contour line extracted by edge detection, the yellow box represents the contour points that make up the contour region (using mask 1), and the gray box represents pixels outside the contour region (using mask 0).
[0044] This embodiment can quickly and accurately obtain the contour region of the reference image and mark the pixels within the contour region.
[0045] In some embodiments of the training method for the lithography prediction model of the present invention, such as Figure 4 As shown, the contour region is determined in the reference image, including: S221, perform edge detection in the reference image and extract multiple closed contour lines; S223, Obtain the contour point set of each contour line respectively, and generate the initial contour region based on the contour point set respectively; S225, expand and / or shrink each initial contour region to form a contour region.
[0046] This embodiment improves upon the scheme of the previous embodiment. Specifically, after generating an initial contour region from the contour point set of the contour line, the initial contour region is further expanded outward by one or more pixels, and / or expanded inward by one or more pixels to obtain the final contour region. In practical use, the expansion or contraction, or both, can be determined as needed, and the extent of the expansion can be determined as required.
[0047] Please see Figure 5 , Figure 5 In order to be in Figure 3 Based on this, a structural diagram of the contour region is obtained. Specifically, after generating an initial contour region from the contour point set of the contour line, the initial contour region is expanded outward by 1 pixel and inward by 1 pixel, ultimately resulting in a ring-shaped contour region with a pixel width of 3. This contour region includes the pixel where the contour line is located, as well as its neighboring pixels.
[0048] By expanding the initial contour region, the final contour region includes not only the pixels where the contour line is located, but also the pixels adjacent to the contour line, thereby further expanding the scope of the loss function and preventing the lithography prediction model from underfitting near the contour line.
[0049] In some embodiments of the training method for the lithography prediction model of the present invention, such as Figure 6 As shown, each initial contour region is expanded outward and / or shrunken inward to form a contour region, including: S231, Obtain the geometric features of the initial design layout; S233, based on geometric features, obtain the extended range parameters; S235, based on the extended range parameter, expands and / or shrinks each initial contour region to form a contour region.
[0050] In this embodiment, the extended range parameter can be represented by the number of pixels n, for example... Figure 5 The number of pixels n equals 1. In practical applications, the values of the corresponding extension range parameters can be flexibly determined based on the geometric characteristics of the initial design layout. Specifically, geometric characteristics include the complexity of the feature graphics, minimum line width, minimum spacing, etc.
[0051] For example, a larger initial outline area can be used for an initial design layout with simple geometry and a large minimum linewidth and minimum spacing. Conversely, a smaller initial outline area can be used for an initial design layout with complex geometry and a small minimum linewidth and minimum spacing.
[0052] For example, the initial design layout can also be divided into multiple blocks based on the geometric features of each location, with different types of blocks using different extension range parameters.
[0053] This embodiment achieves a flexible local contour region attention mechanism by using extended range parameters, making this training method applicable to various types of semiconductor layouts.
[0054] In some embodiments of the training method for the lithography prediction model of the present invention, the loss function includes a global branch loss function and a local branch loss function. The global branch loss function is constructed by building a first ground truth value based on the pixel values of each pixel in the reference image and a first predicted value based on the pixel values of each pixel in the predicted image of the initial lithography prediction model. The local branch loss function is constructed by building a second ground truth value based on the pixel values of each pixel in the reference image located within the contour region and a second predicted value based on the pixel values of each pixel in the predicted image of the initial lithography prediction model located within the region corresponding to the contour region.
[0055] In this embodiment, the loss function adopts the form of a joint loss function to comprehensively represent the overall loss of the global and contour regions. For example, the loss function can be expressed as follows: ,in, , , In the above formula, Indicates the total loss. Represents the global branch loss function. Represents the local branch loss function; This represents the total number of pixels in the predicted or reference image. This represents the predicted value of a pixel in the image. This represents the true value of a pixel in the reference image. Used to iterate through all pixels To predict the total number of pixels in the contour region of an image or reference image, This represents the predicted value of pixels in the contour region of the predicted image. This represents the true value of the pixels in the contour region of the reference image. Used to iterate through all pixels of the outline region. This represents the additional weights of the local branch loss function. Greater than 0.
[0056] In this embodiment, both the global branch loss function and the local branch loss function are constructed based on the mean absolute error function. The mean absolute error function is more robust to outliers and can better preserve the edges and details of the image.
[0057] As can be seen from the above formula, the loss in the contour region is calculated once, thus enhancing the model's attention to edge details, thereby improving prediction accuracy and reducing contour distortion.
[0058] In some embodiments of the training method for the lithography prediction model of the present invention, the first true value weight of the global branch loss function is less than or equal to the second true value weight of the local branch loss function.
[0059] In this embodiment, the loss function It can be greater than or equal to 1, for example, between 1 and 5. This further increases the loss weight for the contour region, and further enhances the model's attention to edge details.
[0060] In practical applications, additional weights of the local branch loss function can be flexibly set according to the geometric characteristics of the initial design layout to prevent overfitting and underfitting, thereby improving the overall prediction accuracy of the lithography prediction model.
[0061] The flowchart provided in this embodiment is not intended to indicate that the operations of the method will be performed in any particular order, or that all operations of the method are included in every case. Furthermore, the method may include additional operations. Within the scope of the technical concept provided by the method in this embodiment, additional variations can be made to the above method.
[0062] It should be understood that in some embodiments, the components may be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods may be implemented using software or firmware stored in memory and executed by a suitable instruction execution system.
[0063] This invention also provides a computer program product 10, a computer-readable storage medium 20, and a computer device 30. Figure 7 This is a schematic diagram of a computer program product 10 according to an embodiment of the present invention. Figure 8 This is a schematic diagram of a computer-readable storage medium 20 according to an embodiment of the present invention. Figure 9 This is a schematic diagram of a computer device 30 according to an embodiment of the present invention. The computer program product 10 includes a computer program 11, which, when executed by the processor 32, implements the steps of the training method for the lithography prediction model described above. A computer-readable storage medium 20 stores the computer program 11 thereon, which, when executed by the processor 32, implements the steps of the training method for the lithography prediction model described above. The computer device 30 may include a memory 31, a processor 32, and the computer program 11 stored in the memory 31 and running on the processor 32.
[0064] The computer program 11 used to perform the operations of this invention may be assembly instructions, Instruction Set Architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages and procedural programming languages. The computer program 11 may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network, including a Local Area Network (LAN) or Wide Area Network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, to perform aspects of this invention, electronic circuits, including, for example, programmable logic circuits, Field-Programmable Gate Arrays (FPGAs), or Programmable Logic Arrays (PLAs), may execute computer-readable program instructions to personalize the electronic circuits by utilizing state information from computer-readable program instructions.
[0065] For the purposes of this embodiment, computer program product 10 is a related product that includes computer program 11.
[0066] For the purposes of this embodiment, the computer-readable storage medium 20 is a tangible device capable of holding and storing a computer program 11. It can be any device capable of containing, storing, communicating, propagating, or transmitting the computer program 11 for use by or in conjunction with an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable storage medium 20 include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable optical disc read-only memory (CD-ROM), digital versatile disc (DVD), memory stick, floppy disk, mechanical encoding device, and any suitable combination thereof.
[0067] Computer device 30 can be, for example, a server, desktop computer, laptop computer, tablet computer, or smartphone. In some examples, computer device 30 can be a cloud computing node. Computer device 30 can be described in the general context of computer system executable instructions (such as program modules) executed by a computer system. Typically, program modules can include routines, programs, object programs, components, logic, data structures, etc., that perform specific tasks or implement specific abstract data types. Computer device 30 can be implemented in a distributed cloud computing environment where tasks are performed by remote processing devices linked through a communication network. In a distributed cloud computing environment, program modules can reside on local or remote computing system storage media, including storage devices.
[0068] Computer device 30 may include a processor 32 adapted to execute stored instructions and a memory 31 that provides temporary storage space for the operation of said instructions during operation. The processor 32 may be a single-core processor, a multi-core processor, a computing cluster, or any other configuration. The memory 31 may include random access memory (RAM), read-only memory, flash memory, or any other suitable storage system.
[0069] Computer device 30 may also include a network adapter / interface and an input / output (I / O) interface. The I / O interface allows external devices that can be connected to the computer device to input and output data. The network adapter / interface provides communication between the computer device and a network, typically represented as a communication network.
[0070] Therefore, those skilled in the art should recognize that although numerous exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Thus, the scope of the present invention should be understood and construed as covering all such other variations or modifications.
Claims
1. A training method for a lithography prediction model, characterized in that, include: Acquire training samples, which include the initial design layout to be predicted and a benchmark image used to characterize the ideal prediction result; The contour region is determined in the reference image; A loss function is obtained, which is constructed to build a true value based on the pixel value of each pixel in the reference image, and the true value weight of the pixels within the contour region is greater than the true value weight of the pixels outside the contour region. The initial lithography prediction model is trained using the training samples and the loss function to obtain the target lithography prediction model.
2. The training method according to claim 1, characterized in that, Determining the contour region in the reference image includes: Edge detection is performed on the reference image to extract multiple closed contour lines; Obtain the contour point set for each contour line, and generate the contour region based on the contour point set.
3. The training method according to claim 1, characterized in that, Determining the contour region in the reference image includes: Edge detection is performed on the reference image to extract multiple closed contour lines; Obtain the contour point set for each contour line, and generate an initial contour region based on the contour point set; Each of the initial contour regions is expanded outward and / or shrunken inward to form the contour region.
4. The training method according to claim 3, characterized in that, The step of expanding and / or shrinking each of the initial contour regions to form the contour region includes: Obtain the geometric features of the initial design layout; Based on the aforementioned geometric features, the extended range parameters are obtained; Based on the expansion range parameter, each of the initial contour regions is expanded outward and / or shrunken inward to form the contour region.
5. The training method according to claim 1, characterized in that, The loss function includes: The global branch loss function is constructed to build a first true value based on the pixel values of each pixel in the reference image and to build a first predicted value based on the pixel values of each pixel in the predicted image of the initial lithography prediction model. The local branch loss function is constructed to build a second true value based on the pixel values of each pixel in the contour region of the reference image, and to build a second predicted value based on the pixel values of each pixel in the region corresponding to the contour region of the predicted image of the initial lithography prediction model.
6. The training method according to claim 5, characterized in that, The first true value weight of the global branch loss function is less than or equal to the second true value weight of the local branch loss function.
7. The training method according to claim 5, characterized in that, The global branch loss function is constructed based on the mean absolute error function; and / or The local branch loss function is constructed based on the mean absolute error function.
8. The training method according to claim 1, characterized in that, The initial design layout includes the mask design layout; The lithography prediction model is used to simulate and predict the input mask design layout, and outputs the corresponding optical spatial image, photoresist development image and / or wafer etching image.
9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the training method for the lithographic prediction model as described in any one of claims 1 to 8.
10. A computer device, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the training method for the lithography prediction model according to any one of claims 1 to 8.