Mask plate correction method and device, computer device and readable storage medium

Abstract

Embodiments of the present application provide a mask correction method and device, computer equipment and a readable storage medium. The method comprises: obtaining a pattern structure image corresponding to a mask to be corrected; dividing the pattern structure image to obtain a plurality of image segments, and generating a to-be-analyzed structure image based on segment features of each image segment and common features between the image segments; inputting the to-be-analyzed structure image into a target model, the target model classifying based on the segment features and the common features in the to-be-analyzed structure image to obtain a plurality of correction decision labels corresponding to each image segment, each correction decision label containing a correction offset and a corresponding probability value; for each image segment, selecting a target correction offset of the image segment based on the probability value in each correction decision label corresponding to the image segment; and correcting the mask according to the target correction offset of each image segment to obtain a target mask. In this way, the efficiency and accuracy of mask correction can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical correction, and in particular to a mask correction method and device, a computer device and a readable storage medium. BACKGROUND

[0002] A mask refers to a hollow template containing circuit design patterns used in the semiconductor manufacturing process. In the lithography process, light emitted by a light source passes through the image on the mask, and is projected onto a wafer covered with photosensitive material to form a projected pattern on the wafer. Due to the optical proximity effect, the shape of the actual projected pattern on the wafer often deforms, which directly affects the manufacturing accuracy and quality of the chip.

[0003] In related technologies, in order to overcome the optical proximity effect, the image structure of the mask is generally divided into multiple image segments, and the edge placement error and the process variation bandwidth are corrected for each image segment in turn. However, this may make the correction result of the edge placement error at the previous moment invalid during the process variation bandwidth correction, at which time the correction program needs to be restarted, resulting in low efficiency and accuracy of the mask correction. Moreover, when correcting each image segment, only a single image segment is considered, and the influence of each structure in the mask is not considered, further reducing the accuracy of the mask correction. SUMMARY

[0004] The main purpose of the embodiments of the present application is to provide a mask correction method, device, computer device and readable storage medium, which can improve the efficiency and accuracy of mask correction.

[0005] In some embodiments, the first aspect of the embodiments of the present application provides a mask correction method, which comprises:

[0006] obtaining a pattern structure image corresponding to a mask to be corrected;

[0007] dividing the pattern structure image to obtain multiple image segments, and generating a to-be-analyzed structure graph based on the segment features of each image segment and the common features between the multiple image segments;

[0008] inputting the to-be-analyzed structure graph into a target model, the target model classifying based on the segment features and the common features in the to-be-analyzed structure graph to obtain multiple correction decision labels corresponding to each image segment, each correction decision label containing a correction offset and a corresponding probability value;

[0009] The target model is obtained by a preset model based on a sample to-be-analyzed structure graph and an adjusted sample target image through maximum learning of a correction index; the correction index is determined by a first edge placement error and a first process variation bandwidth corresponding to the sample to-be-analyzed structure graph, and a second edge placement error and a second process variation bandwidth corresponding to the sample target image; the sample target image is obtained by adjusting the sample to-be-analyzed structure graph according to an indication of a plurality of sample correction decision labels output by the preset model;

[0010] For each image segment, a target correction offset of the image segment is selected based on a probability value in each correction decision label corresponding to the image segment;

[0011] The mask is corrected according to the target correction offset of each image segment to obtain a target mask.

[0012] Correspondingly, a second aspect of the embodiment of the application provides a mask correction device, the device comprising:

[0013] The acquisition module is configured to acquire a pattern structure image corresponding to the mask to be corrected.

[0014] The generation module is configured to divide the pattern structure image to obtain a plurality of image segments, and generate a to-be-analyzed structure graph based on a segment feature of each image segment and a common feature between the plurality of image segments.

[0015] The classification module is configured to input the to-be-analyzed structure graph into a target model, the target model classifies the segment feature and the common feature in the to-be-analyzed structure graph to obtain a plurality of correction decision labels corresponding to each image segment, each correction decision label containing a correction offset and a corresponding probability value; wherein the target model is obtained by a preset model based on a sample to-be-analyzed structure graph and an adjusted sample target image through maximum learning of a correction index; the correction index is determined by a first edge placement error and a first process variation bandwidth corresponding to the sample to-be-analyzed structure graph, and a second edge placement error and a second process variation bandwidth corresponding to the sample target image; the sample target image is obtained by adjusting the sample to-be-analyzed structure graph according to an indication of a plurality of sample correction decision labels output by the preset model;

[0016] The selection module is configured to, for each image segment, select a target correction offset of the image segment based on a probability value in each correction decision label corresponding to the image segment.

[0017] The correction module is configured to correct the mask according to the target correction offset of each image segment to obtain a target mask.

[0018] In some embodiments, the selecting module is further configured to:

[0019] obtain a reference layout corresponding to the mask plate;

[0020] determine a positional deviation relationship of each of the image segments relative to the reference layout;

[0021] for each of the image segments, adjust a probability value in each of the correction decision labels corresponding to the image segment based on the positional deviation relationship, to obtain a correction offset and a modified probability value of the probability value after adjustment of each of the image segments under each of the correction decision labels;

[0022] select a target correction offset in the image segment based on the modified probability value in each of the correction decision labels corresponding to the image segment.

[0023] In some embodiments, the selecting module is further configured to:

[0024] select a target measurement point for each of the image segments and the reference layout;

[0025] calculate an edge placement error of the image segment based on a positional deviation relationship of the target measurement point relative to the reference layout;

[0026] obtain a reference value, and form an error interval based on the reference value and the edge placement error;

[0027] select an equal number of error reference values as the correction decision labels for the error interval, and determine a corresponding relationship between each of the correction decision labels and the error reference values based on the positional deviation relationship of the target measurement point relative to the reference layout;

[0028] adjust a probability value in each of the correction decision labels corresponding to the image segment based on the corresponding relationship, to obtain a modified probability value of the probability value after adjustment of each of the image segments under each of the correction decision labels.

[0029] In some embodiments, the selecting module is further configured to:

[0030] obtain a target modulation function corresponding to the mask plate;

[0031] calculate each of the error reference values according to the target modulation function, to obtain a reference probability value corresponding to each of the error reference values;

[0032] For each of the image segments, the modified probability value is multiplied by a reference probability value corresponding to the error reference value based on the corresponding relationship, to obtain a modified probability value of each of the image segments under the probability value adjustment of each of the correction decision labels.

[0033] In some embodiments, the generating module is further configured to:

[0034] Calculate the Euclidean distance between the segment feature corresponding to each of the image segments and the adjacent segment feature;

[0035] Determine the image segments with common features based on the Euclidean distance between the segment features;

[0036] Image encode each of the image segments to obtain each of the encoded image segments;

[0037] Generate a structure diagram to be analyzed based on the encoded image segments and the common features between the image segments.

[0038] In some embodiments, the generating module is further configured to:

[0039] Obtain a preset neighborhood construction scale, and construct a square neighborhood for each of the image segments based on the neighborhood construction scale;

[0040] In each of the square neighborhoods, divide according to the graphic edge of the image segment to obtain a plurality of grids after the square neighborhood is divided;

[0041] Binary each of the grids in each of the square neighborhoods to obtain a topological matrix corresponding to the square neighborhood;

[0042] In the horizontal direction of the square neighborhood, obtain the length of each of the grids, and aggregate the horizontal lengths of all the grids to generate a horizontal matrix of the square neighborhood;

[0043] In the vertical direction of the square neighborhood, obtain the length of each of the grids, and aggregate the vertical lengths of all the grids to generate a vertical matrix of the square neighborhood;

[0044] Based on the topological matrix, the horizontal matrix and the vertical matrix, obtain each of the encoded image segments.

[0045] In some embodiments, the classification module is further configured to:

[0046] obtain a sample to-be-analyzed structure diagram, and input the sample to-be-analyzed structure diagram into a preset model to obtain a plurality of sample correction decision labels corresponding to each sample image segment; wherein the sample to-be-analyzed structure diagram is obtained by encoding a plurality of sample image segments;

[0047] For each sample image segment, a sample target correction offset of each sample image segment is selected based on a sample probability value in each sample correction decision label corresponding to each sample image segment;

[0048] According to the sample target correction offset of each sample image segment, the sample to-be-analyzed structure diagram is adjusted to obtain an adjusted sample target image;

[0049] The first edge placement error and the first process variation bandwidth corresponding to the sample to-be-analyzed structure diagram, and the second edge placement error and the second process variation bandwidth corresponding to the sample target image are calculated, and a correction index of the preset model is calculated based on the first edge placement error, the first process variation bandwidth, the second edge placement error and the second process variation bandwidth.

[0050] The parameters of the preset model are updated based on the correction index until the number of training times of the preset model reaches a preset iteration number, and a target model is obtained.

[0051] To achieve the above object, a third aspect of embodiments of the present application provides a computer device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the mask correction method of any one of the first aspect of the embodiments of the present application when executing the computer program.

[0052] To achieve the above object, a fourth aspect of embodiments of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the mask correction method of any one of the first aspect of the embodiments of the present application.

[0053] The embodiment of the present application obtains a pattern structure image corresponding to a mask to be corrected; divides the pattern structure image to obtain a plurality of image segments, and generates a to-be-analyzed structure graph based on segment features of each image segment and common features between the plurality of image segments; inputs the to-be-analyzed structure graph into a target model, and the target model classifies based on the segment features and the common features in the to-be-analyzed structure graph to obtain a plurality of correction decision labels corresponding to each image segment, each correction decision label containing a correction offset and a corresponding probability value; wherein the target model is obtained by a preset model based on sample to-be-analyzed structure graphs and sample target images adjusted to maximize a correction index; the correction index is determined by a first edge placement error and a first process variation bandwidth corresponding to the sample to-be-analyzed structure graph, and a second edge placement error and a second process variation bandwidth corresponding to the sample target image; the sample target image is obtained by adjusting the sample to-be-analyzed structure graph according to the indication of a plurality of sample correction decision labels output by the preset model; for each image segment, the target correction offset of the image segment is selected based on the probability value in each correction decision label corresponding to the image segment; and the mask is corrected according to the target correction offset of each image segment to obtain a target mask. In this way, not only the features of a single image segment can be considered, but also the common features between image segments can be considered, that is, the influence between various structures in the mask is considered, so that the probability value corresponding to the output correction offset can fit the actual result of the mask, thereby selecting a more accurate target correction offset. At the same time, the target model of the present application can be obtained by maximizing the correction index, and the correction index is calculated based on the edge placement error and the process variation bandwidth of the sample to-be-analyzed structure graph and the sample correction image before correction, so that the performance learned can be directly used to accurately adjust each image segment in the process of correction of the target model, without having to adjust the image segment in two stages, thereby improving the efficiency and accuracy of the mask correction. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 FIG. 1 is a structural schematic diagram of a mask correction system provided by an embodiment of the present application;

[0055] Figure 2 FIG. 2 is a flowchart of a mask correction method provided by an embodiment of the present application;

[0056] Figure 3 FIG. 3 is an encoding graph of an image segment provided by an embodiment of the present application;

[0057] Figure 4 FIG. 4 is a target modulation function mapping graph provided by an embodiment of the present application;

[0058] Figure 5 FIG. 5 is a functional structural schematic diagram of a mask correction device provided by an embodiment of the present application;

[0059] Figure 6 Fig. 1 is a schematic diagram of a hardware structure of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application and not to limit the present application.

[0061] It should be noted that although the functional modules are divided in the device schematic diagram, and the logical order is shown in the flowchart, in some cases, the steps shown or described can be performed in a manner different from the module division in the device or the order in the flowchart. The terms "first", "second", and the like in the specification and claims and the above-described drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this application is for the purpose of describing embodiments of this application only and is not intended to be limiting of this application.

[0063] A mask is a hollow template containing circuit design patterns used in the semiconductor manufacturing process. In the photolithography process, light emitted by a light source passes through the image on the mask, and is projected onto a wafer covered with a photosensitive material to form a projected pattern on the wafer. Due to the optical proximity effect, the actual projected pattern on the wafer often deforms, which directly affects the manufacturing accuracy and quality of the chip.

[0064] In the related art, in order to overcome the optical proximity effect, the image structure of the mask is generally divided into multiple image segments, and the edge placement error and the process variation bandwidth are corrected for each image segment in turn. However, this may cause the correction result of the edge placement error at the previous moment to be invalid in the process of correcting the process variation bandwidth, at which time the correction program needs to be restarted, resulting in low efficiency and accuracy of the mask correction. Moreover, when correcting each image segment, only a single image segment is considered, and the influence of each structure in the mask is not considered, further reducing the accuracy of the mask correction.

[0065] Based on this, the embodiments of the present application provide a mask correction method and device, a computer device and a readable storage medium, which can improve the efficiency and accuracy of the mask correction.

[0066] The mask plate correction method, device, computer device and readable storage medium provided by the embodiments of the present application are described in detail through the following embodiments. First, the mask plate correction system in the embodiments of the present application is described.

[0067] Please refer to Figure 1 In some embodiments, the mask plate correction system can include a client 11, a server 12 and a controller 13. Specifically, the client 11 can be a software application on a computer or other device, used for image processing and data analysis. For example, the client 11 can also be an interface for users to interact with the correction system. Users can operate the correction system through the client 11, such as setting correction parameters, sending correction tasks, monitoring correction progress, etc. The client 11 provides an intuitive and friendly user interface, allowing users to easily use the mask plate correction system.

[0068] In some embodiments, the server 12 can be a remote computer system or a central processing unit, used for storing and processing large amounts of data, receiving requests from the client 11 and assigning tasks, and distributing tasks to the controller 13 for execution. Specifically, the server 12 can include an algorithm module, a data processing module, etc. When a preset model or target model on the server 12 needs to be trained, the algorithm module can be used to train and optimize the model, so that the target model obtained by training can efficiently complete the mask plate correction task.

[0069] In some embodiments, the controller 13 can be a software program or a hardware device, which can be used to control the operation of the entire system and coordinate the communication between various hardware devices. When the controller 13 receives the task allocation information from the server, it can perform specific correction operations according to the instructions, and feed back the state information during execution and the correction results obtained by execution to the server 12, to ensure that each component works according to the predetermined process.

[0070] The mask plate correction method in the embodiments of the present application can be described as follows.

[0071] It should be noted that in each specific embodiment of the present application, when it is necessary to process data related to the identity or characteristics of the user, such as user information, user behavior data, user history data and user location information, the user's permission or consent will be obtained first. Moreover, the collection, use and processing of such data will comply with relevant laws, regulations and standards. In addition, when the embodiments of the present application need to obtain sensitive personal information of the user, the separate permission or separate consent of the user will be obtained through a pop-up window or a jump to a confirmation page, and after obtaining the separate permission or separate consent of the user, the necessary user-related data for the normal operation of the embodiments of the present application will be obtained.

[0072] In the embodiments of the present application, the dimensions of the mask correction device will be described, which can be integrated in a computer device. Referring to Figure 2 , Figure 2 An optional flowchart of the mask correction method provided by the embodiments of the present application is described. The embodiments of the present application are taken as an example that the mask correction device is integrated in a terminal or a server side. When the processor of the terminal or the server side executes the program instructions corresponding to the mask correction method, the specific process is as follows:

[0073] In step 101, a pattern structure image corresponding to the mask to be corrected is obtained.

[0074] It can be understood that, since the mask may contain background noise and non-critical information, directly using the mask may increase the complexity of the correction process. Therefore, the pattern structure image corresponding to the mask to be corrected can be obtained, so that the pattern structure image is directly analyzed, and the correction process is simplified.

[0075] Wherein, the mask to be corrected can be a silicon wafer (i.e. a wafer) used in the semiconductor manufacturing process, and the mask is full of tiny electronic component patterns. Due to the errors in the manufacturing process, the image projected onto the wafer through the mask will also be deviated. In order to ensure the accuracy and consistency in the chip manufacturing process, the mask can be corrected to improve the success rate of chip manufacturing and product quality, reduce production cost, and ultimately improve overall production efficiency.

[0076] Wherein, the pattern structure image can be a structure image extracted based on the mask, which is usually obtained by digital camera, optical or electronic microscope and other tools. The pattern structure image can be used to analyze the microstructure on the mask, so as to correct and optimize.

[0077] The pattern structure image can be obtained by the above method, which can facilitate subsequent division of the pattern structure image to obtain a plurality of image segments, extraction of segment features of each image segment, generation of a structure to be analyzed based on the common features between each segment feature and the image segments, and improvement of the correction accuracy of the mask.

[0078] In step 102, the pattern structure image is divided to obtain a plurality of image segments, and a structure to be analyzed is generated based on the segment features of each image segment and the common features between the plurality of image segments.

[0079] Specifically, in order to analyze the local features between the mask plates and the local correlations between the local features, the mask plates can be divided into multiple image segments, and when there are common features between the image segments, the common features between the image segments are extracted, so that the segment features of the image segments can be analyzed individually, and the connections between the image segments can be analyzed, so that the generated to-be-analyzed structure graph can better reflect the real structure of the mask plate.

[0080] The image segment can be a plurality of image blocks or image regions obtained by dividing the pattern structure image according to a preset division rule, and each image segment can be used to represent a local feature of the mask plate, such as a structure, a shape, and a texture, etc. For example, the pattern structure image can be divided according to a certain division scale by using an image segmentation algorithm, or a sliding window technique is used to divide the local region of the pattern structure image, and the embodiments of the present application do not make specific limitations.

[0081] The segment feature can be a feature that describes a specific attribute of the image segment, which can exist in the form of a vector or a numerical value, and the segment feature of each image segment can be obtained by using an image processing and feature extraction algorithm. For example, a feature descriptor can be used to describe the segment feature of the image segment, and the segment feature represents the quantitative information of the image segment, so as to facilitate further processing of the image segment.

[0082] The common feature between the image segments can be used to represent the similarity or correlation between the multiple image segments in some attributes. For example, the Euclidean distance between the segment features of the two adjacent image segments can be calculated, and when the segment features corresponding to the two image segments are less than a preset threshold, it can be determined that the two image segments have a common feature, which also indicates that the two image segments have an edge.

[0083] The segment feature of each image segment can be encoded, and the to-be-analyzed structure graph can be generated based on the common features between the image segments, that is, the edges between the image segments. The encoding mode can be matrix-based encoding, or Squish Pattern (SP) encoding, etc. The to-be-analyzed structure graph can provide more abundant information and details, and by analyzing the to-be-analyzed structure graph, the deviations, errors, and deformations in the mask plate can be identified.

[0084] In some embodiments, the division of the image segments can be selected. It can be understood that due to the different angles in the photolithography, some specific regions of the mask may not have errors, such as the middle region of the mask, and thus, when the image segments are divided, the middle region can be masked, or the division scale of the middle region is set to be larger than that of the edge, and the like, so that the adjusted gravity center is more concentrated in the region where the error actually exists and is adjusted, and the speed of the mask correction is improved.

[0085] In some embodiments, when the pattern structure image is divided into multiple image segments, different regions can be set to different division scales, or all regions can be set to the same division scale, and the division scale can be adjusted according to the actual situation, which is not limited in the embodiments of the present application.

[0086] By dividing the pattern structure image into image segments, and generating the to-be-analyzed structure graph by extracting the segment features and the common features, the characteristics and the structure of the mask can be more accurately represented, so that the target model can make more accurate decisions based on the to-be-analyzed structure graph, which is beneficial to improve the accuracy of the mask correction and improve the yield of the mask.

[0087] In some embodiments, in order to improve the accuracy of the mask correction, the to-be-analyzed structure graph can be generated based on the segment features of each image segment and the common features between multiple image segments, so that the target model can make more accurate decisions according to the segment features of each image segment and the connection relationship between the image segments, so as to determine the specific adjustment scale of each image segment. For example, the “generating the to-be-analyzed structure graph based on the segment features of each image segment and the common features between multiple image segments” in step 102 can include:

[0088] (102.1), calculating the Euclidean distance between the segment features corresponding to each image segment and the adjacent segment features;

[0089] (102.2), determining the image segments having the common features based on the Euclidean distance between the segment features;

[0090] (102.3), image encoding each image segment to obtain the encoded image segments;

[0091] (102.4), generating the to-be-analyzed structure graph based on the encoded image segments and the common features between the image segments.

[0092] In some embodiments, to determine whether there is an edge between each image segment and the adjacent image segment, a Euclidean distance between the image segment and the adjacent image segment can be calculated. Specifically, assuming that the feature vector corresponding to the segment feature of the image segment 1 is (x1, y1), and the feature vector corresponding to the segment feature of the image segment 2 is (x2, y2), the Euclidean distance d between the image segment 1 and the image segment 2 is:

[0093]

[0094] It can be understood that based on the Euclidean distance between the segment features, the image segments with common features can be determined. Specifically, the preset threshold of the Euclidean distance can be set to 250 nanometers, or the preset threshold can be set to other values according to the actual size of the pattern structure image or the density of the pattern structure image. If the Euclidean distance calculated for two image segments is less than the preset threshold, it indicates that the difference between the two image segments is small, and the two image segments have common features, that is, there is an edge between the two image segments, so that the local relationship in each pattern structure image can be accurately obtained.

[0095] In some embodiments, to provide more abundant state information and provide better decision basis for the target model, each segment can be encoded. For example, each image segment can be encoded by matrix-based encoding, Squish Pattern (SP) encoding, vector quantization encoding, and the like.

[0096] In some embodiments, based on the encoded image segments and the common features between the image segments, a generative model can be used to generate the image to be analyzed. Specifically, each encoded image segment can be abstracted as a vertex, and the common features between the image segments can generate an edge between two vertices, thereby forming a structure diagram to be analyzed. For example, the complete structure diagram to be analyzed can be generated by the generative model based on the encoded image segments and the common features. The generative model can be a generative adversarial network or a variational autoencoder, etc.

[0097] In some embodiments, to help the target model extract useful features in the structure diagram to be analyzed, each image segment can be encoded to determine the size, shape, and other features of the image segment, thereby providing more detailed basis for the target model to make decisions. For example, (102.3) can include:

[0098] (102.3.1) obtaining a preset neighborhood construction scale, and constructing a square neighborhood for each image segment based on the neighborhood construction scale;

[0099] (102.3.2) In each square neighborhood, the square neighborhood is divided into multiple grids according to the graphic edges of the image segment;

[0100] (102.3.3) Each grid in each square neighborhood is binarized to obtain a topological matrix corresponding to the square neighborhood.

[0101] (102.3.4) In the horizontal direction of the square neighborhood, the length of each grid is obtained, and the horizontal lengths of all grids are summarized to generate a horizontal matrix of the square neighborhood.

[0102] (102.3.5) In the vertical direction of the square neighborhood, the length of each grid is obtained, and the vertical lengths of all grids are summarized to generate a vertical matrix of the square neighborhood.

[0103] (102.3.6) Based on the topological matrix, the horizontal matrix and the vertical matrix, each encoded image segment is obtained.

[0104] The neighborhood construction scale can be the size of the scale used to construct the square neighborhood. It can be understood that the neighborhood construction scale can be adjusted according to the image density of the pattern structure image. For example, for a pattern structure image with a larger image density, a smaller neighborhood construction scale can be set to achieve more accurate encoding, and for a pattern structure image with a smaller image density, a larger neighborhood construction scale can be set to avoid unnecessary waste of computing resources. In some embodiments, the neighborhood construction scale can also be adjusted according to the lithography system parameters. For example, when the lithography system parameter is the minimum resolution of 100 nm (nanometer), the size of the square neighborhood should be less than or equal to this minimum resolution to ensure that the smallest size of the graphic element details can be processed. In some embodiments, when the lithography system parameter is the floating error, if the floating error is large, a certain conservative margin can be added to the size of the square neighborhood to ensure that the lithography system can process the graphics in this size range. For example, the square neighborhood can be expanded to 550 nm*550 nm. The specific neighborhood construction scale can be adjusted according to the actual situation, and the present application does not make specific limitations thereto.

[0105] The square neighborhood can be a square region centered on a certain pixel of the image segment and constructed around the center. Generally, the center pixel of the image segment can be selected as the center of the constructed square region. Further, the size of the square neighborhood is determined by the neighborhood construction scale, for example, the square neighborhood size can be 500 nanometers*500 nanometers, etc.

[0106] The graphic edge of the image segment can be the boundary line of the image shape contained in the image segment. The edge information of the image segment can be extracted by image segmentation or edge detection algorithms, and the image segment can be divided in the horizontal and vertical directions along the graphic edge to obtain at least one grid corresponding to the image segment.

[0107] The topology matrix can be a matrix representing the topological structure of the grid cells obtained by binarizing each grid cell in the square neighborhood. Specifically, the topology matrix is ​​a two-dimensional matrix where each element indicates whether a grid cell belongs to an image segment. In the topology matrix, a grid cell is marked with 1 if it belongs to an image segment, and marked with 0 if it does not. For example, in... Figure 3 In this context, the topological matrix is ​​denoted by M, specifically:

[0108]

[0109] The horizontal matrix can be obtained by taking the length of each grid cell in the horizontal direction of the square neighborhood and summing the horizontal lengths of all grid cells to obtain the horizontal matrix corresponding to the square neighborhood. For example, in Figure 3 In the middle, the horizontal matrix is ​​represented by δ. x Indicates, specifically, δ x =[8,7,63,5,65,20]

[0110] The vertical matrix can be obtained by taking the lengths of each grid cell in the vertical direction of the square neighborhood and summing the vertical lengths of all grid cells, resulting in the vertical matrix corresponding to the square neighborhood. For example, in Figure 3 In the middle, the horizontal matrix is ​​represented by δ. y Indicates, specifically, δ y =[64,70,35,28] T

[0111] In some implementations, since the dimensions of image segments encoded using compressed graphics cannot be customized by the user, it is necessary to convert the image segments to the ideal input dimensions. Specifically, the original topological matrix, horizontal matrix, and vertical matrix obtained by compressed graphics encoding can be adjusted one by one according to the target dimensions through interpolation or scaling to obtain the encoded image segments.

[0112] Take adjusting the topology matrix as an example (both the horizontal matrix and the vertical matrix can be adjusted in the following way), assuming that the topology matrix obtained after the image segment is encoded in the graph compression mode is a 1-dimensional vector, which is represented as: [1, 0, 0, 1, 1, 0, 0, 0, 1, 0]. Assuming that the target dimension is 10. First, according to the target dimension and the length of the topology matrix, the scaling ratio (scaling factor) of the two is calculated. In this example, the target dimension is 10, and the length of the original topology matrix is 10.

[0113] Specifically, the scaling ratio = target dimension / length of the original topology matrix = 10 / 10 = 1. After calculating the scaling ratio, the elements in the original topology matrix are interpolated using the scaling factor to fill the blank area of the target dimension. In this example, the interpolated topology matrix is: [1, 0, 0, 1, 1, 0, 0, 0, 1, 0];

[0114] It can be understood that if the target dimension is longer than the length of the original topology matrix, the final topology matrix will be the extended length; if the target dimension is shorter than the length of the original topology matrix, the final topology matrix will be the truncated length. For example, assuming that the target dimension is 5, the final topology matrix of the image segment is: [1, 0, 0, 1, 1].

[0115] In this way, the image segment can be encoded, and based on the encoded image segment and the common features between the image segments, a to-be-analyzed structure graph is generated, so that the target model can more accurately analyze the mask based on the to-be-analyzed structure graph, and obtain a correction decision label corresponding to each image segment.

[0116] Step 103, inputting the to-be-analyzed structure graph into the target model, the target model classifying based on the segment features and the common features in the to-be-analyzed structure graph to obtain a plurality of correction decision labels corresponding to each image segment, each correction decision label containing a correction offset and a corresponding probability value;

[0117] The target model is obtained by a preset model based on sample to-be-analyzed structure graphs and adjusted sample target images to maximize the learning of the correction index; the correction index is determined by the first edge placement error and the first process variation bandwidth corresponding to the sample to-be-analyzed structure graph, and the second edge placement error and the second process variation bandwidth corresponding to the sample target image; the sample target image is obtained by adjusting the sample to-be-analyzed structure graph according to the indication of the plurality of sample correction decision labels output by the preset model.

[0118] It can be understood that, in order to realize more accurate and efficient judgment on the position of each image segment in the structure diagram to be analyzed, the target model can directly obtain the correction offset and the corresponding probability value by inputting the structure diagram to be analyzed into the pre-trained target model, so as to determine whether each image segment needs to be moved and the scale of the movement according to the probability value.

[0119] The target model is trained by the preset model, and the target model and the preset model are both models in series connection of a graph neural network and a recurrent neural network. The graph neural network can extract features from the structure diagram to be analyzed, and then input the extracted features to the recurrent neural network for classification and decision-making. The target model can output a plurality of correction decision labels corresponding to each image segment based on the segment features and the common features in the structure diagram to be analyzed, so as to indicate the adjustment direction of each image segment. The preset model can perform maximum correction index learning, so as to gradually optimize the parameters of the model, and finally obtain the trained target model.

[0120] The correction decision label can be a correction offset of the image segment and a probability value corresponding to each correction offset, so as to determine the correction offset with the maximum probability value, and provide more fine classification and description for the adjustment of the image segment. Similarly, the sample correction decision label can be a sample correction offset of the sample image segment and a sample probability value corresponding to each sample correction offset.

[0121] The correction offset can be a plurality of correction actions corresponding to each image segment. For example, the correction offset can be moving inward by 2 nm, moving inward by 1 nm, keeping still, moving outward by 1 nm, moving outward by 2 nm, etc. During the training process of the preset model, different correction offsets can be trained, so that the target model has the ability to analyze and obtain the corresponding probability value under any different correction offset. Generally, when the target model analyzes the structure diagram to be analyzed, the correction offset can be selected by human, or the target model can select the correction offset according to the size of the structure diagram to be analyzed, or the correction offset can be randomly selected. After selecting the corresponding correction offset, the target model can directly output the probability value under each correction offset according to the structure diagram to be analyzed.

[0122] The sample structure diagram to be analyzed can be an image sample used for training and testing the preset model. The sample structure diagram to be analyzed can be classified by inputting into the preset model, so as to generate the corresponding sample correction decision label.

[0123] The sample target image can be obtained by inputting the sample structure to be analyzed into a preset model, obtaining a plurality of sample correction decision labels output by the preset model, and selecting a target correction offset according to a sample correction offset and a sample probability value in the sample correction decision labels to adjust the sample structure to be analyzed.

[0124] The correction index can be a reward value for adjusting the parameters of the preset model, and the correction index is determined by the edge placement error and the process variation bandwidth of the sample structure to be analyzed and the sample target image. It can be understood that by calculating the correction index, the training effect of the preset model can be intuitively displayed, so as to determine to punish or reward the preset model according to the correction index, and improve the classification ability of the preset model. In some embodiments, the calculation formula of the correction index is as follows:

[0125]

[0126] wherein r t is the correction index, EPE t is the first edge placement error, EPE t+1 is the second edge placement error, PVB t is the first process variation bandwidth, PVB t+1 is the second process variation bandwidth, and ε and β are constants which can be set by the user.

[0127] The first edge placement error and the second edge placement error are both edge placement errors and are calculated in the same way. The first process variation bandwidth and the second process variation bandwidth are both process variation bandwidths and are calculated in the same way. Specifically, the edge placement error is the positioning error of the edge of each image. The smaller the edge placement error, the more accurate and stable the positioning of the mask edge is, which meets the expected standard. The process variation bandwidth is the range or amplitude of the process variation in the mask. A smaller process variation bandwidth indicates that the process variation in the mask is smaller and the process consistency is better.

[0128] In some embodiments, taking the first edge placement error and the first process variation bandwidth as an example, when calculating the first edge placement error, the position difference of the edge can be compared according to the sample reference layout and the sample structure to be analyzed, and the position information of the edge can be obtained by using image processing algorithms such as edge detection and edge matching, and the deviation or deflection angle between the edges can be calculated as the first edge placement error. When calculating the first process variation bandwidth, the feature variation between different process layers in the mask can be compared according to the sample reference layout and the sample structure to be analyzed. Specifically, the range or amplitude of the process variation can be calculated as the first process variation bandwidth by analyzing the changes of circuit size and interlayer distance.

[0129] By the above mode, the segment features and common features in the to-be-analyzed structure diagram are classified by inputting the to-be-analyzed structure diagram into the trained target model, so that a more rapid and accurate classification result is obtained according to the local features in the to-be-analyzed structure diagram, and the final correction decision label is multiple, which avoids the problem that a single correction decision label cannot accurately indicate the adjustment direction, and is beneficial to subsequently selecting a more accurate target correction offset.

[0130] In some embodiments, in order to improve the accuracy of feature recognition and classification of the target model on the to-be-analyzed structure diagram, the preset model can be continuously trained, and the parameters of the preset model are adjusted according to the calculated correction index in the training process, so as to continuously improve the performance of the preset model, and finally obtain a target model with high fitness and high classification accuracy. The "target model" in step 103 can be trained by the following mode:

[0131] (A.1) Obtain a sample to-be-analyzed structure diagram, and input the sample to-be-analyzed structure diagram into the preset model to obtain a plurality of sample correction decision labels corresponding to each sample image segment; wherein the sample to-be-analyzed structure diagram is obtained by encoding a plurality of sample image segments;

[0132] (A.2) For each sample image segment, based on the sample probability value in each sample correction decision label corresponding to each sample image segment, a sample target correction offset of each sample image segment is selected;

[0133] (A.3) According to the sample target correction offset of each sample image segment, the sample to-be-analyzed structure diagram is adjusted to obtain an adjusted sample target image;

[0134] (A.4) Calculate the first edge placement error and the first process variation bandwidth corresponding to the sample to-be-analyzed structure diagram, and the second edge placement error and the second process variation bandwidth corresponding to the sample target image, and based on the first edge placement error, the first process variation bandwidth, the second edge placement error and the second process variation bandwidth, calculate the correction index of the preset model;

[0135] (A.5) Update the parameters of the preset model based on the correction index until the training times of the preset model reach the preset iteration times, and obtain the target model.

[0136] It can be understood that the sample to-be-analyzed structure diagram and the to-be-analyzed structure diagram have the same encoding mode, that is, based on the image segments, and the to-be-analyzed image is generated according to the encoded image segments and the common features between the image segments, which will not be described here.

[0137] The preset model is an untrained deep neural network model, and the deep neural network model is a model obtained by connecting a graph neural network and a recurrent neural network in series. By inputting the sample structure graph to be analyzed into the untrained preset model, the preset model can extract features and classify the sample structure graph to be analyzed, thereby obtaining multiple sample correction decision labels for each sample image segment. Specifically, each sample correction decision label can include a sample correction offset and a corresponding sample probability value.

[0138] Specifically, each sample image segment corresponds to output multiple sample correction decision labels. Since the sample probability values contained in the sample correction decision labels are not considered in combination with the technical knowledge of the correction corresponding field of the mask, a modulator can be configured to adjust each sample probability value, so that each sample probability value can more accurately reflect the direction of movement of each sample image segment.

[0139] Specifically, the first edge placement error of each sample image segment can be calculated based on the reference layout, and the position of the sample image segment relative to the reference layout is determined based on the first edge placement error, so that the probability of the sample image segment moving inward that deviates from the outside of the reference layout is greater, and the probability of the sample image segment moving outward that deviates from the inside of the reference layout is greater, thereby obtaining the adjusted sample probability value. Then, the sample correction offset corresponding to the maximum sample probability value is selected as the sample target correction offset from the adjusted sample probability value.

[0140] It can be understood that the maximum sample probability value often represents the maximum probability of the sample image segment executing the corresponding sample correction offset. For example, the sample correction offset of the sample image segment can be: moving inward by 2 nm, moving inward by 1 nm, keeping still, moving outward by 1 nm, moving outward by 2 nm, and the sample probability value corresponding to the sample correction offset is: 56%, 25%, 8%, 7%, 4%. Then, moving inward by 2 nm can be selected as the sample target correction offset.

[0141] Specifically, after the sample target correction offset is selected, each sample image segment in the sample structure graph to be analyzed can be adjusted according to the sample target correction offset of each sample image segment, to obtain an adjusted sample target image, so as to determine the accuracy of the preset model adjustment in subsequent steps.

[0142] For example, to determine the accuracy of the preset model adjustment, the first edge placement error and the first process variation bandwidth of the sample to-be-analyzed structure diagram can be calculated, and the second edge placement error and the second process variation bandwidth of the sample target image can be calculated. If the second edge placement error and the second process variation bandwidth of the sample target image are smaller than the first edge placement error and the first process variation bandwidth of the sample to-be-analyzed structure diagram, it indicates that the preset model can effectively reduce the original error after adjusting the sample to-be-analyzed structure diagram, and the preset model should be rewarded. If the second edge placement error and the second process variation bandwidth of the sample target image are larger than the first edge placement error and the first process variation bandwidth of the sample to-be-analyzed structure diagram, it indicates that the preset model makes the original error larger after adjusting the sample to-be-analyzed structure diagram, and the preset model should be punished. The calculation method of the first edge placement error, the first process variation bandwidth, the second edge placement error and the second process variation bandwidth has been described above, and will not be repeated here.

[0143] Specifically, the reward or punishment of the preset model can be determined by calculating a correction index. The calculation formula of the correction index is as follows:

[0144]

[0145] wherein, r t is the correction index, EPE t is the first edge placement error, EPE t+1 is the second edge placement error, PVB t is the first process variation bandwidth, PVB t+1 is the second process variation bandwidth, and ε and β are both constants and can be set by the user.

[0146] In some embodiments, when the correction index is positive, the preset model should be rewarded, and when the correction index is negative, the preset model should be punished. By setting a reasonable reward and punishment mechanism, the preset model can efficiently perform reinforcement learning, improve the feature extraction performance and classification performance of the preset model, and improve the accuracy of subsequent mask correction.

[0147] In some embodiments, after each training of the preset model is completed, the corresponding correction index is calculated, and the parameters of the preset model are adjusted based on the correction index, until a preset iteration number is reached. The preset iteration number can be adjusted according to actual conditions, for example, for some mask application scenarios with high precision, the preset iteration number can be appropriately increased, and for some mask application scenarios with low precision, the preset iteration number can be appropriately reduced, and the like.

[0148] In some embodiments, the training of the preset model can be stopped when the number of times that the second edge placement error is 0 exceeds a preset iteration number. Specifically, for example, when the second edge placement error is 0 within the preset iteration number, it indicates that the correction result of the preset model has no error and the model performance is better, at which time the training can be stopped to obtain the target model. For example, the preset iteration number can be set to 3 times, and the training of the preset model can be stopped to obtain the target model when the number of times that the second edge placement error is 0 exceeds 3 times.

[0149] By training the preset model in the above manner, the accuracy of the obtained target model can be higher and the performance can be better, which is beneficial to improving the efficiency and accuracy of the correction in the subsequent correction process of the mask.

[0150] In step 104, for each image segment, a target correction offset of the image segment is selected based on the probability value in each correction decision label corresponding to the image segment.

[0151] It can be understood that, in order to determine the final adjustment manner of the image segment, the target correction offset of the image segment can be selected based on the probability value in each correction decision label corresponding to the image segment, so as to quickly correct the image segment according to the target correction offset.

[0152] Specifically, in the correction decision label output by the target model, each correction offset corresponds to a probability value, and each probability value represents the possibility that the corresponding correction offset is selected as the target correction offset. For example, the correction decision label has an inward movement of 2 nm, an inward movement of 1 nm, no movement, an outward movement of 1 nm, and an outward movement of 2 nm, and the corresponding probability values are 40%, 31%, 11%, 10%, and 8%, respectively. Therefore, the probability value of the inward movement of 2 nm is the largest, and the inward movement can be selected as the target correction offset of the corresponding image segment.

[0153] In the above manner, the target correction offset can be selected based on the probability value in the correction decision label, so as to realize quick, individualized and adaptive correction adjustment, and improve the accuracy and effect of the adjustment of each image segment.

[0154] In some embodiments, in order to improve the accuracy of the correction of the mask, the probability value in each correction decision label can be adjusted in combination with the knowledge in the art, so as to realize more accurate correction of the mask. For example, step 104 can include:

[0155] (104.1) obtaining a reference layout corresponding to the mask;

[0156] (104.2) determining a positional deviation relationship of each image segment relative to the reference layout;

[0157] (104.3) For each image segment, based on the position deviation relationship, the probability value in each correction decision label corresponding to the image segment is adjusted to obtain the correction offset and the modified probability value of the probability value after adjustment of each image segment under each correction decision label.

[0158] (104.4) Based on the modified probability value in each correction decision label corresponding to each image segment, a target correction offset in the image segment is selected.

[0159] The reference layout is a standard image drawn on the wafer after accurate correction of the mask, and is used for correction and adjustment of the mask. The reference layout can also be referred to as a design pattern. The reference layout can be created by means of a computer-aided design software according to the related parameters and requirements of the mask, or can be obtained by other means. The embodiments of the present application do not make specific limitations in this regard.

[0160] The position deviation relationship can be the deviation direction of each image segment relative to the reference layout. For example, the position deviation relationship can be that the image segment is on the outside of the reference layout, the image segment is on the inside of the reference layout, and the like.

[0161] The modified probability value can be a value obtained by adjusting the probability value in combination with the knowledge in the art. The modified probability value is more accurate and more consistent with the actual situation than the probability value. Based on the modified probability value obtained by adjusting each probability value, the correction offset with the maximum modified probability value can be selected as the target correction offset.

[0162] In some embodiments, the position deviation relationship of each image segment relative to the reference layout can be determined by calculating the edge placement error of each image segment relative to the reference layout. For example, when the edge placement error is negative, it indicates that the contour of the image segment is on the inside of the reference layout; when the edge placement error is positive, it indicates that the contour of the image segment is on the inside and outside of the reference layout. For example, when the edge placement error is -5, it indicates that the contour of the image segment is located 5 nm inward of the reference layout.

[0163] It can be understood that when the position deviation relationship is larger, that is, the edge placement error is larger, it indicates that the probability of the corresponding image segment moving inward or outward is larger. Therefore, when the position deviation relationship is larger, the modulator can be used to adjust the moving scale, and the larger the moving scale, the larger the corresponding probability value, and the smaller the moving scale, the smaller the corresponding probability value, so as to accurately obtain the correction offset and the modified probability value of the probability value after adjustment of each image segment under each correction decision label.

[0164] Further, the correction offset with the maximum correction probability value can be selected as a target correction offset based on the correction probability values in each correction decision label corresponding to each image segment, and the corresponding image segment is accurately corrected based on the target correction offset.

[0165] In some embodiments, in order to accurately adjust the probability value according to the positional deviation relationship, the target modulation function can be used to calculate how the probability value corresponding to each correction offset should be adjusted under the corresponding positional deviation relationship, so as to combine the probability value obtained by classifying the features of the actual structure to be analyzed with the probability value obtained by combining the knowledge in the field, and obtain a more accurate correction probability value, thereby improving the accuracy of the mask correction. For example, (104.3) can include:

[0166] (104.3.1) selecting a target measurement point for each image segment and the reference layout;

[0167] (104.3.2) calculating the edge placement error of the image segment based on the positional deviation relationship of the target measurement point relative to the reference layout;

[0168] (104.3.3) obtaining a reference value and forming an error interval based on the reference value and the edge placement error;

[0169] (104.3.4) selecting an equal number of error reference values for the error interval as the correction decision label, and determining the corresponding relationship between each correction decision label and the error reference value according to the positional deviation relationship of the target measurement point relative to the reference layout;

[0170] (104.3.5) adjusting the probability value in each correction decision label corresponding to the image segment based on the corresponding relationship to obtain the correction probability value of the probability value adjustment of each image segment under each correction decision label.

[0171] The target measurement point is a reference point for each image segment to calculate the edge placement error with the reference layout. In order to ensure that the calculated edge placement error can correctly reflect the position of the image segment relative to the reference layout, the point closest to the reference layout in the image segment should be selected as much as possible. Alternatively, the center point of each image segment can also be selected, and so on. Regardless of the specific position of the target measurement point in the image segment, the selection method of each image segment should be the same, for example, the point closest to the reference layout should be selected as the target measurement point for each image segment to ensure the accuracy of the correction probability value.

[0172] The reference value is a value used to determine an ideal or standard position of the image segment. Generally, 0 is selected as the reference value because it is necessary to make the edge placement error as close to 0 as possible in the process of correcting the mask to ensure the accuracy of the correction.

[0173] The error interval represents the error range of the target measurement point selected by the image segment relative to the reference layout, and the error reference value is a value selected from the error interval and equal in number to the correction decision label, which represents the error degree of each correction decision label corresponding to the error reference value.

[0174] The corresponding relationship can be a mapping relationship between each correction decision label and the corresponding error reference value. After determining the mapping relationship between the correction decision label and the corresponding error reference value, the mapping relationship can be used for subsequent probability value adjustment.

[0175] In some embodiments, the correction offset of the image segment 1 corresponds to moving inward by 2 nm, moving inward by 1 nm, keeping still, moving outward by 1 nm, and moving outward by 2 nm, a total of 5 indicators. The edge placement error of the image segment 1 is -4, which is calculated. Therefore, the error interval can be determined based on the reference value and the edge placement error. When the reference value is 0, the error interval is [-4, 0]. Then, the error reference value is a value selected uniformly from the error interval and equal in number to the correction decision label, that is, the error reference value is -4, -3, -2, -1, and 0. The error reference value can be substituted into the preset target modulation function in turn to directly obtain the reference probability value. As the edge placement error indicates that the image segment 1 is located on the inner side of the reference layout, the probability of moving outward of the image segment 1 needs to be adjusted to be greater to make the edge placement error 0.

[0176] Please refer to Figure 4 , Figure 4 is a target modulation function mapping diagram. According to Figure 4 It can be seen that when the edge placement error is negative, the smaller the edge placement error (EPE), the greater the reference probability value. When the edge placement error is positive, the greater the edge placement error, the greater the reference probability value. Therefore, the corresponding relationship between the correction offset and the error reference value is that moving outward by 2 nm: -4; moving outward by 1 nm: -3; keeping still: -2; moving inward by 1 nm: -1; and moving inward by 2 nm: 0. Since the correction offset corresponds to the error reference value, the correction offset also corresponds to the reference probability value calculated from the error reference value, that is, when the edge placement error is -4, the probability of moving outward by 2 nm of the image segment is greater, thereby combining the image classification result and the knowledge in the art to make more accurate adjustments to each image segment.

[0177] Specifically, the probability value in the correction decision label can be adjusted based on the correspondence relationship, so as to select the correction offset corresponding to the maximum modified probability value as the target correction offset according to the modified probability value, and then adjust the corresponding image segment according to the target correction offset, thereby improving the accuracy of subsequent mask correction.

[0178] In some embodiments, in order to adjust the probability value in the correction decision label, the reference probability value corresponding to each error reference value can be calculated, and the modified probability value can be obtained by multiplying the probability value and the reference probability value, so that the obtained modified probability value can be more accurate, instead of only considering the features of the image to be detected. For example, (104.3.5) can include:

[0179] (104.3.5.1) obtaining a target modulation function corresponding to the mask;

[0180] calculating each error reference value according to the target modulation function to obtain a reference probability value corresponding to each error reference value;

[0181] (104.3.5.2) for each image segment, multiplying the modified probability value and the reference probability value corresponding to the error reference value based on the correspondence relationship to obtain the modified probability value of each image segment after adjusting the probability value under each correction decision label.

[0182] The target modulation function can be a function for modulating the error reference value, and the target modulation function can be used to calculate the corresponding reference probability value based on the error reference value, so as to multiply the probability value and the reference probability value to obtain the modified probability value. The specific form of the target modulation function is not limited, as long as it satisfies the following conditions: when the edge placement error is less than 0, the reference probability value corresponding to the correction offset outward is large; when the edge placement error is greater than 0, the reference probability value corresponding to the correction offset inward is large. For example, the target modulation function f(x) can be f(x) = -0.02x 4 +1.

[0183] The reference probability value is the reference probability value corresponding to the error reference value calculated by the target modulation function. The reference probability value represents the predicted probability value of the correction decision label under a specific error reference value.

[0184] In some embodiments, the error reference value can be directly substituted into the target modulation function to obtain the reference probability value corresponding to each target modulation function. For example, when the error reference value is -2, f(x) = -0.02*(-2) 4 +1 = 0.68, that is, the reference probability value is 0.68.

[0185] Further, the softmax function can be used to normalize each reference probability value to obtain a normalized reference probability value. Since each correction decision label has a one-to-one correspondence with the error reference value, each correction decision label also has a one-to-one correspondence with the reference probability value. Therefore, the correction probability value can be multiplied by the corresponding reference probability value based on the correspondence to obtain the adjusted correction probability value of each image segment under each correction decision label. The correction decision label corresponding to the maximum correction probability value is selected as the target correction offset, and the mask is corrected to obtain a more accurate target mask.

[0186] Step 105, correcting the mask according to the target correction offset of each image segment to obtain a target mask.

[0187] In some embodiments, after determining the target correction offset of each image segment in the mask, the corresponding segment of the image segment in the mask is adjusted one by one to obtain the target mask. For example, there are three image segments in the mask (only as an example), if the target correction offset of image segment 1 is to move inward by 2 nm, then the image segment 1 in the mask is moved inward by 2 nm, and the moving mode of the other image segments is the same as that of the image segment 1, which will not be described here.

[0188] The embodiment of the present application obtains a pattern structure image corresponding to a mask to be corrected; divides the pattern structure image to obtain a plurality of image segments, and generates a to-be-analyzed structure image based on segment features of each image segment and common features between the plurality of image segments; inputs the to-be-analyzed structure image into a target model, and the target model classifies based on the segment features and the common features in the to-be-analyzed structure image to obtain a plurality of correction decision labels corresponding to each image segment, each correction decision label including a correction offset and a corresponding probability value; wherein the target model is obtained by a preset model based on sample to-be-analyzed structure images and sample target images adjusted based on a maximum modification index learning; the modification index is determined by a first edge placement error and a first process variation bandwidth corresponding to the sample to-be-analyzed structure image, and a second edge placement error and a second process variation bandwidth corresponding to the sample target image; the sample target image is obtained by adjusting the sample to-be-analyzed structure image according to the indication of a plurality of sample correction decision labels output by the preset model; for each image segment, a target correction offset of the image segment is selected based on the probability value in each correction decision label corresponding to the image segment; and the mask is corrected according to the target correction offset of each image segment to obtain a target mask. In this way, not only the features of a single image segment can be considered, but also the common features between image segments can be considered, that is, the influence between various structures in the mask is considered, so that the probability value corresponding to the output correction offset can fit the actual result of the mask, thereby selecting a more accurate target correction offset. Meanwhile, the target model of the present application can be obtained by maximum modification index learning, and the modification index is calculated based on the edge placement error and the process variation bandwidth of the sample to-be-analyzed structure image and the sample correction image before correction, so that the performance learned during the target model correction process can be directly used to accurately adjust each image segment, without having to adjust the image segment in two stages, thereby improving the efficiency and accuracy of the mask correction.

[0189] Please refer to Figure 5 The embodiment of the present application also provides a mask correction device, which can implement the above-mentioned mask correction method, and the mask correction device comprises:

[0190] The obtaining module 51 is configured to obtain a pattern structure image corresponding to a mask to be corrected;

[0191] The generating module 52 is configured to divide the pattern structure image to obtain a plurality of image segments, and generate a to-be-analyzed structure image based on segment features of each image segment and common features between the plurality of image segments;

[0192] The classification module 53 is configured to input the structure diagram to be analyzed into a target model, and the target model classifies the structure diagram to be analyzed based on segment features and common features to obtain a plurality of correction decision labels corresponding to each image segment, each of the correction decision labels including a correction offset and a corresponding probability value. The target model is obtained by a preset model based on sample structure diagrams to be analyzed and sample target images adjusted according to a maximum correction index. The correction index is determined based on a first edge placement error and a first process variation bandwidth of the sample structure diagrams to be analyzed and a second edge placement error and a second process variation bandwidth of the sample target images. The sample target images are obtained by adjusting the sample structure diagrams to be analyzed according to the indication of the plurality of sample correction decision labels output by the preset model.

[0193] The selection module 54 is configured to select, for each image segment, a target correction offset of the image segment based on the probability value in each correction decision label corresponding to the image segment.

[0194] The correction module 55 is configured to correct the mask based on the target correction offset of each image segment to obtain a target mask.

[0195] The specific implementation of the mask correction device is basically the same as that of the above-mentioned mask correction method, and will not be repeated here. The mask correction device can also be provided with other functional modules to realize the mask correction method in the above-mentioned embodiments, as long as the requirements of the embodiments of the present application are met.

[0196] The embodiments of the present application also provide a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to realize the above-mentioned mask correction method. The computer device can be any intelligent terminal, such as a tablet computer or a vehicle-mounted computer.

[0197] Please refer to Figure 6 , Figure 6 The hardware structure of the computer device of another embodiment is illustrated, and the computer device includes:

[0198] The processor 61 can be implemented in the form of a general-purpose CPU (Central Processing Unit), a microprocessor, an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits, and is configured to execute a related program to implement the technical solutions provided by the embodiments of the present application.

[0199] The memory 62 can be implemented in the form of a Read Only Memory (ROM), a static storage device, a dynamic storage device, or a Random Access Memory (RAM), etc. The memory 62 can store an operating system and other application programs. When the technical solutions provided by the embodiments of the present application are implemented by software or firmware, the related program codes are stored in the memory 62 and are called and executed by the processor 61 to implement the mask plate correction method of the embodiments of the present application.

[0200] The input / output interface 63 is configured to realize information input and output.

[0201] The communication interface 64 is configured to realize communication interaction between the device and other devices. The communication can be realized by a wired manner (for example, a USB, a network cable, etc.) or a wireless manner (for example, a mobile network, a WI-FI, a Bluetooth, etc.).

[0202] The bus 65 is configured to transmit information between various components (for example, the processor 61, the memory 62, the input / output interface 63, and the communication interface 64) of the device.

[0203] The processor 61, the memory 62, the input / output interface 63, and the communication interface 64 are connected to each other through the bus 65 to realize communication connection between the device.

[0204] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the above-mentioned mask plate correction method.

[0205] The memory is a non-transitory computer readable storage medium, which can be used to store a non-transitory software program and a non-transitory computer executable program. In addition, the memory can include a high-speed random access memory and can also include a non-transitory memory, for example, at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory remotely arranged relative to the processor. These remote memories can be connected to the processor through a network. Examples of the network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0206] The embodiments described in the embodiments of the present application are used to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art can know that, with the evolution of technology and the appearance of new application scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0207] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation to the embodiments of the present application, and can include more or fewer steps than the figures, or combine certain steps, or different steps.

[0208] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, that is, can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments.

[0209] Those skilled in the art can understand that all or some steps in the above disclosed method, the function modules / units in the system and the device can be implemented as software, firmware, hardware and their appropriate combinations.

[0210] The terms "first", "second", "third", "fourth" and the like (if any) in the specification of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0211] It should be understood that in the present application, "at least one" and "several" refer to one or more, and "multiple" refers to two or more. "And / or" is used to describe the association between the associated objects, which means that there can be three relationships, for example, "A and / or B" can represent three cases: only A, only B, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0212] In several embodiments provided in the present application, it should be understood that the disclosed system and method can be implemented in other manners. For example, the system embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There can be another division manner for the actual implementation, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be electrical, mechanical or in other forms.

[0213] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0214] In addition, each functional unit in the various embodiments of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0215] If the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes multiple instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program storage media.

[0216] The preferred embodiments of the embodiments of the present application are described above with reference to the accompanying drawings, but this does not limit the scope of the rights of the embodiments of the present application. Any modifications, equivalent replacements and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall be within the scope of the rights of the embodiments of the present application.

Claims

1. A mask calibration method, characterized in that, The method includes: Obtain the pattern structure image corresponding to the mask to be corrected; The pattern structure image is divided into multiple image segments, and a structure diagram to be analyzed is generated based on the segment features of each image segment and the common features among the multiple image segments; The target model is input into the structural diagram to be analyzed. The target model classifies the segment features and common features in the structural diagram to be analyzed, and obtains multiple correction decision labels corresponding to each image segment. Each correction decision label contains a correction offset and a corresponding probability value. The target model is obtained by learning a preset model to maximize the correction index based on the sample structure diagram to be analyzed and the adjusted sample target image; the correction index is determined by the first edge placement error and the first process change bandwidth corresponding to the sample structure diagram to be analyzed, and the second edge placement error and the second process change bandwidth corresponding to the sample target image; the sample target image is obtained by adjusting the sample structure diagram to be analyzed according to the indications of multiple sample correction decision labels output by the preset model. For each image segment, the target correction offset of the image segment is selected based on the probability value in each correction decision label corresponding to the image segment; The mask is corrected based on the target correction offset of each image segment to obtain the target mask.

2. The mask correction method according to claim 1, characterized in that, The step of selecting the target correction offset for each image segment based on the probability value in each correction decision label corresponding to the image segment includes: Obtain the reference layout corresponding to the mask; Determine the positional offset of each image segment relative to the reference layout; For each image segment, based on the positional deviation relationship, the probability value in each correction decision label corresponding to the image segment is adjusted to obtain the correction offset of each image segment under each correction decision label and the adjusted probability value. The target correction offset in the image segment is selected based on the correction probability value in each correction decision label corresponding to each image segment.

3. The mask correction method according to claim 2, characterized in that, For each image segment, based on the positional deviation relationship, the probability value in each correction decision label corresponding to the image segment is adjusted to obtain the adjusted corrected probability value of each image segment under each correction decision label, including: For each image segment and the reference map, a target measurement point is selected; Based on the positional deviation of the target measurement point relative to the reference map, the edge placement error of the image segment is calculated; Obtain reference values ​​and form an error range based on the reference values ​​and the edge placement error; For the error interval, select an equal number of error reference values ​​as the correction decision labels, and determine the correspondence between each correction decision label and the error reference values ​​based on the positional deviation of the target measurement point relative to the reference map; Based on the correspondence, the probability value in each correction decision label corresponding to the image segment is adjusted to obtain the corrected probability value of each image segment under each correction decision label.

4. The mask correction method according to claim 3, characterized in that, The step of adjusting the probability value in each correction decision label corresponding to the image segment based on the correspondence to obtain the adjusted corrected probability value of each image segment under each correction decision label includes: Obtain the target modulation function corresponding to the mask; The reference probability value corresponding to each error reference value is obtained by calculating each error reference value according to the target modulation function; For each image segment, the corrected probability value is multiplied by the reference probability value corresponding to the error reference value based on the correspondence, to obtain the adjusted corrected probability value of each image segment under each correction decision label.

5. The mask correction method according to claim 1, characterized in that, The generation of the structure map to be analyzed based on the segment features of each image segment and the common features among multiple image segments includes: Calculate the Euclidean distance between the segment features corresponding to each of the image segments and the adjacent segment features; Based on the Euclidean distance between the features of each segment, image segments with common features are identified; Each of the image segments is image encoded to obtain the encoded image segments; Based on the encoded image segments and the common features among the image segments, a structural diagram to be analyzed is generated.

6. The mask correction method according to claim 5, characterized in that, The step of encoding each of the image segments to obtain the encoded image segments includes: Obtain a preset neighborhood construction scale, and construct a square neighborhood for each image segment based on the neighborhood construction scale; Within each square neighborhood, the image fragment is divided according to its graphic edges to obtain multiple grids after the square neighborhood is divided. Binarize each grid in each of the square neighborhoods to obtain the topology matrix corresponding to the square neighborhood; In the horizontal direction of the square neighborhood, the length of each grid is obtained, and the horizontal lengths of all the grids are summarized to generate the horizontal matrix of the square neighborhood; In the vertical direction of the square neighborhood, the length of each grid is obtained, and the vertical lengths of all the grids are summed to generate the vertical matrix of the square neighborhood; Based on the topology matrix, the horizontal matrix, and the vertical matrix, the encoded image segments are obtained.

7. The mask correction method according to claim 1, characterized in that, The target model is trained in the following way: A sample structure diagram to be analyzed is obtained, and the sample structure diagram to be analyzed is input into a preset model to obtain multiple sample correction decision labels corresponding to each sample image segment; wherein, the sample structure diagram to be analyzed is obtained by encoding multiple sample image segments; For each of the sample image segments, based on the sample probability value in each sample correction decision label corresponding to each of the sample image segments, the sample target correction offset of each of the sample image segments is selected; Based on the sample target correction offset of each sample image segment, the sample structure diagram to be analyzed is adjusted to obtain the adjusted sample target image; Calculate the first edge placement error and the first process variation bandwidth corresponding to the sample structure diagram to be analyzed, and the second edge placement error and the second process variation bandwidth corresponding to the sample target image, and calculate the correction index of the preset model based on the first edge placement error, the first process variation bandwidth, the second edge placement error and the second process variation bandwidth; The parameters of the preset model are updated based on the correction index until the preset model has been trained for a preset number of iterations, thus obtaining the target model.

8. A mask calibration device, characterized in that, The device includes: The acquisition module is used to acquire the pattern structure image corresponding to the mask to be corrected; The generation module is used to divide the pattern structure image into multiple image segments, and generate a structure diagram to be analyzed based on the segment features of each image segment and the common features among the multiple image segments; A classification module is used to input the structure image to be analyzed into a target model. The target model classifies the structure image based on the segment features and common features, obtaining multiple correction decision labels for each image segment. Each correction decision label includes a correction offset and a corresponding probability value. The target model is obtained by a preset model through maximizing correction metrics based on the sample structure image to be analyzed and the adjusted sample target image. The correction metrics are determined by the first edge placement error and the first process variation bandwidth corresponding to the sample structure image to be analyzed, and the second edge placement error and the second process variation bandwidth corresponding to the sample target image. The sample target image is obtained by adjusting the sample structure image to be analyzed according to the instructions of the multiple sample correction decision labels output by the preset model. The selection module is used to select the target correction offset of each image segment based on the probability value in each correction decision label corresponding to the image segment; The correction module is used to correct the mask according to the target correction offset of each image segment to obtain the target mask.

9. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the mask correction method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the mask correction method according to any one of claims 1 to 7.

Images