Wafer image quality evaluation method and device based on position coding, equipment and storage medium

Abstract

The application discloses a wafer image quality evaluation method based on position coding, which comprises the following steps: determining a plurality of cut embedding vectors corresponding to a plurality of cut target image blocks and a plurality of sliding window embedding vectors corresponding to a plurality of sliding window target image blocks based on a target wafer image; adding position coding to each cut embedding vector to form an initial sequence; adding a data capture mark at the starting position of the initial sequence to capture global information of the target wafer image, so as to obtain global features of the target wafer image; determining a plurality of sliding window local features based on each sliding window embedding vector; determining global splicing features of the target wafer image based on the global features and the plurality of sliding window local features; outputting a quality prediction index of the target wafer image based on a quality evaluation model; and determining a quality evaluation result of the target wafer image based on the quality prediction index and a quality parameter index. The application combines local features and global features of an image, and improves the comprehensiveness and accuracy of image quality evaluation.

Description

Technical Field

[0001] This invention belongs to the field of wafer quality assessment technology, specifically relating to a wafer image quality assessment method, apparatus, device, and storage medium based on location coding. Background Technology

[0002] Wafer images are commonly used to detect defects and contamination during chip manufacturing. Therefore, accurate and efficient image quality assessment (IQA) is crucial for ensuring product quality and production efficiency. In practical applications, many wafer images lack corresponding high-quality reference images, rendering reference-based quality assessment methods inapplicable. Existing image quality assessment methods mainly include traditional methods based on image features and deep learning methods based on convolutional neural networks. However, these methods often focus on local regions and lack global context modeling capabilities, preventing them from comprehensively understanding the overall and detailed information of the image during quality assessment. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide a method, apparatus, device, and storage medium for wafer image quality assessment based on location encoding. This solves the problem of low accuracy in wafer image quality assessment in existing technologies.

[0004] According to one aspect of this application, a wafer image quality assessment method based on position encoding is disclosed, the method comprising:

[0005] Acquire the target wafer image;

[0006] Based on the target wafer image, multiple segmentation target image blocks and multiple sliding window target image blocks are determined;

[0007] Based on the multiple segmentation target image blocks, multiple segmentation embedding vectors are determined;

[0008] Based on the multiple sliding window target image blocks, multiple sliding window embedding vectors are determined;

[0009] Add positional encoding to each of the segmented embedding vectors to form an initial sequence;

[0010] A global feature determination module is used to add a data capture marker at the beginning position of the initial sequence to capture global information of the target wafer image according to the initial sequence, thereby obtaining global features of the target wafer image. The global features include image segmentation features of each segmented target image block and position information of each segmented target image in the target wafer image.

[0011] Based on each sliding window embedding vector, multiple sliding window local features are determined, and each sliding window target image block corresponds to one sliding window local feature;

[0012] Based on the global features and multiple sliding window local features, the global stitching features of the target wafer image are determined;

[0013] The global stitching features are input into the quality assessment model, and after parsing the global stitching features based on the quality assessment model, the quality prediction index of the target wafer image is output.

[0014] Obtain quality parameter indicators;

[0015] Based on the quality prediction index and the quality parameter index, the quality assessment result of the target wafer image is determined.

[0016] In some embodiments, determining multiple segmentation embedding vectors based on multiple segmentation target image patches includes:

[0017] Each of the segmented target image blocks is subjected to dimensional transformation to obtain multiple one-dimensional first transformed image blocks;

[0018] Each of the one-dimensional first transformed image blocks is linearly projected onto the target dimension to obtain multiple segmentation embedding vectors.

[0019] In some embodiments, adding positional encoding to each of the segmented embedding vectors to form an initial sequence includes:

[0020] Based on the sine and cosine functions, global position information is appended to each segmented embedding vector to obtain the position encoding of each segmented embedding vector relative to the global position information. The position encoding formula includes:

[0021] Row index:

[0022]

[0023] Column index:

[0024]

[0025] Where n is the dimension index of the position encoding, and the value of n ranges from 1 to 2. d model This represents the total dimension of the positional encoding, where i and j are the row and column positions of the current segmented target image patch. For the row dimension encoding of row position i of each segmented target image patch, even-numbered dimensions 2n are determined using a sine function, and odd-numbered dimensions 2n+1 are determined using a cosine function. Similarly, for the column dimension encoding of column position j of each segmented target image patch, even-numbered dimensions 2n are determined using a sine function, and odd-numbered dimensions 2n+1 are determined using a cosine function.

[0026] The initial sequence is:

[0027] POS (i，j) =[POS (r)，2n POS (r)，2n+1 POS (c)，2n POS (c)，2n+1 ].

[0028] In some embodiments, determining the global stitching features of the target wafer image based on the global features and multiple sliding window local features includes:

[0029] The global features and multiple sliding window local features are stitched together based on a stitching formula to determine the global stitching features of the target wafer image. The stitching formula is as follows:

[0030]

[0031] Where Z represents the global stitching feature, E h，q For local features of the sliding window, h is the row encoding of each sliding window target image block, and q is the column encoding of each sliding window target image block.

[0032] In some embodiments, determining a plurality of sliding window target image blocks based on the target wafer image includes:

[0033] Determine the target step size;

[0034] Based on the target step size, the target wafer image is sampled using a sliding window to obtain multiple sliding window target image blocks;

[0035] Multiple sliding window target image blocks include:

[0036] N = N r ×N c ;

[0037]

[0038] Where N is the number of target image patches in the sliding window, N r Where Nc is the number of rows in the target wafer image, H and W are the height and width of the target wafer image, respectively, and P is the number of columns in the target wafer image. H and P W These represent the height and width of the window during the sliding sampling process, respectively.

[0039] In some embodiments, determining multiple sliding window embedding vectors based on multiple sliding window target image patches includes:

[0040] Each sliding window target image block is subjected to dimensional transformation to obtain multiple one-dimensional second transformed image blocks;

[0041] Each of the one-dimensional second transformed image blocks is linearly projected onto the target dimension to obtain multiple sliding window embedding vectors.

[0042] In some embodiments, prior to acquiring the target wafer image, the method further includes:

[0043] Obtain the raw wafer image;

[0044] The original wafer image is processed to obtain the target wafer image.

[0045] According to another aspect of this application, a wafer image quality assessment device based on position coding is also disclosed, the device comprising:

[0046] The target wafer image acquisition module is used to acquire images of the target wafer.

[0047] The target image block determination module is used to determine multiple segmentation target image blocks and multiple sliding window target image blocks based on the target wafer image;

[0048] The segmentation embedding vector determination module is used to determine multiple segmentation embedding vectors based on multiple segmentation target image blocks;

[0049] The sliding window embedding vector determination module is used to determine multiple sliding window embedding vectors based on multiple sliding window target image blocks;

[0050] An initial sequence forming module is used to add positional encoding to each of the segmented embedding vectors to form an initial sequence;

[0051] At the beginning of the initial sequence, a data capture marker is added to capture global information of the target wafer image according to the initial sequence, thereby obtaining global features of the target wafer image. The global features include image segmentation features of each segmented target image block and position information of each segmented target image in the target wafer image.

[0052] The sliding window local feature determination module is used to determine multiple sliding window local features based on each sliding window embedding vector, wherein each sliding window target image block corresponds to one sliding window local feature;

[0053] A global stitching feature determination optical module is used to determine the global stitching features of the target wafer image based on the global features and multiple sliding window local features;

[0054] The indicator prediction module is used to input the global stitching features into the quality assessment model, and output the quality prediction indicator of the target wafer image after parsing the global stitching features based on the quality assessment model.

[0055] The quality parameter index acquisition module is used to acquire quality parameter indexes;

[0056] The quality assessment result determination module is used to determine the quality assessment result of the target wafer image based on the quality prediction index and the quality parameter index.

[0057] According to another aspect of this application, an electronic device is also disclosed, the electronic device including a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to cause the electronic device to perform various steps of the location-encoded wafer image quality assessment method as described in any of the preceding claims.

[0058] According to another aspect of this application, a computer-readable storage medium is also disclosed, on which instructions are stored, which, when executed by a processor, implement the steps of the location-encoded wafer image quality assessment method as described in any of the preceding claims.

[0059] The present invention includes, but is not limited to, the following beneficial effects: (1) This solution, on the one hand, by segmenting the target image block and adding position encoding, can effectively capture the global information of the wafer image, including the image features of each image block and its position information in the wafer image, which can comprehensively reflect the overall structure and distribution characteristics of the wafer image. On the other hand, by extracting local features through sliding window operation, it can capture the detailed information of the wafer image, especially the defect or abnormal pattern in the local area. Furthermore, by splicing the global features with the local features of the sliding window to form a global splicing feature, it can integrate and utilize global and local information to improve the comprehensiveness and accuracy of wafer quality assessment; (2) This solution, by setting a preset step size, can cover more image areas, especially overlapping areas, through flexible window division and step size setting of the target image block of the sliding window, thereby improving the ability to capture local details, thus enabling better detection of small defects and local anomalies in the wafer image, thereby improving the precision of the assessment; (3) By adding data capture markers at the beginning position of the initial sequence, it can provide a clear target for the extraction of global features. (4) This scheme transforms the original image block into a one-dimensional representation by performing dimensional transformation on each segmented target image block, which simplifies the data structure and reduces computational complexity. The embedding vector obtained by linear projection after dimensional transformation can be input into the global information capture module in a fixed dimension form, ensuring that the data of high efficiency and accuracy of global feature extraction is more suitable for subsequent linear projection operations, thereby improving computational efficiency and adapting to the needs of efficient batch processing. (5) This scheme can capture the spatial position information of the segmented target image block in the wafer image by adding position encoding to each segmented embedding vector. The introduction of position encoding can make up for the problem of insufficient global feature expression ability caused by the loss of spatial position information in the traditional image block segmentation method. Furthermore, the position encoding adopts the periodic change mode of sine and cosine functions to provide unique position information representation for embedding vectors of different dimensions. The periodic design of sine and cosine functions enables the model to perceive the relative positional relationship between image blocks, thereby effectively capturing the spatial structural features of the wafer image. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0061] Figure 1 This is a flowchart of a wafer image quality assessment method based on position encoding according to an embodiment of this application;

[0062] Figure 2 This is another flowchart of a wafer image quality assessment method based on position encoding according to an embodiment of this application;

[0063] Figure 3 This is another flowchart of a wafer image quality assessment method based on position encoding according to an embodiment of this application;

[0064] Figure 4 This is another flowchart of a wafer image quality assessment method based on position encoding according to an embodiment of this application;

[0065] Figure 5 This is a structural block diagram of a location-coded image quality assessment device according to an embodiment of this application;

[0066] Figure 6 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention;

[0067] Figure 7 This is a schematic diagram of the model framework for wafer image quality assessment based on position encoding, according to an embodiment of the present invention. Detailed Implementation

[0068] This invention discloses a wafer image quality assessment method based on position encoding. The method includes: acquiring a target wafer image; determining multiple segmented target image blocks and multiple sliding window target image blocks based on the target wafer image; determining multiple segmentation embedding vectors based on the multiple segmented target image blocks; determining multiple sliding window embedding vectors based on the multiple sliding window target image blocks; adding position encoding to each segmentation embedding vector to form an initial sequence; adding a data capture marker at the beginning of the initial sequence to capture global information of the target wafer image according to the initial sequence, thereby obtaining global features of the target wafer image. This method includes image segmentation features for each segmented target image block and the positional information of each segmented target image within the target wafer image. Based on each sliding window embedding vector, multiple sliding window local features are determined, with each sliding window target image block corresponding to one sliding window local feature. Based on global features and multiple sliding window local features, global stitching features of the target wafer image are determined. These global stitching features are input into a quality assessment model, which, after parsing the global stitching features, outputs a quality prediction index for the target wafer image. Quality parameter indicators are obtained. Based on the quality prediction index and the quality parameter indicators, the quality assessment result of the target wafer image is determined. This method integrates local and global image features, improving the comprehensiveness and accuracy of image quality assessment.

[0069] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0070] To facilitate understanding, the following will be combined with Figure 7 The principle and process of this invention are described in detail in the specific process of the embodiments of the invention. Figure 1 A flowchart of a location-encoded wafer quality assessment method, such as Figure 1 As shown, it includes the following steps:

[0071] S100, Obtain the target wafer image.

[0072] Specifically, in some examples, raw wafer images can be acquired using a high-resolution camera, or through computed tomography (CT) or X-ray imaging: the acquired 3D image of the interior of the raw wafer is converted into a 2D image, and this converted 2D image is used as the target wafer image. Furthermore, the target wafer image can be directly obtained from an image database, which pre-stores multiple wafer images. It is understood that the original wafer image is a high-resolution 2024×2024 image, resized to 224×224 pixels.

[0073] S102. Determine multiple slicing target image blocks and multiple sliding window target image blocks based on the target wafer image.

[0074] Specifically, the target wafer image can be directly segmented to form multiple target image blocks.

[0075] Furthermore, in one example, Figure 2 This is an exemplary illustration of determining multiple sliding window target image blocks based on a target wafer image, such as... Figure 2 As shown, it includes the following steps:

[0076] S200, Determine the target step size.

[0077] Specifically, the target step size can be preset, for example, it can be 124, 256, or 224, etc. In this example, the step size is 224. It can be understood that the window size or position can be adaptively adjusted according to the target's position or size. The specific step size can be set based on actual needs; no specific limitation is made here.

[0078] S202. Based on the target step size, perform sliding window sampling on the target wafer image to obtain multiple sliding window target image blocks.

[0079] In this example, multiple sliding window target image patches can be determined based on the following formula:

[0080] N = N r ×N c ;

[0081]

[0082] Where N is the number of target image patches in the sliding window, N r Where Nc is the number of rows in the target wafer image, H and W are the height and width of the target wafer image, respectively, and P is the number of columns in the target wafer image. H and P W These represent the height and width of the window during the sliding sampling process, respectively.

[0083] Understandably, in this example, by setting a preset step size, the sliding window target image block can cover more image areas, especially overlapping areas, through flexible window division and step size settings, thereby improving the ability to capture local details and thus better detecting minute defects and local anomalies in the wafer image, thereby improving the precision of the evaluation.

[0084] S104. Based on multiple segmented target image blocks, determine multiple segmentation embedding vectors.

[0085] In one example, Figure 3 This is a detailed explanation of step S104, which determines multiple segmentation embedding vectors based on multiple segmentation target image patches. See [link to relevant documentation]. Figure 3 It includes the following steps:

[0086] S300. Each segmented target image block undergoes dimensional transformation to obtain multiple one-dimensional first transformed image blocks.

[0087] Specifically, each segmented target image block can be dimensionality reduced to obtain multiple one-dimensional first-transformation image blocks.

[0088] S302. Linearly project each one-dimensional first transformed image block onto the target dimension to obtain multiple segmentation embedding vectors.

[0089] Specifically, the one-dimensional first transformed image patch obtained after dimensionality reduction can be mapped to a fixed-dimensional vector space through a linear projection layer (usually a fully connected layer) to obtain multiple segmentation embedding vectors.

[0090] This scheme simplifies the data structure and reduces computational complexity by performing dimensional transformation on each segmented target image block, converting the original image block into a one-dimensional representation. The embedding vector obtained by linear projection after dimensional transformation can be input into the global information capture module in a fixed-dimensional form, ensuring that the data with high efficiency and accuracy of global feature extraction is more suitable for subsequent linear projection operations, thereby improving computational efficiency and adapting to the needs of efficient batch processing.

[0091] S106. Based on multiple sliding window target image blocks, determine multiple sliding window embedding vectors.

[0092] In one example, Figure 4 This is a detailed explanation of S106, which determines multiple sliding window embedding vectors based on multiple sliding window target image patches. See [link to relevant documentation]. Figure 4 It includes the following steps:

[0093] S400. Perform dimensional transformation on each sliding window target image block to obtain multiple one-dimensional second transformed image blocks.

[0094] Specifically, each sliding window target image block can be dimensionality reduced to obtain multiple one-dimensional second-transformed image blocks.

[0095] S402. Linearly project each one-dimensional second-transformed image block to the target dimension to obtain multiple sliding window embedding vectors.

[0096] Specifically, the one-dimensional second-transformed image patch obtained after dimensionality reduction can be mapped to a fixed-dimensional vector space through a linear projection layer (usually a fully connected layer) to obtain multiple sliding window embedding vectors. In this example, the fixed dimension can be the same as the fixed dimension in step S302, for example, the fixed dimension can be 768×512.

[0097] S108. Add positional encoding to each segmented embedding vector to form the initial sequence.

[0098] In one example, global positional information can be appended to each segmented embedding vector based on sine and cosine functions to obtain the positional encoding of each segmented embedding vector relative to the global positional information. The positional encoding formula includes:

[0099] Row index:

[0100]

[0101] Column index:

[0102]

[0103] Where n is the dimension index of the position encoding, and the value of n ranges from 1 to 2. d model This represents the total dimension of the positional encoding, where i and j are the row and column positions of the current segmented target image patch. For the row dimension encoding of row position i of each segmented target image patch, even-numbered dimensions 2n are determined using a sine function, and odd-numbered dimensions 2n+1 are determined using a cosine function. Similarly, for the column dimension encoding of column position j of each segmented target image patch, even-numbered dimensions 2n are determined using a sine function, and odd-numbered dimensions 2n+1 are determined using a cosine function.

[0104] The initial sequence is:

[0105] POS (i，j)， =[POS (r)，2n POS (r)，2n+1 POS (c)，2n POS (c)，2n+1 ].

[0106] S110. At the beginning position of the initial sequence, add a data capture marker to capture global information of the target wafer image according to the initial sequence, and obtain the global features of the target wafer image.

[0107] Specifically, the global features include the image segmentation features of each segmented target image block and the positional information of each segmented target image within the target wafer image. In one example, the data capture token can be a ClassificationToken ([CLS]token), which can provide the model with a vector representing the overall information of the input sequence. In one example, steps S100-S110 are all implemented based on a position-encoded concatenated VisionTransformer (ViT) architecture, where the [CLS]token is obtained through L layers of VisionTransformer, such as the 10th layer.

[0108] By adding positional encoding to each segmentation embedding vector, the spatial positional information of the segmentation target image block in the wafer image can be captured. The introduction of positional encoding can make up for the problem of insufficient global feature representation caused by the loss of spatial positional information in traditional image block segmentation methods. Furthermore, the positional encoding adopts the periodic variation of sine and cosine functions to provide unique positional information representation for embedding vectors of different dimensions. The periodic design of sine and cosine functions enables the model to perceive the relative positional relationship between image blocks, thereby effectively capturing the spatial structural features of the wafer image.

[0109] S112. Based on each sliding window embedding vector, determine multiple local features of the sliding window.

[0110] Specifically, in this example, the feature corresponding to each sliding window vector is the local feature of each sliding window target image patch. That is, each sliding window target image patch corresponds to a local feature of the sliding window.

[0111] S114. Based on global features and multiple sliding window local features, determine the global stitching features of the target wafer image.

[0112] Specifically, global features and multiple sliding window local features can be stitched together based on a stitching formula to determine the global stitching features of the target wafer image. The stitching formula is as follows:

[0113]

[0114] Where Z represents the global stitching feature, E h，q For local features of the sliding window, h is the row encoding of each sliding window target image block, and q is the column encoding of each sliding window target image block.

[0115] Understandably, steps S112-S114 continue to be based on the Transformer architecture.

[0116] S116. Input the global stitching features into the quality assessment model, and output the quality prediction index of the target wafer image after parsing the global stitching features based on the quality assessment model.

[0117] In one example, the quality assessment model could be a multilayer perceptron (MLP) model, which outputs an image quality score after passing through a ReLU activation function. Understandably, the quality assessment model is pre-trained. During training, the MLP output is compared with the actual human mean opinion score (MOS) to calculate the mean squared error (MSE). The MSE is then used as a loss function to optimize the model's prediction accuracy, making the predicted image quality score closer to the actual human scores in the dataset.

[0118] The loss function can be,

[0119] y is the predicted value of the i-th sample (the result output by the model). i is the true value of the i-th sample, and N is the number of samples.

[0120] S118. Obtain quality parameter indicators.

[0121] Specifically, the quality parameter index can refer to the quality score, which can be preset, such as 8 points, 7 points or 5 points, etc.

[0122] S120. Based on quality prediction indicators and quality parameter indicators, determine the quality assessment result of the target wafer image.

[0123] Specifically, the predicted index can be compared with the quality parameter index. If the quality score corresponding to the predicted index is greater than the quality score corresponding to the quality parameter index, the quality of the target wafer image is considered to be qualified; otherwise, it is considered unqualified.

[0124] Furthermore, the effectiveness of the image quality assessment model can be verified by calculating PLCC and SRCC between the image quality score predicted by the model and the human score. The formula for PLCC is as follows:

[0125]

[0126] in and Let y be the average of the true quality score and the predicted quality score of the i-th sample image, respectively. i and Let be the true quality score and the predicted quality score of the i-th sample image, respectively.

[0127] The formula for SRCC is as follows:

[0128]

[0129] Where d i It is the difference between the true ranking and the predicted ranking of the i-th sample.

[0130] Based on the above verification, it can be seen that the prediction effect of this scheme is good.

[0131] Understandably, this scheme, on the one hand, by segmenting the target image blocks and adding positional encoding, can effectively capture global information of the wafer image, including the image features of each image block and its position information in the wafer image, which can comprehensively reflect the overall structure and distribution characteristics of the wafer image. On the other hand, by extracting local features through sliding window operation, it can capture detailed information of the wafer image, especially defect or abnormal patterns in local areas. Furthermore, by splicing global features with sliding window local features to form global spliced ​​features, it can integrate and utilize global and local information to improve the comprehensiveness and accuracy of wafer quality assessment. Furthermore, by adding data capture markers at the beginning position of the initial sequence, it can provide a clear target for the extraction of global features and improve the quality of global feature representation.

[0132] Furthermore, Figure 5 For a location-encoded wafer image quality assessment device, such as Figure 5 As shown, the device includes:

[0133] The target wafer image acquisition module is used to acquire images of the target wafer.

[0134] The target image block determination module is used to determine multiple segmentation target image blocks and multiple sliding window target image blocks based on the target wafer image;

[0135] The segmentation embedding vector determination module is used to determine multiple segmentation embedding vectors based on multiple segmentation target image patches;

[0136] The sliding window embedding vector determination module is used to determine multiple sliding window embedding vectors based on multiple sliding window target image patches;

[0137] The initial sequence forming module is used to add positional encoding to each segmented embedding vector to form an initial sequence;

[0138] The global feature determination module is used to add data capture markers at the beginning of the initial sequence to capture global information of the target wafer image according to the initial sequence, and obtain the global features of the target wafer image. The global features include the image segmentation features of each segmented target image block and the position information of each segmented target image in the target wafer image.

[0139] The sliding window local feature determination module is used to determine multiple sliding window local features based on each sliding window embedding vector, with each sliding window target image patch corresponding to one sliding window local feature;

[0140] The global stitching feature determination optical module is used to determine the global stitching features of the target wafer image based on global features and multiple sliding window local features;

[0141] The indicator prediction module is used to input global stitching features into the quality assessment model, and output the quality prediction indicators of the target wafer image after parsing the global stitching features based on the quality assessment model.

[0142] The quality parameter index acquisition module is used to acquire quality parameter indexes;

[0143] The quality assessment result determination module is used to determine the quality assessment result of the target wafer image based on quality prediction indicators and quality parameter indicators.

[0144] The application of the relevant modules of the device in this example can be referred to the relevant introduction of the method principle above, and will not be repeated here.

[0145] According to another aspect of this application, this application also discloses an electronic device, which includes a memory and at least one processor, wherein instructions are stored in the memory; the at least one processor invokes the instructions in the memory to cause the electronic device to perform the various steps of the position-encoded wafer image quality assessment method described above.

[0146] The above figure describes in detail the wafer image quality assessment device based on position encoding in the embodiment of the present invention from the perspective of modular functional entities. The following describes in detail the electronic device in the embodiment of the present invention from the perspective of hardware processing.

[0147] Figure 6 This is a schematic diagram of the structure of an electronic device 600 provided in an embodiment of the present invention. The electronic device 600 can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) for storing application programs 733 or data 632. The memory 620 and storage media 630 can be temporary or persistent storage. The program stored in the storage media 630 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the electronic device 600. Furthermore, the processor 610 may be configured to communicate with the storage media 630 and execute the series of instruction operations in the storage media 630 on the electronic device 600.

[0148] Electronic device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input / output interfaces 660, and / or one or more operating systems 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 7 The illustrated electronic device structure does not constitute a limitation on electronic devices and may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0149] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of a position-encoded wafer image quality assessment method.

[0150] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0151] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0152] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A wafer image quality evaluation method based on position coding, characterized in that, The method comprises: acquiring a target wafer image; determining a plurality of cut target image blocks and a plurality of sliding window target image blocks based on the target wafer image; determining a plurality of cut embedding vectors based on a plurality of the cut target image blocks; determining a plurality of sliding window embedding vectors based on a plurality of the sliding window target image blocks; adding position encoding to each of the cut embedding vectors to form an initial sequence; adding a data capture mark at a starting position of the initial sequence to capture global information of the target wafer image according to the initial sequence, to obtain a global feature of the target wafer image, the global feature comprising image cut features of each of the cut target image blocks and position information of each of the cut target image blocks in the target wafer image; determining a plurality of sliding window local features based on each of the sliding window embedding vectors, each of the sliding window target image blocks corresponding to a sliding window local feature; determining a global stitching feature of the target wafer image based on the global feature and a plurality of the sliding window local features; inputting the global stitching feature into a quality evaluation model, and outputting a quality prediction index of the target wafer image after analyzing the global stitching feature based on the quality evaluation model; acquiring a quality parameter index; determining a quality evaluation result of the target wafer image based on the quality prediction index and the quality parameter index. 2.The wafer image quality evaluation method based on position coding according to claim 1, wherein, The method comprises: performing dimension conversion on each of the cut target image blocks to obtain a plurality of one-dimensional first converted image blocks; linearly projecting each of the one-dimensional first converted image blocks to a target dimension to obtain a plurality of the cut embedding vectors. 3.The wafer image quality evaluation method based on position coding according to claim 1, wherein, The method comprises: applying global position information to each of the cut embedding vectors based on a sine function and a cosine function to obtain position encoding of each of the cut embedding vectors relative to the global position information, wherein the position encoding formula comprises: a row index: a column index: wherein n is the dimension index of the position coding, n is in the range of d model is the total dimension of the position coding, i and j are the row position and column position of the current split target image block, for the row dimension coding of each split target image block row position i, the even dimension 2n is determined using a sine function, and the odd dimension 2n+1 is determined using a cosine function; similarly, for the column dimension coding of each split target image block column position j, the even dimension 2n uses a sine function, and the odd dimension 2n+1 uses a cosine function; an initial sequence: POS (i，j) = [POS (r)，2n , POS (r)，2n+1 , POS (c),2n , POS (c),2n+1 ]. 4.The wafer image quality evaluation method based on position coding according to claim 1, wherein, The method comprises: splicing the global feature and a plurality of the sliding window local features based on a splicing formula to determine the global stitching feature of the target wafer image, the splicing formula being: wherein Z is a global stitching feature, E h,q is a sliding window local feature, h is the row encoding of each sliding window target image block, and q is the column encoding of each sliding window target image block. 5.The wafer image quality evaluation method based on position coding according to claim 1, wherein, The method comprises: determining a target step; performing sliding window sampling on the target wafer image based on the target step to obtain a plurality of sliding window target image blocks; The method comprises: N = N r x N c ; where N is the number of sliding window target image blocks, N r is the number of rows that the target wafer image is divided into, Nc is the number of columns that the target wafer image is divided into, H and W are the height and width of the target wafer image, respectively, P H and P W are the height and width of the window when sliding through sampling, respectively. 6.The wafer image quality evaluation method based on position coding according to claim 1, wherein, performing dimension conversion on each of the sliding window target image blocks to obtain a plurality of one-dimensional second converted image blocks; linearly projecting each of the one-dimensional second converted image blocks to a target dimension to obtain a plurality of the sliding window embedding vectors. The method comprises: 7.The wafer image quality evaluation method based on position coding according to claim 1, wherein, acquiring an original wafer image; performing data processing on the original wafer image to obtain the target wafer image. The device comprises:

8. A wafer image quality evaluation apparatus based on position encoding, characterized by, ​ An object wafer image acquisition module is configured to acquire an object wafer image. An object image block determination module is configured to determine a plurality of split object image blocks and a plurality of sliding window object image blocks based on the object wafer image. A split embedding vector determination module is configured to determine a plurality of split embedding vectors based on the plurality of split object image blocks. A sliding window embedding vector determination module is configured to determine a plurality of sliding window embedding vectors based on the plurality of sliding window object image blocks. An initial sequence formation module is configured to add position encoding to each of the split embedding vectors to form an initial sequence. A global feature determination module is configured to add a data capture marker at a starting position of the initial sequence to capture global information of the object wafer image according to the initial sequence, thereby obtaining a global feature of the object wafer image, wherein the global feature includes image split features of each of the split object image blocks and position information of each of the split object image blocks in the object wafer image. A sliding window local feature determination module is configured to determine a plurality of sliding window local features based on each of the sliding window embedding vectors, wherein each of the sliding window object image blocks corresponds to one sliding window local feature. A global stitching feature determination module is configured to determine a global stitching feature of the object wafer image based on the global feature and the plurality of sliding window local features. An index prediction module is configured to input the global stitching feature into a quality evaluation model, and output a quality prediction index of the object wafer image based on analysis of the global stitching feature by the quality evaluation model. A quality parameter index acquisition module is configured to acquire a quality parameter index. A quality evaluation result determination module is configured to determine a quality evaluation result of the object wafer image based on the quality prediction index and the quality parameter index.

9. An electronic device, comprising: The electronic device includes a memory and at least one processor, the memory having instructions stored therein; the at least one processor invokes the instructions in the memory to cause the electronic device to perform the steps of the wafer image quality evaluation method based on position encoding according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: The instructions are executed by the processor to implement the steps of the wafer image quality evaluation method based on position encoding according to any one of claims 1-7.

Images