Defect classification method and device and storage medium
By extracting and calculating the background similarity of semiconductor chip defect images, the problem of time-consuming and insufficient recognition capabilities in the prior art defect classification is solved, and fast, efficient and accurate defect classification is achieved, which is suitable for defect traceability in semiconductor chip manufacturing analyze.
Patent Information
- Application Number
- CN202310778570.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-28
- Publication Date
- 2024-12-31
AI Technical Summary
The existing defect classification methods have problems such as manual judgment time-consuming, poor robustness of automatic classification algorithms and insufficient recognition capabilities for different process conditions in semiconductor chip manufacturing, which is difficult to meet the needs of fast, efficient and accurate defect classification.
By extracting the background part of the defective image to be classified, the similarity between any two defective images is calculated, and the same category is divided according to the preset threshold. The similarity is calculated using histogram algorithm, cosine similarity and other algorithms, and the brightness and contrast are adjusted. Factors are calculated to affect similarity, and target defect categories are set for classification.
It realizes fast, efficient and accurate defect classification, can effectively trace the source analysis and filter defects, is suitable for classification of small or large number of defects, does not rely on a large amount of training data, and improves the accuracy and efficiency of classification.
Smart Images
Figure CN119229155A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of semiconductor technology, and in particular to a defect classification method, device and storage medium. Background Art
[0002] Semiconductor chip manufacturing requires multiple processes to achieve the required structure. No matter which process step, there is a risk of defects, which will affect the product yield. These defects come from machine contamination, imperfect process, unreasonable design structure, unqualified process materials, etc. Therefore, multiple defect detections will be interspersed in these processes to monitor defects on the chip. In order to improve the product yield, the source of the defect will be traced. The method is to obtain defect images through defect detection or re-inspection equipment, and then separate the defect images, and then classify them according to the defect images, analyze the various causes of the defects, and thus guide the optimization and improvement of the process.
[0003] Taking common defect classification as an example, it mainly includes manual classification of defect images through manual judgment and automatic classification based on identification of defect features. Among them, manual classification of defect images through manual judgment requires engineers to identify and judge the defect images one by one, which consumes a lot of manpower and takes a long time to classify. The traditional defect detection classification algorithm in automatic classification has weak robustness, and has poor recognition ability for defects caused by different process conditions, and cannot meet the current defect recognition requirements well. The classification algorithm based on machine learning in automatic classification can solve pattern classification and target recognition problems in different fields after better training, but the above defect classification requires a large amount of data model training for a long time, which is difficult to meet the large demand for deep convolutional neural network training data in actual production. Summary of the invention
[0004] The object of the present invention is to provide a defect classification method, device and storage medium to classify defects quickly, efficiently and accurately.
[0005] In order to solve the above technical problems, the present invention provides a defect classification method, comprising:
[0006] Providing a plurality of defect images to be classified, wherein the defect images include defect parts and background parts;
[0007] Acquire a background portion of the defect image to be classified;
[0008] The similarity between the background parts of any two defect images in the defect images to be classified is compared, and the defect images with the similarity greater than or equal to a preset threshold are classified as the same type of defects.
[0009] Optionally, all defect images to be classified are acquired using the same type of defect detection equipment.
[0010] Optionally, the area of the background portion in the defect image accounts for greater than or equal to two thirds.
[0011] Optionally, one or any of a histogram algorithm, cosine similarity, content feature method, key point matching, peak signal-to-noise ratio, structural similarity algorithm, hash algorithm, ORB algorithm or scale-invariant feature transformation algorithm is used to calculate the similarity between the background parts of the two defect images.
[0012] Optionally, when comparing the similarity between the background parts of any two defect images in the defect images to be classified, the algorithm is adjusted to reduce the influence of the maximum brightness, minimum brightness and maximum contrast of the defect images on the similarity calculation.
[0013] Optionally, the background portion of the defective image includes a first background image and a second background image that are connected, and the defective portion is located between the first background image and the second background image.
[0014] Acquire the first background image and the second background image;
[0015] Compare the first background image with the background parts of other defect images to obtain a first similarity, and compare the second background image with the background parts of other defect images to obtain a second similarity, take the larger of the first similarity and the second similarity as the similarity between the defect image and the background parts of the other defect images, and use this to classify the types of defects.
[0016] Optionally, the background part of the defect image includes a first background image and a second background image that are connected, and the defect image is located in the first background image or the second background image, and the first background image or the second background image where the defect image is located is used as the background part of the defect image.
[0017] Optionally, it also includes setting several target defect categories, each of which includes at least one preset defect image, and the background part of the preset defect image corresponds to the target defect category. The background part of the defect image to be classified is compared with the background part of the preset defect image for similarity to perform defect classification.
[0018] According to another aspect of the present invention, a defect classification device is also provided, comprising:
[0019] A background image acquisition module is used to extract the background part of the defect image to be classified;
[0020] A similarity calculation module, used for calculating the similarity between the background parts of any two defect images in the defect images to be classified;
[0021] The defect category determination module is used to determine whether two defect images are of the same type of defect according to whether the similarity between any two defect images is greater than or equal to a preset threshold.
[0022] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by an executor, the defect classification method as described above is implemented.
[0023] In summary, the defect classification method provided by the present invention extracts the background image of the defect image to be classified, and then calculates the similarity between the background images of any two defects in the defect image to be classified, and classifies the defects with a similarity greater than or equal to a preset threshold into the same category, thereby efficiently and accurately classifying defects with the same or similar background images into one category, that is, classifying the defects with the same or similar film layer and the area on the corresponding film layer into one category, which is very helpful for tracing the source of defects and filtering. Moreover, the number of defects does not affect the defect classification method provided by the present invention, and the defect classification method provided by the present invention can be used to quickly, efficiently and accurately classify a small number of defects or a large number of defects. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Those skilled in the art should understand that the drawings are provided for a better understanding of the present invention, but do not constitute any limitation on the scope of the present invention.
[0025] Figure 1 is a flow chart of the defect classification method provided in Example 1;
[0026] Figure 2 Four defect images to be classified provided in Example 1;
[0027] Figure 3 Four defect images to be classified are provided in Example 2.
[0028] In the attached figure:
[0029] 11-first background image; 12-second background image; 13-boundary area. DETAILED DESCRIPTION
[0030] In order to make the purpose, advantages and features of the present invention clearer, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the drawings are all in a very simplified form and are not drawn to scale, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In addition, the structure shown in the drawings is often a part of the actual structure. In particular, the emphasis of each drawing is different, and sometimes different scales are used.
[0031] As used in the present invention, the singular forms "one", "an", and "the" include plural objects, the term "or" is usually used to include the meaning of "and / or", the term "several" is usually used to include the meaning of "at least one", and the term "at least two" is usually used to include the meaning of "two or more". In addition, the terms "first", "second", and "third" are used only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined as "first", "second", and "third" may explicitly or implicitly include one or at least two of the features, unless the content clearly indicates otherwise.
[0032] Embodiment 1
[0033] Embodiment 1 provides a defect classification method.
[0034] Figure 1 This is a flow chart of the defect classification method provided in Example 1.
[0035] like Figure 1 As shown, the defect classification method provided in this embodiment includes:
[0036] S01: providing a plurality of defect images to be classified, wherein the defect images include defect parts and background parts;
[0037] S02: Acquire the background part of the defect image to be classified;
[0038] S03: comparing the similarity between the background parts of any two defect images to be classified, and classifying the defect images with the similarity greater than or equal to a preset threshold as the same type of defects.
[0039] First, step S01 is performed to provide a number of defect images to be classified, where the defect images include defect parts and background parts.
[0040] The defect image to be classified in this embodiment may be a defect image (original defect image) identified by a defect detection device, which includes a background portion located at the defect portion and surrounding (or partially surrounding) the defect portion. Among them, the defect detection device may be any suitable detection device, such as an automatic optical detection device, an electron beam detection device, a scanning electron microscope, a transmission electron microscope, etc. Preferably, the above-mentioned defect images to be classified are all obtained through the same type of defect detection equipment and under relatively close detection conditions, so as to minimize or avoid the impact of detection differences on defect classification. Of course, if the above-mentioned defect images to be classified include defect images obtained through multiple defect detection devices, the defect images can be divided into different groups (for example, n) according to the type of defect detection equipment (for example, n types, n is a positive integer), and defects are classified within the groups.
[0041] It is worth mentioning that the present embodiment utilizes the background part in the defect image to classify the defects, that is, the defects are classified according to the background information of the defects such as the film layer to which the defects belong and the area where the defects are located on the corresponding film layer. Therefore, the area ratio of the background part (background image) in the above-mentioned defect image to be classified is preferably greater than or equal to two-thirds, so as to provide sufficient background information. Among them, the area of the background image can be the area of the remaining part after the defect is selected by the minimum rectangular frame. In one example, when the defect image is obtained by using a defect detection device, the area ratio of the background image can be increased by adjusting the detection conditions (parameters), or the defect can be moved to the edge of the field of view and only partially located in the field of view to increase the area ratio of the background image. In another example, the defect image can also be processed to increase the area ratio of the background image.
[0042] Next, step S02 is executed to obtain the background portion of the defect image to be classified.
[0043] Specifically, based on the defect recognition parameters of the defect detection equipment, such as morphology, position, center of mass, etc., the smallest rectangular frame or arbitrary curve frame can be used to select the defect along the outline of the defect in the defect image as the defect part, shield or remove the defect part, and extract the remaining part of the defect image as the background image (background part).
[0044] In one example, if Figure 2 As shown in the defect image (c) in FIG. 1 , the defect is, for example, a foreign object and is located in a blank area. The extracted background portion of the defect may include the blank area and other graphic structures on the blank area.
[0045] In another example, if Figure 3 As shown in the defect image (d) in FIG. 1 , the defect is, for example, a graphic defect, and the background image of the extracted defect may be a normal graphic structure around the graphic defect.
[0046] Next, step S03 is executed to compare the similarity between the background parts of any two defect images in the defect images to be classified, and classify the defects with similarity greater than or equal to a preset threshold into the same type of defects.
[0047] The specific steps include: using an image similarity comparison algorithm to calculate the similarity between the background images (background parts) of any two defective images to be classified, taking the completely similar similarity as 1 as an example, the calculated similarity can be 0 to 1. Among them, the image similarity comparison algorithm may include one or any of the following: histogram algorithm, cosine similarity, content feature method, key point matching, peak signal-to-noise ratio, structural similarity algorithm, hash algorithm, ORB (Oriented FAST and Rotated BRIEF) algorithm or scale-invariant feature transformation algorithm. Of course, the standard algorithm of the above algorithm can be used for similarity comparison (calculation), or partial improvements can be made on the basis of the standard algorithm to optimize the actual effect.
[0048] Preferably, when comparing the background parts of two defective images to obtain similarity, the algorithm can also be adjusted to reduce the influence of factors such as the maximum brightness, minimum brightness and maximum contrast of the background image (external environment, detection conditions, etc.) on the calculation of similarity. In other words, the influence of the image structure itself in the background part on the similarity calculation is increased. In one example, before calculating the similarity, the defective image can be preprocessed to reduce the brightness above the preset maximum value to the preset maximum value and increase the brightness below the preset minimum value to the preset minimum value, thereby reducing the influence of external factors such as detection conditions on the similarity calculation.
[0049] Next, based on the similarity between any two defect images obtained above, the defects whose background image similarity is greater than or equal to a preset threshold are classified as the same defect. If the similarity between the background part of a defect image to be classified and the background parts of two or more defect images is greater than the preset threshold, and at the same time, the other two or more defect images (classified) do not belong to the same category of defects, the defect to be classified can be classified into the category of defects with the greatest similarity. In addition, if the similarity between the background part of a defect image and the background part of any other defect image is less than the preset threshold, the defect image can be listed as a separate category of defects.
[0050] In a specific example, if Figure 2 The four defect images (a) to (d) shown are all defect images detected by semiconductor processes. The preset threshold can be 0.6 (complete similarity is 1, and complete dissimilarity is 0). The background image after removing the defect image has a high similarity (all greater than 0.6), and the above four defect images can be classified as the same type of defects.
[0051] Embodiment 2
[0052] Embodiment 2 provides a defect classification method.
[0053] The defect classification method provided in the second embodiment is basically similar to the defect classification method provided in the first embodiment in terms of principle and steps, and the main difference is that in the second embodiment, the background portion of at least one defect image has two background images. Next, the defect classification method of the second embodiment is described by taking the defect image including the first background image and the second background image connected as an example, and the defect portion can be located on the first background image and / or the second background image.
[0054] In one example, the defect image is located on the first background image or the second background image, and the first background image or the second background image where the defect image is located can be used as the background image (background part) of the defect image instead of using the entire background part to participate in the similarity calculation.
[0055] In another embodiment, the defect image is located on the first background image and the second background image, that is, between the first background image and the second background image. To this end, the steps of extracting background images and comparing similarities may include: obtaining the first background image and the second background image of the defect image; obtaining the similarity between the first background image and the background part of the other defect images and taking the maximum value thereof as the first similarity, and obtaining the similarity between the second background image and the background part of the other defect images and taking the maximum value thereof as the second similarity, taking the larger of the first similarity and the second similarity as the similarity between the defect image and the other defect images, and classifying the types of defects accordingly.
[0056] Of course, if there are more than two background images in the defective image, that is, the defective part of the defective image is located between multiple background images (surrounded by multiple background images), it is also feasible to use the methods in the above two examples. The area where two or more background images meet is the boundary area, and the image features of the boundary area are more obvious and special. In practice, it is possible to detect whether the defective image has multiple background images by identifying whether there is a boundary area in the defective image, and it is also possible to determine the background image where the defective part is located by identifying the relative position of the defective image and the boundary area.
[0057] In a specific example, if Figure 3The four defect images (a) to (d) shown are all defects detected by the semiconductor process. Defect image (a) has a relatively standard background image (e.g., the first background image 11), defect image (b) has a background image (e.g., the first background image 11) and a boundary area 13, defect image (c) and defect image (d) have two background graphics (e.g., the first background image 11 and the second background image 12) and their defective parts are both adjacent to the boundary area 13. Therefore, by using the similarity comparison of the above-mentioned background images and setting a suitable preset threshold, defect images (a) to (d) can be classified into one type of defects (mainly located on the first background image 11).
[0058] In addition, based on the background image similarity comparison provided in this embodiment, additional feature comparison of the defect part or the background part is added, and the defect category of the defect image to be classified can be subdivided, for example, into several subcategories. Figure 3 Taking the four defect images (a) to (d) as an example, if the additional features include the distance between the defect part and the boundary area (or whether it is connected to the boundary area), the defect images (a) and (b) can be classified as the first sub-category defects, and the defect images (c) and (d) can be classified as the second sub-category defects. Of course, when subdividing the defects into sub-categories, it is also feasible to simply increase the preset threshold.
[0059] Embodiment 3
[0060] Embodiment 3 provides a defect classification method.
[0061] The defect classification method provided in Example 3 is basically similar to the defect classification method provided in Example 1 or Example 2 in principle and steps. The main difference is that in Example 3, several target defect categories can be pre-set, each target defect category includes at least one preset defect image, and the background image of each preset defect image corresponds to the corresponding target defect category, that is, each defect category corresponds to a background image. When performing defect classification, firstly use the defect image to be classified and the preset defect image to perform a similarity comparison of the background part (obtain several similarities), and divide the defect image to be classified into the target defect category with the highest similarity exceeding the preset threshold value for defect classification. Of course, if there are still some defect images to be classified that do not belong to the above-mentioned target defect categories, the remaining defect images to be classified can be automatically classified with reference to Example 1 or Example 2.
[0062] Embodiment 4
[0063] Embodiment 4 provides a defect classification device.
[0064] The defect classification device provided in this embodiment includes a background image acquisition module, a similarity calculation module and a defect category determination module. The background image acquisition module is used to extract the image of the background part of the defect image to be classified; the similarity calculation module is used to calculate the similarity between the background parts of any two defect images in the defect image to be classified; the defect category determination module is used to determine whether the defect images are of the same type of defect according to whether the similarity of any two defect images is greater than or equal to a preset threshold.
[0065] Embodiment 5
[0066] Embodiment 5 provides a computer-readable storage medium.
[0067] The computer-readable storage medium provided in this embodiment stores a computer program, and when the computer program is executed by a processor, the steps of the defect classification method described in any of the above embodiments are implemented. The computer-readable storage medium involved in this embodiment includes a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the technical field.
[0068] In summary, the defect classification method provided by the present invention extracts the background part of the defect image to be classified, and then calculates the similarity between the background parts of any two defect images in the defect image to be classified, and classifies the defects with similarity greater than or equal to the preset threshold into the same category of defects, thereby efficiently and accurately classifying defects with the same or similar background images into one category, that is, classifying the defects with the same or similar film layer and the area on the corresponding film layer into one category, which is very conducive to tracing the source of defects and filtering, and has the characteristics of being fast, efficient and accurate. Moreover, the defect classification method provided by the invention does not rely on a large amount of training data to improve its accuracy. The defect classification method established by the present invention using a small number of defect images can also have a high accuracy.
[0069] The above description is only a description of the preferred embodiments of the present invention, and is not intended to limit the scope of the present invention. Any changes or modifications made by a person skilled in the art in the field of the present invention based on the above disclosure shall fall within the scope of protection of the claims.
Claims
1. A defect classification method, characterized in that: include: Providing a plurality of defect images to be classified, wherein the defect images include defect parts and background parts; Acquire a background portion of the defect image to be classified; The similarity between the background parts of any two defect images in the defect images to be classified is compared, and the defect images with the similarity greater than or equal to a preset threshold are classified as the same type of defects.
2. The defect classification method according to claim 1, characterized in that: All defect images to be classified are acquired using the same type of defect detection equipment.
3. The defect classification method according to claim 2, characterized in that: The area of the background part in the defect image accounts for more than or equal to two thirds.
4. The defect classification method according to claim 1, characterized in that: The similarity between the background parts of the two defect images is calculated by using one or any of the following: histogram algorithm, cosine similarity, content feature method, key point matching, peak signal-to-noise ratio, structural similarity algorithm, hash algorithm, ORB algorithm or scale-invariant feature transformation algorithm.
5. The defect classification method according to claim 4, characterized in that: When comparing the similarity between the background parts of any two defect images in the defect images to be classified, the algorithm is adjusted to reduce the influence of the maximum brightness, minimum brightness and maximum contrast of the defect images on the similarity calculation.
6. The defect classification method according to claim 1, characterized in that: The background portion of the defective image includes a first background image and a second background image that are connected, and the defective portion is located between the first background image and the second background image. Acquire the first background image and the second background image; Compare the first background image with the background parts of other defect images to obtain a first similarity, and compare the second background image with the background parts of other defect images to obtain a second similarity, take the larger of the first similarity and the second similarity as the similarity between the defect image and the background parts of the other defect images, and use this to classify the types of defects.
7. The defect classification method according to claim 1, characterized in that: The background part of the defect image includes a first background image and a second background image connected to each other. The defect image is located in the first background image or the second background image. The first background image or the second background image where the defect image is located is used as the background part of the defect image.
8. The defect classification method according to claim 1, characterized in that: It also includes setting up several target defect categories, each of which includes at least one preset defect image, and the background part of the preset defect image corresponds to the target defect category. The background part of the defect image to be classified is compared with the background part of the preset defect image for similarity to perform defect classification.
9. A defect classification device, characterized in that: include: A background image acquisition module is used to extract the background part of the defect image to be classified; A similarity calculation module, used for calculating the similarity between the background parts of any two defect images in the defect images to be classified; The defect category determination module is used to determine whether two defect images are of the same category of defects according to whether the similarity between any two defect images is greater than or equal to a preset threshold.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by an executor, the defect classification method as described in any one of claims 1 to 8 is implemented.