Visual detection method and device suitable for wafer surface image, equipment and storage medium

Abstract

The invention relates to a visual detection method and device suitable for a wafer surface image, equipment and a storage medium, and is applied to the field of wafer defect detection, and the method comprises the steps: dividing a to-be-focused wafer into a plurality of virtual unit image grids according to the sheet of a preset collected unit image, and determining the focusing focal length of the unit image grids; obtaining unit color images of the plurality of unit image grids under the focusing focal length; sequentially processing the plurality of unit color images by adopting a preset positive sample detection model and a preset negative sample detection model, and generating a positive sample defect image and a negative sample defect image; and fusing the positive sample defect image and the negative sample defect image, and generating a wafer unit defect image. The wafer defect detection method has the technical effects that the wafer defect detection rate and the detection efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wafer defect detection, in particular to a visual detection method and device suitable for wafer surface images, equipment and storage medium. BACKGROUND

[0002] Wafer defect detection can detect defects, flaws and the like existing in wafer surface images, and is a key link to realize high yield and high reliability, directly affecting the quality of chips.

[0003] At present, the detection of wafer defects usually adopts a negative sample model to detect the wafer, and the negative sample model needs to be trained with a large number of defects in negative sample wafers to detect wafer defects in actual application. However, as the wafer manufacturing process matures, the production yield of wafers is very high, which means that the number of defect samples in actual production is small, and the number of positive samples, that is, normal wafers, is large. Such a situation will result in a small number of samples used for training the negative sample model, thereby making it difficult to detect some rare wafer defects and reducing the detection rate of wafer defects. SUMMARY

[0004] In order to help solve the problem that the number of samples used for training the negative sample model is small, it is difficult to detect some rare wafer defects, and the detection rate of wafer defects is reduced, the present application provides a visual detection method and device suitable for wafer surface images, equipment and storage medium.

[0005] In the first aspect, the present application provides a visual detection method suitable for wafer surface images, which adopts the following technical scheme: the method comprises: According to the map of the preset collected unit image, the wafer to be focused is divided into a plurality of virtual unit image grids, and the focusing focal length of the unit image grid is determined; Obtain a plurality of unit color images of the unit image grid under the focusing focal length; A preset positive sample detection model and a negative sample detection model are used to process a plurality of unit color images in sequence and generate positive sample defect images and negative sample defect images; The positive sample defect images and the negative sample defect images are fused to generate wafer unit defect images.

[0006] In one specific implementation, the determination of the focusing focal length of the unit image grid comprises: The clear focal length of each sampling point in a preset sampling point set on the wafer to be focused is measured in sequence; According to the coordinates of each sampling point and the corresponding measured clear focal length, a predicted wafer warping model is generated; The focal length corresponding to the unit image to be acquired is determined based on the wafer warping model.

[0007] In one specific implementation scheme, determining the focal length corresponding to the unit image to be acquired based on the wafer warpage model includes: The reference focal lengths corresponding to multiple preset reference points on the wafer to be focused are calculated based on the wafer warping model. Match the nearest reference point corresponding to the center point of the unit image to be acquired, and determine the initial focus range of the center point of the unit image based on the reference focal length corresponding to the nearest reference point; Based on the focal length of the center point of the previous unit image in the preset travel direction, and the focal length corresponding to the center point of the unit image to be acquired is adjusted according to the preliminary focusing range, the adjusted focal length is set as the focal length corresponding to the unit image to be acquired.

[0008] In one specific implementation, fusing the positive sample defect image and the negative sample defect image to generate a wafer cell defect image includes: Determine whether there are defect regions in the positive sample defect image and the negative sample defect image; If the positive sample defect image has a defect region and the negative sample defect image does not have a defect region, then the positive sample defect image is retained and set as a wafer cell defect image; If the positive sample defect image does not have a defect region, but the negative sample defect image does have a defect region, then the negative sample defect image is retained and set as a wafer cell defect image; If both the positive sample defect image and the negative sample defect image contain defect regions, then the defect regions in the positive sample defect image and the defect regions in the negative sample defect image are fused together to generate a wafer cell defect image.

[0009] In one specific implementation, fusing the defect regions in the positive sample defect image and the defect regions in the negative sample defect image to generate a wafer cell defect image includes: Sequentially determine whether there is an overlap between the defect region in the positive sample defect image and the defect region in the negative sample defect image; If the defect regions in the positive sample defect image and the defect regions in the negative sample defect image do not overlap, then the defect regions in the positive sample defect image and the defect regions in the negative sample defect image are retained respectively, and the retained defect regions are set as independent defect regions. If the defect region in the positive sample defect image and the defect region in the negative sample defect image overlap, then the defect region that needs to be retained is determined based on the overlapping region. The independent defect regions and the defect regions determined based on the overlapping regions are fused together to generate a wafer cell defect image.

[0010] In one specific implementation scheme, determining the defective region to be retained based on the overlapping region includes: Calculate the overlap rate of the overlapping regions and determine whether the overlap rate is greater than a preset overlap rate threshold. If the overlap rate of the region is greater than the preset overlap rate threshold, then the defect region in the negative sample defect image is retained, and the defect region in the negative sample defect image is set as a completely overlapping defect region. If the overlap rate of the regions is not greater than the preset overlap rate threshold, then the union region of the defect region in the positive sample defect image and the defect region in the negative sample defect image is retained, and the union region is set as a partially overlapping defect region. The step of fusing the independent defect regions and the defect regions determined based on the overlapping regions to generate a wafer cell defect image includes: The independent defect region, the completely overlapping defect region, and the partially overlapping defect region are merged to generate a wafer cell defect image.

[0011] In one specific implementation, after fusing the positive sample defect image and the negative sample defect image to generate a wafer cell defect image, the method further includes: The source of the detection results for all defect regions in the wafer unit defect image is determined sequentially; If the current defect area originates from the detection result of the positive sample detection model, then a manual re-inspection is notified to determine the defect type of the current defect area; If the current defect region originates from the detection result of the negative sample detection model, then the defect type of the current defect region is output according to the detection result of the negative sample detection model. If the current defect region originates from the detection results of the positive sample detection model and the negative sample detection model, then the defect type of the current defect region is output based on the detection result of the negative sample detection model.

[0012] Secondly, this application provides a visual inspection device suitable for wafer surface images, employing the following technical solution: the device includes: The focus focal length determination module is used to divide the wafer to be focused into several virtual unit image grids according to the image size of the preset acquired unit image, and determine the focus focal length of the unit image grids. A color image acquisition module is used to acquire a unit color image of a plurality of the unit image grids at the focal length; The defect model detection module is used to process several unit color images sequentially using a preset positive sample detection model and a negative sample detection model to generate positive sample defect images and negative sample defect images; The defect image generation module is used to fuse the positive sample defect image and the negative sample defect image to generate a wafer unit defect image.

[0013] Thirdly, this application provides a computer device that adopts the following technical solution: it includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as any of the above-mentioned visual inspection methods suitable for wafer surface images.

[0014] Fourthly, this application provides a computer-readable storage medium that stores a computer program capable of being loaded by a processor and executing any of the aforementioned visual inspection methods suitable for wafer surface images.

[0015] In summary, this application has the following beneficial technical effects: By simultaneously using both positive and negative sample detection models and fusing their results to generate a final wafer defect image, the detection rate of wafer surface defects can be improved. In actual inspection, the positive sample detection model is effective at detecting both rare and significant defects on the wafer surface, ensuring that no significant or rare defects are missed. The negative sample detection model can detect common defects and output their corresponding defect types. Furthermore, the negative sample model can effectively detect critical color differences and diffusion features, ensuring the accuracy of wafer defect detection and further improving wafer yield.

[0016] Furthermore, by dividing the wafer plane into several unit image grids and focusing each unit image sequentially, the wafer plane can be refined, making the focal length of each unit image as clear as possible, thereby improving the overall clarity of the wafer image. At the same time, dividing the wafer plane into several grid images can reduce the problems of poor focal length and unclear images caused by wafer warpage and unevenness of the wafer plane, thus improving the clarity of the wafer image. Attached Figure Description

[0017] Figure 1This is a flowchart of a visual inspection method suitable for wafer surface images in the embodiments of this application; Figure 2 This is a schematic diagram used to illustrate the independent defect region in the embodiments of this application; Figure 3 This is a schematic diagram illustrating the completely overlapping defect region in the embodiments of this application; Figure 4 This is a schematic diagram illustrating partially overlapping defect areas in an embodiment of this application; Figure 5 This is a schematic diagram of a visual inspection device suitable for wafer surface images in an embodiment of this application; Figure 6 This is a schematic diagram used to illustrate a computer device in the embodiments of this application.

[0018] Reference numerals: 501, Focusing and focal length determination module; 502, Color image acquisition module; 503, Defect model detection module; 504, Defect image generation module. Detailed Implementation

[0019] The following combination Figures 1-6 This application will be described in further detail.

[0020] This application discloses a visual inspection method suitable for wafer surface images. This method enables defect detection on wafers by simultaneously inspecting the wafer image using both positive and negative sample models, and determining the presence and type of defects based on a combination of the results from both models.

[0021] Wafer defect detection, which identifies flaws and defects in wafer surface images, is a crucial step in achieving high yield and reliability, directly impacting chip quality. Currently, wafer defect detection typically employs negative sample models, which require training with a large number of defective wafers in negative sample mode to detect defects in practical applications. However, with increasingly mature wafer manufacturing processes and very high production yields, defective samples in actual production are relatively few, while positive samples, i.e., normal wafers, constitute the majority. This results in a limited number of samples for training negative sample models, making it difficult to detect some rare wafer defects, thus reducing the wafer defect detection rate. Furthermore, the industry currently still commonly acquires wafer images using black and white images. Black and white images lack color information, which may lead to false positives or false negatives of wafer defects, further reducing the accuracy and detection rate. To help improve the accuracy and detection rate of wafer defects, this application provides a visual inspection method suitable for wafer surface images.

[0022] Reference Figure 1 The method includes the following steps: S10: Based on the preset image size of the acquired unit image, divide the wafer to be focused into several virtual unit image grids and determine the focusing focal length of the unit image grid.

[0023] Specifically, the preset image size of the acquired unit image can be understood as a reference image size for dividing the wafer into several unit image grids. The unit image grid can be a 300*300 grid to cover the wafer, or a 500*500 grid. The number of grids and the image size are user-defined and not limited here. By dividing the wafer into several virtual unit image grids, the focal length of each image unit grid is determined. This ensures that the captured unit image has a sharp focal length, resulting in high-resolution wafer images acquired at the correct focal length after focusing. This facilitates the detection and identification of wafer defects based on the wafer images, enabling visual inspection of the wafer.

[0024] S20: Acquire a unit color image of a number of unit image grids at the focusing focal length.

[0025] Specifically, after determining the focal length of each unit image grid, an image acquisition device is used to acquire a color image of the wafer at that focal length. Since the focusing process is performed sequentially according to the unit image grid, the actual acquisition of the wafer color image is also performed sequentially according to the unit image grid, ultimately resulting in unit color images of several wafers.

[0026] S30 uses a preset positive sample detection model and a negative sample detection model to process several unit color images in sequence and generate positive sample defect images and negative sample defect images.

[0027] Specifically, after obtaining several unit color images of the wafer, the unit color images are processed sequentially using a preset positive sample detection model and a negative sample detection model. The same unit color image will obtain the result output by the positive sample detection model, which is a positive sample defect image, and the result output by the negative sample detection model, which is a negative sample defect image.

[0028] It should be noted that in this embodiment, the positive sample detection model and the negative sample detection model run simultaneously. The same unit color image is processed by both models simultaneously, outputting both a positive sample defect image and a negative sample defect image. This can also be understood as two defect detection results obtained from the same unit color image using two different detection models. The negative sample detection model is trained on common wafer defects to accurately detect common defects in wafer surface images and determine the defect type, such as scratches or dirt. However, for rare defects, due to the limited number of defect samples, the negative sample model struggles to accurately detect them. The positive sample detection model, on the other hand, is trained on defect-free wafers. By comparing the image of the wafer to be detected with an image of a defect-free wafer, it can determine whether the wafer to be tested has defects. The positive sample detection model is trained on defect-free wafers, eliminating the need for a large number of negative samples (i.e., defective samples), thus compensating for the disadvantage of the negative sample detection model in training and recognizing rare defect types. By combining positive sample detection models and negative sample detection models, wafer defects can be detected, thereby improving the detection rate and accuracy of wafer defect detection.

[0029] S40 fuses the positive sample defect image and the negative sample defect image to generate a wafer cell defect image.

[0030] Specifically, after each unit color image is detected by two detection models in sequence, the corresponding positive sample defect image and negative sample defect image are output. Considering that the positive sample detection model and the negative sample detection model work on different principles, the defect results detected by the two models may be different. Therefore, in order to detect as many defects as possible, the two defect images are fused to generate the final wafer unit defect image, so that the staff can use the final wafer unit defect image for re-inspection and re-judgment.

[0031] In this application, by simultaneously using a positive sample detection model and a negative sample detection model, and fusing the detection results of the two models to generate the final wafer defect image, the detection rate of wafer surface defects can be improved. During actual inspection, the positive sample detection model is effective in detecting both rare and significant defects on the wafer surface, ensuring that no significant or rare defects are missed. The negative sample detection model can detect common defects and output the corresponding defect type. Simultaneously, the negative sample model can effectively detect critical color differences and diffusion features, ensuring the accuracy of wafer defect detection and further improving wafer yield.

[0032] Furthermore, by dividing the wafer plane into several unit image grids and focusing each unit image sequentially, the wafer plane can be refined, making the focal length of each unit image as clear as possible, thereby improving the overall clarity of the wafer image. At the same time, dividing the wafer plane into several grid images can reduce the problems of poor focal length and unclear images caused by wafer warpage and unevenness of the wafer plane, thus improving the clarity of the wafer image.

[0033] In one embodiment, the method for determining the focal length of the unit image grid can be specifically performed as follows: First, the sharp focal length of each sampling point within a preset set of sampling points on the wafer to be focused is measured sequentially. Specifically, the sharp focal length of the preset sampling points on the wafer to be focused is measured so that a wafer warpage model can be established based on the measured sharp focal length. The sharp focal length measurement can be performed as follows: First, initial images of several sampling points at an initial focal length are acquired sequentially, and the initial sharpness value of the initial image is calculated using a preset sharpness algorithm. Wafer images can be acquired through an image acquisition module, which can be a camera, liquid lens, etc., and the user can choose according to actual needs; no restrictions are placed here. When acquiring wafer images, an initial focal length of the camera is first set, and the initial image of the wafer is acquired at the initial focal length. After acquiring the initial image, the sharpness value of the initial image is calculated using a sharpness algorithm, which can directly use existing algorithms for calculation; after calculating the sharpness, it is determined whether the initial sharpness value of the initial image is greater than a preset sharpness threshold. If the initial sharpness value of the initial image is greater than the preset sharpness threshold, the focal length corresponding to the initial image is set as the sharpness focal length of the current sampling point; or, if the sharpness value of the calculated wafer image reaches the user's preset sharpness threshold, the focal length of the currently captured wafer image is set as the sharpness focal length of the current sampling point.

[0034] If the initial sharpness value of the initial image is not greater than the preset sharpness threshold, the initial focal length is adjusted successively according to the preset focusing rules to obtain the optimized focal length. Alternatively, if the image sharpness of the captured wafer image does not reach the preset sharpness threshold, the optimized focal length is obtained according to the preset adjustment rules. The focal length can be adjusted in preset increments, such as 0.1µm each time; or it can start with a larger increment, and if the preset sharpness requirement is not met, the increment is reduced. For example, if the first focal length adjustment involves a 0.6µm increment, and the sharpness requirement is not met, then the increment is reduced to 0.3µm.

[0035] When an optimized focal length is obtained or updated, an optimized image of the current sampling point at the optimized focal length is acquired. Once the sharpness value of the optimized image at the current sampling point at the optimized focal length exceeds a preset sharpness threshold, the optimized focal length is set as the sharpness focal length of the current sampling point. Alternatively, if the sharpness value of the initial image does not reach the preset threshold, the focal length is cyclically adjusted to obtain an optimized image. Once the sharpness value of the optimized image reaches the preset sharpness threshold, the loop stops. Finally, when the sharpness value of the optimized image reaches the preset threshold, the corresponding focal length is set as the sharpness focal length of the current sampling point.

[0036] After the sharp focal length is measured, a predicted wafer warpage model is generated based on the coordinates of each sampling point and the corresponding measured sharp focal length. Specifically, the wafer warpage model is a three-dimensional coordinate model, which includes three-dimensionally distributed X-axis, Y-axis, and Z-axis; the plane containing the X-axis and Y-axis in the wafer warpage model is the plane containing the wafer to be focused; the Z-axis in the wafer warpage model is perpendicular to the plane containing the wafer to be focused.

[0037] Each preset sampling point has a planar coordinate (x, y) on the wafer plane. The above steps obtain the focal length z of each sampling point by taking a clear image of the preset sampling point on the wafer. The three-dimensional coordinates (x, y, z) of each preset sampling point can be obtained by using the planar coordinates and focal length. A wafer warping model can be constructed by using the three-dimensional coordinates of several preset sampling points. The focus length of each point on the wafer can be predicted by using the wafer warping model.

[0038] Finally, the focal length corresponding to the unit image to be acquired is determined based on the wafer warp model. Specifically, based on the constructed wafer warp model, the focal length of each unit image on the wafer plane is determined; specifically, it can be the focal length of the center point of each unit image on the wafer plane. Dividing the wafer plane into several unit images ensures that the focal length of each small grid unit image is sharp. Furthermore, dividing the wafer plane reduces the influence of wafer warp and unevenness, preventing warp and unevenness from affecting wafer focusing and thus avoiding poor image sharpness.

[0039] In this application, a wafer warpage model is constructed by measuring the sharp focal length of preset sampling points on the wafer. This model can predict the focusing focal length of detection points on the wafer plane. Secondly, by dividing the wafer plane into several unit image grids, and performing focusing processing on each unit image sequentially, the wafer plane can be refined, ensuring that the focal length of each unit image is as sharp as possible, thereby improving the overall clarity of the wafer image. Simultaneously, dividing the wafer plane into several grid images reduces the problems of poor focus and unclear images caused by wafer warpage and potential unevenness of the wafer plane, thus improving the clarity of the wafer image. Furthermore, by directly adjusting the sharp focal length of the wafer during the focusing process, the complex image processing methods used after image acquisition are avoided, reducing computational load, saving computation time, and improving the efficiency of acquiring sharp wafer images, thereby improving the efficiency of wafer inspection.

[0040] In one embodiment, the method of determining the focal length corresponding to the cell image to be acquired based on the wafer warpage model can be specifically executed as follows: First, the reference focal lengths corresponding to multiple preset reference points on the wafer to be focused are calculated based on the wafer warpage model. After the wafer warpage model is completed, the user can preset several reference points. It should be noted that the reference points set after modeling are not the sampling points mentioned above; the purpose of the sampling points mentioned above is to obtain the sharp focal length of the sampling points to construct the three-dimensional coordinates of these sampling points, thereby constructing the wafer warpage model; while the reference points mentioned here are reset points, and their focusing focal lengths are not obtained by measurement, but by prediction through the constructed model, so as to determine the focusing focal length of the center point of each unit image. The selection of the reference point positions and the number of reference points are determined by the user. Generally speaking, the more reference points selected, the more accurate the focusing focal length, but the more computational load will be; conversely, the fewer the number, the lower the focusing accuracy, but the higher the processing speed and work efficiency. Users can set the reference points according to the actual application requirements.

[0041] Next, the nearest reference point corresponding to the center point of the unit image to be acquired is matched, and the preliminary focus range of the unit image center point is determined based on the reference focal length corresponding to the nearest reference point. Specifically, the determination of the preliminary focus range can be performed as follows: first, calculate the reference distance between the center point of the unit image to be acquired and all reference points; calculate the set of reference distances, and set the reference point with the shortest reference distance as the nearest reference point; finally, add a preset error range to the reference focal length corresponding to the nearest reference point to obtain the preliminary focus range corresponding to the unit image center point. For example, suppose the obtained nearest reference distance is... The preset error can also be understood as the preset focal length error tolerance. The initial focus range is ,in, , .

[0042] Finally, based on the focal length of the center point of the previous unit image along the preset travel direction and the initial focusing range, the focal length corresponding to the center point of the unit image to be acquired is adjusted, and the adjusted focal length is set as the focal length corresponding to the unit image to be acquired. Here, the preset travel direction is the direction of the moving path of the wafer focusing device. (Refer to...) Figure 2 The wafer focusing device moves in an S-shape to traverse the grid image, and the travel direction is also S-shaped. After dividing the wafer plane into several virtual unit image grids, each grid image can be focused sequentially in an S-shape. The S-shaped movement method can save time and improve efficiency. However, the traversal method is not unique. Users can set the direction of the wafer focusing device according to actual needs, and there is no restriction here.

[0043] After determining the initial focus range Then, based on the focal length of the center point of the previous unit image in the preset travel direction and the initial focusing range, the focal length corresponding to the center point of the unit image to be acquired is adjusted. The final focal length can be determined as follows:

[0044]

[0045]

[0046]

[0047]

[0048] in, This represents the predicted focal length of the center point of the current cell image. This indicates the focal length of the center point of the previous unit image. This indicates the reference focal length corresponding to the nearest reference point. This represents the adjustment coefficient. , This indicates the focal length of the final determined unit image. Indicates the lower limit of the initial focus range. This indicates the upper limit of the initial focus range. This indicates the preset error range, also known as the preset focal length error tolerance. This represents the distance between the center point of the current image unit and the center point of the previous image unit, and D represents the average spacing between adjacent reference points.

[0049] In this application, the initial focusing range of each unit image center point is determined first by using the nearest preset reference point. Then, the focusing focal length is adjusted based on the already determined focusing focal length of the previous unit image center point in the travel direction. This improves the accuracy of the focusing focal length, resulting in a clear image of the wafer. Simultaneously, it can predict defocusing trends along the scanning path, achieving micron-level position correction and reducing sensor physical deviations. Furthermore, it improves efficiency, avoiding extensive computation and image processing, thereby enabling high-precision defect identification and dimensional measurement.

[0050] In one embodiment, fusing positive sample defect images and negative sample defect images to generate wafer cell defect images can be specifically performed as follows: First, determine whether defect regions exist in the positive and negative sample defect images. If a defect region exists in the positive sample defect image but not in the negative sample defect image, retain the positive sample defect image and set it as the wafer cell defect image. Alternatively, if a defect region is detected only in the positive sample defect image and not in the negative sample defect image, then directly set the positive sample defect image as the wafer cell defect image.

[0051] If the positive sample defect image does not contain a defect region, but the negative sample defect image does, then the negative sample defect image is retained and set as the wafer cell defect image. Alternatively, if a defect region is detected only in the negative sample defect image, and not in the positive sample defect image, then the negative sample defect image is directly set as the wafer cell defect image.

[0052] If both the positive and negative sample defect images contain defect regions, the defect regions in the positive and negative sample defect images are fused together to generate a wafer cell defect image. In other words, if both the obtained positive and negative sample defect images contain defect regions, the defects in the corresponding regions of the positive and negative sample defect images are fused together, and the resulting image is the wafer cell defect image.

[0053] In one embodiment, the method of fusing the defect regions in the positive sample defect image and the defect regions in the negative sample defect image to generate a wafer cell defect image can be specifically performed as follows: First, determine whether there are overlapping regions between the defect regions in the positive sample defect image and the defect regions in the negative sample defect image. The positive and negative sample defect images are obtained by processing the same unit color image using the positive and negative sample detection models respectively. Each image may contain one or more defect regions, and these regions may or may not overlap. Then, determine whether there is any overlap between the two images corresponding to each defect region.

[0054] If the defect regions in the positive sample defect image and the defect regions in the negative sample defect image do not overlap, then the defect regions in both images are retained, and these retained defect regions are designated as independent defect regions. In other words, if the current defect region exists only in the positive sample defect image or only in the negative sample defect image, and the corresponding location in the other image does not detect this defect region, then the detected defect is directly retained. (Refer to...) Figure 2 In the positive and negative sample defect maps, there are two different locations, defect A and defect B. In the actual fusion process, defect region A and defect region B are retained respectively, and defect region A and defect region B are independent defect regions.

[0055] If there is an overlap between the defect region in the positive sample defect image and the defect region in the negative sample defect image, the defect region to be retained is determined based on the overlap region. Finally, the independent defect regions and the defect regions determined based on the overlap region are fused together to generate a wafer unit defect image.

[0056] Specifically, the method of determining the defective areas to be retained based on overlapping areas can be implemented as follows: First, calculate the overlap rate of the overlapping regions and determine if the overlap rate exceeds a preset threshold. If the overlap rate exceeds the preset threshold, retain the defect region in the negative sample defect image and set it as a completely overlapping defect region. (Refer to...) Figure 3 In the positive and negative sample defect maps, two defect regions C1 and C2 are detected at the same location. The actual overlap rate of the two defect regions is calculated, and the actual overlap rate is greater than a preset overlap rate threshold. Furthermore, considering that the negative sample detection model is trained based on the actual defects and their corresponding defect types, the output of the negative sample model can directly output the defect type. Therefore, the defect regions in the negative sample defect map are merged and retained, and the retained regions are set as completely overlapping defect regions. Figure 3 The defective region C2. The overlap rate threshold can be set according to actual conditions; no restrictions are imposed here.

[0057] If the regional overlap rate is not greater than a preset overlap rate threshold, the union region of the defect region in the positive sample defect image and the defect region in the negative sample defect image is retained, and the union region is set as the partially overlapping defect region. (Refer to...) Figure 4 In the positive and negative sample defect maps, two defect regions D1 and D2 are detected at the same location. The actual overlap rate of the two defect regions is calculated. If the actual overlap rate is small and does not exceed a preset overlap rate threshold, then the union of defect regions D1 and D2 is retained during the actual overlap process, and this union is set as the partially overlapping defect region. Figure 4 The region between D1 and D2.

[0058] Finally, the independent defect regions and the defect regions determined based on the overlapping regions are fused to generate a wafer cell defect image. This includes fusing independent defect regions, completely overlapping defect regions, and partially overlapping defect regions to generate a wafer cell defect image. (Refer to...) Figures 2-4 If defect regions A, C1, and D1 are detected simultaneously in the positive sample defect image, and defect regions B, C2, and D2 are detected simultaneously in the negative sample defect image, then the independent defect regions A and B, the completely overlapping defect region C2, and the partially overlapping region D1+D2 that are retained after fusion will be fused to generate the final wafer unit defect image.

[0059] In this application, the defect regions detected in positive sample defect images and negative sample defect images are fused to obtain the final wafer cell defect image. The positive sample detection model can ensure that significant and rare defects are not missed as much as possible, while the negative sample detection model can ensure that conventional defects, critical color differences, and diffusion features can be effectively detected, thereby maximizing the detection rate of wafer defects.

[0060] In one embodiment, considering that after detecting a wafer defect, the wafer anomaly needs to be processed according to the specific type of the wafer defect, the following steps can be performed after fusing the positive sample defect image and the negative sample defect image to generate a wafer cell defect image: First, the source of detection results for all defect regions in the wafer cell defect image is determined sequentially. The final wafer cell defect image may contain several defect regions; therefore, the detection source for each defect region in the image is determined sequentially, including sources from the positive sample detection model and the negative sample detection model. If the current defect region originates from the positive sample detection model, manual review is notified to determine the defect type. Specifically, considering that the positive sample detection model uses normal images for defect detection, while it can detect significant and rare wafer defects, it cannot directly determine the type of wafer defect. Therefore, if the defect region originates from the positive sample detection model, the system directly notifies manual review to confirm the defect type.

[0061] If the current defect region originates from the detection result of the negative sample detection model, then the defect type of the current defect region is output based on the detection result of the negative sample detection model. Specifically, the negative sample detection model is trained for defect identification and defect type identification during the training process; therefore, the defect type can be directly output for defects detected by the negative sample detection model.

[0062] If the current defect region originates from the detection results of both the positive and negative sample detection models, then the defect type of the current defect region is output based on the detection result of the negative sample detection model. Considering that the defect region after fusing the two defect results contains both the detection results of the positive and negative sample models, and the detection result of the negative sample model can directly output the defect type, the detection result of the negative sample detection model can be directly used as the defect type and output.

[0063] In this application, considering that after a wafer defect is detected, the wafer anomaly needs to be handled according to the specific type of the wafer defect, the detection source of different regions is distinguished to output the type of defect. Only defects detected by the positive sample detection model need to be manually judged, while defects detected by the negative sample detection model can directly output the defect type. Subsequently, the manual judgment can be directly performed based on the known defect type, which can save detection time and improve the efficiency of defect detection.

[0064] Figure 1 This is a flowchart illustrating a visual inspection method suitable for wafer surface images in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows; unless explicitly stated herein, there is no strict order requirement for the execution of these steps, and they can be executed in other orders; and Figure 1At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0065] Based on the above method, this application also discloses a visual inspection device suitable for wafer surface images.

[0066] Reference Figure 5 The device includes the following modules: The focus focal length determination module 501 is used to divide the wafer to be focused into several virtual unit image grids according to the preset image size of the acquired unit image, and determine the focus focal length of the unit image grid. The color image acquisition module 502 is used to acquire a unit color image of a number of unit image grids at the focusing focal length; The defect model detection module 503 is used to process several unit color images sequentially using a preset positive sample detection model and a negative sample detection model to generate positive sample defect images and negative sample defect images; The defect image generation module 504 is used to fuse positive sample defect images and negative sample defect images to generate wafer cell defect images.

[0067] In one embodiment, the focus distance determination module 501 is specifically used to sequentially measure the sharp focal length of each sampling point in a preset sampling point set on the wafer to be focused; generate a predicted wafer warpage model based on the coordinates of each sampling point and the corresponding measured sharp focal length; and determine the focus distance corresponding to the unit image to be acquired based on the wafer warpage model.

[0068] In one embodiment, the focus focal length determination module 501 is specifically used to calculate the reference focal length corresponding to multiple preset reference points on the wafer to be focused based on the wafer warping model; match the nearest reference point corresponding to the center point of the unit image to be acquired, and determine the preliminary focus range of the center point of the unit image based on the reference focal length corresponding to the nearest reference point; adjust the focus focal length corresponding to the center point of the unit image to be acquired based on the focus focal length of the previous unit image center point in the preset travel direction and the preliminary focus range, and set the adjusted focus focal length as the focus focal length corresponding to the unit image to be acquired.

[0069] In one embodiment, the defect image generation module 504 is specifically used to determine whether there are defect regions in the positive sample defect image and the negative sample defect image; if there are defect regions in the positive sample defect image and no defect regions in the negative sample defect image, then the positive sample defect image is retained and set as a wafer unit defect image; if there are no defect regions in the positive sample defect image and there are defect regions in the negative sample defect image, then the negative sample defect image is retained and set as a wafer unit defect image; if there are defect regions in both the positive sample defect image and the negative sample defect image, then the defect regions in the positive sample defect image and the defect regions in the negative sample defect image are fused to generate a wafer unit defect image.

[0070] In one embodiment, the defect image generation module 504 is specifically used to sequentially determine whether there is an overlapping region between the defect region in the positive sample defect image and the defect region in the negative sample defect image; if there is no overlapping region between the defect region in the positive sample defect image and the defect region in the negative sample defect image, then the defect regions in the positive sample defect image and the defect regions in the negative sample defect image are retained respectively, and the retained defect regions are set as independent defect regions; if there is an overlapping region between the defect region in the positive sample defect image and the defect region in the negative sample defect image, then the defect regions to be retained are determined according to the overlapping regions; the independent defect regions and the defect regions determined according to the overlapping regions are fused together to generate a wafer unit defect image.

[0071] In one embodiment, the defect image generation module 504 is specifically used to calculate the regional overlap rate of the overlapping regions and determine whether the regional overlap rate is greater than a preset overlap rate threshold; if the regional overlap rate is greater than the preset overlap rate threshold, the defect region in the negative sample defect image is retained, and the defect region in the negative sample defect image is set as a completely overlapping defect region; if the regional overlap rate is not greater than the preset overlap rate threshold, the union region of the defect region in the positive sample defect image and the defect region in the negative sample defect image is retained, and the union region is set as a partially overlapping defect region; fusing the independent defect regions and the defect regions determined according to the overlapping regions to generate a wafer unit defect image includes: fusing the independent defect regions, the completely overlapping defect regions and the partially overlapping defect regions to generate a wafer unit defect image.

[0072] In one embodiment, the defect image generation module 504 is further configured to sequentially determine the source of the detection results of all defect regions in the wafer cell defect image; if the current defect region originates from the detection result of the positive sample detection model, then a manual re-inspection is notified to determine the defect type of the current defect region; if the current defect region originates from the detection result of the negative sample detection model, then the defect type of the current defect region is output according to the detection result of the negative sample detection model; if the current defect region originates from the detection results of both the positive sample detection model and the negative sample detection model, then the defect type of the current defect region is output according to the detection result of the negative sample detection model.

[0073] The visual inspection device suitable for wafer surface images provided in this application embodiment can be applied to the visual inspection method suitable for wafer surface images provided in the above embodiments. For relevant details, please refer to the above method embodiments. The implementation principle and technical effect are similar, and will not be repeated here.

[0074] It should be noted that the visual inspection device suitable for wafer surface images provided in this embodiment is only illustrated by the above-described division of functional modules / units when performing visual inspection of wafer surface images. In practical applications, the above functions can be assigned to different functional modules / units as needed, that is, the internal structure of the visual inspection device suitable for wafer surface images can be divided into different functional modules / units to complete all or part of the functions described above. Furthermore, the implementation method of the visual inspection method suitable for wafer surface images provided in the above method embodiments and the implementation method of the visual inspection device suitable for wafer surface images provided in this embodiment belong to the same concept. The specific implementation process of the visual inspection device suitable for wafer surface images provided in this embodiment is detailed in the above method embodiments and will not be repeated here.

[0075] This application also discloses a computer device.

[0076] Specifically, such as Figure 6As shown, the computer device can be a desktop computer, laptop computer, handheld computer, or cloud server, etc. The computer device may include, but is not limited to, a processor and memory. The processor and memory can be connected via a bus or other means. The processor can be a Central Processing Unit (CPU). The processor can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, graphics processing units (GPUs), embedded neural network processing units (NPUs) or other dedicated deep learning coprocessors, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0077] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the methods in the above embodiments of this application. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the methods in the above embodiments. The memory may include a program storage area and a data storage area, wherein the program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0078] This application also discloses a computer-readable storage medium.

[0079] Specifically, the computer-readable storage medium is used to store a computer program, which, when executed by a processor, implements the methods described in the above-described method embodiments. Those skilled in the art will understand that implementing all or part of the processes in the methods described in the above-described embodiments of this application can be accomplished by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.

[0080] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A visual inspection method suitable for wafer surface images, characterized in that: The method includes: Based on the preset image size of the acquired unit image, the wafer to be focused is divided into several virtual unit image grids, and the focusing focal length of the unit image grid is determined. At the specified focusing focal length, a unit color image of several unit image grids is acquired; Several unit color images are processed sequentially using a preset positive sample detection model and a negative sample detection model to generate positive sample defect images and negative sample defect images; The positive sample defect image and the negative sample defect image are fused together to generate a wafer cell defect image.

2. The method according to claim 1, characterized in that: Determining the focal length of the unit image grid includes: Sequentially measure the sharp focal length of each sampling point within the preset sampling point set on the wafer to be focused; A predicted wafer warpage model is generated based on the coordinates of each sampling point and the corresponding measured sharp focal length. The focal length corresponding to the unit image to be acquired is determined based on the wafer warping model.

3. The method according to claim 2, characterized in that: The step of determining the focal length corresponding to the unit image to be acquired based on the wafer warpage model includes: The reference focal lengths corresponding to multiple preset reference points on the wafer to be focused are calculated based on the wafer warping model. Match the nearest reference point corresponding to the center point of the unit image to be acquired, and determine the initial focus range of the center point of the unit image based on the reference focal length corresponding to the nearest reference point; Based on the focal length of the center point of the previous unit image in the preset travel direction, and the focal length corresponding to the center point of the unit image to be acquired is adjusted according to the preliminary focusing range, the adjusted focal length is set as the focal length corresponding to the unit image to be acquired.

4. The method according to claim 1, characterized in that: The step of fusing the positive sample defect image and the negative sample defect image to generate a wafer cell defect image includes: Determine whether there are defect regions in the positive sample defect image and the negative sample defect image; If the positive sample defect image has a defect region and the negative sample defect image does not have a defect region, then the positive sample defect image is retained and set as a wafer cell defect image; If the positive sample defect image does not have a defect region, but the negative sample defect image does have a defect region, then the negative sample defect image is retained and set as a wafer cell defect image; If both the positive sample defect image and the negative sample defect image contain defect regions, then the defect regions in the positive sample defect image and the defect regions in the negative sample defect image are fused together to generate a wafer cell defect image.

5. The method according to claim 4, characterized in that: The step of fusing the defect regions in the positive sample defect image and the defect regions in the negative sample defect image to generate a wafer unit defect image includes: Sequentially determine whether there is an overlap between the defect region in the positive sample defect image and the defect region in the negative sample defect image; If the defect regions in the positive sample defect image and the defect regions in the negative sample defect image do not overlap, then the defect regions in the positive sample defect image and the defect regions in the negative sample defect image are retained respectively, and the retained defect regions are set as independent defect regions. If the defect region in the positive sample defect image and the defect region in the negative sample defect image overlap, then the defect region that needs to be retained is determined based on the overlapping region. The independent defect regions and the defect regions determined based on the overlapping regions are fused together to generate a wafer cell defect image.

6. The method according to claim 5, characterized in that: The step of determining the defective areas to be retained based on the overlapping areas includes: Calculate the overlap rate of the overlapping regions and determine whether the overlap rate is greater than a preset overlap rate threshold. If the overlap rate of the region is greater than the preset overlap rate threshold, then the defect region in the negative sample defect image is retained, and the defect region in the negative sample defect image is set as a completely overlapping defect region. If the overlap rate of the regions is not greater than the preset overlap rate threshold, then the union region of the defect region in the positive sample defect image and the defect region in the negative sample defect image is retained, and the union region is set as a partially overlapping defect region. The step of fusing the independent defect regions and the defect regions determined based on the overlapping regions to generate a wafer cell defect image includes: The independent defect region, the completely overlapping defect region, and the partially overlapping defect region are merged to generate a wafer cell defect image.

7. The method according to claim 1, characterized in that: After fusing the positive sample defect image and the negative sample defect image to generate a wafer cell defect image, the method further includes: The source of the detection results for all defect regions in the wafer unit defect image is determined sequentially; If the current defect area originates from the detection result of the positive sample detection model, then a manual re-inspection is notified to determine the defect type of the current defect area; If the current defect region originates from the detection result of the negative sample detection model, then the defect type of the current defect region is output according to the detection result of the negative sample detection model. If the current defect region originates from the detection results of the positive sample detection model and the negative sample detection model, then the defect type of the current defect region is output based on the detection result of the negative sample detection model.

8. A visual inspection device suitable for wafer surface images, characterized in that: The device includes: The focusing focal length determination module (501) is used to divide the wafer to be focused into several virtual unit image grids according to the preset image size of the acquired unit image, and determine the focusing focal length of the unit image grids. A color image acquisition module (502) is used to acquire a unit color image of a plurality of unit image grids at the focal length; The defect model detection module (503) is used to process several unit color images sequentially using a preset positive sample detection model and a negative sample detection model to generate positive sample defect images and negative sample defect images; The defect image generation module (504) is used to fuse the positive sample defect image and the negative sample defect image to generate a wafer unit defect image.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed according to any one of claims 1 to 7.

Images