Wafer appearance defect detection method, device and equipment

Abstract

The application discloses a wafer appearance defect detection method, device and equipment, which comprises the following steps: determining the starting column center coordinates and the ending column center coordinates of each row of crystal grains of a target wafer; traversing each row of crystal grains according to the starting column center coordinates and the ending column center coordinates, and controlling a height measuring sensor to measure the height of each crystal grain in the traversing process to obtain a height data set of each crystal grain; determining the height compensation data of the corresponding crystal grain according to the height data set of each crystal grain; obtaining a moving path and the height compensation data according to the path planning of each row of crystal grains, controlling a target camera to move relatively, adjust the height and take a picture to obtain a target image set; and identifying defects based on a pre-trained differential model and an AI model to obtain a first detection result. The application can improve the accuracy and efficiency of finished wafer appearance defect detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of semiconductor technology, and in particular to a wafer appearance defect detection method, device and equipment. BACKGROUND

[0002] In the production process of semiconductor products, appearance defect detection is performed on finished wafers to ensure product quality. In related technologies, AOI (Automatic Optical Inspection) equipment is mostly used for appearance defect detection. However, the existing AOI equipment still needs to be improved in terms of detection accuracy and detection efficiency. SUMMARY

[0003] The present application aims to at least solve one of the technical problems existing in the prior art. To this end, the present application provides a wafer appearance defect detection method, device and equipment, which can improve the accuracy and efficiency of finished wafer appearance defect detection.

[0004] In one aspect, the present application provides a wafer appearance defect detection method, comprising:

[0005] determining the starting column center coordinates and the ending column center coordinates of each row of dies of a target wafer;

[0006] traversing each row of dies according to the starting column center coordinates and the ending column center coordinates, and controlling the height sensor to measure the height of each die during the traversal process to obtain a height data set of each die;

[0007] determining height compensation data of the corresponding die according to the height data set of each die;

[0008] controlling the target camera to move, adjust height and take pictures according to the movement path obtained by path planning for each row of dies and the height compensation data, to obtain a target image set;

[0009] performing defect recognition on the target image set based on a pre-trained differential model and an AI model to obtain a first detection result.

[0010] According to some embodiments of the present application, the determination of the starting column center coordinates and the ending column center coordinates of each row of dies of the target wafer further comprises:

[0011] obtaining the Map coordinates and the theoretical mechanical coordinates of a preset mark of the target wafer;

[0012] adjusting the center of the field of view of the target camera to align with the center of the preset mark according to the theoretical mechanical coordinates, and determining the actual mechanical coordinates of the preset mark;

[0013] According to the Map coordinate and the actual mechanical coordinate of the preset mark, a coordinate system mapping transformation matrix is determined.

[0014] According to some embodiments of the present application, the determination of the starting column center coordinate and the ending column center coordinate of each row of dies of the target wafer comprises:

[0015] The first starting column center coordinate and the first ending column center coordinate of the Map of each row of dies of the target wafer are determined.

[0016] The traversal of each row of dies according to the starting column center coordinate and the ending column center coordinate comprises:

[0017] The first starting column center coordinate and the first ending column center coordinate are converted into the second starting column center coordinate and the second ending column center coordinate of the mechanical coordinate system according to the coordinate system mapping transformation matrix.

[0018] Each row of dies is traversed according to the second starting column center coordinate and the second ending column center coordinate.

[0019] According to some embodiments of the present application, the determination of the height compensation data of the corresponding die according to the height data set of each die comprises:

[0020] The data of the height data set of each die is sorted and evenly divided into N parts to obtain a first height data subset, N being a positive integer greater than 2.

[0021] The first 1 / N and the last 1 / N of the data of the first height data subset are removed to obtain a second height data subset.

[0022] The high-frequency data of the second height data subset is removed, and the average value of the remaining data is calculated to obtain the height compensation data of the corresponding die.

[0023] According to some embodiments of the present application, the defect identification of the target image set based on the pre-trained differential model and the AI model to obtain a first detection result comprises:

[0024] A first target image of a target die is obtained from the target image set.

[0025] The first target image is regionally identified based on a preset ROI region, and Blob analysis is performed to intercept a second target image corresponding to the region where the target die is located.

[0026] The second target image is aligned and adjusted based on a preset detection template, and a detection region is identified to obtain a third target image.

[0027] Performing defect recognition on the third target image based on a pre-trained difference model to determine a first defect area;

[0028] Taking a partial screenshot of the third target image based on the first defect area to obtain a defect image;

[0029] The defect image is classified based on a pre-trained defect classification model to obtain a first detection result.

[0030] According to some embodiments of the present invention, the performing of region recognition on the first target image based on a preset ROI region and performing Blob analysis to intercept a second target image corresponding to the region where the target grain is located may also include:

[0031] Performing region recognition on the first target image based on a preset ROI region to obtain a fourth target image;

[0032] Performing a clarity detection on the fourth target image, and outputting a second detection result if the clarity detection fails.

[0033] According to some embodiments of the present invention, the third target image includes a photosensitive area image and a conductor area image, and the performing defect recognition on the third target image based on a pre-trained differential model to determine the first defect area includes:

[0034] performing defect recognition on the photosensitive area image based on a pre-trained first difference model to obtain a photosensitive defect area;

[0035] performing defect recognition on the wire region image based on a pre-trained second difference model to obtain a wire defect region;

[0036] The photosensitive defect area and the wire defect area are merged to obtain a first defect area.

[0037] According to some embodiments of the present invention, taking a partial screenshot of the third target image based on the first defect area to obtain a defect image includes:

[0038] Expanding the area based on the first defect area to obtain a second defect area;

[0039] A partial screenshot of the third target image is taken according to the second defect area to obtain a defect image.

[0040] On the other hand, an embodiment of the present invention provides a wafer appearance defect detection device, comprising:

[0041] A first determination module is used to determine the center coordinates of the starting column and the ending column of each row of grains on the target wafer;

[0042] a height measurement module, configured to traverse each row of grains according to the center coordinates of the starting column and the center coordinates of the ending column, and control a height measurement sensor to measure the height of each grain during the traversal process to obtain a height data set for each grain;

[0043] A second determination module is configured to determine height compensation data of a corresponding die according to the height data set of each die;

[0044] An image acquisition module is used to control the target camera to perform relative movement, height adjustment, and take pictures based on the movement path obtained by path planning for each row of grains and the height compensation data, so as to obtain a target image set;

[0045] The defect detection module is used to identify defects in the target image set based on a pre-trained differential model and an AI model to obtain a first detection result.

[0046] On the other hand, an embodiment of the present invention provides a wafer appearance defect detection device, including a processor and a memory, wherein the memory stores a computer program, and when the processor runs the computer program, it is used to implement the wafer appearance defect detection method as described above.

[0047] The embodiments of the present invention have at least the following beneficial effects:

[0048] By traversing the target wafer row by row, the height of each grain can be measured and the height compensation data can be determined. Path planning and height compensation are used to control the movement, height adjustment and photo taking of the target camera. This can optimize the movement path of the target camera and ensure that the image obtained is clear. Defects in the target image set can be identified based on pre-trained differential models and AI models, which can improve the accuracy and efficiency of appearance defect detection on finished wafers.

[0049] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments with reference to the following drawings, in which:

[0051] Figure 1 Schematic diagram of a wafer appearance defect detection method according to an embodiment of the present invention;

[0052] Figure 2 Schematic diagram of the structure of a detection device according to an embodiment of the present invention;

[0053] Figure 3 is a schematic diagram of a partial structure of a target wafer according to an embodiment of the present invention;

[0054] Figure 4 A schematic diagram of path planning for a single row of dies according to an embodiment of the present invention;

[0055] Figure 5 is a schematic diagram of a target wafer according to an embodiment of the present invention;

[0056] Figure 6 for Figure 5 A partial enlarged view of the area marked with M1;

[0057] Figure 7 Schematic diagram of equal division of a height measurement data set of a target grain according to an embodiment of the present invention;

[0058] Figure 8 Schematic diagram of a detection template according to an embodiment of the present invention;

[0059] Figure 9 A partially enlarged view of a target wafer according to an embodiment of the present invention;

[0060] Figure 10 Schematic diagram of a photosensitive area image and a conductive line area image of a target wafer according to an embodiment of the present invention;

[0061] Figure 11 The template image of the first differential model and the image of the photosensitive area with defects according to the embodiment of the present invention;

[0062] Figure 12 The template image of the second differential model and the image of the defective wire region according to the embodiment of the present invention;

[0063] Figure 13 A schematic diagram of various defect types according to an embodiment of the present invention;

[0064] Figure 14 1 is a principle block diagram of a wafer appearance defect detection device according to an embodiment of the present invention;

[0065] Figure 15 This is a principle block diagram of a wafer appearance defect detection device according to an embodiment of the present invention.

[0066] Reference numerals:

[0067] XY stage 10, Z-axis lifting mechanism 20, industrial camera 21, telecentric lens 22, point light source 23, height sensor 24, target wafer 210, photosensitive area 211, wire area 212, first determination module 310, height measurement module 320, second determination module 330, image acquisition module 340, defect detection module 350, processor 410, memory 420. DETAILED DESCRIPTION

[0068] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0069] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they cannot be understood as limitations on the present invention.

[0070] In the description of the present invention, "several" means one or more, "multiple" means more than two, "greater than," "less than," and "exceed" are understood to exclude the number itself, and "above," "below," and "within" are understood to include the number itself. The use of terms such as "first" and "second" is solely for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly specifying the number of the indicated technical features, or implicitly specifying the order of the indicated technical features.

[0071] In the description of the present invention, unless otherwise clearly defined, words such as “setting”, “installation” and “connection” should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above words in the present invention based on the specific content of the technical solution.

[0072] Please refer to Figure 1 This embodiment discloses a method for detecting wafer appearance defects, including steps S100 to S500. It should be noted that the numbering of the steps in this embodiment is only for the convenience of review and understanding, and does not limit the execution order of the steps. Before describing the contents of each step, the hardware structure of the detection equipment involved in the detection method is briefly described. Figure 2 The detection equipment includes an XY stage 10 and a Z-axis lifting mechanism 20. The XY stage 10 is used to support the target wafer 210 (i.e., the wafer to be detected) and can move in both directions along the X and Y directions in the XY plane. The Z-axis lifting mechanism 20 is installed with an industrial camera 21 and a height sensor 24. The Z-axis lifting mechanism 20 is located above the XY stage 10 and can move linearly along the Z direction. The industrial camera 21 is connected to a telecentric lens 22 and a point light source 23. The industrial camera 21 is used for image acquisition. The height sensor 24 adopts a spectral confocal height sensor 24 with a height measurement interval of 10μm and an overlap accuracy of 0.8μm. The height data of different positions on the grain surface of the target wafer 210 are accurately measured.

[0073] The following is a detailed description of each step:

[0074] S100, determining the center coordinates of the starting column and the ending column of each row of dies on the target wafer 210;

[0075] For example, please refer to Figure 3 , Figure 3 The figure shows the local structure of the target wafer 210. The target wafer 210 is provided with a plurality of grains distributed in an array along the horizontal and vertical directions, as shown by the Die mark in the figure, and each grain has a certain length and width. The first complete grain in each row is taken as the starting grain, and the last complete grain in each row is taken as the ending grain. The outer edge of the starting grain is taken as the starting column, and the outer edge of the ending grain is taken as the ending column, so that the center coordinates of the starting column and the center coordinates of the ending column can be determined. Among them, the starting grain can be arranged on the left or on the right. If the starting grain is arranged on the left, the outer edge of the starting grain refers to the left edge of the starting grain, and the outer edge of the ending grain refers to the right edge of the ending grain; if the starting grain is arranged on the right, the outer edge of the starting grain refers to the right edge of the starting grain, and the outer edge of the ending grain refers to the left edge of the ending grain. For example, Figure 3 Point A in FIG. 1 shows the center of the starting column, point B shows the center of the ending column, and the line L between point A and point B is the scanning path for height measurement.

[0076] S200, traversing each row of grains according to the center coordinates of the starting column and the center coordinates of the ending column, and controlling the height measurement sensor 24 to measure the height of each grain during the traversal process to obtain a height data set of each grain;

[0077] For example, the relative positions of the target wafer 210 and the height sensor 24 are adjusted according to the center coordinates of the starting column and the center coordinates of the ending column, so that the height sensor 24 can move along the scanning path, thereby traversing each row of grains, and measuring the height of each grain during the traversal process, thereby obtaining a height data set for each grain, wherein the height data set for each grain includes height data at multiple positions on the surface of the grain. It should be noted that the target wafer 210 of this embodiment is placed on the XY stage 10, and the target wafer 210 can be driven by the XY stage 10 to move relative to the height sensor 24, thereby achieving traversal of each grain by the height sensor 24; in some application examples, for example, the height sensor 24 is installed on an XY moving mechanism, and the height sensor 24 can be driven by the XY moving mechanism to move in the X and Y directions, thereby traversing each grain in each row.

[0078] S300, determining height compensation data of the corresponding grain according to the height data set of each grain;

[0079] For example, during the production process, the target wafer 210 may have an uneven surface. Based on the height data set of each grain, the height compensation data of the corresponding grain is determined to facilitate subsequent photography compensation of each grain, thereby solving the problem of defocusing caused by the uneven surface of the target wafer 210 and the temperature drift of the target camera lens. This is beneficial to improving the clarity of grain image acquisition and thereby improving the accuracy of image recognition.

[0080] S400, according to the movement path and height compensation data obtained by path planning for each row of grains, controlling the target camera to perform relative movement, height adjustment and photo taking, to obtain a target image set;

[0081] For example, please refer to Figure 4 , Figure 4 : A schematic diagram of the photographing path and running direction of a row of grains on the target wafer 210 is shown in the figure. The mark Die in the figure represents a grain, wherein the photographing path and running direction are as shown by the arrows in the figure, and the starting point of the arrow is the photographing position. By planning the photographing path, it is beneficial to increase the photographing speed, reduce the idle time of the axis, and improve the photographing efficiency. It should be noted that in this embodiment, the control of the target camera (i.e., the industrial camera 21 mentioned above) to move relative to the target camera can be achieved by driving the target wafer 210 to move relative to the target camera in the XY plane by the XY stage 10, or, if the target camera is installed on the XY moving mechanism, the target camera can be moved in the X and Y directions under the drive of the XY moving mechanism, thereby achieving relative movement. During the movement process, the height of the target camera is adjusted according to the height compensation data of each grain, which can ensure that the target camera can obtain the clearest image when taking pictures, which is beneficial to improve the accuracy of image recognition.

[0082] S500: Perform defect recognition on the target image set based on the pre-trained differential model and AI model to obtain a first detection result.

[0083] For example, defect identification on a target image set based on a pre-trained differential model and an AI model can significantly reduce labor intensity and improve detection efficiency, accuracy, and stability compared to manual inspection. The AI ​​(artificial intelligence) model can be trained using existing deep neural networks or fine-tuned from a larger model, such as the defect classification model described below.

[0084] Through the above scheme, by traversing the target wafer 210 row by row, the height of each grain can be measured and the height compensation data can be determined, and the target camera movement, height adjustment and photo taking can be controlled through path planning and height compensation. The moving path of the target camera can be optimized and the image obtained by taking pictures can be ensured to be clear. In addition, the target image set can be identified for defects based on the pre-trained differential model and AI model, which can improve the accuracy and efficiency of the appearance defect detection of the finished wafer.

[0085] In some application examples, before step S100, the following steps are further included:

[0086] Obtaining the map coordinates and theoretical mechanical coordinates of the preset marks of the target wafer 210;

[0087] According to the theoretical mechanical coordinates, the center of the field of view of the target camera is adjusted to align with the center of the preset mark to determine the actual mechanical coordinates of the preset mark;

[0088] Determine the coordinate system mapping transformation matrix based on the preset marked map coordinates and the actual machine coordinates.

[0089] For example, a map is a visual graphic used to represent the position and status of each grain on a wafer during the semiconductor manufacturing process. The map coordinates record the relative position of the grain on the target wafer 210, while the mechanical coordinates record the relative position of the grain relative to the coordinate origin of the detection device after the target wafer 210 is placed on the detection device (such as the XY stage 10). In order to facilitate the positioning of each grain on the target wafer 210, some features of the target wafer 210 are selected as alignment marks, i.e., preset marks, before production. For example, the edge of the target wafer 210 is used as a preset mark, such as Figure 5 As shown in the figure, M1, M2 and M3 are marked in the figure. The enlarged view of the area marked M1 can be referred to Figure 6 , Figure 6 The middle mark M1 -C represents the center of the mark M1 , and is positioned by the edge mark without the need for additional alignment marks, thus preventing damage to the target wafer 210 and providing strong compatibility.

[0090] Since there may be a certain positional deviation between the actual placement position of the target wafer 210 and the theoretical placement position during the actual production process, resulting in a decrease in the positioning accuracy of the theoretical mechanical coordinates of the target wafer 210, it is necessary to recalibrate the mechanical coordinates of the target wafer 210. According to the theoretical coordinates of the preset mark, the center of the field of view of the target camera and the center of the preset mark (such as Figure 6The target wafer 210 and the target camera are aligned (as shown by the mark M1-C). For example, based on the theoretical coordinates of the preset mark, the relative position of the target wafer 210 and the target camera is adjusted so that the preset mark enters the field of view of the target camera. The distance between the preset mark and the center of the field of view of the target camera is then fine-tuned to align the center of the field of view of the target camera with the center of the preset mark. At this time, the mechanical coordinates of the preset mark are the actual mechanical coordinates. Relative movement between the target wafer 210 and the target camera can be achieved by driving the target wafer 210 relative to the target camera in the XY plane via the XY stage 10. Alternatively, if the target camera is mounted on an XY motion mechanism, the target camera can be driven by the XY motion mechanism to move in the X and Y directions, thereby achieving relative movement.

[0091] Based on the preset map coordinates and the actual machine coordinates, a coordinate system mapping transformation matrix can be calculated between the map coordinate system and the machine coordinate system. Since the map coordinates corresponding to each die on the target wafer 210 are known, the coordinate system mapping transformation matrix can be used to convert the map coordinates of any point on the target wafer 210 into machine coordinates, facilitating calculation and motion control.

[0092] Accordingly, step S100 includes: determining the first starting column center coordinates and the first ending column center coordinates of the map of each row of dies on the target wafer 210;

[0093] In step S200, each row of grains is traversed according to the center coordinates of the starting column and the center coordinates of the ending column, including:

[0094] Converting the first starting column center coordinate and the first ending column center coordinate into the second starting column center coordinate and the second ending column center coordinate of the mechanical coordinate system according to the coordinate system mapping transformation matrix;

[0095] Each row of grains is traversed according to the center coordinates of the second starting column and the center coordinates of the second ending column.

[0096] For example, the map records the map coordinates of each die on target wafer 210. By querying the map, the center coordinates of the first starting column and the first ending column of each row of die on target wafer 210 can be easily determined. The center coordinates determined in this case are all map coordinates. Because map coordinates are relative coordinates, they are not convenient for calculation and motion control. Coordinate transformation is performed based on the coordinate system mapping transformation matrix, that is, converting map coordinates into mechanical coordinates, which facilitates calculation and motion control.

[0097] The target camera's shooting height compensation has a significant impact on the clarity of the image obtained by shooting. In order to improve the clarity of the photographed image, step S300 includes:

[0098] Sort the data of the height data set of each grain and divide it into N parts, to obtain a first height data subset, N is a positive integer greater than 2;

[0099] Remove the first 1 / N and the last 1 / N of the data of the first height data subset, to obtain a second height data subset;

[0100] Remove the high frequency data of the second height data subset, and calculate the average value of the remaining data to obtain the height compensation data of the corresponding grain.

[0101] For example, the height measurement interval of the height measurement sensor 24 is 10 μm, which is smaller than the surface area of a single grain. When measuring the height of a single grain, the height data of different positions on the grain surface can be measured. Please refer to Figure 7 Sort the height data of a single grain and divide it into N parts, for example, 6 parts. Remove the first 1 / 6 and the last 1 / 6 of the height data (i.e. invalid data), and keep the middle 4 / 6 of the height data (i.e. valid data) to obtain data with higher accuracy. Then remove the high frequency data therein to obtain the remaining data. By calculating the average value of the remaining data, the height compensation data of the grain is obtained.

[0102] Automatic defect detection is performed through the difference model and the AI model, which can greatly improve the detection efficiency and the accuracy of detection. Step S500 includes:

[0103] Obtain a first target image of a target grain from a target image set;

[0104] Perform region recognition on the first target image based on a preset ROI region, and perform Blob analysis to obtain a second target image corresponding to the region where the target grain is located;

[0105] Align and adjust the second target image based on a preset detection template, and identify the region to be detected to obtain a third target image;

[0106] Perform defect recognition on the third target image based on a pre-trained difference model to determine a first defect region;

[0107] Perform local screenshot on the third target image based on the first defect region to obtain a defect image;

[0108] Classify the defect image based on a pre-trained defect classification model to obtain a first detection result.

[0109] Exemplarily, the target image set includes images of all the grains on the target wafer 210, and defects are identified for each grain one by one. A first target image of the target grain is obtained from the target image set, and then the area where the target grain is located is identified from the first target image and a local image of the area is captured for precise detection. Blob in computer vision refers to a connected area in the image. Blob analysis is to extract and mark the connected domain of the binary image after foreground / background separation. For example, the content of Blob analysis includes pre-processing of grayscale threshold, length, width, aspect ratio, opening operation, and closing operation. During the photo shooting process, the placement posture of the target wafer 210 may deviate from the theoretical posture to a certain extent, resulting in a certain deviation between the contour and angle of the image and the preset detection template. Therefore, it is necessary to align and adjust the second target image and identify the area to be detected. By performing defect recognition on the third target image based on the pre-trained differential model, it is possible to determine whether there are defects in the third target image and the area where the defects are located. By taking a local screenshot of the third target image according to the area where the defects are located, the local image containing the defects can be accurately obtained, reducing the interference features in the defect image, which is conducive to the defect classification model to accurately identify and classify the defect type, thereby obtaining an accurate first detection result.

[0110] The size of the preset detection template is the same as that of the first target image, and is configured with multiple detection areas. The detection template is as follows: Figure 8 As shown in the figure, the squares represent the detection area. A positioning mark is placed at the center of the detection template to facilitate acquisition of image coordinate and angle data. During detection, the positioning mark at the center of the detection template is aligned with the center of the current image to determine the location of the current detection area. The current image is then rotated and translated to fully overlap the detection template, achieving alignment. Blob analysis can be used to identify the area to be detected.

[0111] In some application examples, region recognition is performed on the first target image based on a preset ROI region, and blob analysis is performed to intercept a second target image corresponding to the region where the target die is located. Previously, the following steps were also included:

[0112] Performing region recognition on the first target image based on a preset ROI region to obtain a fourth target image;

[0113] A clarity detection is performed on the fourth target image, and if the clarity detection fails, a second detection result is output.

[0114] For example, ROI refers to a region of interest (ROI), which selects a specific area from an image to reduce interference from irrelevant information. Regional recognition is performed on the first target image based on the ROI region, focusing on the area to be analyzed to obtain a fourth target image. The fourth target image is then subjected to clarity testing, for example, by using a Laplace image processing algorithm to calculate the clarity of the fourth target image. If the clarity test fails, a second test result is output. For example, if the target die is defective, if the clarity test passes, regional recognition is performed on the first target image based on the preset ROI region, and blob analysis is performed to capture a second target image corresponding to the area where the target die is located. This ensures that the clarity of the image to be tested meets requirements and avoids misjudgments due to unclear images.

[0115] Please refer to Figure 9 and Figure 10 The target wafer 210 includes a photosensitive area 211 and a wire area 212. The third target image includes a photosensitive area image and a wire area image. The photosensitive area image is as follows: Figure 10 As shown in (a), the wire area image is as follows Figure 10 As shown in (b), defect recognition is performed on the third target image based on the pre-trained differential model to determine the first defect area, including:

[0116] Defect recognition is performed on the photosensitive area image based on the pre-trained first difference model to obtain a photosensitive defect area;

[0117] For example, the first differential model needs to be pre-trained before detection. For example, a defect-free photosensitive area image is selected, and a differential template with the same size as the first target image and the entire detection area is prepared for model training to obtain the first differential model. Defect recognition is performed on the photosensitive area image based on the first differential model, wherein the template image of the first differential model is as follows: Figure 11 As shown in (a), the image of the defective photosensitive area is as follows Figure 11 As shown in (b), the defect recognition conditions are used for screening to obtain the length, width and area of ​​the defect area, and then compared with the preset parameter threshold. When the parameter exceeds the standard, the corresponding area is marked as a photosensitive defect area, as shown by the mark Q1.

[0118] Defect recognition is performed on the wire area image based on the pre-trained second difference model to obtain the wire defect area;

[0119] For example, the training and recognition principle of the second differential model is the same as that of the first differential model, except that the training image of the second differential model is an image of a non-defective wire area. After the defect recognition of the second differential model, the wire defect area can be obtained. For example, the template image of the second differential model is as follows: Figure 12As shown in (a), the image of the defective wire area is as follows Figure 12 As shown in (b), the wire defect area is marked as Q2. By using the first differential model and the second differential model to identify defects in the photosensitive area 211 and the wire area 212 respectively, accurate identification can be performed based on the different characteristics of the photosensitive area 211 and the wire area 212, which is conducive to improving the accuracy of identification.

[0120] The photosensitive defect area and the wire defect area are merged to obtain a first defect area.

[0121] Exemplarily, the types of defects include scratches, color anomalies, dirt, and other types. In most cases, the location of the defect will not appear in a single location, for example, it will not appear only in the photosensitive area 211 or only in the wire area 212. In other words, the defect may appear simultaneously in adjacent photosensitive areas 211 and wire areas 212, or it may simultaneously span multiple photosensitive areas 211 and wire areas 212. The defects identified by the first differential model and the second differential model are discrete defect areas. For example, the photosensitive defect area marks the defect of the photosensitive area 211, but does not mark the defect of the wire area 212. If the defect spans multiple photosensitive areas 211 and wire areas 212, only the corresponding photosensitive area 211 is marked in the photosensitive defect area. Therefore, the discrete photosensitive defect areas and wire defect areas need to be merged to obtain a complete defect area, i.e., the first defect area.

[0122] After obtaining the first defect area, it can be determined whether there is a defect. If so, a partial image of the defect is captured to facilitate defect classification. The process of capturing a partial image of the third target image based on the first defect area to obtain a defect image includes:

[0123] Expanding the area based on the first defect area to obtain a second defect area;

[0124] A partial screenshot of the third target image is taken according to the second defect area to obtain a defect image.

[0125] For example, the first defect area is a precise area containing defects. The field of view of the image may be relatively narrow and contain less feature content. Based on the first defect area, the area range is expanded. For example, the length and width of each area are expanded by 20 pixels respectively to obtain a second defect area with a wider field of view and relatively more feature content. For example, the boundary of the defect is clearer, so as to facilitate the identification and classification of the defect type through the defect classification model. A partial screenshot of the third target image is taken according to the second defect area to obtain a defect image, such as Figure 13 Various types of defect images are shown in FIG.

[0126] The embodiment can detect defects of the photosensitive area 211 and the wire area 212 respectively, can adapt to characteristics of different areas, can improve detection accuracy, can merge the detected photosensitive defect area and the wire defect area, can expand the area range, can obtain a second defect area with a wider visual range, can take a local screenshot according to the second defect area, can focus on a local image containing defect characteristics, and can perform defect classification based on a defect classification model. In this way, the detection and classification of defects are performed in steps, which is beneficial to taking into account the accuracy and efficiency of defect identification.

[0127] Please refer to Figure 14 The embodiment of the present application also provides a wafer appearance defect detection device, which comprises:

[0128] The first determination module 310 is configured to determine the starting column center coordinates and the ending column center coordinates of each row of dies of the target wafer 210.

[0129] The height measurement module 320 is configured to traverse each row of dies according to the starting column center coordinates and the ending column center coordinates, and control the height measurement sensor 24 to measure the height of each die in the traversal process, to obtain a height data set of each die.

[0130] The second determination module 330 is configured to determine height compensation data of each die according to the height data set of each die.

[0131] The image acquisition module 340 is configured to control the target camera to move, adjust the height and take a picture according to the movement path and the height compensation data obtained by path planning of each row of dies, to obtain a target image set.

[0132] The defect detection module 350 is configured to perform defect identification on the target image set based on a pre-trained differential model and an AI model, to obtain a first detection result.

[0133] The inventive concept of the wafer appearance defect detection device embodiment is the same as that of the wafer appearance defect detection method embodiment described above. The content not involved in the wafer appearance defect detection device embodiment can refer to the wafer appearance defect detection method embodiment described above, which will not be described here again. By traversing the target wafer 210 row by row, height measurement and determination of height compensation data of each die are realized, and the movement, height adjustment and photographing of the target camera are controlled through path planning and height compensation, which can optimize the movement path of the target camera and ensure that the photographed image is clear. In addition, defect identification is performed on the target image set based on a pre-trained differential model and an AI model, which can improve the accuracy and efficiency of wafer appearance defect detection of finished products.

[0134] Please refer to Figure 15, an embodiment of the present invention also provides a wafer appearance defect detection device, including a processor 410 and a memory 420, wherein a computer program is stored in the memory 420, and the processor 410 is used to implement the wafer appearance defect detection method as described above when running the computer program. The content of the wafer appearance defect detection method can be referred to above and will not be repeated here. By traversing the target wafer 210 row by row, the height of each grain is measured and the height compensation data is determined, and the target camera movement, height adjustment and photography are controlled through path planning and height compensation. The moving path of the target camera can be optimized and the image obtained by the photography can be ensured to be clear, and the target image set can be identified for defects based on the pre-trained differential model and AI model, thereby improving the accuracy and efficiency of the finished wafer appearance defect detection.

[0135] The embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary technicians in the relevant technical field without departing from the scope of the present invention.

Claims

1. A method for detecting wafer appearance defects, characterized in that: include: Determine the center coordinates of the starting column and the ending column of each row of the target wafer; wherein the first complete grain in each row is used as the starting grain, the last complete grain in each row is used as the ending grain, the outer edge of the starting grain is used as the starting column, and the outer edge of the ending grain is used as the ending column, thereby determining the center coordinates of the starting column and the ending column; Traversing each row of grains according to the center coordinates of the starting column and the center coordinates of the ending column, and controlling a height measuring sensor to measure the height of each grain during the traversal process to obtain a height data set for each grain; wherein the height data set for each grain includes height data of multiple positions on the surface of the grain; Determining height compensation data of the corresponding grain according to the height data set of each grain; According to the movement path obtained by path planning for each row of grains and the height compensation data, the target camera is controlled to perform relative movement, height adjustment, and photographing to obtain a target image set; Perform defect recognition on the target image set based on the pre-trained differential model and the AI ​​model to obtain a first detection result; Determining the height compensation data of the corresponding die according to the height data set of each die includes: sorting the data of the height data set of each grain and dividing the data into N parts to obtain a first height data subset, where N is a positive integer greater than 2; Removing the first 1 / N and last 1 / N of the data from the first height data subset to obtain a second height data subset; removing high-frequency data of the second height data subset and calculating an average value of the remaining data to obtain height compensation data of the corresponding grains; The pre-trained differential model and AI model are used to perform defect recognition on the target image set to obtain a first detection result, including: Acquire a first target image of a target grain from the target image set; Performing region recognition on the first target image based on a preset ROI region, and performing Blob analysis to intercept a second target image corresponding to the region where the target grain is located; Based on a preset detection template, aligning and adjusting the second target image and identifying the area to be detected to obtain a third target image; Performing defect recognition on the third target image based on a pre-trained difference model to determine a first defect area; Taking a partial screenshot of the third target image based on the first defect area to obtain a defect image; The defect image is classified based on a pre-trained defect classification model to obtain a first detection result.

2. The wafer appearance defect detection method according to claim 1, characterized in that: The step of determining the starting column center coordinates and the ending column center coordinates of each row of the target wafer further includes: Obtain the map coordinates and theoretical mechanical coordinates of the preset marks on the target wafer; According to the theoretical mechanical coordinates, the center of the field of view of the target camera is adjusted to align with the center of the preset mark to determine the actual mechanical coordinates of the preset mark; A coordinate system mapping transformation matrix is ​​determined according to the Map coordinates of the preset mark and the actual machine coordinates.

3. The wafer appearance defect detection method according to claim 2, characterized in that: Determining the starting column center coordinates and the ending column center coordinates of each row of grains on the target wafer includes: Determine the center coordinates of the first starting column and the first ending column of the map of each row of dies on the target wafer; The traversing each row of grains according to the center coordinates of the starting column and the center coordinates of the ending column includes: Converting the first starting column center coordinates and the first ending column center coordinates into the second starting column center coordinates and the second ending column center coordinates of the mechanical coordinate system according to the coordinate system mapping transformation matrix; Each row of grains is traversed according to the second starting column center coordinates and the second ending column center coordinates.

4. The wafer appearance defect detection method according to claim 1, characterized in that: The method further includes: performing region recognition on the first target image based on a preset ROI region, performing Blob analysis, and intercepting a second target image corresponding to the region where the target grain is located. Performing region recognition on the first target image based on a preset ROI region to obtain a fourth target image; Performing a clarity detection on the fourth target image, and outputting a second detection result if the clarity detection fails.

5. The wafer appearance defect detection method according to claim 1, characterized in that: The third target image includes a photosensitive area image and a conductor area image, and the performing defect recognition on the third target image based on the pre-trained differential model to determine the first defect area includes: performing defect recognition on the photosensitive area image based on a pre-trained first difference model to obtain a photosensitive defect area; performing defect recognition on the wire region image based on a pre-trained second difference model to obtain a wire defect region; The photosensitive defect area and the wire defect area are merged to obtain a first defect area.

6. The wafer appearance defect detection method according to claim 5, characterized in that: Taking a partial screenshot of the third target image based on the first defect area to obtain a defect image includes: Expanding the area based on the first defect area to obtain a second defect area; A partial screenshot of the third target image is taken according to the second defect area to obtain a defect image.

7. A wafer appearance defect detection device, characterized in that: include: A first determination module is configured to determine the center coordinates of the starting column and the ending column of each row of grains on the target wafer; wherein the center coordinates of the starting column and the ending column are determined by taking the first complete grain in each row as the starting grain, the last complete grain in each row as the ending grain, the outer edge of the starting grain as the starting column, and the outer edge of the ending grain as the ending column; a height measurement module, configured to traverse each row of grains according to the center coordinates of the starting column and the center coordinates of the ending column, and control a height measurement sensor to measure the height of each grain during the traversal process to obtain a height data set for each grain; wherein the height data set for each grain includes height data of multiple positions on the surface of the grain; A second determination module is configured to determine height compensation data of a corresponding die according to the height data set of each die; An image acquisition module is used to control the target camera to perform relative movement, height adjustment, and take pictures based on the movement path and the height compensation data obtained by path planning for each row of grains to obtain a target image set; A defect detection module, configured to perform defect recognition on the target image set based on a pre-trained differential model and an AI model to obtain a first detection result; Determining the height compensation data of the corresponding die according to the height data set of each die includes: sorting the data of the height data set of each grain and dividing the data into N parts to obtain a first height data subset, where N is a positive integer greater than 2; Removing the first 1 / N and last 1 / N of the data from the first height data subset to obtain a second height data subset; removing high-frequency data of the second height data subset and calculating an average value of the remaining data to obtain height compensation data of the corresponding grains; The pre-trained differential model and AI model are used to perform defect recognition on the target image set to obtain a first detection result, including: Acquire a first target image of a target grain from the target image set; Performing region recognition on the first target image based on a preset ROI region, and performing Blob analysis to intercept a second target image corresponding to the region where the target grain is located; Based on a preset detection template, aligning and adjusting the second target image and identifying the area to be detected to obtain a third target image; Performing defect recognition on the third target image based on a pre-trained difference model to determine a first defect area; Taking a partial screenshot of the third target image based on the first defect area to obtain a defect image; The defect image is classified based on a pre-trained defect classification model to obtain a first detection result.

8. A wafer appearance defect detection device, comprising a processor and a memory, wherein the memory stores a computer program, characterized in that: When the processor runs the computer program, it is used to implement the wafer appearance defect detection method according to any one of claims 1 to 6.

Images