Wafer edge contour image acquisition method and detection method

Abstract

The embodiment of the invention discloses a wafer edge contour image acquisition method and detection method, and the image acquisition method comprises the steps: enabling a wafer to be loaded on an objective table, and enabling the center of the wafer to be aligned with the rotation center of the objective table; the objective table drives the wafer to rotate around a rotation center, in the rotation process, a scanning sensor located on one side of the objective table continuously obtains local edge contour images of the wafer, and each rotation angle of the wafer corresponds to one local edge contour image; and arranging the plurality of local edge contour images according to the change sequence of the rotation angles, and splicing the images to form a complete edge contour image. According to the embodiment of the invention, the complete edge contour image is formed through automatic continuous sampling and image splicing, so that the edge contour of the wafer can be detected more comprehensively, and the problems of missing detection and contingency are reduced.

Description

Technical Field

[0001] This invention relates to the field of semiconductor inspection technology, and in particular to a method for acquiring and inspecting wafer edge contour images. Background Technology

[0002] The quality of the wafer edge profile (such as chipping, cracks, roughness, and chamfer shape) is a key indicator affecting its mechanical strength and reliability. Currently, edge profile measurement is typically performed using the following methods:

[0003] Manual sampling inspection: Operators manually search for and observe several points on the edge of the wafer (such as 0°, 90°, 180°, and 270°) under a microscope. This method is highly subjective, inefficient, and cannot reflect the edge quality of the entire wafer circumference.

[0004] Semi-automatic single-point measurement: A motorized stage moves a single, specific point on the edge of the wafer under a measurement sensor (such as a white light interferometer) for measurement. Although quantification is achieved, it is essentially still "the blind men and the elephant," unable to obtain continuous and complete information about the entire circumference.

[0005] The main drawbacks of existing technology:

[0006] Incomplete detection: Existing technologies can only select to obtain edge information of a single point on the wafer, and cannot obtain continuous and complete information of the entire circumference edge. The measurement results obtained based on the edge information of a single point are random and may miss major defects existing in other positions on the circumference, and cannot represent the overall edge quality of the wafer. Summary of the Invention

[0007] This invention provides a method for acquiring and detecting wafer edge contour images, which solves the problem that existing technologies can only select to acquire edge information of a single point on the wafer, and cannot acquire continuous and complete information of the entire circumferential edge, resulting in missed detections and randomness in the detection results, and failing to represent the overall edge quality of the wafer.

[0008] The present invention provides a method for obtaining a wafer edge contour image, comprising:

[0009] The wafer is mounted on a stage, with the center of the wafer aligned with the rotation center of the stage;

[0010] The stage drives the wafer to rotate around the rotation center. During the rotation, a scanning sensor located on one side of the stage continuously acquires local edge contour images of the wafer, so that each rotation angle of the wafer corresponds to one local edge contour image.

[0011] Multiple local edge contour images are arranged and stitched together according to the order of the rotation angle changes to form a complete edge contour image.

[0012] Furthermore, the step of stitching images together to form a complete edge contour image includes:

[0013] In the overlapping area of ​​adjacent contour features in two adjacent local edge contour images, feature points of the contour features in each local edge contour image are extracted respectively.

[0014] Using the feature points as a reference, the contour feature data of the overlapping region are subjected to weighted averaging to form a smooth and continuous complete edge contour image.

[0015] Furthermore, before performing weighted averaging on the contour feature data of the overlapping region, the method further includes:

[0016] The feature values ​​of the feature points in two adjacent local edge contour images are obtained respectively. When the two feature values ​​are not equal, the contour features in the local edge contour images are shifted and adjusted according to the difference between the two feature values ​​so that the positions of adjacent contour features are aligned.

[0017] Furthermore, the displacement adjustment of the contour features in the local edge contour image further includes:

[0018] First, a rigid body transformation is performed on the feature points in the subsequent local contour image to align them with the feature points in the preceding local contour image. Then, the rigid body transformation is applied to each point on the contour feature in the subsequent local contour image to achieve displacement adjustment of the contour feature in the subsequent local contour image.

[0019] Furthermore, the optimal rigid body transformation is calculated using the least squares method.

[0020] Furthermore, the feature points include extreme points of the contour feature height.

[0021] This invention also provides a method for detecting complete edge contour images obtained by the wafer edge contour image acquisition method, comprising: firstly, filtering out complete edge contour images that do not meet the uniformity requirements through uniformity evaluation, and then performing defect identification and defect localization on the complete edge contour images that do not meet the uniformity requirements through image processing to complete the detection.

[0022] Furthermore, the method for evaluating uniformity includes:

[0023] From the complete edge contour image, a contour feature is extracted at every first preset angle, and an analysis parameter is calculated from each contour feature to obtain multiple analysis parameters. The data dispersion of the multiple analysis parameters is calculated to obtain a uniformity evaluation.

[0024] Furthermore, the analysis parameters include edge tilt angle and / or chamfer height.

[0025] Furthermore, the method for defect identification and localization via image processing includes:

[0026] The complete edge contour image is processed to obtain a two-dimensional grayscale image. Edge detection or region growing algorithms are used to identify defect regions that have significant grayscale differences from normal contours. The geometric features of the defect differences are calculated, and the defects are identified and classified according to preset rules.

[0027] The defect is located by converting the pixel column where the centroid of the defect area is located into an angle value or angle range.

[0028] As can be seen from the above technical solutions, the present invention has the following advantages: The embodiments of the present invention achieve continuous acquisition of local edge contour images of the wafer instead of single-point selection acquisition through the cooperation of the stage and the scanning sensor. By arranging and stitching the local edge contour images in the order of rotation angle changes, a complete edge contour image is formed. The automatic continuous sampling and image stitching to form a complete edge contour image is conducive to more comprehensive detection of the wafer edge contour, reducing the problems of missed detection and randomness. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 A schematic flowchart illustrating a method for acquiring a wafer edge contour image according to an embodiment of the present invention;

[0031] Figure 2 This is a flowchart illustrating a wafer edge contour detection method provided in an embodiment of the present invention. Detailed Implementation

[0032] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0033] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the present application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0034] This invention provides a method for acquiring wafer edge contour images, such as... Figure 1 As shown, it includes:

[0035] S01. Load the wafer onto the stage, aligning the center of the wafer with the rotation center of the stage;

[0036] S02. The stage drives the wafer to rotate around the rotation center. During the rotation, the scanning sensor located on one side of the stage continuously acquires local edge contour images of the wafer, and obtains a local edge contour image corresponding to each rotation angle of the wafer.

[0037] S03. Arrange the local edge contour images according to the order of rotation angle changes and stitch the images together to form a complete edge contour image.

[0038] Understandably, in practice, this embodiment achieves continuous acquisition of local edge contour images of the wafer through the cooperation of the stage and scanning sensor, rather than single-point selective acquisition. The local edge contour images are arranged and stitched together according to the rotation angle change sequence to form a complete edge contour image. Therefore, the automatic continuous sampling and image stitching to form a complete edge contour image in this embodiment facilitates more comprehensive detection of the wafer's edge contour, reducing the problems of missed detections and randomness.

[0039] It should be noted that the stage rotates the wafer 360° around its center. Scanning sensors are positioned on the sides of the stage. The wafer is a thin, circular sheet, with a thickness much smaller than its diameter. The scanning area covers the curved surface of the wafer's side, the junction of the curved surface with the upper surface, and the junction of the curved surface with the lower surface. A local edge contour image can be captured by the scanning sensors every second preset rotation angle. Therefore, by arranging and stitching the local edge contour images according to the rotation angle (0°~360°), the resulting complete edge contour image is a contour line (or contour band). The corresponding local edge contour line can be indexed using the rotation angle.

[0040] In a more specific embodiment, high-precision rotational motion control and synchronously triggered sampling are used to achieve comprehensive data acquisition of the entire circumferential edge of the wafer, and image stitching technology is used to generate a full-circumferential contour image that can be used for quantitative analysis. The method specifically includes the following steps:

[0041] Step S100: Wafer loading and center positioning;

[0042] The wafer under test is automatically loaded onto a stage that can rotate 360 ​​degrees with precision. The wafer outline is identified by a vision system, and the center position of the wafer is automatically calculated and calibrated so that the center of the wafer coincides with the rotation center of the stage.

[0043] Step S200: Full-circle scan path planning and synchronization trigger settings;

[0044] Set the scanning parameters, including the starting angle, ending angle (typically 0° to 360°), and total number of scan points (or angular spacing Δθ). The system automatically plans the scanning path based on these parameters. The key innovation lies in establishing a strict synchronization relationship between the triggering of the rotation angle θ measurement sensor.

[0045] Step S300: Continuous circumferential scanning and data acquisition;

[0046] The stage is controlled to rotate the wafer at a constant angular velocity or in a stepping motion. During rotation:

[0047] When a line scan sensor is used, the sensor works continuously, and the system performs real-time, uninterrupted data acquisition to obtain continuous edge contour lines.

[0048] When using a point scanning sensor, the system employs a step-measurement mode: the stage rotates by a small angle Δθ (e.g., 0.1°) and then pauses, triggering the sensor to measure the contour data of the current point, before continuing to rotate to the next point. This cycle repeats until the entire circumference is measured.

[0049] In a more specific embodiment, arranging the local edge contour images according to the order of rotation angle changes includes:

[0050] Align the one-dimensional contour line (or two-dimensional contour band) collected at each angle θ with its angle information as the index;

[0051] All aligned one-dimensional contour lines are arranged sequentially (θ from 0° to 360°) to form a complete two-dimensional panoramic image. The X-axis of this image represents the circumferential angular coordinates of the wafer, and the Y-axis represents the contour height information from the wafer surface to the edge.

[0052] In a more specific implementation, this invention employs a transformation model based on a precision mechanical coordinate system and data coordinates. The specific process is as follows:

[0053] Before scanning, the deviation between the wafer center and the rotary table spindle center is accurately calibrated by a vision positioning system, and this deviation is used as a system error to be compensated in real time during coordinate transformation.

[0054] During data stitching, a coordinate transformation formula is applied to each data point: it is transformed from a local coordinate system with the sensor as the origin to a global cylindrical coordinate system with the wafer center as the origin and a rotation angle θ as the polar angle through a rotation matrix and a translation vector. This ensures that all data points are aligned in the same reference frame.

[0055] It is understandable that, in practical implementation, the rotation angle of this invention actually refers to the rotation angle of the wafer, because the center of the wafer and the center of the rotating stage are not necessarily completely coincident each time the wafer is placed. Assuming no coordinate transformation is performed: if the sensor measures a point P at a position of 30°, it is directly assumed that this point is on the circumference in the 30° direction. However, in reality, because the wafer center is offset, this point may actually be in the 30.5° direction.

[0056] In a more specific embodiment, the coordinate transformation method includes:

[0057] Sensor local coordinate system: When a sensor is measuring, its coordinate system origin is located at a fixed point on the sensor (e.g., the emission point of a laser beam or the center of the receiving lens). The data acquired by the sensor during each measurement is relative to this origin.

[0058] Global Coordinate System: We aim to establish a unified coordinate system with the center of the wafer as the origin. Since the wafer is circular, a cylindrical coordinate system (polar coordinates) can be used, where:

[0059] Origin: Center of the chip;

[0060] Polar angle θ: The angle measured from a reference direction (such as a notch or positioning edge of a wafer);

[0061] Radial distance r: the distance from the center of the wafer to the edge point;

[0062] Height z: The height of the edge point (relative to the height of the wafer surface);

[0063] During each measurement, the sensor can only acquire data in a local coordinate system. Furthermore, the wafer rotates on a rotary table, resulting in a different wafer position for each measurement point. Therefore, coordinate transformation is required to convert the local coordinate data obtained from each measurement to a global coordinate system.

[0064] Conversion principle:

[0065] First, transform the point (x1, y1, z1) in the sensor's local coordinate system to the global Cartesian coordinate system. Note that the origin of the global Cartesian coordinate system is the center of the wafer.

[0066] Let the transformation from the sensor coordinate system to the global Cartesian coordinate system be:

[0067] [X2]=R*[x1,y1,z1]^T+T

[0068] Where R is the rotation matrix and T is the translation vector.

[0069] Then, convert the global Cartesian coordinates (X2, Y2, Z2) to cylindrical coordinates (r, θ, z):

[0070] r = sqrt(X²^2 + Y²^2);

[0071] θ = atan2(Y2, X2) (θ is the angle in the Cartesian coordinate system);

[0072] After converting to cylindrical coordinates, we obtain (r, θ, z) for each point. Since we scan along the edge of the wafer, each point will have an angle θ and a radius r. Then, we generate an image using θ as the x-coordinate, z as the y-coordinate, and the value of r as the pixel value. In this way, the image can display the edge shape at the height z corresponding to each angle θ.

[0073] It should be noted that in the continuous local edge images obtained in the above embodiments, there will be overlapping parts of the contour features in adjacent local edge images. Therefore, when stitching adjacent local edge images, it is necessary to perform data fusion on the overlapping parts of the two images.

[0074] In a more specific embodiment, image stitching to form a complete edge contour image includes:

[0075] In the overlapping area of ​​adjacent contour features in adjacent local edge contour images, feature points of contour features in each image are extracted respectively;

[0076] Based on feature points, the contour feature data of overlapping areas are processed by weighted averaging to form a smooth and continuous complete edge contour image.

[0077] It is understood that, in specific implementation, this embodiment extracts feature points in the overlapping area from two adjacent edge contour images. The feature points in the two adjacent edge contour images correspond one-to-one with each other. Based on the feature points, the contour feature data of the overlapping area is weighted and averaged to form a smooth and continuous complete edge contour image.

[0078] It should be noted that in the above embodiments, adjacent contour features should be identical and continuous in the overlapping area. However, due to errors in the mechanical movement of the stage holding the wafer, when the scanning sensor measures different positions corresponding to different rotation angles of the wafer, the measurement data of the contour features in the two local edge contour images will have an overall deviation. If we directly switch from one contour feature to another at a certain position in the overlapping area, the feature value of the contour feature may undergo a sudden change, i.e., a jump, at this switching point.

[0079] In a more specific embodiment, the process further includes the following steps before performing weighted averaging on the contour feature data of the overlapping regions:

[0080] The feature values ​​of feature points in two adjacent local edge contour images are obtained respectively. When the feature values ​​of feature points in the two local edge contour images are not equal, the contour features in the local edge contour images are shifted and adjusted according to the difference between the two feature point values ​​so that the positions of adjacent contour features are equal.

[0081] Understandably, in practice, when two adjacent local edge contour images jump, the feature point values ​​corresponding to adjacent contour features in the overlapping area are not equal. The error between the two feature points is calculated, and the edge features in the local edge contour image are shifted and adjusted according to the error so that the adjacent contour features are aligned and the jump is eliminated.

[0082] In a more specific embodiment, adjusting the displacement of the local edge contour image further includes:

[0083] First, perform a rigid body transformation on the feature points in the subsequent local contour image to align them with the feature points in the preceding local contour image. Then, apply the rigid body transformation to each point on the contour feature in the subsequent local contour image to obtain the displacement-corrected subsequent local contour image.

[0084] Understandably, in practical implementation, rigid body transformation is used to translate and rotate the contour features in the local contour image while maintaining the shape, size, and internal distance of the graphic. Since this embodiment arranges the local edge contour images according to the order of rotation angle changes and then stitches them together to form a complete edge contour image, rigid body transformation is applied to the contour features in the subsequent local contour image to align them with the contour features in the preceding local contour image. If feature values ​​are unequal between any two adjacent local contour images, the contour features in the subsequent local contour image are adjusted to align with those in the preceding local contour image, eliminating the need for additional global correction and significantly improving stitching efficiency.

[0085] In a more specific embodiment, the optimal rigid body transformation is calculated using the least squares method.

[0086] Understandably, in practical implementation, the least squares method is suitable for real-time, low-computing-power, and multi-source data fusion. It has fast calculation speed, no iterative convergence problem, and reproducible results, meeting real-time requirements and significantly improving image stitching efficiency and computational efficiency.

[0087] In a more specific embodiment, the feature points include extreme points of the contour feature height. It is understood that by using extreme points of the contour feature height as feature points, it is possible to accurately reflect whether there are abrupt changes between contour features in adjacent local edge contour images.

[0088] It should be noted that the contour features in the above embodiments are contour lines or contour bands.

[0089] In a more specific embodiment, the present invention employs a feature-point-assisted weighted average fusion algorithm, the specific process of which is as follows:

[0090] Within the overlapping area of ​​adjacent contour lines, extract stable feature points (such as inflection points and extreme points of the contour).

[0091] Using these feature points as a reference, minute translations and rotations are made to achieve sub-pixel-level precision alignment of adjacent contour lines.

[0092] After alignment, the data in the overlapping areas are processed by weighted averaging, with the weight coefficients gradually changing from 0 to 1, thereby achieving a smooth transition of the data and generating a seamless continuous surface.

[0093] It is understandable that the full-circumference contour map generated in this embodiment is not only a "visualized image," but also a high-precision, quantifiable "data map." The coordinates (θ, h) of any pixel on the image strictly correspond to the physical location of a point on the wafer edge, thus laying a solid foundation for the precise quantitative analysis in step S500. The core innovation of this step lies in solving how to accurately stitch together multiple contour line data into a seamless, quantifiable panoramic scientific image.

[0094] In a more specific embodiment, the extreme points of the contour height of the 20% edge region in two adjacent local contour images are mainly used as feature points.

[0095] In a more specific embodiment, the method for aligning adjacent contour lines includes:

[0096] First, feature point extraction is performed: within the overlapping area, stable feature points are extracted from the two contour lines (here we take the extreme points of the contour height of the 20% region of the edge of the two images as feature points).

[0097] Feature point matching: Matching feature points on two contour lines to find corresponding point pairs.

[0098] Transformation scheme: Calculate the minute transformations (translation and rotation) between two contour lines by matching feature point pairs.

[0099] Contour alignment: Apply a calculated transformation to adjust one contour line so that it aligns with the other contour line in the overlapping area.

[0100] In a more specific embodiment, during conventional stitching: A section of contour line A is measured at angle θ = 45°, and an adjacent contour line B is measured at angle θ = 47°. Upon direct stitching, a 3μm jump in contour height is observed at 46°. The reason for this jump in conventional stitching is:

[0101] Suppose we have two adjacent contour lines A and B, which should be continuous in the overlapping area. However, due to errors in the mechanical movement of the stage holding the wafer when the sensor measures different positions of the wafer, the measurement data of the two contours will have an overall deviation. If we switch directly from one contour line to the other at a certain position in the overlapping area, the height value of the contour may have a sudden change, i.e., a jump, at this switching point.

[0102] Examples of jump transitions:

[0103] Contour line A = data from position 1 to position 3, contour line B = data from position 4 to position 6, and the overlapping area is positions 3 and 4;

[0104] At position 3: the height measured by contour line A is 100 μm, and the height measured by contour line B is 103 μm;

[0105] At position 4: the height measured by contour line A is 110 μm, and the height measured by contour line B is 113 μm;

[0106] That is, the contour height value jumps by 3um.

[0107] The adjusted stitching scheme 1 in this embodiment (extracting feature points and considering angle changes):

[0108] Feature point extraction: It was found that there is a clear contour extreme high point in the overlapping area of ​​the two contour lines A and B.

[0109] Due to equipment error, the coordinates (angle and height) of this extreme point differ slightly between the two contour lines. We need to use this feature point to correct contour line B, aligning it with contour line A.

[0110] step:

[0111] Feature point extraction: Extreme points are detected in the overlapping area of ​​contour lines A and B. Assuming that an extreme point P1 is detected in contour line A and an extreme point P2 is detected in contour line B, we consider P1 and P2 to correspond to the same physical point on the wafer edge.

[0112] Transformation calculation: We can calculate a translation vector using the coordinate difference between P1 and P2.

[0113] Apply the transformation: Apply the calculated transformation to all points of contour line B to align contour line B with contour line A.

[0114] Blending: After alignment, weighted blending is performed in the overlapping areas to obtain a smooth transition of the complete outline.

[0115] In a more specific embodiment, the method for calculating the translation vector includes:

[0116] Transform the extreme points of contour lines A and B into a rectangular coordinate system (the angle and height, which were originally polar coordinates, can be regarded as x and the height as y, thus forming a two-dimensional point set).

[0117] The optimal rigid body transformation (R, T) is calculated using the least squares method, which aligns the feature points of contour line B with the feature points of contour line A.

[0118] Apply this transformation to each point of the contour line B to obtain the corrected contour line B.

[0119] In a more specific embodiment, this embodiment provides an adjusted splicing scheme 2 (mean elimination of jumps, disregarding angle changes), specifically as follows:

[0120] Calculate the average jump: In the overlapping region, calculate the average height difference (3μm) between contour B and contour A.

[0121] Overall translation correction: Subtract 3μm from all height values ​​of contour B.

[0122] Weighted fusion: In the overlapping area, gradient weights are used to fuse the corrected contour B with contour A.

[0123] In a more specific embodiment, the weighted blending method is as follows: a gradual blending is performed in the overlapping region (46±0.5°):

[0124] ① At 45°-45.5°: Use the data from contour line A 100%;

[0125] ② At 45.5°-46.5°: 50% outline A + 50% outline B;

[0126] ③ At 46.5°-47.5°: Use the data from contour line B 100%.

[0127] In summary, the rotating full-circumference scanning scheme provided by this invention involves "controlling the wafer to rotate around its central axis, while simultaneously acquiring continuous or high-density contour data along the wafer circumference during rotation." This ensures that the wafer edge contour data is obtained through dynamic rotating scanning, rather than static single-point measurement. The measurement is comprehensive and thorough: it achieves 360-degree measurement of the wafer edge without blind spots, avoiding the randomness of sampling inspection and greatly reducing the risk of missed detections.

[0128] This invention also provides a method for detecting a complete edge contour image obtained based on any of the above embodiments, comprising:

[0129] First, images of complete edge contours that do not meet the uniformity requirements are selected through uniformity evaluation. Then, image processing is used to identify and locate defects in these images to complete the detection.

[0130] Understandably, in practice, this embodiment first screens out wafers that do not meet quality standards through uniformity evaluation, and then accurately locates defects through image processing and recognition. There is no need to perform image recognition processing on wafers with good uniformity, thus improving the efficiency of batch wafer inspection.

[0131] In a more specific embodiment, the method for evaluating uniformity includes:

[0132] From the complete edge contour image, a contour feature is extracted at every first preset angle. Analysis parameters are calculated from each contour feature to obtain multiple analysis parameters. The data dispersion of the multiple analysis parameters is calculated to obtain a uniformity evaluation.

[0133] Understandably, in practice, the uniformity of a wafer is evaluated by analyzing the fluctuations in the wafer's edge contour. The higher the degree of dispersion, the lower the uniformity evaluation and the better the wafer quality. Conversely, the lower the degree of dispersion, the higher the uniformity evaluation and the more likely the wafer is to have defects.

[0134] In a more specific embodiment, the analysis parameters include edge tilt angle and / or chamfer height.

[0135] In a more specific embodiment, the data dispersion includes standard deviation and / or range.

[0136] Understandably, in practice, the edge tilt angle directly reflects the degree of tilt deviation between the wafer edge and the reference plane, and the chamfer height corresponds to the dimensional stability of the edge transition area. The standard deviation and / or range of the two can transform the macroscopic uniformity of the contour shape into a quantifiable numerical indicator, avoiding the subjectivity and ambiguity of qualitative evaluation.

[0137] In a more specific embodiment, the method for defect identification and defect localization via image processing includes:

[0138] A two-dimensional grayscale image is obtained by performing grayscale processing on the complete edge contour image. Edge detection or region growing algorithms are used to identify defect regions that have significant grayscale differences from normal contours. The geometric features of the defect differences are calculated, and the defects are identified and classified according to preset rules.

[0139] The defect is located by converting the pixel column where the centroid of the defect area is located into an angle value or angle range.

[0140] Understandably, in practice, by identifying defective areas, calculating the geometric features of defect differences, classifying curves, and automatically converting the pixel column where the centroid of the defective area is located into angle values ​​or angle ranges for defect localization, the system can automatically locate the angular position of the defect on the wafer circumference (e.g., "a chipping defect exists at 120°"). Standardized angular position outputs facilitate accurate identification of defect locations and execution of corresponding operations through a unified angular benchmark (e.g., avoiding bonding processes in defective angle areas during packaging). When troubleshooting quality issues later, the processing records of the wafer in each process can be traced back through the angular parameters (e.g., cutting pressure and grinding speed at the corresponding angles), quickly locating the root cause of equipment failures or process deviations, and shortening the problem troubleshooting cycle, etc.

[0141] In a more specific embodiment, the method for analyzing a full-circumference contour image includes:

[0142] Step S500: Based on the analysis of the panoramic image, automatically analyze the generated full-circle contour image:

[0143] Defect identification and location: Through image processing algorithms, abnormal protrusions or depressions (i.e., chips or cracks) in the image are automatically identified, and their circumferential angle positions are accurately marked;

[0144] Parameter extraction: Automatically extract parameters such as edge tilt angle, chip size, and roughness in batches from any location or specified area on the panoramic image.

[0145] Uniformity evaluation: The uniformity of edge processing is quantitatively evaluated by observing the fluctuation of the entire circumferential contour.

[0146] In a more specific embodiment, the analysis process includes:

[0147] Uniformity evaluation process:

[0148] A statistical uniformity analysis method based on panoramic images is employed. On the panoramic image, a contour line is extracted at regular angles (e.g., 2°) along the angular axis. Key parameters, such as edge tilt angle and chamfer height, are calculated from each contour line. This results in 180 (or more) tilt angle and height data points. The standard deviation and range of these datasets are then calculated. The standard deviation and range are quantitative indicators of global uniformity. A smaller standard deviation indicates better processing consistency across the entire circumference.

[0149] Precise Defect Localization: Defect recognition and localization algorithms based on image processing and geometric feature analysis.

[0150] Recognition: Treat the panoramic image as a two-dimensional grayscale image, and use edge detection or region growing algorithms to identify regions that have significant grayscale differences (i.e., height differences) from the normal contours.

[0151] Location: The X coordinate (angular coordinate) of the defect is directly determined by the pixel column where the centroid of the region is located, and the system can automatically convert it into a specific angle value such as "123.5°".

[0152] Classification and Measurement: Calculate the geometric features of the defect area, such as area, perimeter, roundness, and depth. Automatically classify defects (chipping, cracks, etc.) according to preset rules (e.g., depth greater than a threshold, narrow shape is considered a crack).

[0153] Understandably, in practical implementation, the present invention, through step S500, achieves a leap from "determining whether a certain point is qualified" to "assessing the quality level of the entire wafer edge and accurately locating the root cause of the problem." The generated result is no longer a simple qualified / unqualified conclusion, but a comprehensive diagnostic report containing uniformity indicators, defect distribution maps, defect types, and precise locations. This allows process engineers to directly check the corresponding processing equipment based on the angular locations in the report (e.g., "chipping is concentrated near the 90°-270° axis"), greatly improving the efficiency of process optimization.

[0154] This invention also uses edge detection or region growing algorithms to identify regions with significant grayscale differences (i.e., height differences) from normal contours, thereby accurately locating the position of each defect. This allows process engineers to directly check the corresponding processing equipment based on the angular position in the report (e.g., "chipping is concentrated near the 90°-270° axis"), greatly improving the efficiency of process optimization.

[0155] In summary, this embodiment provides an analysis scheme based on panoramic images: it not only covers the method of "generating images" but also extends to the means of "using images" for novel analysis. For example, it can automatically locate the angular position of defects on the circumference (e.g., "there is a chip at 120°"), a capability that did not exist in the era of single-point measurement.

[0156] The following two examples illustrate the wafer edge contour detection process:

[0157] Example 1:

[0158] Suitable for 100% online inspection of edge contours on semiconductor production lines, where extremely high inspection speed is required.

[0159] System Configuration:

[0160] Measurement sensor: Line scan laser profilometer. This sensor can acquire the contour data of a line (containing hundreds of points) in an instant, with an extremely fast scanning speed (up to tens of thousands of contour lines per second).

[0161] Motion system: The wafer is driven to rotate at high speed by a high-precision air bearing spindle.

[0162] Specific implementation steps:

[0163] Wafer transfer and initiation: The transfer mechanism delivers the wafer at high speed to the spindle ejector pin and clamps it. The spindle then begins to accelerate to a preset constant high speed (e.g., 2000 RPM).

[0164] Synchronous scanning: Once the spindle speed stabilizes, the system sends a trigger signal. The line-scan laser profilometer is continuously triggered at a fixed high frequency (e.g., 50kHz) to acquire profile data of the wafer edge passing in front of it. Because the wafer is rotating at high speed, it is equivalent to the sensor performing a continuous and uninterrupted scan of the entire circumference.

[0165] Data association and image generation: The system records the spindle encoder angle value corresponding to each contour line data, strictly mapping the angle information to the contour data. Hundreds of thousands of contour lines can be acquired in a single rotation.

[0166] Real-time analysis and sorting: The data processing unit calculates key parameters (such as maximum chipping depth) in real time. After scanning, the grade is immediately determined. Based on the results, the system automatically sorts the wafers into "qualified" and "unqualified".

[0167] Example 2:

[0168] A measurement system using a high-precision air-bearing rotary table and a white light interferometer as sensors is employed.

[0169] Loading and positioning: The robotic arm places an 8-inch wafer onto a rotating stage. The vision system takes a picture, calculates the wafer's center, and automatically calibrates, with a concentricity error of less than 1μm.

[0170] Parameter settings: Set the scanning range to 0° to 360°, use the step-measurement mode, set the angular spacing Δθ to 0.5°, and scan a total of 720 points.

[0171] Scanning and Acquisition: The stage begins to rotate. It pauses for 10 milliseconds after every 0.5° rotation, during which time a white light interferometer is triggered to perform a rapid 3D scan of the current edge point, acquiring a contour line data approximately 100 μm long. After completion, the stage continues to rotate to the next point.

[0172] Image generation: After scanning, the system stitches together 720 contour lines in angular order to generate a full-circumference contour image with a resolution of 720 (angle) x 1000 (height point).

[0173] Results Analysis: The software automatically analyzed the panoramic image and found a 15μm deep chip at an angle of 123.5°, which was marked as non-conforming. The system generated a report and stored the entire circumference contour image for process engineers to analyze the cause of the chip.

[0174] As can be seen from the above, the embodiments of the present invention have the following advantages:

[0175] 1. Comprehensive and thorough measurement: It enables 360-degree measurement of the wafer edge without blind spots, avoiding the randomness of sampling inspection and greatly reducing the risk of missed detection.

[0176] 2. Intuitive and efficient: The generated panoramic image is very intuitive, allowing engineers to view the entire area like a map. Figure 1 This allows for rapid assessment of the overall quality of the edge and precise location of defects.

[0177] 3. Abundant data and strong analytical capabilities: It has obtained massive amounts of continuous contour data, which supports in-depth statistical analysis of edge uniformity, periodic defects, etc. of the entire circumference, providing unprecedented data support for process optimization.

[0178] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for acquiring a wafer edge contour image, characterized in that, include: The wafer is mounted on the stage, with the center of the wafer aligned with the rotation center of the stage; The stage drives the wafer to rotate around the rotation center. During the rotation, the scanning sensor located on one side of the stage continuously acquires local edge contour images of the wafer, so that each rotation angle of the wafer corresponds to a local edge contour image. Multiple local edge contour images are arranged and stitched together according to the order of rotation angle changes to form a complete edge contour image.

2. The method for acquiring a wafer edge contour image according to claim 1, characterized in that, The process of stitching images together to form a complete edge contour image includes: In the overlapping area of ​​adjacent contour features in two adjacent local edge contour images, feature points of the contour features in each local edge contour image are extracted respectively. Using the feature points as a reference, the contour feature data of the overlapping region are subjected to weighted averaging to form a smooth and continuous complete edge contour image.

3. The method for acquiring a wafer edge contour image according to claim 2, characterized in that, Before performing weighted averaging on the contour feature data of the overlapping region, the following steps are also included: The feature values ​​of the feature points in two adjacent local edge contour images are obtained respectively. When the two feature values ​​are not equal, the contour features in the local edge contour images are shifted and adjusted according to the difference between the two feature values ​​so that the positions of adjacent contour features are aligned.

4. The method for acquiring a wafer edge contour image according to claim 3, characterized in that, The displacement adjustment of the contour features in the local edge contour image further includes: First, a rigid body transformation is performed on the feature points in the subsequent local contour image to align them with the feature points in the preceding local contour image. Then, the rigid body transformation is applied to each point on the contour feature in the subsequent local contour image to achieve displacement adjustment of the contour feature in the subsequent local contour image.

5. The method for acquiring a wafer edge contour image according to claim 4, characterized in that, The optimal rigid body transformation is calculated using the least squares method.

6. A method for acquiring a wafer edge contour image according to claim 4 or 5, characterized in that, The feature points include the extreme points of the contour feature height.

7. A method for detecting a complete edge contour image obtained based on the wafer edge contour image acquisition method according to any one of claims 1-6, characterized in that, include: First, images of complete edge contours that do not meet the uniformity requirements are selected through uniformity evaluation. Then, image processing is used to identify and locate defects in these images to complete the detection.

8. The detection method according to claim 7, characterized in that, The method for evaluating uniformity includes: From the complete edge contour image, a contour feature is extracted at every first preset angle, and an analysis parameter is calculated from each contour feature to obtain multiple analysis parameters. The data dispersion of the multiple analysis parameters is calculated to obtain a uniformity evaluation.

9. The detection method according to claim 8, characterized in that, The analysis parameters include edge tilt angle and / or chamfer height.

10. The detection method according to claim 9, characterized in that, The method for defect identification and defect localization through image processing includes: The complete edge contour image is processed to obtain a two-dimensional grayscale image. Edge detection or region growing algorithms are used to identify defect regions that have significant grayscale differences from normal contours. The geometric features of the defect differences are calculated, and the defects are identified and classified according to preset rules. The defect is located by converting the pixel column where the centroid of the defect area is located into an angle value or angle range.

Images