Wafer post-scribe product cutting defect detection method, device and medium

Abstract

The application provides a wafer post-scribing product cutting defect detection method, device and medium, and the method comprises the following steps: image preprocessing is performed on a to-be-detected picture; image segmentation is performed on the preprocessed to-be-detected picture through an image segmentation algorithm, the to-be-detected region is divided into a cutting track region and a background region, and a skeleton region of the image is extracted; a plurality of intersection points of horizontal and vertical cutting tracks are detected through an intersection point recognition algorithm; two intersection points in a preset direction are tracked through a path tracking algorithm to obtain a cutting path; and quantitative measurement and qualitative identification of a wave defect are performed through a peak-valley algorithm. The application does not require a large number of defect sample pictures and does not require professional knowledge related to images; can effectively reduce the influence of dirt, light and blur on the detection result; can effectively reduce the misrecognition rate of a film scratch; can effectively measure the degree of chip collapse caused by cutting deviation, can reduce the misjudgment rate of a slightly collapsed product, and can increase the warehousing rate of the product.

Description

Technical Field

[0001] This invention relates to the field of wafer inspection technology, and in particular to a method, equipment and medium for detecting dicing defects in products after wafer dicing. Background Technology

[0002] Currently, defect detection methods for specific areas generally fall into two categories: one uses traditional algorithms. For example, before detection, a reference path for the cutting trajectory, the cutting width, and an error tolerance value are defined. During detection, a feature localization algorithm aligns the detection image with the template image, and then aligns it with the reference path. Defects are identified by comparing the cutting path of the detection image with the reference path. The accuracy of this method depends on the precision of the localization and the reasonableness of the manually preset reference path. Another approach involves manually designing defect features based on the characteristic morphology of the defect, and using quantitative and qualitative methods to precisely control the defect.

[0003] Another type uses supervised deep learning methods, such as YOLO-based object detection, to identify and detect abnormal defects. This method doesn't require quality control operators to have specialized knowledge of the algorithms, but its biggest problem is that it uses supervised deep learning algorithms, requiring a large number of defect samples to achieve a satisfactory detection result. However, in actual production environments, most defect samples are very difficult to collect, which poses a significant challenge to the use of this defect detection method. Furthermore, deep learning-based methods can only perform qualitative identification and cannot accurately perform quantitative measurement. Summary of the Invention

[0004] To achieve the above-mentioned objectives and other advantages of the present invention, a first objective of the present invention is to provide a method for detecting dicing defects in products after wafer dicing, comprising the following steps:

[0005] Image preprocessing is performed on the images to be detected;

[0006] The image segmentation algorithm is used to segment the preprocessed image to be detected, dividing the detection area into the cutting edge area and the background area, and extracting the skeleton area of ​​the image.

[0007] Multiple intersections of horizontal and vertical cutting paths are detected using an intersection point recognition algorithm;

[0008] The cutting path is obtained by tracking two intersection points in a preset direction using a path tracing algorithm;

[0009] Quantitative measurement and qualitative identification of wave defects are performed using peak-valley algorithm.

[0010] Furthermore, the image preprocessing of the image to be detected includes the following steps:

[0011] Stretch the overall grayscale to the preset range;

[0012] Gabor filters at 0° and 90° are used to filter out scratch interference in any direction and enhance the edge texture features of the cut path.

[0013] The processed grayscale values ​​are linearly normalized to the range [0, 255].

[0014] Furthermore, the image segmentation of the preprocessed image to be detected using an image segmentation algorithm includes image segmentation using a binary segmentation algorithm;

[0015] The extracted skeleton region of the image includes the skeleton region of the binary image obtained by image thinning algorithm.

[0016] Furthermore, the detection of multiple intersections of horizontal and vertical cutting paths using the intersection recognition algorithm includes the following steps:

[0017] Iterate through all contours and filter candidate rectangular contours based on their area and rectangularity.

[0018] Select a candidate contour and take the point that is furthest from the starting point of the candidate contour as the first intersection point;

[0019] Among the candidate contours, the point furthest from the first intersection point is selected as the second intersection point;

[0020] Among the candidate contours, the point that is furthest from the sum of the first and second intersection points is selected as the third intersection point;

[0021] In the candidate contours, the point with the greatest sum of distances to the first and second intersections is selected as the first candidate intersection. In the second and third intersections, the point with the greatest sum of distances to the second and third intersections is selected as the second candidate intersection. In the third and first intersections, the point with the greatest sum of distances to the third and first intersections is selected as the third candidate intersection. Finally, the point with the largest sum of distances among the first, second, and third candidate intersections is selected as the fourth intersection.

[0022] Based on clockwise sorting, we obtain the new first intersection point, second intersection point, third intersection point, and fourth intersection point;

[0023] Repeat the steps of selecting candidate contours, taking the point with the maximum preset distance from the starting point of the candidate contour as the first intersection point, and sorting them clockwise to obtain new first, second, third, and fourth intersection points, to obtain one or more sets of rectangular contour structures composed of clockwise intersection point information.

[0024] Furthermore, the step of tracking two intersections in a preset direction to obtain the cutting path through the path tracing algorithm includes using a breadth-first search algorithm to traverse the shortest path of adjacent intersections sorted clockwise, thereby tracking the cutting path and obtaining the cutting paths of the four sides of the chip.

[0025] Furthermore, the shortest path traversal of adjacent intersections sorted clockwise includes the following steps:

[0026] Starting from the current intersection point among the adjacent intersection points sorted clockwise, and using the current intersection point as the starting point, find the coordinate point that is immediately adjacent to the next intersection point in the direction from the current intersection point and has not yet been visited;

[0027] Starting from the found coordinates, continue searching for adjacent, unvisited coordinates until the next intersection is found, then end the search;

[0028] Backtracking from the predecessor point of the next intersection point to the current intersection point, the backtracking route is the shortest path from the current intersection point to the next intersection point.

[0029] Furthermore, the quantitative measurement and qualitative identification of wave defects using the peak-to-valley algorithm includes the following steps:

[0030] If the difference between the peaks and troughs of the cutting path in a single direction is greater than a preset threshold, it is judged as a defective wavy edge cutting.

[0031] If the length of the cutting path in a single direction is not within the preset range, it is judged as a defective wavy edge cutting.

[0032] Furthermore, the quantitative measurement and qualitative identification of wave defects using the peak-to-valley algorithm includes the following steps:

[0033] Based on the coordinates (x, y) of the points on the path, a one-dimensional sequence T is established with the path point number as the X-axis and the corresponding y-coordinate as the Y-axis.

[0034] One-dimensional Gaussian filtering is used to remove possible spikes in sequence T;

[0035] Find consecutive pairs of local maxima and minima;

[0036] The severity of the wavy edge defect is quantitatively measured by the difference between the maximum and minimum values ​​of each point pair, and qualitatively judged based on whether it exceeds a preset threshold.

[0037] A second objective of the present invention is to provide an electronic device comprising: a memory having program code stored thereon; a processor connected to the memory, and, when the program code is executed by the processor, implementing a method for detecting dicing defects in products after wafer dicing.

[0038] A third objective of this invention is to provide a computer-readable storage medium having program instructions stored thereon, which, when executed, implement a method for detecting dicing defects in products after wafer dicing.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] This invention provides a method, equipment, and medium for detecting dicing defects in products after wafer dicing. It utilizes a manually designed defect feature algorithm, requiring no large number of defect sample images or specialized image processing knowledge; only a basic understanding of the defect's process and quality control specifications is needed. The preprocessing method, designed specifically for the customer's scenario, effectively reduces the impact of dirt, lighting, and blur on the detection results, thereby improving the robustness of the detection algorithm. The intersection point recognition algorithm effectively reduces the false recognition rate of thin-film scratches, accurately and effectively identifying the die dicing path. The peak-valley localization algorithm effectively measures the degree of chip damage caused by dicing deviation, reducing the false positive rate of slightly damaged products and increasing the product acceptance rate.

[0041] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0042] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0043] Figure 1 This is a schematic diagram of the cutting defects in Example 1;

[0044] Figure 2 This is a flowchart of the method for detecting dicing defects in products after wafer dicing, as described in Example 1.

[0045] Figure 3 This is a flowchart of the image preprocessing process for the image to be detected in Example 1;

[0046] Figure 4 This is a schematic diagram of the pretreatment process in Example 1;

[0047] Figure 5 This is a schematic diagram of the 90-degree filtering process in Example 1;

[0048] Figure 6 This is a schematic diagram of the 0-degree filtering process in Example 1;

[0049] Figure 7 This is a schematic diagram of the detailed processing of Example 1;

[0050] Figure 8 This is a flowchart of Example 1 showing the detection of multiple intersections of horizontal and vertical cutting channels using an intersection recognition algorithm;

[0051] Figure 9 This is a schematic diagram of the cutting path in Example 1;

[0052] Figure 10 This is a flowchart of the shortest path for traversing adjacent intersections in clockwise order in Example 1.

[0053] Figure 11 This is a flowchart of the quantitative measurement and qualitative identification of wave defects using the peak-valley algorithm in Example 1.

[0054] Figure 12 This is a flowchart for the quantitative measurement of wavy edge cutting defects in the cutting path of Example 1.

[0055] Figure 13 This is a schematic diagram of the electronic device in Example 2;

[0056] Figure 14 This is a schematic diagram of the storage medium in Example 3. Detailed Implementation

[0057] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0058] Example 1

[0059] This invention provides a method for detecting dicing defects in products after wafer dicing, primarily applied to the inspection of surface defects in semiconductor wafers after dicing without film expansion. It addresses wavy dicing defects (such as those caused by wafer dicing equipment) presenting at the chip edges. Figure 1 As shown, quantitative measurement of peaks and valleys and qualitative identification of defects are performed.

[0060] This invention utilizes image preprocessing and image segmentation techniques to obtain the cutting area, and obtains the sub-pixel level cutting path through intersection recognition, path tracing, and sub-pixel fitting algorithms. It also achieves quantitative measurement and qualitative identification of wavy cutting defects through peak-valley search algorithms.

[0061] Specifically, a method for detecting dicing defects in products after wafer dicing, such as... Figure 2 As shown, it includes the following steps:

[0062] S1. Perform image preprocessing on the image to be detected; such as... Figure 3 As shown, the specific steps include:

[0063] S11. An image stretching algorithm is used to stretch the overall grayscale to a preset range to ensure the consistency of subsequent processing. The stretching formula is shown in Equation 1.

[0064]

[0065] Where max_value is the maximum value of the grayscale value, and min_value is the minimum value of the grayscale value.

[0066] S12. Use Gabor filters in the 0° and 90° directions to filter out scratch interference in any direction and enhance the edge texture features of the cut edge. The Gabor filter is shown in Equation 2.

[0067]

[0068] S13. The grayscale values ​​after the above processing are linearly normalized to the range of [0, 255].

[0069] The image before preprocessing is as follows Figure 4 As shown, the preprocessed image is as follows Figure 5 (90-degree filter) Figure 6 As shown in (0-degree filtering).

[0070] In this embodiment, the image to be detected is first processed using preprocessing methods such as contrast stretching, directional filtering, and image enhancement to specialize the cutting area, so as to ensure that the scratches and cutting lines in the area have sufficient distinguishability.

[0071] S2. The preprocessed image to be detected is segmented using an image segmentation algorithm, dividing the area to be detected into the cutting edge area (bright area) and the background area (dark area), and the skeleton area of ​​the image is extracted.

[0072] This embodiment uses a binary segmentation algorithm for image segmentation, which is shown in Equation 3.

[0073]

[0074] Then, an image thinning algorithm is used to extract the skeleton region of the binary image in advance, such as... Figure 7 As shown.

[0075] This embodiment uses image segmentation technology to divide the area to be detected into a bright area (i.e., the cutting area) and a dark area (i.e., the background area), thereby reducing interference for subsequent operations.

[0076] S3. Multiple intersections of horizontal and vertical cutting paths are detected using an intersection recognition algorithm; in this embodiment, the intersection recognition algorithm is used to detect four intersections of horizontal and vertical cutting paths.

[0077] like Figure 8 As shown, detecting multiple intersections of horizontal and vertical cutting paths using an intersection recognition algorithm includes the following steps:

[0078] S31. Traverse all contours and filter candidate rectangular contours C{s0…….si} based on contour area and rectangularity.

[0079] S32. Select candidate contour si, and take the point si[k1] that is the furthest from the starting point si[0] of candidate contour si as the first intersection point k1;

[0080] S33. Among the candidate contours si, select the point si[k2] that is farthest from the first intersection point k1 as the second intersection point k2;

[0081] S34. Among the candidate contours si, select the point si[k3] that is farthest from the first intersection point k1 and the second intersection point k2 as the third intersection point k3;

[0082] S35. Among the candidate contours si, select the point si[k4-1] that is farthest from the first intersection point k1 and the second intersection point k2 in the first intersection and the second intersection point [k1,k2] respectively as the first candidate intersection point k4-1. Select the point si[k4-2] that is farthest from the second intersection point k2 and the third intersection point k3 in the second intersection and the third intersection point [k2,k3] as the second candidate intersection point k4-2. Select the point si[k4-3] that is farthest from the third intersection point k3 and the first intersection point k1 in the third intersection and the first intersection point [k3,k1] as the third candidate intersection point k4-3. Select the point with the largest sum of distances among the first candidate intersection point k4-1, the second candidate intersection point k4-2, and the third candidate intersection point k4-3 as the fourth intersection point K4.

[0083] S36. According to the clockwise order, we get the new first intersection point k1, second intersection point k2, third intersection point k3, and fourth intersection point k4;

[0084] S37. Repeat steps S32 to S36 to obtain one or more rectangular outline structures composed of clockwise intersection information.

[0085] S4. The cutting path is obtained by tracking the two intersection points in the preset direction using a path tracing algorithm. This includes using a breadth-first search algorithm to traverse the shortest paths of adjacent intersection points k1-k2, k2-k3, k3-k4, and k4-k5 in clockwise order, thereby tracking the cutting path and obtaining the cutting paths for the four sides of the chip. The effect is as follows: Figure 9 As shown.

[0086] Let's take the search of k1-k2 as an example to illustrate, such as... Figure 10 As shown, the shortest path that iterates through adjacent intersections in clockwise order includes the following steps:

[0087] S41. Starting from the current intersection point k1 among the adjacent intersection points sorted clockwise, with the current intersection point k1 as the starting point, find the coordinate point that is adjacent to the current intersection point k1 in the direction to the next intersection point k2 and has not yet been visited, that is, the upper right corner point, the right side point, and the lower right corner point of the eight neighborhood of the current intersection point k1.

[0088] S42. Then, starting from these found coordinate points, continue to search for adjacent, unvisited coordinate points until the next intersection point k2 is found, and end the search.

[0089] S43. Backtracking back to the current intersection point k1 based on the predecessor point of the next intersection point k2, the backtracking route is the shortest path from the current intersection point k1 to the next intersection point k2.

[0090] S5. Quantitative measurement and qualitative identification of wave defects are performed using peak-to-valley algorithms. For example... Figure 11 As shown, the specific steps include:

[0091] S51. Determine whether the difference between the peaks and troughs of the cutting path in a single direction is greater than a preset threshold.

[0092] S52. Determine whether the length of the cutting path in a single direction is within the preset range;

[0093] S53. If the difference between the peak and trough of the cutting path in a single direction is greater than the preset threshold, it is determined to be a defective wavy edge cutting.

[0094] If the length of the cutting path in a single direction is not within the preset range, it is judged as a defective wavy edge cutting.

[0095] S54. If the difference between the peaks and troughs of the cutting path in a single direction is not greater than the preset threshold, it is determined to be a defective cut without wavy edges.

[0096] If the length of the cutting path in a single direction is within the preset range, it is judged as a defective cut without wavy edges.

[0097] The quantitative measurement of wavy edge cutting defects along cutting paths k1-k2 will be used as an example for illustration. Figure 12 As shown, the quantitative measurement and qualitative identification of wave defects using the peak-to-valley algorithm includes the following steps:

[0098] S511. Based on the coordinates (x, y) of the points on the path, establish a one-dimensional sequence T with the path point number as the X-axis and the corresponding y-coordinate as the Y-axis.

[0099] S512. Use a one-dimensional Gaussian filter to remove any glitch points that may exist in sequence T.

[0100] S513. Find consecutive pairs of local maxima and minima;

[0101] S514. The severity of the wavy edge defect is quantitatively measured based on the difference between the maximum and minimum values ​​of each point pair, and qualitatively judged based on whether it exceeds a preset threshold.

[0102] This embodiment uses an intersection point recognition algorithm to detect four intersection points of the horizontal and vertical cutting paths. Then, it uses path tracing technology to track the two intersection points in the clockwise direction to obtain the cutting path. Finally, it uses a peak-valley algorithm to perform quantitative measurement and qualitative identification of wave defects.

[0103] This invention provides a method for detecting dicing defects in products after wafer dicing. It utilizes a manually designed defect feature algorithm, requiring no large number of defect sample images or specialized image processing knowledge; only a basic understanding of the defect's process and quality control specifications is needed. The preprocessing method, designed specifically for the customer's scenario, effectively reduces the impact of dirt, lighting, and blur on the detection results, thereby improving the robustness of the detection algorithm. The intersection point recognition algorithm effectively reduces the false recognition rate of thin-film scratches, accurately and effectively identifying the die dicing path. The peak-valley localization algorithm effectively measures the degree of chip damage caused by dicing deviation, reducing the false positive rate of slightly damaged products and increasing the product acceptance rate.

[0104] Example 2

[0105] An electronic device, such as Figure 13 As shown, it includes: a memory storing program code; and a processor connected to the memory, which, when executed by the processor, implements a method for detecting dicing defects in products after wafer dicing. For a detailed description of the method, please refer to the corresponding description in the above method embodiments, which will not be repeated here.

[0106] Example 3

[0107] A computer-readable storage medium, such as Figure 14 As shown, it stores program instructions, which, when executed, implement a method for detecting dicing defects in products after wafer dicing. For a detailed description of the method, please refer to the corresponding description in the above method embodiments; it will not be repeated here.

[0108] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0109] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0110] The above are merely embodiments of this specification and are not intended to limit the scope of one or more embodiments of this specification. Various modifications and variations can be made to one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of one or more embodiments of this specification.

Claims

1. A method for detecting dicing defects in products after wafer dicing, characterized in that, Includes the following steps: Image preprocessing is performed on the images to be detected; The image segmentation algorithm is used to segment the preprocessed image to be detected, dividing the detection area into the cutting edge area and the background area, and extracting the skeleton area of ​​the image. Multiple intersections of horizontal and vertical cutting paths are detected using an intersection point recognition algorithm; The cutting path is obtained by tracking two intersection points in a preset direction using a path tracing algorithm; Quantitative measurement and qualitative identification of wave defects are performed using peak-to-valley algorithms. The detection of multiple intersections of horizontal and vertical cutting paths using the intersection recognition algorithm includes the following steps: Iterate through all contours and filter candidate rectangular contours based on their area and rectangularity. Select a candidate contour and take the point that is furthest from the starting point of the candidate contour as the first intersection point; Among the candidate contours, the point furthest from the first intersection point is selected as the second intersection point; Among the candidate contours, the point that is furthest from the sum of the first and second intersection points is selected as the third intersection point; In the candidate contours, the point farthest from the first and second intersection points is selected as the first candidate intersection point. The point farthest from the second and third intersection points is selected as the second candidate intersection point. The point farthest from the third and first intersection points is selected as the third candidate intersection point. Finally, the point with the largest sum of distances among the first, second, and third candidate intersection points is selected as the fourth intersection point. Based on clockwise sorting, we obtain the new first intersection point, second intersection point, third intersection point, and fourth intersection point; Repeat the steps of selecting candidate contours, taking the point with the maximum preset distance from the starting point of the candidate contour as the first intersection point, and sorting them clockwise to obtain new first, second, third, and fourth intersection points, to obtain one or more sets of rectangular contour structures composed of clockwise intersection point information.

2. The method for detecting dicing defects in products after wafer dicing as described in claim 1, characterized in that: The image preprocessing of the image to be detected includes the following steps: Stretch the overall grayscale to the preset range; Gabor filters at 0° and 90° are used to filter out scratch interference in any direction and enhance the edge texture features of the cut path. The processed grayscale values ​​are linearly normalized to the range [0, 255].

3. The method for detecting dicing defects in products after wafer dicing as described in claim 1, characterized in that: The image segmentation of the preprocessed image to be detected by the image segmentation algorithm includes image segmentation by the binary segmentation algorithm; The extracted skeleton region of the image includes the skeleton region of the binary image obtained by image thinning algorithm.

4. The method for detecting dicing defects in products after wafer dicing as described in claim 1, characterized in that: The method of tracking two intersections in a preset direction to obtain the cutting path includes using a breadth-first search algorithm to traverse the shortest path of adjacent intersections sorted clockwise, thereby tracking the cutting path and obtaining the cutting paths of the four sides of the chip.

5. The method for detecting dicing defects in products after wafer dicing as described in claim 4, characterized in that: The process of traversing the shortest path between adjacent intersections in clockwise order includes the following steps: Starting from the current intersection point among the adjacent intersection points sorted clockwise, and using the current intersection point as the starting point, find the coordinate point that is immediately adjacent to the next intersection point in the direction from the current intersection point and has not yet been visited; Starting from the found coordinates, continue searching for adjacent, unvisited coordinates until the next intersection is found, then end the search; Backtracking from the predecessor of the next intersection point to the current intersection point, the backtracking route is the shortest path from the current intersection point to the next intersection point.

6. The method for detecting dicing defects in products after wafer dicing as described in claim 5, characterized in that: The quantitative measurement and qualitative identification of wave defects using the peak-to-valley algorithm includes the following steps: If the difference between the peaks and troughs of the cutting path in a single direction is greater than a preset threshold, it is judged as a defective wavy edge cutting. If the length of the cutting path in a single direction is not within the preset range, it is judged as a defective wavy edge cutting.

7. The method for detecting dicing defects in products after wafer dicing as described in claim 6, characterized in that: The quantitative measurement and qualitative identification of wave defects using the peak-to-valley algorithm includes the following steps: Based on the coordinates (x, y) of the points on the path, a one-dimensional sequence T is established with the path point number as the X-axis and the corresponding y-coordinate as the Y-axis. One-dimensional Gaussian filtering is used to remove possible spikes in sequence T; Find consecutive pairs of local maxima and minima; The severity of the wavy edge defect is quantitatively measured by the difference between the maximum and minimum values ​​of each point pair, and qualitatively judged based on whether it exceeds a preset threshold.

8. An electronic device, characterized in that, include: A memory that stores program code; A processor connected to the memory, which, when the program code is executed by the processor, implements the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that, It stores program instructions that, when executed, implement the method as described in any one of claims 1 to 7.

Images