A cutting and positioning method and system for a semiconductor slicing machine

Abstract

This invention relates to the field of cutting and positioning technology, and more particularly to a cutting and positioning method and system for a semiconductor slicing machine. The method includes: acquiring an initial image of the wafer to be positioned; converting the initial image into a grayscale image; calculating a robust score for each pixel in the grayscale image; calculating a first anomaly index and a second anomaly index to characterize the degree of image contamination based on the grayscale image; calculating an adaptive mask threshold based on the first and second anomaly indices; comparing the robust score of each pixel with the adaptive mask threshold to generate a mask for shielding contaminated areas; and using the mask to perform template matching between the grayscale image and a preset template image to determine the position of the positioning mark. This invention improves the cutting success rate of the final product and meets the needs of wafer cutting.

Description

Technical Field

[0001] This invention relates to the field of cutting and positioning technology, and in particular to a cutting and positioning method and system for a semiconductor slicing machine. Background Technology

[0002] In the automated production process of semiconductor wafers, dicing is one of the key processes. This process uses a high-precision dicing machine to cut along pre-set scribe lines on the wafer surface, thus dividing a complete wafer into individual chips. The positioning accuracy of the dicing directly determines the yield and performance of the chips. During wafer processing, it is necessary to determine the coordinates of the positioning marks, and then determine the position of the scribe lines based on the coordinates of the positioning marks. The positioning marks are generally cross-shaped. During semiconductor manufacturing, after the wafer has undergone multiple processes, the positioning marks are prone to becoming blurred, incomplete, or even covered by foreign objects (such as cooling water droplets), and the cooling water during the dicing process accelerates the blurring of the positioning marks. If the positioning marks are blurred or damaged, the accuracy of the dicing will be lower, reducing the yield of the wafers.

[0003] Chinese patent document CN104484868B discloses a method for aerial tracking of moving targets that combines template matching and image contours, including the following steps: 1. Perform template matching to obtain the target position, and use correlation coefficient matching to perform template matching; 2. Determine the detection range based on the center of the circle obtained from the previous frame; 3. Perform image binarization to obtain the contour map of the image, and obtain a set of candidate circles by creating circles that surround the closed contour; 4. Obtain the circle with the radius closest to half the side length of the target in the detection area, obtain the target circle, and convert it to the coordinates of the full image frame.

[0004] During wafer dicing, if the positioning marks become blurred or damaged, a template matching algorithm can be used to determine their precise location by comparing a pre-set template image with the positioning marks in a real-time acquired grayscale image. To improve the accuracy and robustness of the matching, this method is usually combined with masking technology to shield non-target areas in the image that interfere with the matching results. However, the actual production environment is complex and variable, and the wafer surface is highly susceptible to various contaminants such as dust, slurry splashes, cooling water mist, resin residue, and bubbles. These contaminants can severely interfere with the template matching algorithm, and since the mask threshold is usually a fixed threshold, it is easy to cause misjudgments and missed detections, resulting in low matching accuracy and, further, large errors in the position coordinates of the acquired positioning marks, affecting the yield of the wafer dicing. Summary of the Invention

[0005] To address the problem that a fixed mask threshold cannot meet the requirements for wafer dicing, this invention provides a dicing positioning method and system for semiconductor slicing machines.

[0006] In a first aspect, the present invention provides a cutting and positioning method for a semiconductor slicing machine, which adopts the following technical solution:

[0007] An initial image of the wafer to be located is obtained, and the initial image is converted into a grayscale image. The robustness score of each pixel in the grayscale image is calculated. The robustness score is used to characterize the degree of deviation of the grayscale value of the pixel from the overall grayscale distribution. Based on the grayscale image, a first anomaly index and a second anomaly index are calculated to characterize the degree of contamination of the grayscale image.

[0008] Based on the first and second anomaly indicators, an adaptive mask threshold for the grayscale image is calculated, wherein the larger the values ​​of the first and second anomaly indicators, the smaller the adaptive mask threshold; the robust score of each pixel is compared with the adaptive mask threshold to generate a mask for shielding contaminated areas.

[0009] The grayscale image and the preset template image are matched using a mask to determine the location of the positioning mark.

[0010] By quantifying the deviation of individual pixels using robust scores, and further calculating the first and second anomaly indices to comprehensively evaluate the overall contamination status of the image, an adaptive mask threshold negatively correlated with the degree of contamination is dynamically generated. This effectively filters out interference from various noises and contaminants, and can significantly improve the accuracy and robustness of template matching even under harsh conditions such as blurred or contaminated positioning marks, ultimately improving the positioning accuracy and yield of semiconductor cutting.

[0011] Preferably, before calculating the robustness score of each pixel in the grayscale image, the method further includes: calculating the robustness standard deviation of the pixels in the grayscale image, expressed as:

[0012]

[0013] In the formula, This represents the robust standard deviation of pixels in a grayscale image. The scaling factor is represented by med, the median function is represented by med, and I represents the gray value of a pixel in the grayscale image. This represents the median of grayscale values ​​in a grayscale image.

[0014] By using the median and median absolute deviation (med(|Iu|)) which are insensitive to outliers to calculate robust standard deviation, we can better resist the impact of extreme outliers in the image (such as strongly reflective noise or dark smudges) on the overall statistical characteristics. This makes the calculated robust score more realistically reflect the degree of deviation of the pixels, providing a more reliable and robust data foundation for the subsequent generation of high-quality adaptive masks.

[0015] The preferred expression for the robust score is:

[0016]

[0017] In the formula, This represents the robust score of pixel x in a grayscale image. This represents the grayscale value of pixel x in a grayscale image. This represents the median of grayscale values ​​in a grayscale image. This represents the robust standard deviation of pixels in a grayscale image. This represents hyperparameters.

[0018] Preferably, pixels with robustness scores greater than a preset robustness threshold are used as feature pixels, and connected component analysis is performed on the feature pixels to obtain feature regions.

[0019] Preferably, the expression for the first anomaly indicator is:

[0020] ;

[0021] ;

[0022] In the formula, q represents the first anomaly index of the grayscale image. This indicates the compactness of feature region i. Let represent the area of ​​feature region i, tanh represent the hyperbolic tangent function, M represent the number of feature regions in the grayscale image, and i represent the index of the feature region. This represents the perimeter of feature region i.

[0023] By weighting and summing the compactness and area of ​​the identified abnormal regions, it is possible to more accurately identify contaminants with near-circular (high compactness) characteristics (such as water droplets and bubbles), and to more accurately quantify the degree of image pollution caused by contaminants of specific shapes, thus providing a more targeted basis for adjusting the adaptive threshold.

[0024] Preferably, before calculating the second anomaly index, the method further includes: converting the initial image into an HSV color space image, obtaining the region corresponding to the feature region in the grayscale image, using pixels in the region whose V value is greater than or equal to a preset brightness threshold as marker pixels, mapping the marker pixels to the grayscale image, constructing a first minimum bounding circle for the feature region, and constructing a second minimum bounding circle for the marker pixels.

[0025] By analyzing the bright areas, the system can effectively identify and quantify the loss of detail caused by reflections, making up for the types of contamination that may not be fully covered by robust scores alone (for example, the gray values ​​of some reflective areas may be close to those of normal areas but the brightness is extremely high), making the system's comprehensive assessment of the degree of image contamination more comprehensive and accurate.

[0026] Preferably, the expression for the second anomaly indicator is:

[0027]

[0028] In the formula, This indicates the second anomaly index for grayscale images. This represents the ratio of the area of ​​the second smallest circumcircle to the area of ​​the first smallest circumcircle. Let represent the area of ​​feature region i, tanh represent the hyperbolic tangent function, M represent the number of feature regions in the grayscale image, and i represent the index of the feature region.

[0029] By calculating the area ratio of the circumcircle of the bright area to the circumcircle of the feature area, and combining this with a weighted average of the area, the severity and distribution range of bright reflective pollution are effectively quantified. This forms a comprehensive assessment of the pollutant's morphology and reflectivity, enabling the system to distinguish different types of pollution and adjust the masking strategy more precisely, thereby further improving the anti-interference capability of the positioning method.

[0030] Preferably, the expression for the adaptive mask threshold is:

[0031]

[0032] In the formula, This represents the adaptive mask threshold for a grayscale image. The reference mask threshold for a grayscale image. , Let represent the minimum and maximum values ​​of the grayscale image mask threshold, respectively, and q represent the first anomaly index of the grayscale image. This represents the second anomaly indicator of a grayscale image, and clip represents the clipping function.

[0033] By using the clip function to limit the threshold to a preset maximum and minimum value, the threshold overflow problem caused by extreme values ​​of abnormal indicators is avoided, thus ensuring the stability of the algorithm and the validity of the final result.

[0034] Preferably, the expression for the mask is:

[0035]

[0036] In the formula, The mask representing pixel x in a grayscale image. This represents the robust score of pixel x in a grayscale image. This represents the adaptive mask threshold for a grayscale image.

[0037] Based on the comparison between the robust scores of pixels and the adaptive mask threshold, a binarized mask is generated. By setting normal regions to 1 (preserved) and abnormal regions to 0 (masked), the subsequent matching algorithm is directly guided, ignoring pixel information in contaminated regions, thereby ensuring that only effective features are used for matching.

[0038] Secondly, the present invention provides a cutting and positioning system for a semiconductor slicing machine, which adopts the following technical solution:

[0039] A cutting and positioning system for a semiconductor slicing machine includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a cutting and positioning method for a semiconductor slicing machine as described above is implemented.

[0040] A computer program is generated using the above-mentioned semiconductor slicing machine with a cutting and positioning method and stored in a memory so that it can be loaded and executed by a processor. Thus, a system can be made based on the memory and processor for convenient use.

[0041] The present invention has the following technical effects:

[0042] By calculating the robustness score of pixels and anomaly indicators that characterize the morphology of contaminants and the intensity of reflection, an adaptive mask that matches the real-time contamination level of the image is dynamically generated. This mask can accurately identify and block various complex interferences such as water stains, scratches, and reflections, significantly improving the recognition accuracy and robustness of positioning marks under harsh working conditions, increasing the success rate of the final product cutting, and meeting the needs of wafer cutting. Attached Figure Description

[0043] Figure 1 This is a flowchart of a cutting and positioning method for a semiconductor slicing machine according to the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0045] This invention discloses a cutting and positioning method for a semiconductor slicing machine, referring to... Figure 1 This includes the following steps:

[0046] S1: Acquire initial images during the wafer fabrication process and perform preprocessing.

[0047] During wafer slicing in a semiconductor slicing machine, an initial image of the wafer is captured using an industrial camera before each slice to determine the positions of positioning marks and scribe lines. The initial image is in RGB format. It is then converted to grayscale and subjected to Gaussian filtering to obtain a grayscale image, eliminating noise. The Otsu thresholding method is used to segment the grayscale image to obtain the wafer region. The initial image is then converted to an HSV color space image, and the V value of each pixel is obtained. A one-to-one correspondence is established between pixels in the grayscale image and pixels in the HSV color space image.

[0048] S2: Calculate the robust score of pixels in a grayscale image.

[0049] For the acquired grayscale image, calculate the robust standard deviation of the pixels in the grayscale image, expressed as:

[0050]

[0051] In the formula, This represents the robust standard deviation of pixels in a grayscale image. This represents the scaling factor, with a value of 1.5; med represents the median function; and I represents the grayscale value of a pixel in the grayscale image. This represents the median of grayscale values ​​in a grayscale image.

[0052] This represents the median absolute deviation (MAD). According to the normal distribution theory, the relationship between the median absolute deviation (MAD) and the standard deviation (σ) is as follows: ,therefore, Robust standard deviation represents the degree of fluctuation in the grayscale values ​​of pixels in a grayscale image from the median. A larger robust standard deviation indicates greater fluctuation in grayscale values ​​from the median, more dispersed grayscale data, greater variation, and potentially more outliers. Conversely, a smaller robust standard deviation indicates more concentrated data, smaller deviation from the median, denser data distribution, less fluctuation, and greater consistency. It can be understood that one robust standard deviation corresponds to one grayscale image.

[0053] The robustness score for each pixel is calculated based on the robustness standard deviation of pixels in the grayscale image, expressed as follows:

[0054]

[0055] In the formula, This represents the robust score of pixel x in a grayscale image. This represents the grayscale value of pixel x in a grayscale image. This represents the median of grayscale values ​​in a grayscale image. This represents the robust standard deviation of pixels in a grayscale image. This represents a hyperparameter with a value of 0.1 to prevent the denominator from being 0.

[0056] A robust score indicates that the greater the deviation between the grayscale value of a pixel and the overall data, the closer it may be to an outlier or noise.

[0057] S3: Calculate the feature metrics of the grayscale image.

[0058] Pixels with robustness scores greater than a preset robustness threshold are selected as feature pixels. Connectivity analysis is performed on these feature pixels to obtain feature regions. The robustness threshold is set manually based on the actual situation. Feature indicators are calculated based on the feature regions, including a first anomaly indicator and a second anomaly indicator.

[0059] S31: Calculate the first anomaly index for grayscale images.

[0060] The compactness of the feature region is calculated using the following expression:

[0061]

[0062] In the formula, This indicates the compactness of feature region i. This represents the area of ​​feature region i. This represents the perimeter of feature region i. Here, the number of pixels within the feature region is used as the area of ​​the feature region, and the number of pixels at the edge of the feature region is used as the perimeter.

[0063] During wafer dicing, contaminants on the wafer surface, such as cooling water droplets, resin residue, and bubbles, appear as circular structures in grayscale images. Therefore, if the feature area... The larger the value, the closer the shape of the feature area is to a circle, and the greater the likelihood that it is a pollutant area.

[0064] The expression for the first abnormal indicator is:

[0065]

[0066] In the formula, q represents the first anomaly index of the grayscale image. This indicates the compactness of feature region i. denoted by , i represents the area of ​​feature region i, tanh represents the hyperbolic tangent function used for normalization, M represents the number of feature regions in the grayscale image, and i represents the index of the feature region.

[0067] Since contaminants on the wafer surface are generally circular, and the larger the area of ​​the feature region, the greater the impact on the template matching algorithm, the normalized result of the product of the compactness and area of ​​the feature region is used as the first anomaly index. The first anomaly index comprehensively reflects the possibility of contaminants in the wafer surface image. The larger the value, the greater the possibility of contaminants in the wafer.

[0068] S32: Calculate the second anomaly index for grayscale images.

[0069] In the HSV color space image, the region corresponding to the feature region in the grayscale image is obtained. Pixels in the region with a V value greater than or equal to a preset brightness threshold are used as marker pixels. The marker pixels are mapped to the grayscale image. Then, a first minimum circumcircle is drawn for the feature region so that the feature region is inside the first minimum circumcircle. A second minimum circumcircle is drawn for the marker pixels so that the marker pixels are inside the second minimum circumcircle.

[0070] The expression for the second anomaly indicator is:

[0071]

[0072] In the formula, This indicates the second anomaly index for grayscale images. This represents the ratio of the area of ​​the second smallest circumcircle to the area of ​​the first smallest circumcircle. Let represent the area of ​​feature region i, tanh represent the hyperbolic tangent function used for normalization, M represent the number of feature regions in the grayscale image, and i represent the index of the feature region. When there are no marker pixels in the HSV color space image, let The value is 0.

[0073] In HSV color space images, the V value represents brightness. The larger the value, the brighter the pixel area. The brightness threshold is set manually according to the actual situation. For example, the brightness threshold is 80%. When there are contaminants such as cooling water droplets, resin residue, and bubbles on the wafer surface, strong light reflection occurs in the contaminant area when the image is captured, which easily leads to the problem of highlighting in the HSV color space image, and further leads to the loss of details in the corresponding area.

[0074] S4: Calculate the adaptive mask threshold for the grayscale image.

[0075] The expression for the adaptive mask threshold is:

[0076]

[0077] In the formula, This represents the adaptive mask threshold for a grayscale image. The reference mask threshold for a grayscale image. , Let represent the minimum and maximum values ​​of the grayscale image mask threshold, respectively, and q represent the first anomaly index of the grayscale image. This represents the second anomaly indicator of a grayscale image; clip represents the clipping function used to... Limited to Within the range, if If the value exceeds the range of the minimum and maximum values, then it is clipped to the nearest boundary value. The value is less than the minimum value Then let The final value is ,Right now The value is greater than the maximum value. Then let The final value is .

[0078] When the first and second anomaly indices in the grayscale image are large, it indicates that there is a greater possibility of contaminant regions in the grayscale image. In this case, the smaller the adaptive mask threshold, the stricter the requirements for subsequent mask generation.

[0079] S5: Generate a mask based on the calculated adaptive threshold.

[0080] The expression for the mask is:

[0081]

[0082] In the formula, The mask representing pixel x in a grayscale image. This represents the robust score of pixel x in a grayscale image. This represents the adaptive mask threshold for a grayscale image.

[0083] When the robustness score of a pixel is less than or equal to the adaptive mask threshold, it indicates that the corresponding pixel region is a normal region. In this case, the mask of the corresponding pixel region is set to 1. When the robustness score of a pixel is greater than the adaptive mask threshold, it indicates that the corresponding pixel region is an abnormal region. In this case, the mask of the corresponding pixel region is set to 0, that is, the pixel is masked when the template matching algorithm is used to match the grayscale image and the template image in the future.

[0084] S6: Based on the calculated mask, the grayscale image and the template image are matched using a template matching algorithm to obtain the location of the positioning marker.

[0085] The matching score between the grayscale image and the template image is:

[0086]

[0087] In the formula, This represents the matching score of the grayscale image at the matching position (a, b). The mask representing pixel x in a grayscale image. This represents the grayscale value of pixel x at the matching position (a, b) in the grayscale image. This represents the weighted mean of the grayscale image under the masking effect. This represents the weighted average of the template image under the mask effect, and x represents the index of the grayscale image pixel.

[0088] The matching position of the grayscale image with the highest matching score is obtained. Based on the matching position, the precise position of the positioning mark is obtained, and the wafer is sliced ​​using the positioning mark.

[0089] This invention also discloses a cutting and positioning system for a semiconductor slicing machine, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a cutting and positioning method for a semiconductor slicing machine according to the present invention.

[0090] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0091] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A cutting and positioning method for a semiconductor slicing machine, characterized in that, Including the following steps: The robust standard deviation of pixels in a grayscale image is calculated using the following expression: In the formula, This represents the robust standard deviation of pixels in a grayscale image. The scaling factor is represented by med, the median function is represented by med, and I represents the gray value of a pixel in the grayscale image. This represents the median of grayscale values ​​in a grayscale image. Obtain an initial image of the wafer to be positioned, convert the initial image to a grayscale image, and calculate the robustness score of each pixel in the grayscale image. The expression for this score is: In the formula, This represents the robust score of pixel x in a grayscale image. This represents the grayscale value of pixel x in a grayscale image. The hyperparameters are represented by robust scores, which characterize the degree of deviation of pixel gray values ​​from the overall gray distribution. The first and second anomaly indices are calculated based on the gray image to characterize the degree of gray image contamination. Pixels with robustness scores greater than a preset robustness threshold are taken as feature pixels, and connected component analysis is performed on the feature pixels to obtain feature regions. The expression for the first abnormal indicator is: ； ； In the formula, q represents the first anomaly index of the grayscale image. This indicates the compactness of feature region i. Let represent the area of ​​feature region i, tanh represent the hyperbolic tangent function, M represent the number of feature regions in the grayscale image, and i represent the index of the feature region. Represents the perimeter of feature region i; The initial image is converted to an HSV color space image. The region corresponding to the feature region in the grayscale image is obtained. Pixels in the region with a V value greater than or equal to a preset brightness threshold are used as marker pixels. The marker pixels are mapped to the grayscale image. A first minimum bounding circle is drawn for the feature region, and a second minimum bounding circle is drawn for the marker pixels. The expression for the second anomaly indicator is: ; In the formula, This indicates the second anomaly index for grayscale images. This represents the ratio of the area of ​​the second smallest circumcircle to the area of ​​the first smallest circumcircle; Based on the first and second anomaly indicators, the adaptive mask threshold for the grayscale image is calculated, including: In the formula, This represents the adaptive mask threshold for a grayscale image. The reference mask threshold for a grayscale image. , represents the minimum and maximum values ​​of the grayscale image mask threshold, respectively, and clip represents the clipping function. The larger the values ​​of the first and second anomaly indices, the smaller the adaptive mask threshold. A mask for masking contaminated areas is generated by comparing the robust score of each pixel with the adaptive mask threshold, including: In the formula, The mask representing pixel x in a grayscale image. Represents the adaptive mask threshold for grayscale images; The grayscale image and the preset template image are matched using a mask to determine the location of the positioning mark.

2. A cutting and positioning system for a semiconductor slicing machine, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a cutting and positioning method for a semiconductor slicing machine according to claim 1.

Images