Image processing method

Abstract

Provided is an image processing method for detecting defects existing in an image by excluding the influence of wiring when image inspection is performed on a semiconductor chip or the like having a wiring pattern. Specifically, the image processing method detects an unshaped image by performing image processing on a captured image on which a straight line extending in one direction is displayed, and comprises the steps of: obtaining a first processed image by performing smoothing filtering processing on the captured image by using a section, the section is defined by a first number a of pixels arranged in one direction and a second number b of pixels arranged in another direction different from the one direction; a second step of smoothing and filtering the captured image using a section defined by a third number c of pixels arranged in one direction and a fourth number d of pixels arranged in the other direction to obtain a second processed image; obtaining a third processed image obtained by subtracting the first processed image from the second processed image; and a detection step of detecting an amorphous image using the third processed image, a being larger than any one of b, c, d.

Description

Technical Field

[0001] The present invention relates to an image processing method. Background Art

[0002] Images are used to detect defects in the manufacturing process of various products.

[0003] For example, Patent Document 1 discloses a highly sensitive defect inspection method for inspecting patterns to detect defects in the memory mat and peripheral circuits of semiconductor wafers. Semiconductor manufacturing involves various processes, each of which produces various defects. Among these processes, cleaning is frequently performed, and defects caused by residual cleaning fluid often occur in many cleaning processes. Therefore, inspection is necessary to eliminate defective products or to provide feedback on process manufacturing conditions.

[0004] Since semiconductor chips contain wiring patterns, defect detection requires distinguishing between the wiring patterns and defects. To distinguish between the wiring patterns and defects, an image of only the wiring patterns is prepared and compared with the image to be detected for defects, and image processing is performed to detect defects.

[0005] Conventionally, multiple images of defect-free samples (qualified products) containing only wiring patterns are prepared and superimposed through image processing to create statistical images of qualified products. These statistical images of qualified products are then compared with inspection images and processed for defect detection. While it has been considered to use design drawings (CAD drawings) as patterns for only wiring, in real-world products, variations in wiring width and position occur due to component misalignment and alignment errors in manufacturing equipment. Therefore, using design drawings as patterns for only wiring can result in products being judged as defective even though they have no defects due to variations in the wiring pattern. For this reason, inspections have traditionally used statistical images of qualified products.

[0006] Prior art literature

[0007] Patent Literature

[0008] Patent Document 1: Japanese Patent No. 3625236 Summary of the Invention

[0009] Problems to be solved by the invention

[0010] However, if statistically qualified product images are used as defect-free images, the width of the wiring increases due to manufacturing variations, which reduces the brightness of the wiring. If the wiring being inspected is narrow (high in brightness), surrounding areas may be considered defects. Furthermore, a single defect that spans a wiring may be detected as separate defects on both sides of the wiring, making it impossible to determine the actual number and size of defects.

[0011] The present invention has been completed in view of the above-mentioned problems, and its purpose is to provide an image processing method that eliminates the influence of wiring patterns when inspecting semi-finished products / products such as semiconductor chips with wiring patterns using images to detect defect images existing in the image.

[0012] Means for solving problems

[0013] An image processing method according to the present invention performs image processing on a captured image to detect an irregular shape image displayed on the captured image, the captured image displaying at least one straight line extending in one direction. The method includes the following steps: obtaining a first processed image for the captured image by filtering the captured image using a smoothing filter within a segment defined by a first predetermined number a of pixels arranged in the one direction and a second predetermined number b of pixels arranged in a direction different from the one direction; obtaining a second processed image for the captured image by filtering the captured image using a smoothing filter within a segment defined by a third predetermined number c of pixels arranged in the one direction and a fourth predetermined number d of pixels arranged in the other direction; subtracting the first processed image from the second processed image to obtain a third processed image as a difference image; and detecting the irregular shape image using the third processed image, wherein the number a is greater than any of the numbers b, c, and d. The first predetermined number a may be any number less than the total number of pixels arranged in the one direction and greater than b, c, or d. The second predetermined number b, the third predetermined number c, and the fourth predetermined number d are preferably 1 / 5 or less, and more preferably 1 / 10 or less, of the total number of pixels arranged in one direction. Preferably, binarization is performed in the detection step.

[0014] Preferably, in the detecting step, closing processing is performed after binarization processing to combine the plurality of indefinite shape images divided by the straight lines.

[0015] Preferably, in the detecting step, the marking process is performed after the closing process.

[0016] Effects of the Invention

[0017] By performing image processing on one inspection image (captured image) using different smoothing filters as described above to generate two processed images and generating a difference image between the two processed images, defects can be detected by a simple method without the influence of the wiring pattern. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a diagram showing a captured image for inspection according to the embodiment.

[0019] Figure 2 This figure shows a first processed image which is a result of performing the averaging filter process 1 on the captured image for inspection.

[0020] Figure 3 This figure shows a second processed image which is a result of performing the averaging filter process 2 on the captured image for inspection.

[0021] Figure 4 This figure shows an image obtained by binarizing and closing a third processed image, which is a difference image obtained by subtracting the first processed image from the second processed image.

[0022] Figure 5 This is an example of a flowchart of an implementation method. DETAILED DESCRIPTION

[0023] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. The following description of the preferred embodiments is merely illustrative in nature and is not intended to limit the present invention, its applications, or uses.

[0024] (Implementation 1)

[0025] Embodiment 1 relates to image processing performed on captured images used for inspection of wafers or individual chips in the mid-stage or finished stage of manufacturing semiconductor devices (semiconductor chips).

[0026] The captured image of the wafer after a predetermined process in the manufacture of a semiconductor device is shown in FIG. Figure 1 On the wafer, there are five wiring patterns: wiring patterns 10, 12, 14, 16, and 18. There are also defects 20, 22, 24, 26, and 28, where some of the cleaning solution was not completely removed. Furthermore, the captured image shows tiny dots 30 and 31, which are caused by noise during imaging. Wiring patterns 10, 12, 14, 16, and 18 extend horizontally in the captured image.

[0027] Defects 20, 22, 24, 26, and 28 are caused by residual cleaning liquid, so their shapes are irregular, and the edge portions 20a, 22a, 24a, 26a, and 28a are more raised than the central portions 20b, 22b, 24b, 26b, and 28b.

[0028] It is necessary to remove the wiring pattern from such captured images and detect only defects, thereby determining whether the product is acceptable or making corrections and improvements to the manufacturing process. When using the detection results for acceptance or process correction, image processing is required to appropriately utilize the detection results for determination, or to develop a more refined determination method. The following describes image processing and determination methods.

[0029] exist Figure 1 In the image, defects and wiring are represented by clear lines, but in actual captured images, the boundaries are often blurred or there are changes in brightness and darkness, making it difficult to directly determine. Figure 5 The image processing is performed according to the flow shown.

[0030] exist Figure 5 In the illustrated process, the captured image data is first subjected to a lookup table conversion. In this conversion, a table (allocation table) is used to assign output values ​​to input data (here, brightness values). This enhances contrast. Alternatively, other conversions (filtering) that enhance contrast can be performed instead of a lookup table conversion.

[0031] The data converted using the lookup table undergoes smoothing filter processing 1 and smoothing filter processing 2, respectively. Smoothing filter processing is a low-pass filter that reduces and smoothes the variations in brightness values ​​between adjacent pixels. Specifically, image processing is performed on blocks of multiple adjacent pixels called segments (for example, square partitions of 4 pixels vertically and 4 pixels horizontally). Smoothing filter processing is performed to reduce noise or emphasize features such as edges. Averaging filter processing uses various filters, but the appropriate filter can be selected based on the purpose of the determination.

[0032] In the averaging filter process 1, the number of pixels arranged along the direction in which the wiring pattern extends (horizontally in the figure) is set to a first specified number a, and the number of pixels arranged along the direction orthogonal to the direction in which the wiring pattern extends (vertically in the figure) is set to a second specified number b, and a smoothing filter process is performed on the a×b segment. This segment is a horizontally long rectangular segment with a>b. When the number of pixels arranged in the horizontal direction of the captured image is X, a can be selected from X, X / 2, X / 4, X / 8, etc., but can also be any other number. There is no special limitation on b as long as it is smaller than a, and it is preferably less than 1 / 4 of a, and more preferably less than 1 / 10 of a. In addition, b can be greater than 1 and less than 30. Alternatively, b is preferably a number of pixels equivalent to less than 1 / 2 of the width of the image of the wiring pattern.

[0033] The larger the number of pixels in a segment, the longer the averaging filter process takes. Therefore, the number of pixels in a segment can be selected based on the purpose. Furthermore, when performing averaging filtering on one segment and then transferring to an adjacent segment and performing averaging filtering on that segment, for example, if the shift to the adjacent segment is made by a / m pixels (m is an arbitrary integer) in the horizontal direction, the time required to process the entire image increases as m increases. Therefore, in this case, the amount of shift can be selected based on the purpose.

[0034] The first processed image obtained by performing the averaging filter process 1 is shown in FIG. Figure 2 Averaging filter processing 1 eliminates fine noise-caused dots 30 and 31, as well as defects 20, 22, 24, 26, and 28, leaving only wiring patterns 10, 12, 14, 16, and 18. Specifically, since the segments are horizontally elongated rectangles, wiring patterns 10, 12, 14, 16, and 18 extending horizontally from end to end remain, while the rest of the image disappears. In particular, when b is a number of pixels equal to or less than half the width of the wiring pattern image, the images of wiring patterns 10, 12, 14, 16, and 18 are output as images of approximately the same width as the original images.

[0035] Next, we will explain averaging filter processing 2. In averaging filter processing 2, the number of pixels arranged along the extension direction of the wiring pattern (horizontally in the figure) is set to a third predetermined number c, and the number of pixels arranged in a direction perpendicular to the extension direction of the wiring pattern (vertically in the figure) is set to a fourth predetermined number d. Smoothing filter processing is performed on a segment of c×d. This segment is a>c and a>d. There are no particular restrictions on c and d as long as they are less than a. They are preferably 1 / 4 or less of a, and more preferably 1 / 10 or less of a. Furthermore, c and d can be greater than 1 and less than 30. d is preferably the same as b.

[0036] As with averaging filter processing 1, averaging filter processing 2 takes longer as the number of pixels in a segment increases. Therefore, the number of pixels in a segment can be selected based on the purpose. Furthermore, when averaging filter processing is performed on a segment and then moved to an adjacent segment and averaging filter processing is performed on that segment, for example, if the adjacent segment is shifted horizontally by c / n pixels (n is an arbitrary integer), increasing n increases the time required to process the entire image. Therefore, in this case, the amount of segment movement can be selected based on the purpose.

[0037] Figure 3 2 shows a second processed image obtained by performing averaging filter processing 2. Averaging filter processing 2 eliminates fine dots 30 and 31 caused by noise, and wiring patterns 10, 12, 14, 16, and 18 and defects 20, 22, 24, 26, and 28 remain.

[0038] Next, a difference process (image subtraction) is performed by subtracting the first processed image from the second processed image, resulting in a difference image, or third processed image. In this third processed image, wiring patterns 10, 12, 14, 16, and 18 have disappeared, leaving only defects 20, 22, 24, 26, and 28. Therefore, there is no risk of false defects. This third processed image can also be used directly in the detection step for irregular defects.

[0039] However, in the third processed image, the portion where defects 20, 22, 24, 26, and 28 overlapped with the wiring pattern disappears, resulting in an image of defects 20, 22, 24, 26, and 28 separated by the wiring pattern. Furthermore, the edge portions 20a, 22a, 24a, 26a, and 28a of defects 20, 22, 24, 26, and 28 are displayed with a different brightness than the center portions 20b, 22b, 24b, 26b, and 28b. Even this third processed image can be used without problems depending on the inspection purpose. However, if the precise number, size (area), and other factors of defects are desired, further image processing can be performed during the inspection step using the third processed image before defect inspection.

[0040] As further image processing, such as Figure 5As shown, first, let's look at binarization. Binarization is the process of converting the display of each pixel to white or black using a specified threshold value for the luminance value of each pixel. For example, this process can make the defect white and the rest black. When the third processed image is binarized, there is no luminance difference between the edges 20a, 22a, 24a, 26a, 28a of the defects 20, 22, 24, 26, and 28 and the center 20b, 22b, 24b, 26b, and 28b. Only the difference between the defects 20, 22, 24, 26, and 28 and the rest of the image is displayed.

[0041] Then, a closing process can also be performed. Closing processing is a process performed on a binary image. For example, it is a process of filling in the gaps between adjacent defects (white parts) after they are divided by the wiring pattern and combining the divided parts. In the case where a defect is divided into multiple parts by the wiring pattern in the third processed image, this process returns it to the original defect. In addition, other image processing that can obtain the same result can also be used. The image obtained by the closing process after the binary image is shown in FIG. Figure 4 It is believed that this image almost accurately represents only the detected defect image.

[0042] After the closing process, image processing corresponding to the purpose of defect detection / processing can also be applied. For example, Figure 5 The labeling process shown in Figure 1 is a process that searches for white pixels in a binarized image. If other white pixels are adjacent to a white pixel, both white pixels are considered to be the same object and assigned the same label number. Label numbers can be used to identify defects separated by labeling as distinct defects, allowing the number and location of defects to be determined.

[0043] Size filtering can also be performed. Size filtering refers to the process of picking up only images of a specified size (prescribed length range). Figure 5 While three size filtering processes are shown, examples of other processes include picking out blocks of pixels (defects) whose horizontal lengths fall within a predetermined range, picking out blocks of pixels (defects) whose vertical lengths fall within a predetermined range, and picking out blocks of pixels (defects) whose areas fall within a predetermined range. Thus, for example, defects larger than a predetermined size need to be treated as defective, but defects smaller than that can be determined to be non-problematic as they will disappear in subsequent processes.

[0044] The image processing of this embodiment allows for simple methods to avoid / restore suspected defects that may arise from the presence of a wiring pattern, as well as image fragmentation and distortion caused by the wiring pattern, when performing defect detection / determination on a captured image displaying at least one straight line extending in one direction. Furthermore, since image processing is performed by applying two different averaging filters to a single captured image, image processing can be completed in a short time. By applying conventional image processing methods to obtain a difference image from the two different averaging filters, accurate defect detection / determination can be performed.

[0045] In this embodiment, image processing is performed on a captured image showing horizontally extending wiring. However, the same method can be used to process a captured image showing vertically extending wiring. In this case, the segment used in the averaging filter process 1 can be a vertically elongated rectangle.

[0046] (Other embodiments)

[0047] The above-mentioned embodiments are illustrative of the present invention, and the present invention is not limited to these examples. Known technologies, conventional technologies, and publicly known technologies may be combined in these examples, or a portion thereof may be replaced. In addition, modifications that are easily conceivable to those skilled in the art are also included in the present invention.

[0048] The present invention is intended to process images that display at least one straight line extending in one direction, and is not limited to images of semiconductor devices. For example, images of wiring boards are also intended to process images. Furthermore, the present invention is not limited to images displaying only straight lines extending in one direction; images displaying multiple straight lines extending in two or more directions may also be used.

[0049] Description of Reference Numerals

[0050] 10, 12, 14, 16, 18 wiring patterns (straight line)

[0051] Defects 20, 22, 24, 26, and 28

Claims

1. An image processing method, comprising: performing image processing on a photographic image to detect an amorphous image displayed on the photographic image, wherein the photographic image displays at least one straight line extending in one direction, wherein: The image processing method comprises the following steps: Obtaining a first processed image by filtering the captured image using a smoothing filter using a segment defined by a first predetermined number a of pixels arranged in the one direction and a second predetermined number b of pixels arranged in another direction different from the one direction; For the captured image, obtaining a second processed image by filtering using a smoothing filter using a segment defined by a third predetermined number c of pixels arranged along the one direction and a fourth predetermined number d of pixels arranged along the other direction; subtracting the first processed image from the second processed image to obtain a third processed image as a difference image; as well as a detection step of detecting the amorphous image using the third processed image, The number a is greater than any one of the number b, the number c, and the number d.

2. The image processing method according to claim 1, wherein: In the detection step, a binarization process is performed.

3. The image processing method according to claim 2, wherein: In the detecting step, a closing process is performed after the binarization process to combine the plurality of the irregular shape images divided by the straight lines.

4. The image processing method according to claim 3, wherein: In the detecting step, the marking process is performed after the closing process.

Images