JP2010145145A - Device and method for inspecting circuit pattern, and test pattern - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem that, in a method for inspecting the flaw of a circuit pattern, operation for expanding and contracting the image data of a pattern becoming a standard is performed to form a master pattern having allowable maximum and minimum dimensions and the image data obtained from an article to be inspected is compared with the master pattern to discover a flaw but, in expansion and contraction image processing for forming the master pattern, processing for adding and subtracting peripheral pixels at every pixel is performed but it is difficult to perform expansion and contraction in a definite ratio by the ratio of line widths in the case of the standard pattern containing various line widths. <P>SOLUTION: The distance data from a background is imparted to the pixels belonging to the standard pattern and pixels are eliminated until the distance data larger than oneself are missing in the peripheral pixels to form a skeleton pattern representing the skeleton of the standard pattern. The pixels present within a range wherein the distance data having the pixels belonging to the skeleton pattern is multiplied by a predetermined magnification are set as the master pattern. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a circuit pattern inspection method for automatically inspecting printed circuit board circuit patterns and LSI circuit pattern defects by image processing.

  The demand for a circuit pattern in which a conductor pattern is formed on an insulating member is rapidly increasing with the spread of electronic devices such as mobile phones. Since the circuit pattern affects the performance and durability of an electronic device using the circuit pattern, the circuit pattern is not only electrically connected as designed, but also requires a predetermined dimensional accuracy.

  Therefore, all the circuit patterns are inspected after being manufactured, and it is confirmed that a predetermined specification is satisfied. By the way, since the types and number of circuit patterns have become enormous, it is necessary to complete the inspection in as short a time as possible.

  Conventionally, several methods have been proposed for this inspection. A method called a line width sub-pixel measurement method is a method in which the line width is measured for every fixed length, the line width between them is determined from several measurement points before and after, and the quality with respect to the reference width is judged. This method can be inspected with a resolution of about 1/10 of the pattern pitch.

  The image comparison method is a method of obtaining a difference image between the binarized imaging data and the non-defective product data and obtaining a difference portion having a certain area or more as a defect. Since this method binarizes data, the calculation speed can be increased.

  On the other hand, as a defect determination method, a conventionally proposed method is called a 1/3 rule. According to this rule, if the thickness of the lead portion, which is the conductor of the circuit pattern, is larger than 1/3 of the reference thickness, the lead is judged to have a protrusion, and the lead is thinner than 1/3 of the reference thickness. It is determined that there is a risk of disconnection.

In particular, Patent Document 1 discloses a method using an image comparison method, which is a standard image in which a reference pattern portion is 4/3 times and a standard image in which the size is 2/3 times. Are prepared, and each reference image is compared with an image of a circuit pattern to be inspected.
JP 63-19541 A

  An image comparison method for comparing a standard image with an image to be inspected is useful because it can easily find a defect in a circuit pattern. The accuracy of defect extraction by this method depends on the method of creating a standard image enlarged or reduced with respect to the reference pattern.

  Patent Document 1 discloses a method for creating an enlarged or reduced pattern of a reference pattern by taking a logical sum with adjacent pixels as an enlargement / reduction method. Since this method increases or decreases the same number of pixels for all pixels, a standard image can be created without any problem when the reference pattern is composed of patterns having substantially the same width. However, the following problem occurs in a pattern in which patterns having greatly different line widths are mixed.

That is, when the image is reduced to fit the thick line width pattern, the thin line pattern disappears, and when the image is enlarged to fit the thin line width pattern, the thick line pattern cannot be enlarged by a predetermined amount. In other words, there is a problem that it is difficult to create a standard image enlarged or reduced according to the line width.

  The pattern inspection method and apparatus of the present invention have been conceived to solve the above problems. That is, in the present invention, a skeleton portion of a pattern is extracted from the reference pattern, and a master pattern is created by adding pixels by a distance that is a predetermined multiple from the skeleton. Then, the master pattern and the image of the inspection object are compared to detect a defect.

  In the present invention, since a pixel located within a predetermined scaled distance from the skeleton part of the reference pattern is used as a master pattern, each part is expanded and enlarged at the same magnification even in a pattern in which a thick part and a thin part are mixed. A contracted master pattern can be obtained. Accordingly, it is possible to perform defect inspection of a product in which combinations of patterns having more various line widths are drawn on one substrate.

  Embodiments of the present invention will be described below, but the present invention is not limited to these embodiments, and can be modified and changed within the scope of the gist of the present invention.

(Embodiment 1)
FIG. 1 shows a configuration diagram of a pattern inspection apparatus 1 of the present invention. The pattern inspection apparatus of the present invention includes an image capturing unit 20 that captures a circuit pattern of an object to be inspected 90, a binarization unit 30 that binarizes the captured image, and a reference image (this is referred to as a “master pattern”). MP ”) and an alignment unit 50 that aligns the positions of the images to be inspected, an arithmetic unit 60 that performs a logical operation on the pixels of the aligned images, and a determination unit 70 that determines the operation results. Moreover, the memory | storage part 80 which memorize | stores a master pattern and the master pattern preparation part 10 which produces a master pattern are also included.

  The image capturing unit 20 includes at least a camera 21 and an illumination 22. The camera 21 converts the pattern image of the inspection object 90 into image data Vd and outputs it. Therefore, a device using a photoelectric conversion element such as a CCD is preferable. The CCD may be one in which pixels are arranged two-dimensionally or one-dimensionally arranged like a line sensor. Moreover, in order to image | photograph the to-be-inspected object 90 from upper direction, what mounted the telecentric type lens is more preferable. In addition to using the wavelength of normal visible light, light in the near infrared or ultraviolet wavelength band may be detected.

  The image data Vd is a set of pixel information converted into an electrical signal by the photoelectric conversion element. Specifically, a set of values (gradation values) obtained by converting light received by one pixel of the CCD into 8-bit or 10-bit gradation and pixel positions (coordinates, etc.) is collected for one image. is there. As an example, if the number of effective pixels of a CCD is 5 million pixels and converted to a gradation value of 10 bits per pixel, the image data Vd for one image is only 50 million bits only with the gradation value. (50 Mbit).

  The illumination 22 is used to obtain contrast by shining light when photographing the inspection object 90. There is no particular limitation, and ring-shaped fluorescent lamps, LEDs, optical fiber light guides, and the like can be suitably used.

  In addition to illuminating the inspection object 90 from above, the transmitted light may be imaged by the camera 21 illuminating from below the inspection object 90.

After the binarization unit 30, image processing is performed on the image data Vd. Each component can also create dedicated hardware. However, considering that it is easy to change and the current computer processing speed is sufficiently fast, software processing by the computer is the main.

  Therefore, the binarization unit 30, the alignment unit 50, the calculation unit 60, the determination unit 70, and the master pattern creation unit 10 will be described as elements of the control device 85 in order to represent that they are executed by a computer and software. These elements can be executed by software and computers.

  The storage unit 80 is illustrated as being included in the control device 85. Specifically, a semiconductor memory is preferably used. However, a secondary storage medium such as a hard disk installed outside the control device 85 may be included.

  Here, the pattern will be described. The pattern is image data obtained from a product in which a circuit or a part of the circuit is formed on a substrate such as polyimide. A product includes a portion where a pattern is formed by a conductor or the like, and a portion where a pattern is not formed. Image data of a portion where the pattern is formed is called a product pattern pd, and image data of a portion where the pattern is not formed is called a background pattern bd. That is, the image data Vd photographed by the image photographing unit 20 in FIG. 1 is a pattern, and the pattern includes a product pattern pd and a background pattern bd. The pattern has pixels, and each pixel has a gradation value and coordinates. The pixel of the product pattern is Epd, and the pixel of the background pattern is Ebd.

  The reference pattern Vsd is image data obtained from products that have passed various inspections and CAD data at the time of design (collectively referred to as “reference products”).

  The reference pattern Vsd further has information such as coordinates (x, y), pattern discrimination information (Ev), pixel value (I), distance from the boundary (d), and skeleton information (B) for each pixel. is doing. Moreover, you may have master pattern information (mp) showing whether it becomes the pixel which comprises a master pattern.

  The coordinates are information representing the position of the pixel in the image data. The pixel value (I) is a gradation value indicating the luminance of the pixel. The pattern distinction information (Ev) is information indicating whether the pixel is a product pattern pixel or a background pattern pixel. The boundary between the product pattern and the background pattern is also simply referred to as “boundary”.

  The pixel value is a gradation value. The distance (d) from the boundary is a pixel belonging to the product pattern pd, and indicates the distance from the nearest background pattern bd by the number of pixels. The distance (d) may be called distance data (d). The skeleton information (B) is information that indicates whether or not the pixel belongs to the product pattern pd and has become a skeleton as a result of the skeletonization process described later. The master pattern information (mp) is information indicating whether or not a pixel of the master pattern is obtained as a result of an inverse distance conversion process described later.

  It is sufficient that the reference pattern initially has coordinates, pattern distinction information, and pixel values. This is because the other parameters are produced by the master pattern production unit.

  These may be displayed side by side, and specifically (x, y, Ev, I, d, B, mp) may be described for a certain pixel.

In FIG. 2, the processing flow of the master pattern preparation part 10 is shown. When the reference pattern Vsd is input, distance conversion processing (S201), pixel deletion processing (S202), thinning processing (S20)
3) The reference pattern is processed by the synthesis process (S204), the pruning process (S205), and the reverse distance conversion process (S206), and the master pattern is output as a result (S208). Pixel deletion processing, thinning processing, composition processing, and pruning processing are collectively referred to as skeletonization processing. A master pattern obtained by contracting the reference pattern is called a contracted master pattern, and a master pattern obtained by expanding the master pattern is called an expanded master pattern. Each process will be described below. In the flow of FIG. 2, the process of returning to the main routine (S210) at the end is added and shown as a subroutine for the process of the main routine, but may be an independent process.

  The distance conversion process (S201 in FIG. 2) will be described in more detail with reference to FIG. FIG. 3A shows that there is a product pattern pd and a background pattern bd in one image data. The distance conversion process is a process of giving the distance from the boundary between the background pattern bd of the input reference pattern Vsd and the product pattern pd as the pixel distance data of the product pattern pd.

  First, for all the pixels Epd of the product pattern pd, the adjacent pixels are examined, and distance data “1” is given to the pixels adjacent to the background pattern. In FIG. 3B, the pixel of the distance data “1” is indicated by a black square. Next, the distance data “2” is assigned to the pixels adjacent to the pixel of the distance data “1” for all the pixels of the product pattern pd. In FIG. 3B, pixels to which the distance data “2” is assigned are represented by black circles. As described above, the distance data (d) is obtained by giving the distance from the background pattern by the number of pixels to the product pattern portion of the reference pattern.

  FIG. 3C shows a state in which distance data (d) is assigned to all pixels belonging to the product pattern. This is called a reference distance pattern. The pixel of the distance data “3” is indicated by a black triangle, and the distance data “4” is indicated by a cross. In the reference distance pattern, distance data is given to all the pixels Epd belonging to the product pattern.

  FIG. 4 shows a flow of this distance conversion process. When the reference data is input, the process starts (S220). An end determination is made (S222), and if it is an end (Y branch of S222), the process is returned to the main (step S202 in FIG. 2) (S234). Here, the end determination is made based on whether distance data has been assigned to all the pixels of the product pattern. Since the distance data is represented by the number of pixels from the background pattern, it is preferable to perform in advance a process of giving data of distance zero to all the pixels belonging to the reference pattern in advance. By doing so, the distance data of the background pattern can be all zero.

  Next, it is determined whether or not the following processing has been completed for the pixels Epd belonging to all product patterns (S224). If the process has already been completed, the argument k representing the distance is incremented (S226), and the process returns to step S222 to determine the end. The argument k is given “1” as an initial value immediately after the distance conversion process is started.

  If the following processing has not been completed for pixels belonging to all product patterns (N branch in S224), the following processing is performed. First, attention is paid to an area of 9 pixels centered on a certain pixel. FIG. 4B shows the region. This is hereinafter referred to as a reference area. The reference area is also an area that defines a “periphery” for a certain pixel in the pattern. The middle pixel is denoted by A, and the surrounding pixels (eight) are denoted by Ap1 to Ap8. If the distance data of any one of the peripheral pixels Ap is k−1 (Y branch of S228), the distance data k is given, assuming that the pixel A is a pixel to which the distance is given (S230). Here, the distance data of the central pixel is represented as A (d), and the distance data of the surrounding pixels is Ap (d).

For example, when searching for a pixel of distance data “1” (when k = 1), a pixel belonging to the product pattern is defined as the center of the reference region (pixel A), and any pixel around the pixel A If there is a pixel belonging to the background pattern of the distance “0”, the distance data “1” is given to the pixel A. If there is no pixel whose distance data is zero in the peripheral pixel Ap (N branch of S228), the process skips step S230, moves to the pixel Epd belonging to the next product pattern (S232), and returns to step S224. To move to the next pixel means to move the center pixel of the reference area to the next pixel. By repeating the above processing, distance data can be given to the pixels Epd belonging to all product patterns.

  Next, the skeletonization process will be described. The skeletonization processing is processing for performing pixel deletion processing on the image data subjected to distance conversion processing, combining the result of thinning processing separately, and pruning it to obtain a reference skeletonization pattern.

  This will be described in more detail with reference to FIG. FIG. 5A shows the reference distance pattern of FIG. 3C again. In the pixel deletion process, a reference distance pattern shown in FIG. 5A in which distance data is assigned to the pixel Epd belonging to the product pattern is input. The pixel deletion process includes a process of deleting pixels until there are no pixels having distance data larger than the distance data of the pixel belonging to the product pattern. The pixel deletion process is a process that does not actually delete a pixel but extracts a skeleton pixel from the pixel Epd belonging to the product pattern. Therefore, the pixel deletion process is a process in which a value of 1 is given to the skeleton information (B) in advance for all the pixels Epd, and this skeleton information is set to zero. The result obtained by the pixel deletion processing is referred to as a pixel deletion processing image.

  FIG. 5B shows the result of deleting pixels until the pixel A in the middle of the reference area has no distance data of peripheral pixels in the vertical and horizontal diagonal directions larger than the distance data of the pixel A. That is, FIG. 5B shows the distance data of the pixel Epd belonging to the product pattern whose skeleton information (B) is 1, with symbols such as black squares and black circles. For example, the pixel 111 is a pixel whose distance data is 2 and whose skeleton information is 1. As described above, there is a case where one connected reference skeleton pattern cannot be obtained in the pixel deletion process.

  Next, a thinning-processed image is created by deleting the surrounding pixels until the line thickness becomes 1 by thinning processing in parallel with the pixel deletion processing. The thinning process is performed based on the gradation value. Therefore, the reference pattern Vsd is input to the thinning process. FIG. 5C shows a result of generating an image by thinning processing. The result obtained by the thinning process is called a thinning process image.

  Next, a logical sum operation is performed on the result of the pixel deletion processing image and the thinning processing image. This is a synthesis process. FIG. 5D shows the result. By this arithmetic processing, it is possible to obtain a skeletonized pattern in which the portion where the skeleton line segment is divided in the pixel deletion processing is stored by the thinning processing. That is, the skeleton information (B) of the remaining pixels in the thinned image is set to 1, and the skeleton information (B) is logically summed with the pixel deletion processed image.

  Then, a pruning process is performed on the pattern information obtained as a result of the synthesizing process to delete a portion that becomes a branch part. Here, the branch portion is a portion having a tip and continuous at an angle of 45 ° on the image data. In FIG. 5D, portions 113 to 117 correspond to branches. The branch portion is generated by processing the corner of the pattern as the starting point of the line segment by the thinning process.

The branch portion may generate an unfavorable portion in the master pattern used for defect inspection by an inverse distance conversion process performed later, and needs to be deleted. Further, the pruning process may be performed on a pixel whose skeleton information (B) is 1. FIG. 5E shows a reference skeleton pattern obtained as a result of performing the pruning process and deleting the branches. It can be seen that one continuous skeleton pattern is obtained by the pixel 112 as compared to the pixel deletion processing image of FIG.

  Next, each process is illustrated using a flow. FIG. 6 shows a flow of pixel deletion processing. When the reference distance pattern is input and the pixel deletion process is started (S250), the end of the pixel deletion process is determined (S252). The end determination is the presence or absence of a pixel to be deleted. Therefore, it is sufficient to determine the pixels belonging to the product pattern. When there are no more pixels to be deleted, the process proceeds to the next process (S204) (S264). Otherwise, it is determined whether or not the inspection has been completed for all the pixels belonging to the product pattern whose distance data is m (S254). If completed, the parameter m indicating the distance data to be deleted is incremented. m is set to 1 in the initial setting.

If the processing has not been completed for all the pixels whose distance data belonging to the product pattern is m (N branch in S254), the distance data of the middle pixel in the reference area is compared with the distance data of the surrounding pixels (S258). ). With reference to FIG. 4B, the peripheral pixels are adjacent pixels (Ap1, Ap2, Ap3, Ap4, Ap5, Ap6, Ap7, Ap8) in the eight directions around the middle pixel A. In FIG. 6, the distance data of the i-th peripheral pixel is represented as “Ap ⁱ (d)”.

  If any of the distance data of the peripheral pixels is larger than the distance data of the middle pixel in the reference area (Y branch in S258), the skeleton information (B) of the pixel A is set to zero (S260). The skeleton information (B) being zero means that the pixel A currently in the middle of the reference region cannot be a skeleton pattern. Then, the process proceeds to the next pixel (S262) and returns to step S254.

  By repeating this process, the skeleton information (B) becomes zero from the pixels close to the background pattern, and finally the information shown in FIG. This process corresponds to a pixel deletion process.

  Next, the thinning process (S203) will be described. The thinning process is a process of changing the brightness value so that the bright part becomes the dark part until the thickness of the bright part (bright part) or the dark part (dark part) becomes 1 pixel based on the brightness value of the image. As a result, an image consisting of a light line having a thickness of 1 or a dark line is obtained. Several methods have been proposed for this. Here, one typical method will be described. Note that the background pattern bd is a dark pixel (a pixel with a low gradation value), and the product pattern pd is assumed to be composed of a bright pixel (a pixel with a high gradation value).

  In the thinning process, eight types of reference areas (called “thinning elements”) are applied to the image data, and it is determined whether or not each pixel can be deleted. That is, if the same area as the following thinning element is present in the image data, the luminance value of the center pixel is replaced with a dark part.

  FIG. 7 shows the thinning element. Let each be EL1 to EL8. For example, in EL1, when Ap1, Ap2, and Ap3 of the reference region are dark portions (Z), and Ap6, AP7, and Ap8 are bright portions (W), if the central pixel (A) is a bright portion, Indicates when it can be changed. Here, “X” may be a dark part or a bright part. Here, when it is changed to a dark part, it seems that the pixel is deleted regarding the luminance value.

  EL1 changes the pixel in contact with the boundary with the dark part to the dark part by one pixel when the pixels are arranged horizontally and continuously. By this processing, a portion that is long in the horizontal direction of the screen is deleted from the upper direction of the screen. However, since there are three reference areas in the horizontal direction, when the image data becomes 2 pixels or less in width, the pixels are not deleted from above.

EL2 is a process of changing the central pixel (A) corresponding to the corner to a dark part while leaving the oblique portions of Ap4 and Ap7. EL3 to EL8 are different directions of EL1 and EL2. Accordingly, the image data is deleted from the upper, right, lower, and left directions by EL1, EL3, EL5, and EL7. However, EL1 and EL5 are not applied when the number of pixels in the horizontal direction of the pixels becomes 2 or less. EL3 and EL7 are not applied when the number of vertical pixels becomes 2 or less. Further, the pixel connection in the oblique direction remains by EL2, EL4, EL6, and EL8.

  By continuing the above processing until the luminance value is not replaced, it is possible to execute a thinning process in which the thickness of the image data is 1. FIG. 5C shows the result of this processing.

  8 and 9 show the flow of this process. When the thinning process (S203) starts (S400), variables are initialized (S402). Two variables, “flag” and “EL”, are used. “Flag” is a flag for determining whether or not gradation values are to be replaced. If this value is zero, the processing is terminated on the assumption that gradation values have not been replaced for all pixels. The initial value is zero. “EL” represents the thinning element described in FIG. For example, EL = 1 represents EL1. The initial value is “1”.

  Two types of image data “in” and “out” are prepared so as to correct and update the image data. These image data may be the gradation value and coordinates of the reference pattern Vsd. In initialization, in and out may be the same.

  Next, it is determined whether EL = 9 (S404). This is to know whether or not eight types of thinning elements have been applied to the image data. When there are still thinning elements to be applied (N branch in S404), gradation value replacement processing C is performed (S406). Details of step S406 will be described with reference to FIG.

  Then, EL is incremented (S408), and the process returns to the determination of whether EL is 9 (S404). By performing this process, eight thinning elements are sequentially applied to all the pixels, and the gradation value is replaced by the process C. If EL = 9 in step S404, it is determined that all eight thinning elements have been applied, and an end determination is made (S410).

  The end determination is made by determining whether or not flag is zero. If completed (Y branch of S410), a thinned image is created (S412), and the process returns to the main routine composition processing (S204) described in FIG. 2 (S414). If not (N branch of S410), the process returns to variable initialization (S402). Note that the thinned image is image data obtained by extracting the same coordinates as the pixels of the bright portion obtained by the thinning process from the reference distance pattern. That is, a pixel having the same coordinates as the bright pixel obtained by the thinning process is extracted from the reference distance pattern, and the skeleton information (B) is set to 1. By this process, the image data of FIG. 5C can be obtained.

  Next, the process C will be described with reference to FIG. In process C, the positional relationship of the same pixel as one of the eight thinning elements currently set is found and applied. Therefore, first, the center of the reference area is set at the head of the image data (S420). Then, the pixels are sequentially examined until a pixel relationship with the thinning element currently designated in each pixel of the reference pixels is found. Then, it is determined whether or not the gradation value of the central pixel of the reference pixel is larger than the maximum value (MAX (Z)) of the dark portion and smaller than the minimum value (MIN (W)) of the bright portion ( S422). That is, an area that matches the thinning element currently applied is found, and it is determined whether or not the central pixel of the reference area is a bright part.

If it is a thinning element and the central pixel of the reference area is a bright part (W), the pixel is converted into the maximum value of the gradation value of the dark part. That is, it is set as a dark part. The flag is set to 1. This is because pixel gradation values are replaced.

  It is determined whether or not the pixel is the final pixel (S428). If it is not the final pixel, the process proceeds to the next pixel (S432), and returns to the determination of the pixel value (S422). When the process reaches the last pixel, the content of the image data in is replaced with the image data out. This is because the replacement of the gradation value is performed on the image data out. Then, the process returns to process C (S406). As a result, FIG. 5C can be obtained as a thinning process result based on the gradation value.

  Next, the synthesis process will be described. For the result of pixel deletion in FIG. 5B and the thinned image in FIG. Accordingly, in FIG. 5 (b) and FIG. 5 (c), only one having the skeleton information of 1 is extracted, and the distance (d) is indicated by distance data marks such as black circles and black squares in FIG. 5 (d). is there.

  Next, a pruning process is performed on the image data resulting from the synthesis process. Reference is made to the flow of FIG. First, it is determined whether or not the pruning process is finished (S272). The termination determination here is the presence or absence of a pixel to be pruned (deleted) as a branch. More specifically, when the parameter indicating the distance data is “l” (el) and the maximum value of the distance data among the pixels of the reference pattern is lmax, the skeleton information (B) is not zero and the distance is It is determined whether or not there is a pixel that can be deleted with respect to a pixel Epd whose data has l = 1 to l = lmax-1.

  The end determination of the pruning process is also the end determination of the skeletonization process. When there are no more pixels to delete, the skeletonization process ends. Otherwise, it is determined whether the processing has been completed for a predetermined pixel among the pixels belonging to the product pattern (S274). More specifically, it is determined whether or not the processing is completed for a pixel whose skeleton information (B) is 1 and whose distance data (d) is 1 among the pixels Epd. If completed, the parameter l indicating the distance to be deleted is incremented. The parameter l is given 1 as an initial value.

  If the inspection for the predetermined pixel Epd as described above has not been completed (N branch of S274), first, the skeleton of the middle pixel in the reference area is set to the peripheral pixels Ap1 to Ap8 other than the middle pixel in the reference area. The number of pixels whose information (B) is 1 is confirmed, and the distance data of the middle pixel is compared with the distance data of Ap1, Ap3, Ap6, and Ap8 only when there is one (S278). Ap1, Ap3, Ap6, and Ap8 are pixels having an oblique relationship with respect to the middle pixel A in the reference region with reference to FIG. 4B.

It should be noted that the number of pixels having the skeleton information (B) of 1 in the peripheral pixels Ap1 to Ap8 other than the pixel in the middle of the reference region is “SAp ⁱ (B)”. In addition, Ap1, Ap3, Ap6, the distance data of Ap8 as "Ap ^o (d)", was Ap2, Ap4, Ap5, Ap7 skeletal information of the (B) and "Ap ^e (B)".

  Here, if any of the distance data of Ap1, Ap3, Ap6, and Ap8 is larger than the distance data of the pixel A in the middle of the reference region, and the skeleton information (B) of Ap2, Ap4, Ap5, and Ap7 is zero (S278) ), The skeleton information (B) of the middle pixel is set to zero (S280). The current distance data is 1 (el), there is only one skeleton information (B) of 1 in the peripheral pixels, the pixel exists only in an oblique direction with respect to the pixel A, and the distance data is This is because when the pixel is larger than the pixel A, it can be determined that the pixel is a branch. Then, the process proceeds to the next pixel and returns to step S274 (S282).

By repeating this process, all the branches in the pattern in which the result of the pixel deletion process and the result of the thinning process are combined are deleted, and finally the pattern shown in FIG. The process in FIG. 10 corresponds to a pruning process. In this way, a pattern indicating the skeleton of the product pattern and having no branches is referred to as a reference skeleton pattern td. A pixel belonging to the reference skeleton pattern and having skeleton information (B) of 1 is represented by a skeleton pixel Tpd. For example, the skeleton pixel Tpd is a pixel indicated by a black circle, a black triangle, or a cross in FIG. In this embodiment, pixel deletion processing and pruning processing are performed on the reference distance pattern to obtain the reference skeleton pattern, but the method for obtaining the reference skeleton pattern is not limited to this.

  Next, the reverse distance conversion process (S206 in FIG. 2) will be described. In FIG. 11A, FIG. 5E is shown again. In the reverse distance conversion process, the reference skeleton pattern shown in FIG. The reverse distance conversion process is a process of generating a master pattern by adding pixels corresponding to a distance obtained by multiplying the distance data of the skeleton pixel Tpd by a predetermined magnification to the periphery of the skeleton pixel. For example, if the skeleton pixel has the distance data d and the contraction master pattern is obtained by multiplying the reference pattern by 0.8, all the pixels at a distance of 0.8d from the skeleton pixel are all stored in the contraction master pattern. Let it be a pixel of the pattern.

  With reference to FIG. 11B, the reverse distance conversion process will be described in more detail. A pixel 120 in FIG. 11A is one of the skeleton pixels Tpd. The distance data d of this pixel 120 was 4. If the magnification is 0.8, 4 × 0.8 = 3 pixels from the pixel 120 is the range of the contraction master pattern. In addition, since the pixel must be an integer, the decimal part is rounded off.

  Therefore, a pixel in a range 124 of 5 × 5 = 25 pixels in total of 3 pixels vertically and horizontally with the pixel 120 as the center is a master pattern. Similarly, since the distance data of the pixel 121 is 3 and the distance data of the pixel 122 is 2, the pixels included in the contraction master pattern are 9 pixels (125) and 9 pixels centering on each pixel. (126).

This process is performed on the pixels Tpd belonging to all the reference skeleton patterns, and the master pattern information mp of the pixels that are included in the master pattern is set to 1. A pixel with mp = 1 is referred to as a master pattern pixel. In addition, the master pattern information mp is set to zero for pixels that are not included. Further, the pixel value may be changed without using the master pattern information. For example, the pixels of the master pattern imparted to Imax is the maximum value of the pixel values I, the other pixels imparting minimum value I ₀ of the pixel values I. In this way, a master pattern can be obtained as a binary image.

  FIG. 11C shows the result of performing this process on the pixels belonging to all the reference skeleton patterns to obtain the contraction master pattern 127. In the portion 128 where the pixel 120 is present, one pixel from the background pattern is not included in the master pattern, but in the narrow portion 129 where the pixel 122 is present, the entire product pattern remains. As described above, in the master pattern manufacturing according to the present invention, the same number of pixels is not deleted or expanded uniformly for all parts of the product pattern. In other words, whether the image is contracted or expanded depends on the positional relationship in which the pixel exists in the product pattern.

  FIG. 12 shows a flow of reverse distance conversion processing. A reference skeleton pattern is input to the inverse distance conversion process. First, the end determination (S292) is performed in the same manner as the processing so far. Here, the end determination is whether or not the reverse distance conversion has been performed on the pixels Tpd belonging to all the reference skeleton patterns. If reverse distance conversion is completed (Y branch of S292), it will move to post-processing (S300).

Otherwise (N branch of S292), the result obtained by multiplying the distance data d of the pixel Tpd (also referred to as the pixel A because it is in the center of the reference region) by the magnification α is obtained as z (S294). The master pattern information mp of the pixel (Apz) included in the distance z is set to 1 (
S296). Note that mp is given an initial value of zero.

  Further, a process of putting a predetermined value into the pixel value I of Apz may be added (S296). The predetermined value is, for example, a value of Imax, which is the maximum pixel value. Here, peripheral pixels in the range of the distance z centered on the pixel A are represented by Apz, the master pattern information is Apz (mp), and the pixel value is represented by Apz (I).

The process moves to a pixel belonging to the next reference skeletonization pattern (S298), and repeats from the end determination (S292). When the processing is completed for all the reference skeleton pattern pixels Tpd (Y branch of S292), the pixel value of the pixel Epd belonging to the product pattern, the value of which is not converted to Imax, is the same value as the pixel of the background pattern. (For example, I ₀ ). This processing is necessary when acquiring the master pattern as binary image data. There is no need to create a master pattern as binary image data. Then, the process ends (S302).

  By the above operation, a master pattern in which any part of the reference pattern is uniformly multiplied by α can be obtained. Here, if α is set to a value larger than 1.0, a master pattern in which the reference pattern is expanded can be obtained, and if it is set to a value smaller than 1.0, a master pattern in which the reference pattern is contracted is obtained. Obtainable. Further, when it is desired to obtain a plurality of master patterns, steps S290 to S302 may be performed a predetermined number of times while changing the multiple α.

  As described above, the master pattern creation unit 10 creates the master pattern MP based on the input reference pattern. The binarized expansion / contraction master patterns are an expansion master pattern Bcdx and a contraction master pattern Bcds. That is, Bcdx and Bcds are one form of the master pattern MP. These master patterns are stored in the storage unit 80.

  Next, a pattern inspection method using the master pattern of the present invention will be described with reference to the flowcharts of FIGS. It is assumed that the expansion master pattern and the contraction master pattern are recorded in advance in the storage unit 80 based on the above description. When the inspection is started (S1000), an end determination is made (S1002). Here, the end determination is the presence or absence of an inspection object to be inspected. Of course, even if there is an object to be inspected, it may be terminated by interruption processing. If completed (Y branch of S1002), the inspection is terminated (S1016).

  When the inspection object is inserted, the image capturing unit 20 captures the inspection object and acquires image data Vd (S1004). The image data Vd is converted into binary image data Bcd by the binarization unit 30 (S1006). This is for easy comparison with the master pattern. Note that before the image data is binarized, interpolation processing such as enlargement or reduction to a predetermined size may be performed.

  The binary image data Bcd is aligned with the master pattern MP sent from the storage unit 80 by the alignment unit 50 (S1008). This is for comparing two images. The two images that have been aligned are calculated for each corresponding pixel by the calculation unit 60, and the magnitude comparison with the master pattern is performed (S1010).

  More specifically, the pixel value of the corresponding pixel belonging to the expansion master pattern Bcdx is subtracted from the image data of the pixel belonging to the binary image data Bcd of the object to be inspected, and the number Cdx when positive and the contraction master pattern The binary image data Bcd of the inspection object is subtracted from Bcds to obtain the number Cds when it becomes positive.

When the pixel value is larger than the pixel of the expansion master pattern among the pixels in the image data of the inspection object, it means that the inspection object has a portion that is thicker than a predetermined reference. That is, there is a risk of a protrusion defect. Further, the fact that the contraction master pattern has a pixel value larger than the binary image data of the inspection object means that the inspection object has a very thin portion. That is, there is a risk of chip defects.

  The determination unit 70 compares the number Cdx of pixels protruding from the expansion master pattern obtained by the calculation unit 60 and the number Cds of pixels thinner than the contraction master pattern with predetermined threshold values Thx and Ths, and determines whether or not it is acceptable. Is determined (S1012). Here, the threshold Thx is called an expansion threshold, and the threshold Ths is called a contraction threshold.

  That is, if Cdx is larger than Thx, an unacceptably thick portion exists in the pattern of the inspection object. Therefore, it is determined as a protrusion defect. Similarly, if Cds is larger than Ths, it means that an unacceptably thin portion exists in the pattern of the inspection object. Therefore, it is determined as a chip defect.

  Further, the determination unit outputs these results Con (S1014). The output result Con may be only a determination that a defect with respect to the inspection object is acceptable or not, or may directly indicate the number of Cdx and Cds that are differences from the master pattern. The output result Con may be displayed on a display device or the like, or may be recorded as it is or transmitted to another device. Reference numeral 88 in FIG. 1 indicates an output destination of the result Con.

  A more specific example of pattern inspection will be described with reference to FIGS. Referring to FIG. 14, as already described, the reference skeleton pattern td is generated from the reference pattern Vsd. The contraction master pattern Bcds is generated by the master pattern preparation unit 10 from the reference skeleton pattern td. Assume that image data Vcd1 is obtained from the inspection object. It is assumed that this inspection object has one disconnection 131 in the product pattern. In Vsd, the region of the product pattern pd, the region of the skeleton information B = 1 in the reference skeleton pattern td, and the region of the master pattern information mp = 1 and the pixel value Imax in Bcds are represented in white.

When the calculation unit 60 subtracts the pixel value of the corresponding image data Vcd1 from the pixel value of the pixel belonging to the contraction master pattern Bcds, the image data 130 is obtained. If the image data 130 includes a pixel having a pixel value equal to or greater than I ₀ , it can be determined that the defect is a chip defect. This is because there is a thinner part than the contraction master pattern. In FIG. 14, the region 132 corresponds.

Reference is now made to FIG. The master pattern is the expansion master pattern Bcdx. Here, it is assumed that the image data Vd2 is obtained from the inspection object. It is assumed that this inspection object has a portion 133 protruding in the product pattern. By subtracting the pixel value of the corresponding image data Vd2 from the pixel value of the pixel belonging to the expansion master pattern, the image data 134 is obtained. If there is a pixel having a pixel value equal to or greater than I ₀ in the image data 134, it can be determined that it is a protrusion defect. This is because there is a thicker part than the expansion master pattern. In FIG. 15, this is a region 135. In FIG. 15, the white area is shown in FIG. However, Bcds is replaced with Bcdx.

  As described above, by using the master pattern MP of the present invention, the pattern inspection of the inspection object can be performed. In particular, it is useful to use the master pattern MP according to the present invention when the product pattern includes a portion having a large thickness difference ratio.

(Embodiment 2)
In Embodiment 1, among the pixels of the product pattern, the master pattern information mp is set to 1 and the pixel value maximum value Imax is given to the pixel to be the master pattern MP. The minimum value I ₀ of the pixel value was given to the pixel (see the flow in FIG. 12). This means that the master pattern is obtained as binarized image data.

  Therefore, the difference between the master pattern and the image data Vd of the inspection object is obtained as the number of pixels (Cdx or Cds), and the degree of defect cannot be determined quantitatively. For example, even if the portion Cdx that protrudes from the expansion master pattern is less than the expansion threshold Thx and the determination by the determination unit is acceptable, the portion that is almost the same as the reference pattern or thicker than the reference pattern Therefore, it is impossible to make a judgment as to whether it was accepted because it did not slightly exceed the expansion threshold value Thx.

  Therefore, in the present embodiment, an embodiment in which the difference from the reference pattern can be evaluated more quantitatively will be described.

  In the present embodiment, when creating a master pattern, a pixel value corresponding to a distance from the reference skeleton pattern is given to pixels belonging to the master pattern MP.

  This will be described in more detail with reference to FIG. While changing the magnification α with respect to the reference skeleton pattern, reverse distance conversion processing is performed to produce master patterns 150, 151, 152, and 153. For example, in the master pattern 150, α is zero, 151 is 0.5, 152 is 1.0, 153 is 1.3, and the like. Clearly, the master pattern 150 is a reference skeleton pattern, the master pattern 151 is a contraction master pattern, the master pattern 152 is a reference pattern, and the master pattern 153 is an expansion master pattern. Then, a pixel value corresponding to the distance from the reference skeleton pattern is given to the pixels belonging to each master pattern.

  For example, Imax for pixels belonging to the master pattern 150, 0.8 * Imax for the master pattern 151, 0.5 * Imax for the master pattern 152, 0.3 * Imax for the master pattern 152, and the like. Then, the pixels whose master pattern information mp of these pixel data is 1 are overwritten in order from the one with the largest magnification. A pixel whose master pattern information mp is 1 is called a master pattern pixel. Here, overwriting means replacing a pixel that already has a pixel value with a new pixel value. In FIG. 14, overwriting is performed in the order of reference numerals 155, 156, 157, and 158.

  As a result, the synthesized master pattern 160 includes a background pattern 161, a region 162 having a pixel value of 0.3 * Imax, a region 163 having a pixel value of 0.5 * Imax, and a region 164 having a pixel value of 0.8 * Imax. , A region 165 having a pixel value Imax. This master pattern 160 is called a multi-master pattern.

  This multi-master pattern can acquire not only the approval / disapproval performed by the determination unit, but also information on how much it differs from the reference pattern.

  A more specific description will be given with reference to FIG. The image data 130 was image data of the inspected object having the cut 131 (from FIG. 14). Further, the image data 134 is the image data of the inspected object with the protrusion 135 (from FIG. 15).

When these image data and the multi-master pattern 160 are compared, the cut portion corresponds to the region 171 and the protruding portion corresponds to the region 173. Then, by checking the pixel value and the number of pixels belonging to this region, it can be shown how much the reference pattern is different. Specifically, for the areas 171 and 173 obtained as differences from the master pattern, the total number of pixels and the number of pixels for each pixel value are displayed. These are the determination processing (
S1012) may be included.

  FIG. 18 shows a flow for producing a multi-master pattern. The description will be made assuming that the master pattern manufacturing unit 10 performs processing. The production of the multi-master pattern may be the same process up to the skeletonization process S204 in the process of the master pattern production unit 10 shown in FIG. In order to produce a multi-master pattern, the reverse distance conversion process may be changed to the flow from step S310 shown in FIG.

  Assuming that the flow of FIG. 18 is performed, the number of master patterns to be overwritten, the magnification α of each master pattern, and the pixel value I for each magnification α are determined. Here, it is assumed that the pixel value I decreases as the magnification α increases, and this is expressed as I (α), and the description will be continued. First, when the reverse distance conversion process S310 is started (S310), initial setting is performed (S322). In the initial setting, the magnification α is set to αmax which is the maximum value of α, and the multi-master pattern MMP is set to the reference pattern Vsd. Note that the multi-master pattern is overwritten sequentially, so it does not matter.

  Next, end determination is performed (S324). The end determination here is whether or not a predetermined number of master patterns have been overwritten. The determination may be made based on whether or not the magnification α is the minimum value. When a predetermined number of master patterns are overwritten (Y branch of S324), the process is returned to the main (S338).

  If not (N branch of S324), it is determined whether or not the processing has been performed on the pixels Tpd of all the reference skeleton patterns (S326). If not performed (N branch of S326), the result obtained by multiplying the distance data d of the pixel Tpd (referred to as the pixel A because it is in the center of the reference region) by the magnification α is obtained as z (S328). The master pattern information mp of the pixel (Apz) included in the distance z as the center is set to 1. Note that mp is given an initial value of zero.

  Further, a pixel value I (α) determined in advance according to the magnification is substituted for the pixel value of the pixel Apz. Then, the process proceeds to the next pixel (S332), and step S326 is repeated. By repeating this process, the master pattern MP (α) having the magnification α is completed.

  When the above processing is completed for all the reference skeleton pattern pixels Tpd, the multi-master pattern MMP is overwritten with the multi-pattern MP (α). Overwriting is a process that leaves the master pattern pixel, and in particular leaves the master pattern pixel to be written later.

  More specifically, the following rules are followed when overwriting the pixel y of MP (α) corresponding to the pixel x of MMP. That is, the values of the master pattern information mp of the pixel x and the pixel y are compared, and if both are zero, the pixel value of either the pixel x or the pixel y is adopted. When the mp of the pixel x is zero and the mp of the pixel y is 1, the pixel value of the pixel y is adopted. If mp of pixel x is 1 and mp of pixel y is zero, nothing is done. If both the MP of the pixel x and the mp of the pixel y are 1, the pixel y is adopted.

  Then, the magnification α is set to the next magnification, and the process returns to step S324. By performing the above processing flow, it is possible to produce a multi-master pattern overwritten in sequence from a master pattern having a large magnification α. The comparison between the multi-master pattern and the image data of the inspection object is performed by the calculation unit 10 in FIG. 1 and can be included in the result Con by the determination unit.

(Embodiment 3)
In the inspection method of the present invention, a master pattern corresponding to the pattern width of the inspection object is produced. Therefore, it would be convenient if there was a method for confirming whether the inspection method of the present invention is functioning correctly. Here, a test pattern for verification of the inspection method of the present invention will be described.

  FIG. 19 shows a test pattern for the inspection method of the present invention. This test pattern is configured such that a thick portion 180 and a thin portion 182 are included in the same field of view of the camera 20. Then, defects having the same size are formed in the thick part and the thin part. In FIG. 19, the protrusion 184 and the chip 186 at the thick part 180 and the chip 188 and the protrusion 190 at the thin part 182 correspond to this. The ratio of the thick portion 180 and the thin portion 182 is preferably 100: 20 to 100: 50. This is because the inspection method of the present invention is suitable for inspection of patterns having a difference in thickness within this range within the same visual field.

  The thick portion 180 has defects 184 and 186 for this line width 192. In this case, since the line width is sufficiently thicker than the size of the defect, these defects are acceptable defects. However, the defects 188 and 190 in the narrow portion 182 are unacceptable defects because they are sufficiently large with respect to the line width 193 even if the defects have the same size. The size of these defects can vary depending on how large the line width can be tolerated. It is possible to verify whether or not the inspection method of the present invention is correctly performed by inputting such a test pattern and accepting or detecting these defects.

  FIG. 20 shows a reference pattern for this test pattern. There is no defect in this pattern. Based on this reference pattern, an expansion and contraction master pattern is produced.

  FIG. 21 shows an output example of an unacceptable protrusion portion 194 obtained from the test pattern and the reference pattern, and FIG. 22 shows an output example of an unacceptable defect portion 196. These are obtained by comparing the expansion and contraction master pattern produced from the reference pattern of FIG. 20 with the test pattern. That is, the defect 188 and the protrusion 190 in FIG. 19 are detected by the expansion and contraction master pattern. That is, if a test pattern is input to the inspection apparatus of the present invention and the results shown in FIGS. 21 and 22 can be obtained, it can be verified that the inspection apparatus is operating normally.

  23 and 24 show an expansion master pattern and a reduction master pattern, respectively. These are produced with a 30% increase / decrease relative to the thick and thin portions of the reference pattern. Since these master patterns are expanded / reduced at a predetermined ratio to the line width of the pattern, the difference in thickness from the reference pattern is not constant. However, the ratio is constant.

  The defect formed in the test pattern may be a defect having a boundary size that can be detected depending on the location. For example, assuming that the defect detection threshold for the entire pattern is s% for each pattern location, at least one of a defect that does not exceed s% and a defect that exceeds s% is formed in a thick part and a thin part, respectively. Good. By doing in this way, it can be confirmed immediately whether the inspection as intended is performed.

  Specifically, if the thick part is 80 pixels and the thin part is 20 pixels, and the threshold for inspection of protrusions and dents is 30%, the thick part has 24 pixels as the detection boundary. Therefore, an artificial defect having a size equal to or smaller than 23 pixels and an artificial defect having a size equal to or larger than 24 pixels are formed in the thick portion. If the threshold is 30%, a defect of 7 pixels or less is allowed for a thin portion. If these defects are detected as acceptable and unacceptable defects, respectively, it can be determined that the defect inspection method of the present invention is correctly performed.

  FIG. 25 shows a test pattern in which a protrusion 200 and a chip 202 having 24 pixels or more are added to a thick part 180 and a protrusion 206 and a chip 204 having 7 pixels or less are added to a thin part 182. As described above, by including defects larger and smaller than the threshold in the test pattern, it is possible to more accurately verify the validity of the operation of the inspection method and inspection apparatus of the present invention.

  The present invention can be suitably used for automatic defect inspection of circuit patterns.

It is a figure which shows the structure of the pattern inspection apparatus of this invention. It is a figure which shows the flow of a master pattern preparation part. It is a figure explaining the process which provides distance data to a reference | standard pattern. It is a figure which shows the flow of a distance conversion process. It is a figure explaining the process which produces a reference | standard frame | skeleton pattern. It is a figure explaining the flow of an ancestor deletion process among skeletonization processes. It is a figure which shows the thinning element used for a thinning process. It is a figure which shows the flow of a thinning process. It is a figure which shows the flow of the process C among the flows of a thinning process. It is a figure which shows the flow of a pruning process. It is a figure explaining a reverse distance conversion process. It is a figure which shows the flow of a reverse distance conversion process. It is a figure which shows the flow of an inspection method. It is a figure explaining the process which detects a chip | tip defect by a shrinkage | contraction master pattern. It is a figure explaining the process which detects a protrusion defect by the expansion master pattern. It is a figure explaining the structure of a multi master pattern. It is a figure explaining the process which detects a chip | tip and a processus | protrusion using a multi master pattern. It is a figure which shows the flow of the reverse distance conversion process at the time of producing a multi master pattern. It is a figure which illustrates the test pattern for inspection. It is a figure which illustrates the reference image for test patterns. It is a figure which shows the example which detected the processus | protrusion which is unacceptable. It is a figure which shows the example which detected the processus | protrusion which can be accepted. It is a figure which illustrates the master pattern which increased line width by 30%. It is a figure which illustrates the master pattern which reduced line width 30%. It is a figure which illustrates the test pattern for a test | inspection which added the chip | tip and protrusion corresponding to line | wire width.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Pattern test | inspection apparatus 10 Master pattern preparation part 20 Image photographing part 30 Binarization part 50 Positioning part 60 Calculation part 70 Judgment part 80 Storage part 85 Control apparatus

Claims (11)

A pattern inspection method for detecting defects by comparing a master pattern and image data of an inspection object,
A first step of obtaining a reference distance pattern by subjecting a reference pattern having a background pattern and a product pattern to a distance conversion process;
A second step of skeletonizing the reference distance pattern to obtain a reference skeleton pattern;
A third step of obtaining a master pattern by performing a reverse distance conversion process at a distance by which the reference skeleton pattern is multiplied by a predetermined coefficient;
A pattern inspection method comprising a fourth step of detecting a difference by comparing the master pattern and an inspection object pattern. The first step includes
2. The pattern inspection method according to claim 1, which is a process of assigning distance data from a boundary between the background pattern and the product pattern to pixels belonging to the product pattern belonging to the reference pattern. The second step includes
A pixel deletion process that sequentially deletes the pixels until there is no longer any surrounding pixel in which the distance data of the pixels around the pixel is larger than the distance data of the pixel, and the reference distance pattern The thinning process for obtaining the thinned image by deleting the pixels until the line width is minimized, the synthesizing process for synthesizing the pixel deleted image and the thinned image, and the pixels belonging to the product pattern The pattern inspection method according to claim 1, further comprising a pruning process for deleting branches from pixels remaining as a result of the synthesis process. The method according to any one of claims 1 to 3, wherein the third step is a step of obtaining the master pattern both at least when the predetermined coefficient is less than 1.0 and when the predetermined coefficient is greater than 1.0. The pattern inspection method described. 5. The third step of claim 1, wherein all pixels in a region determined by a distance obtained by multiplying the distance data of a skeleton pixel belonging to the reference skeleton pattern by the predetermined coefficient are master pattern pixels belonging to the master pattern. A pattern inspection method according to any one of the claims. The said 3rd process gives the pixel value according to the said predetermined coefficient to the pixel value of the master pattern pixel which belongs to the said master pattern produced with the said predetermined coefficient, The claim of any one of Claim 1 thru | or 5 Pattern inspection method. The pattern inspection method according to claim 6, wherein the third step further includes a process of overwriting master pattern pixels given pixel values according to the coefficient to obtain a multi-master pattern. The pattern inspection method according to claim 7, further comprising a fifth step of further comparing the difference detected in the fourth step with the multi-master pattern. A master pattern creation unit that receives a reference pattern having a background pattern and a product pattern and outputs a master pattern;
A storage unit for recording the master pattern;
An imaging unit for imaging the object to be inspected and obtaining image data;
A binarization unit for binarizing the image data;
An arithmetic unit that compares the master pattern and the binarized image data and outputs a difference;
The pattern inspection apparatus which has a determination part which outputs a determination result based on the output of the said calculating part. The master pattern production unit
Provide a distance data from the boundary between the background pattern and the product pattern to the pixels belonging to the product pattern to create a reference distance pattern,
Create a reference skeleton pattern without branches from the reference distance pattern,
10. The pattern inspection apparatus according to claim 9, wherein a master pattern is created by using, as a master pattern pixel, a pixel between a distance multiplied by a predetermined coefficient to the distance data of a pixel belonging to the reference skeleton pattern. Having different patterns in the same field of view,
A test pattern in which a defect or protrusion having a predetermined size is formed in a portion having a different thickness.