JP4734625B2 - Master pattern creation device - Google Patents

Description

  The present invention relates to a master pattern creation device, and more particularly to a master pattern creation device that creates a master pattern used in a detection device such as a pattern defect detection device that detects a defect in a pattern such as a substrate using image processing.

  A pattern or the like is formed on a substrate or the like on which an electric element such as an IC is mounted in order to electrically connect between terminals of each electric element and other devices. Many substrates having such a pattern are manufactured, but the same substrate may not be manufactured depending on manufacturing conditions and environmental conditions. For example, when a pattern having a predetermined width and a predetermined length is manufactured, a pattern having a defect in which a part of the pattern is missing or a part of the pattern swells may be manufactured.

  For this reason, conventionally, a master pattern as a reference is prepared in advance, and the master pattern is compared with the manufactured pattern to determine whether or not they match. For this comparison, a matching method is known in which image data obtained by capturing a master pattern and image data obtained by capturing a production pattern are compared using image processing. In this matching method, each pixel when the master pattern is imaged and each pixel when the production pattern is imaged are compared for each pixel, and it is determined whether the patterns match each other. To do. In addition, a plurality of patterns in one production pattern can be compared using the master pattern as a template. By comparing in this way, a defect in the production pattern can be detected. As such a technique, a pattern defect detection apparatus is known (see Patent Document 1).

However, since the creation of a master pattern used to detect a defect becomes a reference for detection, a master pattern with high detection accuracy must be prepared in advance. For this reason, a master pattern is created and used from drawings such as design drawings and production drawings.
Japanese Patent Laid-Open No. 10-83452

  However, as described above, when a master pattern is created from a drawing, a manufacturing error must be taken into consideration, and a complicated operation is required. Further, since the master pattern image data must be stored in advance for all the pixels, a huge amount of processing is required to create the master pattern.

  An object of the present invention is to obtain a master pattern creation device that can quickly create a master pattern used in a detection device such as a pattern defect detection device in consideration of the above facts.

In order to achieve the above object, a master pattern creating apparatus according to a first aspect of the present invention is directed to an imaging unit that divides and images a plurality of manufactured objects into a plurality of vertical and horizontal pixels, and is captured by the imaging unit. Pixel value setting means for setting a predetermined numerical value as a pixel value of the pixel for each pixel of each of the plurality of captured images, and the captured image in which the pixel value of each pixel is set Generating means for generating a run-length code in which a pixel value of one pixel is sequentially read and a length in which the same pixel value continues in a predetermined direction is represented by the number of pixels in which the same pixel value continues. for each of a plurality of images captured by, on the basis of run-length code generated by said generating means, determining characteristics of the sum of the same pixel value for each of the predetermined direction of the line as an X-correlation In each of the lines having a large degree of coincidence of the obtained X correlation, each position of the run length code having a large degree of coincidence of the Y correlation obtained by obtaining the characteristic of the sum of the same pixel values for each line in the predetermined direction as the Y correlation. And a correlation calculation means for obtaining a correlation between the position and length of the run-length code between a plurality of captured images picked up by the image pickup means, and a calculated correlation is a predetermined value. Master setting means for setting the above run length code as master pattern data.

  According to a second aspect of the present invention, in the master pattern creating apparatus according to the first aspect, the pixel value setting means classifies the pixel density as either white when the predetermined density is less than a predetermined density or black when the density is higher than the predetermined density. By performing binarization, a pixel value representing white data or a pixel value representing black data is set for each pixel.

In the master pattern creation device, the correlation calculation means obtains a characteristic of the sum of the same pixel values for each line in the predetermined direction as an X correlation, and for the lines having a large degree of coincidence of the obtained X correlations, Ru can be found a correlation between the position and length of the run length code between objects.

In the master pattern generation device, the correlation calculation means obtains the characteristic of the sum of the same pixel values for each line in the predetermined direction as a Y correlation for the line having a high degree of coincidence of the X correlation, and obtains the calculated Y correlation. Ru can be determined each position and length correlation of the degree of coincidence of the large run length code.

  In the present invention, the object is imaged by being divided into a plurality of vertical and horizontal pixels by the imaging means. In this captured image, the pixel value is set for each pixel by the pixel value setting means. The pixel density of the captured image exists within a continuous density range from the lowest density such as white to the highest density such as black. Therefore, a continuous density range from the lowest density to the highest density such as black is classified into a plurality, and a predetermined numerical value is associated with each of the classified density ranges. Accordingly, for each pixel, a predetermined numerical value based on the density of the pixel is set as the pixel value of the pixel. Thus, the captured image is represented by multivalued data based on pixel values for each pixel.

  The pixel value setting means, as described in claim 2, provides white data for each pixel by binarization that classifies the pixel density as either white when a predetermined density is less than a predetermined density or black when the density is higher than the predetermined density. Or a binarizing means for setting a pixel value representing black data. By this binarization means, pixel data can be binarized into white data or black data. The object picked up by the image pickup means is multi-valued, and the data is sequentially read by the pixel by the generation means, and the length of the same pixel value in a predetermined direction is set to the number of pixels in which the same pixel value is continuous. Is generated.

  A plurality of objects are imaged by the imaging means, and a run length code is generated for each object by the generating means. For example, a binarized object image is generated as a run-length code represented by the number of pixels in which white data or black data continues in a predetermined direction. Therefore, the length in the predetermined direction of the image of the object imaged by the imaging means is represented by the number of pixels in which the same pixel value continues in the predetermined direction. Based on the run length codes of each of the plurality of objects, the correlation calculating means obtains the correlation between the positions and lengths of the run length codes between the plurality of objects. That is, a correlation is obtained between run length codes of a plurality of objects.

  The above correlation can be expressed by the degree of coincidence with the position and length of the run-length code, and the position and length of the run-length code match, that is, the larger the overlapping area when superimposed as an image, The value becomes larger. The run-length code having the calculated correlation equal to or greater than a predetermined value is set by the master setting means as master pattern data. For a plurality of objects that are desired to be substantially the same, the correlation between the position and length of the run-length code is high (correlation is equal to or greater than a predetermined value). Indicates that it is a length code. Therefore, if a run-length code having a high correlation is used as master pattern data, a master pattern can be created from a plurality of objects without requiring complicated operations and enormous processing.

  In a detection device such as a pattern defect detection device, the run length codes generated from the captured image of the object can be compared using the run length code of the created master pattern. Based on the comparison result, that is, when the compared run length codes are different, the number of pixels having the same pixel value, that is, the length is different, and an object having a short or long length is imaged. It can be determined that it is defective.

  When obtaining a correlation between each of a plurality of objects, there may be a positional deviation between the objects even in the same pattern due to a positional state such as a positional deviation at the time of imaging or an error at the time of manufacturing. is there. In addition, slight enlargement / reduction may occur due to an error during imaging. For this reason, as described in claim 3, the correlation calculating means obtains the characteristic of the sum of the same pixel values for each line in a predetermined direction as an X correlation, and a line with a large degree of coincidence of the obtained X correlation. , The correlation between the position and length of the run-length code among a plurality of objects can be obtained. In this way, the correlation can be obtained for each line in the predetermined direction, and the correlation can be obtained by eliminating the positional deviation and the deviation at the time of enlargement / reduction.

  Even if the correlation can be obtained for each line in a predetermined direction, this cannot be resolved if there is a positional shift in the direction in which the pixels on the line continue, that is, in the predetermined direction. For this reason, as described in claim 4, the correlation calculation means obtains the characteristic of the sum of the same pixel values for each line in a predetermined direction as the Y correlation for the line having a large degree of coincidence of the X correlation. It is preferable to further obtain the correlation between the positions and lengths of the run length codes having a high degree of coincidence of the Y correlation. In this way, a correlation can be obtained for each line in a predetermined direction, and by eliminating the deviation on the line, the positional deviation and the deviation at the time of enlargement / reduction can be eliminated with high accuracy. Can be sought.

  As described above, according to the present invention, a run-length code represented by the number of pixels having the same pixel value continuous in a predetermined direction is generated for each of a plurality of objects by the generation unit, and each of the plurality of objects is calculated by the correlation calculation unit. Since the correlation between the position and the length of the run length code between them is obtained, and the run length code whose correlation is equal to or greater than a predetermined value is set as master pattern data by the master setting means, it is complicated from a plurality of objects themselves. There is an effect that a master pattern can be created without requiring a lot of work and enormous processing.

  Further, the correlation calculation means obtains the characteristic of the sum of the same pixel values for each line in a predetermined direction as X correlation, and for the line having a large degree of coincidence of the obtained X correlation, the run length code between a plurality of objects is calculated. By obtaining the correlation between the position and the length, the correlation can be obtained for each line in a predetermined direction, and the correlation can be obtained by eliminating the positional deviation and the deviation at the time of enlargement / reduction. .

  Further, the correlation calculation means obtains the characteristic of the sum of the same pixel values for each line in a predetermined direction as the Y correlation for the line having a large degree of coincidence of the X correlation, and the run length code having a large degree of coincidence of the obtained Y correlation. By further determining the correlation between each position and length, it is possible to determine the correlation for each line in a predetermined direction, and by eliminating the shift on that line, the position shift and enlargement / reduction can be performed with high accuracy. There is an effect that the correlation can be obtained by eliminating the time lag.

  Hereinafter, an example of an embodiment of the present invention will be described in detail with reference to the drawings. In the present embodiment, the present invention is applied to a defect detection apparatus that detects defects in a pattern (manufactured pattern) formed on an IC substrate.

  As shown in FIG. 1, a defect detection apparatus 10 according to the present embodiment includes a camera 12 for imaging an IC substrate, and an A / D conversion apparatus 14 that converts an analog signal that is an output signal from the camera 12 into a digital signal. The motor 16 for moving the IC board relative to the camera 12 in the direction along the surface of the IC board, the driver 18, the monitor 20 capable of displaying at least an image, and the arithmetic unit 22. The arithmetic unit 20 is composed of a microcomputer comprising a CPU 24, a RAM 26, a ROM 28, and an input / output port (I / O) 30, and each is connected by a bus 32 so that commands and data can be exchanged. The ROM 28 stores a processing routine which will be described later. The camera 12 is connected to the input / output port 30 via the A / D converter 14, and the motor 16 is connected via the driver 18. A monitor 20 is connected to the input / output port 30 and a keyboard 34 for inputting data, commands, and the like.

  The A / D conversion device 14 is for converting an analog signal output from the camera 12 into a binary digital signal. However, the A / D converter 14 converts the analog signal output from the camera 12 into a binary digital signal ( (Binary data) may be output. That is, it may be configured as a binarization device for converting an analog signal that is an output signal from the camera 12 into a binary digital signal.

  Next, the operation of the present embodiment will be described.

  When the power is turned on in the arithmetic unit 22, the processing routine of FIG.

  In step 100 of FIG. 2, a production pattern installation process is executed. This production pattern setting process is a process for determining a position at which the camera 12 captures a plurality of production patterns as samples. That is, in order to obtain the shape (length, width, angle, and position data) of the production pattern, at least all of the production patterns must be imaged by the camera 12. Therefore, the camera 12 capable of capturing each of the production patterns and the relative position of the production pattern (including the magnification of the camera 12) are set and stored. This facilitates setting of the position of the IC substrate on which the production pattern is formed when measuring the shape (length, width, angle, and position data) of each of the plurality of production patterns.

  For example, as shown in FIG. 8A, the installation position where the image 50 obtained by capturing the pattern 40 is obtained is stored. The installation position may be obtained by attaching an encoder to the motor 16 and storing data represented by the encoder as position data. In the present embodiment, a plurality (a predetermined number) of production patterns among all the production patterns are set as samples, but all the production patterns may be set as samples.

  In the next step 102, a master pattern manufacturing process that is a reference for detecting a defect in the manufacturing pattern is executed using the sample in which the installation position is stored. As will be described later, this master production process produces a master pattern that is composed of master data that is statistically generated by run length codes in the vertical and horizontal directions using the production pattern (see FIG. 3 for details later).

  In the next step 104, a production pattern setting process is executed. This production pattern installation process is a process for determining a production pattern for detecting a defect at a position for imaging. In the next step 106, run length codes for the vertical and horizontal directions related to the manufacturing pattern are generated, and defects in the manufacturing pattern are detected by comparing the generated data with the master data (see FIG. 7 for details later).

  In the next step 108, it is determined whether or not a production pattern for performing defect detection remains. If it remains, an affirmative determination is made in step 108 and the process returns to step 104 to repeat the above-described process again to detect defects. On the other hand, when the defect detection is completed for all the production patterns, a negative determination is made in step 108, and this routine is terminated. The determination in step 108 may be determined in advance by the number of production patterns, or may be input using the keyboard 34.

  Next, details of step 102 in FIG. 2 will be described. When the master production process is executed in step 102 in FIG. 2, the data production routine in FIG. 3 is executed, and the process proceeds to step 200.

  In step 200 of FIG. 3, the sample 40 which is a production pattern is imaged by reading a digital signal obtained by converting the signal output from the camera 12 by the A / D converter 14 (FIG. 8 (1)). )reference). In the next step 202, the sample image 50 is binarized. In the next step 204, run length processing is performed to read the binarized data for each pixel and create a run length code (see FIG. 4 for details later). In the next step 206, it is determined whether or not the run-length process has been completed for all the pixels of the sample image 50, and the run-length code obtained by the run-length process when the run-length process has been completed for all the pixels of the sample image 50. Is stored as sample data i (i = 1, 2,...) In the next step 208, and in the next step 210, it is determined whether or not the above-described processing for a predetermined number of samples has been completed. If the determination is negative, the process returns to step 200, the next sample is set, and the above process is repeated.

  On the other hand, if the determination at step 210 is affirmative, the routine proceeds to step 212. In this step 212, one sample is set as a temporary master pattern among the predetermined number of samples subjected to the run length processing as described above. That is, one of the samples of the production pattern is set as a temporary master pattern.

  Next, in step 214, a sample different from the sample set as the temporary master pattern is set as a matching sample j (j = 1, 2,..., J ≠ i), and in the next step 216, matching processing is performed. Do. Although details will be described later, this matching processing is processing for obtaining a correlation between a sample set as a temporary master pattern and a matching sample j (see FIG. 5).

  In the next step 218, it is determined whether or not the matching processing for obtaining the correlation has been completed for all the predetermined number of samples. If samples still remain, the determination is negative, and the processing returns to step 214. On the other hand, if the matching process is completed for all samples, the process proceeds to step 220, where a master pattern is generated. The master pattern is generated by setting a run length code having a high correlation between a predetermined number of samples. The details of the master production process in step 220 will be described after the description of the matching process.

  The sample image 50 may be displayed on the monitor 20 as a gradation image or may be displayed as a binarized image.

  Next, details of step 204 in FIG. 3 will be described. When the run length process is executed in step 204 of FIG. 3, the run length process routine of FIG. 4 is executed.

  In describing the run length process, the definition of the run length code in the present embodiment will be described.

As shown in FIG. 9, the tip portion of the partial pattern 60 constituting the pattern 40 composed of black data is located near one corner of the sample image 50 (near the upper left corner of FIG. 8A). . This sample image 50 is composed of an image having a pixel number H ₀ in one direction (horizontal direction, horizontal direction in FIG. 9) and a pixel number V _{0 in} a direction intersecting one direction (vertical direction, vertical direction in FIG. 9). The In the following description, the horizontal position is represented by a pixel position i (1 ≦ i ≦ H ₀ ), the vertical position is represented by a pixel position j (1 ≦ j ≦ V ₀ ), and Is represented by (i, j).

Since the sample image 50 is binarized data, each pixel data is white data or black data. In the present embodiment, the number of pixels in which white data or black data, which is each pixel data, continues is a run-length code. For example, in the example of FIG. 9, the data of the region 62 in which the eighth pixel (j = 8) in the vertical direction is continued in the horizontal direction is white data up to the eighth pixel (i = 8) in the horizontal direction. Black data continues from the 9th pixel to the 12th pixel, and the 13th and subsequent pixels are white data. Therefore, the data in this area 62 can be expressed as “8, 5,...” By the number of continuous pixels of white data or black data. Thus, the white data can be represented as “0”, the black data as “1”, and the data in the area 62 can be represented as “0: 8, 1: 5, 0: n. A continuous group of pixels can be compressed and expressed by “k: n” (k is 0 or 1, n is the number). The continuous pixel data in the horizontal direction is (1, 1) by sequentially reading in the horizontal direction (arrow S direction in FIG. 9) and then sequentially reading in the vertical direction (so-called every scanning line). One image can be represented in the horizontal direction by a run length code represented by “k: n” for one group of pixels from a pixel at a position of (H ₀ , V ₀ ) to a position of (H ₀ , V ₀ ). This data is defined as a horizontal run length code _DH .

Similarly to the horizontal direction, in the example of FIG. 9, the data of the region 64 in which the ninth pixel (i = 9) in the horizontal direction continues in the vertical direction is white data up to the sixth pixel (i = 6) in the vertical direction. And the seventh pixel and thereafter become black data. Therefore, the data in the area 64 can be expressed as “6,...” By the number of continuous pixels of white data or black data. As a result, the data in the region 64 can be expressed as “0: 6, 1: m,...”, And “g: m” (g is 0 for a group of pixels in which white data or black data continues. Or, 1 and m can be expressed in compression. Note that in the vertical direction, since the position of the pixel to be read out must be specified, it is necessary to add the position i of the pixel in the horizontal direction. For this reason, in the present embodiment, the data of the region 64 is represented by “9, 0: 6, 1: m,... That is, to generate a run-length code for each cut out by one pixel in the longitudinal region, the data added with the position i in the horizontal direction of the pixels and the vertical run length codes D _V.

  As described above, in the present embodiment, a run-length code is composed of a horizontal run-length code and a vertical run-length code having the number of pixels in which white data or black data continues as data.

In the following description, the pixels at the position (i, j) = (1, 1) are sequentially read in the horizontal direction (arrow S direction in FIG. 9) and then in the vertical direction (arrow t direction in FIG. 9). It is assumed that pixel data up to a pixel at a position of (H ₀ , V ₀ ) is sequentially read out sequentially (so-called for each scanning line).

  First, in step 120 of FIG. 4, it is determined whether or not this routine is executed for the first time. If it is executed for the first time, initialization processing is performed in the next step 122. That is, the pixel position i in the horizontal direction and the pixel position j in the vertical direction are set to “1”, the number n of pixel groups in which white data or black data continues in the horizontal direction, and the vertical direction continues. The number m of pixel groups to be set is set to “1”.

In the next step 124, one pixel data is read, and it is determined in the next step 126 whether or not the read pixel data matches the previous pixel data (white data or black data) in the horizontal direction. Is affirmed and the routine proceeds to step 132. If they do not match, a negative determination is made in step 126, and in step 128, the horizontal run length code D _H (n) up to the previous time is stored. Then, in step 130, the number n of consecutive pixel groups is incremented and the horizontal run length is increased. The length code D _H (n) is reset.

In the next step 132, it is determined whether or not the pixel data read in step 124 matches the previous pixel data (white data or black data) in the vertical direction. If they do not match, a negative determination is made in step 132, the vertical run length code D _V (i, m) up to the previous time is stored in step 134, and then the number m of the pixel groups continuous in the vertical direction is incremented in step 136. Reset the vertical run length code D _V (i, m).

In the next step 138, the horizontal run length code D _H (n) and the vertical run length code D _V (i, m) are incremented. Thus, each of the horizontal run length code D _H (n) and the vertical run length code D _V (i, m) becomes data representing the number of consecutive white data or black data.

In the next step 140, it is determined whether i = H ₀ or not, and it is determined whether or not the number of read pixels has reached the number of horizontal pixels. The pixel position i in the direction is reset (i = 1), and the pixel position j is incremented in the vertical position. On the other hand, if the determination in step 140 is negative, the pixel still remains in the horizontal direction, so the position i of the pixel in the horizontal direction is incremented.

In the next step 146, it is determined whether or not j <V _{0 by} determining whether or not the number of read pixels is less than the number of vertical pixels. If the determination is affirmative, the next process is performed for the next processing. In step 148, the current data is stored and this routine is terminated. On the other hand, if a negative determination is made in step 146, the processing for all the pixels in the vertical direction has been completed. Therefore, an end flag indicating that the processing has been completed for all the pixels is set, and this routine is terminated.

  As described above, the horizontal run length code and the vertical run length code are generated by reading pixel by pixel.

  Next, details of step 216 in FIG. 3 will be described. When the matching process is executed in step 216 of FIG. 3, the matching process routine of FIG. 5 is executed. First, in step 230, an X correlation representing a correlation for each line in the predetermined direction (the direction of arrow A in FIG. 10) of each of the temporary master pattern and the matching sample j is obtained. In the present embodiment, the line corresponds to the scanning line of the camera 12. When a two-dimensional CCD sensor or a line CCD sensor is used as the camera 12, a cell group in which the minimum cells (elements) constituting a pixel are arranged in one continuous direction can be associated. When this line CCD sensor is used, in order to obtain a two-dimensional image, it is necessary to pick up an image while moving the pattern and the camera in a direction intersecting the direction in which the elements of the line CCD sensor are continuous. The above-described X correlation represents the characteristics of the direction intersecting with a predetermined direction (direction intersecting with the arrow A in FIG. 10) by obtaining the total number of pixels corresponding to black included in the line for each line.

  The X correlation may be obtained using a pixel value for each pixel, or a run length may be used. When a run length code is used, the run length code may be used as it is, or the run length code may be converted into a pixel value.

  FIG. 10 shows images (sample images 50 and 52) in which the substantially matching patterns are slightly shifted. FIG. 10A shows a line on the sample image 50 set as the temporary master pattern as a conceptual image, and a characteristic 70 obtained by X correlation is shown on the right side of FIG. 10A. FIG. 10B shows a line on the matching sample image 52 as a conceptual image, and a characteristic 72 obtained by X correlation is shown on the right side of FIG. 10B. As can be understood from the X correlation characteristics 70 and 72 in FIGS. 10A and 10B, the characteristics are substantially the same and have a relationship shifted in the direction intersecting the arrow A in FIG.

  FIG. 11 shows a sample image 50 and a matching sample image 54 in which a pattern substantially matching the sample image 50 is slightly reduced. It can be understood from the X correlation characteristic 74 on the right side of FIG. 11B that the characteristics are substantially similar. Thereby, the line which shifted | deviated to the direction which cross | intersects the direction along a line is respond | corresponded.

  In the next step 232, a Y correlation representing the correlation in the direction intersecting the predetermined direction of each of the temporary master pattern and the matching sample j (direction intersecting the arrow A direction in FIG. 10) is obtained. This Y correlation represents the total number of pixels corresponding to black in the direction intersecting with the above-mentioned line for each predetermined direction, and represents the characteristic in the predetermined direction (the direction of arrow A in FIG. 10). A characteristic 76 obtained by the Y correlation for the sample image 50 set as the temporary master pattern is shown below the paper surface of FIG. 10 (1), and a matching sample image 52 is shown by the Y correlation on the lower surface of FIG. 10 (2). The obtained characteristic 78 is shown. As understood from the Y correlation characteristics 76 and 78 in FIGS. 10A and 10B, the characteristics are substantially the same and are in a relationship shifted in the direction of arrow A in FIG.

  In the next step 234, a corresponding line is set between each of the temporary master pattern and the matching sample j. That is, in step 230, the line having the highest coincidence in the X correlation characteristics is associated with the temporary master pattern and the matching sample. In the example of FIG. 10, since only the positional relationship is relatively different between each line, all the lines can correspond one-to-one. In the example of FIG. 11, the matching sample image 54 is slightly reduced and therefore does not correspond to a line at a one-to-one position. That is, when the number of lines is the same, the line 80 of the temporary master pattern image 50 should correspond to the line 82 of the sample image 54 for matching. Will be supported.

  Next, in step 238, rough alignment in the line direction is performed using the characteristic of the Y correlation characteristic, and a run-length code for performing a matching check on the line set in step 236 is set. In the next step 240, an overlapping area between the temporary master pattern and the sample is derived for the set run length code. The overlapping area is an area that overlaps when the run-length code on the set line is overlapped with the temporary master pattern and the sample. That is, this step 240 is because it is assumed that there is an overlapping area in the run-length code to be matched on the line.

  In the next step 242, it is determined whether or not the derived overlapping area exists. If there is no overlapping area (see FIG. 12 (2)), the run length code that matches between the temporary master pattern and the sample (No in step 242), the process proceeds to step 246 and matching NG is determined. On the other hand, if there is an overlapping area, an affirmative determination is made in step 242, and in the next step 244, it is determined whether or not there are a plurality of overlapping areas for the temporary master pattern. If there are a plurality of overlapping regions, the result is affirmative in step 244 and the process proceeds to step 246. A case where there are a plurality of overlapping areas of a sample with respect to the temporary master pattern means that a plurality of run-length codes belong to one run-length code of the temporary master pattern as shown in FIG. It is. That is, since a plurality of run-length codes correspond to run-length codes that should originally correspond one-to-one, matching is not performed. For this reason, it is determined as matching NG in step 246.

  On the other hand, as shown in FIG. 12A, when there is only one overlapping region, the result is negative in step 244, and the matching degree is calculated in step 248. This degree of matching is the degree of coincidence of lengths by run length codes, and in this embodiment, the ratio of the length of the sample to the length of the temporary master pattern is used. When the calculation of the matching degree is completed, the process proceeds to step 250, and it is determined whether or not the above process has been completed for all the run length codes. If the processing has not been completed for all the run length codes, the result in Step 250 is negative, and the processing returns to Step 238 to repeat the above processing. On the other hand, when the processing is completed for all run length codes, the result in step 250 is affirmative, and the process proceeds to the next step 252 where it is determined whether the matching processing is completed for all lines. If the line still remains, the result is negative in step 252 and the process returns to step 236 to repeat the above processing. On the other hand, when all the lines have been completed, the result in step 252 is affirmative. In the next step 254, the matching degree is stored, and this routine is terminated.

  When the matching process is completed for a predetermined number of samples in this way, the matching degree for each sample is obtained. A master pattern is generated using these multiple matching degrees (step 220 in FIG. 3).

  Next, details of the master pattern generation process (step 220 in FIG. 3) will be described. When step 220 of FIG. 3 is started, the master data determination routine of FIG. 6 is executed. First, in step 260, a run-length code for determining as master data among the sample run-length codes set as the temporary master pattern is set. In the next step 262, the run length code x0 of the set temporary master pattern and the run length codes x1, x2,..., Xj of each sample with respect to the run length code x0 are read (see FIG. 13), and the next step In H.264, the variation of each matching is obtained. The variation can be obtained by statistical processing such as standard deviation. In the above calculation of the matching degree (step 248 in FIG. 5), since the length ratio is obtained, the length ratio may be used to obtain the variation. The reliability of the sample run-length code x0 set as the temporary master pattern can be determined from the obtained variation. That is, if the variation is large, the reliability is low, and if the variation is small, the reliability is high.

  In the next step 266, it is determined whether or not the variation value obtained in step 264 is less than a predetermined value, thereby determining whether or not the variation is small. When the value is, the reliability is low and the result is negative in step 266 and the process proceeds to step 272. In step 272, the process proceeds to step 276 without determining the sample run-length code set as the temporary master pattern as master data, that is, in an undecided state (see FIG. 14).

  On the other hand, when the variation value is less than the predetermined value and the result is affirmative in step 266, the process proceeds to step 268, where the sample run-length code set as the temporary master pattern by the statistical process is determined as the master data. Store at 270. This statistical processing can use an average value of all run length codes or an average value of specific run length codes extracted. For the extraction of a specific run length code, it is preferable to extract a predetermined portion of the code having a small variation, for example, a run length code of 1/2 or 1/3 from the center distribution.

  In step 274, it is determined whether or not the above process has been completed for all the run length codes. If there is still an unprocessed run length code, the result in step 274 is negative, and the process returns to step 260 to repeat the above process. On the other hand, when the above processing is completed for all the run length codes, an affirmative determination is made in step 274, and it is determined whether or not there is a run length code not set in the next step 276. If there is no run-length code that has not been set, since master data has been set for all run-length codes, the result in Step 276 is negative, and this routine ends.

  If there is a run length code that has not been set, the determination at step 276 is affirmative, and the routine proceeds to the next step 278. In step 278, data corresponding to the run-length code of the unset temporary master pattern is obtained by statistical processing, determined as master data, and updated in the next step 280. The statistical processing in step 278 is performed on the run-length codes x1, x2,..., Xj of each sample excluding the run-length code x0 of the temporary master pattern. First, variation (standard deviation, etc.) of run length codes x1, x2,..., Xj is obtained, and when the variation is small, it is assumed that run length code x0 of the temporary master pattern is defective, and run length codes x1, x2,. ..., Xj average value or average value obtained by extracting a specific run length code is determined as master data. On the other hand, when the obtained variation is large, the target run-length code is removed from the master data, and since the run-length code varies greatly for all samples, data indicating that the master data cannot be produced is output as defective To do. When the target run-length code is removed from the master data, it is also removed from the comparison target without using the master data in the comparison process described later. In this case, master preparation may be performed again using another sample as a temporary master pattern.

  Next, details of step 106 in FIG. 2 will be described. When the comparison process is executed in step 106 of FIG. 2, the comparison process routine of FIG. 7 is executed.

  In step 150 of FIG. 7, the master data created as described above is read (see FIG. 3). In the next step 152, image data obtained by imaging the prepared production pattern 52 (see FIG. 8B) is read and binarized in the next step 154.

  In the next step 156, the above-described run length processing is performed on one pixel of the image 52 obtained by imaging the production pattern (see FIG. 4), and the run length code and the master data run length code for the image 52 obtained by imaging the production pattern are processed. Are compared in the next step 158.

For example, as shown in FIG. 8 (1), in the image stored as the master data 40, the length represented by the horizontal run length code D _H (n) for the pixel position V _{1 in the} vertical direction is the first If u is a continuous group of white data, D _H (u) = A ₁ , D _H (u + 1) = A ₂ , D _H (u + 2) = A ₃ , D _H (u + 3) = A ₄ , DH (u + 4 ) = become A _5. Similarly, the length represented by the horizontal run length code D _H (n) is D _H (u) = a ₁ , D _H (u + 1) = a _{2 with} respect to the vertical pixel position V _{1 in the} production pattern 42. , D _H (u + 2) = a ₃ , D _H (u + 3) = a ₄ , D _H (u + 4) = a ₅ . As a result, the correspondence relationships of the lengths are A ₁ ≈a ₁ , A ₂ ≈a ₂ , A ₃ > a ₃ , A ₄ <a ₄ , and A ₅ ≈a ₅ . Therefore, in step 158, these lengths are compared. For example, the length of the production pattern is subtracted from the master data. This subtraction value is used as a comparison result. As can be understood from FIG. 8, a defect can be detected in the portions of A ₃ > a ₃ and A ₄ <a ₄ in the correspondence relationship of the lengths. In this part, if a value equal to or greater than a predetermined value is obtained as a subtraction value, a permissible range is determined in advance, and a value exceeding this value is determined as a defect, and a defect can be detected.

In the image stored as the master data 40, the length represented by the vertical run-length code D _V (H ₁ , m) for the pixel position H _{1 in the} horizontal direction is DV (H ₁ , 1) = B. ₁ , D _V (H ₁ , 2) = B ₂ and D _V (H ₁ , 3) = B ₃ . Similarly, the length represented by the vertical run-length code D _V (H ₁ , m) is only D _V (H ₁ , 1) = b1 with respect to the horizontal pixel position H _{1 in the} production pattern 42. As a result, the correspondence relationship between the lengths corresponds only to B ₁ <b ₁ , and data corresponding to B ₂ and B ₃ is not detected. Therefore, if a subtraction value obtained by subtracting the length of the production pattern from the master data is used as a comparison result, a defect can be detected in each portion of B ₁ , B ₂ , and B ₃ . In this case, when there is no target data, it is possible to estimate the missing pattern.

  Therefore, in step 160 of FIG. 7, it is determined whether or not there is a defect depending on whether or not the absolute value of the subtraction value exceeds a predetermined value (allowable range). And returns to step 156. On the other hand, since no defect is detected in the case of negative determination, the above processing is repeatedly executed until the run length processing is completed for all pixels (until an affirmative determination is made in step 164).

  When an affirmative determination is made in step 164 and the run length processing is completed for all pixels, it is determined in the next step 166 whether or not there is a defect. If one or more defects exist, all defects are displayed in step 170. Then, this routine is finished. On the other hand, if no defect is detected, a defect-free display indicating that a production pattern corresponding to the master data has been produced is displayed in step 168, and this routine is terminated.

  In the present embodiment, a case has been described in which a defect is detected by comparing a master pattern (master data) and a production pattern (run-length code) using the run-length code itself obtained after binarization processing. A process for producing a master pattern (master data) may be used. That is, applying the process of FIG. 3, the setting of the temporary master pattern in step 212 is replaced with the setting of the master pattern. Also, the master generation process in step 220 is replaced with a determination process. In this determination process, whether the master pattern (master data) and the production pattern (run-length code) substantially match using the matching NG set in step 246 in FIG. 5 or the matching degree obtained in step 248. Determine. The determination may be that the matching is OK when the matching degree is equal to or higher than a predetermined value, the matching is NG when the degree of matching is less than the predetermined value or when the matching is NG, and a defect is detected when the matching is NG.

  As described above, in the present embodiment, master data of the master pattern is generated using the fabrication pattern itself. Accordingly, the generated master data is optimal master data for inspecting the production pattern from the production pattern produced by reflecting an error at the time of production and an allowable range. For this reason, the complicated process of adjusting the master data generated from the design drawing or the like again with the data of the error or the allowable range becomes unnecessary. For this reason, it becomes easy to generate master data.

  In the present embodiment, a run-length code (master data) is produced for the master pattern, and a run-length code is produced for each pixel of the production pattern to be detected by the defect, and compared with the run-length code of the master pattern. By doing so, defect detection is performed. Therefore, since the master data is compressed by storing it as a run-length code, it can be stored with a smaller capacity than a method of detecting defects by comparing pixel data for each pixel, such as ROM and RAM. The capacity of the storage device can be reduced.

  A run length code is created or updated every time pixel data of one pixel is read. For this reason, when detecting a defect, the run length codes may be compared with each other, and high-speed determination is possible.

  Furthermore, the run-length code is produced or updated for each pixel in both the horizontal direction and the vertical direction. Therefore, every time pixel data of one pixel is read, both horizontal and vertical defects can be detected simultaneously.

  In the above-described embodiment, a case has been described in which run-length codes in two perpendicular directions (vertical direction (longitudinal direction) and horizontal direction (lateral direction)) are generated and used. As described above, the pattern defect can be detected with any shape pattern by using only the run length codes in both the horizontal direction and the vertical direction. For example, in the master pattern 40 shown in FIG. 8A, the partial pattern 60 having a constant width is bent in a 45 degree direction at the middle part. For this reason, conventionally, when detecting a pattern width, the axis of width measurement has to be inclined according to the shape of the pattern. In the present embodiment, the horizontal run length code and the vertical run length code include components in all directions, and the width of the pattern can be detected.

For example, as shown in FIG. 8A, the image stored as the master data 40 includes a partial pattern bent in a 45-degree direction at the pixel position V _{2 in the} vertical direction. In the present embodiment, a comparison can be made in the same manner as in the case of the vertical pixel position V ₁ described above. That is, the horizontal length is E ₁ , E ₂ , E ₃ , E ₄ , E ₅ , E ₆ ,... For the master pattern, and e ₁ , e ₂ , e ₃ ,. e ₄ , e ₅ , e ₆ ,... As a result, the correspondence relationships of the lengths are E ₁ ≈e ₁ , E ₂ ≈e ₂ , E ₃ ≈e ₃ , E ₄ ≈e ₄ , E ₅ > e ₅ , E ₆ <e ₆ . Therefore, a defect can be detected from the subtracted value obtained by subtracting the length of the production pattern from the master data.

  In the above embodiment, the case where the position of the substrate including the master pattern and the position of the substrate including the fabrication pattern is first aligned has been described. However, the lengths are compared using the vertical and horizontal run length codes as described above. Sometimes, this is an effective detection method for lateral shift and expansion / contraction of an image. That is, when comparing pixel data for each pixel, it is essential to align the master pattern with the production pattern. Even when the image is shifted even by one pixel or when the size of the image changes due to expansion / contraction, even the same pattern is detected as a defect. In the present embodiment, run length codes in two directions perpendicular to the horizontal direction and the vertical direction are generated, and these run length codes are compared. For this reason, if an allowable range of each run length code is set, a defect can be detected with high accuracy when a lateral shift or an image expansion or contraction occurs during image capture. In addition, if only the patterns are compared, that is, if, for example, run length codes based only on the black data are compared, the pattern in the white data can be treated as a relative position, and the lateral shift at the time of image capture can be reduced. You don't have to consider it.

  In the above-described embodiment, the case where the run-length code is generated using the binarized image obtained by binarizing the captured image has been described. However, the present invention is not limited to binarization, and ternary Or you may make it convert into the multi-value image by the value beyond it. For example, ternary data can be obtained by adding gray data as intermediate data to binary data of white data and black data. These three values are classified into three from the lowest density to the highest density of the original image data (analog data or digital data) for each of the white data, gray data, and black data. The pixel value is determined, and the pixel value in the classified density range including the pixel density may be set as the pixel value of the pixel. If generalized, image data (analog data or digital data) is classified into n items from the lowest density to the highest density, pixel values m are set for each of the classified density ranges, and the pixel density is included. The pixel value in the density range may be set as the pixel value of that pixel.

It is a block diagram which shows schematic structure of the defect detection apparatus concerning embodiment of this invention. It is a flowchart which shows the flow of a process of a defect detection apparatus. It is a flowchart which shows the flow of a master data preparation process. It is a flowchart which shows the flow of a run length process. It is a flowchart which shows the flow of a matching process. It is a flowchart which shows the flow of a master production | generation process. It is a flowchart which shows the flow of a comparison process. It is an image figure which showed the picked-up image, (1) shows the reference | standard image containing a master pattern, (2) has shown the preparation image containing a preparation pattern. It is explanatory drawing for demonstrating a run length code. It is a diagram which shows each image figure of a temporary master pattern and a production pattern, the characteristic of X correlation, and the characteristic of Y correlation. It is a diagram which shows each image figure of the temporary master pattern and the production pattern different from FIG. 11, the characteristic of X correlation, and the characteristic of Y correlation. It is explanatory drawing for demonstrating matching of a run length code. It is an image figure which shows the correspondence of the run length code of a temporary master pattern, and the run length code of a some production pattern. It is an image figure which shows the run length code which is not set in a temporary master pattern.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Defect detection apparatus 12 Camera 14 A / D conversion apparatus 22 Arithmetic apparatus

Claims (2)

For each of the plurality of produced objects, an imaging unit that divides the image into a plurality of vertical and horizontal pixels, and
Pixel value setting means for setting each of a plurality of picked-up images picked up by the image pickup means for each pixel, and a predetermined numerical value based on the density of the pixel as a pixel value of the pixel;
The captured image in which the pixel value of each pixel is set is sequentially read from the pixel value of one pixel, and a length in which the same pixel value continues in a predetermined direction is represented by the number of pixels in which the same pixel value continues. Generating means for generating a length code;
For each of a plurality of captured images captured by the imaging unit, based on the run length code generated by the generating unit, the characteristic of the sum of the same pixel values for each line in the predetermined direction is obtained as an X correlation, For the line having a high degree of coincidence of the X correlation obtained, the position and length of each run-length code having a large degree of coincidence of the Y correlation obtained by obtaining the characteristic of the sum of the same pixel values for each line in the predetermined direction as the Y correlation. by obtaining the the correlation, the correlation calculation means for obtaining a correlation between the position and length of the run length code among the plurality of images captured by the imaging means,
Master setting means for setting a run length code having a calculated correlation value equal to or greater than a predetermined value as master pattern data;
A master pattern creation device.   The pixel value setting means represents a pixel value or black data representing white data for each pixel by binarization that classifies the pixel density as either white when a predetermined density is lower than a predetermined density or black when the density is higher than the predetermined density. 2. The master pattern creating apparatus according to claim 1, wherein a pixel value is set.

Images