JP2010164487A - Defect inspecting apparatus and defect inspecting method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve a defect inspection in detection sensitivity for a minute defect while suppressing generation of pseudo defects. <P>SOLUTION: A defect inspecting apparatus 1 is provided which includes: a gray level determining means 15 for determining a gray level difference of corresponding pixels between an image to be inspected and a reference image; and a defect determining means 16 for determining whether or not an inspection object pixel is a defect candidate by using a pattern recognition which carries out a classification in accordance with a value of a discriminant function whose feature values are respective gray level differences in the inspection object pixel and its adjacent pixel. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a defect inspection apparatus and a defect for detecting a difference portion as a defect by comparing an inspection image obtained by imaging each of the same patterns in a repetitive pattern appearing on a sample surface with a reference image. It relates to the inspection method.

  The manufacture of semiconductor devices such as semiconductor wafers, photomask substrates, and liquid crystal display panel substrates is made up of a large number of man-hours. Inspects the occurrence of defects in the final and intermediate processes and provides feedback to the manufacturing processes. It is important to improve the yield. In order to detect defects in the middle of the manufacturing process, images obtained by imaging the pattern formed on the surface of a sample such as a semiconductor wafer, a photomask substrate, a liquid crystal display panel substrate, or a liquid crystal device substrate are obtained. A pattern defect inspection for detecting defects present on the surface of a sample by inspecting the surface is widely performed (for example, Patent Document 1 below).

  In the following description, a semiconductor wafer defect inspection apparatus that inspects a defect of a pattern formed on a semiconductor wafer will be described as an example. However, the present invention is not limited to this, and can be widely applied to defect inspection apparatuses for inspecting semiconductor devices such as a semiconductor device photomask substrate, a liquid crystal device substrate, and a liquid crystal display panel substrate. .

  FIG. 1 is a diagram showing an array of dies (chips) on a semiconductor wafer (hereinafter referred to as “wafer”) to be inspected (sample). A plurality of dies 3 are repeatedly arranged in a matrix on the wafer 2 in the X and Y directions. Since the same pattern is formed on each die, the images obtained by imaging these dies should be essentially the same, and the pixel values of the corresponding portions of the captured images of each die are essentially the same values.

  Therefore, the two dies are imaged to obtain the pixel value (gray level value) of each pixel, and the pixel value difference (gray level difference signal) between corresponding portions that should be essentially the same in the captured images of the two dies. Is detected, the gray level difference signal is greater when one die is defective than when both dies are not defective.

  Therefore, using one of the two dies as an inspection image and the other as a reference image, the gray level difference between corresponding pixels between these images is detected, and this gray level difference is compared with a predetermined detection threshold. Thus, defects existing on the die can be detected by detecting a gray level difference exceeding the detection threshold (die-to-die comparison).

  Further, when a repeated pattern such as a memory cell is formed in one die, a defect is detected even if a gray level difference is detected between images obtained by capturing corresponding portions that should be the same in the repeated pattern. Can be detected (cell-to-cell comparison).

  In die-to-die comparison, it is common to compare images obtained by imaging two adjacent dies (single detection). This does not tell which die is defective. Therefore, a comparison is further made with a die adjacent to a different side, and it is determined that the die is defective when the gray level difference of the same portion becomes larger than a predetermined detection threshold again (double detection). The same applies to the cell-to-cell comparison.

  Then, a defect classification process is performed in which the portion determined to be a defect is examined in more detail to determine whether it is a true defect that affects the yield. In the defect classification process, it is necessary to examine the defect portion in detail, and a long processing time is required. Therefore, when determining a defect, it is required that a true defect is not leaked and that other than the true defect is not determined as a defect as much as possible.

  Therefore, the threshold setting is a big problem. If the threshold value is decreased, the number of pixels (pixels) that are determined to be defective increases, and a portion that is not a true defect, that is, a “pseudo defect” is determined to be a defect, and the time required for the defect classification process increases. Arise. On the other hand, if the threshold value is too large, it is determined that the true defect is not a defect, which causes a problem that the inspection is insufficient.

  2A to 2C are explanatory diagrams of a detection threshold value determining method in Patent Document 1 below. First, a gray level difference between pixels corresponding to each other between the base image and the reference image is calculated, and a gray level difference histogram as shown in FIG. 2A is created. Next, the accumulated frequency of gray level differences is calculated from the created histogram. Further, assuming that the gray level difference follows a predetermined distribution, the cumulative frequency is converted so that the cumulative frequency has a linear relationship with the gray level value. The gray level value may be assumed to follow, for example, a normal distribution, a Poisson distribution, or a chi-square distribution. This conversion cumulative frequency is shown in FIG.

An approximate straight line (y = ax + b) indicating the relationship between the gray level value and the conversion cumulative frequency is derived according to the converted conversion cumulative frequency. This approximate straight line is shown in FIG. The detection threshold Thd is determined from the parameters a and b of the approximate line and the sensitivity setting parameter (fixed value). Here, VOP and HO are set as fixed sensitivity setting parameters in the approximate straight line of the gray level value and the conversion cumulative frequency, and the cumulative frequency P1 (p is multiplied by the number of samples corresponding to the cumulative probability (p). )) Is obtained, and the gray level value obtained by moving VOP in the vertical axis direction and HO in the horizontal axis direction from that point is set as a detection threshold Thd. Therefore, the detection threshold value Thd is a predetermined calculation formula Thd = (P1−b + VOP) / a + HO.
Is calculated by

JP 2004-177397 A

  With the recent miniaturization of design rules, the defect size affecting the yield has become smaller. For this reason, a defect inspection apparatus is required to be able to detect a defect stably even for a micro defect having a small gray level difference between an inspection image and a reference image. In order to detect a defect with a small gray level difference, it is necessary to use a small detection threshold Thd. However, if the detection threshold Thd is decreased, a large number of pseudo defects are detected. The reason for this will be described with reference to FIG.

  FIG. 3 is a diagram illustrating a histogram of gray level differences between corresponding pixels between the inspection image and the reference image. A solid line 50 shows a histogram of gray level differences of all pixels, a broken line 51 shows a histogram of gray level differences of pixels other than the defective part, and a one-dot chain line 52 shows a histogram of gray level differences of pixels of the defective part.

  The inspection image and the reference image for detecting the gray level difference have variations in the gray level value due to variations in characteristics of the image pickup device that photographs the sample and variations in illumination that illuminates the sample. Further, as the defect size that affects the yield is miniaturized, the gray level difference itself that may indicate a true defect is also reduced. For this reason, as shown in FIG. 3, the same gray level difference may be detected both in the pixel showing the true defect portion and the pixel showing the pseudo defect. Therefore, there is a possibility that the approach of optimal determination of the threshold value that has been performed so far may not be a fundamental solution.

  In view of the above problems, an object of the present invention is to improve the detection sensitivity of a minute defect while suppressing the occurrence of a pseudo defect.

  According to one aspect of the embodiment of the present invention, the inspection image obtained by imaging the patterns that should be identical to each other in the repetitive pattern appearing on the sample surface is compared with the reference image, and the difference is determined as a defect. A defect inspection apparatus for detecting as follows. The defect inspection apparatus includes: a gray level difference determining unit that determines a gray level difference between corresponding pixels between the inspection image and the reference image; and each gray level difference between the inspection target pixel and its adjacent pixels. Defect determining means for determining whether or not the pixel to be inspected is a defect candidate by pattern recognition that performs classification according to the value of the discriminant function having a feature amount as.

  According to another aspect of the embodiment of the invention, the difference between the inspection image and the reference image obtained by imaging the patterns which should be identical to each other in the repetitive pattern appearing on the sample surface is determined as a defect. A defect inspection method to detect is given. In this defect inspection method, a gray level difference between corresponding pixels is determined between the inspection image and the reference image, and each gray level difference between the inspection target pixel and its adjacent pixels is identified as a feature amount. It is determined whether or not the pixel to be inspected is a defect candidate by pattern recognition that performs classification according to the value of the function.

  According to the present invention, since pattern recognition is performed in consideration of not only the pixel to be inspected but also the gray level difference between adjacent pixels, it is possible to detect a minute defect with higher accuracy.

It is a figure which shows the arrangement | sequence of the die | dye on semiconductor wear. (A)-(C) are explanatory drawings of the detection threshold value determination method. It is a figure which shows the histogram of a gray level difference. It is explanatory drawing of the defect of a subpixel size, and the positional relationship of a pixel. It is a schematic block diagram of the defect inspection apparatus by the Example of this invention. 3 is a flowchart illustrating a defect inspection method according to an embodiment of the present invention. It is a figure which shows the order number defined about each adjacent pixel of a test object pixel. It is a flowchart which shows the defect detection process shown in FIG. It is explanatory drawing of the defect detection process shown in FIG.

  A minute defect whose size on the inspection image is smaller than one pixel (hereinafter referred to as “sub-pixel size defect”) is determined between the inspection image and the reference image depending on the positional relationship between the sub-pixel size minute defect and the pixel. The gray level difference may change. FIG. 4 is an explanatory diagram of the positional relationship between a pixel having a sub-pixel size defect and a pixel.

  In FIG. 4, each square in the grid represents each pixel on the inspection image. Reference numerals 60 to 62 indicate sub-pixel size defects. As shown, the defect 60 exists only in the pixel 70, the defect 61 exists over the pixels 71 and 72, and the defect 62 exists over the pixels 73 to 76. Therefore, since the signal indicating the brightness difference between the defect 61 and the defect 62 and the surroundings is distributed to a plurality of pixels, the signal is smaller than a signal generated by the defect 60 existing only in one pixel 70. This is considered to be one of the reasons why the gray level difference between the inspection image and the reference image becomes small for a minute defect.

  Therefore, in the present invention, pattern recognition is performed in consideration of a gray level difference between a pixel to be inspected and an adjacent pixel adjacent to the pixel to be inspected instead of paying attention to each pixel to be inspected. Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. FIG. 5 is a schematic configuration diagram of a defect inspection apparatus according to an embodiment of the present invention.

  The defect inspection apparatus 1 is provided with a stage 11 movable in a three-dimensional direction, and a sample stage (chuck stage) 12 is provided on the upper surface of the stage 11. A wafer 2 as a sample to be inspected is placed on the sample stage 12 and fixed.

  Above the sample stage 12, an imaging unit 13 for capturing an optical image of the surface of the wafer 2 is provided. As the imaging unit 13, an image sensor such as a one-dimensional or two-dimensional CCD camera (preferably a TDI camera) is used. The imaging unit 13 converts an optical image of the surface of the wafer 2 formed on the light receiving surface into an electric signal. In this embodiment, a TDI camera which is a one-dimensional sensor is used as the imaging unit 13. By moving the imaging unit 13 and the wafer 2 relatively by moving the stage 11, the imaging unit 13 is scanned in the X direction and the Y direction with respect to the wafer 2 to obtain a two-dimensional image of the surface of the wafer 2. . Note that either the bright field optical system or the dark field optical system may be used for capturing an image of the surface of the wafer 2 by the imaging unit 13. The image signal output from the imaging unit 13 is converted into a multi-value digital signal (gray level signal) and then stored in the image storage unit 14.

  The difference detection unit 15 uses one image obtained by imaging two dies in the captured image stored in the image storage unit 14 as an inspection image and the other image as a reference image, and the gray level of corresponding pixels between these images. The difference is detected and output to the defect detection unit 16. The defect detection unit 16 detects defect candidates based on the gray level difference detected between the inspection image and the reference image, and further determines which die each defect candidate is in which die by double detection. To do. The defect detection unit 16 generates and outputs defect information indicating the position of the detected defect.

  FIG. 6 is a flowchart illustrating a defect inspection method according to an embodiment of the present invention. In step S <b> 1, the imaging unit 13 images the surface of the wafer 2. The image signal output from the imaging unit 13 is stored in the image storage unit 14 after being converted into a gray level signal. In step S <b> 2, the difference detection unit 15 uses a corresponding portion in the captured image stored in the image storage unit 14, that is, one of the portions captured by the two dies as an inspection image, and the other as a reference image. The gray level difference between the pixels to be detected is detected.

  In step S3, the defect detection unit 16 detects a defect candidate based on the gray level difference detected in step S2. As will be described below, the defect detection unit 16 determines the gray level difference X1 in a pixel to be inspected as a defective pixel (hereinafter referred to as “inspection target pixel”) and a predetermined detection threshold Thd. With respect to the inspection target pixel whose absolute value | X1-Thd | is less than a predetermined margin σ1, pattern recognition is performed by using a gray level difference between the inspection target pixel and its adjacent pixels as a feature amount. Determine whether or not.

  In order to explain pattern recognition by the defect detection unit 16, a notation method used for the inspection target pixel and adjacent pixels is shown below. FIG. 7 is a diagram showing sequence numbers defined for each adjacent pixel of the pixel to be inspected. The cell 80 indicates the pixel to be inspected, and the cells 81 to 88 indicate the adjacent pixels. Each adjacent pixel 81-88 is given a sequence number from 2 to 9, respectively. For example, the second adjacent pixel is the pixel 81 and the fifth adjacent pixel is the pixel 84.

  Further, the gray level difference detected between the inspection image and the reference image for the inspection target pixel is denoted as a first gray level difference X1. For the second adjacent pixel 81 to the ninth adjacent pixel 88, the gray level differences detected between the inspection image and the reference image are denoted as second gray level difference X2 to ninth gray level difference X9, respectively. Here, the i-th gray level difference (i is an integer of 1 to 9) is denoted as Xi.

  FIG. 8 is a flowchart showing the defect detection process shown in FIG. In step S10, the defect detection unit 16 determines whether or not the absolute value | X1−Thd | of the difference between the first gray level difference X1 and the predetermined detection threshold Thd is less than the predetermined margin σ1. When | X1-Thd | <σ1 (step S10: Y), the defect detection unit 16 advances the process to step S14. When | X1-Thd | <σ1 is not satisfied (step S10: N), the defect detection unit 16 advances the process to step S11.

In step S11, the defect detection unit 16 determines whether or not the first gray level difference X1 satisfies the following conditional expression (1).
X1-Thd-σ1> 0 (1)

  When the first gray level difference X1 satisfies the conditional expression (1) (step S11: Y), in step S12, the defect detection unit 16 determines that the inspection target pixel is a defect candidate. When the first gray level difference X1 does not satisfy the conditional expression (1) (step S11: N), in step S13, the defect detection unit 16 determines that the defect detection unit 16 does not identify the inspection target pixel as a defect candidate. Thereafter, the defect detection unit 16 ends the defect determination process for the inspection target pixel.

  FIG. 9 is an explanatory diagram of the defect detection process shown in FIG. In the upper histogram of the figure, a hatched region 53 indicates a portion where the absolute value | X1−Thd | of the difference between the first gray level difference X1 and the predetermined detection threshold Thd is less than the predetermined margin σ1. This range includes both non-defective pixels indicated by the broken line 51 and defective pixels indicated by the alternate long and short dash line 52. Therefore, in steps S11 to S13, the defect detection unit 16 determines whether or not a pixel other than the region 53 is a defect candidate. For the pixel in the region 53, pattern recognition processing is performed in step S14 and subsequent steps. Is used to determine whether or not it is a defect candidate.

  In step S <b> 14, the defect detection unit 16 initializes the value of the index variable i that refers to the sequence number of adjacent pixels to 2. The index variable i has an integer value from 2 to N. N is a set value of the sequence number of adjacent pixels referred to in the pattern recognition processing performed in step S14 and the subsequent steps, and may be an arbitrary value of 2 or more and 9 or less.

  In step S15, the defect detection unit 16 determines whether or not the next product-sum operation value (2) is less than a predetermined margin σi. Here, variable j is an index variable having an integer value from 1 to variable i, and aij and bi are predetermined coefficients. The margin σi is determined in advance by the operator according to the desired detection sensitivity.

  When the product-sum operation value (2) is less than the predetermined margin σi (step S15: Y), the defect detection unit 16 moves the process to step S19. When the product-sum operation value (2) is not less than the predetermined margin σi (step S15: N), the defect detection unit 16 moves the process to step S16. In step S16, the defect detection unit 16 determines whether or not the next discrimination function (3) is greater than zero. The discriminant function is a function that gives an index as to which class each pattern belongs to by inputting a feature quantity representing each pattern in order to identify the patterns. As shown in the following equation, the discriminant function (3) is a discriminant function having the gray level difference Xi between the pixel to be inspected and its adjacent pixels as a feature quantity.

  When the identification function (3) is greater than 0 (step S16: Y), in step S17, the defect detection unit 16 determines that the inspection target pixel is a defect candidate. When the identification function (3) is 0 or less (step S16: N), in step S18, the defect detection unit 16 determines that the inspection target pixel is not a defect candidate. Thereafter, the defect detection unit 16 ends the defect determination process for the inspection target pixel. That is, in steps S16 to S18, the defect detection unit 16 determines whether or not the previous inspection target pixel is a defect candidate by performing classification according to the value of the identification function (3).

  In the lower part of FIG. 9, each pixel to be inspected as a sample is arranged on an identification plane (feature space) in which the abscissa is determined by the value of the first gray level difference X1 and the ordinate is determined by the value of the second gray level difference. The distribution of the sample is shown schematically. Non-defective pixels are distributed in the range indicated by the solid line 90, and defective pixels are distributed in the range indicated by the broken line 91.

For example, when the value of the index variable i is 2, the discrimination function (3) is given by the following equation (4).
a21 × X1 + a22 × X2 + b2-σ2 (4)
At this time, in steps S15 to S18, the defect detection unit 16 determines the inspection target pixel having the first gray level difference X1 larger than the hyperplane 92 (a21 × X1 + a22 × X2 + b2 + σ2 = 0) as the defect candidate, and determines the hyperplane 93 ( It is determined that the inspection target pixel having the first gray level difference X1 smaller than a21 × X1 + a22 × X2 + b2−σ2 = 0) is not a defect candidate.

  Therefore, in steps S15 to S18 when the value of the index variable i is 2, the defect detection unit 16 determines the pixels in the portion of the region 53 that has not been determined as a defect candidate in the determinations in steps S10 to S13. A pixel in the region 94 is newly determined as a defect candidate, and a pixel in the region 95 is determined not to be a defect candidate.

  In step S19, the defect detection unit 16 increases the value of the index variable i by one. In step S20, the defect detection unit 16 determines whether or not the value of the index variable i is larger than (N−1). When the value of the index variable i is equal to or smaller than (N−1) (step S20: N), the defect detection unit 16 returns the process to step S15. For this reason, the defect detection unit 16 determines that the product-sum operation value (2) and the discriminant function (2) until the product-sum operation value (2) becomes equal to or larger than the margin σi or the index variable i reaches N in the determination in step S15. Steps S15 to S20 are repeated while increasing the number of adjacent pixels considered in 3).

When the value of the index variable i becomes larger than (N−1) (step S20: Y), in step S21, the defect detection unit 16 determines whether or not the next discrimination function (5) is larger than zero.

  When the identification function (5) is greater than 0 (step S21: Y), in step S22, the defect detection unit 16 determines that the inspection target pixel is a defect candidate. When the identification function (5) is 0 or less (step S21: N), in step S23, the defect detection unit 16 determines that the inspection target pixel is not a defect candidate. Thereafter, the defect detection unit 16 ends the defect determination process for the inspection target pixel.

  Please refer to FIG. In step S4, the defect detection unit 16 determines in which die each defect candidate is a defect by double detection. The defect detection unit 16 generates and outputs defect information indicating the position of the detected defect.

  In the conventional defect inspection apparatus, as described with reference to FIG. 2, after calculating the gray level difference between the inspection image and the reference image, after calculating the detection threshold Thd, the histogram of the gray level difference is calculated, And other calculations for calculating statistics. On the other hand, according to the present embodiment, by using a predetermined value obtained experimentally in advance as the detection threshold Thd, it is not necessary to calculate the detection threshold Thd for each inspection image, and is used for calculating the detection threshold Thd. The calculation time which has been done can be saved.

  In this embodiment, a calculation process for calculating the value of the discrimination function is added. By appropriately setting the detection threshold value Thd and the margin σ1, the number of pixels to be inspected for performing the pattern recognition process after step S14. By adjusting, the calculation amount of the discriminant function can be reduced as compared with the calculation amount of the detection threshold value Thd performed by the conventional visual inspection apparatus.

  The above-described coefficients aij and bi used in the pattern recognition process can be determined using a support vector machine, a neural network, and a decision tree using a known learning sample. In this embodiment, before performing the pattern recognition process after step S14, the determination is completed for the pixel to be inspected with | X1-Thd |> σ1 by the process of steps S10 to S13, and the target sample is subjected to pattern recognition. Can be removed from. For this reason, since it can be expected that the number of samples subjected to pattern recognition can be reduced, the coefficients aij and bi can be determined more easily.

  In this embodiment, the product-sum operation value of the i-th gray level difference (i = 1 to N) is used in the determination formula and the discrimination function in step S15. Instead, a nonlinear function such as a polynomial of the i-th gray level difference or a Gaussian function may be used in the determination formula and the discrimination function in step S15. By using a non-linear function, it is possible to determine whether or not a defect candidate is highly accurate even when a pixel to be inspected cannot be linearly separated.

  In addition to the i-th gray level difference (i = 1 to N), the determination formula and the discrimination function in step S15 have the gray level of the inspection image or the reference image in the inspection target pixel or adjacent pixels as a feature amount. May be. In the inspection image or the reference image, a portion where the gray level value itself is large has a large gray level difference, and a portion where the gray level value itself is small has a small gray level difference. Therefore, by adding the gray level of the inspection image or the reference image to the feature amount, it is possible to realize pattern recognition with higher accuracy.

  The present invention is applicable to a defect inspection apparatus and a defect inspection method for detecting a defect based on an inspection image obtained by imaging a sample surface.

DESCRIPTION OF SYMBOLS 1 Defect inspection apparatus 2 Wafer 11 Stage 12 Sample stand 13 Imaging part 14 Image memory | storage part 15 Difference detection part 16 Defect detection part

Claims (8)

In the defect inspection apparatus for comparing the inspection image obtained by imaging each of the patterns that should be identical to each other in the repetitive pattern appearing on the sample surface and the reference image, and detecting a difference as a defect
Gray level difference determining means for determining a gray level difference between corresponding pixels between the inspection image and the reference image;
Defect determination means for determining whether or not the inspection target pixel is a defect candidate by pattern recognition that performs classification according to the value of an identification function having the gray level difference between the inspection target pixel and each adjacent pixel as a feature quantity When,
A defect inspection apparatus comprising:   The defect inspection apparatus according to claim 1, wherein the defect determination unit performs the pattern recognition only on an inspection target pixel in which the gray level difference is within a predetermined range. The predetermined range is a range less than a first threshold and greater than a second threshold;
The defect inspection apparatus according to claim 2, wherein the defect determination unit determines that the inspection target pixel is a defect candidate when the gray level difference is equal to or greater than the first threshold.   The defect inspection apparatus according to claim 1, wherein the identification function further uses a gray level of the inspection image or the reference image in the inspection target pixel or the adjacent pixel as a feature amount. In a defect inspection method for detecting a different portion as a defect by comparing an inspection image obtained by imaging each of the same pattern in a repetitive pattern appearing on the sample surface and a reference image,
Determining a gray level difference between corresponding pixels between the inspection image and the reference image;
Defect inspection that determines whether or not the pixel to be inspected is a defect candidate by pattern recognition that performs classification according to the value of an identification function having the gray level difference between the pixel to be inspected and each adjacent pixel as a feature quantity Method.   The defect inspection method according to claim 5, wherein the pattern recognition is performed only on an inspection target pixel in which the gray level difference is within a predetermined range. The predetermined range is a range less than a first threshold and greater than a second threshold;
The defect inspection method according to claim 6, wherein when the gray level difference is equal to or greater than the first threshold, the inspection target pixel is determined as a defect candidate.   The defect inspection method according to claim 5, wherein the identification function further uses a gray level of the inspection image or the reference image in the inspection target pixel or the adjacent pixel as a feature amount.

Images