JP5765713B2 - Defect inspection apparatus, defect inspection method, and defect inspection program - Google Patents

Description

  The present invention relates to a defect inspection apparatus, a defect inspection method, and a defect inspection program. More particularly, the present invention relates to a defect inspection apparatus, a defect inspection method, and a defect inspection program for determining a defect type.

  Patent Document 1 discloses a method for detecting a granular point in an image. In patent document 1, in order to detect the defect of a printed circuit board or a semiconductor wafer, the image is imaged with SEM. Then, defects are detected by comparing the images. An FIR (Finite Impulse Response) bandpass filter is used.

Japanese Patent Laid-Open No. 10-171996

  In macro inspections that inspect the entire sample for unevenness, uneven defects due to image unevenness due to changes in film pressure, etc., and spot-like defects represented by particles, etc., may be detected and must be properly classified. There is. However, the method of Patent Document 1 has a problem that the type of defect cannot be appropriately classified.

  An object of the present invention is to provide a defect inspection apparatus, a defect inspection method, and a defect inspection program that can detect a defect with high sensitivity and further determine the type of the defect.

  A defect inspection apparatus according to a first aspect of the present invention includes a light source that illuminates a sample, a detector that detects light from the sample illuminated by the light source, and a sample image acquired by the detector. A processing device that performs processing, wherein the processing device uses the first filter for the sample image to generate a first filter image, and the sample image, the first filter image, A first difference image corresponding to the difference is generated, and a first defect is detected based on the first difference image. By doing in this way, a defect can be discriminate | determined appropriately.

  A defect inspection apparatus according to a second aspect of the present invention is the defect inspection apparatus described above, wherein a second filter having a characteristic different from that of the first filter is used for the first difference image. A second filter image is generated, a second difference image corresponding to the difference between the first difference image and the second filter image is generated, and a second difference image is generated based on the second difference image. A defect is detected. By doing in this way, a defect can be discriminate | determined appropriately.

  A defect inspection apparatus according to a third aspect of the present invention is the above-described defect inspection apparatus, wherein the processing apparatus has a characteristic different from that of an nth (n is an integer of 1 or more) filter (n + 1). Is used for the nth difference image, so that the (n + 1) th filter image is generated, the nth difference image, and the (n + 1) th filter image according to the difference. A difference process for generating a (n + 1) th difference image, and repeatedly increasing the n, and detecting the (n + 1) th defect based on the (n + 1) th difference image. is there. Thereby, various types of defects can be discriminated.

  A defect inspection apparatus according to a fourth aspect of the present invention is the defect inspection apparatus described above, wherein the nth filter is an FIR filter. Thereby, a defect can be detected appropriately.

  A defect inspection apparatus according to a fifth aspect of the present invention is the defect inspection apparatus described above, wherein a plurality of images are generated by using the nth filter in different operation directions, and the plurality of images are By using the minimum value filter, the nth filter image is generated, and the nth defect is detected based on the nth filter image. By doing in this way, a defect can be discriminate | determined appropriately.

  A defect inspection method according to a sixth aspect of the present invention is a defect inspection method for performing defect inspection of a sample by receiving light from a sample with a detector, and the defect inspection method is performed on the sample image acquired by the detector. On the other hand, by using a first filter, a step of generating a first filter image, a step of generating a first difference image according to a difference between the sample image and the first filter image, Detecting a first defect based on the first difference image. By doing in this way, a defect can be discriminate | determined appropriately.

  A defect inspection method according to a seventh aspect of the present invention is the above-described defect inspection method, wherein a second filter having characteristics different from the first filter is used for the first difference image. A second filter image is generated, a second difference image corresponding to the difference between the first difference image and the second filter image is generated, and a second difference image is generated based on the second difference image. A defect is detected. By doing in this way, a defect can be discriminate | determined appropriately.

  A defect inspection method according to an eighth aspect of the present invention is the above-described defect inspection method, wherein the (n + 1) th filter having characteristics different from those of the nth (n is an integer of 1 or more) filter is (n + 1) th difference corresponding to the difference between the filter process for generating the (n + 1) th filter image, the nth difference image, and the (n + 1) th filter image by using the n difference image. The difference process for generating an image is repeatedly performed to sequentially increase the n, and the (n + 1) th defect is detected based on the (n + 1) th difference image. Thereby, various types of defects can be discriminated.

  A defect inspection method according to a ninth aspect of the present invention is the defect inspection method described above, wherein the nth filter is an FIR filter. Thereby, a defect can be detected appropriately.

  A defect inspection method according to a tenth aspect of the present invention is the above-described defect inspection method, wherein a plurality of images are generated by using the nth filter in different action directions, and the plurality of images are subjected to the defect inspection method. By using the minimum value filter, the nth filter image is generated, and the nth defect is detected based on the nth filter image. By doing in this way, a defect can be discriminate | determined appropriately.

  A defect inspection program according to an eleventh aspect of the present invention is a defect inspection program for performing defect inspection of a sample by receiving light from the sample with a detector, and is acquired by the detector with respect to a computer. Generating a first filter image by using a first filter for the sample image, and generating a first difference image corresponding to a difference between the sample image and the first filter image. And a step of detecting a first defect based on the first difference image. By doing in this way, a defect can be discriminate | determined appropriately.

  A defect inspection program according to a twelfth aspect of the present invention is the above-described defect inspection program, wherein a second filter having characteristics different from the first filter is used as the first difference image with respect to the computer. By using this, a second filter image is generated, a second difference image corresponding to the difference between the first difference image and the second filter image is generated, and based on the second difference image The second defect is detected. By doing in this way, a defect can be discriminate | determined appropriately.

  A defect inspection program according to a thirteenth aspect of the present invention is the above-described defect inspection program, wherein the computer has a characteristic (n + 1) different from that of the nth (n is an integer of 1 or more) filter. ) Filter is used for the nth difference image, so that the (n + 1) th filter image is generated, the nth difference image, and the (n + 1) th filter image according to the difference. The n is sequentially increased by repeatedly executing the difference process for generating the (n + 1) th difference image and the detection process for detecting the (n + 1) th defect based on the (n + 1) th difference image. The (n + 1) th defect is detected based on the (n + 1) th difference image. Thereby, various types of defects can be discriminated.

  A defect inspection program according to a fourteenth aspect of the present invention is the defect inspection program described above, wherein the nth filter is an FIR filter. Thereby, a defect can be detected appropriately.

  A defect inspection program according to a fifteenth aspect of the present invention causes the computer to generate a plurality of third filter images by using the first filter in different action directions in the defect inspection program. The first filter image is generated by using the plurality of third filter images as the minimum value filter, and the third defect is detected based on the first filter image. . By doing in this way, a defect can be discriminate | determined appropriately.

  According to the present invention, it is possible to provide a defect inspection apparatus, a defect inspection method, and a defect inspection program that can detect a defect with high sensitivity and further determine the type of the defect.

It is a perspective view which shows typically the whole structure of the defect inspection apparatus concerning this Embodiment. It is a block diagram which shows the processing apparatus of the defect inspection apparatus concerning this Embodiment. It is a figure which shows an example of the image acquired with the defect inspection apparatus. It is a figure which shows an example of the image which the brightness nonuniformity generate | occur | produced. It is a figure which shows an example of the image which the frame-shaped nonuniformity generate | occur | produced. It is a figure which shows an example of the image which the dotted | punctate nonuniformity generate | occur | produced. It is a figure which shows an example of the image which striped unevenness generate | occur | produced. It is a figure which shows an example of the image which the inclination unevenness generate | occur | produced. It is a figure which shows the sample image A by a dummy defect image. It is a figure which shows the spectrum of a FIR filter. It is a figure which shows the FIR filter calculated | required from the frequency component. It is a figure which shows the filter image B produced | generated using the 1st FIR filter in the X direction. It is a figure which shows the filter image C produced | generated using the 1st FIR filter in the Y direction. It is a figure which shows the filter image D produced | generated using the 1st FIR filter in the diagonal direction. It is a figure which shows the filter image E produced | generated using the 1st FIR filter in the diagonal direction. It is a figure which shows the filter image F produced | generated using the minimum value filter. It is a figure for the process for producing | generating the difference image G according to the difference of the sample image A and the filter image F. FIG. It is a figure which shows the filter image H produced | generated using the 2nd FIR filter. It is a figure for the process for producing | generating the difference image G according to the difference of the difference image G and the filter image H. It is a figure for demonstrating the profile data for detecting frame-shaped nonuniformity. It is a figure which shows the detection result of the defect by frame shape nonuniformity. It is a figure which shows the detection result of the defect by a dotted | punctate nonuniformity. It is a figure which shows the detection result of the defect by striped unevenness. It is a figure which shows the detection result of the defect by inclination unevenness. It is a figure which shows the process in the case of test | inspecting a wafer with a pattern. It is a flowchart which shows the test | inspection method which test | inspects a wafer with a pattern. It is a flowchart which shows another inspection method which test | inspects a wafer with a pattern.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. The following description shows preferred embodiments of the present invention, and the scope of the present invention is not limited to the following embodiments. In the following description, the same reference numerals indicate substantially the same contents.

  The configuration of the inspection apparatus according to this embodiment will be described with reference to FIG. FIG. 1 is a perspective view schematically showing the overall configuration of the inspection apparatus. The inspection apparatus includes a light source 10, a detector 11, a stage 30 and a processing device 50. Further, in FIG. 1, an XYZ three-dimensional orthogonal coordinate system is shown for clarity of explanation. The Z direction is the vertical direction, and the X direction and the Y direction are horizontal directions.

  A light source 10 and a detector 11 are arranged on the sample 20. The light source 10 is a linear light source that irradiates linear illumination light. Alternatively, the light source 10 may irradiate planar illumination light. The light source 10 irradiates illumination light such as visible light or infrared light, for example. The light source 10 illuminates the sample 20 from an oblique direction, that is, a direction inclined from the Z axis. The illumination light uniformly illuminates a line-shaped region along the Y direction on the surface of the sample 20. On the surface of the sample 20, the length of the illumination light in the Y direction extends over the entire inspection region of the sample 20.

  The illumination light from the light source 10 illuminates a line-shaped or planar area on the surface of the sample 20. Then, the reflected light from the illuminated line-shaped region is detected by the detector 11. The detector 11 is a line sensor camera in which light receiving pixels are arranged in the Y direction. Therefore, since the reflectance changes according to the surface state of the sample 20, the amount of light received by the detector 11 changes.

  The pixel size of the detector 11 is, for example, about several μm to several tens of μm, and the pixel size on the sample 20 is about the same. As the detector 11, a line sensor in which photodiodes such as InGaAs are arranged in a line can be used. The detector 11 and the light source 10 are inclined with respect to the Z direction. In addition, the illumination angle of the illumination light from the light source 10 in the XZ plane and the angle of the detector 11 may be the same or different. Of course, the angle between the light source 10 and the detector 11 may be variable.

  A sample 20 to be inspected is placed on the stage 30. The sample 20 is a substrate such as a semiconductor wafer, a TFT substrate, or a mask, for example. The surface of the sample 20 is orthogonal to the Z direction. The stage 30 is movable in the X direction. The detector 11 detects the reflected light from the area illuminated by the light source 10 while moving the stage 30 in the X direction. Then, detection data corresponding to the luminance of the light detected by the detector 11 is input to the processing device 50. Further, the processing device 50 controls the drive of the stage 30. Then, the processing device 50 visualizes the luminance change of the light detected by the detector 11. By doing so, a reflected image of the entire surface of the sample 20 can be acquired.

  The light source 10 illuminates the entire sample 20 at once in the Y direction. That is, an optical system having a low magnification and a wide field of view is used. As described above, since inspection can be performed with high throughput by the low magnification optical system, inspection at 10 to 150 sheets / h is possible. Since the throughput is high, all wafers can be inspected during the manufacturing process.

  For example, the processing apparatus 50 acquires a reflection image of a semiconductor wafer or the like coated with a resist. Then, based on the reflected image, the resist coating unevenness, particles, foreign matter, and the like are inspected. If film thickness unevenness or particles or foreign matter is detected, the resist is removed before patterning and the resist is applied again. By doing so, the productivity of the semiconductor device can be improved.

  A light source 10 illuminates a semiconductor wafer having a resist coated on the entire surface. Then, the amount of light received by the detector 11 changes according to the film thickness of the resist. Thereby, the film thickness nonuniformity of a resist can be test | inspected. Alternatively, the amount of received light varies depending on particles or foreign matter on the resist. Thereby, particles and foreign matters can be detected. The inspection apparatus performs unevenness inspection (macro inspection) of the semiconductor wafer that is the sample 20. When an abnormality in film thickness or particles or foreign matter is detected, the resist is removed and the resist is applied again. By doing so, the pattern of the semiconductor wafer can be appropriately formed.

  Here, the processing device 50 is an information processing device such as a personal computer, for example, and performs processing on the acquired reflected image. And the defect is detected based on the processing result. For example, the defect is detected by comparing the reflection image data with a threshold value. For example, when the luminance value of the reflected image is between the upper limit value and the lower limit value, it is assumed to be normal. Further, when the luminance value of the sample image is larger than the upper limit value or smaller than the lower limit value, an error of the resist film thickness, or a particle or foreign matter is assumed.

  The configuration of the processing device 50 will be described with reference to FIG. FIG. 2 is a block diagram showing the configuration of the processing device 50. The processing device 50 includes a first storage unit 51, a second storage unit 52, a first filter image generation unit 53, a first difference image generation unit 54, and a second filter image generation unit 55. , A second difference image generation unit 56, a profile data generation unit 57, and a comparison unit 58.

  The first storage unit 51 stores a first FIR filter. The second storage unit 52 stores a second FIR filter. The first filter image generation unit 53 generates a first filter image for the sample image using the first FIR filter. The first difference image generation unit 54 generates a first difference image corresponding to the difference between the sample image and the first filter image.

  The 2nd filter image generation part 55 produces | generates a 2nd filter image with respect to a 1st difference image using a 2nd FIR filter. The second difference image generation unit 56 generates a second difference image corresponding to the difference between the first difference image and the second filter image.

  The profile data generation unit 57 generates profile data in the X direction and profile data in the Y direction from the sample image. The comparison unit 58 compares each of the first filter image, the first difference image, the second difference image, and the profile data with a threshold value. Thereby, the processing apparatus 50 detects a defect. Further, the comparison unit 58 classifies the defect type according to the comparison defect. That is, the defects detected in the first filter image, the first difference image, the second difference image, and the profile data are set as different types of defects. The processing device 50 determines the type of defect by an inspection algorithm based on a defect inspection program stored in advance.

  Here, the types of defects will be described with reference to FIGS. FIG. 3 shows an image of the sample 20 in which the average luminance unevenness, the frame-like unevenness, the dot-like unevenness, the striped unevenness, and the inclined unevenness are all generated. In the images shown in the following drawings, the horizontal direction and the vertical direction are the X direction and the Y direction, respectively.

  FIG. 4 is a diagram illustrating an example of an image in which the average luminance unevenness has occurred. The average luminance unevenness is caused by resist application unevenness over the entire wafer. In the average brightness unevenness, the brightness value of the entire sample image is different from the brightness of the other sample images. Such average luminance unevenness can be extracted by obtaining an average value of sample image data.

  FIG. 5 is a diagram illustrating an example of an image in which frame-like unevenness has occurred. The frame-shaped unevenness is unevenness that occurs along the pixel area. That is, it appears along the edge of the rectangular sample image. In FIG. 5, as shown by the dotted frame, frame-shaped unevenness occurs on the upper and left sides of the sample image. That is, the frame-shaped unevenness is a line-shaped defect that occurs at the end of the sample image.

  FIG. 6 is a diagram illustrating an example of an image in which dot-like unevenness occurs. In FIG. 6, as shown by a dotted line frame, three defects of dot-like unevenness have occurred. The dot-like unevenness is a minute unevenness with high contrast. For example, it occurs due to foreign matter adhering to the sample 20 or lack of application of the resist. A spot-like unevenness is a defect that occurs in a minute region of an image.

  FIG. 7 is a diagram illustrating an example of an image in which striped unevenness occurs. Striped unevenness is caused by uneven brushing or striation when applying a resist. In FIG. 7, as shown by the dotted frame, striped unevenness extending diagonally from the lower right corner to the upper left occurs. Striped unevenness is a line-like defect that occurs in a partial region of an image.

  FIG. 8 is a diagram illustrating an example of an image in which inclined unevenness occurs. The inclined unevenness is low in contrast and is uneven coating over the entire chip. For example, in FIG. 8, the brightness gradually increases from the lower right corner toward the upper left corner as indicated by the arrow. The inclined unevenness is a defect that occurs in the entire image area. In the sample image in which the inclined unevenness has occurred, the luminance gradually changes from one end of the image to the other end.

  FIG. 9 shows a dummy defect image in which such five types of defects occur simultaneously. FIG. 9 is an image having a size of 150 × 120 pixels. FIG. 9 is an image in which five types of defects are intentionally generated by simulation. That is, a dummy defect image is created by combining a dummy image for each defect type and noise. In FIG. 9, each pixel is represented by 8 bits, that is, 256 gradations.

  The dummy defect image shown in FIG. 9 is used as the sample image A acquired by the detector 11 to determine the type of defect. Hereinafter, data processing for determining the type of defect will be described. In the following description, five types of defects are discriminated by performing processing on the sample image A shown in FIG. In each of the images shown below, contrast is emphasized for clarity of explanation.

(FIR filter)
First, an FIR filter used to determine the type of defect will be described with reference to FIGS. 10 and 11 are diagrams for explaining the FIR filter according to the present embodiment. Compared with the image size of the sample image A, an FIR filter that removes a constant term in a narrow region is created. Therefore, as shown in FIG. 10, only the constant term (0th order) is 0, and the other terms (-n-order to -1st order and 1st-order to n-th order) are 1 spectra. When the spectrum shown in FIG. 10 is Fourier-transformed, the FIR filter shown in FIG. 11 is obtained. By using an FIR filter in which only the 0th order term is 0, a component having a large differential value can be extracted in a region smaller than the filter size. The FIR filter shown in FIG. 11 is a first FIR filter. The first storage unit 51 stores a first FIR filter.

  Here, the size of the sample image A to be filtered is 150 × 120. The size of the first FIR filter is 25 × 3 pixels. That is, the first FIR filter is a two-dimensional filter having a size of 25 pixels in the X direction and 3 pixels in the Y direction. The first FIR filter has a longitudinal direction and a lateral direction, and is a directional filter. In the first FIR filter, only one pixel at the center of 25 × 3 is about 0.987, and the other pixels are about −0.013. Of course, the size of the image and the size of the first FIR filter are not particularly limited. The first FIR filter may be a finite size smaller than the size of the sample image A.

  A first FIR filter filters the sample image A. An image generated using the first FIR filter is defined as a filter image. Specifically, for each pixel of the first FIR filter, a product of the data of the first FIR filter and the image data of the corresponding pixel is obtained. Then, the sum of the products of all the 25 × 3 pixels constituting the first FIR filter is obtained. This sum is a value of one pixel of the filter image. A two-dimensional filter image can be obtained by moving the first FIR filter in the X direction and the Y direction.

(Extraction of dotted irregularities)
A process for extracting the dot-like unevenness will be described with reference to FIGS. The first FIR filter is used to detect the spot-like unevenness as a defect. As shown in FIG. 12, the sample image A is filtered by the first FIR filter whose longitudinal direction is the X direction. Here, a filter image when the first FIR filter is used in the X direction is a filter image B.

  Next, as shown in FIG. 13, the sample image A is filtered by the first FIR filter rotated by 90 °. That is, a first FIR filter having a size of 25 pixels in the Y direction and 3 pixels in the X direction is applied to the sample image A. A filter image generated by using the first FIR filter in the Y direction is defined as a filter image C.

  Next, the first FIR filter is used in an oblique direction between the X direction and the Y direction, that is, a direction inclined by + 45 ° from the X axis and a direction inclined by −45 ° from the X axis. As shown in FIG. 14, an image after using the first FIR filter in an oblique direction (+ 45 °) between the X direction and the Y direction is defined as a filter image D. That is, the filter image D is generated by using, as the sample image A, the first FIR filter whose longitudinal direction is a direction inclined by 45 ° from the X axis.

  As shown in FIG. 15, an image after using the first FIR filter in a diagonal direction (−45 °) between the X direction and the Y direction is defined as a filter image E. That is, the filter image E is generated by using, as the sample image A, the first FIR filter whose longitudinal direction is the direction inclined by −45 ° from the X axis. Note that the filter image D and the filter image E are generated by using the first FIR filter in directions different by 90 °.

  In this way, the sample image A is filtered using the first FIR filter in various action directions. Thereby, a plurality of filter images B to E are generated. Here, filter images B to E are generated using the first FIR filter in the X direction, the Y direction, and the oblique direction. That is, the first FIR filter is used in a total of four directions, ie, the X direction, the Y direction, and the two directions between the X direction and the Y direction. Accordingly, four filter images B to E are generated.

  Then, the processing device 50 uses the minimum value filter for the filter images B to E. That is, for each of 150 × 120 pixels, a value that is the minimum value is extracted from the filter images B to E. In this way, a filter image generated using the minimum value filter for the four filter images B to E is defined as a filter image F. The filter image F is as shown in FIG.

  The first filter image generation unit 53 generates filter images B to E using the first FIR filter. Furthermore, the first filter image generation unit 53 generates the second filter image F using the minimum value filter. The comparison unit 58 compares the data of the filter image F with a threshold value. By doing so, dot-like unevenness can be detected as a defect. A point where the data of the filter image F exceeds the upper limit value or a pixel that falls below the lower limit value is detected as a defect.

  A plurality of filter images are generated by using the first FIR filter in various directions. Then, a minimum value filter is used for a plurality of filter images. By doing so, it is possible to extract point-like defects having no directivity from striped unevenness depending on directionality and inclined unevenness. Therefore, it is possible to efficiently classify the point-like defects.

  In the above description, the first FIR filter is used in four directions of two oblique directions orthogonal to the X direction (horizontal direction) and the Y direction (vertical direction). However, the direction of operation using the first FIR filter is used. And the number thereof are not particularly limited. Of course, it is preferable to use the first FIR filter along the arrangement direction of the pixels of the sample image A.

(Striped unevenness extraction)
Next, processing for extracting striped unevenness will be described with reference to FIG. FIG. 17 is a diagram illustrating processing for extracting striped unevenness. When extracting striped unevenness, it is possible to use the filter image used when the dot-shaped unevenness is extracted. In the process of extracting striped unevenness, the sample image A and the filter image F are used. As shown in FIG. 13, the difference between the sample image A and the filter image F is obtained. Here, an image obtained by the difference between the sample image A and the filter image F is referred to as a difference image G. By subtracting the filter image F from the sample image A, dot-like unevenness and high-frequency noise components typified by camera noise are removed.

  Based on the difference image G corresponding to the difference between the sample image A and the filter image F, striped unevenness is detected as a defect. Stripe unevenness can be detected as a defect by comparing the data of the difference image G with a threshold value. That is, a point where the data of the difference image G exceeds the upper limit value, or a pixel that falls below the lower limit value is detected as a defect. The first difference image generation unit 54 generates a difference image G based on the filter image F and the sample image A. And the comparison part 58 can detect striped unevenness by comparing the difference image G and a threshold value.

  Although the difference image G is obtained from the filter image F in the above processing, the difference image G may be calculated using any one of the filter images B to E instead of the filter image F. For example, the difference image G may be calculated using the filter image B instead of the filter image F. Further, the difference image G may be obtained by a value obtained by subtracting the filter image C, the filter image D, or the filter image E from the sample image A. Furthermore, a filter image obtained by applying a minimum value filter to two filter images or three filter images among the filter images B to E may be used. That is, the difference image G only needs to be generated based on one or more filter images generated using the first FIR filter.

(Inclined unevenness)
The process of extracting the inclined unevenness will be described with reference to FIG. In order to extract the inclined unevenness, a second FIR filter having a characteristic different from that of the first FIR filter is used. Here, the size of the second FIR filter is larger than that of the first FIR filter, for example, a size of 51 × 51 pixels. That is, a second FIR filter having a size of 51 pixels in the X direction and the Y direction is used. In the second FIR filter, only one pixel at the center of 51 × 51 is about 0.987, and the other pixels are about −0.013. Here, a square FIR filter that does not depend on directionality is used. Then, the difference image G is filtered using the second FIR filter. The filter image at this time is defined as a filter image H. The filter image H is as shown in FIG. The second filter image generation unit 55 generates a filter image H based on the difference image G. Note that the first FIR filter has a characteristic of passing a higher frequency region than the second FIR filter. That is, the second FIR filter has a characteristic that does not pass higher frequency components than the first FIR filter.

  The filter image H becomes a high-pass filter, and the striped unevenness is extracted by filtering the inclined unevenness. Then, a difference image I corresponding to the difference between the difference image G and the filter image H is calculated. Thereby, the striped unevenness of the filter image H is drawn from the striped unevenness of the difference image G, and only the inclined unevenness can be extracted. The second difference image generation unit 56 generates a difference image I based on the difference image G and the filter image H.

  And the comparison part 58 can detect inclination unevenness by comparing the difference image I and a threshold value. In this way, the second FIR filter having a different size is used for the difference image G. Then, by repeating the difference, the inclined unevenness can be detected.

(Frame unevenness)
Next, processing for extracting frame-like unevenness will be described with reference to FIG. FIG. 20 is a diagram for explaining processing for extracting frame-shaped unevenness. In order to extract the frame-shaped unevenness, the data in the X direction and the Y direction are combined to generate profile data in the Y direction and the X direction. For example, by obtaining an average value of data having the same position in the X direction of the sample image A, one-dimensional profile data in the Y direction is generated. In other words, the profile data in the horizontal direction is obtained based on the sum of the data in one vertical column. Similarly, one-dimensional profile data in the X direction is generated by obtaining an average value of data having the same position in the Y direction of the sample image A.

  As described above, the profile data in the Y direction is obtained by the sum of the data in the X direction, and the profile data in the X direction is obtained by the sum of the data in the Y direction. Then, based on the profile data in the X direction and the Y direction, the frame-shaped unevenness is detected. For example, when a frame-like unevenness in which data increases at the outer peripheral portion of the sample image occurs, the luminance value increases at both ends in the X direction as shown in FIG. Similarly, when a frame-shaped unevenness in which data increases at the outer periphery of the sample image occurs, the luminance value increases at both ends in the Y direction as shown in FIG. Therefore, the frame-shaped unevenness can be detected by comparing the profile data with the threshold value.

  The profile data generation unit 57 generates one-dimensional profile data in the Y direction and the X direction by combining the data in the X direction and the Y direction. If the one-dimensional profile data exceeds the upper limit value or falls below the lower limit value, frame-shaped unevenness is detected. By doing in this way, frame-shaped nonuniformity can be detected simply. Of course, the sum for some pixels may be obtained without obtaining the sum for all the pixels of the sample image A. That is, one-dimensional data in the second direction may be generated by synthesizing data of some pixels in the first direction of the sample image A.

  By using the inspection method described above, it is possible to reliably discriminate between spot-like unevenness, frame-like unevenness, striped unevenness, and inclined unevenness. For example, defects detected based on different images are determined as different types. Furthermore, since each defect is classified by comparing each image with a threshold value, a higher-sensitivity inspection can be performed. That is, a threshold value suitable for each image can be set. Thereby, even if the threshold value is set strictly in order to entrust the sensitivity, it is possible to reliably determine the pseudo defect and the actual defect. Therefore, it is possible to detect a defect with high sensitivity and accuracy. Of course, the order of determination is not particularly limited. For example, you may test | inspect in the procedure which discriminate | determines an inclined unevenness first.

(simulation result)
A simulation result of the above inspection algorithm will be described. Here, the sample image A having a noise level of 7.3 gradations (3 sigma) was used. The dummy defect image is set with 25 gradations for setting the frame-shaped unevenness, 23 gradations for setting the striped unevenness, 40 gradations for setting the dotted unevenness, and 25 gradations for setting the uneven unevenness. Was generated. And the processing apparatus 50 performed said process by making the dummy defect image into the sample image A. FIG. The set luminance is a difference in luminance between the defective pixel and the normal pixel.

  As a result, for the frame-shaped unevenness, the set luminance level was 25 gradations, whereas the detection level was 19 gradations and the pseudo level was 1 gradation. The detection results of the frame-shaped unevenness are shown in FIG. Regarding the dot-like unevenness, the set luminance level is 40 gradations, the detection level is 37 gradations, and the pseudo level is 11 gradations. The detection result of the dot-like unevenness is shown in FIG.

  For the striped unevenness, the set luminance level is 23 gradations, whereas the detection level is 15 gradations and the pseudo level is 5 gradations. With regard to the detection result of the striped unevenness, with respect to the inclined unevenness shown in FIG. 23, the set luminance level is 25 gradations, whereas the detection level is 23 gradations, and the pseudo level is 0 gradations. there were. The detection result of the inclined unevenness is shown in FIG. The detection level is a threshold level at which a defect can be detected, and the pseudo level indicates a threshold level at which a pseudo defect occurs. That is, the difference in luminance between the defective pixel and the normal pixel after the above filter processing or difference processing is performed becomes the detection level.

  As described above, the detection level can be made higher than the pseudo level for any type of defect. Therefore, even if the threshold level is set strictly to increase the sensitivity, detection of pseudo defects can be prevented. Therefore, highly sensitive defect inspection can be performed accurately. Furthermore, the type of defect can be determined. In addition, since defects are classified by different images, even if two or more defects overlap, inspection can be performed accurately.

  The above inspection method can also be applied to a patterned wafer. In the case of a wafer with a pattern, die-to-die processing may be performed as preprocessing. For example, as shown in FIG. 25, sample images A1 and A2 of two wafers having the same pattern are acquired. And the difference image A3 is calculated | required by taking the difference of sample image A1 and sample image A2. And said process is performed with respect to difference image A3 according to a difference. By so doing, it is possible to appropriately determine the defect type even for a patterned wafer. Of course, the sample 20 is not limited to a semiconductor wafer, but can be a photomask, a TFT substrate, or a color filter substrate.

  In the above description, the FIR filter is used in the filter processing, but a filter other than the FIR filter may be used. For example, a digital filter such as an IIR (Infinite Impulse Response) filter can be used. That is, a filter that passes a predetermined frequency band can be used. Of course, a plurality of filters may be combined.

(Inspection method 1)
The generalized inspection procedure will be described below using the flowchart of FIG. FIG. 26 shows a method of repeating the filtering process by the FIR filter and the process of obtaining the difference to obtain the difference image Nmax (Nmax is an integer of 2 or more) times.

  When the detection process is started, a die-to-die process is performed on the two images to extract defects (step S101). In the case of a wafer with a pattern, the sample image Image_Original is obtained by obtaining the difference between the two images. If the wafer is not a patterned wafer, step S101 is omitted. In other words, the image detected by the detector 11 becomes the sample image Image_Original as it is.

Next, after setting n = 1 (n is an integer equal to or greater than 1) (step S102), a first filter image Image_FIR ₁ is generated from the sample image Image_Original (step S103). That is, the sample image Image_Original is filtered by the first FIR filter. Thus, the first filter image Image_FIR ₁ is generated by the first filtering process. In this step S103, as shown in FIGS. 12 to 16, the first filter image Image_FIR ₁ may be generated using the first FIR filter and the minimizing filter in different directions of action. . Of course, it is also possible to generate the _first filter image Image_FIR ₁ by using the first FIR filter only in one action direction. Then, defect enhancement processing is performed on the first filter image Image_FIR ₁ (step S104).

  Then, defect detection is performed using the image Image_Calc0 after the defect enhancement processing in step S104 is performed (step S105). For example, binarization is performed by comparing a threshold value with image data. This binarized data indicates the presence or absence of defects. In step S104, defect enhancement processing using a differential filter or the like is performed. Note that if the filtering process by the first FIR filter also serves as the defect enhancement process, the process of step S104 may be omitted. Further, step S104 and step S105 can be omitted.

Next, the first difference image Image_Sub ₁ is generated by the first difference process (step S106). That is, the first difference image Image_Sub ₁ can be obtained by obtaining the difference between the sample image Image_Original and the first filter image Image_FIR ₁ . Then, defect enhancement processing is performed on the first difference image Image_Sub ₁ (step S107), and then defect detection is performed (step S108). For example, binarization is performed by comparing the data of the image Image_Calc ₁ after the defect enhancement processing with a threshold value. This binarized data indicates the presence or absence of defects. Thereby, the first defect detection after the difference process is performed. Since step S107 and step S108 are the same as step S105 and step S106, description thereof is omitted. Note that if the filter processing by the first FIR filter also serves as defect enhancement processing, the processing in step S107 may be omitted.

Next, n is incremented (step S109). Here, n = 2, but in order to generalize the description, the description will be made with n. the n-th filtering, the filtered image Image_FIR _n of the n is generated (step S110). That is, the (n-1) th difference image Image_Sub _n-1 an FIR filter of the n of, by filtering, the filtered image Image_FIR _n of the n is generated. The nth FIR filter has different characteristics from the (n−1) th FIR filter. Also in step S110, as in step S103, the nth FIR filter may be applied in different action directions, and processing using a minimum value filter may be performed.

Next, the n-th difference image Image_Sub _n is generated by performing the n-th difference process (step S111). That is, by obtaining the difference between the (n−1) th difference image Image_Sub _n−1 and the nth filter image Image_FIR _n , the nth difference image Image_Sub _n can be acquired.

Then, defect enhancement processing is performed on the nth difference image Image_Sub _n (step S112), and then defect detection is performed (step S113). That is, by comparing the data of the image Image_Calc _n after defect emphasizing process is performed with a threshold, performs binarization. This binarized data indicates the presence or absence of defects. Thereby, the nth defect detection after a difference process is performed. Since step S112 and step S113 are the same as step S105 and step S106, description thereof is omitted. Note that when the filtering process also serves as the defect enhancement process by the nth FIR filter, the process of step S112 may be omitted.

  Then, it is determined whether n = Nmax (step S114). When n has not reached Nmax (NO in step S114), the process returns to step S109. That is, n is incremented (step S109), and processing from step S110 is performed. Until N = Nmax is reached (YES in step S114), the processes in steps S109 to S114 are repeated. By doing so, the filter process and the difference process are repeatedly performed Nmax times. That is, the filtering process and the difference process are repeated, and n is sequentially increased to 1, 2, 3,... Nmax. Then, after the difference process, the nth defect is detected. In this way, various types of defects can be detected and classified.

  Here, when the filtering process is performed twice or more, it is desirable to use a filter that transmits a high frequency first. That is, the filter of the Nth filter process has a characteristic that allows a higher frequency component to pass than the filter of the (N + 1) th filter process. By doing in this way, it is possible to detect in order from a small defect. That is, a larger defect is extracted as the number of repetitions in the filter process and the difference process is increased.

(Inspection method 2)
Another inspection method will be described with reference to FIG. FIG. 27 is a flowchart showing another inspection method. In the inspection method shown in FIG. 27, the defect detection is performed by comparing the filter image with the threshold value without comparing the difference image with the threshold value. In addition, about the content which overlaps with said description, it abbreviate | omits suitably.

  When the detection process is started, a die-to-die process is performed on the two images to extract defects (step S201). Thereby, in the case of a wafer with a pattern, the sample image Image_Original is obtained by obtaining a difference between the two images. If the wafer is not a pattern, step S101 is omitted. In other words, the image detected by the detector 11 becomes the sample image Image_Original as it is.

Next, after setting n = 1 (n is an integer equal to or greater than 1) (step S202), a first filter image Image_FIR ₁ is generated from the sample image Image_Original (step S203). That is, the sample image Image_Original is filtered by the first FIR filter. Thus, the first filter image Image_FIR ₁ is generated by the first filtering process. In this step S103, as shown in FIGS. 12 to 16, the first filter image Image_FIR ₁ may be generated using the first FIR filter and the minimizing filter in different directions of action. . Of course, it is also possible to generate the _first filter image Image_FIR ₁ by using the first FIR filter only in one action direction. Then, defect enhancement processing is performed on the first filter image Image_FIR ₁ (step S204).

Then, defect detection is performed using the image Image_Calc ₁ after the defect enhancement processing in step S204 is performed (step S205). Thereby, the first defect detection is performed. For example, a defect can be extracted by comparing a threshold value with image data. In step S204, defect enhancement processing using a differential filter or the like is performed. Note that when the filtering process by the first FIR filter also serves as the defect enhancement process, the process of step S204 may be omitted.

Next, the first difference image Image_Sub ₁ is generated by the first difference process (step S206). That is, the first difference image Image_Sub ₁ can be obtained by obtaining the difference between the sample image Image_Original and the first filter image Image_FIR ₁ . Next, n is incremented (step S207). Here, n = 2, but in order to generalize the description, the description will be made with n. the n-th filtering, the filtered image Image_FIR _n of the n is generated (step S208). That is, the (n-1) th difference image Image_Sub _n-1 an FIR filter of the n of, by filtering, the filtered image Image_FIR _n of the n is generated. The nth FIR filter has different characteristics from the (n−1) th FIR filter. In step S208 as well, similarly to step S203, the nth FIR filter may be applied in different directions of action and processing using a minimum value filter may be performed.

Then, defect enhancement processing is performed on the nth filter image Image_FIR _n (step S209), and then defect detection is performed (step S210). Thereby, the nth defect detection is performed. That is, a defect is detected by comparing the data of the _nth filter image Image_FIR _n that has been subjected to defect enhancement processing with a threshold value. Note that when the filter processing by the first FIR filter also serves as defect enhancement processing, the processing in step S209 may be omitted.

  Then, it is determined whether or not n = Nmax (step S212). When n has not reached Nmax (NO in step S212), the process returns to step S207. That is, n is incremented (step S207), and the processing from step S208 is performed. Until N = Nmax is reached (YES in step S212), the processes in steps S207 to S212 are repeated. By doing so, the filter process and the difference process are repeatedly performed Nmax times. That is, the filtering process and the difference process are repeated, and n is sequentially increased to 1, 2, 3,... Nmax. Then, after the filtering process, n-th defect detection is performed. In this way, various types of defects can be detected and classified.

  Here, when the filtering process is performed twice or more, it is desirable to first use a filter that transmits high frequencies. That is, the filter of the Nth filter process has a characteristic that allows a higher frequency component to pass than the filter of the (N + 1) th filter process. By doing in this way, it is possible to detect in order from a small defect. That is, a larger defect is extracted as the number of repetitions in the filter process and the difference process is increased.

  As described above, in FIG. 27, the filter process and the difference process are repeatedly performed, and the defect is detected based on the difference image and the filter process. That is, a defect can be detected by comparing the filter image with a threshold value. Even if it does in this way, while being able to detect a defect with high sensitivity, the classification of a defect can be discriminate | determined appropriately.

  The control in the processing device 50 may be executed by a computer program. The control program described above can be stored using various types of non-transitory computer readable media and supplied to a computer. Non-transitory computer readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (for example, flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (for example, magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / W, semiconductor memory (for example, mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (Random Access Memory)) are included. The program may also be supplied to the computer by various types of transitory computer readable media. Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. The temporary computer-readable medium can supply the program to the computer via a wired communication path such as an electric wire and an optical fiber, or a wireless communication path.

  In addition to the case where the function of the above-described embodiment is realized by the computer executing the program that realizes the function of the above-described embodiment, this program is not limited to the OS ( The case where the functions of the above-described embodiment are realized in cooperation with the Operating System) or application software is also included in the embodiment of the present invention.

DESCRIPTION OF SYMBOLS 10 Light source 11 Detector 20 Wafer 30 Stage 50 Processing apparatus 51 1st memory | storage part 52 2nd memory | storage part 53 1st filter image generation part 54 1st difference image generation part 55 2nd filter image generation part 56 1st 2 difference image generation unit 57 profile data generation unit 58 comparison unit

Claims (12)

A light source for illuminating the sample; a detector for detecting light from the sample illuminated by the light source;
A processing device for processing the sample image acquired by the detector,
The processing device is
Generating a first filtered image by using a first filter on the sample image;
Generating a first difference image corresponding to a difference between the sample image and the first filter image;
Detecting a first defect based on the first difference image ;
A second filter image is generated by using a second filter having characteristics different from those of the first filter for the first difference image,
Generating a second difference image corresponding to a difference between the first difference image and the second filter image;
A defect inspection apparatus for detecting a second defect based on the second difference image . The processing device is
A filter process for generating an (n + 1) th filter image by using an (n + 1) th filter having a characteristic different from that of an nth (n is an integer of 1 or more) filter for the nth difference image;
Difference processing for generating a (n + 1) th difference image corresponding to a difference between the nth difference image and the (n + 1) th filter image;
A detection process for detecting a (n + 1) th defect based on the (n + 1) th difference image;
The defect inspection apparatus according to claim 1, wherein the (n + 1) -th defect is sequentially detected by repeatedly performing n and sequentially increasing n . The defect inspection apparatus according to claim 2 , wherein the nth filter is an FIR filter. By using the nth filter in different action directions, a plurality of images are generated,
Generating the nth filtered image by using a minimum value filter for the plurality of images;
The defect inspection apparatus according to claim 2, wherein the n-th defect is detected based on the n-th filter image. A defect inspection method for detecting defects from a sample by receiving light from the sample with a detector,
Generating a first filtered image by using a first filter on the sample image acquired by the detector;
Generating a first difference image according to a difference between the sample image and the first filter image;
Detecting a first defect based on the first difference image ;
Generating a second filter image by using a second filter having a characteristic different from that of the first filter for the first difference image;
Generating a second difference image according to a difference between the first difference image and the second filter image;
And a step of detecting a second defect based on the second difference image . A filter process for generating an (n + 1) th filter image by using an (n + 1) th filter having a characteristic different from that of an nth (n is an integer of 1 or more) filter for the nth difference image;
Difference processing for generating a (n + 1) th difference image corresponding to a difference between the nth difference image and the (n + 1) th filter image;
A detection process for detecting a (n + 1) th defect based on the (n + 1) th difference image ;
The defect inspection method according to claim 5 , wherein the (n + 1) -th defect is sequentially detected by repeatedly performing the above and sequentially increasing n . The defect inspection method according to claim 6 , wherein the nth filter is an FIR filter. By using the nth filter in different action directions, a plurality of images are generated,
Using the minimum value filter for the plurality of images to generate the nth filter image;
The defect inspection method according to claim 6, wherein an nth defect is detected based on the nth filter image. A defect inspection program for detecting defects from a sample by receiving light from the sample with a detector,
Against the computer,
Generating a first filtered image by using a first filter on the sample image acquired by the detector;
Generating a first difference image according to a difference between the sample image and the first filter image;
Detecting a first defect based on the first difference image ;
Generating a second filter image by using a second filter having characteristics different from those of the first filter for the first difference image;
Generating a second difference image according to a difference between the first difference image and the second filter image;
Detecting a second defect based on the second difference image;
Defect inspection program to execute. For the computer
A filter process for generating an (n + 1) th filter image by using an (n + 1) th filter having a characteristic different from that of an nth (n is an integer of 1 or more) filter for the nth difference image;
Difference processing for generating an (n + 1) th difference image corresponding to a difference between the nth difference image and the (n + 1) th filter image;
A detection process for detecting a (n + 1) th defect based on the (n + 1) th difference image;
The defect inspection program according to claim 9, wherein the (n + 1) -th defect is sequentially detected by repeatedly executing n and sequentially increasing n . The defect inspection program according to claim 10 , wherein the nth filter is an FIR filter. For the computer
By using the nth filter in different action directions, a plurality of images are generated,
By using a minimum value filter for a plurality of the images, the nth filter image is generated,
The defect inspection program according to claim 10 or 11 , wherein an nth defect is detected based on the nth filter image.