JP4865204B2 - Image processing method, image processing apparatus, and semiconductor inspection apparatus - Google Patents

Description

  The present invention relates to an image processing method and apparatus for recognizing a target pattern from a captured image including random noise.

  In recent years, the performance of imaging devices such as digital cameras, video cameras, and electron microscopes, and the performance of computing devices equipped with microprocessors, FPGAs, ASICs, etc. An image processing apparatus equipped with a function for recognizing a pattern (hereinafter referred to as “pattern matching”) has been put into practical use. For example, a traffic jam monitoring system equipped with a car number recognition function, a security device equipped with a face recognition function, and a semiconductor inspection system equipped with an automatic inspection function for semiconductor wafers.

  Pattern matching is a technique for recognizing (detecting) a specific pattern and its position from an image photographed by an imaging device (hereinafter also referred to as “searched image”). In pattern matching, a pattern to be recognized is registered in advance as a template image. Then, image areas in the image to be searched are sequentially selected with a predetermined size (the same size as the template image), and a matching degree (matching) with the template image is obtained for each selected image area by scanning. In this way, an image with a high degree of coincidence or the position of the image (hereinafter referred to as “matching position”) can be recognized from the searched image.

  Specifically, as shown in FIG. 18, an image region having the same size as the template image 3 (hereinafter referred to as “matching candidate image 2”) is cut out from the searched image 1, and the template image 3 and the matching candidate image 2 are extracted. The brightness value of each pixel is compared, and the degree of coincidence between the template image 3 and the matching candidate image 2 is obtained.

  Normally, the matching candidate image 2 is shifted in units of one pixel from the start point to the end point of the searched image 1, and the matching degree with the template image 3 is obtained for each matching candidate image 2 by scanning. Thereby, the matching candidate image 4 having the highest degree of matching in the searched image 1 is determined. The coordinate position of the matching candidate image in the searched image becomes the matching position.

Pattern matching methods include a luminance difference method that simply compares the luminance between images, or a Hough transform method (Maresca, M) that converts the coordinates of each pixel into a polar coordinate system and recognizes a line figure based on the polar coordinate shape. , Lavin, M., and Li, H., Parallel HoughTransform Algorithms on Polymorphic Torus Ar-Cityure Levidi, S. eds., Multicomputer Vi-sion Academ. normalized normalized correlation method to determine the correlation between the image after (R.L.Lillestrand: T echniques for change detection.IEEETrans.Compute , Vol.c-21, No.7, pp.654-659,1972.) And the like have been proposed. Among them, the normalized correlation method is widely used as a technique that is not affected by “brightness fluctuation between images due to fluctuations in ambient light” that is likely to occur in an actual environment.

  However, all such pattern matching methods have a problem that they are easily affected by random noise included in the image. Random noise is a signal with irregular frequency and amplitude. In an imaging device such as a digital camera that images reflected light from an object with a photoelectric conversion element, random noise enters a captured image due to the influence of a dark current of a photodiode, charge detection amplifier noise, and the like.

  Even in an imaging device such as a scanning electron microscope that irradiates a sample with an electron beam, collects secondary electrons emitted from the irradiated surface of the sample, and amplifies and modulates brightness to form an image. Random noise is mixed in the photographed image due to the influence of noise generated due to excessive thermal motion and noise generated from peripheral circuits.

  In such an image containing random noise, the original pattern information of the subject is lost, and in the pattern matching with the template, the matching / mismatching relationship between the pixels cannot be accurately detected, which is incorrect. It becomes a cause of detecting a matching position.

  As a pattern matching method for reducing random noise included in a photographed image, there is a pattern matching method using a mask as described in Japanese Patent Application Laid-Open No. 2000-342558.

  This technique masks the area other than the pattern information and does not perform the matching calculation in the masked area, thereby reducing the influence of noise included in the area other than the pattern information on the matching result. Is. However, it is very difficult to extract a mask area from an image including random noise, and the influence of noise overlapping with pattern information cannot be suppressed. In addition, there are many techniques for performing pattern matching after performing filtering on a search target image and reducing noises as described in Patent Document 2 (Japanese Patent Laid-Open No. 5-28273) and Patent Document 3 (Japanese Patent Laid-Open No. 2000-260699). Although proposed, it is very difficult to restore a pattern that is missing due to noise.

  As an apparatus that targets an image including random noise for pattern matching, for example, there is a semiconductor inspection apparatus. A semiconductor inspection apparatus specifies an inspection position on a semiconductor wafer from an image of a semiconductor wafer taken with a scanning electron microscope, and inspects a pattern shape and a defect.

  As a pre-inspection process, an inspection position (which may include the vicinity of the inspection position) of a reference semiconductor wafer is photographed with a scanning electron microscope, and the photographed image is registered as a template image. In the inspection process, a semiconductor wafer to be inspected is imaged, and an image position that matches the template image is specified from the captured image by pattern matching, and then the pattern shape and defects are inspected.

  As described above, the scanning electron microscope irradiates a local region of a sample with an electron beam, collects and amplifies secondary electrons emitted from the irradiation surface of the sample, and modulates the luminance on the screen based on the amplification. The secondary electron image is displayed. At this time, the image signal that can be acquired by the electron beam contains a lot of random noise due to its characteristics, so that a clear image cannot be obtained by scanning the subject only once. There is a problem that the subsequent inspection cannot be performed accurately.

  For this reason, a semiconductor wafer is imaged multiple times with the imaging position fixed for the purpose of reducing random noise, and the S / N ratio of the captured image is increased by weighted averaging of the captured image to obtain a clear image. After that, pattern matching is performed to identify and inspect the inspection position. The prior art using the weighted average processing is described in, for example, Japanese Patent Application Laid-Open No. 2002-15692. Random noise appears in different positions and different modes for each captured image. Therefore, if the weighted average is performed, the signal intensity of random noise is weakened, while the normal pattern is enhanced, so that the S / N can be increased.

  In order to reduce random noise, a method using a weighted average is applied not only to a semiconductor inspection apparatus but also to an imaging apparatus equipped with a photoelectric conversion element such as a CCD or a CMOS.

JP 2000-342558 A JP-A-5-28273 JP 2000-260699 A JP 2002-15692 A Maresca, M .; Lavin, M .; , And Li, H .; , Parallel Hull Transform Algorithms on Polymorphic Torus Ar-City Levidi, S. eds. , Multicomputer Vi-sion Academic Press, pp. 9-21) R. L. Lillestrand: Techniques for change detection. IEEE Trans. Computer, Vol. c-21, no. 7, pp. 654-659, 1972.

  Since an increase in the weighted average number leads to an increase in the photographing time, for example, in a semiconductor inspection apparatus, the weighted average number is set small for the purpose of shortening the inspection time. Further, since semiconductors are finely processed, they are easily damaged by an electron beam, and there is an increasing need for inspection with a weighted average number minimized. However, if the weighted average number is reduced, the S / N is reduced, and pattern matching may fail due to the influence of random noise.

  FIG. 9 shows an example of a pattern matching result using the normalized correlation method. Correlation image 404 of image A pattern is obtained by converting the degree of coincidence between each pixel calculated in the process of comparing template image 401 with image A pattern 402 and image B pattern 403 included in the searched image. This is a correlation image 405 of the image B pattern. The sum of the degrees of coincidence of each pixel becomes the correlation value of each pattern image, and a pattern that matches the template image 401 and its matching position can be obtained from the magnitude relationship.

  Pixels with high (bright) luminance values indicate matching with the template image 401, and pixels with low (dark) luminance values indicate mismatching with the template image 401. Since the image A pattern 402 has more pixels with higher luminance values than the image B pattern 403, the position (coordinates) of the image A pattern can be detected as a matching position that matches the template image 401.

  However, the above-described example is a result when noise is not superimposed on the image A pattern (matching candidate A image) 402 and the image B pattern (matching candidate B image) 403. An image in which random noise is superimposed on the image A pattern 402 and the image B pattern 403 is a noise superimposed image A pattern 406 and a noise superimposed image B pattern 407. A correlation image 408 of the noise superimposed image A pattern is obtained by converting the matching degree between each pixel calculated in the process of comparing the template image 401 with the noise superimposed image A pattern 406 and the noise superimposed image B pattern 407 into luminance information. And a correlation image 409 of the noise superimposed image B pattern.

  In the correlation images 408 and 409, pixels having a low degree of coincidence are seen everywhere compared to the correlation image 404 and the correlation image 405 on which no noise is superimposed. This is because the original patterns such as the image A pattern 402 and the image B pattern 403 are lost due to the influence of random noise. In the noise superimposed image B pattern 407, there is a pixel whose degree of coincidence of pixel positions that should not originally coincide is high due to the influence of random noise.

  If random noise is superimposed on the image to be searched in this way, it is impossible to obtain an accurate degree of coincidence. When there are a plurality of matching candidates similar to the template image in the searched image, the position of the matching candidate that is different from the matching candidate that should be originally detected may be detected as the matching position due to the random noise.

  The weighted average processing is effective as such a false detection (false recognition) prevention means. However, when the number of captured images used for the weighted average is reduced, there may be a problem in preventing the erroneous detection.

  An object of the present invention is to perform image processing that can obtain a good pattern matching result even if the number of times of photographing of a photographed image (searched image) including random noise is reduced in weighted average processing and pattern matching. Implementing a method and apparatus.

  In order to achieve the above object, the present invention provides a weighted average process in addition to an image (matching candidate image) that is originally subjected to a weighted average when an image including random noise is subjected to a weighted average process. An image was prepared, and the weighted average processing was performed by adding the weighted average auxiliary image.

  FIG. 1 shows a comparison between the present invention of the pattern matching method with a weighted average process and a conventional image processing method.

FIG. 1B shows a conventional method. In this conventional method, the same image pattern as the search target is registered in advance as a template image. N photographed images 1A-1 N photographed N times from the subject to be inspected as the image to be searched are photographed (step S1), and in the region having the same coordinates among the photographed images among the N photographed images. In addition, N image areas having the same size as the template image are picked up as matching candidate images 2A-2N. The matching candidate images are picked up by sequentially selecting each of the captured images 1A-1N while shifting them in units of one pixel from the start point to the end point. The picked-up matching candidate images 2A- 2N are subjected to weighted average processing (step S2), and then pattern matching between the weighted average matching candidate image 2 'and the template image 3 is executed (step S3).

FIG. 1A shows image processing according to the present invention. The inspected object of the present invention is a product that can be mass-produced, such as a semiconductor wafer. In this case, L (L <N; L is 1 or more) images to be searched 1A-1L are photographed from the subject to be inspected (step S1), and the matching candidate images 2A-2L picked up from the images are searched. A weighted average process is performed by adding M (M ≦ L; M is 1 or more) weighted average auxiliary images 5A-5M registered in advance (step S2) . The weighted average image for the auxiliary are the same size as the object to be inspected, but the same shape, are previously registered shot to the inspection object reference having the same pattern, it is an image containing a pattern to be searched, The template image has the same size, the same coordinates, and the same image pattern. For example, it can be obtained by photographing a subject having the same pattern as the subject used to create the template image and registering it in advance prior to the weighted average. The weighted average auxiliary image can also serve as the template image. Subsequent pattern matching is performed in the same manner as in step S3.

  According to this configuration, even if the number of captured images of matching candidate images is reduced, the weighted average auxiliary image compensates for the reduction. In other words, if an appropriate amount of weighted average auxiliary image is used so as not to be too much, it increases the intensity of the pattern detection signal only when it matches the pattern in the searched image (the pattern of the matching candidate image). Therefore, the S / N ratio can be increased.

  According to the present invention, a good pattern matching result can be obtained even if the number of shots of a shot image (searched image) including random noise is reduced in the weighted average process and the image process for performing pattern matching. . In particular, when pattern-matching a large number of objects to be inspected, the weighted average auxiliary image can be used in common for those objects to be inspected, so that the effect is great.

  First, an outline of an image processing apparatus used in an embodiment of the present invention will be described with reference to FIG.

  As shown in FIG. 2A, the image processing apparatus according to the present invention includes, for example, an image input device 76 such as a video camera and an electron microscope, an information input device 77 such as a keyboard and a mouse, an information processing device 7000, and a CRT. Information output device (display) 78. The information processing device 7000 is the center of the image processing device, and an image input device 76, an information input device 77, and an information output device 78 are connected to the information processing device 7000. An example of the appearance including the information processing device 7000 and peripheral devices is shown in FIG.

  The information processing device 7000 includes a CPU 71, a signal input IF (interface) 72, a RAM 73, a ROM 74, a signal output IF (interface) 75, a pattern matching device 100, and an image data processing unit 79.

  The CPU 71 controls the signal input IF 72, the signal output IF 75, the pattern matching device 100, the image data processing unit 79, the RAM 73, and the like based on a control program stored in the ROM 74. With these controls, the image data from the image input device 76 is input via the IF 72 and stored in the RAM 73.

  The RAM 73 stores data for pattern matching processing and pattern matching results in addition to image data.

  The pattern matching apparatus 100 has a calculation function, performs weighted average processing on data of a matching candidate image of an input search target image (inspected image) and a weighted average auxiliary image, and the weighted averaged image and the template image Is checked. Further, a pattern having the highest degree of coincidence with the pattern of the template image and its position (matching position) are searched from the searched image data through this collation. The pattern matching result is stored in the RAM 73. Specific examples of the weighted average process and pattern matching will be described later.

  Information relating to image processing, for example, information relating to weighted average and pattern matching condition setting, is input from the information input device 77 to the information processing device 7000 via the signal input IF 72 and stored in the RAM 703.

  The pattern matching result is output to the information output device 78 via the signal output IF 75.

  Note that the pattern matching apparatus 100 is hardware such as an ASIC or FPGA configured by blocks described in the embodiments described later.

  The signal input IF 72 is an interface such as USB, IEEE 1394, Centronics, memory card, PCI, Ethernet (registered trademark). The image data from the image input device 76 and the data from the information input device 77 are input to the information processing device 7000.

  The RAM 73 is an SDRAM, SRAM, DRAM, memory card, hard disk, or the like, and accumulates image data and stores parameters used for pattern matching.

  The ROM 74 is configured by a flash ROM or the like, and stores a control program used by the CPU 71 and the like.

  The parameters used for pattern matching may be stored in the ROM 73.

  The signal output IF 75 is an interface such as USB, IEEE 1394, Centronics, memory card, PCI, Ethernet, and the like, and outputs a pattern matching result to the signal output device 78.

  The image data processing unit 79 uses the image stored in the RAM 73 as an input, performs image processing to reduce information such as random noise and luminance unevenness included in the image, and stores the processing result image in the RAM 73. To do. The image data processing unit 79 includes image processing hardware such as an FPGA and an ASIC that is activated, executed, and terminated under the control of the CPU 71. The image data processing unit 79 does not necessarily need to be hardware, and the image processing function that reduces information such as random noise and luminance unevenness is programmed and stored in the RAM 73 and the ROM 74. Software processing can also be performed under the control of the CPU 71. For example, the configuration may be such that random noise reduction processing is performed by software processing, and luminance unevenness is reduced by hardware processing. Further, the function of the image processing unit 79 is not limited.

  In FIG. 2A, the function of the pattern matching apparatus 100 is configured by hardware, but can also be configured by a software program. In FIG. 2B, such a software program is installed in the information processing apparatus 7001. In FIG. 2B, the same reference numerals as those in FIG. 2A indicate the same or common elements as the constituent elements in FIG.

  2B, the information processing apparatus 7001 includes a CPU 71, a signal input IF 72, a memory (RAM 73, ROM 74), a signal output IF 75, and an image data processing unit 79. In the present embodiment, a function corresponding to the pattern matching apparatus 100 in FIG. 2A is executed by software processing of the CPU 71.

  Since the signal input IF 72, the signal output IF 75, and the image data processing unit 79 are the same as those shown in FIG. 2A, the description thereof will be omitted, and the CPU 71, RAM 73, and RM 74 will be specifically described here.

  A program for executing the functions of the pattern matching apparatus 100 in FIG. 2A by software processing is stored in the RAM 73 or the ROM 74, and the CPU 71 executes the pattern matching program. Parameters used for pattern matching may be stored in the ROM 73.

  The pattern matching apparatus 100 of the present invention can be realized by hardware as described above, or can be realized by software. It is also possible to realize some functions of the pattern matching apparatus by software and realize the remaining functions by hardware.

Hereinafter, an example in which the pattern matching apparatus 100 according to the present invention is realized by hardware will be described based on Examples 1 and 2, and an example in which the pattern matching apparatus 100 is realized by software will be described by Example 3. An example in which the image processing technique of the present invention is applied to a semiconductor inspection apparatus will be described with reference to a fourth embodiment.
[Example 1]
FIG. 3 is a block diagram showing the basic configuration of the pattern matching apparatus 100 used in the present invention. The relationship among the pattern matching apparatus 100, the CPU 71, and the RAM 73 is shown.

  The RAM 73 captures an image of a subject on which random noise is superimposed (searched image 12; corresponding to the searched image 2A-2L in FIG. 1), template image 10 (corresponding to the template image 3 in FIG. 1), weighted average. The auxiliary image 11 (corresponding to the weighted average auxiliary image 5A-5M in FIG. 1) is stored. The purpose of the pattern matching apparatus 100 is to find a pattern that matches the template image 10 from the searched images 12 on which random noise is superimposed, and to detect a position (matching position) 17 where the pattern exists. Hereinafter, each component will be described.

  As described above, the RAM 73 stores image data (template image 10, weighted average auxiliary image 11, and searched image 12) used for pattern matching. The CPU 71 functions as a pattern matching control unit, and controls activation and termination of the pattern matching apparatus 100, input of image data and parameters to the pattern matching apparatus 100, and output of a matching result (matching position 17).

  The pattern matching procedure is shown below.

  The pattern matching apparatus 100 reads the searched image 12 from the RAM 73 after activation. A plurality of searched images 12 are acquired. That is, the searched image 12 is obtained by photographing the same subject a plurality of times at the same position.

  A matching candidate image 18 (corresponding to the matching candidate images 2A-2L in FIG. 1) to be compared with the template image 10 is cut out from each searched image 12 and input to the weighted average means 14.

  The weighted average auxiliary image 11 is added to the matching candidate image 18 to perform a weighted average process, and an image 18 ′ (corresponding to the image 2 ′ in FIG. 1) after the weighted average is input to the pattern matching unit 15.

  The pattern matching unit 15 performs pattern matching between the template image 10 and the weighted average image 18 ′, and inputs a matching value (image pattern matching degree) 19 to the matching value comparison unit 16. The matching value comparison unit 16 detects a matching candidate having a large matching value and outputs a matching result 17.

  The above-described extraction of the matching candidate image 18 is performed by extracting an image having the same size as the template 10 from the searched image 12 and is basically performed sequentially from the start point to the end point of the searched image 12. The weighted average auxiliary image 11 and the template image 10 are fixed in the pattern matching process from the start point to the end point of the searched image 12. Here, the weighted average auxiliary image 11 is an image that is used to assist image data required for the weighted average of the matching candidate images and includes the same pattern as the template image 10. The template image 10 and the weighted average auxiliary image 11 are registered in the RAM 73 by the designation by the information input device 77 from the images photographed by the image input device 76 as a pre-process of pattern mung.

  Registration of the template image 10 and the weighted average auxiliary image 11 can be realized by using a template registration program 90 shown in FIG.

  The template registration procedure will be described with reference to FIGS.

  The template registration program 90 is stored in the RAM 73 or the ROM 74, and the CPU 71 executes the program.

  When the template registration program 90 is executed, if the photographed image 80 serving as the template is not stored in the RAM 73, the subject serving as the template image is photographed by the image input device 76 (step 900) and stored in the RAM 73.

  Next, the captured image 80 stored in the RAM 73 is displayed on the information output device (display) 78 (step 901).

  As shown in FIG. 12, the operator designates the coordinates of an image area including a desired pattern in the captured image 80 using the information input device 77 while viewing the screen of the information output device 78. In response to the designation of the coordinate information, the CPU 71 registers the image area 81 in the coordinate range as the template image 10 (step 902). Next, the weighted average auxiliary image 11 is registered (step 905). The weighted average auxiliary image 11 is an image including the same pattern as the template image 10.

  Here, the pattern refers to luminance information indicating the shape of the subject. That is, when an optical imaging device is used for imaging a subject, the intensity of reflected light or transmitted light from the subject is luminance information indicating the shape of the subject, and an electron microscope is used for imaging the subject. When used, luminance information indicating the shape of the subject is obtained by collecting and amplifying secondary electrons and reflected electrons emitted from the subject.

  Normally, random noise is superimposed on luminance information indicating the shape of the subject during the shooting process, but the image registered as the template image 10 and the weighted average auxiliary image 11 is not affected by the random noise, and the shape of the subject is changed. An image having luminance information that accurately indicates it is desirable.

  From the above, the template image 10 may be registered as the weighted average auxiliary image 11, or the same pattern as the template image 10 may be cut out and registered from a captured image other than the captured image 80 from which the template image 10 is extracted.

  In this example, the subject to be registered as the weighted average auxiliary image 11 is photographed (step 903), stored in the RAM 73, and then the photographed image is displayed on the information output device 78 (step 904). After confirmation, the image is registered as the weighted average auxiliary image 11.

  When the template image 10 is used as the weighted average auxiliary image 11, steps 903 and 904 are omitted.

  As described above, the weighted average means 14 performs the weighted average processing by adding the weighted average auxiliary image 11 to the matching candidate image 18.

  The weighted average is obtained by multiplying the luminance value of each pixel of a plurality of images to be weighted (in this embodiment, the matching candidate image 18 and the weighted average auxiliary image 11) by a weighting coefficient, thereby weighting the plurality of images. An average image (hereinafter referred to as a weighted average image) is obtained and can be expressed by Equation 1. Although the weighting coefficient k1 = k2 = 0.5 is a normal value, this is not limited.

Dst (i, j) = k1 × Img1 (i, j) + k2 × Img2 (i, j)...
k1 + k2 = 1.0
Dst (i, j): Weighted average processed image Img1 (i, j): Matching candidate image Img2 (i, j): Weighted average auxiliary image
Here, (i, j) are image coordinates. As is apparent from Equation 1, when the weighted average auxiliary image 11 is added to the matching candidate image 18 and the weighted average processing is performed, the image coordinates (i, j) of both are matched.
FIG. 5 shows a hardware configuration of the weighted average means 14, and a multiplier 141 for multiplying the luminance value of each pixel of the matching candidate image Img1 (i, j) by a weighting coefficient k1, and a weighted average auxiliary image Img2. A multiplier 142 for multiplying the luminance value of each pixel of (i, j) by a weighting coefficient k2, and an adder 143 for adding the luminance value of each pixel of the weighted matching candidate image and the weighted average auxiliary image. Is done.

  The effect of the weighted average will be described using the luminance profile of FIG.

  The luminance profile represents the luminance value of the pixel at each coordinate on the vertical axis with the coordinates on the image as the horizontal axis. A broken line 1 indicates a pattern (image signal) on which random noise is not superimposed, a broken line 2 indicates random noise, and a solid line indicates an image signal on which random noise is superimposed.

  FIG. 8 shows the luminance file 301 of the template image 10, the luminance file 302 of the matching candidate A image 18-1 in the searched image, the luminance file 303 of the matching candidate B image 18-2, and the luminance of the weighted average auxiliary image 11. A file 304 is illustrated. Among the two matching candidate images, the luminance file 302 is a luminance file at a matching position in the searched image (that is, an image at a position where the same pattern as the template image exists). The luminance file 303 is a luminance file at a non-matching position in the searched image. Note that each of the matching candidate images 18-1 and 18-2 is actually cut out from a plurality of search target images photographed a plurality of times for the same subject at the same position. There are a plurality of 1 and 18-2, but here, for convenience, only one of each is illustrated as a representative.

  The matching candidate images 18-1 and 18-2 are both affected by random noise, and the pattern is missing. If only one image is pattern-matched with the template image, incorrect matching occurs in the matching process. The degree may be detected and mismatched.

  In order to eliminate such inconvenience, a plurality of matching candidate images are acquired and weighted averaged by the weighted averaging means 14, but in this embodiment, each matching image is reduced in order to reduce the number of times the searched image is captured. The weighted average auxiliary image 11 is added to the candidate image to perform a weighted average process. That is, the weighted average processing of the matching candidate A image 18-1 and the weighted average auxiliary image 11 and the weighted average processing of the matching candidate B image 18-2 and the weighted average auxiliary image 11 are performed. Here, if the number of weighted average auxiliary images is too large compared to the matching candidate images, the image data after the weighted average of each matching candidate image will be similar, so it is necessary to limit to an appropriate amount. Considering the above, the number of matching candidate images (the number of searched images) L and the number M of weighted average auxiliary images have a relationship of L> M.

  According to such weighted average processing, random noise is reduced for the matching candidate A image 18-1, and the addition of the weighted average auxiliary image 11 makes it clearer the pattern to be matched with the template image. I can plan. The luminance file of the image 18′-1 after the weighted average in this case is indicated by 305 in FIG.

  On the other hand, for the matching candidate B image 18-2 (an image that does not match the template image 11), the pattern of the weighted average auxiliary image 304 is included in the ratio of the weighting coefficient k2 in the image 18′-2 after the weighted average. It will be. For example, when k1 = 0.5 and k2 = 0.5, an image including 1/2 of the weighted average auxiliary image 11 pattern is obtained. In such a case, the matching candidate image 18′-2 after the weighted average tends to be similar to the template. However, random noise is reduced by the weighted average. Further, even if the weighted average auxiliary image 11 having the same pattern as the template image is added to the matching candidate B image 18-2 at the non-matching position, it is a small number of the weighted average number of processed images in each image. To dilute the pattern. Therefore, original pattern information can be extracted well. Reference numeral 306 denotes a luminance file of the image 18′-2. For this reason, it is possible to detect the difference in matching degree between the matching candidates by reflecting the pattern difference 30 after the weighted average processing in the matching result.

  In FIG. 3, the pattern matching means 15 includes a template image 10 (corresponding to the luminance file 301 in FIG. 8) and a weighted average image 18 ′ (corresponding to the luminance files of the weighted average images 18′-1 and 18′-2 in FIG. 8). Pattern matching is performed, and the degree of coincidence (hereinafter referred to as matching value 19) is measured. The pattern matching technique may be any technique that can determine the degree of coincidence based on the pattern information difference 30 of each weighted average image 18 ′, and the specific technique is not limited. That is, any pattern matching method may be used as long as the degree of coincidence is measured by using a difference between a pattern that has been changed by the weighted average (mismatched matching candidate) and a pattern that has not been changed by the weighted average (matching matching candidate).

  Examples of the pattern matching method as described above include a luminance difference method for simply comparing luminance between images, and a normalized correlation method for obtaining correlation between images after normalization using the standard deviation of images. is there.

The matching value comparison means 16 detects the highest matching value from the matching values 19 of each matching candidate image, and matches the image position (positional coordinate) in the searched image of the matching candidate image having the matching value. Output as position 17.
[Example 2]
FIG. 4 is a block diagram of the pattern matching apparatus 100 according to the second embodiment of the present invention.

  In the present embodiment, the same reference numerals as those in the first embodiment indicate the same or common elements as those used in the first embodiment.

  In this embodiment, the difference from the first embodiment is that an image conversion means 20 is provided after the weighted average means 14.

  The image conversion means 20 generates an image 21 (herein referred to as “weighted average converted image”) obtained by separating the pattern and the random noise component from the weighted average image 18 ′. There are various methods for separating a pattern and random noise, such as a method using a Fourier transform and a method using a filter coefficient, but the method is not limited as long as it has a noise reduction effect.

  An example of image conversion is shown using FIG.

  For the matching candidate A image 18-1 and the matching candidate B image 18-2, an image obtained by weighted averaging the weighted average auxiliary image 11 with a weighting coefficient of k1 = 0.5 and k2 = 0.5 is a weighted average A image 18′-1. And a weighted average B image 18'-2. An image obtained by applying the edge extraction filter shown in Expression 2 to the weighted average A image 18′-1 and the weighted average B image 18′-2 using the image conversion unit 20 is obtained as a weighted average A converted image 509 and a weighted average. This is a B-converted image 510.

  The images denoted by reference numerals 504 and 505 are comparative examples for the present embodiment, and they are directly expressed by Equation 2 without performing weighted averaging on the matching candidate A image 18-1 and the matching candidate B image 18-2, respectively. It is the image which performed the filter process shown in FIG. It can be seen that the edge component and the noise component are well separated in the image subjected to the weighted average as compared with the image not subjected to the weighted average. By applying such image conversion means 20, pattern matching can be performed with an image in which the influence of random noise is further reduced.

i, j: Image coordinates
Dst2 (i, j): Weighted average processing conversion image
Dst (i, j): Weighted average processed image
Fx (i, j): Horizontal edge component
Fy (i, j): Vertical edge component In Expression 2, Fx (i, j) is obtained as an edge where there is a gradient of light and shade between adjacent pixels in the horizontal direction of the image, and Fy (i, j ) Is obtained as an edge where there is a gradient of light and shade between adjacent pixels in the vertical direction of the image.

  FIG. 6 is an example showing a hardware configuration of the image conversion means 20 (edge extraction filter) for executing the above expression 2, and subtracters 151 and 152, adders 153, 154, 155, 156, 157, 158, Multipliers 159 and 160, subtracters 161 and 162, square calculators 163 and 164, an adder 165, and a root calculator 166.

FIG. 7 (a) shows image coordinate data to which the above equation 2 is applied, and FIG. 7 (b) shows coefficients applied to each pixel when the horizontal edge component is obtained (when image conversion is performed). Yes, FIG. (C) shows coefficients applied to each pixel when the edge component in the vertical direction is obtained for the weighted average processing conversion (when image conversion is performed).
Example 3
FIG. 16 is a flowchart of a pattern matching program for executing the functions of the pattern matching apparatus 100 of the present invention shown in the first and second embodiments by software.

  The pattern matching program is stored in the RAM 703 and the ROM 704 of the information processing apparatus 7001 shown in FIG. 2B, and the CPU 701 that has received an instruction from the information input apparatus 706 performs activation, execution, and termination necessary for pattern matching. Control is performed.

After starting the pattern matching program, the template image 10 is registered using the image stored in the RAM 703 (step 1201), and the weighted average auxiliary image 11 is registered (step 1202). The matching candidate image 18 is cut out from the searched image 12 stored in the RAM 703, and the weighted average auxiliary image 11 and the template image 10 are weighted averaged to generate a weighted average image 18 '(step 1203). Pattern matching with the template image 10 is performed on the weighted average image 18 ′ or the weighted average converted image 21 obtained by image conversion of the weighted average image 18 ′ (step 1204) (step 1205). When the pattern matching is completed up to the end point of the searched image 12, the image position of the matching candidate having the highest matching value is detected (step 1207), and is output to the RAM 703 and the information output device 708 (step 1208). These weighted average processing, image conversion processing, and pattern matching are performed by the calculation of the CPU 701.
Example 4
FIG. 14 is a block diagram in which the image processing apparatus (information processing apparatus) 7000 shown in FIG. 2 is applied to a semiconductor inspection apparatus 1001.

  The apparatus 1001 of this example inspects a semiconductor wafer using an image taken with a scanning electron microscope 76. The CPU 71, the signal input IF 72, the RAM 73, the ROM 74, the signal output IF 75, the image data processing unit 79, the information output device 77, and the pattern matching device 100, which are constituent elements of the block diagram, are the image processing device 7000 shown in FIG. Since it is equivalent to the component of, the description is omitted.

  Depending on the cost and required performance of the semiconductor inspection apparatus 1001, the pattern matching apparatus 100 of this example may be configured by the hardware described in the first and second embodiments, or may be configured by the software described in the third embodiment. Alternatively, a part may be constituted by hardware and the rest may be constituted by software.

  The entire inspection process is controlled by the CPU 71 using an inspection control program stored in the RAM 73 and the ROM 74. The processing flow of the inspection control program is shown in FIG.

  As a pre-inspection process, the template registration program shown in the first embodiment is started to register the inspection pattern on the semiconductor wafer, or the inspection pattern and its neighboring pattern as the template image 10, and include the same pattern as the template image 10. The weighted average auxiliary image 11 is registered (step 1301). The creation of the template image 10 and the weighted average auxiliary image 11 has been described above and will be omitted. Subsequently, the inspection position on the semiconductor wafer is registered (step 1302).

  FIG. 15 shows the relationship between the semiconductor wafer 1101, the image 1102 obtained by photographing a predetermined position on the semiconductor wafer 1101 with the scanning electron microscope 76, the template image 1103 registered by the template registration program, and the index 1104 for specifying the inspection position. . The inspection position can be specified by operating the index 1104 by the information input device 77 at an arbitrary position on the image 1102 displayed on the information output device 77.

  After step 1302, an inspection method such as a scanning electron microscope scanning method, length measurement inspection, and defect inspection is registered (step 1303).

  After that, an inspection process enters. In the inspection process, a semiconductor wafer to be inspected is photographed with an electron microscope (step 1304), and a photographed image (corresponding to the searched image 12 described above) is stored in the RAM 73. Data relating to a matching candidate image from the template 10, the weighted average auxiliary image 11, and the searched image 12 is input to the pattern matching apparatus 100, and a matching position is obtained by performing weighted average processing and pattern matching (step 1305). .

  Step 1305 includes execution 1204 of image conversion shown in FIG. Subsequently, since the coordinates of the matching position can be known, the inspection position 1104 is specified in the image of the matching position, the inspection processing apparatus 1002 is activated, and the inspection designated by the inspection method registration 1303 is performed (step 1306). Thereafter, the inspection result is output to the RAM 73 and the information output device 77. As soon as the wafer inspection is completed, the inspection control program is terminated. Note that the inspection processing apparatus 1002 may be configured by hardware such as FPGA and ASIC, like the image processing unit 79, or the function of the inspection processing apparatus 1102 is programmed and registered in the RAM 73 and the ROM 74, and the CPU 71 You may comprise with the software started, executed, and complete | finished by control of this.

(A) is the principle figure regarding the image processing of this invention, (b) is the principle figure of the image processing of a conventional system. (A) is a block diagram showing an information processing apparatus (image processing apparatus) equipped with the pattern matching apparatus according to the first embodiment of the present invention, and (b) is a block diagram of the information processing apparatus of the second embodiment. 1 is a block diagram showing a pattern matching device used in Embodiment 1. FIG. FIG. 5 is a block diagram showing a pattern matching device used in Embodiment 2. FIG. 3 is a diagram illustrating a configuration of a weighted average processing unit according to the first embodiment. FIG. 6 is a diagram illustrating a configuration of a weighted average processing unit according to the second embodiment. FIG. 6 is a diagram illustrating data used for a weighted average process according to the second embodiment. The figure which showed the brightness | luminance profile of the image utilized with the pattern matching apparatus of this invention. The figure which showed the image utilized with the pattern matching apparatus of this invention. The figure which showed the image utilized with the pattern matching apparatus of this invention. The figure which showed the external appearance of the information processing apparatus and peripheral device which mount the pattern matching apparatus of this invention. The figure which showed the image utilized with the external appearance of an information output device, and the pattern matching apparatus of this invention. The flowchart of a template registration program. 1 is a block diagram showing a semiconductor inspection apparatus equipped with an image processing apparatus of the present invention. The figure which showed the relationship between a semiconductor wafer, a picked-up image, and a template image. The flowchart which shows the function of the pattern matching apparatus of this invention. The flowchart which shows the semiconductor test process including the function of the pattern matching apparatus of this invention. The figure which showed the procedure of the pattern matching using a template.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Template image, 11 ... Weighted average auxiliary image, 12 ... Searched image, 14 ... Weighted average processing means, 15 ... Pattern matching means, 16 ... Matching value comparison means, 17 ... Matching position, 18 ... Matching candidate image, 18 '... weighted average image, 20 ... image conversion means, 21 ... weighted average converted image, 100 ... pattern matching device, 301 ... brightness profile of template image, 302 ... brightness profile of matching candidate A image, 303 ... matching candidate B image Luminance profile of the weighted average auxiliary image, 305... The luminance profile of the weighted average A image, 306... The luminance profile of the weighted average B image, 30. 402 ... Matching candidate A image 403 Matching candidate B image 404 ... Correlation image of matching candidate A image 405 ... Correlation image of matching candidate B image 406 ... Matching candidate A image with superimposed noise 407 ... Matching candidate B image with superimposed noise 408 ... Noise ,...,...,...,... CPU, 72... Signal input IF, 73... RAM, 74... ROM, 75. ... Image input device, 77 ... Information input device, 78 ... Information output device, 79 ... Image processing unit, 7000 ... Information processing device, 701 ... CPU, 702 ... Signal input IF, 703 ... RAM, 704 ... ROM, 705 ... Signal Output IF, 706 ... Image input device, 707 ... Information input device, 708 ... Information output device, 709 ... Image processing unit, 700 DESCRIPTION OF SYMBOLS 1 ... Information processing apparatus, 80 ... Photographed image, 81 ... Template image, weighted average auxiliary image, 1001 ... Semiconductor inspection apparatus, 1002 ... Inspection processing section, 1101 ... Semiconductor wafer, 1102 ... Photographed image of semiconductor wafer, 1103 ... Template image, 1104 ... inspection position.

Claims (8)

In order to detect an image pattern to be searched from among the captured images of the inspection target product including random noise, the same image pattern as the search target is registered as a template image in advance, and the captured image of the inspection target product is the same pattern plurality acquired by performing a plurality of times of photographing by position, by weighted average processing the plural photographed images, which creates an object search image, with respect to the object to be searched image, using the template image In an image processing method for performing matching ,
As an image for assisting the weighted average processing, a subject to be an inspection reference having the same size, shape, and pattern as the inspection target product is previously photographed, and the same size as the template image from the photographed image of the reference subject, register extracts an image for the weighted average aid having the same image pattern, the weighted average processing, a and the template image in the area having the same coordinates of the captured image with each other from the front Kifuku number of captured images An image area of the same size is selected by shifting from the start point to the end point by one pixel unit, and the weighted average auxiliary image is added to the selected image area , and the image subjected to the weighted average process is performed. An image processing method characterized by pattern-matching the template image with the template image.   The image processing method according to claim 1, wherein the template image also serves as the weighted average auxiliary image.   The image processing method according to claim 1, wherein the image processing method is used for inspecting a surface state of a semiconductor wafer that is an inspection object, and the captured image including the random noise is obtained by imaging the surface of the inspection object a plurality of times. A plurality of obtained captured images, wherein the weighted average auxiliary image is registered in advance by capturing a reference subject having the same pattern as the inspection object prior to inspection, and includes the random noise. An image processing method in which the number L of photographed images of the object to be inspected and the number M of the weighted average auxiliary images have a relationship of L> M. Means for previously registering the same image pattern as the search target as a template image in order to detect the image pattern to be searched from among the captured images of the inspection target product including random noise;
The captured image of the inspected products plurality obtained by a plurality of times of photographing, by a weighted averaging process the plural photographed image to create an object search image, with respect to the object to be searched image, the template image an image processing apparatus comprising means for performing a pattern matching using the,
An image region having the same size and the same image pattern as the template image is extracted from an image obtained by photographing a subject as an inspection reference having the same size, shape and pattern as the inspection target product, and the weighted averaging process is assisted. Means for registering as a weighted average auxiliary image for
Select while shifting by one pixel and an image area of the template image and the same size from the start point in the region to an end point having the same coordinates of the photographing between from the front Kifuku number of captured images, the selected image area respect, the image processing apparatus characterized by comprising, means for performing a weighted averaging process also added image for the weighted average aid.   5. The image processing apparatus according to claim 4, further comprising means for reducing random noise from the weighted averaged image.   5. The display according to claim 4, wherein a display for displaying the image of the subject to be imaged as an inspection standard, information input means for designating the template image and the weighted average auxiliary image from the image displayed on the display, An image processing apparatus comprising: a data registration unit that registers the template image and the weighted average auxiliary image in a data storage unit based on an instruction from the information input unit. In a semiconductor inspection apparatus for inspecting a semiconductor by inputting surface image data of a semiconductor to be inspected obtained by photographing with a scanning electron microscope,
As a means for processing semiconductor surface image data, a plurality of photographic images of a semiconductor to be inspected are obtained, subjected to weighted average processing, and the weighted average processed image and the template image are subjected to pattern matching so that the template matches the pattern. An image processing device for detecting a matching position in an image, and means for inspecting a pattern of matching positions on a semiconductor surface,
7. The semiconductor inspection apparatus according to claim 4, wherein the image processing apparatus is the image processing apparatus according to any one of claims 4 to 6.   8. The brightness unevenness of the photographic image according to claim 7, the means for controlling the semiconductor inspection process, the data storage means for storing a photographed image, template image, weighted average auxiliary image, etc. from the scanning electron microscope. Means for performing image processing such as removal, an information input device for designating the template image, the weighted average image, and the inspection position in the photographed image from the scanning electron microscope, and receiving instructions from the information input device A means for registering the template image and the weighted average image in a data storage means; an information output device for displaying a photographed image of the scanning electron microscope, a template image, a weighted average auxiliary image, a pattern matching result, and the like; A semiconductor inspection apparatus comprising:

Images