JP2007078511A - Mark detector - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reliably detect a mark on an object to be inspected by accurately verifying a registration image with a verification image even when noise is mixed in the verification image to some extent. <P>SOLUTION: Fourier transform is performed on a registration image GR to acquire Fourier image data FR, and Fourier transform is performed on a verification image GC to acquire Fourier image data FC. FR and FC are synthesized to suppress its amplitude to acquire amplitude-suppressed synthetic Fourier image data FCRr. Complex number components of FCRr are each multiplied by a coefficient proportional to an amplitude component of FR, Fourier transform is further performed to acquired correlation data DCR. The registration image GR is collated with the verification image GC on the basis of the intensity of a correlation component of every individual pixel of a correlation component area SCR determined on the correlation data DCR. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a mark detection device that detects a positioning mark or the like attached to an inspection object such as a semiconductor wafer or a glass substrate.

  The present applicant has previously proposed a pattern matching apparatus for matching patterns such as fingerprints based on spatial frequency characteristics (see, for example, Patent Document 1). In this pattern matching device, the matching image is subjected to Fourier transform to generate matching Fourier pattern data. Then, the registered Fourier pattern data of the registered image created by performing the same processing as the verification Fourier pattern data is synthesized, and after performing amplitude suppression processing (log processing, etc.) on the synthesized Fourier pattern data. , Fourier transform is performed to obtain correlation data.

  Although this correlation data is created based on data in which the amplitude in the frequency space is suppressed, it can be considered as data obtained by convolving the collation image and the registration image. It represents a correlation. The registered image and the collation image are collated based on the intensity of the correlation component for each pixel in the predetermined correlation component area determined on the correlation data. This matching algorithm is called amplitude suppression correlation method.

  In this pattern matching apparatus, as one form of the amplitude suppression correlation method, the matching Fourier pattern data and the registered Fourier pattern data are expressed by complex numbers based on the amplitude and the phase, respectively, and the phase only correlation method in which all the amplitude information is set to 1. (Phase-Only Correlation: POC). This phase-only correlation method does not require amplitude information in the Fourier transform data, and the same algorithm can be applied regardless of the type of collation pattern, and even if an illuminance difference or misalignment occurs during registration / verification It has the advantage that it is less affected by accuracy and boasts higher collation accuracy than other collation algorithms.

  This pattern matching device is used, for example, for automatic positioning of a semiconductor wafer, a glass substrate or the like in a semiconductor or liquid crystal manufacturing process. In this case, a positioning mark is added as an inspection mark by marking or printing on an inspection target such as a semiconductor wafer or a glass substrate. The inspection mark attached to the inspection object is imaged at a predetermined position, and a collation image (“inspection mark” + “background”) including the imaged inspection mark and a registered image including an inspection mark registered in advance (“ The registered inspection mark "+" background ") is collated using the phase only correlation method. Based on the collation result, the inspection mark attached to the inspection target is detected, and the position of the inspection target is adjusted so that the positional deviation of the inspection mark becomes zero. Note that the displacement of the inspection mark attached to the inspection object is detected based on the position of the correlation peak having the highest intensity among the correlation components for each pixel in the correlation component area in the correlation data.

Japanese Patent Laid-Open No. 9-22406 Introduction to computer image processing (Nippon Industrial Technology Center, published by Soken Publishing Co., Ltd., pages 44-45)

  However, in the manufacturing process of semiconductors and liquid crystals, inspection targets such as semiconductors and glass substrates undergo various chemical processes, so the inspection marks that are initially marked are also deformed, the contrast is reduced, and noise such as large amplitude is superimposed. Receive. Typical examples include a mark formed by transparent electrode (ITO) printing in a liquid crystal manufacturing process, a mark viewed through an opaque conductive sheet (ACF), and the like.

  In general, inspection marks are often simple geometric shapes (cross, star, polygon, etc.). In a simple figure such as an inspection mark, the amplitude (energy) when Fourier transformed is biased to a specific frequency component. For this reason, it can be said that an inspection mark having a unique frequency characteristic is used.

  On the other hand, since noise generally has a wide range of frequency components, when noise is superimposed on an image including an inspection mark, the phase is disturbed by the noise in the frequency component having a small amplitude in the image after Fourier transform. In simple figures, the frequency with the smaller amplitude occupies the majority.

  For this reason, for example, when noise is superimposed on the collation image, in the synthesized Fourier image obtained by synthesizing the Fourier-transformed registered image and the Fourier-transformed collation image, the phase due to noise with a frequency component having a small amplitude Disturbance increases.

  In the phase-only correlation method, the amplitude of all frequencies is set to 1 for this synthesized Fourier image, and then Fourier transform is performed to obtain correlation data. Therefore, phase disturbance due to noise in the synthesized Fourier image is greatly increased in the correlation data. Affect. As a result, the correlation peak is easily broken, and there is a possibility that positioning cannot be performed, and collation accuracy is also lowered. In addition, if there are a lot of noises whose phase is abruptly disturbed, the inspection mark cannot be recognized and the collation accuracy may be lowered.

  The present invention has been made to solve such a problem, and the object of the present invention is to accurately check a registered image and a matching image even if noise is mixed in the matching image. An object of the present invention is to provide a mark detection device capable of reliably detecting a mark on an object.

  In order to achieve such an object, the first invention (the invention according to claim 1) images a mark on an inspection object, and includes a collation image including the imaged mark and a preregistered mark. In a mark detection device that collates an image and detects a mark on an inspection object based on the collation result, a registered Fourier pattern data creating unit that performs Fourier transform on the registered image to create registered Fourier pattern data, and a collation image A Fourier transform is performed on the resultant Fourier pattern data creating means, and the registered Fourier pattern data and the collated Fourier pattern data are synthesized, and amplitude suppression processing is performed on the resultant Fourier pattern data. Synthesizing Fourier to create composite Fourier pattern data that has been subjected to amplitude suppression processing The turn data creation means and the complex number component of the synthesized Fourier pattern data subjected to amplitude suppression processing are respectively weighted according to the spatial frequency intensity of the registered Fourier pattern data, and further, either Fourier transform or inverse Fourier transform is performed. Pattern processing means for performing correlation data, and a registered image and a collation image based on the intensity of the correlation component for each individual pixel in a predetermined correlation component area defined on the correlation data obtained by the pattern processing means And a collating means for collating.

  According to the present invention, the registered image (image including the registered mark) is subjected to Fourier transform to create registered Fourier pattern data, and the collation image (image including the mark on the inspection object) is subjected to Fourier transform. Thus, collation Fourier pattern data is created, and the registered Fourier pattern data and the collation Fourier pattern data are synthesized to obtain synthesized Fourier pattern data. An amplitude suppression process is performed on the synthesized Fourier pattern data, and the complex number component of the synthesized Fourier pattern data subjected to the amplitude suppression process is weighted according to the spatial frequency intensity of the registered Fourier pattern data, and further subjected to Fourier transform. One of the inverse Fourier transform and the inverse Fourier transform is performed to obtain correlation data. Then, the registered image and the collation image are collated based on the intensity of the correlation component for each pixel in the predetermined correlation component area defined on the correlation data.

  In the present invention, when the registered image does not contain noise, or if it is assumed that the registered image contains a small amount of noise, the spatial frequency of the registered Fourier pattern data is added to the complex number component of the synthesized Fourier pattern data subjected to amplitude suppression processing. By applying weighting according to the intensity, the phase energy is emphasized in the characteristic frequency characteristic portion of the registered mark, and the influence caused by the noise included in the collation image becomes relatively small.

  In the second invention (the invention according to claim 2), the mark on the inspection object is imaged, the collation image including the imaged mark is collated with the registered image including the mark registered in advance, and the collation result is obtained. In the mark detection device that detects the mark on the inspection object based on the Fourier transform of the registered image, the amplitude suppression process is performed on the registered Fourier pattern data obtained thereby, and the registered Fourier pattern data subjected to the amplitude suppression process is obtained. Registered Fourier pattern data creation means to create, and a matching Fourier pattern that performs Fourier transformation on the matching image, performs amplitude suppression processing on the matching Fourier pattern data obtained thereby, and creates matching Fourier pattern data subjected to amplitude suppression processing Data creation means, amplitude suppression processing registered Fourier pattern data and amplitude suppression Synthetic Fourier pattern data creating means for synthesizing the processed collation Fourier pattern data to create synthetic Fourier pattern data, and the space of the registered Fourier pattern data before amplitude suppression processing is performed on the complex number component of the synthesized Fourier pattern data A pattern processing unit that performs weighting according to the frequency intensity, and further performs either Fourier transform or inverse Fourier transform to obtain correlation data, and a predetermined value defined on the correlation data obtained by the pattern processing unit Collation means for collating the registered image and the collation image based on the intensity of the correlation component for each pixel in the correlation component area is provided.

  According to the present invention, a Fourier transform is performed on a registered image (an image including a registered mark), an amplitude suppression process is performed on the registered Fourier pattern data obtained thereby, and a matching image (on an inspection object) The image including the mark) is subjected to Fourier transformation, and the amplitude suppression processing is performed on the matching Fourier pattern data obtained thereby, the registered Fourier pattern data subjected to the amplitude suppression processing, and the matching Fourier pattern data subjected to the amplitude suppression processing, Are combined to obtain combined Fourier pattern data. Then, the complex number component of the synthesized Fourier pattern data is weighted according to the spatial frequency intensity of the registered Fourier pattern data before being subjected to the amplitude suppression process, and further subjected to either Fourier transform or inverse Fourier transform. Correlated data. Then, the registered image and the collation image are collated based on the intensity of the correlation component for each pixel in the predetermined correlation component area defined on the correlation data.

  In the present invention, when the registered image does not contain noise or is small even if it contains noise, the complex number of the synthesized Fourier pattern data (the synthesized Fourier image data subjected to amplitude suppression processing before synthesis). By applying weighting to each component according to the spatial frequency intensity of the registered Fourier pattern data before amplitude suppression processing on the components, the characteristic frequency characteristic portion of the registered mark is emphasized and included in the collation image. The effect caused by noise is relatively small.

  According to a third invention (invention according to claim 3), in the first invention or the second invention, means for capturing an image of a registration target mark, and means for capturing an image including the imaged registration target mark as a registration target image Based on the shape of the registration target mark, the ideal correlation image is used to individually match a plurality of noise-free ideal registration image candidates, including registration candidate marks that are slightly different in shape, using the correlation method. And a means for using the ideal registered image candidate having a high registration image as a registered image.

  According to the present invention, a captured registration target image (an image including a registration target mark) and a plurality of ideal registration image candidates (including registration candidate marks whose shapes are slightly different based on the shape of the registration target mark). Noise-free images) are individually verified by the correlation method, and the ideal registered image candidate having the highest correlation value is set as a registered image (an image including a mark registered in advance). Thereby, an ideal registered image candidate having no noise including a registration candidate mark whose shape most closely matches the registration target mark is set as a registered image.

  In the third invention, using the matching method of the first invention, the registration target image is subjected to Fourier transform to obtain registered Fourier pattern data, and the ideal registered image candidate is subjected to Fourier transform to obtain matching Fourier pattern data. This registered Fourier pattern data And collation Fourier pattern data are subjected to amplitude suppression processing, the complex number component of the synthesized Fourier pattern data subjected to amplitude suppression processing is weighted according to the spatial frequency intensity of the registered Fourier pattern data, and further Fourier transform and inverse Fourier transform One of the above is used as correlation data, and the ideal registered image candidate and the registration target image are collated based on the intensity of the correlation component for each pixel in the predetermined correlation component area defined on the correlation data. It may be. In this case, the registration target image becomes the collation image in the first invention, the ideal registered image candidate becomes the registration image in the first invention, and the collation image and the registration image are collated.

  In the third invention, using the matching method of the second invention, the registration target image is subjected to Fourier transform to be registered Fourier pattern data, the ideal registered image candidate is subjected to Fourier transform to be verified Fourier pattern data, and the amplitude suppression processing is performed. The Fourier pattern data and collation Fourier pattern data subjected to amplitude suppression processing are combined, and the complex number component of the resultant Fourier pattern data obtained by this is weighted according to the spatial frequency intensity of the registered Fourier pattern data before amplitude suppression processing. And applying either Fourier transform or inverse Fourier transform to obtain correlation data, and ideal registered image candidates based on the intensity of the correlation component for each pixel in a predetermined correlation component area defined on the correlation data And the registration target image may be collated. In this case, the registration target image becomes the collation image in the second invention, the ideal registration image candidate becomes the registration image in the second invention, and the collation image and the registration image are collated.

  According to the first invention, the registered image (image including the registered mark) is subjected to Fourier transform to create registered Fourier pattern data, and the verification image (image including the mark on the inspection target) is subjected to Fourier transform. The verification Fourier pattern data is created, the registered Fourier pattern data and the verification Fourier pattern data are synthesized, the amplitude suppression processing is performed on the resultant Fourier pattern data obtained by this, and the amplitude suppression processing of the synthesized Fourier pattern data is performed. Since the complex number component is weighted according to the spatial frequency intensity of the registered Fourier pattern data, it is registered if the registered image does not contain noise or if it contains a small amount of noise. The phase energy is emphasized in the characteristic frequency characteristic part of the mark, The effect caused by the noise contained in the combined image is relatively small, and even if there is some noise in the collation image, the registered image and the collation image are collated accurately, and the mark on the inspection object is reliably detected Will be able to.

  According to the second invention, a registered image (an image including a registered mark) is subjected to Fourier transform, an amplitude suppression process is performed on the registered Fourier pattern data obtained thereby, and a verification image (a mark on the inspection object) Image) is subjected to Fourier transform, the amplitude suppression processing is performed on the matching Fourier pattern data obtained thereby, and the registered Fourier pattern data subjected to the amplitude suppression processing and the matching Fourier pattern data subjected to the amplitude suppression processing are synthesized. The complex number component of the synthesized Fourier pattern data (synthesized Fourier image data that has been subjected to amplitude suppression processing before synthesis) obtained by the above is weighted according to the spatial frequency intensity of the registered Fourier pattern data before amplitude suppression processing. So, the registered image does not contain noise or contains noise However, if the amount is small, the phase energy is emphasized in the characteristic frequency characteristic part of the registered mark, and the effect caused by the noise contained in the collation image becomes relatively small. Even if some are mixed, the registered image and the collation image are collated with high accuracy, and the mark on the inspection object can be reliably detected.

  According to the third invention, the captured registration target image (image including the registration target mark) and a plurality of ideal registration image candidates (including registration candidate marks that are slightly different in shape based on the registration target mark shape). Image with no noise) is individually matched by the correlation method, and the ideal registered image candidate with the highest correlation value is used as the registered image, so the registration candidate mark whose shape matches the mark to be registered most Even if it is unavoidable that noise is unavoidably mixed in the registration target image, the ideal registered image candidate that contains no noise is included in the registration target image. Can be detected.

  Hereinafter, the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram of a mark detection apparatus showing an embodiment of the present invention. In the figure, 1 is a camera (CCD camera) and 2 is a collation unit. The collation unit 2 includes a CPU 2-1, a ROM 2-2, a RAM 2-3, a hard disk (HD) 2-4, a frame memory (FM) 2-5, and an external connection unit (I / F) 2-6. The ROM 2-2 stores a registration program and a verification program as programs unique to the present embodiment.

  The mark detection device is used for positioning an inspection object such as a semiconductor wafer or a glass substrate. In this mark detection apparatus, a cross-shaped inspection mark M is used as an inspection mark for positioning. In addition, the hard disk 2-4 stores a plurality of ideal registered image candidates without noise including the inspection marks of registration candidates whose shape is slightly changed based on the shape of the inspection mark M.

  In the present embodiment, the shape parameters of the inspection mark M are the line width W and the length L of the cross-shaped projection, and the values of the shape parameters W and L are changed little by little. 2 (a)), and the noise-free images (“marks” + “background”) including the inspection marks M1 to Mn of the registration candidates are used as the ideal registration image candidates G1 to Gn, and the image data is stored on the hard disk 2- 4 is stored. It goes without saying that the registration candidate inspection marks M1 to Mn include those having the same shape in which the inspection mark M and its line width W and the length L of the cross-shaped projection are equal.

  Hereinafter, a method for performing amplitude suppression processing after combining is referred to as Embodiment 1, a method for performing amplitude suppression processing before combining is referred to as Embodiment 2, and registration of inspection marks and inspection marks in Embodiments 1 and 2 are described. Will be described.

[Embodiment 1: Method of performing amplitude suppression processing after combining]
[Registering inspection marks]
In the first embodiment, inspection marks are registered as follows. First, the inspection mark M on the inspection target (inspection mark M0 to be registered) is imaged by the camera 1 (FIG. 3: step 101), and the registration target image G0 including the registration target inspection mark M0 (FIG. 4A). Browse). This registration target image G0 is sent to the collation unit 2.

  The CPU 2-1 of the verification unit 2 captures the registration target image G0 from the camera 1 via the frame memory 2-5 (step 102), and performs a two-dimensional discrete Fourier transform (DFT) on the image data of the registration target image G0. (Step 103). Thereby, the image data of the registration target image G0 is set to Fourier image data (collation Fourier pattern data) F0 (see FIG. 4B). Note that the two-dimensional discrete Fourier transform is described in Non-Patent Document 1, for example.

  Next, the CPU 2-1 sets i = 1 (step 104), reads the i = 1st ideal registered image candidate G1 from the hard disk 2-4 (step 105), and adds 2 to the image data of the ideal registered image candidate G1. A dimensional discrete Fourier transform is performed (step 106). Thereby, the image data of the ideal registered image candidate G1 is set to Fourier image data (registered Fourier pattern data) F1 (see FIGS. 4C and 4D).

  Then, the CPU 2-1 synthesizes the Fourier image data F0 of the registration target image G0 obtained in step 103 and the Fourier image data F1 of the ideal registered image candidate G1 obtained in step 106, and produces synthesized Fourier image data (synthesized Fourier pattern). Data) F01 is obtained (step 107: see FIG. 4E).

  In the present embodiment, the synthesized Fourier image data F01 is A · exp (jθ) for the Fourier image data F0 of the registration target image G0 and B · exp (jφ) for the Fourier image data F1 of the ideal registered image candidate G1. In this case, it is expressed by A · B · exp (j (θ−φ)) obtained by multiplying the Fourier image data F0 of the registration target image G0 by the complex conjugate of the Fourier image data F1 of the ideal registered image candidate G1. However, A, B, θ, and φ are functions of the spatial frequency (Fourier) space (u, v).

And A · B · exp (j (θ−φ)) is
A · B · exp (j (θ−φ)) = A · B · cos (θ−φ) + j · A · B · sin (θ−φ) (1)
As A · exp (jθ) = α ₁ + jβ ₁ and B · exp (jφ) = α ₂ + jβ ₂ , A = (α ₁ ² + β ₁ ² ) ^1/2 , B = (α ₂ ² + Β ₂ ² ) ^1/2 , θ = tan ⁻¹ (β ₁ / α ₁ ), and φ = tan ⁻¹ (β ₂ / α ₂ ). By calculating the equation (1), synthesized Fourier image data is obtained.

A · B · exp (j (θ−φ)) = A · B · exp (jθ) · exp (−jφ) = A · exp (jθ) · B · exp (−jφ) = (α ₁ + jβ ₁ ) · (α ₂ −jβ ₂ ) = (α ₁ · α ₂ + β ₁ · β ₂ ) + j (α ₂ · β ₁ −α ₁ · β ₂ ) .

  Then, the CPU 2-1 performs amplitude suppression processing on the synthesized Fourier image data F01 obtained in this way (step 108). This is to suppress variation due to the difference in the amplitudes A and B in the expression (1) among the pixel values. In this embodiment, all amplitudes are set to 1 as the amplitude suppression processing. That is, the amplitudes A and B of A · B · exp (j (θ−φ)), which are the arithmetic expressions of the synthesized Fourier image data described above, are all set to 1 and only the phase.

  As the amplitude suppression processing, the log of the amplitude A · B of A · B · exp (j (θ−φ)) is taken to be log (A · B) · exp (j (θ−φ)). The amplitude A · B may be suppressed to log (A · B) (A · B> log (A · B)). Further, the √ process may be performed, and the process is not limited to the log process and the √ process, and any process may be performed as long as the amplitude can be suppressed. When all the amplitudes are set to 1 in the amplitude suppression process, there are an advantage that the calculation amount can be reduced and data are reduced as compared with the log process and the √ process.

  After performing the amplitude suppression process in step 108, the CPU 2-1 adds the ideal registered image candidate G1 obtained in step 106 to the complex number component of the synthesized Fourier image data F01r (see FIG. 4F) subjected to the amplitude suppression process. F01r ′ is multiplied by a coefficient proportional to the amplitude component of the Fourier image data F1 (weight according to the spatial frequency intensity) to obtain F01r ′ (step 109: FIG. 4 (g)). Conversion (DFT) is performed (step 110). Thereby, correlation data D01 (see FIG. 4 (h)) representing the correlation between the registration target image G0 and the ideal registration image candidate G1 is obtained.

  Next, the CPU 2-1 scans the intensity (amplitude) of each pixel in the predetermined correlation component area S01 defined on the correlation data D01 obtained in step 110, and registers the registration target image G0 and the ideal registration image candidate G1. A histogram of the intensity of the correlation component of each pixel is obtained, the top n pixels having the highest correlation component intensity are extracted from the histogram, and the average of the extracted correlation components of the n pixels is registered as the registration target image G0 and the ideal registered image candidate G1. (Step 111).

  In the same manner, by repeating steps 105 to 113, the ideal registration image candidates G2 to Gn and the registration target image G0 are collated (hereinafter, this collation method is referred to as ACPOC), and the registration target image G0 and the ideal registration are registered. Correlation values with image candidates G2 to Gn are obtained.

  In the comparison between the ideal registered image candidate Gi and the registration target image G0 by this ACPOC, a coefficient proportional to the amplitude component of the Fourier image data Fi of the ideal registered image candidate Gi is added to the complex number component of the synthesized Fourier image data F0ir subjected to the amplitude suppression process. By multiplying each, the phase energy is emphasized in the characteristic frequency characteristic portion of the registration candidate inspection mark Mi included in the ideal registered image candidate Gi, and the influence caused by the noise included in the registration target image G0 is relatively affected. Get smaller.

  Thereby, even if some noise is mixed in the registration target image G0, the registration target image G0 and the ideal registration image candidate Gi can be accurately collated, and the registration target inspection mark M0 and the registration candidate inspection mark Mi. It is possible to reliably detect the degree of similarity. In the ACPOC in the registration process of the inspection mark, the registration target image G0 corresponds to the collation image referred to in the present invention, and the ideal registered image candidate Gi corresponds to the registered image referred to in the present invention.

  The CPU 2-1 obtains a correlation value with the registration target image G0 for the ideal registration image candidates G1 to Gn (YES in step 112), and then the ideal registration image candidate Gx having the maximum correlation value (see FIG. 2B). ) As a registered image GR, the image data is registered in the hard disk 2-4 (step 114).

  As a result, the ideal registration image candidate Gx without noise including the registration candidate inspection mark Mx whose shape most closely matches the inspection mark M0 to be registered is registered as the registration image GR, and is included in the registration target image G0. The influence of noise can be eliminated.

  That is, when the registration target image G0 includes noise, if the registration target image G0 is registered as it is as the registration image GR, the inspection mark in the verification image input during verification and the inspection mark M0 in the registration target image G0 are registered. Even if and match, there is a possibility of misjudging that they do not match due to the influence of noise.

  On the other hand, in the present embodiment, the ideal registration image candidate Gx without noise including the registration candidate inspection mark Mx whose shape most closely matches the inspection mark M0 to be registered is registered as the registration image GR. An erroneous determination due to noise included in the target image G0 can be eliminated.

[Verification of inspection mark]
In the first embodiment, inspection marks are collated as follows. First, an inspection mark M (inspection mark MC to be detected) on the inspection object is imaged by the camera 1 (FIG. 5: step 201), and a collation image GC including the inspection mark MC (see FIG. 6A) is obtained. . This verification image GC is sent to the verification unit 2.

  The CPU 2-1 of the verification unit 2 captures the verification image GC from the camera 1 via the frame memory 2-5 (step 202), and performs two-dimensional discrete Fourier transform (DFT) on the image data of the verification image GC ( Step 203). Thereby, the image data of the collation image GC is set to Fourier image data (collation Fourier pattern data) FC (see FIG. 4B).

  Next, the CPU 2-1 reads the registered image GR from the hard disk 2-4 (Step 204), and performs two-dimensional discrete Fourier transform (DFT) on the image data of the registered image GR (Step 204). Thereby, the image data of the registered image GR is set to Fourier image data (registered Fourier pattern data) FR (see FIGS. 6C and 6D).

  Then, the CPU 2-1 synthesizes the Fourier image data FC of the collation image GC obtained in step 203 and the Fourier image data FR of the registered image GR obtained in step 205 to produce synthesized Fourier image data (synthesized Fourier pattern data) FCR. (Step 206: see FIG. 6E). Then, an amplitude suppression process is performed on the synthesized Fourier image data FCR (step 207). In this embodiment, as amplitude suppression processing, all amplitudes are set to 1, and only the phase is set.

  After performing the amplitude suppression process in step 207, the CPU 2-1 adds the Fourier of the registered image GR obtained in step 205 to the complex number component of the synthesized Fourier image data FCRr (see FIG. 6F) subjected to the amplitude suppression process. Each coefficient (weight according to the spatial frequency intensity) proportional to the amplitude component of the image data FR is multiplied to be FCRr ′ (step 208: FIG. 6G), and this FCRr ′ is once again subjected to a two-dimensional discrete Fourier transform ( DFT) is performed (step 209). Thereby, correlation data DCR representing the correlation between the collation image GC and the registered image GR is obtained.

  Next, the CPU 2-1 scans the intensity (amplitude) of each pixel in the predetermined correlation component area SCR determined on the correlation data DCR obtained in step 209, and determines the pixel of the matching image GC and the registered image GR. A histogram of correlation component strength is obtained, the top n pixels having the highest correlation component strength are extracted from the histogram, and an average of the extracted correlation components of the n pixels is obtained as a correlation value between the matching image GC and the registered image GR. (Step 210).

  Then, the calculated correlation value is compared with a predetermined threshold value (step 211). If the correlation value is larger than the threshold value (YES in step 211), the collation image GC and the registered image GR are (Step 212), that is, it is detected that there is the same inspection mark MC as the inspection mark (registered inspection mark) MR in the registered image GR in the collation image GC, and the correlation component of the n pixels is detected. Based on the position of the correlation peak having the highest intensity among the intensities, the amount of positional deviation between the registered inspection mark MR and the inspection mark MC is obtained (step 213).

  In the collation between the collation image GC and the registered image GR by the ACPOC, by multiplying the complex number component of the composite Fourier image data FCRr subjected to the amplitude suppression process by a coefficient proportional to the amplitude component of the Fourier image data FR of the registration image GR, respectively. The characteristic frequency characteristic portion of the registered inspection mark MR included in the registered image GR is enhanced in phase energy, and the influence caused by noise included in the verification image GC is relatively reduced.

  Thus, even if some noise is mixed in the collation image GC, the collation image GC and the registered image GR are collated with high accuracy, the inspection mark MC is reliably detected, and the amount of positional deviation can be obtained more accurately. It becomes possible. In the ACPOC in the verification mark verification process, the verification image GC corresponds to the verification image referred to in the present invention, and the registered image GR corresponds to the registered image referred to in the present invention.

[Embodiment 2: Method of performing amplitude suppression processing before composition]
In the second embodiment, inspection marks are registered as follows. First, the inspection mark M on the inspection target (inspection mark M0 to be registered) is imaged by the camera 1 (FIG. 7: step 301), and the registration target image G0 including the registration target inspection mark M0 (FIG. 8A). Browse). This registration target image G0 is sent to the collation unit 2.

  The CPU 2-1 of the collation unit 2 captures the registration target image G0 from the camera 1 via the frame memory 2-5 (step 302), and performs two-dimensional discrete Fourier transform (DFT) on the image data of the registration target image G0. (Step 303). Thereby, the image data of the registration target image G0 is set to Fourier image data (collation Fourier pattern data) F0 (see FIG. 8B).

  Then, an amplitude suppression process is performed on the Fourier image data F0 (step 304) to obtain Fourier image data F0r subjected to the amplitude suppression process (see FIG. 8C). In this embodiment, as amplitude suppression processing, all amplitudes are set to 1, and only the phase is set.

  Next, the CPU 2-1 sets i = 1 (step 305), reads the i = 1st ideal registered image candidate G1 from the hard disk 2-4 (step 306), and adds 2 to the image data of the ideal registered image candidate G1. A dimensional discrete Fourier transform is performed (step 307). Thereby, the image data of the ideal registered image candidate G1 is set as Fourier image data (registered Fourier pattern data) F1 (see FIGS. 8D and 8E).

  Then, an amplitude suppression process is performed on the Fourier image data F1 (step 308) to obtain Fourier image data F1r subjected to the amplitude suppression process (see FIG. 8F). In this embodiment, as amplitude suppression processing, all amplitudes are set to 1, and only the phase is set.

  Then, the CPU 2-1 combines the Fourier image data F0r subjected to the amplitude suppression processing of the registration target image G0 obtained in step 304 and the Fourier image data F1r subjected to the amplitude suppression processing of the ideal registered image candidate G1 obtained in step 308. Then, synthetic Fourier image data (synthetic Fourier pattern data) F01r is obtained (step 309: see FIG. 8G).

  Then, the CPU 2-1 multiplies the complex number component of the synthesized Fourier image data F01r by a coefficient proportional to the amplitude component of the Fourier image data F1 before the amplitude suppression processing of the ideal registered image candidate G1 obtained in step 307, respectively. (Step 310: FIG. 8 (h)), this F01r 'is again subjected to two-dimensional discrete Fourier transform (DFT) (step 311). Thereby, correlation data D01 representing the correlation between the registration target image G0 and the ideal registration image candidate G1 is obtained. (See Fig. 8 (i))

  Next, the CPU 2-1 scans the intensity (amplitude) of each pixel in the predetermined correlation component area S01 defined on the correlation data D01 obtained in step 311 to determine the registration target image G0 and the ideal registered image candidate G1. A histogram of the intensity of the correlation component of each pixel is obtained, the top n pixels having the highest correlation component intensity are extracted from the histogram, and the average of the extracted correlation components of the n pixels is registered as the registration target image G0 and the ideal registered image candidate G1. (Step 312).

  Similarly, by repeating steps 306 to 314, ideal registration image candidates G2 to Gn are collated (APPOC) with registration target image G0, and correlation between registration target image G0 and ideal registration image candidates G2 to Gn is performed. Find the value.

  In the comparison between the ideal registered image candidate Gi and the registration target image G0 by this ACPOC, the complex component of the synthesized Fourier image data F0ir is proportional to the amplitude component of the Fourier image data Fi before the amplitude suppression processing of the ideal registered image candidate Gi. By multiplying each coefficient, the phase energy is emphasized in the characteristic frequency characteristic portion of the registration candidate inspection mark Mi included in the ideal registered image candidate Gi, and the influence caused by the noise included in the registration target image G0 is relative. Become smaller.

  Thereby, even if some noise is mixed in the registration target image G0, the registration target image G0 and the ideal registration image candidate Gi can be accurately collated, and the registration target inspection mark M0 and the registration candidate inspection mark Mi. It is possible to reliably detect the degree of coincidence. In the ACPOC in the registration process of the inspection mark, the registration target image G0 corresponds to the collation image referred to in the present invention, and the ideal registered image candidate Gi corresponds to the registered image referred to in the present invention.

  The CPU 2-1 obtains the correlation value with the registration target image G0 for the ideal registration image candidates G1 to Gn (YES in step 313), and then the ideal registration image candidate Gx having the maximum correlation value (see FIG. 2B). ) As a registered image GR, the image data is registered in the hard disk 2-4 (step 315).

  As a result, the ideal registration image candidate Gx without noise including the registration candidate inspection mark Mx whose shape most closely matches the inspection mark M0 to be registered is registered as the registration image GR, and is included in the registration target image G0. The influence of noise can be eliminated.

[Verification of inspection mark]
In the second embodiment, inspection marks are collated as follows. First, an inspection mark M (inspection mark MC to be detected) on the inspection object is imaged by the camera 1 (FIG. 9: step 401), and a collation image GC including the inspection mark MC (see FIG. 10A) is obtained. . This verification image GC is sent to the verification unit 2.

  The CPU 2-1 of the verification unit 2 captures the verification image GC from the camera 1 via the frame memory 2-5 (step 402), and performs two-dimensional discrete Fourier transform (DFT) on the image data of the verification image GC ( Step 403). Thereby, the image data of the collation image GC is set to Fourier image data (collation Fourier pattern data) FC (see FIG. 10B).

  Then, an amplitude suppression process is performed on the Fourier image data FC (step 404) to obtain Fourier image data FCr subjected to the amplitude suppression process (see FIG. 10C). In this embodiment, as amplitude suppression processing, all amplitudes are set to 1, and only the phase is set.

  Next, the CPU 2-1 reads the registered image GR from the hard disk 2-4 (step 405), and performs two-dimensional discrete Fourier transform (DFT) on the image data of the registered image GR (step 406). Thereby, the image data of the registered image GR is set as Fourier image data (registered Fourier pattern data) FR (see FIGS. 10D and 10E).

  Then, an amplitude suppression process is performed on the Fourier image data FR (step 407) to obtain Fourier image data FRr subjected to the amplitude suppression process (see FIG. 10F). In this embodiment, as amplitude suppression processing, all amplitudes are set to 1, and only the phase is set.

  Then, the CPU 2-1 synthesizes the Fourier image data FCr subjected to the amplitude suppression processing of the collation image GC obtained in step 404 and the Fourier image data FR subjected to the amplitude suppression processing of the registered image GR obtained in step 407. Fourier image data (synthetic Fourier pattern data) FCRr is obtained (step 408: see FIG. 10G).

  Then, the complex number component of the synthesized Fourier image data FCRr is multiplied by a coefficient proportional to the amplitude component of the Fourier image data FR before the amplitude suppression processing of the registered image GR obtained in step 406 to obtain FCRr ′ (step 409: FIG. 10 (h)), the FCRr 'is again subjected to a two-dimensional discrete Fourier transform (DFT) (step 410). Thereby, correlation data DCR representing the correlation between the collation image GC and the registered image GR is obtained (see FIG. 10I).

  Next, the CPU 2-1 scans the intensity (amplitude) of each pixel in the predetermined correlation component area SCR determined on the correlation data DCR obtained in step 410, and calculates the pixel of the collation image GC and the registered image GR. A histogram of the correlation component intensity is obtained, the top n pixels having the highest correlation component intensity are extracted from the histogram, and an average of the extracted correlation components of the n pixels is obtained as a correlation value between the collation image GC and the registered image GR. (Step 411).

  Then, the calculated correlation value is compared with a predetermined threshold value (step 412). If the correlation value is larger than the threshold value (YES in step 412), the collation image GC and the registered image GR are (Step 413), that is, it is detected that there is an inspection mark MC identical to the inspection mark (registered inspection mark) MR in the registered image GR in the collation image GC, and the correlation component of the n pixels is detected. Based on the position of the correlation peak having the highest intensity among the intensities, the amount of positional deviation between the registered inspection mark MR and the inspection mark MC is obtained (step 414).

  In the collation between the collation image GC and the registered image GR by the ACPOC, the Fourier before the amplitude suppression processing of the registration image GR is performed on the complex number component of the composite Fourier image data (the synthesis Fourier image data subjected to the amplitude suppression processing before the synthesis) FCRr. By multiplying each coefficient proportional to the amplitude component of the image data FR, the phase energy is emphasized in the characteristic frequency characteristic portion of the registered inspection mark MR included in the registered image GR, and the noise included in the verification image GC The resulting effect is relatively small.

  As a result, even if some noise is mixed in the collation image GC, the collation image GC and the registered image GR can be collated with high accuracy, the detection mark MC can be reliably detected, and the amount of displacement can be accurately obtained. It becomes possible. In the ACPOC in the verification mark verification process, the verification image GC corresponds to the verification image referred to in the present invention, and the registered image GR corresponds to the registered image referred to in the present invention.

  In the first embodiment described above, step 110 shown in FIG. 3 and step 209 shown in FIG. 5 are performed. In the second embodiment described above, step 311 shown in FIG. 7 and step 410 shown in FIG. 9 are performed. However, instead of the two-dimensional discrete Fourier transform, a two-dimensional discrete inverse Fourier transform may be performed. The two-dimensional discrete Fourier transform and the two-dimensional discrete inverse Fourier transform do not change the matching accuracy quantitatively.

  In the first and second embodiments described above, in the registration of the inspection mark, the ideal registered image candidate Gx having the maximum correlation value is registered as the registered image GR, but the image data is registered. Registered Fourier pattern data Fx subjected to two-dimensional discrete Fourier transform may be registered. In this way, in the verification of the inspection mark, by reading the registered Fourier pattern data Fx as the registered Fourier pattern data FR, there is no need to apply a two-dimensional discrete Fourier to the registered image GR, and the processing is simple. Become.

  In the first and second embodiments described above, verification is performed by ACPOC when registering an inspection mark. However, it is not always necessary to use ACPOC. That is, when the inspection mark M0 to be registered is imaged, it is possible to consciously image an image with little noise, and collation may be performed using another correlation method such as normal POC.

  In the first and second embodiments described above, an ideal registered image candidate Gx including the registration candidate inspection mark Mx that images the registration target inspection mark M0 and has the shape most similar to the registration target inspection mark M0 is obtained. Although the search is performed, the inspection mark M0 to be registered may be imaged without including noise, and the captured image may be registered as a registered image.

  In the first and second embodiments described above, the ideal registered image candidates G1 to Gn in which the values of the shape parameters W and L are changed little by little are stored in the hard disk 2-4. The created ideal registered image candidates G1 to Gn may be stored in the hard disk 2-4, and the ideal registered image candidates G1 to Gn are automatically created by changing the values of the shape parameters W and L. You may make it store in 2-4. In addition, ideal registered image candidates may be automatically created each time collation with the registration target image G0 is performed.

Further, in the first embodiment, the inspection mark may be registered by the flowchart according to FIG. 7 described in the second embodiment. In the second embodiment, the flowchart according to FIG. 3 described in the first embodiment. The inspection mark may be registered as described above.
Similarly, in the first embodiment, the inspection mark may be collated by the flowchart according to FIG. 9 described in the second embodiment, and in the second embodiment, according to FIG. 5 described in the first embodiment. You may make it collate an inspection mark with a flowchart.

  In the first and second embodiments described above, the inspection mark M is a cross, but it goes without saying that it may be a star, a polygon, or the like. Further, it can be used not only for positioning of inspection objects such as semiconductor wafers and glass substrates but also for reading characters. In other words, it is possible to recognize a character written on the paper by regarding the character written on the paper as a mark and collating the collation image obtained by capturing the character with a registered image including a pre-registered character. It is.

It is a block block diagram of the mark detection apparatus which shows one embodiment of this invention. It is a figure which illustrates the ideal registered image candidate used in this mark detection apparatus, and the registered image searched out from this ideal registered image candidate. 6 is a flowchart for explaining inspection mark registration according to the first embodiment; 6 is a diagram illustrating a correlation value calculation process in registering an inspection mark according to Embodiment 1. FIG. 6 is a flowchart for explaining inspection mark verification according to the first embodiment; 6 is a diagram for explaining a process of calculating a correlation value when collating inspection marks according to Embodiment 1. FIG. 10 is a flowchart for explaining inspection mark registration according to the second embodiment; FIG. 10 is a diagram for explaining a process of calculating a correlation value when registering an inspection mark according to the second embodiment. 10 is a flowchart for explaining inspection mark verification according to the second embodiment. 6 is a diagram for explaining a process of calculating a correlation value when collating inspection marks according to Embodiment 1. FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Camera, 2 ... Collation part, 2-1 ... CPU, 2-2 ... ROM, 2-3 ... RAM, 2-4 ... Hard disk (HD), 2-5 ... Frame memory (FM), 2-6 ... External connection unit (I / F), M ... inspection mark, M0 ... registration target inspection mark, MC ... detection target inspection mark, G0 ... registration target image, G1-Gn ... ideal registration image candidate, M1-Mn ... registration Candidate inspection marks, GC ... verification image, GR ... registered image.

Claims (3)

Mark detection in which a mark on the inspection object is imaged, a collation image including the captured mark is collated with a registered image including a pre-registered mark, and the mark on the inspection object is detected based on the collation result In the device
A registered Fourier pattern data creating means for creating a registered Fourier pattern data by applying a Fourier transform to the registered image;
A matching Fourier pattern data creating means for creating a matching Fourier pattern data by performing Fourier transform on the matching image;
Synthesizing the registered Fourier pattern data and the matching Fourier pattern data, performing amplitude suppression processing on the resultant Fourier pattern data obtained thereby, and generating synthesized Fourier pattern data that has been subjected to amplitude suppression processing Means,
The complex number component of the composite Fourier pattern data subjected to the amplitude suppression process is weighted according to the spatial frequency intensity of the registered Fourier pattern data, and further subjected to either Fourier transform or inverse Fourier transform to obtain correlation data. Pattern processing means, and
Collation means for collating the registered image with the collation image based on the intensity of the correlation component for each individual pixel in a predetermined correlation component area defined on the correlation data obtained by the pattern processing means. A mark detection device characterized by that. Mark detection in which a mark on the inspection object is imaged, a collation image including the captured mark is collated with a registered image including a pre-registered mark, and the mark on the inspection object is detected based on the collation result In the device
A Fourier transform is performed on the registered image, an amplitude suppression process is performed on the registered Fourier pattern data obtained thereby, and registered Fourier pattern data creating means for creating registered Fourier pattern data subjected to the amplitude suppression process;
A Fourier transform is performed on the collation image, an amplitude suppression process is performed on the collation Fourier pattern data obtained thereby, and collation Fourier pattern data creating means for creating collation Fourier pattern data subjected to the amplitude suppression process;
Synthesizing Fourier pattern data creating means for synthesizing the amplitude-suppressed registered Fourier pattern data and the amplitude-suppressed collation Fourier pattern data to create synthesized Fourier pattern data;
Weighting is applied to the complex number component of the synthesized Fourier pattern data according to the spatial frequency intensity of the registered Fourier pattern data before the amplitude suppression processing, and one of Fourier transform and inverse Fourier transform is further performed. Pattern processing means as correlation data;
Collation means for collating the registered image with the collation image based on the intensity of the correlation component for each individual pixel in a predetermined correlation component area defined on the correlation data obtained by the pattern processing means. A mark collating apparatus characterized by the above. In the mark collating apparatus according to claim 1 or 2,
Means for imaging a mark to be registered;
Means for capturing an image including the imaged registration target mark as a registration target image;
Based on the shape of the registration target mark, a plurality of noise-free ideal registration image candidates including registration candidate marks whose shapes are slightly different from each other and the registration target image are individually collated by the correlation method, and the most correlated A mark detection apparatus comprising: means for setting an ideal registered image candidate having a high value as the registered image.