JP2008040705A - Blur filter design method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To create a detailed blur reference image by extracting an amount of blur of a real image to be examined, and improve the defect detection sensitivity of a pattern edge part. <P>SOLUTION: An optical scanning part 4 scans a wiring pattern and outputs a scan signal 14, and a photoelectric image processing part 5 converts the scan signal 14 to a real image 15 of a multi-valued tone and outputs it. A design data development part 2 develops the design data 11 inputted by a design data inputting part 1 to each pixel of the resolution lower than the examination resolution as a wiring pattern in multi-tone values, and creates a reference data 12. Based on the edge position of the real image 15, a reference image generation part 3 carries out rounding correction of tones at a corner part by the resolution lower than the examination resolution with the width of each wiring pattern of the reference data 12 being kept at multi-tones, executes blur processing by an optical spot spread function, and creates a blur reference image 13. An image memory positioning part 6 carries out positioning of the real image 15 and the blur reference image 13 in the image memory. An image comparison and defect detection part 7 compares the two and conducts the examination. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention obtains a light spot spread function indicating the intensity and spread of blur from the pattern edge of a captured real image, and a reference image by a light spot spread function showing a luminance distribution very similar to a real image obtained from a real optical system. For example, a blur reference image necessary for detecting a defect of a circuit pattern drawn on a photomask LCD (Liquid Crystal Display), a package, etc. by image comparison, for example, a blur filter design method and a blur reference image generation method The present invention relates to a blur filter design method for automatically generating a light spot spread function used for a coefficient of a blur filter for a real image.

  In general, pattern inspection is performed to check whether fine patterns such as semiconductor integrated circuits formed on photomasks, wafers, and liquid crystals are accurately drawn based on the dimensions and shapes of ideal patterns created from design data. It is necessary to check with the equipment. In this type of pattern inspection apparatus, first, design data described by rectangular or trapezoidal position coordinates and line segment length is converted into bit data of “0” and “1” corresponding to the pattern to be inspected with appropriate resolution. Convert.

  Next, the inspection target pattern is scanned with an appropriate imaging system, and a display image of the inspection target obtained through the imaging lens is formed on a light receiving element such as a CCD (Charge Coupled Device), a photo detector, etc. A blur filter that creates a predetermined optical point spread function from an optical pupil that is a transfer function of an optical system including an imaging system from the pattern information converted into, and performs a convolution operation with binary bit data using this function By processing, the design data is converted into multi-gradation data (multi-valued data) to obtain a blurred reference image, which is obtained from the design data in synchronization with the actual image obtained by optical system scanning or imaging system input. Then, the defect reference on the actual pattern is detected by image comparison in which the blurred reference image is read and a mismatch point between the two is detected at the corresponding pixel position.

  It should be noted that the pattern in the actual image has a rounded transfer pattern, a thicker or thinner line pattern compared to the ideal design value due to optical conditions such as lens aberration of the imaging system, optical axis misalignment, and the influence of the manufacturing process. Dimensional errors due to differences in pattern direction blur, etc., especially for defects near fine pattern edges with large luminance fluctuations, it is difficult to analytically determine the transfer function of the optical system including the imaging system. Due to a pseudo error from the blurred reference image obtained from the data, a pseudo defect that is not determined as a defect is likely to occur. Therefore, the blur filter that acts on the design data is necessary for the image comparison inspection with the actual image by performing the blurring process with a luminance distribution similar to the light intensity of the actual optical system according to the blur amount for each direction of the pattern. A method of generating a design reference blur reference image in advance is conceivable.

  In the conventional blur filter design method and blur reference image generation method based on automatic generation of the light spot spread function, when creating a blur reference image close to a defect-free real image from design data, optical blur information close to the real image is generated. The light intensity distribution can be obtained from an optical simulation that approximates the MTF curve in the actual pattern portion of the mounted optical system or a correlation function of the pattern edge portion that is close to the OTF curve in the optical pupil and pattern portion. Assuming that the optical pupil of the edge or the beam intensity is a Gaussian distribution, the parameters of the Gaussian distribution function are obtained by numerical simulation, the pattern edge profile of the actual image is approximated, and the light spot spread that becomes the blur filter coefficient Find a function and design a blur filter that performs a composition operation between design data and filter coefficients It is allowed to act to create a blurred reference image on the over data.

Furthermore, real images and blur fill. In the defect detection hardware that performs image comparison using the blurred reference image obtained by the data design method, the gradation difference between the obtained blurred reference image and the actual image is determined using the appropriate defect algorithm threshold. The defect is detected by determining whether or not (see, for example, Patent Document 1 and Patent Document 2).
JP-A-5-160002 JP 2004-309826A (P2004-309826A)

  However, when such a conventional blur filter design method and blur image automatic generation method are applied to defect inspection by image comparison, the original design data is used by using a correction template close to the corner of the blur reference image created from the design data. The blur reference image that is close to the actual image data of the normal pattern obtained by optical scanning is created by correcting the image and performing blur filter processing again to create the blur reference image. There is a problem that the inspection throughput is reduced by the correction processing on the data, and a gradation distribution very similar to the actual image cannot be obtained due to a gradation difference from the correction template.

  That is, pattern correction of design data is performed by selecting a template close to the shape of the actual image from a plurality of templates having a pattern shape indicated by a bit string, and a bit string obtained by the logical operation is calculated from a point spread function obtained by optical scanning. Since the blurring reference image used for the comparison process with the multi-valued real image is created by the convolution calculation of the light intensity component value and the corrected template image, the amount of calculation to be processed greatly increases as the number of pixels increases. There was a problem that the throughput was lowered.

  Furthermore, the template of the design data is corrected by calculating a bit string in each pixel, and the pattern gradation value after the correction of the blurring reference image is limited to the gradation distribution of the template. Furthermore, since tone correction is performed regardless of the shape of the surrounding pattern, the tone difference from the actual image obtained by scanning increases at the edge portion where the pattern edge is not located at the pixel boundary, and image comparison is performed. When applied to the defect inspection by, there is a problem that pseudo defects are likely to occur.

  Furthermore, in the conventional blur filter design method, it is difficult to accurately determine the optical transfer function between the optical system including the imaging system and the object to be inspected from the actual image necessary for the blur filter process. Assuming that the light intensity distribution is a Gaussian distribution, the blur filter coefficient for the reference data is obtained from the isotropic optical pupil. The gradation distribution of the edge pattern of the blurred reference image may deviate from the blur method indicated by the real optical system. For this reason, a luminance difference from the blurred reference image at a position corresponding to the actual image in the vicinity of the pattern edge is likely to occur, and defect detection by image comparison has a drawback in that the defect detection sensitivity on the edge tends to be lowered. Furthermore, since the gradation correction is performed by the blurring filter that includes the shape of the surrounding pattern, the optical conditions due to the optical axis deviation from the light receiving element to the imaging system, and the sneak in of the light fluctuation, the pattern is not positioned at the pixel boundary. At the edge portion or the rounded portion extending from the edge to the corner portion, there is a problem that the gradation difference from the actual image obtained by scanning becomes large and pseudo defects are likely to occur.

  For this reason, in the pattern portion composed of two arbitrary angled oblique edges where the design pattern does not exist at the pixel separation position, the blurring due to optical error factors such as the optical axis deviation of the imaging system is different. Even if the gradation of the pattern edge is corrected by using the pixel unit, the step of the gradation difference cannot be interpolated. In particular, the defects at the pixel boundaries and the contrast with the surroundings are poor, and the pattern defects on the edges that are finer than the inspection resolution are unlikely to produce a gradation difference from the surroundings. When defect inspection is performed by comparing with an actual image, there is a drawback that the defect detection sensitivity is lowered.

  The present invention has been made in view of such a situation, and an object thereof is to provide an automatic generation method of a light spot spreading function and a blurring reference image generation method capable of solving the above-described problems.

  In order to achieve such an object, the invention described in claim 1 scans a light-shielding portion, a transmission portion, and a fine pattern portion to be inspected having a fine pattern, which are formed based on design data. The transmitted light obtained by passing through the inspection object is imaged on a light receiving element such as a CCD by an objective lens, a real image is generated from the pattern information obtained from the light receiving element, and obtained from the pattern information of the real image. Is a blur filter design method in which the gradation distribution of the blurred image of the obtained reference data accurately reproduces the gradation distribution of the actual image, each pixel having a resolution finer than the inspection resolution. The reference data is created by developing the design data as a pattern consisting of multi-gradation values on the top, and the selected arbitrary coordinates on the pattern edge part from the reference data without any blur information, and this Characterized by a light spot spread function, which is generated by extracting the blur information real image pattern edge corresponding to the pattern edge portion position of the irradiation data by tracing the vertical direction as the filter coefficients.

  Further, the invention of claim 2 is the blur filter design method according to claim 1, wherein the absolute value of the target pixel position on the luminance profile obtained by tracing the pattern edge of the actual image in the vertical direction and the target pixel position + the second luminance difference. Each pixel within a range of a perfect circle or an ellipse determined by the combination of two directions adjacent to each other among the upper, lower, left, and right from the actual image edge inclination. A radius value having a pixel area value in the area as a coefficient is set as an optical coefficient in each direction of the real image, and the area and direction of the blur filter coefficient that is a light spot spread function of the real image is determined from the optical coefficient. It is a thing.

  According to a third aspect of the present invention, in the above-described optical coefficient of the blur filter designing method according to the second aspect, the luminance profile traced in the vertical direction with respect to the pattern edges in the upper, lower, right, and left directions of the real image. Of the circle or ellipse determined by the combination of the radius values of the optical coefficients in the two adjacent directions among the actual image edge slopes that are the maximum value of the luminance difference between the target pixel and the target pixel + k (where k> = 1) The filter coefficient value is determined by associating the pixel area value in each pixel region with the light intensity.

According to a fourth aspect of the present invention, in the blur filter design method according to the third aspect, each radius having an area value occupying each unit pixel within a true circle drawn by a radius value matching the optical coefficient as a blur filter coefficient value. The pattern edge on the blurred reference image is traced in the vertical direction on each blurred reference image obtained by performing successive convolution operations on the multi-valued reference data with the blurring value filter. The maximum value of the edge inclination of the blurred reference image, which is the maximum absolute value of the position and the target pixel position + second luminance difference, is α, and an arbitrary radius value R of the light spot spread function that is the radius value of the optical coefficient is actual. Α = aR ³ + bR ² + cR + d with coefficients a, b, c, d (a, b, c, d are real numbers (a ≠ 0))
This is the only one that is obtained from the relationship of the cubic equation.

  According to a fifth aspect of the present invention, in the blur filter design method according to the fourth aspect, each radius value having an area value of each pixel within a range of a perfect circle of a radius value matching the optical coefficient as a blur filter coefficient is set. The above described edge inclination α on each blurred reference image obtained by sequentially performing convolution operations on the multi-valued reference data with a blur filter is the edge edge of the multi-valued reference data based on the design data. Is the same neighborhood value at any arbitrary position within one pixel.

  According to a sixth aspect of the present invention, in the blurring filter design method according to the fifth aspect, the optical coefficients in the four directions that are traced in the vertical direction with respect to the pattern edges in the upper, lower, right, and left directions of the real image are adjacent. The pixel area value in the area where the radius values of the optical coefficients in the two directions are in contact is used as the light spot spread function in the trace direction, and within the area of a perfect circle or ellipse that is affected by the light intensity of the light spot spread function in the trace direction. A blur filter of the real image is generated with the area ratio in the unit pixel in contact with the filter coefficient in each trace direction, and a new one is generated by inverting the optical coefficient in the direction not applied to the pattern among the respective optical coefficients. A light spot required for blurring reference data for each pixel area value within the range where a perfect circle or ellipse with a radius value corresponding to this optical coefficient is created. It is obtained as the filter coefficients of the gully function.

  According to a seventh aspect of the present invention, in the blurring filter design method according to the sixth aspect, from the new four-direction optical coefficients generated by inverting the blur filter coefficient corresponding to the optical coefficient in each trace direction of the real image, Reference data that is converted into multi-values using a blur filter that has a filter coefficient of the light spot spread function, with the area value occupying a unit pixel within a range where a perfect circle or ellipse shape with a radius corresponding to the optical coefficient touches. Is subjected to a convolution process so that blurred reference images having various luminance profiles with different blurring directions in each direction of the reference data can be created.

  The invention according to claim 8 scans the light-shielding portion formed based on the design data, the transmission portion, and the fine pattern portion of the inspection target having a fine pattern, and transmits the transmitted light obtained through the inspection target. An image is formed on a light receiving element such as a CCD with an objective lens, a real image is generated from pattern information obtained from the light receiving element, and a filter obtained from the pattern information of the real image is applied to the design data. A blur filter design method and an image comparison defect detection hardware in which the gradation distribution of the blurred image of the reference data accurately reproduces the gradation distribution of the actual image, and the actual image is generated by generating the light spot spread function according to claims 1 to 7. Generating a blurred reference image that closely resembles how to blur.

  The invention according to claim 9 scans the light-shielding portion formed based on the design data, the transmission portion, and the fine pattern portion of the inspection target having a fine pattern, and transmits the transmitted light obtained through the inspection target. An image is formed on a light receiving element such as a CCD with an objective lens, a real image is generated from pattern information obtained from the light receiving element, and a filter obtained from the pattern information of the real image is applied to the design data. This is a filter manufactured by the blur filter design method in which the gradation distribution of the blurred image of the reference data accurately reproduces the gradation distribution of the actual image, and a large amount of design data is placed on each pixel having a resolution finer than the inspection resolution. The reference data is created by developing it as a pattern consisting of gradation values. From the reference data that has no blur information, the selected arbitrary coordinates on the pattern edge and the reference data Characterized in that it is manufactured by the blur filter design method with a light spot spread function, which is generated by extracting the blur information real image pattern edge corresponding to the pattern edge portion located by tracing the vertical direction on the filter coefficients.

  The invention of claim 10 scans the light shielding part formed based on the design data, the transmission part, and the fine pattern part of the inspection object having a fine pattern, and transmits the transmitted light obtained by passing through the inspection object. An image is formed on a light receiving element such as a CCD with an objective lens, a real image is generated from pattern information obtained from the light receiving element, and a filter obtained from the pattern information of the real image is applied to the design data. A blur filter design system in which the gradation distribution of the blurred image of the reference data accurately reproduces the gradation distribution of the actual image, and an optical scanning unit that scans the fine pattern of the inspection target and outputs a scanning signal; A photoelectric image processing means for converting the scanning signal output by the optical scanning means into a real image having a multi-value gradation, a design data input means for inputting design data, and the design data. The design data input by the data input means is developed as a pattern of multi-gradation values on each pixel having a resolution finer than the inspection resolution to generate reference data; A reference for generating a blurred reference image by performing rounding gradation correction of a corner portion at a test resolution or less while maintaining the width of each pattern of the reference data based on an edge position, with a multi-gradation, and performing a blurring process using a light spot spread function. Image generator, image memory alignment means for aligning the actual image and the blurred reference image to the corresponding image position address in a predetermined image memory, and comparing the luminance of the actual image and the blurred reference image And an image comparison defect detecting means for detecting a defect of the pattern.

  According to the present invention, the design data in which the gradation value of the pixel is expressed by a multi-gradation value matrix of multiple double precision is scanned with a blur filter having a light spot spread function obtained by inverting the optical coefficient of the actual image. By applying the blurring process, the gray level of the edge position of the reference data can be expressed with high precision regardless of the position of the edge of the reference data in the pixel, not the pixel break, and it was obtained from the inspection target. A highly accurate blurred reference image having a luminance distribution very close to the luminance distribution of the actual image can be created.

  Combining optical coefficients having an arbitrary radius R in each trace direction of the actual pattern to compose a blur filter coefficient of the light spot spread function, it is possible to freely generate a blur reference image having a different blur method depending on the pattern direction. . In addition, even if there is a pattern deviation within one pixel in multi-valued reference data, the optical coefficient does not change, and the optical coefficient can be stably extracted by allowing the pattern deviation, so that the robust image position between the actual image and the reference image As a result, it is possible to realize the alignment and to greatly reduce the cause of the positional deviation error when the defect is detected.

  Further, even if the light spot spread function is not a perfect circle but an untargeted shape, the coefficient value of the point spread function in the one-pixel grid is defined as an elliptical area converted value indicating the light spread in the pixel and the light intensity. Even if the scanning function of the inspection target has an asymmetric beam shape, for example, the shape of a laser beam waist changes, even if the optical pupil has a complicated shape due to the presence of light sneak or light fluctuation Since the light spot spread function of the blur filter generated from the optical coefficients in each direction of the real image is uniquely aggregated into a perfect circle or an elliptical shape at each pixel position of the radius R of the optical coefficient, it is generated from the real image. The optical coefficient can be quantified as a robust blur amount that does not depend on the shape of the light intensity distribution of the scanning system.

  For the reference data with multiple gradations, the light spot spread function that represents the light intensity and range of the blur filter is obtained from the optical coefficients for the different directions of blur from the actual image, so the gradation distribution of the actual image Precise blur reference image can be created according to Further, since the light spot spread function can be obtained in pixel units or sub-pixel units in accordance with the blurring amount of the edge pattern of the actual image, the gradation reference image gradation distribution suitable for the desired inspection accuracy is obtained. be able to.

  In addition, even if the corner of the design data and the edge with an arbitrary angle of the wiring pattern are arranged at any angle, the blur processing is performed in sub-pixel units finer than the pixel resolution. Because it is possible to create a blurred reference image pattern that is very close to the actual image of the pattern to be inspected without depending on the positional relationship between the position of the captured image and the scanning direction of the captured image, the corner portion or pattern edge portion of the actual image of the pattern inspection device There is an effect that it is possible to greatly improve the detection sensitivity of defects in the above.

  The present invention relates to a light spot spread function indicating a light intensity and a blur range from edge information in each direction of a real image pattern obtained by imaging a test object in an image comparison inspection of a fine pattern drawn on a screen, a glass material, etc. And a design means for designing a defect-free blurred reference image 13 (FIG. 1) having a blur amount similar to that of the actual image 15 (FIG. 1). Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram of image comparison defect detection hardware for realizing a blur filter design method and an automatic generation of a blur reference image 13 according to an embodiment of the present invention. The image comparison defect detection hardware that implements this blur filter design method includes an optical scanning unit 4 that outputs a scanning signal 14 by scanning a wiring pattern to be inspected, that is, a pattern to be inspected, and a plurality of scanning signals 14. A photoelectric image processing unit 5 that converts and outputs the actual image 15 of value gradation is provided.

  Also, a design data input unit 1 for inputting design data 11 in which the graphic dimensions of the pattern to be inspected are defined, and design data expansion for expanding the design data 11 into a wiring pattern to create multi-value gradation reference data 12 Unit 2, reference image generation unit 3 that creates blurred reference image 13 by correcting each wiring pattern of reference data 12 to be close to real image 15, and the pattern to be inspected obtained by optical scanning An image memory alignment unit 6 that aligns an address of a corresponding image position in an image memory (not shown) that incorporates the actual image 15 and the blurred reference image 13 created from the design data 11, and an image memory alignment unit 6 An image comparison defect for inspecting a pattern to be inspected by comparing the actual image 15 corresponding to the design data 11 in the image and the blurred reference image 13. A detection unit 7 is provided.

  Next, the operation of the present invention will be described with reference to the drawings. Design data 11 described in a format such as NEBES or raster scan data format is input from the design data input unit 1. Subsequently, in the design data development unit 2, the input design data 11 is developed as a circuit pattern on each pixel arranged in a grid corresponding to the address of the pattern coordinates of the actual image 15.

  FIG. 2 is an explanatory view showing a part of the developed wiring pattern. Each pixel corresponds to the development resolution in the design data development unit 2. In this case, the edge 22 of the pattern 21 is not a pixel break, and for example on the pixel 23, the pattern 22 is developed at a ratio of 3: 1 in the x direction (horizontal direction) and a ratio of 1: 1 in the y direction (vertical direction). It shows that it is expanded to the position.

  Each pixel is provided with a plurality of sub-pixels in order to calculate a gradation value (gray value) of a multi-level gradation with a resolution finer than the inspection resolution. Depending on the number of sub-pixels, The accuracy of the gradation value is determined. For example, when developing data with a maximum gradation value of 255 and a minimum gradation value of 0, one pixel is composed of 16 × 16 subpixels, and each subpixel is “0”, “1”. The binary value is taken.

  FIG. 3 is an enlarged view of the pixel 23 of FIG. 2. In this case, the pattern 21 exists on 8 × 12 sub-pixels out of 16 × 16 sub-pixels. Here, when the sub-pixel belonging to the pattern 21 is represented by “1” and the sub-pixel not belonging to the pattern 21 is represented by “0”, the gradation value of the pixel 23 is 8 × 12 = 96.

  Thereby, the gradation value of the pixel not belonging to the pattern 21 becomes the minimum gradation value (MIN = 0), and the gradation value of the pixel other than the pattern edge among the pixels belonging to the pattern 21 is the maximum gradation value (MAX = 255). Further, the gradation value of the pixel at the pattern edge part partially belonging to the pattern 21 is a gradation value corresponding to the number of sub-pixels belonging to the pattern 21 in the pixel.

  In this way, by integrating the area of the bit string in the pixel, as shown in FIG. 4, the gradation value of the developed pattern 21 can be calculated for each pixel, that is, for each pattern coordinate address of the real image 15. . Therefore, the design data developing unit 2 develops a wiring pattern on each pixel based on the design data 11, calculates a bit-integrated value for each pixel as a gradation value, and outputs it as reference data 12.

  Next, in the reference image generation unit 3, enlargement / reduction correction is performed in an appropriate pixel unit or sub-pixel of each wiring pattern width in the reference data 12 output from the design data development unit 2. Move and correct the edge position. Subsequently, based on the corner radius of the actual image 15 at the position corresponding to the corner portion of the reference data 12, the rounding correction processing of the corner portion of the reference data 12 is performed. Thereafter, a blurring process is performed with a filter having optical characteristics of the actual image 15 obtained by optically scanning the object to be inspected, and a blurred reference image 13 close to the actual image 15 is created.

  It should be noted that the blur strength and the blur range of the filter having the optical characteristics of the actual image 15 mounted on the defect detection hardware that performs image comparison in the image defect inspection unit 7 are stored in the image memory alignment unit 6. An optical coefficient representing the amount of blur is created from the image considered as the representative example of the actual image 15 along the gradation distribution at the position where the vertical and horizontal edge portions in the four directions of the pattern to be inspected exist. A blur filter coefficient is created by synthesizing a light spot spread function in which the coefficient is inverted for each trace direction. Subsequently, the reference image generation unit 3 uses the obtained blur filter coefficient to perform a convolution operation on the reference data 12 output from the design data development unit 2 to create a blur reference image 13 to align the image memory. The image memory alignment unit 6 aligns the actual image 15 corresponding to the position address of the design data 11 and the blurred reference image 13 in the image memory, and the information image defect inspection unit 7 A comparison process between the image 15 and the blurred reference image 13 is performed to detect a non-inspected defect portion.

  First, an optical coefficient (filter radius R) that is a standard of blur intensity and range registered in advance in the reference image generation unit 3 that executes the blur filter operation, and a light spot spread function that serves as a blur filter coefficient The generation method of will be described. FIG. 5 shows the range of blurring when the radius R of the light spot spread function of a perfect circle necessary for determining the optical characteristics of the image defect detection hardware is changed, and the filter coefficient representing the weighting of the light intensity in the unit pixel. It is explanatory drawing.

  The lattice area in FIG. 5A represents a unit pixel equal to the development resolution of the design data 11, and the area of the perfect circle occupied by this unit pixel is the filter coefficient occupied by the radius R that is the optical coefficient of the light spot spread function. Shows the strength of In order to calculate the light intensity and spread in sub-pixel units below the inspection resolution, the region that matches the unit pixel is quantized into 16 × 16 divisions (the maximum value of one pixel area is 256), and this sub-pixel is turned on. It consists of a bit string having an OFF value. For example, as shown in FIG. 5B, the filter coefficient of the light spot spread function having the radius R in each unit pixel is integrated with the number of On bits of gray sub-pixels in each unit pixel occupied by the perfect circle region. To calculate. On the other hand, in the blurring process, the filter coefficient value in each unit pixel is normalized by the sum of all the filter coefficients, and the quantization error is negligibly small. Therefore, in this embodiment, the filter area occupied in the unit pixel is the maximum value. When taking 256, the maximum value of the light intensity of the filter is set to 255.

  FIGS. 6A, 6B, 6C, and 6D are explanatory diagrams showing a method of synthesizing filter coefficients of a light spot spread function from combinations of optical coefficients in the trace direction. The filter coefficient value in each filter area is a perfect circle light having optical coefficients of two sets of radius values (upper, right) (right, lower) (lower, left) (left, upper) shown in FIG. This is a set that represents the area value of a circle that occupies each unit pixel in a quarter-divided region of the point spread function. When the 1/4 divided region determined by the combination of the optical coefficients in the trace direction is determined, the upper left 1/4 divided region stored in advance in the reference image generation unit 3 is sequentially rotated or inverted according to the range of the target region. Thus, the filter coefficient values of the entire region of the light spot spread function are synthesized. However, when synthesizing the filter coefficients, as shown in FIG. 6A, the filter coefficients at the pixel positions corresponding to the respective boundaries of the quarter-divided area are considered in consideration of the overlap with other quarter-divided areas. One half of the area value that should actually be stored is stored, and the coefficient sum generated in the overlapping portion is set to a value not exceeding 255.

  For example, in the embodiment shown in FIG. 7, the light spot spread function 7 obtained by synthesizing a region in which the optical coefficients in the respective directions in the upper, lower, left and right directions of FIG. 6 have a radius R = 2.0 [pixels]. It is a filter coefficient in a x7 pixel range.

In this example, the optical coefficient to be a perfect circle is shown, but in addition to the perfect circle, the filter coefficient value of the light spot spread function is calculated by the same calculation in the area of an ellipse composed of optical coefficients with two different radii. May be.
In this way, the reference image generation unit 3 stores in advance filter coefficients that are formed by a combination of optical coefficients that occupy the upper left quarter-divided region of a light spot spread function that is a perfect circle or an ellipse that is obtained in advance by simulation. The reference image generation unit 3 combines the coefficient groups of the upper left quarter-divided area stored in accordance with the combination of the optical coefficients in the two trace directions by rotating or inverting them, and all the filter coefficients of the light spot spread function Is generated.

  Next, the radius R of the light spot spread function having a perfect circle optical coefficient stored in the reference image generation unit 3 and the blur amount α in the trace direction calculated from the luminance profile will be described. In order to simplify the description, reference data will always be present at pixel boundaries, and an example in which the unit pixel at the edge of the pattern takes a maximum gradation value of 255 will be described.

  FIG. 8 sequentially enlarges the radius R of the light spot spread function to be a perfect circle shown in FIG. 5, and the brightness of the peripheral portion of the edge of the blurred reference image 13 when the reference data 12 is blurred by the filter coefficient corresponding to each radius R. A profile is shown. As the radius R increases, the gradient of the brightness profile of the pattern edge becomes gentler, and the blurring of the edge periphery becomes larger. In this way, by making the radius R of the light spot spread function that becomes a perfect circle variable, the edge inclination of the luminance profile when the reference data is blurred, that is, the amount of blur can be controlled.

FIG. 9 shows the brightness value at each pixel position on the brightness profile of the blurred reference image 13 and the blur amount α when the reference data is blurred at each radius R which is the optical coefficient of the light spot spread function of a perfect circle. Is. For example, when the pixel position of interest on the luminance profile at each radius R of the blurred reference image 13 in FIG. 8 is S (x, y) and the luminance value Ls (x, y), the pattern is traced to the left. The amount of blur α is
α = MAX {Ls (x + k, y) −Ls (x, y)} (where k = 2) (Equation 1)
And calculate.

  A blurred reference image having a light spot spread function that is a perfect circle defined by the maximum value of the absolute value difference between the target pixel of the luminance profile calculated from (Equation 1) from FIG. 9 and the neighboring pixel in the trace direction. It is explained that the blur amount α of 13 takes eigenvalues that are uniquely determined for the optical coefficients of the respective radii R of the light spot spread functions having different sizes.

  Further, these optical coefficients have a property that does not depend on the position in the pixel at the edge of the reference data 12. For example, FIGS. 11A, 11B, 11C, and 11D show the tone distribution of the reference data 12 immediately after being developed with an appropriate resolution and the radius R = 2.0 [pixel] in FIG. This is a gradation distribution when the reference data 12 is blurred by a light spot spread function having an optical coefficient occupied by a perfect circle. This pattern is an example of a pattern having a width in the up-down direction that does not exist in the pixel breaks developed by shifting the edge portion by 1/4 pixel, and the numerical value indicates the gradation value at the pixel position.

11 (a), (b), (c), and (d), in the case of N to N + K of the target pixel at an arbitrary edge position developed within one pixel (but, Here, the maximum value of the difference between the absolute values of the luminance values of K = 2) is close to 150.
From FIG. 11A, the maximum luminance value difference between N and N + 2 is 167-18 = 149.
From FIG. 11B, the maximum luminance value difference between N and N + 2: 185−35 = 150
As shown in FIG. 11C, the maximum luminance value difference between N and N + 2 is 202−52 = 150.
From FIG. 11D, the maximum luminance value difference between N and N + 2: 219−70 = 149
In this way, the maximum difference in luminance value between N and N + K (K = 2) in FIGS. 11A and 11D is 149, which is the case in FIGS. 11B and 11C. Although there is a difference that results in a luminance value difference of 150, this is a rounding error in calculation.

  As described above, the blur amount α in the direction traced in the vertical direction to the edge pattern uniquely determined by each radius R of the optical coefficient has the same value regardless of the position of the edge end in the unit pixel of the developed reference data 12. Take. That is, the optical coefficient of the light spot spread function indicated by a perfect circle having a radius R in the trace direction can be obtained stably by detecting the blur amount α and allowing the positional deviation amount of the edge edge in one pixel. Explained.

The relationship between the optical coefficient of the light spot spread function having the radius R of the perfect circle and the blur amount α can be expanded to a radius of any size. FIG. 10 shows the blur amount defined by the radius R of the radius R optical coefficient shown in FIG. 9 and the maximum value of the absolute value difference between the target pixel position and the pixel position in the trace direction at the K-position ahead. This shows the relationship with α. For example, the blur amount α in the case of K = 2 uses the optical coefficient representing the blur amount, and the coefficient of the polynomial is uniquely determined using the least square method from the relationship between the blur amount α and the optical coefficient shown in FIG. Can decide
α = 1.3261R ³ −0.3449R ² −68.085R + 276.5 (Formula 2)
It is calculated by the following third order polynomial.

  In this way, the blur amount α in the trace direction on the luminance profile can be uniquely calculated from the optical coefficient of the radius R of an arbitrary size that becomes a perfect circle by (Equation 2). Further, when the blur amount α is known, the radius R of the corresponding optical coefficient is obtained using (Equation 1) and (Equation 2), so that the blur strength and the blur range are obtained as in the embodiment of FIG. It is possible to visually and quantitatively obtain a light spot spread function of a perfect circle indicating

  Next, a method for designing the blur amount of the actual image 15 using the optical coefficient R and the blur amount α stored in the reference image generating unit 3 will be described with reference to the drawings. Regarding the blur strength and the blur range of the filter having optical characteristics, the vertical direction in the four directions of the pattern to be inspected is determined from the image regarded as the representative example among the actual images 15 stored in the image memory alignment unit 6. Based on the gradation distribution at the position where the horizontal edge portion exists, tracing is performed in a direction perpendicular to the edge pattern of the actual image 15.

FIG. 12 shows a method of calculating the blur amount β when the edge pattern of the real image 15 is traced in the left direction. When the coordinate position on the luminance profile on the actual image 15 in FIG. 12 is P (x, y) and the luminance value of the pattern at that position is Lp (x, y), tracing is performed in the left direction from the search start point 25. Β = MAX {Lp (x + k, y) −Lp (x, y)} (where k> = 2) (Equation 3)
And calculate. Similar processing is performed for the remaining upper, lower, and right edge patterns, and the blur amount β in each edge pattern direction of the actual image 15 is calculated. However, in order to avoid the influence of noise, light fluctuation, and light wraparound, the trace width 26 in the actual image 15 is set to about 7 pixels, and the luminance average within the trace width 26 is set as a representative luminance value, and (Expression 3) is obtained. Is used to calculate the blur amount β. Subsequently, the blur amount β of the actual image 15 is compared with the blur amount α obtained by the light spot spread function of a perfect circle stored in the reference image generation unit 3, and the radius of the blur amount α when the error amount is the smallest. The value is selected as the optical coefficient in the trace direction of the real image 15.

  FIG. 12 shows the blur state of the actual image 15 obtained by tracing the pattern edges in each direction of the actual image 15 in FIG. The blur intensity and range of the actual image 15 corresponding to the optical coefficient R in each direction in FIG. 12 is a light spot spread function obtained by synthesizing an elliptical filter region divided into quarters.

  Next, a configuration method of a blur processing filter of a light spot spread function that is applied to the reference data 12 using the optical coefficient obtained from the actual image 15 will be described with reference to the drawings. FIG. 14 is an explanatory diagram showing the relationship between the edge portion of the blur reference image 13 and the blur filter coefficient. When an optical coefficient is extracted from the luminance profile of the pattern edge in each direction, the luminance value of each pixel on the pattern edge is affected by the light intensity and spread in the direction in which no pattern exists.

  As shown in FIG. 14, when there is a light spread in a direction in which no pattern exists, the luminance value of the edge portion becomes “smooth”. That is, in order to obtain a blurred reference image 13 that is similar to the luminance distribution of the edge at the corresponding pixel position in the actual image 15, the optical coefficient of the light spot spread function that becomes a blurred filter applied to the position of the reference data 12 is It is necessary to assign an optical coefficient for a direction in which no pattern exists (a direction in which the luminance value decreases). Therefore, as shown in the embodiment of FIG. 15, the light spot spread function necessary for the blurring process on the reference data 12 is an optical coefficient obtained by inverting the optical coefficient in the scanning direction obtained in the actual image 15 of FIG. It is a light spot spread function. A convolution process is performed on the reference data 12 by using a light spot spread function serving as a blur filter having the optical coefficient obtained in this manner, and a blur reference image 13 is created.

  In order to supplement the description of the operation of the light spot spread function that becomes a blurring filter that blurs the reference data 12 described above, here, a description will be given using a representative image that traces the pattern edge. FIG. 16A shows the multi-valued reference data 12 developed in the image memory alignment unit 6, and FIG. 16B shows an implementation in which the blurring process is performed on FIG. 16A based on the conventional Gaussian distribution. For example, FIG. 16C shows a real image 15 obtained by optical scanning of an actual inspection apparatus, and FIG. 16D shows a light spot spread function shown in FIG. 15 extracted from the real image 15 shown in FIG. 4 shows the processing result obtained by performing a convolution operation on the multi-valued reference data 12 using.

  The light spot spread function based on the Gaussian function of FIG. 16B is isotropic, whereas the light spot spread function blurred from the real image 15 of FIG. 16C is blurred. The blur reference image 13 shown in (d) has different blur directions at the top, bottom, left, and right edges. The blur reference image 13 having an asymmetric blur characteristic very similar to that of the actual image 15 can be generated. Has been. In addition, it is shown that the light intensity distribution having a shape similar to the roundness of the actual image 15 is realized in the blurring of the corner portion.

  In order to show more specific examples of the effects of the blur filter design method and the blur generation reference image 13 automatic generation method, here, description will be given using the luminance distribution of the inspection target having a general-purpose wiring pattern. 17A shows an actual image 15 of a wiring pattern having a defect drawn on a glass substrate obtained by optical scanning of image defect inspection hardware, and FIG. 17B shows a resize and a corner suitable for the reference data 12. It is the blurring reference image 13 obtained by performing a blurring process with the light spot spread function with the optical coefficient shown in FIG. 18 extracted from the real image 15 of FIG. The linear pattern shown in FIG. 19 corresponds to the luminance distribution of the real image 15 in the 27 region of FIG. 17A, and FIG. 20 corresponds to 27 region of the real image 15 shown in FIG. 17A of FIG. FIG. 21 shows a luminance distribution of a difference image between the real image 15 in FIG. 19 and the blurred reference image 13 in FIG. As is clear from FIGS. 19 and 20, the blurred reference image 13 is very similar to the luminance distribution of the actual image 15, and is within a difference of about 25 gradations from the difference image of FIG.

  22 to 24 show a comparison between the real image 15 and the blurred reference image 13 regarding the luminance distribution of the wiring pattern including the outer corner and the inner corner shown in FIG. is there. The luminance distribution of the normal pattern without any defects and the luminance distribution of the corner portion of the blurred reference image 13 shown in FIG. 23 is very similar to the luminance distribution of the real image 15 of FIG. 22 at the corresponding position. Further, in the difference image shown in FIG. 24, since only the luminance value of the defective portion is higher than the surrounding luminance value, the defective portion on the edge can be easily detected.

  In this way, the reference image generation unit 3 in FIG. 1 performs resize processing on the input reference data 12 in units of pixels and sub-pixels, performs appropriate corner correction processing, and then obtains the obtained light spot spread. A blurring filter 13 having a function as a filter coefficient is applied to the reference data 12 to create a blurring reference image 13 close to the luminance distribution of the actual image 15. Subsequently, the blur reference image 13 is stored in the image memory of the image memory alignment unit 6, and the actual image 15 and the blur reference image 13 are aligned at the position address corresponding to the design data 11, and then the image comparison is performed. The defect detection unit 7 compares the actual image 15 already stored and inspects whether or not the pattern to be inspected is accurately drawn based on the size and shape of the ideal pattern created from the design data 11 Is done.

  It should be noted that the configuration and operation of the above-described embodiment are examples, and it goes without saying that they can be changed as appropriate without departing from the spirit of the present invention.

  The present invention can also be applied to a case where a blurred reference image is automatically generated.

It is a block diagram of the image comparison defect detection hardware which mounted the blurring filter design method and blurring reference image automatic generation method by one embodiment of this invention. It is explanatory drawing which shows a part of developed | deployed wiring corner pattern. FIG. 3 is an enlarged view of one pixel in FIG. 2. It is explanatory drawing which shows the gradation value of each pixel of the pattern of FIG. It is explanatory drawing which shows the intensity | strength and range of the blur of the light spot expansion function of a perfect circle when the range of the radius R is expanded. It is explanatory drawing which shows the synthetic | combination method of the blurring filter coefficient using the coefficient of the light spot spread function defined in the 1/4 division | segmentation area | region. It is explanatory drawing which shows the light spot spreading function of the perfect circle defined with the circular mask pattern (radius R = 2.5). It is explanatory drawing which shows the relationship between the radius R of the light spot spread function of FIG. 5, and edge inclination. It is explanatory drawing which shows the numerical relationship between the luminance profile of the radius R used as an optical coefficient, and blur amount (alpha). It is explanatory drawing which shows the relational expression of the radius R of the light spot spread function when the filter radius R is expanded, and the blur amount α. It is explanatory drawing which shows the relationship between the edge position in case the edge of the expand | deployed reference data is not a pixel break, and blur amount. It is explanatory drawing which shows the method of tracing the edge pattern of a real image and calculating | requiring an optical coefficient. It is explanatory drawing which shows the light intensity distribution and the range of the light spot spreading function which a real image has when tracing a real image. It is explanatory drawing which shows the relationship between the luminance distribution of the trace direction of reference data, and the luminance profile produced | generated from the coefficient area | region of the light spot spread function obtained from a real image. It is explanatory drawing which shows the light spot spread function used as the blurring filter coefficient required for the blurring process of the reference data produced | generated from the coefficient area | region of the light spot spread function obtained from a real image. It is explanatory drawing of a to-be-inspected image, reference data, and a blurring reference image required in order to produce | generate a light spot spread function. It is explanatory drawing which shows the relationship between the real image of the wiring pattern containing a defect, and the blurring reference image in a corresponding position. It is explanatory drawing which shows the light spot spreading function required for the blurring process of the reference data obtained from the real image pattern. It is explanatory drawing which showed numerically the gradation distribution of the linear pattern of a real image. It is explanatory drawing which showed numerically the gradation distribution of the blurring reference image in the position corresponding to the linear pattern of a real image. It is explanatory drawing which showed numerically the gradation distribution of the difference image of the real image and blurring reference image in a linear pattern. It is explanatory drawing which showed numerically the diagonal pattern with a defect of a real image, and the gradation distribution of a corner part. It is explanatory drawing which showed the gradation distribution of the blurring reference image in the position which the diagonal pattern with a defect of a real image and a corner part respond | correspond numerically. It is explanatory drawing which showed numerically the gradation distribution of the difference image of the blurring reference image in the position corresponding to the real image which has a diagonal pattern and a corner part with a defect.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Design data input part, 2 ... Design data expansion | deployment part, 3 ... Reference image generation part, 4 ... Optical scanning part, 5 ... Photoelectric image processing part, 6 ... Image memory alignment part 7 ... Image comparison defect detection part, DESCRIPTION OF SYMBOLS 11 ... Design data, 12 ... Reference data, 13 ... Blur reference image, 14 ... Scanning signal, 15 ... Actual image

Claims (10)

  Scanning the light-shielding part formed based on the design data, the transmissive part, and the fine pattern part of the inspection object having a fine pattern, and receiving the transmitted light obtained through the inspection object by the objective lens such as a CCD An image is formed on the element, a real image is generated from the pattern information obtained from the light receiving element, a filter obtained from the pattern information of the real image is applied to the design data, and the scale of the blurred image of the obtained reference data is obtained. A gradation filter design method in which the tone distribution accurately reproduces the tone distribution of the actual image, and the design data is developed as a pattern consisting of multi-tone values on each pixel having a resolution finer than the inspection resolution. The selected arbitrary coordinates on the pattern edge part from the reference data that does not have any blur information and the actual image pattern corresponding to the pattern edge part position of this reference data Blur filter design method characterized by a light spot spread function, which is generated by extracting the blur information by tracing an edge in the vertical direction as the filter coefficients.   2. The blur filter design method according to claim 1, wherein the absolute value of the absolute value of the target pixel position and the target pixel position + second luminance difference on the luminance profile obtained by tracing the pattern edge of the actual image in the vertical direction is defined as the actual image edge inclination. From the actual image edge inclination, the area value within the pixel that occupies each pixel area within the range of a perfect circle or an ellipse determined by the combination of the two directions adjacent to each other among the upper, lower, left, and right A blur filter design method characterized in that a radius value of a coefficient is used as an optical coefficient in each direction of a real image, and a blur filter coefficient region and direction that are a light spot spread function of the real image are determined from the optical coefficient.   3. The optical coefficient of the blur filter design method according to claim 2, wherein a target pixel and a target pixel + k (provided that the brightness profile is traced in a vertical direction with respect to the pattern edges in the upper, lower, right, and left directions of the real image. , K> = 1) of the actual image edge inclination that is the maximum value of the luminance difference, it occupies in each pixel area within the range of the circle or ellipse determined by the combination of the radius values of the optical coefficients in the two adjacent directions A blurring filter design method, wherein an area value in a pixel is determined as an optical coefficient corresponding to light intensity, and is used as a blurring filter coefficient. 4. The blur filter design method according to claim 3, wherein each of the blur filters having a radius value having an area value occupied by each unit pixel within a range of a perfect circle drawn by a radius value matching the optical coefficient as a blur filter coefficient value. The pattern edge on the blurred reference image is traced in the vertical direction on each blurred reference image obtained by sequentially performing the convolution operation on the quantified reference data, and the target image position and the target pixel position on the luminance profile + second. Α is the maximum value of the edge inclination of the blurred reference image that is the maximum absolute value of the luminance difference, and the arbitrary radius value R of the light spot spread function that is the radius value of the optical coefficient is the real coefficient a, b, c, α = aR ³ + bR ² + cR + d with d (a, b, c, d are real numbers (a ≠ 0)
A blur filter design method characterized in that it is obtained only by the relationship of the cubic equation.   5. The blur filter design method according to claim 4, wherein each of the blur filters having a radius filter coefficient corresponding to an area value of each pixel within a radius of a perfect circle corresponding to the optical coefficient is a multi-value. The edge slope α described above on each blurred reference image obtained by performing successive convolution operations on the converted reference data is any arbitrary value within the edge of the reference data that is multi-valued based on the design data. A blur filter design method characterized by taking the same neighborhood value at a position.   6. The blur filter design method according to claim 5, wherein the radius of the optical coefficient in two adjacent directions among the four optical coefficients traced in the vertical direction with respect to the pattern edges in the upper, lower, right and left directions of the real image. The pixel area value in the area where the value touches is defined as the light spot spread function in the trace direction, and the light spot spread function in the trace direction occupies in the unit pixel touching the area of a perfect circle or ellipse affected by the light intensity. Generate a blur filter of the real image with the area ratio as the filter coefficient in each trace direction, create a new optical coefficient generated by inverting the optical coefficient in the direction not applied to the pattern among each optical coefficient, Fills the area of each unit pixel within the range where the ellipse touches the perfect circle or the ellipse with the radius corresponding to this optical coefficient. Blur filter design method characterized by a coefficient.   7. The blur filter design method according to claim 6, wherein a radius value corresponding to each optical coefficient is calculated from new four-direction optical coefficients generated by inverting the blur filter coefficient corresponding to the optical coefficient in each trace direction of the real image. The area value occupying the unit pixel within the range where the perfect circle or ellipse shape touches is used as the filter coefficient of the light spot spread function, and the convolution processing is performed on the reference data multi-valued by the blur filter having this filter coefficient. A blur filter design method, characterized in that blur images having various brightness profiles with different blurring methods can be created for each direction of reference data.   Scanning the light-shielding part formed based on the design data, the transmissive part, and the fine pattern part of the inspection object having a fine pattern, and receiving the transmitted light obtained through the inspection object by the objective lens such as a CCD An image is formed on the element, a real image is generated from the pattern information obtained from the light receiving element, a filter obtained from the pattern information of the real image is applied to the design data, and the blurred image of the obtained reference data is obtained. In the blur filter design method in which the tone distribution accurately reproduces the tone distribution of the actual image, a blur reference image that is precisely similar to the blur of the actual image is generated by generating the light spot spread function according to claims 1 to 7. A blur filter design method comprising steps.   Scanning the light-shielding part formed based on the design data, the transmissive part, and the fine pattern part of the inspection object having a fine pattern, and receiving the transmitted light obtained through the inspection object by the objective lens such as a CCD An image is formed on the element, a real image is generated from the pattern information obtained from the light receiving element, a filter obtained from the pattern information of the real image is applied to the design data, and the scale of the blurred image of the obtained reference data is obtained. A gradation filter design method in which the tone distribution accurately reproduces the tone distribution of the actual image, and the design data is developed as a pattern consisting of multi-tone values on each pixel having a resolution finer than the inspection resolution. The selected arbitrary coordinates on the pattern edge part from the reference data that does not have any blur information and the actual image pattern corresponding to the pattern edge part position of this reference data Blur filter manufactured by the blur filter design method with a light spot spread function, which is generated by extracting the blur information by tracing an edge in a direction perpendicular to the filter coefficients. Scanning the light-shielding part formed based on the design data, the transmissive part, and the fine pattern part of the inspection object having a fine pattern, and receiving the transmitted light obtained through the inspection object by the objective lens such as a CCD An image is formed on the element, a real image is generated from the pattern information obtained from the light receiving element, a filter obtained from the pattern information of the real image is applied to the design data, and the scale of the blurred image of the obtained reference data is obtained. A gradation filter design system in which a tone distribution precisely reproduces a gradation distribution of an actual image, and includes an optical scanning unit that scans a fine pattern of the inspection target and outputs a scanning signal, and an optical scanning unit A photoelectric image processing unit that converts the output scanning signal into a multi-valued gray scale actual image and outputs it, a design data input unit that inputs design data, and the design data input unit The design data development means for developing the received design data as a pattern of multi-gradation values on each pixel having a resolution finer than the inspection resolution and generating reference data, and based on the edge position of the actual image And a reference image generation means for performing rounding gradation correction of the corner portion at the inspection resolution or less while maintaining the width of each pattern of the reference data at multiple gradations, and performing blurring processing by a light spot spread function to generate a blurred reference image An image memory alignment means for aligning the actual image and the blur reference image with a corresponding image position address in a predetermined image memory; and comparing the luminance of the actual image and the blur reference image with the pattern A blur filter design system comprising image comparison defect detection means for detecting a defect of the image.

Images