JP4970570B2 - Pattern inspection device, image alignment method and program - Google Patents

Description

  The present invention relates to a pattern inspection apparatus, an image alignment method, or a program for causing a computer to execute the method, and relates to, for example, a pattern inspection technique for inspecting a pattern defect of an object serving as a sample used in semiconductor manufacturing. The present invention relates to an apparatus for inspecting a defect of an extremely small pattern such as a photomask, a wafer, or a liquid crystal substrate used for manufacturing an element or a liquid crystal display (LCD).

  In recent years, the circuit line width required for a semiconductor element has been increasingly narrowed as a large scale integrated circuit (LSI) is highly integrated and has a large capacity. These semiconductor elements use an original pattern pattern (also referred to as a mask or a reticle, hereinafter referred to as a mask) on which a circuit pattern is formed, and the pattern is exposed and transferred onto a wafer by a reduction projection exposure apparatus called a stepper. It is manufactured by forming a circuit. Therefore, a pattern drawing apparatus capable of drawing a fine circuit pattern is used for manufacturing a mask for transferring such a fine circuit pattern onto a wafer. A pattern circuit may be directly drawn on a wafer using such a pattern drawing apparatus. Alternatively, development of a laser beam drawing apparatus for drawing using a laser beam in addition to an electron beam has been attempted.

  In addition, improvement in yield is indispensable for manufacturing an LSI that requires a large amount of manufacturing cost. However, as represented by a 1 gigabit class DRAM (Random Access Memory), the pattern constituting the LSI is about to be in the order of submicron to nanometer. One of the major factors that reduce the yield is a pattern defect of a mask used when an ultrafine pattern is exposed and transferred onto a semiconductor wafer by a photolithography technique. In recent years, with the miniaturization of LSI pattern dimensions formed on semiconductor wafers, the dimensions that must be detected as pattern defects have become extremely small. Therefore, it is necessary to improve the accuracy of a pattern inspection apparatus that inspects defects in a transfer mask used in LSI manufacturing.

  On the other hand, with the development of multimedia, LCDs (Liquid Crystal Display) are increasing in size of the liquid crystal substrate to 500 mm × 600 mm or more, and TFTs (Thin Film Transistors) formed on the liquid crystal substrate. : Thin film transistors) and the like are being miniaturized. Therefore, it is required to inspect a very small pattern defect over a wide range. For this reason, there is an urgent need to develop a sample inspection apparatus for efficiently inspecting defects of a photomask used in manufacturing such a large area LCD pattern and a large area LCD in a short time.

  Here, in a conventional pattern inspection apparatus, an optical image obtained by imaging a pattern formed on a sample such as a lithography mask using a magnifying optical system at a predetermined magnification and an identical pattern on the sample are captured. It is known to perform an inspection by comparing with an optical image. For example, as a pattern inspection method, a “die to die inspection” that compares optical image data obtained by imaging the same pattern at different locations on the same mask, or a drawing apparatus that draws a pattern using pattern-designed CAD data as a mask Drawing data (design pattern data) converted into a device input format for input to the inspection device is input to the inspection device, design image data (reference image) is generated based on the drawing data, and measurement data obtained by imaging the pattern There is a “die to database test” that compares an optical image. In the inspection method in such an inspection apparatus, the sample is placed on the stage, and the stage is moved so that the light beam scans on the sample and the inspection is performed. The sample is irradiated with a light beam by a light source and an illumination optical system. The light transmitted or reflected by the sample is imaged on the sensor via the optical system. The image picked up by the sensor is sent to the comparison circuit as measurement data. The comparison circuit compares the measured data and the reference data according to an appropriate algorithm after the images are aligned, and determines that there is a pattern defect if they do not match.

  Here, the reference image and the optical image are compared for each region having a predetermined size, but in order to compare, the reference image and the optical image need to be aligned with high accuracy. For this reason, the position deviation amount between the reference image and the optical image is first corrected for each pixel by the residual sum of squares calculation, and then the position correction for each sub-pixel is further performed from the position by the least square method. (See, for example, Patent Document 1).

  However, when the position is corrected by the residual sum of squares calculation for each pixel, there is a problem that the calculation may be performed so as to align the position with the graphic next to the graphic that should be aligned. Such a problem is likely to occur in, for example, a pattern in which similar figures are repeatedly arranged, such as a line and space pattern, or a figure pattern having an arbitrary angle side other than 45 degrees. If the alignment correction is performed on the adjacent graphic in units of pixels, the subsequent correction by the least square method results in correction in units of sub-pixels, which makes it difficult to correct the original graphic position to be aligned.

JP 2007-071630 A

  In order to compare the reference image and the optical image, it is necessary to align the reference image and the optical image with high accuracy. However, as the pattern size is reduced, the relative relationship between the reference image and the optical image is increased. It has become difficult to detect displacement with high accuracy. In particular, as described above, there is a problem in that different figures are aligned at the initial pixel-by-pixel alignment stage.

  An object of the present invention is to provide a method and an apparatus for overcoming the above-described problems and performing highly accurate alignment between a reference image and an optical image.

The pattern inspection apparatus according to one aspect of the present invention includes:
An optical image acquisition unit for acquiring an optical image of a test sample on which a pattern is formed;
Based on the design data of the sample to be inspected, a reference image creating unit that creates a reference image to be compared with the optical image;
Using another plurality of reference images shifted in parallel based on the alignment position of the optical image and the reference image, the optical image between the optical image and each reference image of the plurality of other reference images And the residual sum of squares that is the smallest of the calculated residual sums of squares. A residual sum-of-squares calculation unit that acquires the amount of misalignment in the combination for which
By the least-squares method of calculating using a value obtained by spatially differentiating in the x direction and a value obtained by spatially differentiating the reference image acquired by the residual sum of squares computing unit in the x direction A least-squares method computing unit that computes the amount of displacement from the alignment position of the image and the corrected reference image;
A comparison unit that compares the optical image with the reference image corrected using the residual sum of squares and the least squares method;
With
Starting with the calculation by the residual sum of squares calculation unit, the residual sum of squares calculation unit and the least squares method calculation unit alternately repeat the same number of times multiple times,
The comparison unit compares the optical image with a reference image corrected to a position obtained as a result of repeating the calculation a plurality of times.

An image alignment method of one embodiment of the present invention includes:
A first step of substituting a first reference image created for comparison with an optical image of a sample to be inspected with a pattern into a second reference image;
A second step of calculating a position correction amount by a parallel shift that minimizes a residual sum of squares of the second reference image and the optical image;
A third step of obtaining a new second reference image by correcting the position of the first reference image by the first parameter indicating the position correction amount by the parallel shift calculated in the second step;
A fourth step of modeling the parallel shift, expansion / contraction, rotation, and image intensity variation of the optical image with respect to the new second reference image as an unknown second parameter, and calculating the second parameter by a least square method;
A fifth step of image-correcting the first reference image by the second parameter calculated in the fourth step to obtain a new second reference image;
With
Repeat the second to fourth steps N times, the fifth step N-1 times, the second to fifth steps in order,
After the Nth fourth step, the second second reference image used in the Nth fourth step is image-corrected by the second parameter calculated in the Nth fourth step and the third parameter is corrected. A sixth step of obtaining and outputting the reference image is further provided.

The pattern inspection apparatus according to another aspect of the present invention includes:
An optical image acquisition unit for acquiring an optical image of a test sample on which a pattern is formed;
A reference image creating unit that creates a first reference image to be compared with the optical image based on the design data of the sample to be inspected;
Using the first different reference images shifted in parallel based on the first alignment position of the optical image and the first reference image, the optical image and the first different plurality of reference images are used. And calculating a residual sum of squares of the pixel value of the optical image and the pixel value of each reference image of the first plurality of other reference images with each reference image of the reference image. A first calculation unit that obtains a first positional shift amount in a combination in which a minimum residual square sum is calculated from each residual square sum;
The optical image and the first reference image are calculated by a least square method that uses a value obtained by spatial differentiation in the x direction and a value obtained by spatial differentiation in the y direction of the second reference image in which the first positional deviation amount is corrected. A second calculation unit that calculates a second positional deviation amount from the second alignment position with the second reference image;
The (2n−1) th (2n−2) th (where 2 ≦ n ≦ m, where n and m are integers) corrected positional shift amounts in the optical image and the first reference image. Using the plurality of nth different reference images shifted in parallel based on the (2n-1) th matching position with the reference image of the reference image, the optical image and the other plurality of the nth reference images Calculating a residual sum of squares between the pixel value of the optical image and the pixel value of each reference image of the nth other reference images between each reference image of the image, A (2n-1) th arithmetic unit that obtains the (2n-1) th positional deviation amount in the combination in which the minimum residual square sum is calculated from the residual square sums;
Using a value obtained by spatially differentiating in the x direction and a value obtained by spatially differentiating in the y direction, the first reference image obtained by correcting the (2n-1) th positional deviation amount in the first reference image. A (2n) -th computing unit that computes a (2n) -th positional shift amount from the alignment position of the optical image and the (2n) reference image by a least-squares method to be computed;
The (2m + 1) th positional deviation amount corrected with respect to the (2m) th reference image with the (2m-1) th positional deviation amount corrected with respect to the first reference image. A comparison unit for comparing a reference image and the optical image;
It is provided with.

An image alignment method of one embodiment of the present invention includes:
An image alignment method for aligning an optical image and a reference image used for comparative inspection of a test sample on which a pattern is formed,
The first different reference images shifted in parallel based on the first alignment position of the optical image and the first reference image are used, and the optical image and the first other plurality of reference images are used. A residual sum of squares of the pixel value of the optical image and the pixel value of each reference image of the first plurality of reference images is calculated between each reference image of the reference image, and the calculation is performed. A first calculation step of obtaining a first positional deviation amount in a combination in which a minimum residual square sum is calculated from each residual square sum;
The optical image and the first reference image are calculated by a least square method that uses a value obtained by spatial differentiation in the x direction and a value obtained by spatial differentiation in the y direction of the second reference image in which the first positional deviation amount is corrected. A second calculation step of calculating a second positional deviation amount from the second alignment position with the second reference image;
The (2n−1) th (2n−2) th (where 2 ≦ n ≦ m, where n and m are integers) corrected positional shift amounts in the optical image and the first reference image. Using the plurality of nth different reference images shifted in parallel based on the (2n-1) th matching position with the reference image of the reference image, the optical image and the other plurality of the nth reference images Calculating a residual sum of squares between the pixel value of the optical image and the pixel value of each reference image of the nth other reference images between each reference image of the image, A (2n-1) th calculation step of obtaining a (2n-1) th positional shift amount in a combination in which the minimum residual square sum is calculated from the residual square sum;
Using a value obtained by spatially differentiating in the x direction and a value obtained by spatially differentiating in the y direction, the first reference image obtained by correcting the (2n-1) th positional deviation amount in the first reference image. A (2n) th calculation step of calculating a (2n) th positional shift amount from the alignment position of the optical image and the (2n) reference image by a least square method to be calculated;
The (2m + 1) th (2m + 1) th position shift amount is generated by correcting the (2m) th position shift amount with respect to the (2m) th reference image in which the (2m-1) th position shift amount is corrected in the first reference image. ) A reference image and the optical image are aligned and output,
It is provided with.

A program for causing a computer of one embodiment of the present invention to execute is as follows.
A storage process for storing in the storage device the optical image and the first reference image used for the comparative inspection of the test sample on which the pattern is formed;
The optical image and the first reference image are read from the storage device, and a plurality of first different plurality of signals shifted in parallel based on a first alignment position of the optical image and the first reference image. Using a reference image, each reference of the pixel value of the optical image and the first plurality of reference images between the optical image and each reference image of the first other plurality of reference images The first sum of the squares of the residual values with the pixel values of the image is calculated, and the first misregistration amount in the combination in which the minimum residual sum of squares is calculated from among the calculated residual sums of squares. A first calculation process to be acquired;
The optical image and the first reference image are calculated by a least square method that uses a value obtained by spatial differentiation in the x direction and a value obtained by spatial differentiation in the y direction of the second reference image in which the first positional deviation amount is corrected. A second calculation process for calculating a second positional shift amount from the second alignment position with the second reference image;
The (2n−1) th (2n−2) th (where 2 ≦ n ≦ m, where n and m are integers) corrected positional shift amounts in the optical image and the first reference image. Using the plurality of nth different reference images shifted in parallel based on the (2n-1) th matching position with the reference image of the reference image, the optical image and the other plurality of the nth reference images Calculating a residual sum of squares between the pixel value of the optical image and the pixel value of each reference image of the nth other reference images between each reference image of the image, A (2n-1) th calculation process for obtaining a (2n-1) th positional deviation amount in a combination in which the minimum residual square sum is calculated from the residual square sums;
Using a value obtained by spatially differentiating in the x direction and a value obtained by spatially differentiating in the y direction, the first reference image obtained by correcting the (2n-1) th positional deviation amount in the first reference image. A (2n) th calculation process for calculating a (2n) th positional shift amount from the alignment position of the optical image and the (2n) reference image by a least square method to be calculated;
The (2m + 1) th (2m + 1) th position shift amount is generated by correcting the (2m) th position shift amount with respect to the (2m) th reference image in which the (2m-1) th position shift amount is corrected in the first reference image. ) The reference image and the optical image are aligned and output,
It is provided with.

  According to the present invention, even when different figures are aligned by the first residual sum of squares calculation, the positions are corrected so that desired figures are aligned by the subsequent residual sum of squares calculation. it can. Therefore, highly accurate alignment between the reference image and the optical image can be performed. As a result, highly sensitive inspection can be performed.

1 is a conceptual diagram showing a configuration of a pattern inspection apparatus in Embodiment 1. FIG. FIG. 3 is a block diagram illustrating a configuration of an alignment circuit in the first embodiment. FIG. 4 is a flowchart showing main steps of the pattern inspection method in the first embodiment. 6 is a diagram for explaining an optical image acquisition procedure according to Embodiment 1. FIG. 6 is a diagram for explaining an SSD calculation method according to Embodiment 1. FIG. FIG. 6 is a diagram illustrating a model formula for positional deviation amount calculation based on the least square method in the first embodiment. 6 is a diagram illustrating a correlation determinant based on the least square method in Embodiment 1. FIG. It is a figure which shows an example of the alignment condition at the time of performing SSD alignment and least square method alignment for every time. FIG. 6 is a diagram illustrating an example of an alignment situation when a plurality of SSD alignments and least squares alignment are performed in the first embodiment. It is a figure for demonstrating another optical image acquisition method.

Embodiment 1 FIG.
FIG. 1 is a conceptual diagram showing the configuration of the pattern inspection apparatus according to the first embodiment. In FIG. 1, a pattern inspection apparatus 100 that samples a substrate such as a mask or wafer on which a pattern is formed and inspects defects in the pattern on the sample includes an optical image acquisition unit 150 and a control system circuit 160. . The optical image acquisition unit 150 includes an XYθ table 102, a light source 103, an enlargement optical system 104, a photodiode array 105, a sensor circuit 106, a laser length measurement system 122, an autoloader 130, and an illumination optical system 170. In the control system circuit 160, the control computer 110 serving as a computer receives a position circuit 107, a comparison circuit 108, a development circuit 111, a reference circuit 112, a positioning circuit 140, and an autoloader control circuit 113 via a bus 120 serving as a data transmission path. Are connected to a table control circuit 114, a magnetic disk device 109, a magnetic tape device 115, a flexible disk device (FD) 116, a CRT 117, a pattern monitor 118, and a printer 119. The XYθ table 102 is driven by an X-axis motor, a Y-axis motor, and a θ-axis motor. In FIG. 1, description of components other than those necessary for describing the first embodiment is omitted. It goes without saying that the pattern inspection apparatus 100 usually includes other necessary configurations.

FIG. 2 is a block diagram showing a configuration of the alignment circuit in the first embodiment.
2, the alignment circuit 140 includes a reference data memory 302, a measurement data memory, an SSD (Sum of the Squared Difference) alignment calculation unit 310, a least squares alignment calculation unit 320, and memories 311 and 321. . In the SSD alignment calculation unit 310, a plurality of pixel unit SSD calculation units 312 (a,... N,... M) and a plurality of pixel unit alignment units 314 (a,... N,... M ), A plurality of sub-pixel unit SSD calculation units 316 (a,... N,... M), and a plurality of sub-pixel unit alignment units 318 (a,... N,... M). ing. In the least squares alignment calculation unit 320, a plurality of least squares method displacement calculation units 322 (a,... N,... M) and a plurality of least squares alignment units 324 (a,... n,... m) are arranged. (However, n is 2 ≦ n ≦ m and is an integer of 2 or more. M is an integer of 2 or more and is the maximum value that n can take.)

  The alignment circuit 140 receives the reference data from the reference circuit 112 and the measurement data from the optical image acquisition unit 150, and after aligning, outputs the reference data and the measurement data to the comparison circuit 108. Data input into the SSD alignment calculation unit 310 and data calculated within the SSD alignment calculation unit 310 are stored in the memory 311 each time. Data input into the least square method alignment calculation unit 320 and data calculated in the least square method alignment calculation unit 320 are stored in the memory 321 each time.

  FIG. 3 is a flowchart showing main steps of the pattern inspection method according to the first embodiment. In FIG. 3, the sample inspection method performs a series of steps of an optical image acquisition step (S102), a reference data creation step (S104), an alignment step, and a comparison step (S250). As an alignment process as an example of the image alignment method, a storage process (S202), a first SSD alignment process (S210), a first least squares alignment process (S220),. n-th SSD alignment step and n-th least-squares alignment step)... last m-th SSD alignment step (S230) and final m-th least-squares alignment step (S240) ) Is performed.

  In the first SSD alignment step (S210), a pixel unit SSD calculation step (S214), a pixel unit alignment step (S214), a sub pixel unit SSD calculation step (S216), and a sub pixel unit alignment step (S218). Perform a series of internal steps. Similarly, in the m-th SSD alignment step (S230), a pixel unit SSD calculation step (S234), a pixel unit alignment step (S234), a sub pixel unit SSD calculation step (S236), and a sub pixel unit alignment step ( A series of internal processes of S238) are performed.

  In the first least square method alignment step (S220), a series of internal steps of a least square method displacement calculation step (S222) and a least square method alignment step (S224) are performed. Similarly, in the m-th least square method alignment step (S240), a series of internal steps of a least square method displacement calculation step (S242) and a least square method alignment step (S244) are performed.

As the optical image acquisition step (S102), the optical image acquisition unit 150 acquires an optical image on the photomask 101 that is a sample on which the graphic indicated by the graphic data included in the design data is drawn based on the design data. Specifically, the optical image is acquired as follows.
A photomask 101 to be inspected is placed on an XYθ table 102 provided so as to be movable in a horizontal direction and a rotation direction by motors of XYθ axes, and the pattern formed on the photomask 101 includes an XYθ table. Light is emitted by a suitable light source 103 disposed above 102. The light beam emitted from the light source 103 irradiates the photomask 101 serving as a sample via the illumination optical system 170. A magnifying optical system 104, a photodiode array 105, and a sensor circuit 106 are disposed below the photomask 101, and light that has passed through the photomask 101 that is a sample such as an exposure mask passes through the magnifying optical system 104. Then, an image is formed as an optical image on the photodiode array 105 and is incident thereon.

FIG. 4 is a diagram for explaining an optical image acquisition procedure according to the first embodiment.
As shown in FIG. 4, the inspection area is virtually divided into a plurality of strip-shaped inspection stripes having a scan width W in the Y direction, and each of the divided inspection stripes is continuously scanned. Thus, the operation of the XYθ table 102 is controlled, and an optical image is acquired while moving in the X direction. In the photodiode array 105, images having a scan width W as shown in FIG. 4 are continuously input. Then, after acquiring the image of the first inspection stripe, the image of the scan width W is continuously input in the same manner while moving the image of the second inspection stripe in the opposite direction. When an image in the third inspection stripe is acquired, the image moves while moving in the direction opposite to the direction in which the image in the second inspection stripe is acquired, that is, in the direction in which the image in the first inspection stripe is acquired. To get. In this way, it is possible to shorten a useless processing time by continuously acquiring images.

  The pattern image formed on the photodiode array 105 is photoelectrically converted by the photodiode array 105 and further A / D (analog-digital) converted by the sensor circuit 106. A sensor such as a TDI sensor is installed in the photodiode array 105. By continuously moving the XYθ table 102 serving as a stage in the X-axis direction, the TDI sensor images the pattern of the photomask 101 serving as a sample. These light source 103, magnifying optical system 104, photodiode array 105, and sensor circuit 106 constitute a high-magnification inspection optical system.

  The XYθ table 102 is driven by the table control circuit 114 under the control of the control computer 110. It can be moved by a drive system such as a three-axis (XY-θ) motor that drives in the X, Y, and θ directions.

  The measurement data (optical image) output from the sensor circuit 106 is sent to the alignment circuit 140 together with data indicating the position of the photomask 101 on the XYθ table 102 output from the position circuit 107. The measurement pattern data is, for example, 8-bit unsigned data, and represents the brightness gradation of each pixel.

  Then, as a reference data creation step (S104), a reference image creation unit configured by the development circuit 111, the reference circuit 112, and the like is used for comparison with measurement data based on design data of the photomask 101 to be inspected. Create reference data (reference image). The design data is stored in the magnetic disk device 109 or the like.

Next, as an alignment process, alignment is performed in order to compare the measurement data with the reference data.
In the storage step (S202), the reference data is read out and stored in the reference data memory 302 using the control computer 110. Similarly, the measurement data is read out and stored in the measurement data memory 304.

  As the first SSD alignment step (S210), the SSD alignment calculation unit 310, which is an example of the residual sum of squares calculation unit, receives the measurement data (a1) from the measurement data memory 304, and the reference data ( Enter b1). Then, a shift is made in units of pixels, the amount of positional deviation to the position where the residual sum of squares of the pixel value of the measurement data and the pixel value of the reference data is minimized, and the reference is shifted by the amount of positional deviation in units of pixels. Create an image. Then, by shifting in units of sub-pixels, the amount of positional deviation to the position where the residual sum of squares of the pixel value of the measurement data and the pixel value of the reference data is minimized is calculated, and only the positional deviation amount in units of sub-pixels Create a shifted reference image.

FIG. 5 is a diagram for explaining the SSD calculation method according to the first embodiment.
First, as the pixel unit SSD calculation step (S214), the pixel unit SSD calculation unit 312a uses the position information from the position circuit 107 as a base, and obtains an image of a predetermined size as a unit of comparison processing from the reference data memory 302. The reference data of the area (frame) is read out. At this time, the pixel unit SSD calculation unit 312a creates images (shifted images) obtained by shifting the reference data of the frames in parallel on a pixel basis. In FIG. 5, data 1, data 2,..., Data n are shown. The measurement data in such a frame is compared with the reference data. For example, an area of 512 × 512 pixels is preferable as one frame. Then, a residual sum of squares is calculated between the reference data of the plurality of reference data shifted in pixel units and the measurement data in the same size area read from the measurement data memory 304. The residual sum of squares can be obtained by squaring the residual between each pixel value of the reference data and each pixel value of the measurement data, and taking the sum of the entire read area. Then, a residual sum of squares is calculated for each of the plurality of reference data, and a minimum value of the residual sum of squares is calculated. Then, the amount of positional deviation to the position where the minimum value is obtained is calculated.

  In the pixel unit alignment step (S214), the pixel unit alignment unit 314a aligns the measurement data and the reference data at a position where the minimum value is obtained. Specifically, reference data shifted to a position that takes such a minimum value is created. In this way, it is possible to align at the most suitable position when the pixel unit is shifted in parallel in the x and y directions.

  Next, as the sub-pixel unit SSD calculation step (S216), the sub-pixel unit SSD calculation unit 316a shifts the pixel value of the measurement data from the position where the measurement data and the reference data are combined in the pixel unit. And a positional deviation amount to a position where the residual sum of squares between the pixel values of the reference data is minimized. The sub-pixel unit SSD calculation method is the same as that described in FIG. An image (shifted image) shifted in parallel in sub-pixel units is created for the reference data of the size of the region to be compared as described above based on the position combined in pixel units. In FIG. 5, data 1, data 2,..., Data n are shown. For example, 1/8, 1/16, 1/32 or the like of one pixel is used as a sub pixel. For example, when 1/8 of one pixel is used as a sub pixel, ± 1/8 pixel, ± 2/8 pixel, ± 3/8 pixel, ± 4/8 pixel in the x direction and y direction, Reference data of an area having a predetermined size shifted by ± 5/8 pixel, ± 6/8 pixel, and ± 7/8 pixel and reference data having a shift amount of 0 are created. In other words, 256 types of reference data are created that have 16 combinations in the x direction and 16 combinations in the y direction. Then, a residual sum of squares is calculated between each reference data and measurement data. The residual sum of squares can be obtained by squaring the residual between each pixel value of the reference data and each pixel value of the measurement data, and taking the sum of the entire read area. Then, a residual sum of squares is calculated for each of the plurality of reference data, and a minimum value of the residual sum of squares is calculated. In this way, it is possible to obtain the amount of displacement from the position where the minimum value is obtained.

  Then, as the sub-pixel unit alignment step (S218), the sub-pixel unit alignment unit 318a aligns the measurement data and the reference data at a position where the minimum value is obtained. Specifically, reference data shifted to a position that takes such a minimum value is created. In this way, it is possible to align the position most suitable when the parallel shift is performed in the x and y directions in sub-pixel units from the position aligned in pixel units. As described above, the corrected reference data (c2) aligned by the SSD calculation in units of sub-pixels is output to the next least square method alignment calculation unit 320.

  Next, as the first least square method alignment step (S220), the least square method alignment calculation unit 320, which is an example of the least square method calculation unit, receives the measurement data (a2) from the measurement data memory 304 and the reference data memory. The reference data (b2) is input from 302, and the reference data (c2) corrected by the first SSD alignment step (S210) is further input from the SSD alignment calculation unit 310. Then, a positional deviation amount based on the least square method from the above-described alignment position of the measurement data (a2) and the reference data (c2) is calculated. Then, the corrected reference data (c2m-1) aligned is created by applying the obtained parameter for correcting the positional deviation amount to the reference data (b2) that has not been corrected yet. Here, the amount of displacement required for alignment is calculated using a least square method, which is a statistical method.

FIG. 6 is a diagram illustrating a model formula for calculating a displacement amount based on the least square method according to the first embodiment. First, as the least square method displacement calculation step (S222), the least square method displacement calculation unit 322 calculates a displacement (x ₀ , y ₀ ) based on the least square method. As shown in FIG. 6, with respect to reference data U (x, y) serving as a reference image, a positional deviation amount (x in the x and y directions) of measurement data S (x, y) serving as an optical image (actual image). ₀ , y ₀ ) and the image intensity variation rate ε, S (x, y) = (1−ε) U (x−x ₀ , y−y ₀ ), as shown in equation (1). be able to. By linearizing on the assumption that the fluctuation amount is sufficiently small, ε · U + x ₀ · (dU / dx) + y ₀ · (dU / dy) = US is obtained as shown in the equation (2). dU / dx is a partial differential of U by x, and dU / dy is a partial differential of U by y. Then, for each pixel of the two-dimensional image, the value (US) obtained by subtracting the measurement data (a2) that is a real image from the reference data (c2) and the reference data (c2) are spatially differentiated in the x direction. A correlation matrix shown below can be obtained by obtaining a value (dU / dx) and a value (dU / dy) obtained by spatially differentiating the reference data (c2) in the y direction.

FIG. 7 is a diagram showing a correlation determinant based on the least square method in the first embodiment.
The positional deviation (x ₀ , y ₀ ) and the image intensity fluctuation rate ε can be estimated by the least square method by solving the correlation determinant shown in FIG. As a result, it is possible to obtain the amount of displacement (x ₀ , y ₀ ) for aligning with the most suitable position based on the least square method. In this way, parallel shift, expansion / contraction, rotation, and image intensity variation are modeled as unknown parameters, and the unknown parameters are calculated by the least square method.

In the least square method alignment step (S244), the least square method alignment unit 324a performs reference data (b2) on which parameters such as the obtained positional deviation (x ₀ , y ₀ ) and the image intensity variation rate ε have not yet been corrected. ) To create the corrected reference data (c2n-1) aligned. Specifically, the corrected reference is obtained by applying the measurement data S (x, y), the obtained positional deviation (x ₀ , y ₀ ), and the image intensity variation rate ε to the equation (1) in FIG. Data U (x, y) may be obtained. The corrected reference data (c2n-1) is output to the SSD alignment calculation unit 310 in order to perform the next SSD alignment process.

  FIG. 8 is a diagram illustrating an example of an alignment situation when the SSD alignment and the least square method alignment are performed once. In FIG. 8A, when the alignment of the measurement data A and the reference data B is performed, the peak positions of the respective pixel values (the position where the pixel value is large or the position where the image intensity is maximum) are sequentially 1, 2, 3 And numbering. That is, the measurement data A is arranged such that the peak positions are arranged in order of A1, A2, A3,. In the reference data B, the peak positions are arranged in order of B1, B2,. Here, a line and space pattern is shown as an example. For example, FIG. 8A shows a case where the peak position B1 of the reference data B that should be compared with the peak position A1 is located near the middle of the peak positions A1 and A2 of the measurement data A. In this state, when the pixel alignment is performed by the SSD alignment method, the peak position B1 of the reference data B should be calculated so as to be matched with the peak position A1 of the measurement data A, but is matched with the adjacent peak position A2. May occur. Such an error in the pixel unit SSD calculation is likely to occur because the image intensity, image distortion, and the like of the reference data B do not match the measurement data A. As shown in FIG. 8 (a), after positioning to a different figure, the peak position B1 is set to B'1 and the peak position B2 is set to B by performing sub-pixel unit SSD calculation or least square method calculation. '2 corrects the displacement. However, such alignment correction results in alignment at the sub-pixel level at the wrong position, as shown in FIG. 8B. As a result, it is difficult to match the figure that should originally be matched. As described above, an error in the pixel unit SSD calculation is likely to occur because the image intensity, image distortion, and the like of the reference data B do not match the measurement data A.

  Therefore, in the first embodiment, even when the wrong position is set in units of pixels, the amount of positional deviation that matches the image intensity, image distortion, etc. of the reference data B with the measurement data A at that position is obtained by the least square method calculation. . Then, using such a result, alignment is performed on a pixel basis by the SSD alignment method again. Thereby, in the pixel unit SSD calculation of the 2nd time, it can match with the figure which should be matched. Then, after the alignment, the sub-pixel unit SSD calculation or the least square method calculation is performed, thereby enabling alignment with higher accuracy. Hereinafter, in order to facilitate understanding of the description, a case where n = m = 2 will be described.

  As the second (m-th) SSD alignment step (S230), the SSD alignment calculation unit 310 obtains measurement data (a3: indicated by a2m-1 in FIG. 3) from the measurement data memory 304 from the reference data memory 302. Reference data (b3: indicated by b2m-1 in FIG. 3) is input as corrected reference data (c3: indicated by c2m-1 in FIG. 3) from the least squares alignment calculation unit 320. Then, the measurement data (a3: indicated by a2m-1 in FIG. 3) and the corrected reference data (c3: indicated by c2m-1 in FIG. 3) are shifted in units of pixels, A positional shift amount to a position where the residual sum of squares with the pixel value of the reference data is minimized is calculated, and a reference image shifted by a positional shift amount in pixel units is created. Then, by shifting in units of sub-pixels, the amount of positional deviation to the position where the residual sum of squares of the pixel value of the measurement data and the pixel value of the reference data is minimized is calculated, and only the positional deviation amount in units of sub-pixels is calculated. The shifted reference data (c4: indicated by c2m in FIG. 3) is created.

  Specifically, first, as the pixel unit SSD calculation step (S234), the pixel unit SSD calculation unit 312m converts the measurement data (a3: indicated by a2m-1 in FIG. 3) from the measurement data memory 304 to the least square alignment. Reference data after correction (c3: indicated by c2m-1 in FIG. 3) is input from the calculation unit 320. Then, the residual between each reference data of a plurality of reference data shifted in pixel units from the measurement data (a3: indicated by a2m-1 in FIG. 3) of the same size area read from the measurement data memory 304 Calculate the sum of squares. Then, the minimum value of the residual square sum is calculated. Then, the amount of positional deviation to the position where the minimum value is obtained is calculated. The calculation method is the same as in the pixel unit SSD calculation step (S214).

  Then, as the pixel unit alignment step (S234), the pixel unit alignment unit 314m uses the position information from the position circuit 107 as a base, and from the reference data memory 302, an image area of a predetermined size serving as a unit for comparison processing. Reference data (b3: indicated by b2m-1 in FIG. 3) is read out. Then, the pixel unit alignment unit 314m has measurement data (a3: indicated by a2m-1 in FIG. 3) and uncorrected reference data (b3: indicated by b2m-1 in FIG. 3) at the position where the obtained minimum value is obtained. And align. Specifically, reference data shifted to a position that takes such a minimum value is created.

  Next, as a sub-pixel unit SSD calculation step (S236), the sub-pixel unit SSD calculation unit 316m performs measurement data (a3: indicated by a2m-1 in FIG. 3) and reference data (b3: b2m-1 in FIG. 3). The pixel value of the measurement data (a3: indicated by a2m-1 in FIG. 3) and the reference data (b3: indicated by b2m-1 in FIG. 3) are shifted from the position aligned in pixel units to the subpixel unit. ) To calculate a position shift amount to a position where the residual sum of squares with the pixel value is minimum.

  Then, in the sub-pixel unit alignment step (S238), the sub-pixel unit alignment unit 318m aligns the measurement data and the reference data at the position where the minimum value is obtained. Specifically, reference data shifted to a position that takes such a minimum value is created. As described above, the corrected reference data (c4: indicated by c2m in FIG. 3) aligned by the SSD calculation in units of sub-pixels is output to the next least square method alignment calculation unit 320.

  Next, as the second (m-th) least squares alignment step (S240), the least squares alignment calculation unit 320, which is an example of the least squares method calculation unit, receives measurement data (a4: Reference data (c4: indicated by c2m in FIG. 3) corrected by the second (m-th) SSD alignment step (S230) is input from the SSD alignment calculation unit 310. Then, a positional deviation amount based on the least square method from the above-described alignment position of the measurement data (a4: indicated by a2m in FIG. 3) and the reference data (c4: indicated by c2m in FIG. 3) is calculated. Then, the corrected reference data (c2m + 1) aligned is created by applying the obtained parameter for correcting the positional deviation amount to the reference data (c4: indicated by c2m in FIG. 3).

Specifically, as the least square method displacement calculation step (S242), the least square method displacement calculation unit 322m includes parameters such as a displacement (x ₀ , y ₀ ) and an image intensity variation rate ε based on the least square method. Is calculated. The least square method displacement calculation method is the same as the first least square method alignment step (S220).

Then, as the least square method alignment step (S244), the least square method alignment unit 324m uses parameters such as the obtained positional deviation (x ₀ , y ₀ ) and the image intensity variation rate ε as reference data (c4: c2m in FIG. 3). The corrected reference data (c2m + 1) that has been aligned is created. Specifically, the corrected reference is obtained by applying the measurement data S (x, y), the obtained positional deviation (x ₀ , y ₀ ), and the image intensity variation rate ε to the equation (1) in FIG. Data U (x, y) may be obtained. The corrected reference data (c2m + 1) is output to the comparison circuit 108.

  FIG. 9 is a diagram illustrating an example of an alignment state when the SSD alignment and the least square method alignment are performed a plurality of times in the first embodiment. FIG. 9A shows an example of a situation in which the parameters obtained by the first least-squares alignment are applied to the reference data B that has not been corrected yet. In FIG. 9A, when the alignment of the measurement data A and the reference data B is performed, the peak position of each pixel value (the position where the pixel value is large or the position where the image intensity is maximum) is sequentially 1, 2, 3 And numbering. That is, the measurement data A is arranged such that the peak positions are arranged in order of A1, A2, A3,. In the reference data B, the peak positions are arranged in order of B1, B2,. Here, a line and space pattern is shown as an example. As in FIG. 8 described above, for example, in FIG. 9A, the peak position B1 of the reference data B that should be compared with the peak position A1 is located near the middle of the peak positions A1 and A2 of the measurement data A. Shows the case. By correcting the reference data B by applying the parameter obtained by the first least square method alignment, the reference data B ′ in which the image intensity or the image distortion approaches the measurement data A can be created. For example, in a repetitive pattern such as a line-and-space pattern, the shapes of adjacent graphics are the same or approximate, so an image can be obtained from the parameters obtained by aligning the adjacent graphics with the first least squares alignment. The intensity and image distortion can be brought close to the measurement data A.

  Then, the measurement data A and the reference data B ′ are aligned by the second (m-th) SSD alignment from the state shown in FIG. 9A, so that they should be aligned as shown in FIG. 9B. A displacement amount for aligning the peak positions (A1 and B′1) can be obtained. Then, by applying the positional deviation amount obtained by the second (m-th) SSD alignment to the reference data B that has not yet been corrected, the peak positions (A1 and B1) that should originally be matched are determined in pixel units. Can be matched.

  By performing the sub-pixel unit alignment and the least square method alignment on the measurement data A and the reference data B from the state of FIG. 9B in which the peak positions (A1 and B1) that should be originally aligned are aligned in pixel units, Furthermore, reference data whose position matches the measurement data A can be created.

  As described above, the SSD alignment calculation unit 310 (first calculation unit: in particular, the pixel unit SSD calculation unit 312a and the sub-pixel unit SSD calculation unit 316a) has the measurement data (a1) (optical image) and the reference circuit 112. Using the first different reference images shifted in parallel based on the first alignment position (first alignment position) with the reference data (b1) (first reference image) created by The residual square between the pixel value of the optical image and the pixel value of each reference image of the first plurality of reference images between the optical image and each reference image of the first other plurality of reference images. Each sum is calculated. Then, the positional deviation amount (first positional deviation amount) in the combination in which the smallest residual square sum is calculated among the calculated residual square sums is acquired.

  Next, the least square method alignment calculation unit 320 (second calculation unit: in particular, the least square method positional deviation calculation unit 322a) performs reference data (c2) (second) with the first positional deviation amount corrected. The reference position of the measurement data (a2) and the reference data (c2) is calculated by a least square method using a value obtained by spatially differentiating the reference image in the x direction and a value obtained by spatially differentiating in the y direction. Position misalignment (second misalignment amount) from the (alignment position). Then, the obtained parameters are applied to the reference data (b2) to generate reference data (c3: c2n-1).

  Then, the SSD alignment calculation unit 310 ((2n-1) th (odd number) (where 2 ≦ n ≦ m, where n and m are integers equal to or greater than 2): in particular, a pixel unit SSD calculation unit 312n and the sub-pixel unit SSD calculation unit 316n) match the position of the measurement data (a2n-1) and the reference data (c3: c2n-1) ((2n-1) reference image) (second (2n-1)). The reference data of the measurement data (a2n-1) and the other reference images of the nth are used by using the reference images of the nth different reference images shifted in parallel on the basis of Between the pixel values of the measurement data (a2n-1) and the pixel values of the respective reference images of the nth different reference images. Then, a positional deviation amount ((2n-1) th positional deviation amount) in the combination in which the smallest residual square sum is calculated from among the calculated residual square sums is acquired.

  Then, the least square method alignment calculation unit 320 (the (2n) th calculation unit which is an even number: in particular, the least square method positional deviation calculation unit 322n) has the (2n-1) th reference data (b1). By the least square method of calculating using the value obtained by spatially differentiating the reference data (c4: c2n) (the (2n) reference image) in which the displacement amount is corrected in the x direction and the value obtained by spatially differentiating in the y direction, A displacement amount (a (2n) th displacement amount) from the alignment position of the measurement data (a2n) and the reference data (c4: c2n) is calculated.

  As described above, with the calculation by the SSD alignment calculation unit 310 as the first, the SSD alignment calculation unit 310 and the least square method alignment calculation unit 320 alternately repeat the calculation a plurality of times the same number of times. At that time, the SSD alignment calculation unit 310 performs calculation based on the alignment position of the measurement data (a1) and the reference data (b1) that has not been position-corrected in the first calculation (first calculation), For the second and subsequent times, the alignment position of the measurement data (a2n-1) and the reference data (c2n-1) in which the positional deviation amount obtained by using the least square method is corrected for the reference data (b1) whose position is not corrected. Calculate based on.

  In addition, the least square method alignment calculation unit 320 calculates the difference between the measurement data (a2n) and the reference data (c2n) in which the positional deviation amount obtained by using the SSD calculation is corrected to the reference data (b1) that is not position-corrected. The amount of displacement from the alignment position is calculated. However, in the calculation prior to the last calculation (the m-th calculation), the obtained positional deviation amount is corrected to reference data (b1) whose position is not corrected and output.

  Then, the last least-squares method alignment calculation unit 320 (the even-numbered (2m) -th calculation unit: in particular, the least-squares method displacement calculation unit 322m) uses the least-squares method to calculate the measurement data (a2m) and 1 The amount of displacement (second (2m) from the alignment position with the reference data (c2m) (second (2m) reference image) generated by the SSD alignment computing unit 310 of the last (2m-1) th) ) The (first positional deviation amount) is calculated. Then, the least square method alignment unit 324m corrects the reference data (c2m) by applying a parameter that is the (2m) th positional deviation amount. Thus, corrected reference data (c2m + 1) ((2m + 1) th reference image) is generated. As described above, the least square method alignment unit 324m corrects the obtained positional deviation amount in the last calculation (m-th computation) with the final positional deviation amount obtained by using the SSD calculation. The reference data (c2m) is corrected and output.

  The maximum value m of the number of repetitions n may be 2 or more. The larger the position, the higher the accuracy of alignment.

  As described above, in the image alignment method according to the first embodiment, first, the reference image (1) created for comparison with the optical image of the sample to be inspected on which the pattern is formed is substituted into the reference image (2). The first step is performed.

  Then, as a second step, a position correction amount by parallel shift that minimizes the residual sum of squares (SSD) of the reference image (2) and the optical image is calculated by searching. The position correction amounts here are parallel shifts δx and δy. In general, a range of δx = i + j / N (i, j: integer) is possible.

  Subsequently, as a third step, the position of the reference image (1) is corrected by the first parameter indicating the position correction amount by the parallel shift calculated in the second step to obtain a new reference image (2).

  Subsequently, as a fourth step, the parallel shift / expansion / rotation / rotation / image intensity fluctuation of the optical image with respect to the new reference image (2) is modeled as an unknown second parameter, and the second parameter is calculated by the least square method. Is calculated. However, the parallel shift amount is limited to less than one pixel.

  Subsequently, as a fifth step, the reference image (1) is image-corrected by the second parameter calculated in the fourth step to obtain a new second reference image.

  Then, the second to fourth steps are repeated N times, the fifth step is repeated N-1 times, and the second to fifth steps are sequentially repeated. Then, as the sixth process which is the final time, after the Nth fourth process, only the second parameter calculated in the Nth fourth process is used in the Nth fourth process. The reference image (2) is corrected to obtain a reference image (3) and output. This reference image (3) becomes the reference data (c2m + 1) described above.

  As the comparison step (S250), the comparison circuit 108 (comparison unit) compares the measurement data with the reference data corrected using the SSD calculation and the least square method calculation. In the example of FIG. 3, reference data (c2m + 1) and measurement data (a2m + 1) are taken and compared according to a predetermined algorithm, and the presence or absence of a defect is determined for each pixel. By performing data comparison in which alignment is performed with high accuracy as described above, it is possible to suppress false detection of defects and reduce pseudo defects, and to perform high-precision inspection.

FIG. 10 is a diagram for explaining another optical image acquisition method.
In the configuration of FIG. 1, the photodiode array 105 that simultaneously enters the number of pixels of the scan width W (for example, 2048 pixels) is used. However, the present invention is not limited to this, and as shown in FIG. Each time a constant pitch movement is detected by a laser interferometer while scanning at a constant speed in the direction, a laser scanning optical device (not shown) scans the laser beam in the Y direction, detects transmitted light, A technique of acquiring a two-dimensional image for each size area may be used.

  In the above description, what is described as “˜circuit”, “˜unit”, or “˜process” can be configured by a program operable by a computer. Or you may make it implement by not only the program used as software but the combination of hardware and software. Alternatively, a combination with firmware may be used. When configured by a program, the program is recorded on a recording medium such as a magnetic disk device, a magnetic tape device, an FD, or a ROM (Read Only Memory). For example, each circuit in the table control circuit 114, the expansion circuit 111, the reference circuit 112, the comparison circuit 108, the alignment circuit 140, and the like that constitute the arithmetic control unit may be configured by an electrical circuit, You may implement | achieve as software which can be processed by the computer 110 or the computer etc. which are arrange | positioned in each circuit. Moreover, you may implement | achieve with the combination of an electrical circuit and software.

The embodiments have been described above with reference to specific examples. However, the present invention is not limited to these specific examples. For example, in each embodiment, transmitted light is used, but reflected light or transmitted light and reflected light may be used simultaneously. Although the reference image is generated from the design data, data of the same pattern captured by a sensor such as a photodiode array may be used. In other words, a die to die inspection or a die to database inspection may be used. Further, in the above-described example, the parameter obtained by the least square method is the scale of the measurement data that becomes the optical image (actual image) with respect to the gradation value (pixel value) U (x, y) of the reference data that becomes the reference image. Although the positional deviation amount (x ₀ , y ₀ ) and the image intensity fluctuation rate ε in the X and Y directions of the tone value S (x, y) are used, the present invention is not limited to this. Further, a gray level offset δ may be added.

  In addition, although descriptions are omitted for parts and the like that are not directly required for the description of the present invention, such as a device configuration and a control method, a required device configuration and a control method can be appropriately selected and used.

  In addition, all pattern inspection apparatuses, pattern inspection methods, and image alignment methods that include elements of the present invention and that can be appropriately modified by those skilled in the art are included in the scope of the present invention.

DESCRIPTION OF SYMBOLS 100 Pattern inspection apparatus 101 Photomask 102 XY (theta) table 103 Light source 104 Magnification optical system 105 Photodiode array 106 Sensor circuit 107 Position circuit 108 Comparison circuit 109 Magnetic disk apparatus 110 Control computer 111 Expansion circuit 112 Reference circuit 115 Magnetic tape apparatus 140 Position alignment circuit 150 Optical image acquisition unit 302 Reference data memory 304 Measurement data memory 310 SSD alignment calculation unit 311, 321 Memory 312 Pixel unit SSD calculation unit 314 Pixel unit alignment unit 316 Sub pixel unit SSD calculation unit 318 Sub pixel unit alignment unit 320 Least square method Alignment calculation unit 322 Least square method position shift calculation unit 324 Least square method alignment unit

Claims (7)

An optical image acquisition unit for acquiring an optical image of a test sample on which a pattern is formed;
Based on the design data of the sample to be inspected, a reference image creating unit that creates a reference image to be compared with the optical image;
Using another plurality of reference images shifted in parallel based on the alignment position of the optical image and the reference image, the optical image between the optical image and each reference image of the plurality of other reference images And the residual sum of squares that is the smallest of the calculated residual sums of squares. A residual sum-of-squares calculation unit that acquires the amount of misalignment in the combination for which
By the least-squares method of calculating using a value obtained by spatially differentiating in the x direction and a value obtained by spatially differentiating the reference image acquired by the residual sum of squares computing unit in the x direction A least-squares method computing unit that computes the amount of displacement from the alignment position of the image and the corrected reference image;
A comparison unit that compares the optical image with the reference image corrected using the residual sum of squares and the least squares method;
With
Starting with the calculation by the residual sum of squares calculation unit, the residual sum of squares calculation unit and the least squares method calculation unit alternately repeat the same number of times multiple times,
The said comparison part compares the reference image and the said optical image which were correct | amended to the position obtained as a result of repeating calculation several times, The pattern inspection apparatus characterized by the above-mentioned. The residual sum-of-squares calculation unit performs a calculation based on a matching position between the optical image and the reference image that has not been position-corrected at the time of the first calculation, and the second and subsequent times are position-corrected with the optical image. Based on the alignment position with the reference image corrected for the positional deviation amount obtained by using the least square method to the reference image that is not,
The least square method computing unit is configured to detect a positional deviation from a position where the optical image and the reference image not corrected for position are aligned with a reference image obtained by correcting a positional deviation amount obtained by using the residual sum of squares. The amount of misregistration is calculated, and the obtained misregistration amount is corrected to the reference image that has not been corrected in the calculation before the last calculation, and the obtained misregistration amount is obtained in the final calculation. The pattern inspection apparatus according to claim 1, wherein the pattern inspection apparatus corrects and outputs a reference image in which a final positional shift amount obtained using the residual sum of squares is corrected.   The residual sum of squares calculation unit calculates the amount of positional shift to a position where the residual sum of squares is minimized while shifting in units of pixels, and further shifts in units of sub-pixels from the position where the residual sum of squares is minimum. 3. The pattern inspection apparatus according to claim 1, wherein a displacement amount to a position where the difference sum of squares is minimized is calculated. A first step of substituting a first reference image created for comparison with an optical image of a sample to be inspected with a pattern into a second reference image;
A second step of calculating a position correction amount by a parallel shift that minimizes a residual sum of squares of the second reference image and the optical image;
A third step of obtaining a new second reference image by correcting the position of the first reference image by the first parameter indicating the position correction amount by the parallel shift calculated in the second step;
A fourth step of modeling the parallel shift, expansion / contraction, rotation, and image intensity variation of the optical image with respect to the new second reference image as an unknown second parameter, and calculating the second parameter by a least square method;
A fifth step of image-correcting the first reference image by the second parameter calculated in the fourth step to obtain a new second reference image;
With
Repeat the second to fourth steps N times, the fifth step N-1 times, the second to fifth steps in order,
After the Nth fourth step, the second second reference image used in the Nth fourth step is image-corrected by the second parameter calculated in the Nth fourth step and the third parameter is corrected. An image alignment method, further comprising a sixth step of obtaining and outputting the reference image. An optical image acquisition unit for acquiring an optical image of a test sample on which a pattern is formed;
A reference image creating unit that creates a first reference image to be compared with the optical image based on the design data of the sample to be inspected;
Using the first different reference images shifted in parallel based on the first alignment position of the optical image and the first reference image, the optical image and the first different plurality of reference images are used. And calculating a residual sum of squares of the pixel value of the optical image and the pixel value of each reference image of the first plurality of other reference images with each reference image of the reference image. A first calculation unit that obtains a first positional shift amount in a combination in which a minimum residual square sum is calculated from each residual square sum;
The optical image and the first reference image are calculated by a least square method that uses a value obtained by spatial differentiation in the x direction and a value obtained by spatial differentiation in the y direction of the second reference image in which the first positional deviation amount is corrected. A second calculation unit that calculates a second positional deviation amount from the second alignment position with the second reference image;
The (2n−1) th (2n−2) th (where 2 ≦ n ≦ m, where n and m are integers) corrected positional shift amounts in the optical image and the first reference image. Using the plurality of nth different reference images shifted in parallel based on the (2n-1) th matching position with the reference image of the reference image, the optical image and the other plurality of the nth reference images Calculating a residual sum of squares between the pixel value of the optical image and the pixel value of each reference image of the nth other reference images between each reference image of the image, A (2n-1) th arithmetic unit that obtains the (2n-1) th positional deviation amount in the combination in which the minimum residual square sum is calculated from the residual square sums;
Using a value obtained by spatially differentiating in the x direction and a value obtained by spatially differentiating in the y direction, the first reference image obtained by correcting the (2n-1) th positional deviation amount in the first reference image. A (2n) -th computing unit that computes a (2n) -th positional shift amount from the alignment position of the optical image and the (2n) reference image by a least-squares method to be computed;
The (2m + 1) th positional deviation amount corrected with respect to the (2m) th reference image with the (2m-1) th positional deviation amount corrected with respect to the first reference image. A comparison unit for comparing a reference image and the optical image;
A pattern inspection apparatus comprising: An image alignment method for aligning an optical image and a reference image used for comparative inspection of a test sample on which a pattern is formed,
The first different reference images shifted in parallel based on the first alignment position of the optical image and the first reference image are used, and the optical image and the first other plurality of reference images are used. A residual sum of squares of the pixel value of the optical image and the pixel value of each reference image of the first plurality of reference images is calculated between each reference image of the reference image, and the calculation is performed. A first calculation step of obtaining a first positional deviation amount in a combination in which a minimum residual square sum is calculated from each residual square sum;
The optical image and the first reference image are calculated by a least square method that uses a value obtained by spatial differentiation in the x direction and a value obtained by spatial differentiation in the y direction of the second reference image in which the first positional deviation amount is corrected. A second calculation step of calculating a second positional deviation amount from the second alignment position with the second reference image;
The (2n−1) th (2n−2) th (where 2 ≦ n ≦ m, where n and m are integers) corrected positional shift amounts in the optical image and the first reference image. Using the plurality of nth different reference images shifted in parallel based on the (2n-1) th matching position with the reference image of the reference image, the optical image and the other plurality of the nth reference images Calculating a residual sum of squares between the pixel value of the optical image and the pixel value of each reference image of the nth other reference images between each reference image of the image, A (2n-1) th calculation step of obtaining a (2n-1) th positional shift amount in a combination in which the minimum residual square sum is calculated from the residual square sum;
Using a value obtained by spatially differentiating in the x direction and a value obtained by spatially differentiating in the y direction, the first reference image obtained by correcting the (2n-1) th positional deviation amount in the first reference image. A (2n) th calculation step of calculating a (2n) th positional shift amount from the alignment position of the optical image and the (2n) reference image by a least square method to be calculated;
The (2m + 1) th (2m + 1) th position shift amount is generated by correcting the (2m) th position shift amount with respect to the (2m) th reference image in which the (2m-1) th position shift amount is corrected in the first reference image. ) A reference image and the optical image are aligned and output,
An image alignment method characterized by comprising: A storage process for storing in the storage device the optical image and the first reference image used for the comparative inspection of the test sample on which the pattern is formed;
The optical image and the first reference image are read from the storage device, and a plurality of first different plurality of signals shifted in parallel based on a first alignment position of the optical image and the first reference image. Using a reference image, each reference of the pixel value of the optical image and the first plurality of reference images between the optical image and each reference image of the first other plurality of reference images The first sum of the squares of the residual values with the pixel values of the image is calculated, and the first misregistration amount in the combination in which the minimum residual sum of squares is calculated from among the calculated residual sums of squares. A first calculation process to be acquired;
The optical image and the first reference image are calculated by a least square method that uses a value obtained by spatial differentiation in the x direction and a value obtained by spatial differentiation in the y direction of the second reference image in which the first positional deviation amount is corrected. A second calculation process for calculating a second positional shift amount from the second alignment position with the second reference image;
The (2n−1) th (2n−2) th (where 2 ≦ n ≦ m, where n and m are integers) corrected positional shift amounts in the optical image and the first reference image. Using the plurality of nth different reference images shifted in parallel based on the (2n-1) th matching position with the reference image of the reference image, the optical image and the other plurality of the nth reference images Calculating a residual sum of squares between the pixel value of the optical image and the pixel value of each reference image of the nth other reference images between each reference image of the image, A (2n-1) th calculation process for obtaining a (2n-1) th positional deviation amount in a combination in which the minimum residual square sum is calculated from the residual square sums;
Using a value obtained by spatially differentiating in the x direction and a value obtained by spatially differentiating in the y direction, the first reference image obtained by correcting the (2n-1) th positional deviation amount in the first reference image. A (2n) th calculation process for calculating a (2n) th positional shift amount from the alignment position of the optical image and the (2n) reference image by a least square method to be calculated;
The (2m + 1) th (2m + 1) th position shift amount is generated by correcting the (2m) th position shift amount with respect to the (2m) th reference image in which the (2m-1) th position shift amount is corrected in the first reference image. ) The reference image and the optical image are aligned and output,
A program that causes a computer to execute.

Images