JP2020052025A - Pattern inspection device and reference image creation method - Google Patents

Abstract

PURPOSE: To provide an inspection device capable of improving accuracy of filter processing.CONSTITUTION: A pattern inspection device includes: an optical image acquisition section for acquiring an optical image of a patterned sample to be inspected; a design image creation section for creating a design image by performing image development of design pattern data to be a base of pattern formation of the sample to be inspected; a conversion section for converting the design image into frequency data; an extraction section for extracting data of a part of a frequency band out of the frequency data; an inverse conversion section for inversely converting the extracted data of the part of frequency band and generating a space area image; a coefficient calculation section for calculating a coefficient of a filter function for filter processing of the design image using at least a part of space area image, a corresponding part of the optical image, and a corresponding part of the design image; a reference image creation section for creating a reference image by filter processing of the design image using the coefficient; and a comparison section for comparing the optical image with the reference image for each pixel.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a pattern inspection apparatus and a method for creating a reference image. For example, the present invention relates to a pattern inspection technology for inspecting a pattern defect of an object serving as a sample used in semiconductor manufacturing, and relates to a method for inspecting an extremely small pattern defect of a photomask used when manufacturing a semiconductor element or a liquid crystal display (LCD). .

  2. Description of the Related Art In recent years, circuit line widths required for semiconductor devices have become increasingly smaller with the increase in integration and capacity of large-scale integrated circuits (LSIs). These semiconductor elements are formed by exposing and transferring a pattern onto a wafer using a reduction projection exposure apparatus called a stepper, using an original pattern on which a circuit pattern is formed (also referred to as a mask or a reticle; hereinafter, collectively referred to as a mask). It is manufactured by forming a circuit. Therefore, in manufacturing a mask for transferring such a fine circuit pattern onto a wafer, a pattern drawing apparatus using an electron beam capable of drawing a fine circuit pattern is used. A pattern circuit may be directly drawn on a wafer using such a pattern drawing apparatus. Alternatively, development of a laser beam drawing apparatus that draws using a laser beam other than the electron beam has been attempted.

  In addition, for the production of LSIs that require a large production cost, improvement of the yield is indispensable. However, as typified by a 1 gigabit DRAM (random access memory), the pattern forming the LSI is going to be on the order of submicron to nanometer. One of the major factors that reduce the yield is a pattern defect of a mask used when exposing and transferring an ultrafine pattern on a semiconductor wafer by photolithography. 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 for inspecting a defect of a transfer mask used in LSI manufacturing.

  As an inspection method, 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 is compared with design data or an optical image obtained by imaging the same pattern on the sample. A method of performing an inspection by performing the inspection is known. For example, as a pattern inspection method, drawing data (design pattern data) obtained by converting pattern-designed CAD data into a device input format for a drawing device to input is input to an inspection device, and a reference image is generated based on the data. Then, there is a “die-to-database (die-database) inspection” for comparing a reference image with an optical image serving as measurement data obtained by capturing a pattern. In the inspection method in such an inspection apparatus, a sample is placed on a stage, and a light beam scans over the sample as the stage moves, thereby performing an inspection. The sample is irradiated with a light beam by a light source and an illumination optical system. Light transmitted or reflected by the sample is imaged on a sensor via an optical system. The image captured by the sensor is sent to a comparison circuit as measurement data. After the alignment of the images, the comparison circuit compares the measured data with the reference data according to an appropriate algorithm, and if they do not match, determines that there is a pattern defect.

  The pixel data of the optical image picked up from the sample is in a state in which a filter is applied due to the resolution characteristics of the optical system used for the image pickup, in other words, in a continuously changing analog state. The value is different from the design image. Therefore, in the die-database inspection, a comparison process is performed after applying a filter process to a design image to create a reference image close to measurement data. While the improvement in inspection accuracy accompanying the recent miniaturization of patterns is required, it is effective to improve the accuracy of such filter processing in order to improve the inspection accuracy.

  Here, in order to improve the accuracy of the filtering process, it is ideal to cover all the shapes of the pattern formed on the mask substrate, but it is difficult to cover all the shapes from the viewpoint of processing time and the like. is there. Therefore, the coefficients of the filter function are calculated using some of the patterns. On the other hand, an auxiliary pattern (SRAF: Sub Resolution Assist Feature pattern and / or OPC) for improving a lithography margin and correcting a dimensional error due to a proximity effect with miniaturization of an LSI pattern dimension formed on a semiconductor wafer. : Optical proximity correction pattern). Therefore, various complicated-shaped patterns are mixed on the mask substrate. The auxiliary pattern portion in the created reference image tends to have a lower degree of coincidence with the measured image than the main pattern (actual circuit pattern) constituting the circuit. As a result, in the pattern defect inspection of the mask substrate, there is a problem that not only the poorly formed auxiliary patterns but also the well-formed auxiliary patterns generate pseudo defects that are determined to be defective. Therefore, it is required to create a reference image that is not determined to be a defect for a good auxiliary pattern. In order to increase the degree of coincidence between the measurement image and the reference image, it is required that the coefficient of the filter function be calculated using an area including a pattern having a low degree of coincidence that becomes a pseudo defect. However, there is a problem that it is difficult to selectively extract a region including such a pattern with a low degree of coincidence from the design data.

  Here, it has been studied to reduce the number of defects including not only poorly-formed auxiliary patterns but also poorly-formed auxiliary patterns by loosening the inspection sensitivity in the vicinity of the auxiliary pattern (for example, Patent Document 1). However, with such a technique, it is difficult to create a reference image in which a poorly formed auxiliary pattern is not determined to be a defect while excluding poorly formed auxiliary patterns.

JP 2007-071629 A

  Therefore, the present invention provides an inspection apparatus and a reference image creation method capable of improving the accuracy of filter processing.

A pattern inspection apparatus according to one embodiment of the present invention includes:
An optical image acquisition unit that acquires an optical image of a pattern-formed inspection sample,
A design image creating unit that creates a design image by developing design pattern data that is the basis of pattern formation of a sample to be inspected,
A conversion unit that converts a design image into frequency data,
An extracting unit that extracts data of some frequency bands from the frequency data;
An inverse transform unit that inversely transforms the extracted data of some of the frequency bands to generate a spatial domain image;
At least a part of the spatial domain image, a corresponding part of the optical image, and a corresponding part of the design image, using a coefficient calculation unit that calculates a filter function coefficient for filtering the design image,
A reference image creating unit that creates a reference image by filtering the design image using the coefficient,
A comparison unit that compares the optical image and the reference image for each pixel,
It is characterized by having.

  Further, it is preferable that the extraction unit extracts, as data of a part of the frequency band, data of a frequency band having a line width smaller than that of the actual circuit pattern and corresponding to an auxiliary pattern for assisting the formation of the actual circuit pattern.

  Preferably, the conversion unit converts the design image into frequency data by performing a Fourier transform process on the data of the design image.

  Further, it is preferable that, of the spatial region image, a small region including a pixel whose pixel value difference between the corresponding optical image and the design image exceeds a threshold value is used for the calculation of the coefficient.

A method for creating a reference image according to one embodiment of the present invention includes:
A step of obtaining an optical image of the pattern-formed inspection sample,
A step of developing a design image by developing the design pattern data that is the basis of the pattern formation of the sample to be inspected,
Converting the design image into frequency data;
Extracting frequency band data from the frequency data;
Inversely converting the extracted data of some of the frequency bands to generate a spatial domain image,
Using at least a part of the spatial domain image, a corresponding part of the optical image, and a corresponding part of the design image, calculating a coefficient of a filter function for filtering the design image,
Using the coefficients to filter the design image to create a reference image;
It is characterized by having.

  ADVANTAGE OF THE INVENTION According to this invention, the precision of a filter process can be improved. Therefore, a highly accurate reference image can be created. As a result, the inspection accuracy can be improved.

1 is a configuration diagram illustrating a configuration of a pattern inspection device according to a first embodiment. FIG. 3 is a block diagram showing an internal configuration of a learning point area data creation circuit according to the first embodiment. FIG. 3 is a conceptual diagram for describing an inspection area according to the first embodiment. FIG. 4 is a flowchart showing a part of a main step of the pattern inspection method in the first embodiment; 9 is a diagram illustrating an example of a learning point in Embodiment 1. [FIG. 9 is a diagram illustrating an example of a learning point region image according to Embodiment 1. FIG. 6 is a diagram for describing an example of a method of calculating coefficients of a filter function according to Embodiment 1. [FIG. FIG. 5 is a diagram for describing a filtering process according to the first embodiment. FIG. 3 is a diagram for describing a method of frequency data conversion according to the first embodiment. 9 is a diagram illustrating an example of a relationship between frequency and spectrum intensity according to Embodiment 1. [FIG. FIG. 3 is a diagram showing an example of a spatial domain image obtained from data of a part of the frequency bands in the first embodiment. FIG. 3 is a configuration diagram illustrating an example of an internal configuration of a comparison circuit according to the first embodiment; FIG. 5 is a flowchart showing the remaining part of the main steps of the pattern inspection method according to the first embodiment;

Embodiment 1 FIG.
FIG. 1 is a configuration diagram showing a configuration of the pattern inspection apparatus according to the first embodiment. In FIG. 1, an inspection apparatus 100 for inspecting a defect of a pattern formed on a sample, for example, a mask, includes an optical image acquisition mechanism 150 and a control system circuit 160 (control unit).

  The optical image acquisition mechanism 150 includes a light source 103, an illumination optical system 170, a movable XYθ table 102, an enlargement optical system 104, a photodiode array 105 (an example of a sensor), a sensor circuit 106, a stripe pattern memory 123, and It has a laser length measurement system 122. The sample 101 is arranged on the XYθ table 102. The sample 101 includes, for example, an exposure photomask for transferring a pattern to a wafer. Further, a pattern constituted by a plurality of graphic patterns to be inspected is formed on the photomask. The sample 101 is placed on the XYθ table 102 with the pattern forming surface facing downward, for example. In addition to the main pattern (actual circuit pattern) constituting the circuit, an auxiliary pattern (SRAF pattern and / or OPC pattern) having a smaller line width than the main pattern and assisting the formation of the main pattern is provided on the sample 101. Are formed.

  In the control system circuit 160, a control computer 110 that controls the entire inspection apparatus 100 includes a position circuit 107, a comparison circuit 108, a development circuit 111, a reference circuit 112, an autoloader control circuit 113, a table control circuit 114 via a bus 120. It is connected to a coefficient calculation circuit 140, a learning point area data creation circuit 142, 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. Further, the sensor circuit 106 is connected to the stripe pattern memory 123, and the stripe pattern memory 123 is connected to the comparison circuit 108. The XYθ table 102 is driven by an X-axis motor, a Y-axis motor, and a θ-axis motor. Table 102 is an example of a stage.

  In the inspection apparatus 100, a light source 103, an XYθ table 102, an illumination optical system 170, a magnifying optical system 104, a photodiode array 105, and a sensor circuit 106 constitute a high-magnification inspection optical system. Table 102 is driven by table control circuit 114 under the control of control computer 110. It can be moved by a driving system such as a three-axis (XY-θ) motor that drives in the X, Y, and θ directions. As the X motor, the Y motor, and the θ motor, for example, a step motor can be used. The XYθ table 102 can be moved in the horizontal and rotational directions by motors for each axis of XYθ. Then, the moving position of the sample 101 placed on the XYθ table 102 is measured by the laser length measuring system 122 and supplied to the position circuit 107.

  Design pattern data (drawing data) that is a basis for forming a pattern of the sample to be inspected 101 is input from outside the inspection apparatus 100 and stored in the magnetic disk device 109.

  Here, FIG. 1 illustrates components necessary for describing the first embodiment. Needless to say, the inspection apparatus 100 may normally include other necessary components.

  FIG. 2 is a block diagram showing an internal configuration of the learning point area data creation circuit according to the first embodiment. 2, the learning point area data creation circuit 142 includes storage devices 73, 75, 76, 79, 80, 81, 87 such as magnetic disk devices, a learning point setting unit 70, a frame selection unit 71, and a design image creation. Unit 72, division unit 74, extraction unit 77, extraction unit 78, difference calculation unit 82, determination unit 83, frequency data conversion unit 84, band data extraction unit 85, inverse conversion unit 86, difference calculation unit 88, extraction unit 89, An extraction unit 90 and a selection unit 91 are arranged. Learning point setting section 70, frame selection section 71, design image creation section 72, division section 74, extraction section 77, extraction section 78, difference calculation section 82, determination section 83, frequency data conversion section 84, band data extraction section 85, Each of the “-units” such as the inverse conversion unit 86, the difference calculation unit 88, the extraction unit 89, the extraction unit 90, and the selection unit 91 has a processing circuit. Such a processing circuit includes, for example, an electric circuit, a computer, a processor, a circuit board, a quantum circuit, or a semiconductor device. Each of the units may use a common processing circuit (the same processing circuit) or different processing circuits (separate processing circuits). Learning point setting section 70, frame selection section 71, design image creation section 72, division section 74, extraction section 77, extraction section 78, difference calculation section 82, determination section 83, frequency data conversion section 84, band data extraction section 85, Information input / output to / from the inverse conversion unit 86, the difference calculation unit 88, the extraction unit 89, the extraction unit 90, and the selection unit 91 and the information being calculated are stored in a memory (not shown) each time.

  FIG. 3 is a conceptual diagram for describing an inspection area according to the first embodiment. As shown in FIG. 3, the inspection area 10 (the entire inspection area) of the sample 101 is virtually divided into a plurality of strip-shaped inspection stripes 20 having a scan width W, for example, in the Y direction. Then, the inspection apparatus 100 acquires an image (stripe region image) for each inspection stripe 20. For each of the inspection stripes 20, an image of a graphic pattern arranged in the stripe region is captured in the longitudinal direction (X direction) of the stripe region using laser light. The optical image is obtained while the photodiode array 105 is relatively continuously moved in the X direction by the movement of the XYθ table 102. The photodiode array 105 continuously captures an optical image having a scan width W as shown in FIG. In other words, the photodiode array 105, which is an example of the sensor, captures an optical image of the pattern formed on the sample 101 using the inspection light while relatively moving with respect to the XYθ table 102 (stage). In the first embodiment, after capturing an optical image of one inspection stripe 20, the optical image having the scan width W is similarly moved while moving in the Y direction to the position of the next inspection stripe 20 and in the opposite direction. Image continuously. That is, imaging is repeated in the forward (FWD) -back forward (BWD) direction heading in the opposite direction on the outward path and the return path.

  Here, the imaging direction is not limited to repetition of forward (FWD) -backforward (BWD). The image may be taken from one direction. For example, repetition of FWD-FWD may be used. Alternatively, BWD-BWD may be repeated.

  Here, the pixel data of the optical image picked up from the sample 101 is in a state where a filter is actuated due to the resolution characteristics of the optical system used for the image pickup, in other words, in an analog state that changes continuously, so that the image intensity (shading) Value) is different from a later-described developed image (design image) of a digital value. Therefore, a filter process is performed on the developed image so as to approach the measured image data and then perform the comparison process. For this purpose, it is necessary to first calculate the coefficients of a filter function for performing such a filtering process before the inspection process of the sample 101 is performed. However, an auxiliary pattern (SRAF pattern and / or OPC pattern) which is not resolved on a wafer arranged to improve a lithography margin or correct a dimensional error due to a proximity effect is a main pattern (a plurality of patterns) constituting a circuit. Pattern), the degree of coincidence in the case of creating a reference image tends to be lower. As a result, in the pattern defect inspection of the sample 101, there is a problem that a pseudo defect that is determined to be a defect is generated even for a well-formed auxiliary pattern. Therefore, it is required to create a reference image that is not determined to be a defect for a good auxiliary pattern. However, in the conventional method, it is difficult to select a location where the degree of coincidence is low because a learning pattern used for obtaining a filter function coefficient for creating a reference image from a developed image from design data is probabilistically extracted. Met. Therefore, in the first embodiment, a portion where the degree of coincidence is low is extracted so as to be surely included in the learning pattern (learning point).

  FIG. 4 is a flowchart showing a part of main steps of the pattern inspection method according to the first embodiment. Among the main steps of the pattern inspection method according to the first embodiment, a series of steps shown in FIG. 4 include steps of a coefficient calculation method of a filter function for creating a reference image. In FIG. 4, some of the main steps of the pattern inspection method according to the first embodiment include a learning point setting step (S102), a frame selection step (S104), a design image creation step (S106), and a stripe image acquisition. Step (S108), frame division step (S110), learning point area design image extraction step (S112), learning point area optical image extraction step (S114), filter coefficient calculation step (S116), and reference image creation Step (S118), difference calculation step (S120), determination step (S122), frequency data conversion step (S124), band data extraction step (S126), inverse conversion step (S128), learning point area A design image extracting step (S132), a learning point area optical image extracting step (S134), a difference calculating step (S136), and a selecting step (S138). When, performing a series of steps referred to.

  In the learning point setting step (S102), the learning point setting unit 70 sets a learning point for extracting a learning pattern used for obtaining a filter function coefficient for creating a reference image from a developed image from the design data. .

  FIG. 5 is a diagram illustrating an example of a learning point according to the first embodiment. In the example of FIG. 5A, a pattern edge is shown as an example of a learning point. In the example of FIG. 5B, a pattern corner is shown as an example of a learning point. Here, the pattern size is not set.

  In the frame selection step (S104), the frame selection unit 71 selects the frame area 30 from which the learning point area including the learning point is extracted from the inspection area 10 of the sample 101. The selection of the frame region 30 may be arbitrarily performed. Alternatively, the user designates a rough area in which the auxiliary pattern is arranged together with the main pattern from the pattern layout formed on the sample 101, and the frame selecting unit 71 selects a frame from the designated rough area. The region 30 may be arbitrarily selected. Alternatively, the user designates a frame region 30 in which the auxiliary pattern is arranged together with the main pattern from the pattern layout formed on the sample 101, and the frame selecting section 71 selects the designated frame region 30 and Is also good. Here, as shown in FIG. 3, the frame area 30 is an area obtained by dividing the inspection stripe 20 in the x direction by a predetermined size (for example, the same width as the scan width W). For example, it corresponds to a region of 1024 × 1024 pixels (or, for example, 512 × 512 pixels). In the first embodiment, the frame selection unit 71 selects several (for example, 3 to 7) frame regions 30.

  In the design image creation step (S106), the design image creation unit 72 controls the expansion circuit 111 to create a design image of the selected frame region 30. The development circuit 111 (an example of a design image creation unit) develops an image based on design pattern data on which the pattern of the sample to be inspected 101 is formed to create a design image (development image). Specifically, the design data is read from the magnetic disk device 109 through the control computer 110, and each graphic pattern defined in the design data of the selected frame area 30 is converted into binary or multi-valued image data (image development). To create a design image (expanded image).

  Here, the figure defined in the design pattern data is, for example, a rectangle or triangle as a basic figure. For example, the coordinates (x, y) at the reference position of the figure, the length of the side, the rectangle or triangle, etc. Graphic data (vector data) defining the shape, size, position, etc. of each pattern graphic is stored with information such as a graphic code as an identifier for distinguishing the graphic type.

When the information of the design pattern as the graphic data is input to the expansion circuit 111, the data is expanded to data for each graphic, and a graphic code indicating the graphic shape of the graphic data, a graphic size, and the like are interpreted. Then, binary or multi-valued design image data is developed and output as a pattern arranged in a grid having a grid of a predetermined quantization size as a unit. In other words, the design data is read, and the occupation ratio of the figure in the design pattern is calculated for each of the squares formed by virtually dividing the inspection area as a unit having a predetermined dimension as a unit. Output. For example, it is preferable to set one square as one pixel. Assuming that one pixel has a resolution of 1/2 ⁸ (= 1/256), a small area of 1/256 is allocated to the area of the figure arranged in the pixel, and the occupation ratio in the pixel is reduced. Calculate. Then, a design image of 8-bit occupancy data is created for each pixel. The design image is sent to the learning point area data creation circuit 142 together with data indicating the position. The data of the design image (expanded image) output to the learning point area data creation circuit 142 is stored in the storage device 76.

  In the stripe image obtaining step (S108), the optical image obtaining mechanism 150 obtains an optical image of the pattern-formed test sample 101. Here, under the control of the control computer 110, the optical image acquisition mechanism 150 acquires an optical image of the inspection stripe 20 including the selected frame region 30 in the inspection region 10 of the photomask serving as the sample 101. Specifically, it operates as follows.

  First, the XYθ table 102 is moved to a position where the inspection stripe 20 including the selected frame region 30 can be imaged. The pattern formed on the sample 101 is irradiated with laser light (for example, DUV light) having a wavelength equal to or shorter than the ultraviolet region serving as inspection light from the appropriate light source 103 via the illumination optical system 170. The light transmitted through the sample 101 forms an optical image on a photodiode array 105 (an example of a sensor) via a magnifying optical system 104 and enters the photodiode array 105. As the photodiode array 105, for example, a TDI (time delay integration) sensor or the like is preferably used.

  The image of the pattern formed on the photodiode array 105 is photoelectrically converted by each light receiving element of the photodiode array 105 and further A / D (analog / digital) converted by the sensor circuit 106. Then, the pixel data of the inspection stripe 20 to be measured is stored in the stripe pattern memory 123. When capturing such pixel data (striped region image), the dynamic range of the photodiode array 105 is, for example, a dynamic range that maximizes the gradation when the amount of illumination light is 60% incident. Thereafter, the stripe area image is sent to the learning point area data creation circuit 142 together with data indicating the position of the photomask 101 on the XYθ table 102 output from the position circuit 107. The measurement data (pixel data) is, for example, 8-bit unsigned data, and expresses the brightness gradation (light amount) of each pixel. The stripe area image (stripe data) output to the learning point area data creation circuit 142 is stored in the storage device 73.

  In the frame dividing step (S110), the dividing unit 74 reads out the stripe region image from the storage device 73, and cuts out the frame image of the selected frame region 30 by dividing the stripe region image. The generated frame image is stored in the storage device 75.

  In the learning point area design image extracting step (S112), the extraction unit 77 reads the design image of the selected frame area 30 from the storage device 76, and extracts a learning point area design image from the design images.

  FIG. 6 is a diagram illustrating an example of a learning point region image according to the first embodiment. In FIG. 6, a set learning point is searched from the design image of the selected frame area 30. Then, a learning point area design image of the learning point area 32 including the searched learning points is extracted. The learning point area 32 has a size of L1 × L1 centered on the searched learning point. For example, it is composed of an area of 31 pixels × 31 pixels. A learning point area design image of a learning point area 32 of L1 × L1 (for example, 31 × 31 pixels) is extracted from a design image of a frame area 30 of L2 × L2 (for example, 1024 × 1024 pixels). I do. Depending on the contents of the set learning points, as described above, when setting pattern edge points and corner points as learning points, learning point area design of learning point areas 32 of, for example, several thousand points from one design image is performed. Extract the image. The extracted learning point region design image is stored in the storage device 79.

  In the learning point area optical image extracting step (S114), the extraction unit 78 reads the frame image (optical image) of the selected frame area 30 from the storage device 75, and extracts the learning point area optical image from the frame images. . Specifically, the extraction unit 78 extracts the learning point area optical image of the learning point area 32 corresponding to the extracted learning point area design image from the frame image (optical image) of the selected frame area 30. I do. As in the case of the learning point area design image, for example, a learning point area optical image of several thousand learning point areas 32 is extracted from one frame image. The extracted learning point region optical image is stored in the storage device 80.

  Here, as described above, although the characteristic part of the pattern shape is set as the learning point, the size of the pattern is not set, so the learning point area design image and the learning point area optical It is not known whether the pattern in the image is a main pattern or an auxiliary pattern. Therefore, it is not known at this time whether or not an auxiliary pattern that tends to have a low degree of coincidence has been conveniently selected.

  In the filter coefficient calculation step (S116), the coefficient calculation circuit 140 (coefficient calculation unit) reads the learning point region design image from the storage device 79 and reads the learning point region optical image from the storage device 80. The coefficient calculation circuit 140 uses the learning point region design image (part of the design image) and the learning point region optical image (part of the optical image) to filter coefficients of the filter function for filtering the design image. Is calculated.

  FIG. 7 is a diagram for explaining an example of a method of calculating coefficients of a filter function according to the first embodiment. For example, as shown in FIG. 7A, an unknown coefficient matrix a (m, n) (an example of a coefficient) composed of (2k + 1) × (2k + 1) elements smaller than the number of pixels in the learning point region 32 ). For example, a 15 × 15 coefficient matrix a (m, n) is obtained for an image of the learning point area 32 composed of 31 × 31 pixels. With the pixel of interest d (i, j) in the learning point region design image as the center, the sum of the product of the pixel of (2k + 1) × (2k + 1) and the coefficient matrix a (m, n) is the pixel of interest d (i, j). A coefficient matrix a (m, n) closer to the target pixel r (i, j) of the learning point area optical image corresponding to j) is obtained. The relational expression (1) is shown below.

  As shown in FIG. 7B, the relational expression (1) is calculated each time the pixel of interest is moved in the learning point area 32. Then, the coefficient matrix a (m, n) that most satisfies the relational expression (1) defined using the unknown coefficient matrix a (m, n) obtained for each of all the pixels in the learning point region 32. Ask for. The number of elements (2k + 1) × (2k + 1) of the coefficient matrix a (m, n) may be set as appropriate. If the amount is small, the accuracy is deteriorated, and if the amount is too large, the calculation time is long. When the target pixel moves in the learning point area 32, there may be a case where the necessary pixels around the end side do not exist as necessary if the pixel is close to the end. What is necessary is just to calculate by N. Since the learning point area 32 is formed around the learning point, the learning point is configured so that the learning point does not become an end. As described above, since several thousands of learning point region design images are extracted from one set image, for example, in such a case, the relationship between the number of values obtained by multiplying several thousand by the number of pixels of the learning point region design image is obtained. A coefficient matrix a (m, n) that most satisfies Expression (1) is obtained.

  The coefficient matrix a (m, n) (an example of a coefficient) obtained as described above is output to the reference circuit 112 and set as a coefficient of the filter function. Next, it is confirmed whether the obtained coefficient matrix a (m, n) (an example of a coefficient) is appropriate.

  In the reference image creation step (S118), the reference circuit 112 (reference image creation unit) creates a reference image by filtering the design image using the obtained coefficient matrix a (m, n). The image used for extracting the learning point region design image may be used as the design image to be used. One or more design images may be used.

  FIG. 8 is a diagram for describing the filtering process according to the first embodiment. Since the measurement data as an optical image obtained from the sensor circuit 106 is in a state where a filter is actuated due to the resolution characteristics of the magnifying optical system 104 and the aperture effect of the photodiode array 105, in other words, it is in an analog state that changes continuously, By applying a filtering process to the reference design image data, which is image data on the design side of which the image intensity (shade value) is a digital value, it is possible to match with the measurement data. In this way, a reference image to be compared with the frame image (optical image) is created. The created reference image is output to the learning point area data creation circuit 142, and the reference image output to the learning point area data creation circuit 142 is stored in the storage device 81.

  In the difference calculation step (S120), the difference calculation unit 82 calculates the degree of coincidence between the entire pattern in the created reference image and the entire pattern in the corresponding frame image. Specifically, the difference calculation unit 82 reads a reference image from the storage device 81 and reads a corresponding frame image from the storage device 75. Then, the difference calculator 82 calculates a difference between each pixel value of the reference image and a corresponding pixel value of the frame image.

  In the determination step (S122), the determination unit 83 determines whether the maximum difference value indicating the maximum value among the calculated difference values of all the pixels is larger than the threshold value Th. If the maximum difference value is not greater than the threshold Th, it is used as it is as an appropriate coefficient. In such a case, the extracted learning point area design image accidentally includes an auxiliary pattern having a low degree of coincidence in addition to the main pattern. If the maximum difference value is larger than the threshold value Th, it means that the learning point area design image including the auxiliary pattern having the low matching degree has not been extracted. Therefore, in the first embodiment, a portion where the degree of coincidence is low is extracted so as to be surely included in the learning pattern (learning point).

  In the frequency data conversion step (S124), the frequency data conversion unit 84 (the conversion unit) converts the design image into frequency data. The image used for extracting the learning point region design image may be used as the design image to be used. One or more design images may be used. The frequency data conversion unit 84 converts the design image into frequency data by performing a Fourier transform process on the data of the design image. The design image is, for example, an image in a spatial region where the length of the pattern can be defined. An image in such a spatial domain is transformed into a frequency domain.

  FIG. 9 is a diagram for explaining a method of frequency data conversion according to the first embodiment. In the example of FIG. 9, a plurality of graphic patterns having different sizes are arranged in the design image. In the example of FIG. 9, only a few graphic patterns are shown for the sake of convenience, but actually, a large number of graphic patterns are usually arranged. For this design image, for example, a Fourier transform is performed on the value scanned in the x direction at each position moved in the y direction in pixel units. Similarly, a Fourier transform is performed on the value scanned in the y direction at each position moved in the x direction, for example, in pixel units. By this processing, for example, a function (spectrum) of the frequency whose length has been converted is calculated.

  FIG. 10 is a diagram illustrating an example of a relationship between frequency and spectrum intensity according to the first embodiment. In FIG. 10, the vertical axis indicates the spectrum intensity, and the horizontal axis indicates the frequency. When a design image on which a plurality of graphic patterns are arranged is converted into frequency data, as shown in FIG. 10, the spectrum intensity gradually decreases from the low frequency side to the high frequency side and converges. In addition, although not shown, a plurality of small peaks may occur in the process of decreasing the spectrum intensity. In the relationship between the frequency and the spectrum intensity shown in FIG. 10, the spectrum intensity shown for each frequency band (band) has a characteristic. In other words, the size of the graphic pattern corresponding to the spectrum intensity indicated for each frequency band (band) is different. Therefore, unlike the main pattern, the auxiliary pattern formed to a size that is not resolved when the pattern is transferred to the wafer corresponds to the spectrum intensity shown in the specific frequency band.

  In the band data extraction step (S126), the band data extraction unit 85 (extraction unit) extracts data of a part of the frequency band Δ from the frequency data. The band data extraction unit 85 extracts, as data of a part of the frequency band, data of a frequency band having a line width smaller than that of the main pattern (actual circuit pattern) and corresponding to an auxiliary pattern for assisting the formation of the main pattern. In other words, the data of the frequency band (band) Δ corresponding to the auxiliary pattern having the low matching degree is extracted. For example, data of the frequency band of F3 is extracted. Which frequency band corresponds to the size of the auxiliary pattern may be determined in advance by simulation or the like. The width of each frequency band Δ of F1 to F5 shown in FIG. 10 is preferably set to, for example, about several tens of Hz (for example, 50 Hz). The frequency band (band) to be extracted is not limited to one frequency band, but may be a plurality of frequency bands. The main pattern is assumed to correspond to a low-frequency band, but the high-frequency band may include an n-th harmonic (n is an integer of 2 or more) of the frequency corresponding to the main pattern. Therefore, it is preferable that bands including these (for example, F1 and F5 shown in FIG. 10) are excluded from extraction targets.

  In the inverse transform step (S128), the inverse transform unit 86 inversely transforms the extracted data of a part of the frequency band to generate a spatial domain image. The inverse transform unit 86 performs an inverse transform process of the Fourier transform process on the frequency data of the design image, thereby converting the frequency domain image into the spatial domain image. The converted space area image is stored in the storage device 87.

  FIG. 11 is a diagram illustrating an example of a spatial domain image obtained from data of a part of the frequency bands according to the first embodiment. Here, for convenience, it is preferable to set the same size as the design image. Since the graphic pattern obtained from the data of some frequency bands is a graphic pattern smaller in size than the main pattern, including the auxiliary pattern formed to a size that is not resolved, the spatial area image is Is constituted by a small figure pattern.

  In the learning point area design image extraction step (S132), the extraction unit 89 reads the spatial area image generated by the inverse transformation from the storage device 87. Then, the extraction unit 89 searches for the set learning point from the space area image. Then, a learning point area design image of the learning point area 32 including the searched learning points is extracted. The learning point area 32 has a size of L1 × L1 centered on the searched learning point. For example, it is composed of an area of 31 pixels × 31 pixels. The method of extraction may be the same as in the learning point area design image extraction step (S112). Here, a learning point area design image is extracted from a spatial area image in which auxiliary patterns having a low degree of coincidence are selectively arranged. Therefore, it is possible to selectively extract the learning point region design image including the auxiliary pattern having the low matching degree. When a plurality of auxiliary patterns are arranged in the space area image, a plurality of learning point area design images can be extracted.

  In the learning point region optical image extraction step (S134), the extraction unit 90 performs learning corresponding to the learning point region design image extracted from the spatial region image from among the frame images (optical images) corresponding to the spatial region image. A learning point area optical image of the point area 32 is extracted. When a plurality of learning point region design images are extracted, it goes without saying that a plurality of learning point region optical images are similarly extracted.

  In the difference calculation step (S136), the difference calculation unit 88 calculates each pixel value of the learning point area design image corresponding to the learning point area extracted from the spatial area image corresponding to the extracted partial frequency band data. The difference between the corresponding pixel value of the learning point area optical image and the corresponding pixel value is calculated.

  In the selection step (S138), the selection unit 91 selects a learning point region design image and a learning point region optical image from at least one learning point region design image corresponding to the learning point region extracted from the spatial region image. A learning point area design image in which the maximum difference value between the pixel values exceeds the threshold Th is selected. The selected learning point area design image is stored in the storage device 79. Similarly, the learning point area optical image corresponding to the selected learning point area design image is stored in the storage device 80. If the learning point area design image in which the maximum value of the difference value exceeds the threshold value Th is not extracted, it may be determined that there is no selection target. By comparing the design image and the optical image corresponding to the learning point region extracted from the spatial region image, the SRAF or OPC portion having a low degree of coincidence (a large difference) can be selectively added to the learning point region.

  Then, returning to the filter coefficient calculation step (S116), the coefficient calculation circuit 140 (coefficient calculation unit) corresponds to at least a part of the spatial domain image, a corresponding part of the frame image (optical image), and a corresponding part of the design image. The coefficients of the filter function for filtering the design image are calculated using the part. Specifically, the coefficient calculation circuit 140 generates a plurality of learning point region design images (parts of the design image) to which learning point region design images (parts of the design image) including the newly extracted auxiliary patterns are added. And a plurality of learning point region optical images (part of the optical image) to which the learning point region optical image (part of the optical image) including the newly extracted auxiliary pattern is added, and the design image is filtered. Calculate the coefficients of the filter function to be processed. The calculation of the coefficients of the filter function is performed on the design image and the corresponding optical image, and the inversely transformed spatial domain image is used to assist it. By inverting the specific frequency band, a spatial domain image in which SRAF and OPC are emphasized can be generated. A learning point region is extracted from the spatial domain image, and a coefficient is obtained using the corresponding optical image and design image.

  Then, in the determination step (S122), each step from the filter coefficient calculation step (S116) to the learning point area optical image extraction step (S134) is repeated until the maximum difference value does not become larger than the threshold value Th. Alternatively, the process is repeated up to a preset maximum number of repetitions. By repeating, optimization of the filter coefficient can be further advanced.

  As described above, according to the first embodiment, it is possible to reliably include an auxiliary pattern having a low degree of coincidence as a learning point for calculating a filter coefficient. As a result, the accuracy of the obtained filter coefficient can be improved.

  FIG. 12 is a configuration diagram illustrating an example of an internal configuration of the comparison circuit according to the first embodiment. 12, storage devices 50, 52, and 56 such as magnetic disk devices, a frame division unit 54, a positioning unit 58, and a comparison processing unit 59 are arranged in a comparison circuit 108. Each of the “-units” such as the frame division unit 54, the positioning unit 58, and the comparison processing unit 59 has a processing circuit. Such a processing circuit includes, for example, an electric circuit, a computer, a processor, a circuit board, a quantum circuit, or a semiconductor device. Each of the units may use a common processing circuit (the same processing circuit) or different processing circuits (separate processing circuits). Information input to and output from the frame division unit 54, the positioning unit 58, and the comparison processing unit 59 and information being calculated are stored in a memory (not shown) each time.

  FIG. 13 is a flowchart showing the remaining part of the main steps of the pattern inspection method according to the first embodiment. 5 shows a step following a series of steps shown in FIG. 4 among main steps of the pattern inspection method in the first embodiment. In FIG. 13, the remaining main steps of the pattern inspection method according to the first embodiment include an optical image acquisition step (S202), a frame division step (S204), a developed image creation step (S206), and a reference image creation step. A series of steps of (S208), an alignment step (S210), and a comparison step (S212) are performed.

  In the optical image obtaining step (S202), the optical image obtaining mechanism 150 obtains an optical image of a photomask serving as the sample 101. The method of acquiring the stripe image is the same as the above. However, here, stripe images are sequentially acquired as shown in FIG. Then, pixel data is stored in the stripe pattern memory 123 for each inspection stripe 20. Thereafter, the stripe region image is sent to the comparison circuit 108 together with data indicating the position of the photomask 101 on the XYθ table 102 output from the position circuit 107. The measurement data (pixel data) is, for example, 8-bit unsigned data, and expresses the brightness gradation (light amount) of each pixel. The stripe region image output to the comparison circuit 108 is stored in the storage device 50.

  In the frame division step (S204), in the comparison circuit 108, the frame division unit 54 has a stripe region image (optical image) of a predetermined size (for example, the same width as the scan width W) in the x direction for each inspection stripe 20. Is divided into a plurality of frame images (optical images) for each frame region 30. For example, the image is divided into frame images of 1024 × 1024 pixels (or, for example, 512 × 512 pixels). In other words, the stripe region image for each inspection stripe 20 is divided into a plurality of frame images (optical images) with a width similar to the width of the inspection stripe 20, for example, a scan width W. Through this processing, a plurality of frame images (optical images) corresponding to the plurality of frame regions 30 are obtained. The plurality of frame images are stored in the storage device 56. As described above, one image (measured image) data to be compared for the inspection is generated.

  In the developed image creating step (S206), the developing circuit 111 (design image creating unit) develops an image based on the design pattern data on which the pattern of the sample to be inspected 101 is formed to create a design image (developed image). . Specifically, the expansion circuit 111 reads the design data from the magnetic disk device 109 through the control computer 110, and converts each figure pattern defined in the read design data into a binary or multivalued image data. Convert to Then, a design image of 8-bit occupancy data is created for each pixel. The data (image data) of the design image (expanded image) is output to the reference circuit 112.

  In the reference image creation step (S208), the reference circuit 112 (reference image creation unit) uses a filter function in which the calculated filter coefficients are defined, and a plurality of frame images (a small area) of the plurality of frame areas 30 (small areas). Optical images) are created. In other words, using the set coefficient matrix a (i, j) (an example of a filter coefficient), the design image (developed image) of each frame region 30 is filtered to create a reference image. The created reference image of each frame region 30 is output to the comparison circuit 108, and the reference image output to the comparison circuit 108 is stored in the storage device 52.

  In the positioning step (S210), the positioning unit 62 reads the frame image (optical image) to be compared from the storage device 56, and similarly reads the reference image to be compared from the storage device 52. Then, positioning is performed by a predetermined algorithm. For example, alignment is performed using the least squares method.

  In the comparison step (S212), the comparison processing unit 59 (comparing unit) corresponds to the frame image (optical image) of the frame region 30 and the frame image for each of the plurality of frame regions 30 (small regions). The pattern is inspected for defects by comparing the reference image with each pixel. The comparison processing unit 59 compares the two for each pixel according to a predetermined determination condition, and determines whether or not there is a defect such as a shape defect. As a determination condition, for example, the two are compared for each pixel according to a predetermined algorithm, and the presence or absence of a defect is determined. Then, the comparison result is output. The comparison result is output to the magnetic disk device 109, magnetic tape device 115, flexible disk device (FD) 116, CRT 117, or pattern monitor 118. Alternatively, it may be output from the printer 119.

  As described above, in the first embodiment, the optimum value of the filter coefficient (or a value more suitable than the conventional method) can be obtained, and the accuracy of the filter processing can be improved. Therefore, a highly accurate reference image can be created. This makes it possible to create a reference image in which, among auxiliary patterns having a low degree of coincidence, a well-formed auxiliary pattern is not determined as a defect or the possibility of being determined is reduced. As a result, highly accurate pattern defect inspection can be performed, and inspection accuracy can be improved.

  The embodiment has been described with reference to the specific examples. However, the present invention is not limited to these specific examples. For example, in the embodiment, the transmitted illumination optical system using transmitted light is shown as the illumination optical system 170, but the present invention is not limited to this. For example, a reflection illumination optical system using reflected light may be used. Alternatively, the transmitted light and the reflected light may be used simultaneously by combining the transmitted illumination optical system and the reflected illumination optical system.

  In addition, although description is omitted for parts that are not directly necessary for the description of the present invention, such as the device configuration and the control method, the required device configuration and control method can be appropriately selected and used. For example, the description of the control unit configuration that controls the inspection apparatus 100 is omitted, but it goes without saying that the required control unit configuration is appropriately selected and used.

  In addition, all pattern inspection apparatuses and pattern inspection methods which include the elements of the present invention and whose design can be appropriately changed by those skilled in the art are included in the scope of the present invention.

Reference Signs List 10 inspection area 20 inspection stripe 30 frame area 32 learning point area 50, 52, 56 storage unit 54 frame division unit 58 positioning unit 59 comparison processing unit 73, 75, 76, 79, 80, 81, 87 storage unit 70 learning point Setting unit 71 frame selection unit 72 design image creation unit 74 division units 77 and 78 extraction unit 82 difference calculation unit 83 determination unit 84 frequency data conversion unit 85 band data extraction unit 86 inverse conversion unit 88 difference calculation unit 89 extraction units 89 and 90 Extraction unit 91 Selection unit 100 Inspection device 101 Sample 102 XYθ table 103 Light source 104 Magnifying optical system 105 Photodiode array 106 Sensor circuit 107 Position circuit 108 Comparison circuit 109 Magnetic disk drive 110 Control computer 111 Expansion circuit 112 Reference circuit 113 Autoloader control circuit 114 Table control times Road 115 Magnetic tape unit 116 FD
117 CRT
118 Pattern monitor 119 Printer 120 Bus 122 Laser length measurement system 123 Stripe pattern memory 140 Coefficient operation circuit 142 Learning point area data creation circuit 150 Optical image acquisition mechanism 160 Control system circuit 170 Illumination optical system

Claims (5)

An optical image acquisition mechanism for acquiring an optical image of a pattern-formed sample to be inspected,
A design image creating unit that creates a design image by developing a design pattern data based on which a pattern is formed on the sample to be inspected,
A conversion unit that converts the design image into frequency data,
Among the frequency data, an extraction unit that extracts data of some frequency bands,
An inverse transform unit that inversely transforms the extracted data of the partial frequency band to generate a spatial domain image,
Using at least a part of the spatial domain image, a corresponding part of the optical image, and a corresponding part of the design image, a coefficient for calculating a coefficient of a filter function for filtering the design image An operation unit;
A reference image creating unit that creates a reference image by filtering the design image using the coefficient,
A comparison unit that compares the optical image and the reference image for each pixel,
A pattern inspection apparatus comprising:   The extraction unit, as the data of the partial frequency band, is to extract data of a frequency band corresponding to an auxiliary pattern having a smaller line width than an actual circuit pattern and assisting the formation of the actual circuit pattern. The pattern inspection apparatus according to claim 1, wherein   The pattern inspection apparatus according to claim 1, wherein the conversion unit converts the design image into the frequency data by performing a Fourier transform process on the design image data.   4. The spatial region image, wherein a small region including a pixel whose pixel value difference between the corresponding optical image and the design image exceeds a threshold value is used for calculating the coefficient. 5. Pattern inspection equipment. A step of obtaining an optical image of the pattern-formed inspection sample,
A step of developing a design pattern data based on the pattern formation of the sample to be inspected to create a design image,
Converting the design image into frequency data,
A step of extracting data of some frequency bands from the frequency data;
Inversely transforming the extracted data of the partial frequency band to generate a spatial domain image,
Calculating a filter function coefficient for filtering the design image using at least a part of the spatial region image, a corresponding part of the optical image, and a corresponding part of the design image; When,
Using the coefficients, filter the design image to create a reference image,
A method for creating a reference image, comprising:

Images