JP2006091790A - Pattern inspecting method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To generate reference data speedily at a high approximation level from design data, and to enhance an inspection precision and an inspection speed. <P>SOLUTION: When the reference data are generated from the design data, a scan list defined by coordinates of intersections found by scanning graphic data represented as polygons in a line direction in units of sub-pixels obtained by dividing a pixel or bit maps by the units of sub-pixels are generated, and an image intensity distribution corresponding to the transmissivity or phase distribution of a sample is calculated, using a physical model for partial coherent imaging. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a technique for inspecting a pattern formed on a sample such as a semiconductor exposure mask or a liquid crystal substrate, and in particular, a die-to-database method for comparing and collating measurement data obtained from a sample with reference data obtained from design data. The present invention relates to a pattern inspection method.

  As a method for inspecting a sample such as a mask for semiconductor exposure, a die-to-die comparison method in which an image pattern between adjacent dies is compared and a mismatch is detected as a defect, and a die to be inspected There is a die-to-database comparison method that compares the image pattern with design data.

  While the die-to-die comparison method may miss common defects, the die-to-database comparison method is a reliable method, but it is necessary to generate reference image data corresponding to an image pattern from design data. For this purpose, for example, a binary or gray level raster image is generated by painting design data expressed in a polygon such as a rectangle or a triangle in units of pixels, and an optical point response function is convolved. It is widely done. In order to further improve the degree of approximation, it is conceivable to estimate a point response function from an image pattern, or to employ a physical model with a large amount of calculation. It is also conceivable to adopt a partially coherent imaging model, which is a general model in lithography simulation.

  Next, as a technique for improving the inspection speed, a high-speed image processing algorithm using run-length coding is known. In run-length encoding, an image is encoded with a continuous number of white and black binary images on the same line. As this image processing algorithm, a two-dimensional convolution method (for example, see Non-Patent Document 1), a morphology erosion or an opening method, and the like are known (for example, Non-Patent Document 2).

  Based on Non-Patent Document 1, a two-dimensional convolution algorithm will be briefly described. Assume that the output y (t) is obtained by convolving f (t) with the input u (t). When this is expressed in the frequency space, Y (s) = F (s) · U (s). Here, Y (s), F (s), and U (s) are Laplace transforms of y (t), f (t), and u (t), respectively. However, since Y (s) = (1 / s) {F (s) · sU (s)} holds, the input U (s) is first differentiated from this equation, and then f (t) and convolution. Thus, it can be seen that the same result can be obtained even if integration is performed at the end. Since the input u (t) given by the run length is the sum of the step functions, if it is differentiated, it becomes a Dirac delta function, and the convolution of the delta function and an arbitrary function g (t) is simply a parallel of g (t) Since it is a movement, convolution can be omitted. This can significantly reduce the labor of product-sum operation.

If morphological technology is utilized as described in Non-Patent Document 2, resizing and corner rounding processing can be realized. Run-length encoding itself is one of encodings that can be used for binary images such as faxes and can compress information without loss of information (see Non-Patent Document 3, for example). Further, as described above, the processing algorithm can be made more efficient.
Patrice Simard, Leon Bottou, Patrick Haffner, and Yann Le Cun, "Boxlets: a fast convolution algorithm for neural networks and signal processing," in Advances in Neural Information Processing Systems, (Denver), 1999. F. Ercal, F. Bunyak, F. Hao, and L. Zheng, "A Fast Modular RLE-based Inspection Scheme for PCBs", Proc. Of SPIE-Architectures, Networks, and Intelligent Systems for Manufacturing Integration, Pittsburg, Oct. 1997, Vol. 3203, pp. 49-59. Ochi Tomohiro, Kuroda Hideo "Image compression technology understood by illustration", Nihon Jitsugyo Publishing (1999)

  When an image obtained by imaging a sample such as a semiconductor exposure mask is inspected by comparing it with design data, it is important to generate a reference image from the design data with a good approximation. In order to generate a reference image with good approximation, it is necessary to input design data with high resolution first. Next, it is also very effective to perform calculation by an algorithm based on not only an optical image forming model but also a physical model incorporating the transmittance and phase distribution of the sample. Furthermore, in order to actually perform the inspection, not only the degree of approximation but also the processing speed of the algorithm is important. That is, a reference image generation algorithm having a good degree of approximation and a high processing speed is required.

  On the other hand, in the conventional technique, the physical model is not always taken into the algorithm, and the approximation degree of the reference image may be deteriorated particularly in a minute pattern. Even if the pixel size can be reduced to improve the approximation of the reference image, the inspection speed may be reduced.

  The present invention has been made in consideration of the above circumstances, and the object of the present invention is to create reference data at high speed from design data with a high degree of approximation, and to improve inspection accuracy and inspection speed. It is to provide a pattern inspection method that can be measured.

  In order to solve the above problems, the present invention adopts the following configuration.

  That is, according to one aspect of the present invention, inspection data obtained by imaging a two-dimensional pattern formed on a sample is compared with reference data obtained from design data corresponding to the two-dimensional pattern. A pattern inspection method for performing defect inspection of a pattern, the step of inputting graphic data represented by a plurality of polygons from the design data, and line-up the polygons in units of sub-pixels obtained by dividing pixels by a division number N Creating a scan list defined by a start point and an end point when scanning in the direction, M being an integer greater than the division number N, i being an integer from 1 to M, and j being a pixel line number, i + N × jth For the first line and the second line corresponding to the (M + 1−i) + N × jth sub-line on the second line included in the vicinity of the start point and end point of the scan list belonging to the first line Different values are selected depending on whether or not the pixel is included in the scan list belonging to the second line, and the selected values are accumulated in the sub-pixel near the start point and the sub-pixel near the end point of the first line, respectively. The process is repeated while increasing i by 1; the process of cumulatively adding each line subjected to the cumulative process for the subpixels near the start point and the end point; and the cumulative addition for each line that has been cumulatively added , By performing down-sampling every N sub-pixels, and generating gradation value data in pixel units for each line, and convolving a point response function to the gradation value data.

  According to another aspect of the present invention, inspection data obtained by imaging a two-dimensional pattern formed on a sample is compared with reference data obtained from design data corresponding to the two-dimensional pattern. A pattern inspection method for performing a defect inspection of a pattern by the step of inputting graphic data represented by a plurality of polygons from the design data, and the polygon in units of sub-pixels obtained by dividing pixels by a division number N Creating a first scan list defined by a start point and an end point when scanning the image in the line direction, and performing black / white reversal, contraction, or corner rounding on a graphic represented by the first scan list in units of subpixels To create a second scan list, and using the second scan list, M is an integer larger than the division number N, i is an integer from 1 to M, j is a pixel line number, and i + N Sub-pixels on the second line included in the vicinity of the start and end points of the scan list belonging to the first line with respect to the first line corresponding to the jth and the second line corresponding to the (M + 1−i) + N × jth Is selected depending on whether or not is included in the scan list belonging to the second line, and the selected value is accumulated in the sub-pixel near the start point and the sub-pixel near the end point of the first line, respectively. Is repeated while increasing i by one, the step of cumulatively adding each line subjected to the accumulation processing for the subpixels near the start point and the end point, and the cumulative addition line, A step of generating pixel value grayscale data for each line by downsampling every N subpixels, and a step of convolving a point response function to the grayscale data , Characterized in that it comprises a.

  According to another aspect of the present invention, inspection data obtained by imaging a two-dimensional pattern formed on a sample is compared with reference data obtained from design data corresponding to the two-dimensional pattern. A pattern inspection method for performing defect inspection of a pattern by the step of inputting graphic data expressed by a plurality of polygons from the design data, and the sub-pixel unit in which pixels are divided by a division number N × N A step of creating a bitmap from a polygon, and values of two points on the bitmap at a diagonal position centered on a pixel downsampled every N subpixels and having a chessboard distance of M or less. Comparing and accumulating values determined by the distance between the two points to generate gray level data in units of pixels for each line, and combining the point response function with the gray level data Characterized in that it comprises the steps of Shon, the.

  According to the present invention, when creating reference data from design data, scanning defined by coordinates of intersections obtained by scanning graphic data represented by polygons in a line direction in units of sub-pixels obtained by dividing pixels. By generating a list or bitmap and calculating the image intensity distribution according to the transmittance and phase distribution of the sample based on the physical model of partial coherent imaging, the reference data can be quickly transferred from the design data with a high degree of approximation. Can be created. Therefore, the inspection accuracy and inspection speed can be improved.

  The details of the present invention will be described below with reference to the illustrated embodiments. This embodiment is a die-to-database pattern inspection method that compares and collates measurement data obtained by imaging a pattern formed on a mask with reference data obtained from design data, and has a high degree of approximation from design data. This is a method for creating reference data at high speed.

  First, the basic processing flow of this embodiment is shown in FIGS. The example of FIG. 1 includes a step of inputting graphic data (S1), a step of creating a scan list (S2), a step of obtaining gray value data from the scan list (S5), and a step of convolving to gray value data (S7). Consists of.

  In the example of FIG. 2, a step of resizing the scan list (S3) and a step of rounding the scan list (S4) are added to the example of FIG. Further, in the example of FIG. 3, the step of obtaining gray value data from the scan list (S5) is independently performed from the step of inputting graphic data (S1) in the example of FIG. Step (S6) to be added is added.

  Hereinafter, this embodiment will be described in detail.

  First, elements constituting this embodiment will be described. A scan list is adopted as the data structure. The scan list defines graphic data as a segment defined by a start point and an end point intersecting with a scan line in the scan direction (line direction) in a sub-pixel unit obtained by dividing a pixel, and a pointer to an adjacent segment on the same line. . The graphic data is represented by a polygon including an arbitrary triangle or trapezoid. The scan list is sorted so that the coordinate values of the segments are in ascending order, and nodes on the same line can be searched by sequentially tracing the pointers. Here, in order to maintain the positional accuracy, the start point and the end point are defined by the coordinates of the sub-pixel unit obtained by dividing the pixel. The pixel division number is often a power of 2. However, if the division number is large, the image resolution is improved, but the processing speed may be lowered. Therefore, it is necessary to determine an optimal one from the viewpoint of both. .

  Next, a method for synthesizing a scan list by inputting a plurality of polygons will be described. FIGS. 4A and 4B show an example of synthesizing three graphic data. The horizontal axis is the scanning direction, and the vertical axis is the stage traveling direction. As shown in FIG. 4 (a), the coordinates of the start point and end point of the segment of the graphic data 11, 12, 13 to be newly inserted are scanned in order, and the segment is connected to a node in the existing scan list. If so, replace the start or end point of the partner segment so that the two segments are connected. If the segments are not connected, the segment is inserted at a position where the coordinates of the scan list are in ascending order, and the scan list is updated. As a result, as shown in FIG. 4B, the graphic data 11 to 13 are combined into a single graphic data 10.

  Next, a method for creating a scan list at high speed will be described. The graphic data is sorted in advance by quick sort, heap sort, etc. so that the coordinate values are in descending order. As a result, most of the newly inserted segments can be inserted only by determining the connection with the head node of the existing scan list. In FIG. 4, this corresponds to processing in the order of graphic data 13, graphic data 12, and graphic data 11. Here, the sorting of the graphic data is not limited to the descending order, and if the sorting is performed using a rule suitable for the data structure, it is expected that the insertion speed of the segment is increased even for different data structures.

  Next, a process of inputting a scan list and performing black and white inversion will be described. This can be easily realized by performing NOT operation on the scan list. In other words, it is only necessary that the area between the input scan lists is the scan list to be output.

  Next, a process for inputting and resizing the scan list will be described. A morphological erosion algorithm for the run-length data shown in Non-Patent Document 2 may be employed. As shown in FIG. 5, a 5 × 5 template 21 of sub-pixels is translated so that the template 21 is completely included in the input figure. The resized figure is defined by the union of the central pixels of the template 21 in this state. In the figure, a solid line 22 is an input figure, and a dotted line 23 is a resized output figure. This is an example of contraction processing, but in the case of expansion processing, black and white inversion, contraction processing, and black and white inversion may be performed.

  Next, a process of inputting a scan list and rounding corners will be described. Corner rounding can also be realized by a morphological opening algorithm for run-length data. The opening algorithm can be performed in the same manner as the erosion algorithm. As shown in FIG. 6, if the union of the templates 31 that are completely included in the input figure is obtained by translating the template 31 that approximates a circle, the figure after rounding the corner is defined. In the figure, a solid line 32 is an input figure, and a dotted line 33 is a resized output figure. This is an example of convex corner rounding. In the case of concave corner rounding, black and white inversion, convex corner rounding, and black and white inversion may be performed. Thus, the rounding amount of the convex corner and the concave corner can be adjusted independently.

  Next, a method of generating grayscale data by an algorithm that inputs a scan list and incorporates a physical model of partial coherent imaging in consideration of optical characteristics such as transmittance and phase difference of the sample will be described. Although I won't go into details, it is possible to apply the same technique as the convolution described in the background art because the input data is represented by a scan list.

  As a sample, a quartz region having a transmittance of 1, a halftone film having a transmittance of t and a phase difference of φ is assumed. The area included in the scan list is a quartz area. For example, t = 6-50% and φ = 60-180 degrees are typical values.

  The algorithm for calculating the gray value from the scan list may repeat the following flow as shown in FIG. That is, the calculation is performed for each pixel line, and for each pixel line, the calculation is performed for each sub-pixel while sequentially moving the scan list. The calculation loop for each sub-pixel in FIG. 7 corresponds to Step 2 described later, and the calculation loop for each line corresponds to the first step and the third step described later.

  Here, the number of divisions is N, a pixel divided by the number of divisions is defined as a subpixel, and all coordinate units are subpixels. As shown in FIG. 8A, the start point and end point of the segment are read by sequentially tracing the segments of the scan list of the scan line 1. Next, a different value is obtained depending on whether or not the pixels near the start point and end point of the segment of the scan line 1 are included in the scan list of the scan line 2, and are cumulatively added in units of sub-pixels (FIG. 8 ( b)). For the start point, points 1 to 2 are applicable, and for the end point, points 3 to 4 are applicable. The relationship between the scanning line 1 and the scanning line 2 is that when M is an integer greater than N, i is an integer from 1 to M, and the scanning line 1 is the i-th line, the scanning line 2 is (M + 1−i). The second line.

  If the number of pixel lines is j, since the sub-pixel divides the pixel by N, the scanning line 1 corresponds to the (i + N × j) th pixel line number j, and the scanning line 2 is This corresponds to (Mi) + N × jth. For example, if i = 1 and j = 0, scan line 1 corresponds to the first, scan line 2 corresponds to the Mth, and if i = 1 and j = 1, scan line 1 corresponds to 1 + Nth. Therefore, the scanning line 2 corresponds to the M + Mth.

  The processing flow for each pixel line is shown below. FIG. 9 shows the flow of correlation calculation between scan lists. This flow represents an outline, and the coefficient and the like are determined according to the magnitude of M.

  Repeat step 1 from i = 1 to M.

  The array a [k] is cleared with the value 0 from k = 0 to P.

Step 1: {
The start point Y1 and end point Y2 of the current node are taken from the scan list belonging to the i-th line.

    Repeat step 2 from k = -M to M.

Step 2: {
When the point of coordinates (Y1 + k) with respect to the start point Y1 of the node is included in the scan list belonging to the (M + 1−i) th line, A = 1−t cosφ in the array a [Y1 + k / 2].
Is added. If not included, B = t × (cosφ−t) is added.

If the point at the coordinate (Y2 + k) with respect to the end point Y2 of the node is included in the scan list belonging to the (M + 1−i) th line, −A is added to the array a [Y2 + k / 2].

      If not included, -B is added.

}
Proceed to the next node from the scan list belonging to the i-th line. If the next node does not exist, step 1 is skipped.

}
Let val = t ² .

  Repeat step 3 from i = 1 to P.

Step 3:
val = val + a [i]
The val is sampled every N and the gray value is output for each pixel.

  Here, FIG. 8C shows a result of cumulative addition from the end of each line in which accumulation processing is performed on the subpixels near the start point and the end point, and sampling is performed every N subpixels based on this result. Thus, gray value data in units of pixels for each line is generated.

  Next, an algorithm for calculating gray values at high speed will be described. As shown in FIG. 10, among the subpixels near the start point and end point of the segment of scan line 1, the number α included in the scan list of scan line 2 and the number β not included correspond to the subpixels. Obtained for each pixel. Thereafter, the gray value of the pixel is determined by a value determined from α and β. Note that the values of α and β are the same if the remainder obtained by dividing the start point and end point of the target line by N is the same, so that the pixel values corresponding to the remainder are calculated in advance as a database as shown in FIG. In addition, by referring to the pixel value from the remainder obtained by dividing the starting point and the ending point by N, it is possible to perform a higher speed calculation. At this time, the pixel value may be obtained in all cases in advance, or the calculation may be performed only when a new remainder appears adaptively during the calculation of the gray value.

Here, when the sample is a transparent object having a transmittance of 1 and a halftone film having an amplitude transmittance of t and a phase difference of φ, the cosine function is cosφ, and an arbitrary point included in the vicinity of the start point of the scan list is the scan. The values selected depending on whether they are included in the list are A = 1−t × cosφ and B = t × (cosφ−t), and at the same time, any point included in the vicinity of the end point is the scan list. The values selected depending on whether or not they are included are −A and −B, respectively, and the initial value at the end of the line is C = t ² . Here, it can be seen that A + B + C = 1.

  For example, if t = 50% and φ = 180 degrees are substituted, A = 1.5, B = −0.75, and C = 0.25. Here, the gradient value in the line direction represents the gradient of the image shape accompanying the transition from the quartz to the halftone film, and the image shape exhibiting an undershoot due to the phase difference is simulated by this effect. Gray value data can be obtained by accumulating and adding the gradient value in the line direction in the scanning direction with the initial value as C. Finally, a grayscale image is input, and a point response function that is an optical characteristic is convoluted. As a result, an image shape that matches the aberration of the lens and the characteristics of the camera can be provided.

  As described above, according to the present embodiment, when creating reference data from design data, the coordinates of intersections obtained by scanning the graphic data represented by polygons in the line direction in units of sub-pixels obtained by dividing the pixels. A scan list to be defined is generated, a predetermined value is accumulated for the sub-pixels regarding the start point and end point of the scan list, and each line subjected to the accumulation process is sequentially accumulated from the end, and then every N sub-pixels. By generating sampling grayscale data for each line by sampling and convolving a point response function to the grayscale data, reference data can be created at high speed from design data with a high degree of approximation. . Therefore, the inspection accuracy and inspection speed can be improved.

(Second Embodiment)
This method can be applied to a tritone mask in which chromium and a halftone film are mixed. FIG. 12 shows an application example. FIG. 12A is a cross-sectional view showing a mask structure, and FIGS. 12B to 12F are data waveform diagrams.

  First, first graphic data that defines a quartz region with a halftone film as a background and second graphic data that defines a halftone film region with a chrome film as a background are prepared. The advantage of this method is that it can be shared with the drawing data as the first graphic data and the second graphic data.

As a configuration in this case, a first scan list is generated from the first graphic data, resizing, corner rounding, and gray image generation are performed to obtain a first gray image (FIGS. 12B and 12C). ). The values of A, B, and C are the same as in the first embodiment. Independently of the processing of the first graphic data, a second scan list is generated from the second graphic data at the same time or after a while, and the second scan list is similarly generated for resizing, corner rounding, and grayscale image generation. A grayscale image is obtained (FIGS. 12D and 12E). Here, the points are changed to A ′ = t ² , B ′ = 0, and C ′ = − t ² . Here, A ′ + B ′ + C ′ = 0. Assuming that the amplitude transmittance t = 50%, A ′ = 0.25, B ′ = 0, and C ′ = − 0.25. Of course, it is possible to use values obtained by multiplying these values by appropriate coefficients in the implementation.

By obtaining the sum of the first grayscale image and the second grayscale image, the desired image intensity of the tritone mask is obtained (FIG. 12 (f)).

  This method is characterized in that the processing from inputting graphic data to generating a grayscale image can be performed independently of each other. Also, resizing and corner rounding can be performed under different conditions corresponding to the drawing apparatus and mask process conditions.

  However, it is necessary to satisfy the constraint that the edges of the first graphic data and the second graphic data are arranged at a certain distance from each other and that the first graphic data is included in the second graphic data. . An example of violating the constraint condition is shown in FIG.

  As shown in FIG. 13A, the edges of the first layer graphic 51 and the second layer graphic 52 need to be at least a certain distance apart from each other between the quartz halftone boundary and the chrome halftone boundary. This is because the influence cannot be ignored. The minimum value L of the distance needs to be increased in proportion to the size of M. In the case of FIGS. 13B and 13C, the normal phase shift mask is drawn after the glass opening is formed by drawing the first layer graphic 51 and then the second layer graphic 52 is drawn. This is based on the situation where the halftone part is formed by drawing. That is, regardless of whether or not the second layer graphic 52 is drawn, the glass opening drawn by the first layer graphic 51 is not the glass opening regardless of whether or not the second layer graphic is drawn. .

(Modification)
The present invention is not limited to the above-described embodiments. In the embodiment, as the accumulation process for the sub-pixels near the start point and end point of the first line, A or B is added to the range of ± M / 2 before and after the start point, and − Although A or -B is added, the accumulation range of the start point and end point is not limited to ± M / 2, and can be changed as appropriate according to the specification. Similarly, the values of A and B can be appropriately changed according to the specifications. Further, the black-and-white reversal, contraction, or corner rounding processing in units of subpixels is not always necessary and can be omitted. In addition, various modifications can be made without departing from the scope of the present invention.

  In the embodiment, the scan list is created. However, it can be changed to bitmap development as described below. That is, when a figure having a long side in the scanning direction is arranged, the number of runs is small and the effect of run length coding is remarkable. However, when a figure having a short side in the scanning direction is arranged, the number of runs may increase and processing efficiency may decrease. In such a case, instead of using run-length encoding, data obtained by dividing a pixel into N × N sub-pixels and developing it into a bitmap may be used. The model formula that gives the gray value I (x, y) is shown in the following formula.

I (x, y)
= ∬w (x0, y0) f (x-x0, y-y0) f ^* (x + x0, y + y0) dx0dy0
Here, w (x, y) represents a window function, and f (x, y) represents a complex amplitude of the sample. F ^* represents the complex conjugate of f.

  Further, two points on the bitmap where the pixel down-sampled every N sub-pixels is located at a diagonal center and the chessboard distance is M or less are compared and determined according to the distance between the two points. By accumulating the values indicating the obtained image intensities, gray value data in units of pixels for each line is generated. As shown in FIG. 14, 61 is a central pixel, and 62 and 63 are sub-pixels at diagonal positions.

A value indicating the image intensity of the halftone phase shift mask is defined below. There are three combinations of bitmap values: 0 for both, one for 0, the other for 1, and both for 1. The value of intensity I (x, y) is t ² when both the amplitude transmittance is t and the phase difference is φ and both are 0, t × cos φ when one is 0 and the other is 1, and both are 1 1

The figure which shows the basic flow of 1st Embodiment. The figure which shows the basic flow of 1st Embodiment. The figure which shows the basic flow of 1st Embodiment. FIG. 5 is a schematic diagram for explaining a method of combining a scan list by inputting a plurality of polygons. The schematic diagram for demonstrating the method to resize from a scanning list. The schematic diagram for demonstrating the method of rounding a corner from a scanning list. The figure which shows the flow which calculates | requires the grayscale image from a scanning list. The schematic diagram for demonstrating the method to produce | generate the gray value data from a scanning list. The figure which shows the flow for demonstrating the correlation calculation between scanning lists. The schematic diagram for demonstrating the 1st gray value data high speed generation method. The schematic diagram for demonstrating the 2nd gradation value data high-speed generation method. The schematic diagram for demonstrating the tone value data generation method of a tritone mask. The figure which shows the example of violation of constraint conditions. FIG. 5 is a schematic diagram illustrating a gray value generation method based on a bitmap.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Composite graphic data 11, 12, 13 ... Graphic data 21, 31 ... Template 22, 32 ... Input graphic 23, 33 ... Output graphic 51 ... First layer graphic 52 ... Second layer graphic 61 ... Down-sampled pixel 62 63 ... Pairs of sub-pixels at diagonal positions

Claims (7)

Pattern inspection method for performing pattern defect inspection by comparing and comparing inspection data obtained by imaging a two-dimensional pattern formed on a sample and reference data obtained from design data corresponding to the two-dimensional pattern Because
Inputting graphic data represented by a plurality of polygons from the design data;
Creating a scan list defined by a start point and an end point when the polygon is scanned in the line direction in sub-pixel units obtained by dividing the pixel by the division number N;
M is an integer larger than the division number N, i is an integer from 1 to M, j is a pixel line number, and the first line corresponding to i + N × jth and the second line corresponding to (M + 1−i) + N × jth Selecting a different value depending on whether or not sub-pixels on the second line included in the vicinity of the start point and end point of the scan list belonging to the first line are included in the scan list belonging to the second line, Repeating the process of accumulating the selected value in each of the subpixels near the start point and the end point of the first line while increasing i by one;
A step of cumulatively adding each line in which accumulation processing is performed on subpixels near the start point and the end point, sequentially from the end;
Generating gray value data in units of pixels for each line by down-sampling every N sub-pixels for each cumulatively added line;
Convolving a point response function to the gray value data;
A pattern inspection method comprising: Pattern inspection method for performing pattern defect inspection by comparing and comparing inspection data obtained by imaging a two-dimensional pattern formed on a sample and reference data obtained from design data corresponding to the two-dimensional pattern Because
Inputting graphic data represented by a plurality of polygons from the design data;
Creating a first scan list defined by a start point and an end point when the polygon is scanned in the line direction in sub-pixel units obtained by dividing the pixel by the division number N;
Creating a second scan list by performing black-and-white reversal, shrinkage, or corner rounding on the graphics represented by the first scan list in sub-pixel units;
Using the second scan list, M is an integer larger than the division number N, i is an integer from 1 to M, j is a pixel line number, and the first line corresponding to i + N × jth and (M + 1−i) + N × Whether or not sub-pixels on the second line included in the vicinity of the start point and end point of the scan list belonging to the first line are included in the scan list belonging to the second line with respect to the jth second line Selecting a different value depending on whether or not, and repeating the process of accumulating the selected value in the sub-pixel near the start point and the sub-pixel near the end point of the first line while increasing i by 1,
A step of cumulatively adding each line in which accumulation processing is performed on subpixels near the start point and the end point, sequentially from the end;
Generating gray value data in units of pixels for each line by down-sampling every N sub-pixels for each cumulatively added line;
Convolving a point response function to the gray value data;
A pattern inspection method comprising:   As a cumulative process for the subpixels near the start point and near the end point, when a point of coordinates (Y1 + k) with respect to the coordinate Y1 of the start point is included in the scan list belonging to the second line, the first scan line array a A = 1−t × cosφ is added to [Y1 + k / 2], where t is the amplitude transmittance or amplitude reflectance of the pattern, φ is the phase difference, and cosφ is the cosine function. If not, B = t × ( cosφ−t), and when the point of the coordinate (Y2 + k) is included in the scan list belonging to the second line with respect to the coordinate Y2 of the end point, −A is added to the array a [Y2 + k / 2], 3. The pattern inspection method according to claim 1, wherein if it is not included, the process of adding -B is repeated from k = -M to + M.   3. The pattern inspection method according to claim 1, wherein in the step of creating the scan list, the graphic data is sorted in advance so that coordinate values are in descending order.   Of the subpixels in the vicinity of the start point and end point of the scan list of the first line, the number α included in the scan list of the second line and the number β not included are obtained for each corresponding pixel of the subpixel. 3. The pattern inspection method according to claim 1, wherein after that, the gray value of the pixel is determined by a value determined from α and β.   6. The pattern inspection method according to claim 5, wherein a pixel value corresponding to the remainder divided by N is calculated in advance as a database, and the pixel value is referenced from the remainder obtained by dividing the start point and the end point by N. Pattern inspection method for performing pattern defect inspection by comparing and comparing inspection data obtained by imaging a two-dimensional pattern formed on a sample and reference data obtained from design data corresponding to the two-dimensional pattern Because
Inputting graphic data represented by a plurality of polygons from the design data;
Creating a bitmap from the polygon in units of sub-pixels obtained by dividing a pixel by a division number N × N;
Comparing two points that are diagonally centered around the pixel down-sampled every N sub-pixels and the chessboard distance is M or less, and accumulates the values determined according to the distance between the two points. A step of generating grayscale data in units of pixels for each line,
Convolving a point response function to the gray value data;
A pattern inspection method comprising:

Images