JP2013053986A - Pattern inspection method and device thereof - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a semiconductor pattern inspection device which detects a systematic defect without causing many pieces of misinformation.SOLUTION: A semiconductor pattern inspection device is constituted so as to calculate an identification boundary for identifying misinformation and a defect by using feature quantity extracted from an image obtained by picking up an inspection objet pattern, feature quantity of a part corresponding to the image obtained by picking up the inspection object pattern extracted from a design data image generated from design data, and information on teaching data created by using the image obtained by picking up the inspection object pattern and the design data image, to calculate image feature quantity of an inspection area of the inspection object pattern from the image obtained by picking up the inspection area of the inspection object pattern from the image obtained by picking up the inspection area of the inspection object pattern, to create the design data image from the design data corresponding to the inspection area of the inspection object pattern, to calculate feature quantity of the created design data image, and to detect a defect in the inspection area of the inspection object pattern based on the calculated image feature quantity of the inspection area of the inspection object pattern, the feature quantity of the design data image, and the identification boundary.

Description

  The present invention relates to a semiconductor pattern inspection method and apparatus.

  With the miniaturization of semiconductor circuit patterns, the resolution of an optical exposure apparatus has reached its limit, and it is becoming difficult to form a designed pattern on a semiconductor wafer. The pattern formed on the semiconductor wafer is liable to have a defect such as a line width deviating from a design value, degeneration at the pattern tip, or a change in the shape of the base of the pattern. Such a defect is called a systematic defect and occurs in common to all the dies. Therefore, it is difficult to detect by a method of comparing adjacent dies called a die-to-die comparison.

  On the other hand, Japanese Unexamined Patent Application Publication No. 2011-17705 (Patent Document 1) discloses a method of comparing an inspection target pattern with design data instead of comparing with an adjacent die. Specifically, an outline is extracted from the pattern to be inspected, and this is compared with design data represented by a line segment or a curve. If the deviation is large, it is determined as a defect. . Since it is a comparison with the design data, even systematic defects that occur in common on all dies can be detected in principle.

JP 2011-17705 A

N. Dalal, and B. Triggs: “Histograms of Oriented Gradients for Human Detection,” Computer Vision and Pattern Recognition, Vol. 1, pp. 886-893 (2005) T.Kurita, and S.Hayamizu, `` Gesture Recognition using HLAC Features of PARCOR Images and HMM based Recognizer, '' Proc. Of Inter. Conf. On Automatic Face and Gesture Recognition: FG'98, pp.422-427, 1998 . C. M. Bishop: Pattern Recognition and Machine learning (Japanese version, Volume 1): Springer Japan, pp. 185-190 (2006) C. M. Bishop: Pattern Recognition and Machine learning (Japanese edition, second volume): Springer Japan, pp. 35-55 (2006)

  In the pattern transferred on the wafer, there are many shape divergences (such as differences in roundness of corners) from design data that cannot be said to be defects. The invention described in Patent Document 1 is a method for determining a defect if there is a large difference between the contour line extracted from the inspection target pattern and the design data represented by a line segment or curve. It is difficult to discriminate between a discrepancy in shape that cannot be said to be a flaw and a systematic defect. As a result, when a systematic defect is to be detected, there is a problem that a deviation portion of a shape that cannot be said to be a defect is detected, and false information is frequently generated.

  The present invention solves the above-described problems of the prior art and provides a pattern inspection method and apparatus capable of detecting a systematic defect without causing frequent false alarms.

  In order to achieve the above object, in the present invention, a pattern inspection apparatus for detecting a defect in an inspection target pattern by comparing an image obtained by imaging the inspection target pattern with a design data image generated from the design data. An imaging unit that captures an inspection target pattern and obtains an image of the inspection target pattern, and an inspection target that calculates a feature amount of the image of the inspection target pattern from an image obtained by imaging the inspection target pattern by the imaging unit Pattern image feature quantity calculating means, design data image creating means for creating a design data image from design data, and design for calculating the feature quantity of the design data image created from the design data of the pattern to be inspected by this design data image creating means Data image feature quantity calculating means, and image and design data of the inspection target pattern obtained by imaging the inspection target pattern with the imaging means Instructions are given on the display means having a display screen for displaying the design data image of the inspection target pattern generated by the image creation means, and on the screen of the display means on which the image of the inspection target pattern and the design data image of the inspection target pattern are displayed The teaching image data creating means for creating and storing the teaching image data of the defect of the inspection target pattern, and the image feature amount and design data image feature amount calculation of the inspection target pattern image calculated by the inspection target pattern image feature amount calculating means Using the feature amount of the design data image calculated by the means and the teaching image data created by the teaching image data creating means to calculate and store the identification boundary for identifying the false alarm and the defect in the image of the inspection target pattern And the inspection target pattern image feature amount calculating means from the image obtained by imaging the inspection area of the inspection target pattern with the imaging means. From the feature amount of the image of the inspection target pattern and the feature amount of the design data image calculated by the design data image feature amount calculation means from the image created by the design data image creation means from the design data corresponding to the inspection area of the inspection target pattern And a defect determination unit that determines the defect of the inspection target pattern using the identification boundary stored in the identification boundary calculation unit.

  In order to achieve the above object, in the present invention, in a method for detecting defects in an inspection target pattern by comparing an image obtained by imaging the inspection target pattern with a design data image generated from the design data, The sample pattern of the pattern to be inspected is imaged to acquire an image of this sample pattern, the feature amount of the image of this sample pattern is calculated from the image of the sample pattern acquired by acquiring this image, and the sample from the design data of the sample pattern A design data image corresponding to the pattern image is created, a feature amount corresponding to the feature amount of the sample pattern image calculated for the created design data image is calculated, and an image of the sample pattern obtained by imaging is created. Create and calculate teaching image data for teaching defects using design data images For identifying a defect image from a sample pattern image using the feature amount of the design data image corresponding to the calculated feature amount of the sample pattern image and the calculated feature value of the sample pattern image and the created teaching image data The identification boundary is calculated, the image feature amount of the inspection area of the inspection target pattern is calculated from the image obtained by imaging the inspection area of the inspection target pattern, and the design data image is obtained from the design data corresponding to the inspection area of the inspection target pattern. The feature amount of the created design data image is calculated and the inspection region of the inspection target pattern is calculated based on the calculated image feature amount of the inspection region of the inspection target pattern, the feature amount of the design data image, and the calculated identification boundary. Detecting defects inside.

  Furthermore, in order to achieve the above-described object, in the present invention, in a method for detecting defects in an inspection target pattern by comparing an image obtained by imaging the inspection target pattern with a design data image generated from the design data, Capturing feature quantities and inspection target patterns corresponding to images obtained by capturing inspection target patterns extracted from design data images generated from design data and feature quantities extracted from inspection target patterns Using the information of the teaching data created using the image obtained in this way and the design data image, the identification boundary for identifying the false alarm and the defect is calculated, the calculated identification boundary is stored, and the inspection target pattern The image feature quantity of the inspection area of the inspection target pattern is calculated from the image obtained by imaging the inspection area, and the design data is calculated from the design data corresponding to the inspection area of the inspection target pattern. And the feature amount of the created design data image is calculated. Based on the image feature amount of the inspection area of the calculated pattern to be inspected, the feature amount of the design data image, and the stored identification boundary A defect in the inspection area of the inspection target pattern is detected.

  According to the present invention, since a defect part and a normal part are taught in advance and an identification boundary surface for discriminating between the defect and the normal is obtained based on the teaching, a systematic defect can be detected without causing many false reports.

It is a flowchart which shows the flow of the defect determination system based on the Example of this invention. 1 is a block diagram showing a schematic configuration of an electron beam pattern inspection apparatus according to an embodiment of the present invention. It is a block diagram which shows the structure of the image process part of the electron beam type pattern inspection apparatus which concerns on the Example of this invention. It is a design data image. It is a real pattern image acquired by scanning electron microscope (SEM) imaging. It is a figure which shows the image of the operator comprised by 5x5 pixel. It is a figure which shows the image of the operator which comprised the area | region of the same size as the operator of FIG. 3C as 1 pixel. It is a figure which shows the image of the operator comprised by 3x3 pixel. It is the image of the real pattern which displayed the operator comprised by 3x3 pixel superimposed on the real pattern image. It is a curved surface diagram of the identification boundary showing the concept of the identification boundary divided into two spaces of normal and defective on the feature space. It is the top view of the pattern which showed notionally the influence of the charge of the image imaged with the electron beam type pattern inspection apparatus. It is the front view of GUI which showed the example of preparation of the teaching data by GUI from a real pattern image and design data. It is the front view of GUI which showed the example of GUI which displays the result of defect determination. It is a graph explaining LDA and SVM which are the methods of calculating | requiring an identification boundary. It is a graph explaining LDA and SVM which are the methods of calculating | requiring an identification boundary. It is a design data image. It is the image of the real pattern obtained by imaging the location corresponding to the design data image of FIG. 9A with SEM. It is the figure which arranged the design data image, the image of the real pattern, and the teaching image data side by side. It is the figure which arranged the design data image, the image of the real pattern, and the teaching image data side by side. It is the figure which arranged the design data image, the image of the real pattern, and the teaching image data side by side. It is a front view of GUI for teaching for each defect type. It is the graph which showed the distribution of the feature vector, and the identification boundary of one straight line. It is the graph which showed distribution of a feature vector, and an identification boundary of two straight lines. It is a design pattern image. It is the image of the real pattern obtained by imaging with SEM. It is the graph which showed distribution of a feature vector, and an identification boundary of two curves. It is the graph which showed the histogram of a certain feature vector in a normal part and a defective part. It is a front view of GUI for performing sensitivity adjustment at the time of identifying normality and a defect. Although the example which performs defect determination by post-processing from the pixel by which the defect determination was carried out was shown, it is a figure which shows the image. It is a flowchart which shows the flow of the process based on an if-then rule. It is a graph showing the state which showed the defect determination area | region in feature-value space with two straight lines. It is a flowchart which shows the flow of the process based on an if-then rule. It is a graph showing the state which showed the defect determination area | region in feature-value space with four straight lines. It is the graph which represented the defect determination area | region in feature-value space by the linear sum of ax + by. It is the graph which represented the defect determination area | region in feature-value space with a nonlinear function. It is an image of design data before alignment. It is the real image before the alignment corresponding to FIG. 15A. It is an image showing the example which aligns a real image and design data. It is the graph which computed the square sum of a difference, shifting the position at the time of aligning a real image and design data.

  EMBODIMENT OF THE INVENTION Below, the form for implementing this invention is demonstrated using figures.

(1) Overall configuration
FIG. 2A shows the overall configuration of the pattern inspection apparatus according to the present invention. In the present embodiment, an image of the inspection target wafer 212 is acquired by the scanning electron microscope (SEM) 200. The electron optical system 201 of the scanning electron microscope 200 includes an electron beam source 203 that generates an electron beam 221, a condenser lens 204 that converges the electron beam 221, a deflector 206 that deflects the electron beam 221 in the XY directions, and an electron beam 221. Is provided with an objective lens 207 that converges on the target substrate. The sample chamber 202 that holds the inspection target wafer 212 in a vacuum includes an XY stage 210 to which a retarding voltage 211 is applied, and a cassette 213 that sends the inspection target wafer 212 from the outside to the sample chamber.

  The secondary electrons 214 generated from the sample are accelerated to the retarding voltage 211 and pulled up to the detector 215. The detection signal is sent to the A / D converter 216 by the detector 215, converted into a digital image, and stored in the storage device 220. The image processing unit 219 sequentially reads out images from the storage device 220 and performs defect detection by comparison with design data. Setting of inspection conditions and confirmation of inspection results are performed using a GUI on the display means 218.

  FIG. 2B shows a detailed configuration of the image processing unit 219. The image processing unit 219 aligns the design data image created by the design data image creation unit 2191 and the design data image creation unit 2191 that creates an image from the design data and the pattern image of the inspection target substrate obtained by the SEM 200. An alignment processing unit 2192 for performing, an inspection target pattern image feature amount calculating unit 2193 for calculating a feature amount of a pattern image of an inspection target substrate obtained by imaging with the SEM 200, and a design data image generated by the design data image generating unit 2191 A design data image feature value calculation unit 2194 that calculates feature values, a teaching image data storage unit 2195 that stores teaching image data created from the pattern image of the inspection target substrate, and an identification boundary for identifying defect images and non-defective images The identification boundary calculation unit 2196 for calculating the identification boundary and the identification boundary calculated by the identification boundary calculation unit 2196 are used. A defect determination unit 2197 for determining a defect from the inspection target pattern image feature amount calculated by the inspection target pattern image feature amount calculation unit 2193 and the design data image feature amount calculated by the design data image feature amount calculation unit 2194. Yes.

(2) Overall flow of defect detection
FIG. 1 is an overall flow of defect detection performed by the image processing apparatus 219 of FIG. 2A. In the present embodiment, prior to the inspection, as a pre-inspection preparation process S100, a small number of real images 11 obtained by imaging several sampling regions of the inspection target wafer 212 with the scanning electron microscope 200 described in FIG. A design data image 12 (an image created based on the design data) corresponding to the actual image 11 is aligned with the image 11 in units of pixels (S101), and features representing the features of the patterns included in the respective sampling regions (S102, S103), and from this and the teaching data 13 in which the defect coordinates are designated, an identification boundary that is a rule for discriminating between a normal pattern and a defect pattern is calculated (S104).

  In the processing S110 at the time of inspection, the actual image 14 of the inspection target area of the inspection target wafer 212 is aligned with the design data 15 corresponding to the actual image 14 (S111), and the design data that has been aligned with the actual image 14 is registered. 15 is subjected to image processing in the same manner as the feature amount calculating step in S102 and S103 of the pre-inspection preparation process S100, and the feature amount is calculated (S112 and S113). In contrast, the process proceeds to S104 of the pre-inspection preparation process S100. Defect determination (S114) is performed by applying the identification boundary calculated in the above, and the determination result is displayed on the screen (S115).

  Hereinafter, each processing content (feature amount calculation S102, S103, S112, S113, identification boundary calculation S104, defect determination S114, determination result display S115) will be described in detail.

(3) Positioning
In the present embodiment, as described above, feature amounts representing the features of each pattern are calculated from the actual image and design data, and the defect determination is performed by applying a rule for identifying normality and defect to these. Do.

  First, alignment between the actual image and the design data will be described with reference to FIGS. 15A to 15D. Since the actual image after the image pickup and the design data are in a state in which the normal position is shifted, it is necessary to align the position with the actual image by correcting the position information of the design data. In the design data image 1501 in FIG. 15A and the corresponding real image 1502 in FIG. 15B, the coordinates 1503 and the coordinates 1504 correspond to the positions on the image data. It is. Therefore, as shown in FIG. 15C, the origin of the design data image (such as the coordinates of the upper left corner of the image) is shifted along the path 1507 with respect to the actual image 1506, and the square sum of the difference between the actual image and the design data image is shifted. Calculate at the selected position. FIG. 15D is a graph of a waveform 1508 when the axis 1510 is the position of the path 1507 and the axis 1509 is the sum of the absolute values of the differences. In this waveform, correction is performed with the coordinates correctly aligned based on the position of the design data when the minimum value 1511 is taken.

(4) Calculation of feature quantity
Next, a feature amount calculation method will be described. A feature amount (corresponding to S102 and S112 in FIG. 1) calculated from an actual image (hereinafter referred to as an actual pattern image) acquired by imaging the inspection target wafer 212 with the scanning electron microscope 200 will be described.

  The feature amount of the actual pattern image is calculated for each pixel of the actual pattern image. Consider calculating the feature amount of the pixel of interest 304 in the actual pattern image 302 of FIG. 3B. The feature amount of the target pixel 304 is calculated by combining the luminance values in the peripheral pixels of the target pixel 304. In order to finally obtain a good identification boundary, it is desirable that the feature quantity represents the pattern shape information around the target pixel as much as possible, and that the number of dimensions of the feature quantity is small.

  The simplest feature amount is to make each luminance value in the peripheral pixels of the pixel of interest 306 as a feature amount as it is like the operator 305 shown in FIG. 3C. This has complete information of the actual pattern shape by the size of the operator (5 pixels × 5 pixels in the example of FIG. 3C). However, as the operator size is increased, the dimension of the feature amount explosively increases. And the shape information is too verbose. Therefore, it is better to calculate various statistics using the operator size 308 as an area of the same size as the operator 307 that is sufficiently larger than the target pixel 309 such as the operator 307 shown in FIG.

As a statistical feature amount obtained from the actual pattern image 302 of FIG. 3B,
(A) Average brightness value within the operator
(B) Moment feature
(C) HOG (Histograms of Oriented) Gradients) feature quantity (see Non-Patent Document 1)
(D) Higher order local autocorrelation features (see Non-Patent Document 2)
Etc. are effective.

  Further, like the operator 310 shown in FIG. 3E, a region of the cell 312 which is a local region in the vicinity of the pixel of interest 311 is taken, and a statistical feature amount is taken for each cell. A wide range of shape information can be calculated as a feature amount. FIG. 3F shows an example in which the operator 310 shown in FIG. 3E is applied as the operator 315 to the periphery of the target pixel 304 shown in the actual pattern image 302 of FIG. 3B. In this example, by calculating the HOG feature amount of the cell, the edge component in the vertical direction is extracted in the cells 316 and 317, and the edge component in the oblique direction is extracted in the cells 318 and 319, and the target pixel 319 (the target pixel in FIG. 3B) is extracted. The pixel 304, which corresponds to the pixel of interest 311 in FIG. 3E) is expressed as a line corner.

  FIG. 5 shows the influence of charging on the SEM image. Generally, when the SEM image 500 is obtained, it is obtained by scanning an electron beam in the direction 501. Therefore, a shadow 503 due to the influence of charging is generated on the scanning direction side of the region where the edge 502 exists. Such a feature can also be expressed by calculating an average luminance value by the operator 310.

  Next, feature amounts calculated from the design data image 301 (corresponding to the feature amounts calculated in S103 and S113 in FIG. 1) will be described.

  Also from the design data image 301 shown in FIG. 3A, the feature amount of the target pixel 303 corresponding to the target pixel 304 of the actual pattern image 302 is calculated. As the feature amount from the design data image 301, a contour image (design data image) as shown in FIG. 3A may be created from the design data, and the same feature amount as the actual pattern image 302 in FIG. 3B may be obtained.

In addition, the following design data-specific features
(I) Vector component of wiring structure
(Ii) Whether it is on wiring
(Iii) Relationship between line end points and connection points
It is also effective to calculate the above.

  (I) is the degree to which the wiring direction around the target pixel of the design data extends. (Ii) represents a feature amount whether or not the target pixel is on the pattern wiring. (Iii) represents the feature amount of the distribution of the end point of the edge and the location of the connection point when the wiring structure is a graph structure and how it is distributed around the pixel of interest.

(5) Calculation of identification boundary
A method for calculating an identification boundary (calculated in S104 in FIG. 1), which is a rule for identifying normality and defects, will be described. The identification boundary is a hypersurface (a two-dimensional curve) 404 that is divided into a normal space 402 and a defect space 403 in a feature space 401 (see FIG. 4) having the feature amounts obtained in S102 and S103 as axes. That is. The rule for discriminating the feature vector 405 of the normal part and the feature vector 406 of the defective part is the simplest if-then rule such as the example 1401 in FIG. And y represent feature quantities). When determining this rule, it may be obtained analytically or the user may input the threshold values A1 and B1.

  FIG. 14B shows a defect determination area in the feature amount space 1420 as a hatched portion. A black dot 1421 is a feature vector obtained from a normal pixel, and a black penalty point 1422 is a feature vector obtained from a defective pixel. The determination area is defined by a rectangular area having sides perpendicular to the feature amount axis. As shown in FIG. 14D, the accuracy of the identification boundary 1404 can be increased by increasing the if-then rule as indicated by 1403 in FIG. 14C. However, in this case, the complicated boundary is represented by a combination of rectangles, so the number of rules increases and adjustment by the user becomes difficult. When the true identification boundary has a correlation between x and y, as shown in FIG. 14E, the identification boundary 1405 expressed by a linear sum of ax + by or a non-linear function as shown in FIG. 14F A better boundary can be determined for the represented identification boundary 1406.

  It is very difficult to obtain these boundaries by adjusting the parameters by the user. However, by learning from the teaching data as feature vectors that are already normal or defective, the feature vector 405 of the normal part and the defect part are detected. It is possible to obtain a boundary that best separates the feature vector 406. If the user's intention can be reflected in the teaching data, the identification boundary can be drawn along the intention.

First, a method of creating teaching data (13 in FIG. 1) will be described.
FIG. 6 shows an example of a GUI 600 for creating teaching data. The image display unit 601 displays a contour line 602 (solid line) extracted from the already aligned actual pattern and a design data contour line 603 (dotted line). When there are regions having different shapes in the two contour lines 602 and 603, regions surrounded by two contour lines such as 604, 605, 606, 607, and 608 are generated. The user selects the region selection tool 610, and selects a region surrounded by the contour lines 602 and 603 by designating one point in the region desired to be a defect from the image displayed on the image display unit 601 with the pointing device 616. be able to. The portion where the contour line 602 of the actual pattern disappears can be corrected by selecting the pencil tool 611 so that the region surrounded by the contour lines 602 and 603 is closed. In addition, an erased tool 612 can be used to erase an erroneously extracted contour line or a portion that has been corrected with a pencil tool.

  613, 614, and 615 are modes for designating a defective portion when white 613 is selected by a color selection button, a mode for designating a normal portion when black 614 is selected, and a portion where the user does not know whether a defect is normal or not when gray 615 is selected. It becomes the mode to choose. All points that are not clicked in creating the teaching image are regarded as normal parts. In the case of this image, it is determined that 604, 606, and 608 are defective areas, and a teaching image 609 with a white area as a defective portion and a black area as a normal portion can be created and confirmed. After confirmation, when the “LOAD” button 621 is clicked with the pointer 616, the teaching image 609 is loaded into the image processing unit 219. On the other hand, when “SAVE” button 622 is clicked with pointer 616, teaching image 609 is temporarily stored in storage device 220.

  FIG. 9A is a diagram of a pattern 901 created based on design data, and FIG. 9B is an example of an image of a normal actual pattern 902 with respect to design data. If an actual pattern image 903 as shown in FIG. 9C is actually obtained, a defect image 907 in which the image areas 905 and 906 shown in FIG. As a set of teaching data to be used, an image set 904 including a pattern 901 created based on design data as shown in FIG. 9C, an image of a normal actual pattern 902 for the design data, and a defect image 907 can be created. . Here, the defect area images 905 and 906 included in the defect image 907 are converted into the convex defect area image 905 and the design data in which the line of the actual pattern is expanded with respect to the pattern 901 created based on the design data. The type can be classified into the image 906 of the concave defect area in which the line of the actual pattern is degenerated with respect to the pattern 901 created based on the pattern 901.

  Accordingly, as shown in FIG. 9D, a teaching image 909 including only the convex defect region image 905 is combined with a pattern 901 created based on the design data and an image of a normal actual pattern 902 corresponding to the design data into a teaching data set. 9E, and a pattern 901 created based on the design data, and an image of a normal actual pattern 902 corresponding to the design data, as shown in FIG. 9E. A concave defect teaching data set 910 that is combined into a teaching data set is created. As a result, as shown in FIG. 9F, defect detection and defect classification can be performed simultaneously by obtaining the identification boundary 914 from the convex defect teaching data set 908 and the concave defect teaching data set 910, respectively.

  FIG. 9F is a diagram illustrating an example in which the image sets 904, 908, and 910 described with reference to FIGS. 9B, 9 C, and D are arranged on one screen 912 and displayed simultaneously. In the example of the screen 912 shown in FIG. 9F, the convex defect teaching data set 908 is selected by the check box 916 on the screen, and the identification boundary 914 of the convex defect teaching data set 908 is displayed on the graph 913 in the feature space. The displayed state is shown. As will be described later, this has the advantage of improving the determination accuracy and the advantage that the user can adjust the sensitivity for each defect type. In systematic defect detection as intended this time, the defect types may be divided into convex defects, concave defects, bridging, necking, line width expansion, line width degeneration, and the like.

A screen 912 in FIG. 9F is a GUI that can be used when obtaining an identification plane for each defect type.
A data set divided for each defect type, such as image sets 904, 908, and 910, can be selected in the portion 915 shown in the screen 912 of FIG. 9F, and the data set selected in the portion of the feature space graph 913 The distribution on the feature space of the feature space (image set 908 in the example of FIG. 9F) and a histogram (not shown) for each feature amount are displayed and can be confirmed by the user. By clicking the “learn” button 916 displayed on the screen 912 with the mouse, an identification boundary for the selected data set can be obtained and the identification boundary 914 can be displayed on the graph 913 of the feature space.

As a method for calculating the identification boundary 314, a general pattern recognition method, a linear discrimination method (hereinafter referred to as LDA), or a support vector machine (hereinafter referred to as SVM) can be applied (see Non-Patent Document 3 and Non-Patent Document 4). )
The principle of LDA will be described with reference to FIGS. 8A and 8B. For easy understanding, the feature quantity 805 of the design data is treated as one dimension (vertical axis) and the feature quantity 806 of the actual pattern data is treated as one dimension (horizontal axis). In FIG. 8A and FIG. 8B, a penalty point such as 803 is a defect, and a black dot such as 804 is a normal feature vector. A straight line 801 is an identification boundary obtained by linear discriminant analysis (hereinafter referred to as LDA).

  In FIG. 8A, 801 is represented by a straight line, but since the actual feature vector takes three or more dimensions, the identification boundary becomes a hyperplane. Dividing the data with the straight line 801 as the boundary projects all feature vectors onto the axis 807 by calculating the linear sum of the feature values, and with a threshold value 808 (a range between a pair of dotted lines parallel to the straight line 801). It is the same as determining.

In LDA,
Variance of projected normal vector feature vector set: σ _T ²
Defect feature vector:
Variance of all feature vectors:
Represented when
Degree of separation: J = (σ _T ² + σ _N ² ) / σ _A ² (Equation 1)
Find the projection axis that maximizes.

  As features of the LDA, an identification surface and an identification boundary expressed by a linear sum of the feature values are obtained from the statistics of the feature vector set. Therefore, even if the number of feature vectors is large, the learning time is short, and the calculation time for normality / defect determination for one feature vector is also short. In addition, there is an advantage that it is easy to understand which feature quantity is often used for the user.

  A curve 802 in FIG. 8B is an identification boundary obtained by a non-linear support vector machine (hereinafter, non-linear SVM). In the non-linear SVM, a feature vector is once converted into a point on the superspace, and data is divided by obtaining a hyperplane in the superspace. When this hyperplane is expressed on the two dimensions of FIG. 8B, a curve like 802 is obtained. As the identification hyperplane, a hyperplane that passes through the largest margin (gap) existing between the feature vector of the normal part and the defective part in the superspace is obtained. In two dimensions, this margin is represented by 813 between curves 809 and 810, and is determined by a feature vector of white circle points 811 and white penalty points 812 in the figure called support vectors. Similar to LDA, nonlinear SVM essentially projects a vector on one axis and determines a normal defect based on a threshold value.

  As a feature of SVM, an identification boundary on the hypersurface is obtained in the dimension of the original feature amount, so that an identification boundary close to the identification hypersurface 404 shown in FIG. Has judgment accuracy.

  When the feature vector is calculated from the teaching data set 904 of FIG. 9C as described above, the feature vector 1011 of the defective part may be distributed on a diagonal line with the feature vector 1012 of the normal part interposed therebetween as shown in a graph 1001 of FIG. 10A. is there. In the graph of FIG. 10A, the vertical axis represents the design data feature amount, and the horizontal axis represents the actual pattern image feature amount. As described with reference to FIG. 8B, the nonlinear SVM can take a curvilinear identification boundary, and can deal with such a distribution of feature vectors. Since the identification boundary obtained by LDA can be represented only by a straight line, the obtained identification boundary is as shown by the straight line 1002 in FIG. 10A, and there is a limit in obtaining an effective discrimination boundary. However, by dividing the learning data according to the defect type, as shown in FIG. 10B, two identification boundaries of a straight line 1003 and a straight line 1004 can be drawn, and it is possible to make a correct determination. In the graph of FIG. 10B, the vertical axis represents the design data feature amount, and the horizontal axis represents the actual pattern image feature amount.

(6) Processing contents during inspection Using the identification boundary calculated in the identification boundary calculation step S104 described in FIG. 1, the defect detection of the inspection target wafer 212 is performed.

  For the actual pattern image 14 obtained by imaging the inspection target wafer 212 with the SEM 200 described in FIG. 2 and the design data image 15, the same feature amount as that used in the identification boundary calculation step of S 104 is calculated. (S112, S113) The normality / defect determination is performed on the feature boundary obtained in the discrimination boundary calculation step S104 (S114). The feature amount of the actual pattern image 14 in S112 must be calculated after the actual pattern image 14 is captured by the SEM, but the feature amount of the design data image 15 is calculated before the inspection, and the storage device If it is held in advance, the calculation time during inspection will be reduced.

  FIG. 11A to FIG. 11C show an example of actually detecting the defect of the inspection target wafer from the learned identification boundary. FIG. 11A shows a design data image 1101 to be inspected, and FIG. 11B shows an actual pattern image 1102. A feature vector obtained from the corresponding pixel 1103 on the design data image 1101 in FIG. 11A and a corresponding pattern 1103 on the actual pattern image 1102 in FIG. 11B is represented by a point 1108 on the graph shown in FIG. This is represented by the upper point 1109. In the pre-inspection preparatory stage S100, an identification boundary 1107 by SVM is obtained from the feature quantity vector 1106 of the normal part and the feature quantity vector 1105 of the defective part. Based on the identification boundary 1107, the feature vector 1108 is determined as a normal portion, and the feature vector 1109 is determined as a defective portion.

FIG. 13 is a diagram for explaining post-processing after the pixel-by-pixel defect determination. An image 1301 is obtained in which the pixels determined to be normal in the defect determination are black and the pixels determined to be defective are white as shown in FIG. At this time, the defective portion is detected by connecting defective pixels as in a region 1306, but 1305 connected by one pixel or several pixels can be obtained simultaneously by erroneous determination. Such a minimal region is removed as noise. Specifically, as shown in (b), the defective pixel is degenerated by morphological operation or the like (image 1302), and as shown in (c), an image like the image 1303 is obtained by expanding. .
In addition, as shown in (d), there is a method in which the major axis 1307 of the defective pixel region connected to the image 1301 is obtained, and if it is equal to or larger than a certain threshold value, it is determined as a defective region.

  The finally determined defect area is confirmed by the user in the determination result display in S114 of FIG.

  FIG. 7 shows an example of determination result display by the GUI 700. A wafer map is displayed in the display area 701, and the position of the target die and the approximate defect distribution position are displayed. In the display area 702, a die map is displayed, and the position 703 of the wiring pattern of interest is displayed, and a large defect or a region with a high defect frequency is displayed as a point 704. A display area 705 is a portion for displaying a histogram of occurrence frequency for each defect type. A display area 706 is a portion that displays coordinate information of a wiring pattern of interest, display magnification, and the like. A display area 707 is a part for displaying a design data image 7071 and an actual pattern image 7072 of the wiring pattern area of interest. On the actual pattern image 7072, defect determination areas 7073 to 7076 and the like are surrounded by lines and displayed. Reference numeral 708 denotes an operation panel which can switch the position and magnification of a wiring pattern to be viewed, defect display, and the like.

(7) Adjustment of defect detection sensitivity
12A and 12B, the sensitivity determination for defect determination will be described. Determining a defect at an identification boundary obtained by LDA or SVM is to project a feature vector onto one axis such as 1201 as shown in FIG. The operation is divided. The histogram consists of two peaks, a defect peak 1203 and a normal peak 1204. By moving the threshold value 1202 in the direction of the arrow 1205, the sensitivity to defects can be reduced, and by moving in the direction of the arrow 1206, the sensitivity to false information can be reduced and the false alarm detection rate can be reduced.

  An example of sensitivity adjustment for each defect type using this is shown by a GUI 1207 in FIG. 12B. Based on a plurality of identification boundaries obtained by learning for each defect type, display is made for each defect type as in the feature vector histograms 1208-1 to 1308-1. The user adjusts the threshold value for each defect type by changing the slider 1211 using a pointing device or the like. The determination result after adjustment is reflected on the actual pattern image 1212 in real time, and the defect areas 1213 and 1214 are displayed or not displayed. The user can adjust to an appropriate threshold while comparing with the design data image 1211.

  As mentioned above, although the invention made by the present inventor has been specifically described based on the embodiments, it is needless to say that the present invention is not limited to the above embodiments and can be variously modified without departing from the gist thereof. Yes.

  DESCRIPTION OF SYMBOLS 200 ... Scanning electron microscope 212 ... Inspection object wafer 215 ... Detector 216 ... A / D converter 218 ... Display means 219 ... Image processing part 220 ... Memory | storage device.

Claims (15)

An apparatus for detecting defects in an inspection target pattern by comparing an image obtained by imaging an inspection target pattern with a design data image generated from design data,
Imaging means for capturing an inspection target pattern and obtaining an image of the inspection target pattern;
Inspection target pattern image feature quantity calculating means for calculating a feature quantity of the image of the inspection target pattern from an image obtained by imaging the inspection target pattern by the imaging means;
Design data image creation means for creating a design data image from design data;
Design data image feature quantity calculating means for calculating the feature quantity of the design data image created from the design data of the inspection target pattern by the design data image creating means;
Display means comprising a display screen for displaying the image of the inspection target pattern acquired by imaging the inspection target pattern by the imaging means and the design data image of the inspection target pattern generated by the design data image creating means;
Teach image data creation means for creating and storing teaching image data of the defect of the inspection target pattern indicated on the screen of the display means on which the image of the inspection target pattern and the design data image of the inspection target pattern are displayed When,
The inspection target pattern image feature amount calculated by the inspection target pattern image feature amount calculation unit, the design data image feature amount calculation unit calculated by the design data image feature amount calculation unit, and the teaching image data generation unit An identification boundary calculating means for calculating and storing an identification boundary for identifying a false report and a defect in the image of the inspection target pattern using the teaching image data;
The feature amount of the image of the inspection target pattern calculated by the inspection target pattern image feature amount calculating means from the image obtained by imaging the inspection area of the inspection target pattern by the imaging means, and the inspection area of the inspection target pattern Using the identification boundary stored in the identification boundary calculation means from the feature quantity of the design data image calculated by the design data image feature quantity calculation means from the image created by the design data image creation means from the corresponding design data A pattern inspection apparatus comprising: defect determination means for determining a defect of an inspection target pattern.   The pattern inspection apparatus according to claim 1, wherein the imaging unit is a scanning electron microscope.   2. The pattern according to claim 1, wherein the teaching image data creation means creates teaching image data on a screen on which the image of the inspection target pattern and the design data image of the inspection target pattern are displayed in an overlapping manner. Inspection device.   The display means further displays on the display screen information relating to the identification boundary that identifies the false alarm and the defect calculated by the identification boundary calculating means, and the identification boundary is displayed on the display screen from the information relating to the displayed identification boundary. The pattern inspection apparatus according to claim 1, further comprising an identification boundary adjustment unit capable of adjustment.   The identification boundary calculation means calculates an identification boundary for identifying a feature vector of a normal part and a feature vector of a defective part in a feature quantity space represented by a design data image feature quantity and an actual pattern image feature quantity. The pattern inspection apparatus according to claim 1.   The pattern inspection apparatus according to claim 1, wherein the identification boundary calculation unit calculates an identification boundary corresponding to a defect type in a feature amount space represented by a design data feature amount and an actual pattern image feature amount. A method for detecting defects in an inspection target pattern by comparing an image obtained by imaging an inspection target pattern with a design data image generated from design data,
Capture the sample pattern of the pattern to be inspected to obtain an image of the sample pattern,
Calculating the feature amount of the image of the sample pattern from the image of the sample pattern obtained by capturing the image;
Create a design data image corresponding to the sample pattern image from the sample pattern design data,
Calculating a feature amount corresponding to the feature amount of the image of the calculated sample pattern for the created design data image;
Create teaching image data for teaching defects using the image of the sample pattern acquired by imaging and the created design data image,
Using the calculated feature amount of the sample pattern image, the feature amount of the design data image corresponding to the calculated feature amount of the sample pattern image, and the created teaching image data, a defect image is obtained from the sample pattern image. Calculate the identification boundary for identification,
Calculating an image feature amount of the inspection area of the inspection target pattern from an image obtained by imaging the inspection area of the inspection target pattern;
Create a design data image from design data corresponding to the inspection area of the inspection target pattern and calculate the feature amount of the created design data image,
A defect in the inspection region of the inspection target pattern is detected based on the calculated image feature amount of the inspection region of the inspection target pattern, the feature amount of the design data image, and the calculated identification boundary. Pattern inspection method.   Creating teaching image data for teaching a defect from the image of the sample pattern acquired by imaging and the created design data image, the image of the inspection target pattern and the design data image of the inspection target pattern The pattern inspection method according to claim 7, wherein the pattern inspection method is performed on a screen displayed in a superimposed manner.   8. The pattern inspection method according to claim 7, wherein information on the calculated identification boundary is displayed on a screen, and the identification boundary is adjusted on the screen from information on the identification boundary displayed on the screen. A method for detecting defects in an inspection target pattern by comparing an image obtained by imaging an inspection target pattern with a design data image generated from design data,
The feature quantity extracted from the image obtained by imaging the inspection target pattern and the feature quantity at the location corresponding to the image obtained by imaging the inspection target pattern extracted from the design data image generated from the design data, and the inspection target pattern Using the information of the teaching data created using the image obtained by imaging the image and the design data image, the identification boundary for identifying the false alarm and the defect is calculated,
Storing the calculated identification boundary;
Calculating an image feature amount of the inspection area of the inspection target pattern from an image obtained by imaging the inspection area of the inspection target pattern;
Create a design data image from design data corresponding to the inspection area of the inspection target pattern and calculate the feature amount of the created design data image,
Detecting a defect in the inspection region of the inspection target pattern based on the calculated image feature amount of the inspection region of the inspection target pattern, the feature amount of the design data image, and the stored identification boundary. Characteristic pattern inspection method.   11. The pattern inspection method according to claim 10, wherein the teaching image data is created on a screen on which the image of the inspection target pattern and the design data image of the inspection target pattern are displayed in an overlapping manner.   Information on the identification boundary calculated and stored is displayed on the screen, the identification boundary is adjusted on the screen from the information on the identification boundary displayed on the screen, and the adjusted identification boundary is newly stored. The pattern inspection method according to claim 10.   The pattern inspection method according to claim 7 or 10, wherein an image obtained by imaging the sample pattern is an SEM image.   The identification boundary is calculated by identifying an identification boundary that distinguishes between a normal part feature vector and a defective part feature vector in a feature amount space represented by a design data image feature amount and an actual pattern image feature amount. The pattern inspection method according to claim 7 or 10, wherein:   11. The method according to claim 7, wherein calculating the identification boundary includes calculating an identification boundary according to a defect type in a feature amount space represented by a design data image feature amount and an actual pattern image feature amount. The pattern inspection method described.

Images