JP4951591B2 - Matching method corresponding to disappearance of pattern and inspection apparatus using the same - Google Patents

Description

  The present invention relates to a general apparatus for performing inspection by performing template matching, such as a semiconductor inspection apparatus.

  Conventionally, a technique for searching a given specific shape (template) from a target image has been widely used as template matching.

  In pattern measurement on a semiconductor wafer using a scanning electron microscope, template matching is performed to obtain a measurement position. Rough positioning of the measurement position is performed by moving the stage on which the wafer is placed. However, the positioning accuracy of the stage causes a large shift on the image taken at a high magnification of the electron microscope. Template matching is performed in order to correct this deviation and perform measurement at an accurate position. Specifically, a unique pattern close to the measurement position is registered as a template, and the relative coordinates of the measurement position viewed from the template are stored. When obtaining a measurement position from a photographed image, template matching is performed on the photographed image to obtain a matching position, and the position moved by the stored relative coordinates is the measurement position.

  FIG. 1 illustrates template matching in a conventional scanning electron microscope. First, the measurement position of the object (a) is photographed, and a unique pattern included therein is registered as a template. The image (b) photographed at this time is referred to as a template selection image, and the unique pattern (c) selected from the template selection image is referred to as a template.

  Next, another object (a ′) ((a ′) is another part having the same pattern on the same wafer as (a), for example, the same part of a die repeatedly formed on the same wafer. Or a place with the same pattern on different wafers.) When a photograph is taken, a pattern that matches the template is searched from the photographed images. This captured image (b ′) is referred to as a search image. There is a deviation of the positioning error of the stage between the template selection image (b) and the search image (b ′). This deviation is corrected by template matching. As a result of template matching, a location having a high similarity to the template is a candidate for a matching position, and a location most suitable for the matching position is the final matching position among these candidates.

  However, this requires an actual wafer for matching, and temporarily occupies the scanning electron microscope system to perform imaging and registration of various conditions. Therefore, it has been devised to perform matching using design (CAD) data as disclosed in [Patent Document 1].

  FIG. 2 shows an apparatus configuration for realizing matching using CAD data. The template registration unit includes a template image drawing unit that creates a template image from CAD data and a normal template storage unit that stores the image. The matching execution unit is composed of an image search unit that searches while matching the search image using the image, and a candidate table that stores information on positions with a high matching score (matched better) in order of score. Output the position as a matching position.

  In the measurement of a pattern on a semiconductor wafer using a scanning electron microscope, it is necessary to perform template matching of a distorted pattern against CAD data. This is because it is becoming difficult to form a pattern as designed as semiconductor processing becomes finer. In such a case, even if there is some distortion between the template pattern and the target pattern in the search image, it is necessary to allow this distortion by template matching.

  The CAD data is configured as a set of closed figures by line segments, but the template image drawing unit creates a CAD image in which the outlines (edges) of these closed figures are drawn.

  A search image obtained by photographing a pattern on a semiconductor wafer with a scanning electron microscope is also composed of edges.

  For example, [Non-patent Document 1] discloses that a stable matching can be achieved by allowing distortion such as scaling by thickening (swelling and smoothing) an edge image in template matching of the edge image.

  [Patent Document 1] also discloses smoothing of edges of a search image taken with a template and an electron microscope in matching using CAD data.

  Also in the template image drawing unit of FIG. Further, it is assumed that the search image is also an image on which an edge is extracted and thickened in advance.

Japanese Patent Laid-Open No. 2007-5818 "High-speed template matching algorithm using contour point information" Section 2.2 of IEICE Transactions Vol.J74-D2 No.10 pp.1419-1427

  As described above, template matching that allows distortion to some extent can be performed by thickening the edges, but due to recent semiconductor miniaturization, a pattern that exists in CAD data but is not resolved on the wafer is generated. Has happened. That is, it may happen that a certain pattern exists in the template but not in the search image.

  FIG. 3 is an example of an image in which CAD data is drawn. There is no expansion treatment. There are continuous areas with very dense patterns. FIG. 4 is a search image obtained by photographing a pattern on an actual wafer corresponding to the CAD data of FIG. The search image does not resolve a continuous region in which the pattern existing in the template is very dense. FIG. 5 shows an image obtained by extracting edges from the search image and fattening them. The actual matching is performed using the image obtained by fattening the image of FIG. 3 as a template and the image of FIG. 5 as a search image. The presence or absence of a continuous region having a very dense pattern has an adverse effect. However, it actually fails as shown in FIG.

  In addition, even if successful, the matching score is very low due to the presence or absence of a very dense continuous region of the pattern, and false alarms (even though there are no matching parts, different pattern parts are used as matching positions. Output case).

  Even when such a fine pattern disappears, matching has not been performed correctly and stably.

  An object of the present invention is to provide a technique and an inspection apparatus capable of matching correctly and stably (distinguishable from false alarms) even when a fine pattern disappears.

  In order to solve the above-described problems, the present invention obtains a search image, and in an inspection apparatus that applies template matching to the search image using a template created from CAD data, It has a degenerate template that imitates the state in which the pattern has disappeared, performs matching with a plurality of templates, and merges the candidates obtained by each matching.

  Furthermore, the present invention is characterized in that the inspection apparatus has a critical expansion degree determination unit that determines the degree of degeneration and a search control unit that controls matching based on the result.

  Furthermore, the present invention is characterized in that the inspection apparatus has a template comparison unit that compares a plurality of templates and measures the degree of difference between them, and a search control unit that controls matching based on the result.

  Furthermore, the present invention is characterized in that the inspection apparatus refers to a history of candidates in past matching.

  Furthermore, the present invention is characterized in that the inspection apparatus has an interface for setting a degree of expansion by a user.

  Furthermore, the present invention is characterized in that the inspection apparatus has an interface for setting the degree of degeneration by the user.

  Furthermore, the present invention is characterized in that the inspection apparatus has an interface for allowing the user to set a history range and a difference between the score of the highest candidate and the second candidate in the history.

  In order to solve the above problems, the present invention obtains a search image, and in the inspection method for performing template matching on the search image using a template created from CAD data, in addition to a normal template It is characterized by obtaining a degenerate template that imitates a state in which a fine pattern has disappeared, matching with a plurality of templates, merging candidates obtained by each matching, and performing template matching .

  Furthermore, the present invention is characterized in that in the inspection method, the degree of degeneration is determined, and matching is controlled based on the determination result.

  Further, the present invention is characterized in that, in the inspection method, a plurality of templates are compared to measure their different degrees, and matching is controlled in response to the comparison result.

  Furthermore, the present invention is characterized in that in the inspection method, a history of candidates in past matching is referred to.

  Furthermore, the present invention is characterized in that, in the inspection method, the user sets the degree of expansion.

  Furthermore, the present invention is characterized in that, in the inspection method, the user sets the degree of degeneration.

  Furthermore, the present invention is characterized in that, in the inspection method, the user sets a history range and a difference between the score of the highest candidate and the second candidate in the history.

  According to the inspection apparatus or the inspection method of the present invention, in addition to a normal template, there is a degenerate template that imitates a state in which a fine pattern has disappeared, matching is performed with a plurality of templates, and candidates obtained by each matching Thus, even when a fine pattern disappears on an actual wafer, the matching position can be obtained correctly and stably.

  Also, it does determine the critical expansion determination of the degree of degeneration, resulting by controlling the matching receiving, even when not known whether advance how much of the fine pattern is lost is automatically This realizes matching with an optimal number of templates subjected to an optimal degree of degeneration.

  Further, it is possible to avoid unnecessary matching by comparing a plurality of templates, performing template comparison to measure the degree of difference between them, and controlling matching according to the result.

  Further, it is possible to avoid unnecessary matching by referring to the history of candidates in past matching.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

  FIG. 8 shows one embodiment of the image processing method according to the present invention.

  CDA data is input to the template image drawing unit of the template registration unit. The template image drawing unit draws the edge of the closed figure included in the CAD data, outputs the image to the template degeneration processing unit, and displays the image subjected to the processing for thickening the edge as the normal template storage unit And the normal template storage unit stores it.

  The template reduction processing unit performs a reduction process for erasing a fine pattern on the image received from the template image drawing unit and outputs it to the reduction template storage unit. The reduction template storage unit stores this.

  The image search unit uses the template stored in each template storage unit to perform edge extraction and edge thickening in advance on an image photographed by a camera or an electronic microscope, or an image that has been saved once Template matching is performed as a search image subjected to, and the result (position and position of matching candidates) is output to a candidate table in a normal template and a candidate table in a degenerate template, and each candidate table stores it. The candidate table merging unit merges the candidate information of each candidate table and outputs the matching position of the candidate with the maximum score.

  FIG. 9 shows the processing operation of the template registration unit. The image (a) is an example in which the edge of the closed figure included in the CAD data is drawn by the template image drawing unit, and this is output to the template reduction processing unit. (B) is obtained by subjecting (a) to the process of thickening the edge, and this is normally stored in the template storage unit. (C) Subsequent image creation is performed by the template reduction processing unit. (C) is obtained by expanding (a) at a certain degree of expansion. (D) is obtained by shrinking (c) with the same degree of expansion. (E) is an edge extracted from (c). (F) is obtained by expanding (e) with the above-described expansion degree. (G) represents a process of ANDing (d) and (f). This result is (h). That is, in (d) and (f), white pixels become white in (h). In (h), the dense pattern in (a) is degenerated. Finally, the degenerate template (h) is processed to thicken edges and stored in the degenerate template storage unit. The process of thickening the last edge is not a process for degenerating the pattern but a process for allowing distortion. Therefore, smoothing such as Gaussian may be performed, or expansion processing with different expansion degrees may be performed.

  Next, template matching performed by the matching execution unit will be described with reference to FIG. (1) represents a search image, and (2) represents a template (reference image). Template matching is processing for detecting the same part as a template in a search image. Specifically, for example, the normalized correlation is calculated by superimposing the search image while shifting the template, and the place where the highest correlation value is obtained is determined as the place having the same pattern as the template. It is done.

  A CD-SEM is an inspection device that measures a specific location on a semiconductor wafer using a scanning electron microscope. In order to specify a location on the wafer to be measured, a unique pattern is used in the vicinity of the measurement location. A portion having a length is registered as a template, and at the time of length measurement, a length measurement portion is specified based on coordinates obtained by template matching. In the case of FIG. 10, it has the same pattern as the template at the position a in (3). Further, although not the same pattern, the next highest correlation values are the four locations b, c, d, and e in (4). Further, the next highest correlation value is f and g in (5). There is almost no correlation at other locations, and the correlation value is close to zero.

  (6) is a normalized correlation map, which is an image having a correlation value for each coordinate as a pixel value. The value a ′ is taken at a, which is close to 1. b, c, d, e, f, and g also take values b ', c', d ', e', f ', and g', which are lower than a '. If the correlation value is a score, a becomes the highest candidate of the maximum score, and b, c, d, e, f, and g become candidates next to this. These are registered in the candidate table.

  In the case of FIG. 10, an example in which the search image is larger than the template is shown. However, since the matching only needs to know the relative position of the search image with respect to the template, the sizes may be reversed. Search (search) processing is the same as replacing the template and the search image.

  FIG. 11 shows an example of determining the matching position from the normalized correlation map. The horizontal axis represents xy coordinates, and the vertical axis represents correlation values. The position with the largest correlation value is the position that best matches the template. When the maximum correlation value v1 is smaller than the predetermined threshold th1, it is reasonable to consider that there is no part that matches the template. However, when the pattern disappears, the correlation value becomes low even at the correct position of matching, and it becomes impossible to distinguish from the case where there is no matching part. Further, when the difference v1−v2 between the second largest correlation value v2 and v1 is smaller than the predetermined threshold th2, the correlation value at the correct position may be reversed from the next candidate due to the influence of noise or the like. However, when the pattern is lost, the correlation value at the correct position is lowered, so that there are no differences from the correlation values of other candidates and matching often occurs. Therefore, when the pattern is lost, a template corresponding to the lost pattern is created, and the correlation value at the correct position is kept high, so that confusion with failure and misinformation can be avoided.

  Next, a method for creating a degenerate template will be described in detail.

  FIG. 12 shows an image obtained by rendering CAD data. (A) is an image in which CAD data is drawn. (B) to (d) are obtained by adding an expansion filter corresponding to the difference in each degree of expansion.

The expansion filter will be described with reference to FIG. The expansion filter has a square filter area (kernel), and the center pixel is the target pixel. In FIG. 13, a 5 × 5 square area is the kernel. If there is at least one white pixel in the kernel for the current pixel of interest, the pixel of interest is newly white. Here, the degree of expansion indicates the number of pixels from the target pixel to the outermost peripheral pixel of the kernel. The calculation formula is (kernel size-1) / 2
It becomes.

  In the case of FIG. 13, since the kernel size is 5, the degree of expansion is 2.

  In FIG. 12, (b) is applied with an expansion filter with an expansion degree of 1, (c) with an expansion degree of 2, and (d) with an expansion degree of 3.

  FIG. 14 is obtained by applying a contraction filter having the same expansion degree to each expansion image of FIG.

  The contraction filter will also be described with reference to FIG. In the case of the contraction filter, if there is at least one black pixel in the filter area with respect to the current target pixel, the target pixel is newly set to black. In the case of contraction, it should be expressed not as the degree of expansion but as the degree of contraction, but for the sake of simplicity, the term expansion degree is used here for both purposes.

  14 (a) to 14 (d) are obtained by applying a contraction filter to each of FIGS. 12 (a) to 12 (d) with the same degree of expansion as during expansion. Comparing FIG. 12 (d) and FIG. 14 (d), the pattern that stuck to the surrounding pattern at the time of expansion does not return to its original shape even when contracted, and constitutes a large area. You can see that In this case, the comb-shaped fine pattern was not degenerated up to an expansion of 2, but it was degenerated at an expansion of 3. The degenerated portion is a white area in FIG.

  FIG. 15 is obtained by performing edge extraction on each image of FIG. Here, a Sobel filter is applied as edge extraction. However, when the Sobel filter is applied, the pixel value does not become binary, but all pixels other than black are made binary as white. It can be seen that the edge image of (d) that has been shrunk by expansion 3 has very few white pixels as compared to (a) to (c) having a smaller expansion degree. In (d), the portion of the degenerated fine pattern is black.

  FIG. 16 is obtained by applying an expansion filter having an expansion degree of 0 to 3 to each of FIG. FIG. 16 shows degeneration mask information in which the degenerated portion is represented in black.

  When (a) to (d) in FIG. 14 are applied as masks to (a) to (d) in FIG. 16, that is, when processing (AND) is performed so that only white pixels are newly made white, expansion occurs. 12 (a) of the current image up to degree 2, but the portion of the fine comb mold that has been degenerated at the degree of expansion 3 does not return to the comb type, and the outline of the degenerated area as shown in FIG. 9 (h). Is left.

  17A to 17D are obtained by applying another form of edge extraction filter to FIGS. 12A to 12D. This filter will be described with reference to FIG. This filter is a 3 × 3 pixel square filter with the pixel of interest at the center of the kernel. The pixel of interest is newly white only when the pixel of interest is white and there is at least one black pixel other than the pixel of interest. Is black.

  Although the shape is different from that of the Sobel filter of FIG. 15, (a) to (d) of FIG. 19 in which an expansion filter having an expansion degree of 0 to 3 is applied to (a) to (d) of FIG. The degenerated mask information is shown in black.

  When (a) to (d) in FIG. 14 are applied as masks to (a) to (d) in FIG. 19, that is, processing (AND) is performed so that only white pixels are newly white. Similarly to the case of FIG. 16, the current image returns to (a) of FIG. 12 until the degree of expansion 2, but the portion of the fine comb that has been degenerated at the degree of expansion 3 does not return to the comb shape ( As shown in h), the outline of the degenerated area remains.

  In FIG. 15 and FIG. 17, when the expansion degree becomes large and the degeneration occurs, the number of white pixels rapidly decreases. FIG. 20 shows the transition of the number of white pixels in the edge extracted image according to the degree of expansion.

  In addition, when the contracted image after expansion in FIG. 14 is viewed, the number of white pixels increases rapidly when the expansion degree increases and degeneration occurs. FIG. 21 shows the transition of the number of white pixels in the contracted image after expansion according to the expansion degree.

  If the degree of expansion when degeneration occurs for the first time in a fine pattern region is called the critical degree of expansion, in this case, the degree of expansion 3 is the critical degree of expansion. When creating a degenerate template, a difference from a normal template does not appear unless expansion and contraction are performed at an expansion degree higher than the critical expansion degree. The degree of expansion when creating a template may be fixed in the system, but if an interface that can be given from the outside is prepared, versatility is enhanced. In addition, the degree of expansion can be automatically set such that when the change in the number of pixels in FIG. 20 and FIG.

  22 and 23 show examples in which a degenerate template is created using CAD data of an actual semiconductor pattern. (A) is an image in which CAD data is drawn and expanded, (b) is an image in which the CAD data is contracted with the same expansion, (c) is an image in which a Sobel filter is applied to (a), (d ) Is an image obtained by expanding (c) at the same expansion degree, (e) is an image obtained by ANDing (b) and (d), and (f) is an expansion of (e) at an expansion degree 1 to allow distortion. It is the image made to do. FIG. 22 differs from FIG. 23 in that FIG. 22 has a degree of expansion of 1 and FIG.

  Comparing the respective degenerate templates, the fine pattern of the expansion type 1 comb type is degenerated, but when it is set to the expansion degree 2, the scale is small, but degeneracy occurs in several places. I understand. It is conceivable that there are patterns with different densities in the same pattern even though the actual pattern is the same.

  The change in the number of white pixels shown in FIG. 21 does not actually have a single critical expansion, but generally has a plurality of critical expansions as shown in FIG.

  FIG. 25 shows another embodiment corresponding to the case of having a plurality of critical expansion degrees.

  The apparatus in FIG. 8 has only one degenerate template processing system, whereas the apparatus in FIG. 25 has two degenerate template processing systems. In this way, by having a plurality of processing systems for degenerate templates, it is possible to cope with a plurality of critical expansion degrees. The number of degenerate template processing systems to be used may be fixed, but if an interface that can be given from the outside is prepared, versatility is enhanced.

  FIG. 26 shows another embodiment in which the template registration unit has a critical expansion degree determination unit and the matching execution unit has a search control unit. The critical expansion determination unit determines how many degenerative template processing systems to use according to changes in the number of pixels shown in FIG. 24 and the like, and determines which expansion is optimal for each system. The use / non-use and the degree of expansion for each system are set. At the same time, the search control unit is notified of which system is being used, and matching is not performed in unnecessary systems. Also, unused candidate tables are not merged.

  Such optimization can be performed by using a graph such as FIG. 20 and clustering the expansion degree by the number of pixels. If you want to execute with the number of processing systems of the degenerate template of the maximum resource that can be used, or if you want to execute with a fixed number of systems, such as when you want to execute with the number of systems specified by the user, expand the degree of expansion using the k-average method, respectively The degree of expansion of the representative value of the cluster may be set for each system.

  Further, limited clustering can be performed by setting a distance threshold within the cluster. If a surplus system appears, the system can be disabled.

  FIG. 27 shows another embodiment in which the template registration unit has a template comparison unit. If the templates stored in the template storage unit are not significantly different from each other, it is expected that the result of matching with the templates will be the same. Therefore, the template comparison unit is a template stored in each template storage unit. If there is no difference, the system to be used is transmitted to the search control unit in order to reduce the number of templates to be used, and unnecessary matching is reduced.

  FIG. 28 is another embodiment in which the search control unit refers to the past history of candidates merged by the candidate table merging unit and reduces unnecessary matching processing. If the matching candidate of a certain system has always been at the highest level so far, the optimum expansion degree can be specified in the process. Whether it is considered to be optimal if it is continued, or how much it should be different from the second candidate, etc. may be fixed, but if you prepare an interface that can be given from the outside, it will be versatile Rise.

  So far, the description has been given of creating a reduced template based on an image in which the outline (edge) of CAD data is drawn. However, it is also possible to create a reduced template based on an image in which a closed figure included in CAD data is filled. Is possible. FIG. 29 and FIG. 30 show a method of creating a degenerate template based on such a filled image. In FIGS. 29 and 30, the arrangement of black and white is reversed. (A) is a painted image, (b) is an image with expansion at a certain degree of expansion, (c) is an image with (b) contracted with the same expansion, and (d) is FIG. Edges are extracted using the edge extraction filter shown. By changing the arrangement of white and black in this way, there are cases where a contour portion of a continuous region of a high-density pattern appears and does not appear. Since this contour always appears when the image is drawn based on the image of the CAD data edge, if it is used for a pattern in which the contour of the continuation area of the comb pattern does not appear as shown in FIG. May be more appropriate based on

  In addition, it is possible to process differently arranged black and white arrangements in different systems.

  Although the template matching apparatus has been mainly described so far, the entire inspection / measurement system is the same as that disclosed in Japanese Patent Application Laid-Open No. 2007-5818, and the entire system can be applied. Is possible.

  The present invention can be applied to a general apparatus that performs inspection by performing template matching, such as a semiconductor inspection apparatus.

Explanatory drawing of the template matching in the conventional scanning electron microscope. An example of an apparatus configuration that realizes matching using CAD data. An image example in which CAD data is drawn. A search image that captures a pattern on a real wafer. An image obtained by extracting edges from a search image and fattening them. Matching result diagram (success example). Matching result diagram (failure example). 1 shows an embodiment of an image processing method according to the present invention. The figure which showed the processing operation of the template registration part. Explanatory drawing of template matching. An example of determining a matching position from a normalized correlation map. An example of expansion of an image in which CAD data is drawn. The example which gave the expansion filter. The example which gave the contraction filter. An example of performing edge extraction on an image. The example which gave the expansion filter. An example of applying an edge extraction filter. Explanatory drawing of an edge extraction filter. The example which gave the expansion filter. The transition example of the number of white pixels of the edge extraction image by the expansion degree. The transition example of the white pixel number of the shrinkage | contraction image after the expansion | swelling by the expansion degree. An example in which a degenerate template is created using CAD data of an actual semiconductor pattern. An example in which a degenerate template is created using CAD data of an actual semiconductor pattern. Example with multiple critical expansions. Another embodiment. Another embodiment. Another embodiment having a template comparison unit in the template registration unit. An embodiment in which unnecessary matching processing is reduced. The example which produced the degenerate template based on the image of the closed figure contained in CAD data. The example which produced the degenerate template based on the image of the closed figure contained in CAD data.

Claims (10)

In an inspection apparatus that obtains a search image and applies template matching to the search image using a template created from CAD data.
Has a normal template and a degenerate template that has been subjected to expansion and contraction processing on the edge of the pattern,
An inspection apparatus, wherein matching is performed with the plurality of templates, the position and score of a matching candidate of each template are stored, and the position of the candidate with the maximum score is set as a matching position. The inspection apparatus according to claim 1,
Judge the critical expansion from the change in the number of white pixels of multiple templates with different expansions, and perform matching using the template created by setting the appropriate expansion for the appropriate number of templates Inspection apparatus characterized by that. The inspection apparatus according to claim 1,
An inspection apparatus that compares a plurality of templates and reduces the number of templates used when there is no difference. The inspection apparatus according to claim 1,
An inspection apparatus comprising an interface for a user to set an expansion degree when the edge of the pattern is expanded. The inspection apparatus according to claim 2, wherein
An inspection apparatus comprising an interface for a user to set a degeneration degree of the degenerate template. In an inspection method for obtaining a search image and applying template matching to the search image using a template created from CAD data,
Obtaining a normal template and a degenerate template in which the edges of the pattern are expanded and contracted ,
An inspection method, wherein matching is performed using the plurality of templates, the position and score of the matching candidate of each template are stored, and the position of the candidate with the maximum score is set as the matching position. In the inspection method of Claim 6,
Judge the critical expansion from the change in the number of white pixels of multiple templates with different expansions, and perform matching using the template created by setting the appropriate expansion for the appropriate number of templates Inspection method characterized by that. In the inspection method of Claim 6,
An inspection method characterized by comparing a plurality of templates and reducing the number of templates used when there is no difference. In the inspection method of Claim 6,
An inspection apparatus, wherein a user sets an expansion degree when expanding an edge of the pattern. The inspection method according to claim 7,
An inspection method, wherein a user sets a degeneration degree of the degenerate template.

Images