JP2013102234A - Detector and method, exposure device, and device manufacturing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To allow more stable detection of a position of a mark.SOLUTION: A detector that detects a position of a mark in an image by template matching includes a processing unit that performs template matching to an image by using a template. When the degree of correlation obtained by the template matching does not satisfy an allowable condition, the processing unit performs template matching by using a template different from the template described above.

Description

  The present invention relates to a detection apparatus and method for detecting the position of a mark.

  When a reticle pattern is projected and exposed on a wafer, the exposure is performed after aligning the wafer and the reticle. As a positioning technique therefor, there is a method of executing two types of alignment, pre-alignment and fine alignment, using alignment marks provided on the wafer. The role of pre-alignment is to detect the amount of feed deviation that occurs when the wafer is placed on the wafer chuck that is configured on the stage in the exposure apparatus from the wafer transfer device, and within the accuracy that fine alignment can be processed normally. This is to roughly align the position of. Also, the role of fine alignment is to accurately measure the position of the wafer on the stage and precisely align so that the alignment error with the reticle is within an allowable range. Prealignment accuracy is required to be about 3 microns. The precision of fine alignment varies depending on the requirements of wafer processing precision, but for example, 64 MDRAM requires 80 nanometers or less.

  As described above, the pre-alignment must detect a feed deviation that occurs when the transfer device feeds the wafer to the chuck. Therefore, a very wide range of detection is required, and generally a range of 500 μm □ is detected. A pattern matching process is often used as a method for detecting the X and Y coordinates from the alignment marks in such a wide range.

  There are two types of this type of pattern matching processing. One is a method in which the image is binarized and matched with a template that is held in advance, and the position having the highest correlation is set as a mark position. This is a method of performing the correlation calculation. For the latter, a normalized correlation method or the like is often used.

  In the pattern matching process as described above, it is difficult to detect an alignment mark in a low-contrast image, a noise image, or an image having defects generated during wafer processing. The applicant of the present application has proposed a mark detection method that makes it possible to perform stable detection on an image in which it is difficult to detect such an alignment mark. This method is characterized in that the edge of the mark is extracted optimally and at the same time, the direction characteristics of the edge are extracted together, and pattern matching focusing on the edges divided into the direction components is performed.

JP 2000-260699 A Japanese Patent Laid-Open No. 10-97983

  In the case of pre-alignment, a very wide range of detection is performed, but the alignment mark is small. This is because the alignment mark is a pattern that does not become a semiconductor element, and the alignment mark must be reduced in order to make the area of the semiconductor element as large as possible. For this reason, the alignment mark is often placed in a region not used as an element, for example, a scribe line, and generally the mark size is determined by the width of the scribe.

  With the recent increase in efficiency of semiconductor manufacturing and improvement in processing accuracy, the scribe width is narrowed year by year, and the scribe is narrowed to 100 μm or less. Therefore, the detection mark size that fits in such a scribe is also 60 μm or less. In addition, various wafer processing steps have been performed to manufacture high-density semiconductors.

  In the alignment, a mark imprinted on the lower layer is detected. In the manufacturing process of the semiconductor element, the upper layer is gradually formed. For this reason, a change in the shape of the alignment mark may occur. FIG. 9 shows an example of a change in the mark line width of the detection mark. In FIG. 9, a mark made according to the design value in the size of WO can be easily detected if the template is as designed. However, when various films are laminated on the step structure of the detection mark through the process, the mark line width becomes narrower or conversely thicker. FIG. 9 shows an example in which the mark line width WO is narrowed to WP by stacking films. In this case, detection with a template prepared in advance becomes difficult.

  Further, when the illumination method used for mark detection is a bright field, various changes occur in the observation state depending on the difference in reflectance between the mark and its surroundings. FIG. 8 shows an example of such a state. FIG. 8A shows a step structure of a detection mark formed by etching a material with high reflectivity when a material with low reflectivity is used as a base of the detection mark and a material with high reflectivity is laminated thereon. Is shown in (b). The change in the reflectance of the mark WIN part is shown in (c), and the change in the brightness of the WIN part is shown in (d). In the two-dimensional image of the mark, the mark appears black and the periphery appears white as shown in FIG.

  On the other hand, if the reflectance is changed to a material that is reversed from that shown in FIG. 8A, the image shown in FIG.

  Conventional mark detection methods based on template matching include template matching from a binarized image and a normalized correlation method. In these processes, changes in the brightness of the mark cannot be detected. Therefore, in these correlation processes, it is necessary to prepare a template that is adjusted to light and dark beforehand.

  As described above, with the advancement of technology for manufacturing high-density semiconductors, it is very difficult to detect alignment marks that exist in a wide range of detection areas and are deformed depending on the process.

  In order to make up for the defect that the mark cannot be detected due to the deformation of the alignment mark, a method of storing the feature of the mark portion from the image every time a detection error occurs and using the stored image as a template is common. Furthermore, as shown in Patent Document 2, there is only a method of rewriting template information until a mark can be detected manually. None of these methods can complete the alignment without human intervention. Another problem is that detection is not possible when the conditions of the laminated structure change. Particularly in recent years, semiconductor factories have increased in cleanliness, and humans have not been able to operate devices. For this reason, if a person is interposed, the downtime of the semiconductor exposure apparatus becomes longer and the production efficiency becomes worse.

  In the recent semiconductor device manufacturing field, in order to produce high-capacity, high-speed, high-performance semiconductors in high yields, technological developments such as increasing the diameter of silicon, reducing pattern width, and increasing the number of wiring layers Is underway. Along with this, the demand for the flatness of the wafer serving as the substrate has become increasingly severe. Therefore, in order to increase the flatness of the wafer, a repetitive pattern that becomes a dummy having an unspecified size may be formed in the wafer processing process. When a large number of such patterns are arranged on the wafer at equal and narrow intervals, a pattern arrangement similar to the mark edge information disclosed in Patent Document 1 may be obtained.

  Since the purpose of pre-alignment is to search for an alignment mark for determining the position of the wafer from a wide field of view, an unintended pattern may enter the field of view. If the pattern components (edge position and direction) that entered the field of view in this way happen to be the same as the template, the conventional template matching is similar to the mark edge information instead of the mark coordinates as the highly correlated coordinates. The detected pattern may be erroneously detected.

  The above phenomenon will be specifically described with reference to FIG. FIG. 10 shows a view of the wafer surface including the pre-alignment mark 251 and the hole pattern 253. In FIG. 10, a cross-shaped (solid line) pattern is the pre-alignment mark 251, and a circular pattern is the hole-shaped pattern 253. Since such fine hole-shaped patterns 253 are arranged on the wafer at equal intervals, a cross shape similar to the pre-alignment mark 251 is formed as shown at the detection position 252. This shape has a pattern arrangement similar to the mark edge information in terms of image processing.

  When template matching is performed on the wafer of FIG. 10, coordinates in the hole pattern 253 (crosses indicated by dotted lines (detection positions 252)) are not coordinates corresponding to the pre-alignment marks 251 as coordinates having a high degree of correlation. , There is a case where it is erroneously detected.

  An object of the present invention is to enable more stable detection of a mark position.

In order to achieve the above object, the detection method according to the present invention comprises:
A detection device that detects the position of a mark in an image by template matching,
A processing unit that performs the template matching on the image using a template;
The processing unit performs the template matching using a template different from the template when the degree of correlation obtained in the template matching does not satisfy an allowable condition.

In addition, a detection device according to the present invention for achieving the above object is as follows.
A detection method for detecting the position of a mark in an image by template matching,
Performing the template matching on the image using a template;
When the degree of correlation obtained in the template matching does not satisfy an allowable condition, the template matching is performed using a template different from the template.

An exposure apparatus according to the present invention for achieving the above object is
In an exposure apparatus that exposes a substrate,
Imaging means for imaging a mark on the substrate;
The above-described detection device for detecting the position of the mark in the image obtained by the imaging means,
Based on the position of the mark detected by the detection device, the substrate is positioned to expose the substrate.

  According to the present invention, it is possible to detect the position of the mark more stably.

It is a flowchart explaining pre-alignment image processing. It is a figure which shows schematic structure of the semiconductor exposure apparatus by this embodiment. It is a figure which shows schematic structure of the semiconductor exposure apparatus by other embodiment. It is a figure which shows schematic structure of the semiconductor exposure apparatus by other embodiment. It is a figure explaining the detection mark used in embodiment, and its detection signal. It is a figure explaining edge extraction of a detection mark by an embodiment. It is a figure explaining the standard template by embodiment and a matching process. It is a figure explaining the deformation | transformation of the template to XY direction by embodiment. It is a figure explaining the deformation | transformation of the template to XY direction by embodiment. It is a figure explaining the deformation | transformation of the template to the Y direction by embodiment. It is a figure explaining the deformation | transformation of the template to the Y direction by embodiment. It is a figure explaining detection mark detection when a substance with low reflectance is used for the ground of a detection mark. It is a figure explaining detection mark detection when a substance with high reflectance is used for the ground of a detection mark. It is a figure which shows an example of the change of the mark line width of a detection mark. 5 is a diagram illustrating an example of a wafer surface including a pre-alignment mark 251 and a hole-like pattern 253. FIG. It is a flowchart explaining the pre-alignment image processing by 3rd Embodiment. It is a figure explaining the template by 3rd Embodiment. It is a flowchart explaining the pre-alignment image processing by 4th Embodiment. It is a figure which shows the manufacture flow of a semiconductor device. It is a figure which shows the detailed flow of a wafer process.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings.

<First Embodiment>
In the present embodiment, more stable mark detection is realized for an alignment mark image that is difficult to detect due to deformation of the mark line width, change in mark brightness, and the like. For this reason, the deformation of the detection target mark and the brightness and darkness of the pattern matching disclosed in Japanese Patent Laid-Open No. 2000-260699, which has an excellent effect in detecting low contrast, noise images, images having defects generated during wafer processing, etc. Template optimization processing and self-learning function corresponding to changes in

  The evaluation of pattern matching disclosed in Japanese Patent Laid-Open No. 2000-260699 is based on the use of a template that represents a spatial position and the directionality of the pixel at that position. Optimization of the template corresponding to the deformation of the mark according to the present embodiment is performed by repeating template matching while performing an operation of reducing or enlarging a spatial position of the template, particularly a portion corresponding to the line width. In this embodiment, the optimization corresponding to the change in the brightness of the mark is performed by an operation that reverses the directionality of the template. Further, self-learning is a function of storing a template optimized immediately before for performing template matching processing from the next time.

  First, the pre-alignment operation will be described with reference to FIG. 2A. FIG. 2A is a view showing the schematic arrangement of the semiconductor exposure apparatus according to the present embodiment. A pattern to be exposed on the wafer exists on the reticle 1, and is illuminated using an I-line or excimer laser light from an illumination system (not shown), and the projection light is irradiated onto the wafer 5 through the projection lens 2. The wafer 5 is exposed.

  On the other hand, the pre-alignment is placed on the wafer suction chuck 4 on the XY stage 3 by a wafer transfer device (not shown). The wafer on the chuck is loaded with the accuracy of the transfer device. Therefore, precise wafer position measurement cannot be performed. Therefore, a pre-alignment (rough alignment) mark on the wafer is observed with an off-axis scope 6 formed outside the projection lens, photoelectrically converted with a CCD camera 7 or the like, and the position of the mark is detected with a pre-alignment image processing device 8. To detect. The photoelectrically converted video signal is converted into digital information by an A / D converter 81, and a pre-alignment mark position is detected by an image processor 82 having an image memory.

  Since it is advantageous in terms of throughput that the X and Y coordinates can be detected with one mark, the alignment mark has a shape shown as a mark 100 in FIG. The position of the stage when the pre-alignment mark image is captured is accurately measured by the laser interference system 12, and the exposure control device 9 correctly measures the deviation of the mark position and the amount of deviation of the wafer placed from the stage position. To do.

  In the first embodiment, a case of dark field illumination will be described as an example of illumination with an off-axis scope. In the case of dark field illumination, scattered light from the edge position of the mark step is received by a CCD camera or the like. However, it goes without saying that the present embodiment can be applied even to bright field illumination.

  Next, the pre-alignment image processing according to the present embodiment will be described with reference to FIG. Note that the process described below is a process realized by a CPU (not shown) of the image processor 82 executing a control program stored in a memory (not shown) such as a ROM.

  FIG. 1 is a flowchart for explaining pre-alignment image processing. Edge extraction is performed on the image captured by the CCD camera 7 by the image processor 82 to generate edge information (step S100). In the edge extraction, information such as whether the edge is an upward edge, a downward edge, a rightward edge, or a leftward edge is also extracted at the same time. In this embodiment, the four directions are up, down, left and right, but four directions of 45 degrees may be added to form eight directions. Note that whether the edge direction is upward, downward, rightward, or leftward is determined based on whether a threshold value thl, thr, thu, tho, which will be described later with reference to FIG.

  In step S101, an edge search is performed from the edge information. Search means searching for an approximate position of the mark in the image. Since the shape of the mark is known, it is known where and what edge is from the center of the mark. The two-dimensional arrangement of the position of the upward edge, the position of the downward edge, the position of the rightward edge, and the position of the leftward edge is stored as a registered edge arrangement, and the matching between the edge image extracted in step S100 and the registered edge arrangement is performed. Do.

  If the matching result is equal to or greater than the detection determination threshold, the mark search is successful, and the process proceeds from step S102 to step S103. In step S103, the detailed position of the mark is measured.

  On the other hand, if the search in step S101 is not successful, the process proceeds from step S102 to step S104. If the number of searches has not reached the specified number, the process proceeds to step S105, the template shape deformation is adjusted by a predetermined method, and the process returns to step S101 to perform the search again. Since the edge information that has already been acquired can be used, the processing is executed from the search processing in S101. Such template shape deformation and search repetition loops are controlled by a preset number of times or conditions. In short, a detection error occurs when it is determined that the mark cannot be detected even if the above repeating loop is repeated further.

[About Edge Extraction (S100)]
Next, the edge extraction in step S100 described above will be described. 3 and 4 are diagrams for explaining edge extraction according to the present embodiment. The scattered light of the mark 100 is received by the CCD camera 7 and subjected to photoelectric conversion, and then A / D converted by the A / D converter 81 and stored in the memory.

  A signal Xi is obtained for a scanning line in the X direction where the stored image is present. In the signal, the mark edge position shines and the other parts are black. Differentiating signal Xi gives signal Xid. When the scanning line is viewed from the left side, it becomes a positive signal at the rising edge and a negative signal at the falling edge. When a threshold value of thl is set on the rising side of the signal Xid and binarization is performed depending on whether the signal Xid is larger than thl or not, the signal Le indicating an edge is obtained. Similarly, a threshold value thr is set, and if the signal Xid is smaller than thr, if not, the binarization results in a signal Re indicating an edge. The signal Le indicates the leftward edge position of the mark signal, and the signal Re indicates the rightward edge position of the mark signal.

  Similarly, when a signal in a vertical line (Y direction) on the memory is created, a signal Yi is obtained. When Yi is viewed from below and binarized by setting the differentiation and threshold as in the case of X, a signal Ue and a signal Oe are obtained. The signal Ue indicates the downward edge position of the mark signal, and the signal Oe indicates the upward edge position of the mark signal.

  FIG. 4 shows a two-dimensional image in which the right edge position, left edge position, upward edge position, and downward edge position of the mark 100 are obtained. When combined, the result is as shown in FIG. 4A. As independent information, the upward edge position (FIG. 4B), the downward edge position (FIG. 4C), and the left edge position (FIG. 4D). The rightward edge position (FIG. 4E) is stored as an edge position image.

  The search in step S101 is performed by matching calculation between a template stored in advance and the edge position images shown in FIGS. 4B to 4E. The search will be described with reference to FIG. Since the positions of the upper, lower, left, and right edges with respect to the mark center (cross) 51 are known in advance, the characteristic part of the mark is set as the position of the circle, and FIG. An arrangement as shown in (e) is registered as a template. When the upper side (FIG. 5B), lower side (FIG. 5C), left side (FIG. 5D), and right side (FIG. 5E) of the registered template are combined, FIG. 5A is obtained. Become. The position of the circle is called a feature point of interest, and in this embodiment, this set of points of interest is used as a template. The template is composed of a direction feature indicating the position of the feature point of interest and the direction of the edge of the point of interest.

[About Edge Search (S101)]
The matching calculation in the search determines whether or not the edge information shown in FIG. 4B exists at the position of the circle in FIG. 5B with respect to the position of the mark center 51 (cross position). Similar determinations are made for FIGS. 4C and 5C, FIGS. 4D and 5D, and FIGS. 4E and 5E. If edge information is present at all the positions of the circles, the degree of correlation is 100%. If there is a place where there is no edge information at the position of the circle, it falls from 100%. Such a matching operation is performed on the entire edge image. Finally, the coordinates of the mark position having the highest correlation among them are extracted and the search is completed.

  Note that the feature points of interest shown in FIG. 5 are expressed by being thinned. Even if this number is increased, there is no effect unless a feature is expressed. Even if the density is increased, if the correlation decreases, the degree of correlation (correlation coefficient) may drop rapidly, and detection may not be determined. In such a case, there are many cases where the mark is mainly missing. Therefore, a higher detection rate can be maintained by setting the thinned feature points of interest.

  If the maximum correlation mark can be selected in the above search, but the value is lower than the determination threshold, the coordinates of the template from which the maximum correlation is obtained may not indicate the correct mark position. . That is, the template shape may not match the alignment mark shape stamped on the wafer. Therefore, the template shape is deformed to optimize the template. Hereinafter, template optimization will be described.

[Change / Optimization of Template Shape (S105)]
The template shape deformation will be described with reference to FIGS. 6A and 6B. In both FIGS. 6A and 6B, the template shown in FIG. 5A is modified to a standard shape, that is, a shape as designed. FIG. 6A shows a template one pixel larger in the XY direction. FIG. 6B shows a template that is two pixels larger in the XY direction.

  The difference between the template of FIG. 6A and the standard template of FIG. 5 can be seen from (d) by moving the upward edge position one pixel upward with respect to the standard template, as can be seen from FIGS. The edge position facing left with respect to the standard template is shifted to the left by one pixel. Also, the difference between the template of FIG. 6B and the standard template is that the upward edge position is one pixel upward (b), the downward edge position is one pixel downward (c), and the left edge position is relative to the standard template. Is moved to the left by one pixel (d), and the edge position facing right is moved to the right by one pixel (e).

  Although not shown, a template smaller by 1 pixel and a template smaller by 2 pixels than a standard template (FIG. 5) are prepared in the same manner as described above. For example, a template that is smaller by one pixel moves the upward edge position shown in FIG. 5B downward by one pixel, and moves the left edge position shown in FIG. 5D rightward by one pixel. Obtained. Further, a template that is two pixels small has an upward edge position (b), a downward edge position (c), a leftward edge position (d), and a rightward edge position (e) shown in FIG. , Obtained by moving one pixel to the right and left.

Note that the template deformation is not limited to the above four types, and a template that is three pixels larger or smaller than the standard template can be used. And, for example, the following rules:
(1) One pixel larger template
(2) One pixel smaller template
(3) 2 pixel large template
(4) 2 pixel small template
(5) 3 pixel large template
(6) 3 pixel small template::
::
It would be efficient to automatically generate a template deformed according to the above.

  The template deformation and edge search are repeated until the interval corresponding to the line width becomes 1 or less, or until the interval corresponding to the line width exceeds an assumed deformation amount. If the template cannot be matched at this point, a detection error is generated (S104). In addition, it is easy to match the template with the mark deformation because it is highly probable that the deformation of the template is large, small, large, and small. Because it ends.

  Of course, the deformation template automatic generation method need not be limited to the method described above. For example, the detection mark may be deformed only in the X direction, or may be deformed only in the Y direction. Therefore, as shown in FIGS. 7A and 7B, template deformation may be performed in the X direction, or template deformation in the Y direction may be performed in the same manner as FIGS. 7A and 7B. 7A shows a template that is one pixel larger in the X direction, and FIG. 7B shows a template that is two pixels larger in the X direction.

Regarding the automatic generation of such templates,
(1) A template that is one pixel larger in the X (Y) direction
(2) One pixel smaller template in the X (Y) direction
(3) A template 2 pixels larger in the X (Y) direction
(4) A template 2 pixels smaller in the X (Y) direction
(5) A template that is 3 pixels larger in the X (Y) direction
(6) A template that is 3 pixels smaller in the X (Y) direction
:
:
The automatic generation according to the order as described above will be efficient.

  Furthermore, as automatic generation, an optimal template automatic generation procedure can be determined by combining all of XY deformation at the same time, X deformation, and Y deformation.

  When template matching is performed while automatically generating a deformed template and the maximum correlation exceeds the determination threshold, it is determined that the automatically generated template has the same shape as the mark on the wafer, and the optimization is completed. The coordinate at which the maximum correlation is obtained is the correct mark position. The modified template may be stored in advance in a memory and read as necessary. However, if a method of calculating the deformation of the template is used each time, the standard template and the template after learning may be held, so that memory can be saved.

  Note that the automatically generated template that has exceeded the determination threshold is stored (self-learning), and is used as a template to be used in the next and subsequent template matching. Since it is highly likely that the mark on the wafer is deformed in the same way from the next time onward, by performing such self-learning, the probability that the template matching at the next time and below will fall below the determination threshold is reduced. Therefore, it is possible to quickly find the mark position without repeating template optimization.

  If the mark position cannot be detected with the optimum template as a result of the previous self-learning, the template is matched with the design value (standard template), and the template is optimized using the method described above until the judgment threshold is exceeded. And self-learning.

  In the above description, the optimization is completed when the maximum correlation exceeds the determination threshold. However, the present invention is not limited to this. For example, (1) a method of performing optimization to the limit of template deformation and adopting a template that can obtain the largest maximum correlation in the optimization process, and (2) the maximum correlation exceeds a determination threshold. There is also a method of adopting a template having a center line width (a template having an average of the line widths of those templates and a line width closest to the average) among a plurality of templates that are formed. These methods can be said to be templates that are insensitive to mark deformation because of the large size margin of the optimized template.

  Precise detection after the mark search is completed can determine the mark position with an accuracy equal to or lower than the pixel resolution by, for example, a method of obtaining the center of gravity from the luminance distribution with the search center of the AD-converted image as the origin. .

  Further, although the edge extraction is performed after the image is captured, noise removal filtering may be performed before the edge extraction so that unnecessary edge information is not generated by lowering the noise level in advance. The addition of such processing has the effect of increasing the mark search detection rate.

  As described above, according to the first embodiment, by automatically performing template optimization, even when the detection mark cannot be detected by matching the standard template, the process is not simply terminated. It becomes possible to search and set a parameter with a high detection rate. Thus, even an image that has been difficult to detect can be detected, and fine alignment can be executed without stopping by pre-alignment.

Second Embodiment
In the first embodiment, the optimization of the template shape in the case where the line width of the mark is increased or decreased is described. In the second embodiment, template optimization for a case where the brightness of a mark changes as shown in FIGS. 8A and 8B will be described. In the second embodiment, bright field illumination is used. However, the description in the bright field is for convenience of description, and the second embodiment can be applied to the dark field. In the dark field, detection is performed using edge scattered light. However, if the mark portion is flat and is formed of a rough surface other than the mark, the mark may be black and the rough surface may be white. For this reason, white and black may change even in a dark field. In such a case, the second embodiment can be applied.

  FIG. 8A shows an example (Black) in which the background is white and the mark portion is black. Conversely, FIG. 8B shows an example in which the background is black and the mark portion is white (White). The difference in brightness between the mark portions is caused by the difference between the mark on the wafer and the material around the mark. Since the reflectance of light differs depending on the material, in the case of bright field illumination, a difference in brightness occurs between the mark portion and the periphery.

  Next, the mark template shown in FIGS. 8A and 8B will be described. As described in the first embodiment, when the mark contour portion is differentiated and the edge direction component is extracted by thl, thr, thu, thd, the signal shown in (f) of FIGS. 8A and 8B is obtained. 8A and 8B, only the edge components in the leftward and rightward directions are shown for the sake of simplicity, but the same applies to the edges in the upward and downward directions.

  As shown in (f) of FIGS. 8A and 8B, it can be seen that the edge positions are the same by reversing the brightness, but the edge directions, that is, the left direction and the right direction are reversed. Although not shown, the same applies to the upward edge and the downward edge. Using this property, optimization corresponding to the white / black change of the mark due to template deformation is performed.

  For example, when the template prepared in advance corresponds to the white type (FIG. 8B), detection is impossible if the mark to be detected is the black type (FIG. 8A). Therefore, it falls below the detection determination threshold. If it falls below the detection judgment threshold, the template is automatically changed to the black type and the template matching process is performed again. In this case, since it matches the brightness of the mark to be detected, it does not fall below the detection determination threshold. Note that the black type and the white type of the template are changed by switching the direction of the edge to the left and the right and the direction of the upward and the downward while maintaining the position of the edge. Here, the edge direction in the template has been described as switching between the right direction and the left direction, and switching between the upward direction and the downward direction, but the polarity of the image in which the edge is detected may be changed. Change the definition of the right-facing edge image to the left-facing edge image, the left-facing edge image to the right-facing edge image, the upward-facing edge image to the downward-facing edge image, and the downward-facing edge image to the upward-facing edge image, and obtain the correlation with the original template. But the same result can be obtained. That is, it is also possible to cope with the change of Left and Right of the edge detection signal, and the exchange of Over and Down.

  As in the first embodiment, the template of the type determined in the last optimization is self-learned and used for template matching from the next time. If it cannot be detected, it can be automatically returned to the white type again.

  As described above, according to the second embodiment, the brightness of the mark is automatically recognized regardless of the type of the detected brightness of the mark, and the pattern matching is continued.

  It is possible to cope with a wide range of mark deformation by combining the template optimization corresponding to the line width deformation of the mark described in the first embodiment and the template optimization corresponding to the change in mark brightness described in the second embodiment. It is good. Also in this case, if the template whose shape, brightness and darkness have been optimized immediately before is adopted as the template matching template for the next and subsequent times, the next and subsequent matching processing does not require template optimization processing. It is possible to detect.

  Further, the template matching process is not limited to the method shown in the first and second embodiments, and any process using a template describing a two-dimensional space can be applied. The above embodiment can also be applied to matching using normalized correlation, binary pattern matching, and the like.

  As described above, according to the first and second embodiments, the edge of the mark image is optimally extracted, and simultaneously the direction characteristics of the edge are extracted together. Pattern matching focusing on edge position is performed. Furthermore, since the template is automatically optimized with respect to changes in mark deformation and mark brightness, it is possible to detect mark positions for a wide variety of types of marks. That is, more stable mark detection can be performed for images of all kinds of alignment marks for manufacturing high-density semiconductors. As a result, not only the yield of semiconductor manufacturing is improved, but also the time during which the semiconductor manufacturing apparatus is not driven, that is, the downtime can be reduced, the reliability is further increased, and future high-density processing can be supported.

<Third Embodiment>
In the third embodiment, a part where no edge exists is added to a part where an edge exists as an element constituting the template. That is, the template information described in the first embodiment is extended by adding a location where an edge does not necessarily exist, that is, a location where no presence is expected as a feature focus point. For example, the edge pattern does not exist inside the pre-alignment mark 251 shown in FIG. Therefore, by adding a point for evaluating that there is no pattern inside the template, it is possible to prevent erroneous detection even in a place where the pattern on the hole is dense (that is, the detection position in FIG. 10). 252 can be prevented from being erroneously detected).

  The configuration of the exposure apparatus according to the third embodiment is the same as that of the first embodiment (FIG. 2A), and a detailed description thereof is omitted. In addition, from the viewpoint of throughput, the alignment mark has a cross shape (mark 100 in FIG. 3) that can detect both the X and Y coordinates with a single mark, as in the first embodiment.

  In the third embodiment, description will be given using dark field illumination as illumination of the off-axis scope. In dark field illumination, scattered light from the edge position of the mark step is received by a CCD camera or the like. Needless to say, the present invention can be similarly applied to bright-field illumination.

  Hereinafter, pre-alignment image processing by the exposure apparatus of the third embodiment will be described. FIG. 11 is a flowchart for explaining pre-alignment image processing according to the third embodiment.

  The image captured by the CCD camera 7 is subjected to edge extraction by the image processor 82 (step S200). In the edge extraction, information on whether the extracted edge is the upper side, the lower side, the left side, or the right side of the detection mark signal is acquired at the same time. 4B to 4E are two-dimensional images obtained by obtaining the upper edge position, the lower edge position, the left edge position, and the right edge position of the mark 100, respectively. When the positions of the edges are combined, FIG. 4A is obtained. In the third embodiment, as the edge position image, the upper edge position in FIG. 4B, the lower edge position in FIG. 4C, and FIG. ) And the right edge position in FIG. 4E are stored as independent information.

  Next, in step S201, an edge search is performed from the edge information extracted in step S200. Search means searching for an approximate position of the detection mark in the image. Since the detection mark shape is known, the position of the edge is registered and stored in advance. In step S201, matching between the registered template and the edge position image extracted in step S200 is performed. Details of the matching process of the third embodiment will be described later.

  In step S202, it is determined whether or not the mark detection result in step S201 is correctly detected. That is, if the matching result is equal to or greater than the detection determination threshold value, the mark search is successful, and the mark is searched for in detail in step S203 using the mark position searched in step S201. On the other hand, if the matching result is less than or equal to the detection determination threshold, a detection error occurs.

  The details of edge extraction, top and bottom, left and right, edge separation, and template matching are the same as those in the first embodiment, and thus detailed description thereof is omitted.

  Next, template matching according to the third embodiment will be described with reference to FIGS.

  The search is performed by template matching between the templates of FIGS. 12B to 12E stored in advance and the edge position images shown in FIGS. 4B to 4E. Since the positions of the upper, lower, left, and right edge positions with respect to the mark center (cross) are known in advance, an edge always exists as a template, as shown in FIGS. ○ Register the position of the marked part. Next, the positions of the X mark locations shown in FIG.

  The position of the mark ○ and the position of the mark X indicate the characteristics of the mark. When the registered templates are combined, FIG. 12A is obtained. Note that the positions of the ○ mark and the X mark are called feature focus points, and a set of feature focus points is used as a template. Further, there are four types of feature attention points at the positions of the circles, ie, upper side, lower side, left side, and right side. In the present embodiment, these four types will be described as positions where edges exist without being distinguished. The cross positions shown in FIGS. 12B to 12F are the center positions of the marks.

  In template matching in search, edge information exists at positions corresponding to all the positions of the circles in FIG. 12 in the image of FIG. 4, and positions corresponding to all the positions of the x marks shown in FIG. This is done by determining that no edge information exists. At this time, when an edge component is present at the feature target point position indicated by ◯, the correlation is increased, and when an edge component is present at the feature target point position indicated by X, the correlation is decreased. When the edge information exists at all the feature target point positions of the ◯ marks and the edge information does not exist at all the feature target point positions of the X marks, the correlation degree is 100%. On the other hand, if there is a place where the edge information does not exist at the position of the mark ◯ or there is a place where the edge information exists at the position of the mark X, the correlation decreases from 100%. The template matching is performed on the entire edge position image while moving the center position of the mark, and finally the coordinates having the highest correlation are extracted to complete the search.

  The correlation degree by this template matching will be specifically described using the image of FIG. In template matching, whether or not edge information is present at positions corresponding to all the positions of the circles in FIG. 12 in the image of FIG. 10, and whether edge information is present at positions corresponding to all the positions of the x marks. It is done by judging. In the template shown in FIG. 12, the correlation weights of the feature attention points marked with “◯” and the feature attention points marked with “X” are the same for explanation. The following table summarizes the weights.

  The breakdown of the feature attention points of the template shown in FIG. 12 is 24 points for the circle mark and 16 points for the x mark.

  When template matching is performed under the above conditions, in the alignment mark 251 of FIG. 10, there are edges at all the 24 feature-focused points, and there are no edges at all 16 feature-marked points. Therefore, the degree of correlation is 24/24 = 100%. On the other hand, at the detection position 252, there are edges at all 24 feature attention points indicated by ◯, but there are also edges at 16 feature attention points. Therefore, the correlation degree is (24−16) / 24 = 33%. Therefore, when the correlation degree at the detection positions 251 and 252 is compared, a higher value is obtained for the correlation degree at the detection position 251. Therefore, the alignment mark 251 is determined as a correct detection position.

  In the above description, the weights of the feature points of interest for the ◯ mark and the X mark are the same, but they may be different. In this case, the correlation field calculation result changes. For example, if a weight of 1.5 is given with emphasis on the x mark, (24) / 24 = 100% at the detection position 251 and (24-16 × 1.5) / 24 = 0% at the detection position 252. In this way, the discriminating power of the patterns at the detection positions 251 and 252 can be increased by the weight.

  In the above description, attention is paid to the case where an edge exists at the position of the circle and the position of the circle as a template. Similarly, if an edge component exists at the feature point of interest of the circle, the degree of correlation is increased, The matching degree may be increased even when there is no edge component at the feature point of interest marked by X. In this case, the following table summarizes the template matching weights.

  The correlation degree by the template matching will be specifically described using the image of FIG. When template matching is performed using the weighting shown in Table 2, at the detection position 251, there are edges at all 24 feature points of interest marked with ◯ and there are no edges at feature points of interest marked with 16 ×. . Accordingly, the degree of correlation is (24 + 16 / (24 + 16) = 100%) On the other hand, at the position of 252, there are edges at all 24 feature points of interest marked by ○, but feature feature points of 16 × marks. Therefore, the correlation degree is (24 + 0) / (24 + 16) = 60% When the correlation degrees of the detection positions 251 and 252 are compared, a higher correlation degree is obtained at the detection position of 251. 251 is determined as the correct detection position.

  Furthermore, it goes without saying that the template matching described above can be modified within a range conceivable by those skilled in the art. For example, template matching as shown in Table 3 below may be used by combining the weights in Tables 1 and 2 above.

  By the template matching shown in Table 3, the degree of correlation at the detection positions 251 and 252 in FIG. 10 is calculated as follows. That is, (24 + 16) / (24 + 16) = 100% at the detection position 251 and (24−16 / (24 + 16) = 20% at the detection position 252.

  In this way, it can be seen that the introduction of the feature point of interest marked with x marks increases the discrimination ability between the true mark and the similar mark. The template matching method shown in Table 3 has the highest discrimination effect.

  As described above, it has been described that a correct detection position is obtained when template matching is performed using the template of the third embodiment. However, since the number of feature points of the template increases, the time required for calculation increases. When it is necessary to consider such an increase in calculation time, the following method may be used.

  This will be described with reference to the flowchart of FIG. In the first process of step S201, the template is set to only the feature points of interest in the circles in FIGS. 12B to 12E, and the edges at all the circles in FIG. Perform template matching to determine whether there is information. If there are a plurality of detection positions as a result of this matching, the process returns from step S202 to step S201. In the second process of step S201, the template is set as the feature focus point of the ◯ mark and the X mark in FIGS. 12B to 12F, and any of the template matching described above is performed. In this way, the combination of feature points of interest is changed to only ○ marks, and further, such as ○ marks and X marks, and by increasing the number of feature points of interest as necessary, the mark detection rate and the detection speed can be satisfied simultaneously. be able to.

  In this case, the number of feature points of interest marked with X may not be introduced all at once, and the number thereof may be gradually increased. For example, in the above example, in the second processing in step S201, first, eight feature attention points are added out of the 16 feature attention points, and if the detection candidates still cannot be narrowed down, the third and fourth times in step S201 are performed. It may be increased to 12 or 16 by processing. Of course, the number of feature points of interest marked with ○ may be combined and the number of feature points of interest marked with × may be changed (for example, the first time, 16 circles, 8 x marks, the second time) Is 20 circles and 12 crosses).

  Furthermore, it is effective for speeding up detection if the next template matching is performed using a combination of feature points of interest determined by this method.

<Fourth embodiment>
In the fourth embodiment, template matching that does not erroneously detect the on-hole pattern even when a part of the mark to be truly detected is missing and difficult to detect will be described. FIG. 13 is a flowchart for explaining pre-alignment image processing according to the fourth embodiment. In FIG. 13, steps having the same functions as those in the flowchart of FIG.

  After the search is performed, there is a case where the correlation between the mark positions is low due to a part of the mark missing, and a plurality of detection mark candidate points exist, resulting in a detection error. In such a case, as shown in step S212, the position of the feature point of interest marked by X is slightly changed, and the search is performed again. By doing this, the degree of correlation or the detection position of the pattern existing at a position other than the mark, such as the hole-shaped pattern 253, changes according to the position of the feature point of interest indicated by the x mark. On the other hand, since the edge information does not exist at the feature point of interest of the mark X, the true mark is not affected by the movement of the feature point of interest of the mark X, and the correlation degree and position do not change.

  As described above, even if the position of the feature point of interest (x mark) where no edge exists is changed, the point where the degree of correlation and the position are stable can be identified as the true mark position. Therefore, in step S211, after passing through step S212, the function of identifying the correlation degree and the stable position among the plurality of candidate points is also executed.

The detection criteria in step S211 are as follows.
(1) If one detection position exceeds the set threshold, detection is complete. The position is set as a mark detection position.
(2) When the feature point of interest marked with x is changed,
-The position where the degree of correlation changes every time the position of the feature point of interest marked by X is excluded from the determination position.
-The position that appears every time the position of the feature point of interest marked with X is changed is excluded from the judgment position. (3) If there is one detection position that is not excluded from the determination position, that position is set as the mark detection position.

  Furthermore, template optimization is also achieved by performing the next template matching using the position of the feature point of interest determined by this method.

  As described above, according to the third and fourth embodiments, when the template is configured, the feature points of interest are the portion where the edge portion exists and the portion where the edge does not exist. As a result, even if a pattern having characteristics similar to the edge portion such as the hole-like pattern 253 exists on the wafer at the time of template matching, the degree of correlation of the portion can be lowered, and the correct mark position Can be detected. Mark misdetection can be effectively prevented by using a portion where no edge exists as a feature point of interest.

<Fifth Embodiment>
In the first to fourth embodiments, the case of adapting to pre-alignment using an off-axis microscope has been described. However, the mark position search, template optimization, and self-learning function described in the first and second embodiments, In addition, the matching process using the feature point position where no edge exists as described in the third and fourth embodiments is not limited to the pre-alignment process using off-axis. For example, the present invention can be applied to the following position detection system.

  FIG. 2B is a diagram showing an embodiment of a TTR (Through The Reticle) detection system that detects a mark on a wafer or a stage through a reticle. In the case of TTR, the mark is detected by exposure light. For example, in the case of an excimer laser exposure apparatus, the CCD camera 7 and the laser 21 are synchronized by a synchronization signal generator 20. The laser is oscillated only during the light accumulation time of the CCD camera 7. Using the mark image photoelectrically converted in this way, the mark position search, template optimization, and self-learning are performed by the same method as in the first to fourth embodiments. Then, a precise mark position calculation is performed after the search.

  In addition, in the case of I-line instead of excimer laser, image capturing and illumination system synchronization are not necessary, but the others are the same.

  As another example, FIG. 2C shows an embodiment of a TTL (Through The Lens) detection system that detects a mark position on a wafer or stage via a projection lens without using a reticle. Also in this case, the mark search, template optimization, and self-learning function are the same as those in the first to fourth embodiments, except that the mark imaging method is different.

  Further, the position detection in each of the above embodiments can be applied to wafer position detection in an electron beam type (EB) exposure apparatus that does not use a reticle (not shown). In the case of the EB exposure apparatus, mark search and position determination can be performed in the same manner as in the first to fourth embodiments.

  The search using the feature point of interest described in the third and fourth embodiments can also be applied to the template described in the first and second embodiments.

<Other embodiments>
[Application to semiconductor manufacturing equipment]
Next, an embodiment of a device manufacturing method using the above-described exposure apparatus will be described. FIG. 14 shows a manufacturing flow of a microdevice (a semiconductor chip such as an IC or LSI, a liquid crystal panel, a CCD, a thin film magnetic head, a micromachine, etc.).

  In step 101 (circuit design), a semiconductor device circuit is designed. In step 102 (mask production), a mask on which the designed circuit pattern is formed is produced. On the other hand, in step 103 (wafer manufacture), a wafer is manufactured using a material such as silicon. In step 104 (wafer process), called a pre-process, an actual circuit is formed on the wafer by lithography using the prepared mask and wafer. Next, step 105 (assembly) is referred to as a post-process, and is a process for forming a semiconductor chip using the wafer produced in step 104, and is a process such as an assembly process (dicing, bonding), a packaging process (chip encapsulation), or the like. including. In step 106 (inspection), inspections such as an operation confirmation test and a durability test of the semiconductor device manufactured in step 105 are performed. Through these steps, the semiconductor device is completed and shipped (step 107).

  FIG. 15 shows a detailed flow of the wafer process. In step 111 (oxidation), the wafer surface is oxidized. In step 112 (CVD), an insulating film is formed on the wafer surface. In step 113 (electrode formation), an electrode is formed on the wafer by vapor deposition. In step 114 (ion implantation), ions are implanted into the wafer. In step 115 (resist process), a photosensitive agent is applied to the wafer. In step 116 (exposure), the circuit pattern of the mask is printed onto the wafer by the exposure apparatus described above. In step 117 (development), the exposed wafer is developed. In step 118 (etching), portions other than the developed resist image are removed. In step 119 (resist stripping), the resist that has become unnecessary after etching is removed. By repeatedly performing these steps, multiple circuit patterns are formed on the wafer.

  If the exposure apparatus described in the first to fifth embodiments is used in the above manufacturing method, more stable mark detection can be performed for any alignment mark image, and the yield of semiconductor manufacturing is improved. At the same time, downtime is reduced and throughput is improved.

Claims (8)

A detection device that detects the position of a mark in an image by template matching,
A processing unit that performs the template matching on the image using a template;
The detection unit, wherein the template matching is performed using a template different from the template when the degree of correlation obtained in the template matching does not satisfy an allowable condition.   The processing unit executes repetition of template matching when the degree of correlation does not satisfy the allowable condition until the degree of correlation satisfies the allowable condition or the number of repetitions reaches a predetermined number of times. The detection apparatus according to claim 1.   The detection apparatus according to claim 2, wherein the processing unit sets a template in which the degree of correlation satisfies the allowable condition in the repetition as a template used for a next template matching.   The processing unit executes a template matching iteration when the correlation degree does not satisfy the permissible condition for a predetermined number of times, and based on a plurality of correlation degrees respectively obtained using a plurality of templates by the repetition. The detection apparatus according to claim 1, wherein one template is adopted from the plurality of templates.   The detection apparatus according to claim 4, wherein the processing unit sets the adopted template as a template used for a next template matching. A detection method for detecting the position of a mark in an image by template matching,
Performing the template matching on the image using a template;
A detection method characterized by performing the template matching using a template different from the template when the degree of correlation obtained in the template matching does not satisfy an allowable condition. In an exposure apparatus that exposes a substrate,
Imaging means for imaging a mark on the substrate;
The detection device according to any one of claims 1 to 5, which detects a position of a mark in an image obtained by the imaging means.
An exposure apparatus, wherein the substrate is exposed by positioning the substrate based on a mark position detected by the detection device. A step of exposing the substrate using the exposure apparatus according to claim 7;
And a step of developing the substrate exposed in the step.

Images