JP3873373B2 - Image recognition method and image recognition apparatus for semiconductor wafer - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an image recognition method and an image recognition apparatus for a semiconductor wafer. The image recognition method according to the present invention can detect the position of various patterns and perform various inspections on the patterns, measure the length of the patterns, and the like. It can be used in the process of manufacturing various materials (such as electronic materials). In addition, the semiconductor wafer image recognition apparatus according to the present invention needs to know the position of the pattern when inspecting or measuring various patterns formed on the semiconductor wafer, or in the manufacturing process of the semiconductor device. It can be general-purpose in some cases.
[0002]
[Prior art]
Conventionally, the position of various patterns is detected by image recognition. For example, in a length measuring SEM (Scanning Electron Microscope), automatic detection of a point to be measured is performed by image recognition by pattern matching. In such image recognition used in, for example, a length measurement SEM, when the measurement pattern is finished in substantially the same size and shape, the detection accuracy is improved to a level where there is almost no erroneous recognition. For example, for a mass-produced wafer, for measurement and the like for performing QC (quality control) in the process, the measurement pattern is finished in almost the same size and shape, and it can be said that there is almost no erroneous recognition.
[0003]
However, in some cases, the specific pattern to be detected is not properly formed, and thus cannot be recognized. For example, with regard to patterns formed by exposure, in the case of patterns exposed by changing the exposure amount and focus by setting conditions, etc., deterioration occurs in patterns formed under conditions that deviate from the optimum conditions, and the pattern that becomes the reference When the difference is large, the recognition rate is greatly reduced, and measurement may not be possible.
[0004]
For example, the problems of the prior art will be described with reference to an example of a length-measuring SEM that measures the length of a pattern formed through an exposure process on a wafer. In measuring the length, it is usually assumed that a file has already been created and registered for the shot map in the wafer as the object to be detected and the coordinates of the points in the chip.
[0005]
When performing automatic length measurement, first, a length measurement point and its surrounding pattern are captured as image data. In the current technology, in the length measurement SEM, the stage stopping accuracy is about ± 3 μm, so the stage itself can be brought to a predetermined position. Here, the comparison between the reference image data and the actual pattern is performed to calculate the possibility of the same pattern. Here, as shown in FIG. 2, detection is performed using a pattern in which line-shaped patterns are repeatedly arranged as a reference image pattern.
[0006]
The reference image pattern shown in FIG. 2 is a pattern corresponding to a properly formed pattern. Therefore, ideally, a pattern exactly the same as the reference image pattern is formed on the wafer, which is recognized in comparison with the reference image pattern, and its position is detected.
[0007]
That is, in the manufacturing process, all the chips should be exposed in the same way, so that the same pattern as the reference should be recognized unless it is a defective product. However, in the case of condition setting or an experimental wafer, the exposure conditions differ from chip to chip, so that there may be a significant deviation from the optimum conditions. For chips that deviate significantly from the optimum conditions, the pattern shape collapses and the same pattern is no longer recognized.
[0008]
For example, assume that the actual pattern on the wafer is slightly thicker than the reference image pattern as shown in FIG. Such a pattern is formed when a positive resist is used, or under slightly exposed conditions. In the case of the pattern shown in FIG.
[0009]
On the other hand, it is assumed that the actual pattern on the wafer is considerably thinner than the reference image pattern as shown in FIG. Such a pattern is formed when a positive resist is used and overexposure conditions are used. In the case of the pattern of FIG. 4, it is not recognized that they are the same.
[0010]
As described above, the pattern to be detected does not necessarily match the reference image pattern, and therefore, it is not recognized that the pattern is the same. Therefore, the position is not detected. This is likely to happen, and in some cases it is not even a real process.
[0011]
This problem also occurs when exposure is performed using the halftone phase shift method. The halftone phase shift method is a method in which phase shift exposure is performed using a semitransparent film (halftone film) having a light transmittance of, for example, about 10%, but when this is used, a secondary peak is used. May cause the false pattern A to be resolved. As shown in FIG. 7, this is a phenomenon in which a pattern is formed between intended intended patterns by secondary peaks between the patterns. In this case, if a reference without a false pattern is used corresponding to the pattern by appropriate exposure, it is not recognized that they are the same, and the necessary position cannot be detected.
[0012]
In the case of pattern recognition, when there are many similar patterns around, such as holes, for example, a pattern with large features such as numbers and English letters as shown in FIG. May be used. In this case, as shown in FIG. 8B (enlarged view of the corresponding part of the XIII portion in FIG. 8A), when a pattern by a secondary peak is resolved around this, the secondary peak The pattern recognition rate is lowered because the pattern without the reference is used as a reference, and the pattern recognition necessary as a reference is not performed. For this reason, measurement may not be possible.
[0013]
[Problems to be solved by the invention]
The present invention solves the above-mentioned problems of the prior art and performs position detection using pattern recognition, thereby improving the pattern recognition rate when performing, for example, automatic measurement, inspection, manufacturing, etc. It is an object of the present invention to provide an image recognition method and a semiconductor wafer image recognition apparatus with improved performance.
[0014]
[Means for solving the problems]
An image recognition method according to the present invention includes:
For an object to be detected in which a specific pattern exists, in an image recognition method for detecting an existing position in which the specific pattern is present by performing image recognition of the object to be detected and comparing it with a predetermined reference image,
A plurality of the reference images are determined for one type of a specific pattern to be detected , a priority order is determined for the plurality of reference images , and the specific pattern exists by sequentially comparing according to the priority order. In the image recognition for detecting the presence position, the priority can be changed by the detected position or the detection coordinates of the detected object .
[0015]
An image recognition apparatus for a semiconductor wafer according to the present invention includes:
In an image recognition apparatus for detecting a position where a specific pattern exists, by detecting the semiconductor wafer and comparing it with a predetermined reference image for a semiconductor wafer where a specific pattern exists.
A plurality of the reference images are determined for one type of a specific pattern to be detected , a priority order is determined for the plurality of reference images , and the specific pattern exists by sequentially comparing according to the priority order. Image recognition of a semiconductor wafer for detecting a presence position , wherein a program for changing the priority order according to the detected position or detection coordinates of the detected object is sequentially compared according to the priority order changed by the program It is characterized by performing .
[0016]
According to the present invention, when an existing position where a specific pattern exists is detected by comparing with a predetermined reference image by image recognition, a plurality of types are determined for one type of the specific pattern from which the reference image is to be detected. Therefore, even if the pattern that is formed does not match the reference image that should be obtained properly, the next candidate or the reference image as the next candidate can be set to deviate from the appropriate pattern. It will be possible to recognize that pattern. Therefore, the recognition rate can be increased and the reliability of image recognition can be increased.
[0017]
The plurality of reference images are used to sequentially compare and detect the existence position where the specific pattern exists. At this time, the plurality of reference images to be used can be prioritized. Furthermore, this priority can be changed depending on the part to be detected (for example, whether the left half or the right half of the semiconductor wafer is inspected). You may make it changeable with a detection coordinate. In principle, it is preferable to prioritize the reference corresponding to the pattern that is most likely to exist according to the scene of each detection, and to determine the priority in descending order of possibility of existence. By doing so, the time for position detection can be shortened.
[0018]
In other words, priorities can be set for a plurality of reference images, and comparison can be made in order according to the priorities, and the priorities can be determined based on the detected position or detection of the detected object (semiconductor wafer). It can be changed by coordinates.
[0019]
The plurality of reference images include a main reference image (for example, a pattern in the shape of a properly formed pattern as shown in FIG. 2) corresponding to the specific pattern formed properly, and one or more, A sub-reference corresponding to a pattern deviating from the above-mentioned specific pattern formed properly (for example, a pattern with a shape narrower than a properly formed pattern as shown in FIG. 5 or a pattern shown in FIG. A pattern having a shape thicker than a properly formed pattern) image.
[0020]
In this case, as a priority order of image recognition, the first reference can be compared using the main reference image, and the second and subsequent comparisons can be performed using the sub reference image. When a plurality of sub-reference images are used, priorities among the sub-reference images can be given and compared in order. However, when the pattern corresponding to the sub-reference image image has a higher existence probability than the appropriate pattern corresponding to the main reference image, it is natural that the higher existence probability may be given priority. In addition, since priorities are not always assigned, it is not necessary to stick to priorities.
[0021]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, specific examples of preferred embodiments of the present invention will be described with reference to the drawings. However, as a matter of course, the present invention is not limited to the illustrated embodiment.
[0022]
Embodiment 1
In this embodiment, the present invention is an example of a length measurement SEM that measures the length of a pattern on a semiconductor wafer. In particular, a case is described in which a length measurement SEM is used to measure a sample that has been exposed for several tens of points in the wafer by varying the exposure amount and focus, such as in a condition setting or experimental wafer.
[0023]
Any sample can be used. For example, in the case of a resist pattern, the substrate, the type of resist, and the shape are not limited as long as it can be measured with a length measurement SEM when exposed at least under appropriate conditions. Not only the resist pattern but also a pattern after etching may be used as long as it is a measurable pattern. Here, the following sample is considered as an example.
[0024]
A silicon wafer coated with a resist is used. The sample was exposed and developed using a KrF excimer laser (wavelength 248 nm) stepper with a 0.30 μm wide line and space (1.5 μm on the mask and reduced to 1/5). It was created. The optical conditions are NA = 0.50 and σ = 0.60. The exposure is performed by swinging at an appropriate pitch around the optimum exposure.
[0025]
In this example, position detection for length measurement is performed as follows. It is assumed that a file has already been created and registered with respect to the shot map in the wafer, which is the object to be detected, and the coordinates of the points in the chip.
[0026]
In accordance with a normal method for performing automatic length measurement, a length measurement point and its surrounding pattern are captured as image data. In the current length measurement SEM technique, the stage stopping accuracy is about ± 3 μm, so the stage itself can be brought to a predetermined position. Here, the reference image data is compared with the actual pattern to calculate the possibility of the same pattern, and if it is equal to or greater than a certain threshold, it is recognized as the same pattern. When the recognition is made, the length measurement work is started.
[0027]
In this example, the line and space pattern shown in FIG. 2 when obtained by proper exposure is used as the main reference image. If the pattern is formed under ideal appropriate conditions, the position of the pattern can be detected by comparison with the main reference image. This main reference image is referred to as reference 1 in this example.
[0028]
As mentioned above, especially in the case of condition setting or in the case of an experimental wafer, the exposure conditions differ from chip to chip, so the conditions deviate significantly from the optimum conditions, and the pattern to be formed is far from what is obtained under the optimum conditions. There is.
[0029]
Therefore, in this example, a plurality of reference images can be read. Here, a pattern image formed on a chip having different conditions as described above is read. In this example, in addition to the reference 1 shown in FIG. 2, a pattern slightly thicker than the reference 1 shown in FIG. 6 (pattern under a slightly underexposed condition using a positive resist) is stored as the reference 2. Keep it. Further, a pattern thinner than the reference 1 shown in FIG. 5 (a pattern in a case where the positive resist is used and is slightly overexposed) is stored as the reference 3.
[0030]
First, comparison with reference 1 is performed. If the position where the same pattern exists is recognized in comparison with the reference 1, the position detection ends there and the length measurement work is performed. When it is determined that the patterns are not the same in comparison with the reference 1, the comparison with the reference 2 is newly performed and the determination is made. Further, when it is determined that the reference 2 is not the same pattern, the comparison with the reference 3 is newly performed and the determination is performed.
[0031]
As the apparatus, as shown in the block diagram of FIG. 1, a memory for storing a plurality of references (here, references 1 to 3) as indicated by reference numeral 1 is secured. A program is also installed that gives priority to references and compares them in order. Here, as indicated by reference numeral 1 in FIG. 1, a procedure for performing image recognition in the order of reference 1, then reference 2, and then reference 3 is set up.
[0032]
Pattern collapse due to deviation from these optimum conditions varies depending on the exposure amount and defocus direction, so that the reference to be prioritized can be changed depending on the position on the wafer that is the detection target or the chip coordinates. To do.
[0033]
In FIG. 1, reference numeral 2 denotes a process for recombining the priority order in such a case. This priority recombination 2 changes the priority reference. For example, as shown by reference numeral 3 in FIG. 1, the priority is changed in the order of reference 1, then reference 3, and then reference 2.
[0034]
Specifically, in this example, the pattern exposed under the optical conditions described above has a thick line width when the exposure amount is small, and the line width becomes narrower as the exposure amount increases. Therefore, assuming that the exposure amount increases from left to right with the orientation flat on the bottom and the center is exposed to the optimum exposure amount, the left half of the wafer looks like a pattern corresponding to the reference 2 with a large line width. Thus, the right half of the wafer has a pattern corresponding to the reference 3 having a narrow line width.
[0035]
In this way, since the reference 2 and the reference 3 which are patterns deviated from the reference 1 which is an appropriate pattern are also registered, in the left half of the wafer, the priority of the reference 2 having a high existence probability in the region is increased, In the right half of the wafer, image recognition is performed by increasing the priority of the reference 3 having a high existence probability in the region. Specifically, in the position detection in the left half of the wafer indicated by reference numeral 5 in FIG. 1, when pattern matching with reference 1 fails, the program is used to perform the second matching using reference 2. To do. In the position detection on the right half of the wafer indicated by reference numeral 6 in FIG. 1, when the pattern matching at the reference 1 fails, the reference 3 is used to perform the second matching.
[0036]
References can be set in succession, including reference 4, 5, 6... And the reference to be prioritized can be changed according to the measurement situation such as wafer position or chip coordinates. . By adopting the configuration as in this example, even when the pattern deviates from the appropriate pattern due to various factors, the possibility of being detectable is improved.
[0037]
Embodiment 2
In this embodiment, when there are many similar patterns such as holes, when a pattern with large features such as numbers and English letters is used as a reference, this pattern is a halftone phase shift method. The present invention is used for image recognition when a pattern due to a secondary peak is resolved around a pattern.
[0038]
In this example, a single layer mask (quartz substrate) using Mo—Si as a light shielding band is used, a positive resist is applied on a silicon wafer, and exposure and development are performed to form a hole pattern. Under the same conditions as in the case of the normal mask of the first embodiment described above, exposure is performed with the exposure amount and focus being varied. When the exposure fee increases, a false pattern due to a secondary peak (sub-peak) may be resolved.
[0039]
Particularly in the case of a resolution pattern, in order to search for a target pattern by pattern matching, a characteristic pattern (such as a number) in the vicinity is often targeted. However, it is known that in such a relatively large pattern, the secondary peak (sub-peak) becomes large, and a false pattern is generated as shown in FIG. 8B at a relatively low exposure amount.
[0040]
Therefore, in this example, the image of FIG. 8B in which this false pattern exists is taken as the reference 2, and particularly when the right half of the wafer having a large exposure amount is exposed as described in the first embodiment. Referenced in case of detection. Thereby, the recognition rate for the target characteristic pattern shown in FIG. 8A can be improved.
[0041]
Embodiment 3
In this embodiment, when a hole pattern is obtained by performing exposure using a halftone phase shift method, image recognition is performed when a pattern due to a secondary peak is resolved between patterns as shown in FIG. The present invention is used.
[0042]
Even in the case of an original contact hole, even if there is a false pattern due to some secondary peaks (sub-peaks), it can be evaluated symmetrically. Imported to improve the recognition rate.
[0043]
【The invention's effect】
According to the present invention, pattern recognition is used to detect a position, thereby improving the pattern recognition rate and improving the reliability when performing, for example, automatic measurement, inspection, and manufacturing. A method and an image recognition apparatus for a semiconductor wafer can be provided.
[Brief description of the drawings]
FIG. 1 is a configuration diagram illustrating a first embodiment.
FIG. 2 is a diagram illustrating a reference 1 according to the first embodiment.
FIG. 3 is a diagram showing a pattern actually formed, and shows a case of a pattern thicker than an appropriate pattern.
FIG. 4 is a diagram showing a pattern that is actually formed, and shows a case where the pattern is narrower than an appropriate pattern.
FIG. 5 is a diagram illustrating a reference 2 according to the first embodiment.
FIG. 6 is a diagram illustrating a reference 3 according to the first embodiment.
FIG. 7 is a diagram showing a pattern when a halftone phase shift mask is used.
FIG. 8 is a diagram showing a characteristic symbol pattern when a halftone phase shift mask is used.
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Stored reference, 2 ... Priority order recombination, 3 ... Stored reference (after priority order recombination), 5 ... First position detection of detected object, 6 ... Second position detection of the detection target.

Claims (2)

For an object to be detected in which a specific pattern exists, in an image recognition method for detecting an existing position in which the specific pattern is present by performing image recognition of the object to be detected and comparing it with a predetermined reference image,
A plurality of the reference images are determined for one type of a specific pattern to be detected , a priority order is determined for the plurality of reference images , and the specific pattern exists by sequentially comparing according to the priority order. An image recognition method for detecting an existing position, wherein the priority can be changed by a detected position or detected coordinates of the detected object . In an image recognition apparatus for detecting a position where a specific pattern exists, by detecting the semiconductor wafer and comparing it with a predetermined reference image for a semiconductor wafer where a specific pattern exists.
A plurality of the reference images are determined for one type of a specific pattern to be detected, a priority order is determined for the plurality of reference images, and the specific pattern exists by sequentially comparing according to the priority order. Image recognition of a semiconductor wafer for detecting a presence position, wherein a program for changing the priority order according to the detected position or detection coordinates of the detected object is sequentially compared according to the priority order changed by the program An image recognition apparatus for a semiconductor wafer.

Images