Inspection apparatus and inspection method
The inspection device and method ensure accurate alignment of inspection and reference images by registering unique patterns based on similarity thresholds, addressing misalignment issues and false defect detection in inspection processes.
Patent Information
- Application Number
- JP2024011129
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-01-29
- Publication Date
- 2025-08-08
AI Technical Summary
Existing inspection methods face issues with positional deviation between inspection and reference images due to poor positioning accuracy, leading to misalignment and false defect detection, especially when unique registration patterns are not accurately identified.
An inspection device and method that registers a provisional pattern in the reference image, searches for similar candidates, and compares similarity differences with a threshold to ensure the pattern's uniqueness before registering it as a registered pattern, thereby aligning the inspection image accurately with the reference image.
This approach effectively suppresses positional deviation between inspection and reference images, reducing false defect detection and enabling automated, accurate alignment without manual intervention.
Smart Images
Figure 2025116611000001_ABST
Abstract
Description
[Technical Field]
[0001] The present disclosure relates to an inspection device and an inspection method. [Background technology]
[0002] As disclosed in Patent Document 1, an inspection method is known in which an inspection image obtained by imaging the object to be inspected is compared with a reference image to inspect the object to be inspected. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Japanese Patent Application Publication No. 7-325046 Summary of the Invention [Problem to be solved by the invention]
[0004] In this type of inspection method, the object to be inspected is divided into a plurality of fields of view, and the imaging means is moved relative to the object to be inspected, sequentially imaging the object for each field of view using the imaging means, thereby obtaining an image for each field of view.
[0005] The captured image for each field of view is trimmed to obtain an inspection image for each field of view. The inspection image obtained for each field of view is compared with a reference image created in advance for each field of view. In this way, the object to be inspected is inspected, and for example, defects occurring in the object to be inspected are detected.
[0006] Here, if the positioning accuracy of the stage supporting the object to be inspected, the imaging means, etc. is poor, a positional deviation occurs between the inspection image and the reference image to be compared.
[0007] To prevent such misalignment from occurring, a registration pattern for alignment is registered in the reference image for each field of view, and the captured image (field of view) is trimmed using the position of the registration pattern in the reference image as a reference to obtain an inspection image. The obtained inspection image is then aligned with the reference image and compared with the reference image.
[0008] The registration pattern for alignment is selected from among a plurality of patterns formed on the object to be inspected, and is as unique as possible in order to avoid misidentification with other patterns.
[0009] However, even if a unique pattern is selected as a registration pattern for alignment in the reference image, the selected registration pattern may not actually be unique, and similar patterns may exist in addition to the registration pattern.
[0010] In this case, when the captured image (field of view) is trimmed to obtain an inspection image and the inspection image is aligned with the reference image, a similar pattern that is different from the registered pattern on the inspection object may be mistaken for the registered pattern. In this case, the captured image (field of view) is trimmed to obtain an inspection image based on an incorrect reference, and the inspection image is aligned with the reference image, resulting in a misalignment between the inspection image and the reference image.
[0011] Conventionally, an operator manually creates an inspection recipe by registering unique alignment patterns from the object to be inspected. However, when a large number of such alignment patterns must be registered, it becomes difficult to register them manually.
[0012] Therefore, we would like to simplify the process of creating an inspection recipe by registering a unique pattern for alignment from the object to be inspected.In particular, when automatically registering a unique pattern for alignment from the object to be inspected, we would like to automatically suppress misalignment between the inspection image and the reference image by registering a pattern that is more unique than when done manually.
[0013] The present disclosure has been made in consideration of such points, and its purpose is to suppress positional deviation between an inspection image and a reference image when inspecting an object to be inspected by comparing the inspection image obtained by imaging the object to a reference image. [Means for solving the problem]
[0014] An inspection apparatus according to the present disclosure is an inspection apparatus that inspects an object to be inspected by comparing an inspection image obtained by imaging the object to a reference image, wherein in the reference image, a pattern for alignment from among a plurality of patterns formed on the object to be inspected is registered as a registered pattern; in the inspection image, a pattern similar to the registered pattern from among the plurality of patterns formed on the object to be inspected is detected as a detected pattern; the inspection image is aligned with the reference image based on the position of the detected pattern in the inspection image and the position of the registered pattern in the reference image; in the reference image, a provisional registered pattern is provisionally registered from among the plurality of patterns formed on the object to be inspected; at least a first pattern candidate and a second pattern candidate similar to the provisional registered pattern are searched for; a similarity difference, which is the difference between the similarity between the provisional registered pattern and the first pattern candidate and the similarity between the provisional registered pattern and the second pattern candidate, is compared with a threshold; when the similarity difference is smaller than the threshold, the provisional registered pattern is not registered as the registered pattern; and when the similarity difference is greater than the threshold, the provisional registered pattern is registered as the registered pattern.
[0015] According to this configuration, a first pattern candidate and a second pattern candidate similar to the provisionally registered pattern are searched for in the reference image, and a similarity difference between the provisionally registered pattern and the first pattern candidate and the similarity between the provisionally registered pattern and the second pattern candidate is compared with a threshold value.
[0016] When the similarity difference is smaller than the threshold, the provisional registration pattern is not considered to be unique, and therefore the provisional registration pattern is not registered as a registered pattern. When the similarity difference is larger than the threshold, the provisional registration pattern is considered to be unique, and therefore the provisional registration pattern is registered as a registered pattern.
[0017] In this way, only when the provisional registration pattern is unique, the provisional registration pattern is registered as a registration pattern.
[0018] This makes it possible to prevent a pattern that is different from the registered pattern on the object being inspected from being mistaken for the registered pattern when the inspection image is aligned with the reference image, and also to prevent misalignment between the inspection image and the reference image.
[0019] As described above, when inspecting an object to be inspected by comparing an inspection image obtained by imaging the object to a reference image, it is possible to suppress positional deviation between the inspection image and the reference image.
[0020] In particular, this is advantageous for automatically registering a unique pattern for alignment from within an object to be inspected, without manual work.
[0021] In one embodiment, the provisional registration pattern includes at least a first provisional registration pattern and a second provisional registration pattern, and when the similarity difference for a first provisional registration pattern, which is the similarity difference in the first provisional registration pattern, is greater than the threshold value, the first provisional registration pattern is registered as the registration pattern, and when the similarity difference for the first provisional registration pattern is smaller than the threshold value and the similarity difference for a second provisional registration pattern, which is the similarity difference in the second provisional registration pattern, is greater than the threshold value, the second provisional registration pattern is registered as the registration pattern.
[0022] According to this configuration, if the first provisional registration pattern is unique, the first provisional registration pattern is registered as the registration pattern, and if the first provisional registration pattern is not unique and the second provisional registration pattern is unique, the second provisional registration pattern is registered as the registration pattern.
[0023] Even if the first provisional registration pattern is not unique, there is a possibility that the second provisional registration pattern will be registered as a registered pattern, so it is possible to reduce the possibility of a situation occurring in which no registered patterns are registered.
[0024] In one embodiment, the first provisional registration pattern and the second provisional registration pattern are ranked based on a score value associated with a reference registration pattern so that the score value in the first provisional registration pattern is higher than the score value in the second provisional registration pattern.
[0025] According to this configuration, it is possible to suitably set a ranking in which the first provisionally registered pattern has priority over the second provisionally registered pattern based on the score value associated with the reference registered pattern.
[0026] In one embodiment, the score value for the first provisional registration pattern and the score value for the second provisional registration pattern are determined by comparing the standard image showing the first provisional registration pattern and the second provisional registration pattern with a reference image showing the reference registration pattern.
[0027] According to this configuration, the score value of the first provisionally registered pattern and the score value of the second provisionally registered pattern can be easily obtained.
[0028] In one embodiment, when the standard image and the reference image are compared, a plurality of brightness change images with different brightnesses or a plurality of focus change images with different focuses are used as the standard image.
[0029] According to this configuration, by using a plurality of reference images under various conditions, it is possible to set a highly robust first provisional registration pattern and a highly robust second provisional registration pattern.
[0030] In one embodiment, the object to be inspected is a chip mounted on a substrate, and the chip includes one chip and another chip mounted on the substrate, and when the pattern similar to the registered pattern is detected as a detected pattern in the inspection image for one of the chips and the pattern similar to the registered pattern is not detected as a detected pattern in the inspection image for the other chips, the registration of the registered pattern is canceled.
[0031] According to this configuration, if a foreign object is erroneously registered as a registered pattern, this can be detected.
[0032] In one embodiment, before the temporary registration pattern is temporarily registered in the reference image, a filtering process is performed to remove high frequency components in the reference image.
[0033] According to this configuration, it is possible to prevent a repetitive pattern from being registered as a registered pattern.
[0034] In one embodiment, an intersection point is set between an area in the reference image where the component in a first direction is larger than others and an area where the component in a second direction perpendicular to the first direction is larger than others, and the intersection point in the reference image is provisionally registered as the provisional registration pattern.
[0035] According to this configuration, the temporary registration pattern can be made into a unique corner (corner portion).
[0036] an inspection method according to the present disclosure, which inspects an object to be inspected by comparing an inspection image obtained by imaging the object to a reference image, wherein in the reference image, a pattern for alignment from among a plurality of patterns formed on the object to be inspected is registered as a registered pattern; in the inspection image, a pattern similar to the registered pattern from among the plurality of patterns formed on the object to be inspected is detected as a detected pattern; the inspection image is aligned with the reference image based on the position of the detected pattern in the inspection image and the position of the registered pattern in the reference image; in the reference image, a provisional registered pattern is provisionally registered from among the plurality of patterns formed on the object to be inspected; at least a first pattern candidate and a second pattern candidate similar to the provisional registered pattern are searched for; a similarity difference, which is the difference between the similarity between the provisional registered pattern and the first pattern candidate and the similarity between the provisional registered pattern and the second pattern candidate, is compared with a threshold; if the similarity difference is smaller than the threshold, the provisional registered pattern is not registered as the registered pattern; and if the similarity difference is greater than the threshold, the provisional registered pattern is registered as the registered pattern. [Effects of the Invention]
[0037] According to the present disclosure, when inspecting an object to be inspected by comparing an inspection image obtained by imaging the object to a reference image, it is possible to suppress positional deviation between the inspection image and the reference image. [Brief explanation of the drawings]
[0038] [Figure 1] FIG. 1 shows an inspection apparatus that inspects an object by comparing an inspection image obtained by imaging the object with a reference image. [Figure 2] FIG. 2 shows an example of erroneous registration of a registration pattern. [Figure 3] FIG. 3 shows a method for registering a registration pattern. [Figure 4] FIG. 4 shows a comparison between the provisional registration pattern and the reference registration pattern. [Figure 5]FIG. 5 shows the chip mounted on the substrate. [Figure 6] FIG. 6 shows the high frequency filtering of the reference image. [Figure 7] FIG. 7 illustrates the concept of edge strength. [Figure 8] FIG. 8 shows a flow chart. DETAILED DESCRIPTION OF THE INVENTION
[0039]
[0023] An embodiment of the present disclosure will be described in detail below with reference to the accompanying drawings. The following description of the preferred embodiment is merely exemplary in nature and is not intended to limit the present disclosure, its application, or its uses.
[0040] (Inspection equipment) 1 shows an inspection device 1. The inspection device 1 inspects an object A to be inspected by comparing an inspection image J obtained by capturing an image of the object A with a reference image K. The inspection device 1 includes a camera (not shown) as an imaging unit and a computer (not shown) as a control unit. The control unit includes, for example, a processor mounted on a board and a memory device that stores software for operating the processor.
[0041] The object to be inspected A is, for example, a chip C. The chip C is provided on the surface of a substrate W (see FIG. 5). The substrate W is, for example, a semiconductor wafer. The chip C provided on the substrate W will be described later.
[0042] The inspection device 1 detects defects that occur in the inspection object A. Examples of defects include foreign matter, scratches, poor edge rinsing, through-hole abnormalities, uneven coating, peeling, scratches, bonding pad abnormalities, color unevenness, pattern abnormalities, stains, discoloration, and deformation.
[0043] In the inspection device 1, the object to be inspected A is divided into a plurality of fields of view Fj, and the camera is moved relative to the object to be inspected A, and images of the object to be inspected A are sequentially captured by the camera for each field of view Fj, thereby obtaining captured images for each field of view Fj.
[0044] The outer periphery of the captured image for each field of view Fj is trimmed to obtain an inspection image J for each field of view Fj. The inspection image J obtained for each field of view Fj is compared with a reference image K.
[0045] The reference image K is also called a fitting image or a statistically acceptable image. The reference image K is obtained, for example, by averaging images of the inspection object A that has been determined to be acceptable. The reference image K is set in advance. The reference image K is created in advance for each field of view Fk of the camera using the same method as the inspection image J.
[0046] The field of view Fj for obtaining the inspection image J (hereinafter referred to as "inspection field of view Fj") and the field of view Fk for obtaining the reference image K (hereinafter referred to as "reference field of view Fk") are the same size and shape (rectangle in this example).
[0047] The object A to be inspected is inspected by comparing the inspection image J with the reference image K. For example, defects occurring in the object A to be inspected are detected.
[0048] However, if the positioning accuracy of the stage supporting the object A, the camera, etc. is poor, a misalignment occurs between the inspection field Fj and the reference field Fk, which in turn causes a misalignment between the inspection image J and the reference image K. In the part where such a misalignment occurs, false defects (excessive defects) may be detected.
[0049] To prevent such misalignment from occurring, a pattern P for alignment is pre-registered as a registered pattern Pk from among the multiple patterns P formed on the object A for each reference visual field Fk in the reference image K. Then, a pattern P similar to the registered pattern Pk is detected as a detected pattern Pj from among the multiple patterns P formed on the object A for each inspection visual field Fj in the inspection image J.
[0050] In Figure 1, in the inspection field Fj and the reference field Fk, the gray areas represent the base material of the object A to be inspected (plain, for example, the surface of a chip on a semiconductor substrate), and the white or black areas represent areas that may contain some kind of pattern P.
[0051] The pattern P may be, for example, a wiring pattern or a geometric pattern of an integrated circuit.
[0052] The detection pattern Pj in the object to be inspected A is detected in a pattern matching search range D, which is narrower than the inspection field of view Fj and the inspection image J, but wider than the detection pattern Pj. In other words, in the object to be inspected A, the detection pattern Pj is detected in the pattern matching search range D.
[0053] Based on (using as a reference) the position Lj of the detection pattern Pj in the inspection image J (inspection visual field Fj) and the position Lk of the registered pattern Pk in the reference image K (reference visual field Fk), the inspection image J is aligned (overlaid by translation) with the reference image K. In other words, the inspection image J is aligned with the reference image K based on the amount of alignment of the detection pattern Pj with respect to the registered pattern Pk. This alignment method is called template matching.
[0054] Specifically, the outer periphery of the inspection field of view (captured image) Fj is trimmed to obtain an inspection image J, and the obtained inspection image J is aligned with a reference image K. Then, the inspection image J is compared with the reference image K. In this way, the object to be inspected A is inspected.
[0055] 1, an example of the position Lj of the detection pattern Pj in the inspection image J (inspection visual field Fj) is the distance between the lower right corner of the inspection image J (inspection visual field Fj) and the lower right corner of the detection pattern Pj. Similarly, an example of the position Lk of the registration pattern Pk in the reference image K (reference visual field Fk) is the distance between the lower right corner of the reference image K (reference visual field Fk) and the lower right corner of the registration pattern Pk.
[0056] The registration pattern Pk for alignment is selected from among a plurality of patterns P formed on the object A to be inspected, and is as unique as possible in order to avoid misidentification with other patterns P.
[0057] However, even if a unique pattern P is selected as the registered pattern Pk for alignment in the reference image K, the selected registered pattern Pk may not actually be unique, and similar patterns P may exist in addition to the registered pattern Pk.
[0058] FIG. 2 shows an example of erroneous registration of a registered pattern Pk. For simplicity, FIG. 2 shows an inspection object A different from that shown in FIG. 1. As shown in the left diagram of FIG. 2, a pattern P similar to the registered pattern Pk (hereinafter referred to as a "similar pattern Pk'") may exist near the registered pattern Pk in the reference image K. For example, in the reference image K, both the registered pattern Pk and the similar pattern Pk' may exist within the pattern matching search range D.
[0059] In this case, when the inspection field (captured image) Fj is trimmed to obtain an inspection image J and the inspection image J is aligned with the reference image K, a similar pattern Pk' that is different from the registered pattern Pk in the object to be inspected A may be mistakenly recognized as the registered pattern Pk. In this case, the inspection field (captured image) Fj is trimmed to obtain the inspection image J and the inspection image J is aligned with the reference image K based on an incorrect reference, resulting in a misalignment between the inspection image J and the reference image K. Such a misalignment may cause the detection of false defects.
[0060] In this embodiment, when inspecting the object A by comparing the inspection image J obtained by imaging the object A with the reference image K, the positional deviation between the inspection image J and the reference image K is suppressed by using the following techniques.
[0061] (Registering registered patterns) 3 shows a method for registering a registration pattern Pk. In a reference image K, a temporary registration pattern Q is temporarily registered from among a plurality of patterns P formed on an object to be inspected A. The temporary registration pattern Q includes at least a first temporary registration pattern Qa and a second temporary registration pattern Qb.
[0062] In addition to the first provisional registration pattern Qa and the second provisional registration pattern Qb, there are multiple provisional registration patterns Q such as the third provisional registration pattern, the fourth provisional registration pattern, etc. (for example, about 10 in total), but for simplicity, only the first provisional registration pattern Qa and the second provisional registration pattern Qb are shown as examples.
[0063] In this example, the first temporary registration pattern Qa is, for example, rectangular, and the second temporary registration pattern Qb is, for example, diamond-shaped.
[0064] The first provisional registration pattern Qa and the second provisional registration pattern Qb are ranked based on the score value E associated with the reference registration pattern R so that the first score value Ea as the score value E in the first provisional registration pattern Qa is higher than the second score value Eb as the score value E in the second provisional registration pattern Qb.
[0065] The reference registered pattern R is an ideal and unique pattern P, and in this example, has the shape of a corner (angle portion). The reference registered pattern R is set in advance as an ideal and unique pattern P.
[0066] For detecting (searching for) the shape of a corner, known algorithms such as Harris (Harris Corner Detector) and FAST (Features from Accelerated Segment Test) can be suitably applied.
[0067] The score value E indicates the similarity of the provisional registration pattern Q to the reference registration pattern R. In this example, the rectangular first provisional registration pattern Qa is more similar to the corner-shaped reference registration pattern R than the diamond-shaped second provisional registration pattern Qb. Therefore, the first score value Ea of the first provisional registration pattern Qa is higher than the second score value Eb of the second provisional registration pattern Qb.
[0068] The score value E is calculated based on a comparison between the luminance of the provisional registration pattern Q and the luminance of the reference registration pattern R, for example.
[0069] Additionally, a total of about 10 patterns are selected in descending order of score value E, such as a third score value of a third provisionally registered pattern, a fourth score value of a fourth provisionally registered pattern, . . .
[0070] In the reference image K, at least a first pattern candidate T1 and a second pattern candidate T2 similar to the provisionally registered pattern Q are searched for. In addition to the first pattern candidate T1 and the second pattern candidate T2, a third pattern candidate, a fourth pattern candidate, and so on may also be searched for.
[0071] The search for the first pattern candidate T1 and the second pattern candidate T2 in the reference image K is performed in a predetermined area (for example, an area corresponding to the pattern matching search range D).
[0072] In the reference image K, the similarity difference ΔF (=F1-F2), which is the difference between the first similarity F1, which is the similarity between the provisionally registered pattern Q and the first pattern candidate T1, and the second similarity F2, which is the similarity between the provisionally registered pattern Q and the second pattern candidate T2, is compared with the threshold value G.
[0073] The larger the similarity difference ΔF, the more unique the provisionally registered pattern Q. The smaller the similarity difference ΔF, the less unique the provisionally registered pattern Q.
[0074] In the reference image K, when the similarity difference ΔF is smaller than the threshold value G (ΔF < G), the provisional registration pattern Q is not registered as the registration pattern Pk, and when the similarity difference ΔF is larger than the threshold value G (ΔF > G), the provisional registration pattern Q is registered as the registration pattern Pk.
[0075] In the reference image K, when the similarity difference ΔF matches the threshold value G (ΔF = G), whether the provisional registration pattern Q is registered as the registration pattern Pk is appropriately determined according to the user's settings.
[0076] Specifically, in the reference image K, the first pattern candidate Ta1 for the first provisional registration pattern as the first pattern candidate T1 similar to the first provisional registration pattern Qa, and the second pattern candidate Ta2 for the first provisional registration pattern as the second pattern candidate T2 are searched for.
[0077] Similarly, in the reference image K, the first pattern candidate Tb1 for the second provisional registration pattern as the first pattern candidate T1 similar to the second provisional registration pattern Qb, and the second pattern candidate Tb2 for the second provisional registration pattern as the second pattern candidate T2 are searched for.
[0078] In the reference image K, the first similarity difference ΔFa (= Fa1 - Fa2) for the first provisional registration pattern, which is the difference between the first similarity Fa1 for the first provisional registration pattern, which is the similarity between the first provisional registration pattern Qa and the first pattern candidate Ta1 for the first provisional registration pattern, and the second similarity Fa2 for the first provisional registration pattern, which is the similarity between the first provisional registration pattern Qa and the second pattern candidate Ta2 for the first provisional registration pattern, is compared with the threshold value G. The similarity difference ΔFa for the first provisional registration pattern is the similarity difference ΔF in the first provisional registration pattern Qa.
[0079] Similarly, in the reference image K, a similarity difference ΔFb for the second provisional registration pattern (= Fb1 - Fb2), which is the difference between a first similarity Fb1 for the second provisional registration pattern, which is the similarity between the second provisional registration pattern Qb and the first pattern candidate Tb1 for the second provisional registration pattern, and a second similarity Fb2 for the second provisional registration pattern, which is the similarity between the second provisional registration pattern Qb and the second pattern candidate Tb2 for the second provisional registration pattern, is compared with a threshold value G. The similarity difference ΔFb for the second provisional registration pattern is the similarity difference ΔF in the second provisional registration pattern Qb.
[0080] In the reference image K, when the similarity difference ΔFa for the first provisional registration pattern is greater than the threshold value G (ΔFa > G), the first provisional registration pattern Qa is registered as the registration pattern Pk.
[0081] In the reference image K, when the similarity difference ΔFa for the first provisional registration pattern is smaller than the threshold value G (ΔFa < G) and the similarity difference ΔFb for the second provisional registration pattern is greater than the threshold value G (ΔFb > G), the second provisional registration pattern Qb is registered as the registration pattern Pk.
[0082] In the reference image K, when the similarity difference ΔFb for the second provisional registration pattern is smaller than the threshold value G (ΔFb < G), the second provisional registration pattern Qb is not registered as the registration pattern Pk. In this case, in provisional registration patterns Q with lower score values E, such as the third provisional registration pattern, the fourth provisional registration pattern, etc., a comparison is made between the similarity difference ΔF and the threshold value G, and it is determined whether or not to be registered as the registration pattern Pk.
[0083] In the reference image K, when the similarity difference ΔFa for the first provisional registration pattern and / or the similarity difference ΔFb for the second provisional registration pattern matches the threshold value G (ΔFa = G, ΔFb = G), whether or not the first provisional registration pattern Qa and / or the second provisional registration pattern Qb is registered as the registration pattern Pk is appropriately determined according to the user's settings.
[0084] In this example, the first similarity Fa1 for the first provisional registration pattern is 0.99, the second similarity Fa2 for the first provisional registration pattern is 0.95, the first similarity Fb1 for the second provisional registration pattern is 0.99, the second similarity Fb2 for the second provisional registration pattern is 0.70, and the threshold G is 0.15. In this case, the similarity difference ΔFa for the first provisional registration pattern is 0.04, and the similarity difference ΔFb for the second provisional registration pattern is 0.29.
[0085] The similarity is calculated by various known methods, such as a method of comparing the luminance of two patterns to be compared.
[0086] 4 shows a comparison between the temporary registration pattern Q and the reference registration pattern R. By comparing the standard image K, which shows the first temporary registration pattern Qa and the second temporary registration pattern Qb, with the reference image M, which shows the reference registration pattern R, the first score E1 value of the first temporary registration pattern Qa and the second score E2 value of the second temporary registration pattern Qb are obtained.
[0087] As described above, the reference registered pattern R is an ideal and unique pattern P, and has a corner shape. The reference image M in which the reference registered pattern R is projected is set in advance.
[0088] When comparing the standard image K with the reference image M, a plurality of brightness-change images N1 with different brightness or a plurality of focus-change images N2 with different focus are used as the standard image K. In the plurality of brightness-change images N1, the brightness of the standard image K is changed in stages. In the plurality of focus-change images N2, the focus of the standard image K is changed in stages (from clear to blurred).
[0089] For convenience of illustration, the same image is used as the plurality of brightness change images N1 and the plurality of focus change images N2 in FIG.
[0090] (Presence of foreign matter) As shown in the middle diagram of Figure 2, when a foreign substance Z is present in the inspection object A, the foreign substance Z may be erroneously registered as a registered pattern Pk in the reference image K. When the foreign substance Z is erroneously registered as a registered pattern Pk in the reference image K, some kind of countermeasure must be taken.
[0091] FIG. 5 shows a chip C provided on a substrate W. In this example, the object to be inspected A is the chip C. A plurality of chips C are provided on the surface of the substrate W. For simplicity, FIG. 5 illustrates an example in which a first chip C1 to a ninth chip C9 are provided on the substrate W in a 3×3 matrix.
[0092] The plurality of chips C1 to C9 are arranged at equal intervals on the substrate W. The shapes and arrangements of the patterns P formed on the chips C1 to C9 should all be the same unless a foreign substance Z is present.
[0093] The chips C include one chip Ca and other chips Cb provided on the substrate W. In this example, the one chip Ca is the fifth chip C5. The other chips Cb are the second chip C2, the fourth chip C4, the sixth chip C6, and the eighth chip C8. The other chips Cb are arranged adjacent to the one chip Ca in the vertical and horizontal directions.
[0094] The inspection image J for one chip Ca and the inspection image J for another chip Cb are different from each other, whereas the reference image K for one chip Ca and the reference image K for another chip Cb are the same (common).
[0095] When a pattern P similar to a registered pattern Pk is detected as a detected pattern Pj in an inspection image J for one chip Ca, and a pattern P similar to the registered pattern Pk is not detected as a detected pattern Pj in an inspection image J for another chip Cb, the registration of the registered pattern Pk is cancelled.
[0096] In this case, it is determined that the foreign substance Z has been erroneously registered as the registered pattern Pk in the reference image K.
[0097] As described above, the chip C (inspection object A) is divided into multiple fields of view. Each field of view corresponds to an inspection image J. A comparison between one chip Ca and another chip Cb is performed between the same fields of view (inspection images J).
[0098] (high frequency filtering) As shown in the right diagram of Figure 2, when an object to be inspected A contains many repeated patterns P, all of these repeated patterns P become candidates for the registered pattern Pk, making it difficult to register the registered pattern Pk. Ideally, we would like to register a unique pattern P formed at a corner as the registered pattern Pk.
[0099] 6 shows high frequency filtering of a reference image K. Before a temporary registration pattern Q is temporarily registered in the reference image K, a filter process is performed to remove high frequency components in the reference image K. The filter process to remove high frequency components is performed using a known high frequency filter.
[0100] Specifically, first, the reference image K is subjected to an FFT (fast Fourier transform). Then, high-frequency components in the reference image K are removed. Then, the reference image K is restored to its original state by an inverse FFT. This removes the high-frequency components in the reference image K. In other words, the repeated patterns P in the reference image K are removed, leaving only the unique patterns P in the corners.
[0101] (Edge Strength) FIG. 7 illustrates the concept of edge strength. The horizontal line x extending left and right, indicated by the thick line in the left diagram of FIG. 7, is an area in the reference image K where the component in the horizontal direction X, which is the first direction, is greater than in other areas. The horizontal line x has the highest edge strength in the horizontal direction X compared to other areas. In other words, in the reference image K, the variation in brightness on both sides of the horizontal line x (when straddling the horizontal line x vertically) is the highest compared to other areas.
[0102] The vertical line y shown by the thick line on the right side of FIG. 7 is an area in the reference image K where the component in the vertical direction Y as the second direction is larger than in other areas. The vertical direction Y (second direction) is perpendicular to the left-right direction (first direction) X. The vertical line y has the highest edge strength in the vertical direction Y compared to other areas. In other words, in the reference image K, the variation in brightness on both sides of the vertical line y (when straddling the vertical line y from left to right) is the highest compared to other areas.
[0103] In the reference image K, an intersection U is set between a horizontal line x whose horizontal component in the horizontal direction X is larger than the others (highest edge strength in the horizontal direction X) and an upper / lower line y whose vertical component in the vertical direction Y is larger than the others (highest edge strength in the vertical direction Y). The intersection U corresponds to a unique pattern P formed at a corner.
[0104] An intersection U is set between an area in the reference image K where the component in the left-right direction X is larger than others (the left-right line x where the edge strength in the left-right direction X is the highest) and an area in the reference image K where the component in the up-down direction Y is larger than others (the up-down line y where the edge strength in the up-down direction Y is the highest). Then, the intersection U in the reference image K is provisionally registered as a provisional registration pattern Q.
[0105] If there are multiple intersection points U, the first intersection point U is provisionally registered as a first provisional registration pattern Qa, the second intersection point U is provisionally registered as a second provisional registration pattern Qb, the third intersection point U is provisionally registered as a third provisional registration pattern, and so on.
[0106] (flowchart) 8 shows a flowchart. Starting from the start, in a first step S1, high frequency components are removed from the reference image K. As a result, repetitive patterns P in the reference image K are removed.
[0107] In the second step S2, an intersection U is set between an area in the reference image K where the component in the left-right direction X is larger than others (the left-right line x where the edge strength in the left-right direction X is the highest) and an area where the component in the up-down direction Y is larger than others (the up-down line y where the edge strength in the up-down direction Y is the highest).
[0108] In a third step S3, an intersection U from among a plurality of patterns P formed on the object A to be inspected in the reference image K is provisionally registered as a provisional registration pattern Q.
[0109] Specifically, if there are multiple intersection points U, the first intersection point U in the reference image K is temporarily registered as a first temporary registration pattern Qa, and the second intersection point U in the reference image K is temporarily registered as a second temporary registration pattern Qb.
[0110] In the fourth step S4, the first provisional registration pattern Qa and the second provisional registration pattern Qb are ranked based on the score value E associated with the reference registration pattern R so that the first score value Ea of the first provisional registration pattern Qa is higher than the second score value Eb of the second provisional registration pattern Qb.
[0111] In a fifth step S5, the reference image K is searched for a first pattern candidate T1 and a second pattern candidate T2 similar to the temporary registration pattern Q. Specifically, the reference image K is searched for a first pattern candidate Ta1 for the first temporary registration pattern and a second pattern candidate Ta2 for the first temporary registration pattern similar to the first temporary registration pattern Qa. Similarly, the reference image K is searched for a second pattern candidate Tb1 for the second temporary registration pattern and a second pattern candidate Tb2 for the second temporary registration pattern similar to the second temporary registration pattern Qb.
[0112] In a sixth step S6, in the reference image K, a similarity difference ΔF between the first similarity F1 between the provisionally registered pattern Q and the first pattern candidate T1 and the second similarity F2 between the provisionally registered pattern Q and the second pattern candidate T2 is found (calculated).
[0113] Specifically, in the reference image K, a similarity difference ΔFa for the first provisional registration pattern is calculated between a first similarity Fa1 for the first provisional registration pattern between the first provisional registration pattern Qa and a first pattern candidate Ta1 for the first provisional registration pattern, and a second similarity Fa2 for the first provisional registration pattern between the first provisional registration pattern Qa and a second pattern candidate Ta2 for the first provisional registration pattern.
[0114] Similarly, in the reference image K, a similarity difference ΔFb for the second provisional registration pattern is obtained between a first similarity Fb1 for the second provisional registration pattern Qb and the first pattern candidate Tb1 for the second provisional registration pattern, and a second similarity Fb2 for the second provisional registration pattern Qb and the second pattern candidate Tb2 for the second provisional registration pattern.
[0115] In the seventh step S7, the similarity difference ΔF is compared with the threshold value G. Specifically, the similarity difference ΔFa for the first provisional registration pattern is compared with the threshold value G. Similarly, the similarity difference ΔFb for the second provisional registration pattern is compared with the threshold value G (it may be compared for the first time in the ninth step S9).
[0116] When the similarity difference ΔF is smaller than the threshold value G (ΔF < G), the provisional registration pattern Q is not registered as the registration pattern Pk. When the similarity difference ΔF is larger than the threshold value G (ΔF > G), the provisional registration pattern Q is registered as the registration pattern Pk.
[0117] Specifically, when the similarity difference ΔFa for the first provisional registration pattern is larger than the threshold value G (ΔFa > G) in the seventh step S7, the process proceeds to the eighth step S8.
[0118] In the eighth step S8, the first provisional registration pattern Qa is registered as the registration pattern Pk. Then, the process proceeds to the twelfth step.
[0119] When the similarity difference ΔFa for the first provisional registration pattern is smaller than the threshold value G (ΔFa < G) in the eighth step S8, the process proceeds to the ninth step S9.
[0120] In the ninth step S9, the first provisional registration pattern Qa is not registered as the registration pattern Pk. When the similarity difference ΔFb for the second provisional registration pattern is larger than the threshold value G (ΔFb > G) in the ninth step S9, that is, when the similarity difference ΔFa for the first provisional registration pattern is smaller than the threshold value G (ΔFa < G) and (AND) the similarity difference ΔFb for the second provisional registration pattern is larger than the threshold value G (ΔFb > G), the process proceeds to the tenth step S10.
[0121] In the 10th step S10, the second provisional registration pattern Qb is registered as the registration pattern Pk. Then, it proceeds to the 12th step.
[0122] When the similarity difference ΔFb for the second provisional registration pattern is smaller than the threshold value G (ΔFb < G) in the 9th step S9, it proceeds to the 11th step S11.
[0123] In the 11th step S11, the second provisional registration pattern Qb is not registered as the registration pattern Pk. In this case, for provisional registration patterns Q with lower score values E such as the third provisional registration pattern, the fourth provisional registration pattern, etc., a comparison is made between the similarity difference ΔF and the threshold value G, and it is determined whether to register it as the registration pattern Pk. In the 11th step S11, first, it is determined whether the similarity difference for the third provisional registration pattern is larger than the threshold value G.
[0124] In the 12th step S12, while relatively moving the camera with respect to the inspection target A (chip C provided on the substrate W), the inspection target A is divided into a plurality of visual fields Fj, and an inspection image J is acquired for each visual field Fj. Then, in the inspection image J for each visual field Fj, a pattern P similar to the registration pattern Pk is detected from among the plurality of patterns P formed on the inspection target A as the detection pattern Pj.
[0125] In the 13th step S13, when a pattern P similar to the registration pattern Pk is detected as the detection pattern Pj in the inspection image J related to one chip Ca and a pattern P similar to the registration pattern Pk is not detected as the detection pattern Pj in the inspection image J related to another chip Cb, it proceeds to the 14th step S14. Otherwise, it proceeds to the 15th step S15.
[0126] In the 14th step S14, the registration of the registration pattern Pk in the reference image K is cancelled. Then, it reaches the end.
[0127] In the 15th step S15, the registration of the registration pattern Pk in the reference image K is maintained. Then, it proceeds to the 16th step S16.
[0128] In a sixteenth step S16, the inspection image J is aligned with the reference image K based on (the amount of misalignment between) the position Lj of the detection pattern Pj in the inspection image J and the position Lk of the registered pattern Pk in the reference image K. Then, the process ends.
[0129] (Action and effect) According to this embodiment, a first pattern candidate T1 and a second pattern candidate T2 similar to a provisionally registered pattern Q are detected in a reference image K. A similarity difference ΔF between a first similarity Fa between the provisionally registered pattern Q and the first pattern candidate T1 and a second similarity Fb between the provisionally registered pattern Q and the second pattern candidate T2 is compared with a threshold G.
[0130] When the similarity difference ΔF is smaller than the threshold G, it can be said that the provisionally registered pattern Q is not unique, and therefore the provisionally registered pattern Q is not registered as the registered pattern Pk. When the similarity difference ΔF is larger than the threshold G, it can be said that the provisionally registered pattern Q is unique, and therefore the provisionally registered pattern Q is registered as the registered pattern Pk.
[0131] In this way, only when the provisional registration pattern Q is unique, the provisional registration pattern Q is registered as the registered pattern Pk.
[0132] This makes it possible to prevent a similar pattern Pk′, which is different from the registered pattern Pk in the object to be inspected A, from being mistakenly recognized as the registered pattern Pk when the inspection field (captured image) Fj is trimmed to obtain the inspection image J and the inspection image J is aligned with the reference image K. Also, it is possible to prevent misalignment between the inspection image J and the reference image K.
[0133] As described above, when inspecting the object A by comparing the inspection image J obtained by imaging the object A with the reference image K, the positional deviation between the inspection image J and the reference image K can be suppressed.
[0134] If the first provisional registration pattern Qa is unique, the first provisional registration pattern Qa is registered as the registered pattern Pk, and if the first provisional registration pattern Qa is not unique and the second provisional registration pattern Qb is unique, the second provisional registration pattern Qb is registered as the registered pattern Pk.
[0135] Even if the first provisional registration pattern Qa is not unique, there is a possibility that the second provisional registration pattern Qb will be registered as the registered pattern Pk, so it is possible to reduce the possibility of a situation occurring in which no registered pattern Pk is registered.
[0136] Based on the score value E associated with the reference registration pattern R, the ranking can be preferably set such that the first provisional registration pattern Qa has priority over the second provisional registration pattern Qb.
[0137] By comparing a reference image K showing the first provisional registration pattern Qa and the second provisional registration pattern Qb with a reference image M showing the reference registration pattern R, the first score value Ea of the first provisional registration pattern Qa and the second score value Eb of the second provisional registration pattern Qa can be easily determined.
[0138] By using a plurality of reference images (a plurality of brightness change images or a plurality of focus change images) under various conditions, it is possible to set a highly robust first temporary registration pattern Qa and a second temporary registration pattern Qb.
[0139] When a pattern P similar to the registered pattern Pk is detected as a detected pattern Pj in the inspection image J of one chip Ca, and a pattern P similar to the registered pattern Pk is not detected as a detected pattern Pj in the inspection image J of another chip Cb, the registration of the registered pattern Pk is cancelled. This allows detection of an erroneous registration of a foreign substance Z as the registered pattern Pk, and allows the registration of the registered pattern Pk to be redone.
[0140] The high frequency filter can prevent the repetitive pattern P from being registered as the registered pattern Pk.
[0141] By setting the intersection point U based on the edge strength, the provisional registration pattern Q can be made into a unique corner.
[0142] When a chip C (inspection object A) is divided into multiple fields of view (for example, 100 fields of view per chip C), a registered pattern Pk must be registered in the reference image K for each field of view. It is difficult for a person to manually register a registered pattern Pk in the reference image K for each field of view, as has been done in the past. There is a demand for mechanizing the task of registering a registered pattern Pk in the reference image K for each field of view, and applying the inspection device 1 according to this embodiment would be effective.
[0143] Specifically, conventionally, an operator manually creates an inspection recipe by registering a unique pattern P for alignment as a registered pattern Pk from the object to be inspected A. However, when a large number of such registered patterns Pk for alignment need to be registered, it becomes difficult to register them manually.
[0144] Therefore, there has been a demand for simplifying the work of creating an inspection recipe by registering a unique pattern P for alignment from the object to be inspected A as a registered pattern Pk. In particular, when automatically registering a unique pattern P for alignment from the object to be inspected A as a registered pattern Pk, it has been desired to automatically suppress misalignment between the inspection image J and the reference image K by registering a pattern P that is more unique than when done manually as the registered pattern Pk.
[0145] This embodiment is advantageous in that a unique pattern P for alignment from the inspection object A can be automatically registered as a registered pattern Pk without manual work, and misalignment between the inspection image J and the reference image K can be automatically suppressed.
[0146] (Other embodiments) Although the present disclosure has been described above with reference to preferred embodiments, such description is not limiting, and it goes without saying that various modifications, substitutions, or combinations are possible.
[0147] There may be only one provisional registration pattern Q. In this case, it is determined whether or not the one provisional registration pattern Q should be registered as the registration pattern Pk.
[0148] In Figure 1, the position Lj of the detection pattern Pj in the inspection image J (inspection field of view Fj) and the position Lk of the registered pattern Pk in the reference image K (reference field of view Fk) are illustrated as the distance between the bottom right corners, but this is not limited to this and may also be expressed, for example, as the distance between other corners, such as the top left corners, or the distance between the centers.
[0149] The inspection visual field Fj (inspection image J) and the reference visual field Fk (reference image K) are not limited to a rectangle, and may be, for example, a polygon other than a rectangle, or a circle other than a polygon.
[0150] The provisional registration pattern Q is not limited to the above, and various modes are possible (it is preferable that the pattern is unique).
[0151] The selection of one chip Ca and the other chip Cb may be made arbitrarily.
[0152] The object to be inspected A is not limited to a chip C provided on a substrate W. The substrate W is not limited to a semiconductor wafer, and may be, for example, glass, metal, resin, etc. The object to be inspected A does not have to be related to the chip C on the substrate W.
[0153] The inspection method according to the present disclosure inspects the object A to be inspected by comparing an inspection image J obtained by imaging the object A to a reference image K. In the reference image K, a pattern P for alignment is registered as a registered pattern Pk from among multiple patterns P formed on the object A to be inspected. In the inspection image J, a pattern P similar to the registered pattern Pk is detected as a detected pattern Pj from among the multiple patterns P formed on the object A to be inspected. The inspection image J is aligned with the reference image K based on the position Lj of the detected pattern Pj in the inspection image J and the position Lk of the registered pattern Pk in the reference image K.
[0154] In the reference image K, a provisional registration pattern Q is provisionally registered from among multiple patterns P formed on the object to be inspected A, and at least a first pattern candidate T1 and a second pattern candidate T2 similar to the provisional registration pattern Q are searched for. The similarity difference ΔF, which is the difference between the first similarity Fa between the provisional registration pattern Q and the first pattern candidate T1 and the second similarity Fb between the provisional registration pattern Q and the second pattern candidate T2, is compared with a threshold value G. If the similarity difference ΔF is smaller than the threshold value G, the provisional registration pattern Q is not registered as a registered pattern Pk, and if the similarity difference ΔF is larger than the threshold value G, the provisional registration pattern Q is registered as a registered pattern Pk. [Industrial Applicability]
[0155] The present disclosure is applicable to an inspection device and an inspection method, and is therefore extremely useful and has high industrial applicability. [Explanation of symbols]
[0156] 1. Inspection equipment A. Subject to inspection W substrate C-chip Ca chip Cb other chips Fj inspection field (field of view) Fk reference field of view (field of view) J Inspection image K reference image P pattern Pj detection pattern Pk Registration Pattern Pk' similar pattern D Pattern Matching Search Range Lj position Lk position Q Provisional registration pattern Qa First provisional registration pattern Qb 2nd provisional registration pattern R Reference Registration Pattern E-score value Ea 1st score value Eb Second Score Value T1 First pattern candidate T2 Second pattern candidate Ta1 First pattern candidate for the first provisional registration pattern Ta2: Second pattern candidate for the first provisional registration pattern Tb1: First pattern candidate for the second provisional registration pattern Tb2: Second pattern candidate for second provisional registration pattern F1 1st similarity (similarity) F2 Second similarity (similarity) Fa1 First similarity for the first provisionally registered pattern (similarity) Fa2 Second similarity for the first provisionally registered pattern (similarity) Fb1 First similarity for second provisionally registered pattern (similarity) Fb2 Second similarity for second provisionally registered pattern (similarity) ΔF similarity difference ΔFa Similarity difference for first provisional registration pattern ΔFb Similarity difference for second provisional registration pattern G threshold M Reference Image N1 Brightness change image N2 focus change image Z Foreign object x left and right lines X Left / right direction (first direction) y up / down line Y vertical direction (second direction) U intersection
Claims
1. An inspection apparatus that inspects an object to be inspected by comparing an inspection image obtained by imaging the object to a reference image, In the reference image, the pattern for alignment is registered as a registered pattern from among a plurality of patterns formed on the object to be inspected; In the inspection image, a pattern similar to the registered pattern is detected as a detection pattern from among a plurality of patterns formed on the object to be inspected; the inspection image is aligned with the reference image based on the position of the detection pattern in the inspection image and the position of the registration pattern in the reference image; an inspection device in which, in the reference image, a provisional registration pattern is provisionally registered from among the plurality of patterns formed on the object to be inspected; at least a first pattern candidate and a second pattern candidate similar to the provisional registration pattern are searched for; a similarity difference, which is the difference between the similarity between the provisional registration pattern and the first pattern candidate and the similarity between the provisional registration pattern and the second pattern candidate, is compared with a threshold; when the similarity difference is smaller than the threshold, the provisional registration pattern is not registered as the registration pattern; and when the similarity difference is larger than the threshold, the provisional registration pattern is registered as the registration pattern.
2. the provisionally registered patterns include at least a first provisionally registered pattern and a second provisionally registered pattern, when a first provisional registration pattern similarity difference, which is the similarity difference in the first provisional registration pattern, is greater than the threshold value, the first provisional registration pattern is registered as the registration pattern, 2. The inspection device according to claim 1, wherein when the similarity difference for the first provisional registration pattern is smaller than the threshold value and the similarity difference for the second provisional registration pattern, which is the similarity difference in the second provisional registration pattern, is larger than the threshold value, the second provisional registration pattern is registered as the registration pattern.
3. 3. The inspection device according to claim 2, wherein the first provisional registration pattern and the second provisional registration pattern are ranked based on a score value associated with a reference registration pattern so that the score value of the first provisional registration pattern is higher than the score value of the second provisional registration pattern.
4. 4. The inspection device according to claim 3, wherein the score value for the first provisional registration pattern and the score value for the second provisional registration pattern are determined by comparing the standard image in which the first provisional registration pattern and the second provisional registration pattern are displayed with a reference image in which the reference registration pattern is displayed.
5. 5. The inspection device according to claim 4, wherein when the standard image and the reference image are compared, a plurality of brightness change images with different brightnesses or a plurality of focus change images with different focuses are used as the standard image.
6. the object to be inspected is a chip provided on a substrate, the chips include one chip and another chip provided on the substrate, 6. The inspection device according to claim 1, wherein when a pattern similar to the registered pattern is detected as a detected pattern in the inspection image for one of the chips and a pattern similar to the registered pattern is not detected as a detected pattern in the inspection image for another of the chips, the registration of the registered pattern is cancelled.
7. 6. The inspection device according to claim 1, wherein a filtering process is performed to remove high frequency components in the reference image before the provisional registration pattern is provisionally registered in the reference image.
8. an intersection point between an area in the reference image where a component in a first direction is larger than other areas and an area in the reference image where a component in a second direction orthogonal to the first direction is larger than other areas is set; The inspection device according to claim 1 , wherein the intersections in the reference image are temporarily registered as a temporarily registered pattern.
9. An inspection method for inspecting an object to be inspected by comparing an inspection image obtained by imaging the object to a reference image, comprising: In the reference image, the pattern for alignment is registered as a registration pattern from among a plurality of patterns formed on the object to be inspected; In the inspection image, a pattern similar to the registered pattern is detected as a detection pattern from among a plurality of patterns formed on the object to be inspected; aligning the inspection image with the reference image based on the position of the detection pattern in the inspection image and the position of the registration pattern in the reference image; an inspection method comprising: provisionally registering, in the reference image, a provisional registration pattern from among the plurality of patterns formed on the object to be inspected; searching for at least a first pattern candidate and a second pattern candidate similar to the provisional registration pattern; comparing a similarity difference, which is a difference between a similarity between the provisional registration pattern and the first pattern candidate and a similarity between the provisional registration pattern and the second pattern candidate, with a threshold; and not registering the provisional registration pattern as the registered pattern when the similarity difference is smaller than the threshold;
Citation Information
Patent Citations
Appearance inspection method and appearance inspection apparatus
JP1995325046A