Defection classification device

Abstract

To classify a pattern of defection generated in an inspected object with high accuracy.SOLUTION: A defect classification device 1 comprises a classification part 10 that classifies a pattern of a defect X generated in an inspected object W imaged so as to be divided into a plurality of visual fields F. The plurality of visual fields contain a first visual field F1 and a second visual field F2 that are adjacent each other. A first end part F1p on the second visual field side in the first visual field and a second end part F2p on the first visual field side in the second visual field are overlapped each other. In the first visual field, a first defect X1 is contained. In the second visual field, a second defect X2 is contained. In the classification part, a region where the first defect is reflected in the first visual field is cut out as a first defect image C1, and a region where the second defect in the second visual field is reflected is cut out as a second defect image C2. The classification part classifies a defect pattern in a state where the first and second defect images are positioned so that the first and second defects are overlapped when the first defect image contains the first end part and the second defect image contains the second end part.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present disclosure relates to a defect classification device.

Background Art

[0002] For example, in the manufacturing process of semiconductor devices, a large number of circuit patterns are formed on the surface of a semiconductor wafer (substrate) by performing processes such as exposure, development, and etching on the surface of the semiconductor wafer. Then, the semiconductor wafer on which a large number of circuit patterns are formed is diced into a plurality of chip components. The diced chip components are shipped as individual electronic components or incorporated into electrical products after being packaged.

[0003] Here, before the semiconductor wafer is diced into a plurality of chip components, defect detection is performed on the surface of the semiconductor wafer. Defect detection is performed, for example, based on an inspection image captured by a camera connected to an optical microscope or a scanning electron microscope.

[0004] Then, for the semiconductor wafer on which defects have been detected, defect pattern classification is performed based on feature amounts such as the shape, size, and color of the defects. Examples of defect patterns include foreign matter and scratches.

[0005] Conventionally, defect pattern classification has been performed by visual inspection by an operator. However, in visual inspection, variations may occur in the defect pattern classification results depending on the subjectivity of the operator, so the defect patterns could not be classified accurately.

[0006] Therefore, as shown in Patent Document 1, an automatic defect classification device (Automatic Defect Classification: ADC) that automatically classifies defect patterns generated on the surface of a semiconductor wafer is known.

Prior Art Documents

Patent Documents

[0007] [Patent Document 1] Japanese Patent Application Laid-Open No. 2004-47939 [Summary of the Invention] [Problems to be Solved by the Invention]

[0008] However, even when using the automatic defect classification device according to Patent Document 1 to classify defect patterns while excluding the subjectivity of the operator, there are still problems with the classification accuracy of the defect patterns.

[0009] A semiconductor wafer to be inspected is usually larger than the field of view (imaging range) of a camera. Therefore, by dividing the semiconductor wafer into a plurality of fields of view and imaging with a camera, inspection images are obtained for each of the plurality of fields of view. At this time, a defect generated in the semiconductor wafer may straddle a plurality of fields of view.

[0010] For this reason, when attempting to classify a defect pattern based on only one field of view, it may not be possible to accurately classify the defect pattern. For example, when only a part of a defect is shown at the edge of the one field of view, the original feature amounts (such as shape, size, color, etc.) of the defect shown only in part cannot be fully grasped, so an accurate defect pattern cannot be grasped.

[0011] The above problems also apply to other inspection objects other than semiconductor wafers.

[0012] The present disclosure has been made in view of such points, and its object is to accurately classify the pattern of defects generated in the inspection object. [Means for Solving the Problems]

[0013] The defect classification device according to the present disclosure includes a classification unit that classifies the patterns of defects generated in an inspection target imaged so as to be divided into a plurality of visual fields. The plurality of visual fields include a first visual field and a second visual field adjacent to each other. A first end portion that is an end portion on the second visual field side in the first visual field and a second end portion that is an end portion on the first visual field side in the second visual field overlap each other. The first visual field includes a first defect as the defect, and the second visual field includes a second defect as the defect. The classification unit cuts out a region where the first defect in the first visual field is reflected as a first defect image, and cuts out a region where the second defect in the second visual field is reflected as a second defect image. When the first defect image includes the first end portion and the second defect image includes the second end portion, the classification unit classifies the pattern of the defect in a state where the first defect image and the second defect image are aligned so that the first defect and the second defect overlap each other.

[0014] According to the present embodiment, the first defect image cut out from the first visual field and the second defect image cut out from the second visual field are combined so that the first defect and the second defect overlap each other. In a state where the first defect image and the second defect image are combined, the pattern of the defect is classified.

[0015] Even when the defect generated in the inspection target straddles the first visual field and the second visual field, the original feature amount of the defect can be grasped.

[0016] As described above, the pattern of the defect generated in the inspection target can be accurately classified.

[0017] In one embodiment, the classification unit cuts out, as a matching image, a region including the first end portion in the first defect image, sets, as a search range, a region including the second end portion in the second defect image, the classification unit cuts out a plurality of target images corresponding to the matching image from within the search range, and the classification unit aligns the first defect image and the second defect image so that the matching image is superimposed on the target image that is closest to the matching image among the plurality of target images.

[0018] According to such a configuration, by aligning the first defect image and the second defect image so that the matching image overlaps with the target image that is closest to the matching image among the plurality of target images T, the first defect and the second defect can be more correctly superimposed. This is advantageous for accurately classifying the pattern of the defect generated in the object to be inspected.

[0019] In one embodiment, the classification unit obtains the feature amount of the matching image and obtains the feature amount of each of the plurality of target images, the classification unit compares the feature amount of the matching image with the feature amount of each of the plurality of target images, and the classification unit superimposes the matching image on the target image having the feature amount closest to the feature amount of the matching image among the plurality of target images.

[0020] According to such a configuration, by comparing the feature amount of the matching image with the feature amount of each of the plurality of target images, the target image closest to the matching image can be more accurately determined from among the plurality of target images.

[0021] In one embodiment, the classification unit aligns the first defect image and the second defect image in an adjacent direction in which the first visual field and the second visual field are adjacent and in an intersecting direction intersecting the adjacent direction.

[0022] According to such a configuration, the first defect image and the second defect image can be suitably aligned in at least two directions of the adjacent direction and the crossing direction.

Advantages of the Invention

[0023] According to the present disclosure, the pattern of defects generated in the object to be inspected can be accurately classified.

Brief Description of the Drawings

[0024]

Figure 1

Figure 2

Figure 3

Embodiments for Carrying Out the Invention

[0025] Hereinafter, an embodiment of the present disclosure will be described in detail with reference to the drawings. The following description of the preferred embodiments is merely exemplary in nature and is in no way intended to limit the present disclosure, its applications, or its uses.

[0026] (Defect Classification) FIG. 1 schematically shows a defect classification method by a defect classification apparatus 1. The defect classification apparatus 1 is an automatic defect classification (ADC) apparatus. The defect classification apparatus 1 includes a classification unit 10. The classification unit 10 is built into the defect classification apparatus 1 main body. The classification unit 10 is, for example, a computer. The classification unit 10 includes, for example, a processor mounted on a substrate and a memory device that stores software for operating the processor.

[0027] The classification unit 10 of the defect classification apparatus 1 classifies the pattern of a defect X generated in the object to be inspected W. In this example, the object to be inspected W is the surface of a semiconductor wafer (substrate) in a semiconductor manufacturing process.

[0028] Before the process of classifying the pattern of defect X by the defect classification device 1, the defect X is detected by a defect detection device (not shown). When detecting the defect X, the inspection target W is imaged by a camera (not shown) connected to an optical microscope or a scanning electron microscope to obtain an inspection image J, and the inspection image J is compared with a reference image K.

[0029] The reference image K is also called a fitting image or a statistical non-defective image. The reference image K is obtained, for example, by averaging the images of the inspection target W determined to be non-defective. The reference image K is set in advance. When detecting the defect X, the inspection image J is superimposed on the reference image K.

[0030] The classification unit 10 of the defect classification device 1 classifies the pattern of the defect X only for the inspection target W in which the defect X is detected by the defect detection device. Here, examples of the pattern of the defect X include foreign matter, scratches, edge rinsing defects, through-hole abnormalities, coating unevenness, peeling, scratches, bonding pad abnormalities, color unevenness, pattern abnormalities, stains, discoloration, deformation, etc.

[0031] The classification unit 10 of the defect classification device 1 classifies the pattern of the defect X based on a predetermined feature amount of the defect X for the inspection image J of the inspection target W.

[0032] Examples of the feature amount of the defect X include area, brightness, color, luminance, shape, sharpness, length, width, width / length, angle, transparency, etc. The feature amounts of the defect X are registered in advance in the classification unit 10 of the defect classification device 1.

[0033] The classification unit 10 of the defect classification device 1 may classify the pattern of the defect X based on classification conditions generated by AI (Artificial Intelligence) learning for the inspection image J of the inspection target W. Examples of AI learning include machine learning and deep learning. Also, examples of machine learning include "supervised learning", "unsupervised learning", and "reinforcement learning".

[0034] (Field of view segmentation) Here, the object W to be inspected is usually larger than the field of view (imaging range) F of the camera. Therefore, the object W to be inspected is imaged so as to be divided into a plurality of fields of view F. As a result, inspection images J are obtained for each of the plurality of fields of view F.

[0035] Each field of view F (each inspection image J) is adjacent to each other. In this example, each field of view F is rectangular. Each field of view F is arranged in a rectangular grid. In other words, each field of view F is adjacent to each other vertically and horizontally. Each field of view F is further divided into a plurality of pixels.

[0036] Hereinafter, the field of view F may be referred to as FOV (field of view).

[0037] In FIG. 1, as a plurality of fields of view F, the upper left FOV11, the lower left FOV12, the upper right FOV21, and the lower right FOV22 are illustrated.

[0038] FOV11 and FOV12 are adjacent to each other in the vertical direction. The lower end portion 11c, which is the end portion on the FOV12 side in FOV11, and the upper end portion 12a, which is the end portion on the FOV11 side in FOV12, overlap each other (overlap).

[0039] FOV11 and FOV21 are adjacent to each other in the horizontal direction. The right end portion 11b, which is the end portion on the FOV21 side in FOV11, and the left end portion 21d, which is the end portion on the FOV11 side in FOV21, overlap each other (overlap).

[0040] FOV21 and FOV22 are adjacent to each other in the vertical direction. The lower end portion 21c, which is the end portion on the FOV22 side in FOV21, and the upper end portion 22a, which is the end portion on the FOV21 side in FOV22, overlap each other (overlap).

[0041] FOV12 and FOV22 are adjacent to each other in the left - right direction. The right - end portion 12b, which is the end portion on the FOV22 side in FOV12, and the left - end portion 22d, which is the end portion on the FOV12 side in FOV22, overlap (overlap) with each other.

[0042] The above - mentioned "end portion" includes the end in the visual field F and the region slightly inside from the end.

[0043] Four inspection images J11, J12, J21, and J22 are obtained by the camera so as to correspond to the four visual fields F (FOV11, FOV12, FOV21, FOV22).

[0044] The reference image K is also the same as the inspection image J. That is, four reference images K11, K12, K21, and K22 are prepared in advance so as to correspond to the four visual fields F (FOV11, FOV12, FOV21, FOV22).

[0045] Here, the defect X generated in the object under inspection W may straddle multiple visual fields F (inspection images J). In the example of FIG. 1, the defect X generated in the object under inspection W straddles in the left - right direction across FOV11 (inspection image J11) and FOV21 (inspection image J21).

[0046] As shown in FIG. 1, the defect X extends from the middle portion to the right - end portion 11b in FOV11 (inspection image J11) and then reaches the left - end portion 21d in FOV21 (inspection image J21). The defect X has a jagged shape like a sawtooth.

[0047] Here, the defect detection device detects the defect X in FOV11 (inspection image J11) and FOV21 (inspection image J21) of the object under inspection W. On the other hand, the defect detection device does not detect the defect X in FOV12 (inspection image J12) and FOV22 (inspection image J22) of the object under inspection W. Note that the defect X does not exist in any of the four reference images K11, K12, K21, and K22.

[0048] In the classification of the pattern of defect X by the defect classification device 1, only the field of view F (inspection image J) in which the defect X is detected by the defect detection device is extracted and compared with the reference image K. In the example of FIG. 1, as the fields of view F (inspection images J) in which the defect X is detected, FOV11 (inspection image J11) and FOV21 (inspection image J21) are first extracted. Then, FOV11 (inspection image J11) is compared with the reference image K11, and FOV21 (inspection image J21) is compared with the reference image K21.

[0049] If an attempt is made to classify the pattern of defect X based on only one field of view F, it may not be possible to accurately classify the pattern of defect X. For example, in FOV21 (inspection image J21), only a slight defect X is reflected near its left end 21d. Therefore, even if FOV21 (inspection image J21) is compared with the reference image K22, the original feature amounts (such as shape, size, color, etc.) of defect X cannot be sufficiently grasped, and the accurate pattern of defect X cannot be grasped.

[0050] On the other hand, in FOV11 (inspection image J11), defect X is reflected mostly from its middle part to its right end 11b. Therefore, when FOV11 (inspection image J11) is compared with the reference image K11, the original feature amounts of defect X may be grasped, and the accurate pattern of defect X may be grasped. For example, based on the jagged shape of defect X reflected in FOV11 (inspection image J11), the pattern of defect X may be classified as "scratch".

[0051] However, if only a slight defect X is reflected in FOV11 (inspection image J11) as well, similar to FOV21 (inspection image J21), the original feature amounts of defect X still cannot be grasped, and the accurate pattern of defect X cannot be grasped.

[0052] Thus, when the defect X occurring in the object under inspection W straddles multiple fields of view F (inspection images J), it is difficult to accurately classify the pattern of defect X.

[0053] In the defect classification device 1 according to this embodiment, by making the following improvements, even when the defect X generated in the object W to be inspected straddles a plurality of fields of view F (inspection images J), it is possible to accurately classify the pattern of the defect X.

[0054] (Cutting out of the defect image) Hereinafter, FOV11 is defined as the first field of view F1, and FOV21 is defined as the second field of view F2. As shown in FIG. 1, the plurality of fields of view F include FOV11 as the first field of view F1 and FOV21 as the second field of view F2. FOV11 (the first field of view F1) and FOV21 (the second field of view F2) are adjacent to each other in the left-right direction.

[0055] The right end portion 11b as the first end portion F1p, which is the end portion on the FOV21 (second field of view F2) side in FOV11 (first field of view F1), and the left end portion 21d as the second end portion F2p, which is the end portion on the FOV11 (first field of view F1) side in FOV21 (second field of view F2), overlap each other (overlap).

[0056] The right end portion 11b (first end portion F1p) of FOV11 (first field of view F1) includes the right end and a region slightly to the left of the right end in FOV11 (first field of view F1). The left end portion 21d (second end portion F2p) of FOV21 (second field of view F2) includes the left end and a region slightly to the right of the left end in FOV21 (second field of view F2).

[0057] The right end, which is the end on the FOV21 (second field of view F2) side in FOV11 (first field of view F1), is located closer to the right side of FOV21 (second field of view F2) than the left end of FOV21 (second field of view F2). The left end, which is the end on the FOV11 (first field of view F1) side in FOV21 (second field of view F2), is located closer to the left side of FOV11 (first field of view F1) than the right end of FOV11 (first field of view F1).

[0058] As shown in FIG. 1, the FOV11 (first field of view F1) includes a first defect X1 as a defect X. The FOV21 (second field of view F2) includes a second defect X2 as a defect X. The first defect X1 is part of the defect X. The second defect X2 is part of the defect X.

[0059] As shown in FIG. 1, the classification unit 10 of the defect classification device 1 cuts out the area where the first defect X1 in the FOV11 (first field of view F1) is reflected as a first defect image C1. The classification unit 10 of the defect classification device 1 cuts out the area where the second defect X2 in the FOV21 (second field of view F2) is reflected as a second defect image C2.

[0060] The first defect image C1 is smaller than the FOV11 (first field of view F1). The second defect image C2 is smaller than the FOV21 (second field of view F2). The size and position of the first defect image C1 are set so that the first defect X1 is easy to visually recognize. The size and position of the second defect image C2 are set so that the second defect X2 is easy to visually recognize. The size and position of the first defect image C1 and the size and position of the second defect image C2 may be the same as each other or may be different from each other.

[0061] (Combination of defect images) The combination of the first defect image C1 and the second defect image C2 will be described in detail. FIG. 2 shows the combination of the first defect image C1 and the second defect image C2. When the first defect image C1 includes the right end portion 11b (first end portion F1p) of the FOV11 (first field of view F1) and the second defect image C2 includes the left end portion 21d (second end portion F2p) of the FOV21 (second field of view F2), the classification unit 10 of the defect classification device 1 classifies the pattern of the defect X in a state where the first defect image C1 and the second defect image C2 are aligned so that the first defect X1 and the second defect X2 overlap.

[0062] In this example, the first defect image C1 includes the right end portion 11b of FOV11. The second defect image C2 includes the left end portion 21d of FOV21. Therefore, the first defect image C1 and the second defect image C2 are aligned so that the first defect X1 and the second defect X2 overlap. Then, in a state where the first defect image C1 and the second defect image C2 are aligned (a state where the first defect X1 and the second defect X2 overlap), the pattern of the defect X is classified.

[0063] Here, the left - right direction in which FOV11 (the first field of view F1) and FOV21 (the second field of view F2) are adjacent is defined as the adjacent direction A. The adjacent direction A is also the direction in which FOV11 and FOV21 overlap. The up - down direction that intersects (specifically, is orthogonal to) the left - right adjacent direction A is defined as the intersecting direction B. The intersecting direction B is also the direction parallel to the right side of FOV11 and the left side of FOV21.

[0064] The classification unit 10 of the defect classification device 1 aligns the first defect image C1 and the second defect image C2 in the adjacent direction (left - right direction) A and the intersecting direction (up - down direction) B. That is, the first defect image C1 and the second defect image C2 are aligned while being moved in the adjacent direction A and the intersecting direction B so that the first defect X1 and the second defect X2 overlap.

[0065] Conversely, when the first defect image C1 does not include the right end portion 11b (the first end portion F1p) of FOV11 (the first field of view F1) or the second defect image C2 does not include the left end portion 21d (the second end portion F2p) of FOV21 (the second field of view F2), the classification unit 10 of the defect classification device 1 does not align the first defect image C1 and the second defect image C2 (does not overlap the first defect X1 and the second defect X2). In this case, the classification unit 10 of the defect classification device 1 classifies the pattern of the defect X based on the first defect image C1 alone and / or the second defect image C2 alone.

[0066] (Overlay of the first defect and the second defect) The superposition of the first defect X1 and the second defect X2 will be described in detail. FIG. 3 shows the superposition of the first defect X1 and the second defect X2. The classification unit 10 of the defect classification device 1 cuts out a part of the region including the right end portion 11b (the first end portion F1p) of the FOV11 (the first field of view F1) in the first defect image C1 as the matching image M.

[0067] Next, the classification unit 10 of the defect classification device 1 sets the region including the left end portion 21d (the second end portion F2p) of the FOV21 (the second field of view F2) in the second defect image C2 as the search range N.

[0068] The matching image M is smaller than the first defect image C1. The search range N is smaller than the second defect image C2. The matching image M is smaller than the search range N. The matching image M may be cut out after setting the search range N in the reverse order of the above.

[0069] The classification unit 10 of the defect classification device 1 cuts out a plurality of target images T (T1, T2,... Tn) from the search range N set in the second defect image C2. Each target image T cut out from the search range N of the second defect image C2 corresponds to the matching image M cut out from the first defect image C1. The size of each target image T is equal to the size of the matching image M.

[0070] The classification unit 10 of the defect classification device 1 obtains the feature amount Ma of the matching image M. The classification unit 10 of the defect classification device 1 obtains the feature amounts Ta (Ta1, Ta2,... Tan) of each of the plurality of target images T (T1, T2,... Tn).

[0071] The classification unit 10 of the defect classification device 1 compares the feature amount Ma of the matching image M with the feature amounts Ta (Ta1, Ta2,... Tan) of each of the plurality of target images T (T1, T2,... Tn).

[0072] The classification unit 10 of the defect classification device 1 designates the target image T that is closest to the matching image M among the plurality of target images T as the closest target image TC.

[0073] The classification unit 10 of the defect classification device 1 aligns the first defect image C1 and the second defect image C2 so as to superimpose the matching image M on the most recent target image TC (the target image T closest to the matching image M among the plurality of target images T).

[0074] Specifically, the classification unit 10 of the defect classification device 1 designates, as the most recent target image TC, the target image T having the feature amount Ta closest to the feature amount Ma of the matching image M among the plurality of target images T.

[0075] The classification unit 10 of the defect classification device 1 superimposes the matching image M on the most recent target image TC (the target image T having the feature amount Ta closest to the feature amount Ma of the matching image M among the plurality of target images T).

[0076] When comparing the feature amount Ma of the matching image M with the feature amounts Ta of the respective target images T, for example, the SIFT (Scale-Invariant Feature Transform) method is used. Feature points are detected from the matching image M. Feature vectors in the peripheral region of the feature points of the matching image M are obtained. Feature points are detected from each target image T. Feature vectors in the peripheral region of the feature points of each target image T are obtained.

[0077] The Euclidean distance between the feature vector of the matching image M and the feature vectors of the respective target images T is calculated. It is determined that the feature points of the target image T are identical to the feature points of the matching image M on the assumption that the higher the similarity between the feature points of the matching image M and the feature points of the target image T, the smaller the Euclidean distance.

[0078] In applying this example, a part of the region including the right end portion 11b of FOV11 in the first defect image C1 is cut out as the matching image M. A region including the left end portion 21d of FOV21 in the second defect image C2 is set as the search range N. A plurality of target images T (T1, T2,... Tn) are cut out from within the search range N.

[0079] The feature points of the matching image M are obtained to obtain the feature quantity Ma. The feature points of each target image T (T1, T2, … Tn) are obtained to obtain the feature quantities Ta (Ta1, Ta2, … Tan).

[0080] The feature quantity Ma of the feature points of the matching image M is matched with the feature quantities Ta (Ta1, Ta2, … Tan) of the feature points of each target image T (T1, T2, … Tn). The target image T with the largest overlapping area between the feature quantity Ma of the feature points of the matching image M and the feature quantities Ta (Ta1, Ta2, … Tan) of the feature points of each target image T is obtained and taken as the nearest target image TC. From the deviation amount between the matching image M and the nearest target image TC, the overlap value between the first defect image C1 and the second defect image C2 is obtained, and the first defect image C1 and the second defect image C2 are combined.

[0081] In this way, the classification unit 10 of the defect classification device 1 classifies the pattern of the defect X in a state where the first defect image C1 and the second defect image C2 are aligned (a state where the first defect X1 and the second defect X2 overlap). Thereby, the pattern of the defect X generated in the inspection target W is correctly classified as a "scratch" so as to straddle the right end portion 11b of the FOV11 and the left end portion 21d of the FOV21.

[0082] (Function and effect) The first defect image C1 cut out from the FOV11 (the first field of view F1) and the second defect image C2 cut out from the FOV21 (the second field of view F2) are combined so that the first defect X1 and the second defect X2 overlap. In a state where the first defect image C1 and the second defect image C2 are combined, the pattern of the defect X is classified.

[0083] Even when the defect X generated in the inspection target W straddles the FOV11 (the first field of view F1) and the FOV21 (the second field of view F2), the original feature quantity of the defect X can be grasped.

[0084] As described above, the pattern of the defect X occurring in the inspection target W can be accurately classified.

[0085] Rather than directly connecting FOV11 and FOV21, by combining the first defect image C1 cut out from FOV11 and the second defect image C2 cut out from FOV21, the capacity of the image stored in the classification unit 10 of the defect classification device 1 can be saved.

[0086] The first defect image C1 and the second defect image C2 are aligned so that the matching image M overlaps with the target image T (the nearest target image TC) among the plurality of target images T that is closest to the matching image M. As a result, the first defect X1 and the second defect X2 can be more accurately overlapped. And this is advantageous for accurately classifying the pattern of the defect X occurring in the inspection target W.

[0087] By comparing the feature amount Ma of the matching image M with the feature amounts Ta of each of the plurality of target images T, the target image T (the nearest target image TC) that is closest to the matching image M among the plurality of target images T can be determined more accurately.

[0088] The first defect image C1 and the second defect image C2 can be preferably aligned in at least two directions, namely the adjacent direction A and the intersecting direction B.

[0089] (Other Embodiments) As described above, the present disclosure has been described by way of preferred embodiments, but such descriptions are not limiting matters, and of course, various modifications, substitutions, or combinations are possible.

[0090] In the above embodiment, the defect X extends horizontally between the FOV11 as the first field of view F1 and the FOV21 as the second field of view F2, but it is not limited to this. For example, the defect X may extend vertically between the FOV11 as the first field of view F1 and the FOV12 as the second field of view F2. In this case, the vertical direction becomes the adjacent direction A, and the horizontal direction becomes the intersecting direction B. Also, the lower end portion 11c of the FOV11 becomes the first end portion F1p, and the upper end portion 12a of the FOV12 becomes the second end portion F2p.

[0091] Alternatively, the defect X may extend horizontally between the FOV12 as the first field of view F1 and the FOV22 as the second field of view F2, or may extend vertically between the FOV21 as the first field of view F1 and the FOV22 as the second field of view F2.

[0092] For the cases exemplified above, the first field of view F1 and the second field of view F2 may be applied in reverse.

[0093] "First..." and "Second..." are merely named for the convenience of explanation in the specification.

[0094] In the above embodiment, as the plurality of fields of view F, the upper left FOV11, the lower left FOV12, the upper right FOV21, and the lower right FOV22 are exemplified, but a large number of other fields of view may also be included.

[0095] The field of view F is not limited to a quadrilateral, and may be a polygon other than a quadrilateral, a shape other than a polygon, or the like.

[0096] The defect classification device 1 may include a determination unit for determining whether the defect X obtained in a state where the first defect image C1 and the second defect image C2 are aligned (a state where the first defect X1 and the second defect X2 overlap) looks natural.

[0097] When comparing the feature amount Ma of the matching image M and the feature amount Ta of each target image T, other methods (for example, a method using a machine learning model) may be applied instead of the SIFT method.

[0098] The classification unit 10 may be provided not in the main body of the defect classification apparatus 1 but in an external server or the like outside the main body of the defect classification apparatus 1.

[0099] The inspection target W is not limited to a semiconductor wafer, and various targets are conceivable. The inspection target W may be, for example, glass, metal, resin, or the like. The inspection target W does not have to be a plate material, and may be, for example, an electric wire or the like.

Industrial Applicability

[0100] Since the present disclosure can be applied to a defect classification apparatus, it is extremely useful and has high industrial applicability.

Explanation of Reference Numerals

[0101] W Inspection target X Defect X1 First defect X2 Second defect J Inspection image K Reference image F Field of view FOV11 F1 First field of view FOV21 F2 Second field of view 11b Right end F1p First end 21d Left end F2p Second end C1 First defect image C2 Second defect image A Adjacent direction B Crossing direction M Matching image N Search range T Target image TC Nearest target image Ma Feature amount Ta Feature amount 1 Defect classification apparatus 10 Classification unit

Claims

1. A classification unit that classifies the pattern of defects generated in the inspection target imaged so as to be partitioned into a plurality of fields of view, The plurality of fields of view include a first field of view and a second field of view adjacent to each other, A first end portion that is an end portion on the second field of view side in the first field of view and a second end portion that is an end portion on the first field of view side in the second field of view overlap each other, The first field of view includes a first defect as the defect, The second field of view includes a second defect as the defect, The classification unit cuts out a region where the first defect in the first field of view is reflected as a first defect image, and cuts out a region where the second defect in the second field of view is reflected as a second defect image, When the first defect image includes the first end portion and the second defect image includes the second end portion, the classification unit classifies the pattern of the defect in a state where the first defect image and the second defect image are aligned so that the first defect and the second defect overlap each other. A defect classification device.

2. The classification unit cuts out a region including the first end portion in the first defect image as a matching image, and sets a region including the second end portion in the second defect image as a search range, The classification unit cuts out a plurality of target images corresponding to the matching image from the search range, The defect classification device according to claim 1, wherein the classification unit aligns the first defect image and the second defect image so as to overlap the matching image with respect to the target image closest to the matching image among the plurality of target images.

3. The classification unit obtains the feature amount of the matching image and obtains the feature amount of each of the plurality of target images, The classification unit compares the feature amount of the matching image with the feature amount of each of the plurality of target images, The defect classification device according to claim 2, wherein the classification unit overlaps the matching image with respect to the target image having the feature amount closest to the feature amount of the matching image among the plurality of target images.

4. The defect classification device according to any one of claims 1 to 3, wherein the classification unit aligns the first defect image and the second defect image in an adjacent direction in which the first field of view and the second field of view are adjacent and in an intersecting direction intersecting the adjacent direction.

Images