Defect portion recognition device and defect portion recognition method

Abstract

The present application can mechanically recognize the state of a defective portion in a multilayer film substrate and perform correction processing that is not affected by the skill of an operator. A defective portion recognition device includes a microscope that irradiates white light with a drop shape on the surface of a multilayer film substrate to obtain an enlarged image of a unit area in which a defective portion is recognized on the surface, a spectral light-splitting camera that has an imaging surface on which the enlarged image is imaged and that outputs spectral light-splitting information of the enlarged image for each pixel of the imaging surface, and an information processing portion that processes the spectral light-splitting information output from the spectral light-splitting camera, the information processing portion including a mechanical learning portion that performs clustering processing on the spectral light-splitting information for each pixel and a defective portion recognition portion that recognizes a defective portion based on the processing result of the mechanical learning portion, the mechanical learning portion setting a cluster based on the layer structure present in the unit area and generating a histogram in which the number of pixels clustered in the cluster is frequency, and the defective portion recognition portion comparing the frequency distribution of the generated histogram with the frequency distribution of a histogram in which no defective portion is present and recognizing a defective portion by the presence of a cluster having a frequency difference.

Description

Technical Field

[0001] The present invention relates to a device and method for identifying defective parts of a multilayer film substrate. Background Art

[0002] In the manufacturing process of FPD (Flat Panel Display), when a defective pixel is detected during the inspection process, a correction process is performed on a multi-layer film substrate such as a TFT (Thin Film Transistor) by irradiating the defective portion of the defective pixel with laser light.

[0003] Regarding the identification of the defective part at this time, naked eye identification is performed based on the microscope image, or image processing technology is used to compare the image of the defective part with a reference image without the defective part, so as to grasp the position of the defective part, identify the characteristics, etc. (refer to the following patent document 1).

[0004] Previous technical literature

[0005] Patent Literature

[0006] Patent Document 1: Japanese Patent Application Laid-Open No. 2008-188638 Summary of the Invention

[0007] Technical issues to be solved by the invention

[0008] For defects in multilayer film substrates, it's necessary to identify which layer of the film the defect is located in, and further, to determine the underlying layer structure of the defect. This difference allows for the appropriate selection of a laser correction method. Traditionally, visually identifying defects in microscope images relied on the operator's experience and knowledge to determine the layer structure. This led to issues such as varying correction methods depending on the judgment, which affected the quality of the correction.

[0009] Furthermore, conventional techniques for identifying defects using image processing technology limit the information obtained from the defect using a two-dimensional image to its color, size, contrast, and shape, making it impossible to accurately grasp the layer structure of the defect's underlying layer. Consequently, this ultimately relies on the operator's experience and knowledge, making high-quality correction processing impossible even with these conventional techniques.

[0010] The present invention is made to solve such problems. Specifically, the present invention aims to enable mechanical recognition of the state of a defective portion in a multilayer film substrate and to perform correction processing that is not affected by the operator's skills.

[0011] Means for solving technical problems

[0012] In order to solve such problems, the present invention has the following configuration.

[0013] A defect recognition device is characterized in that it comprises: a microscope for irradiating the surface of a multilayer film substrate with white incident light to obtain an enlarged image of a unit area of ​​the surface for identifying a defect; a spectral spectrometer camera having an imaging surface for imaging the enlarged image and outputting spectral spectrum information of the enlarged image for each pixel of the imaging surface; and an information processing unit for processing the spectral spectrum information output from the spectral spectrometer camera, the information processing unit comprising: a machine learning unit for clustering the spectral spectrum information of each pixel; and a defect recognition unit for identifying a defect based on the processing result of the machine learning unit, the machine learning unit setting a cluster based on the layer structure existing in the unit area and generating a histogram with the number of pixels clustered in the cluster as the frequency, the defect recognition unit comparing the frequency distribution of the generated histogram with the frequency distribution of a histogram without defects and identifying the defect by the existence of a cluster with a frequency difference. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is an explanatory diagram of a defective portion identification device.

[0015] Figure 2 This is an explanatory diagram of the information processing unit of the defective part identification device.

[0016] Figure 3 This is an explanatory diagram for explaining clustering processing in the machine learning unit.

[0017] Figure 4 This is an explanatory diagram illustrating the function of the defect recognition unit.

[0018] Figure 5 This is an explanatory diagram illustrating the defect position recognition function of the defect recognition unit ((a) is the coordinate position of a normal pattern in a layer structure, (b) is the coordinate position of pixels clustered in the layer structure where the defective part exists, and (c) is the coordinate position of pixels in the defective area).

[0019] Figure 6 This is an explanatory diagram for explaining the re-clustering process in the defect area.

[0020] Figure 7 This is an explanatory diagram showing the flow of the defect identification method.

[0021] Figure 8 This is an explanatory diagram showing a laser correction device equipped with a defective portion recognition device. DETAILED DESCRIPTION

[0022] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the following description, the same reference numerals in different figures represent parts with the same functions, and repeated descriptions in the figures will be omitted as appropriate.

[0023] like Figure 1 As shown, the defect recognition device 1 is a device for recognizing defects in a multilayer film substrate W, which is a workpiece placed on a stage S, and includes a microscope 10 , a spectroscopic camera 20 , and an information processing unit 30 .

[0024] The microscope 10 is an optical microscope that irradiates the surface Wa of the multilayer film substrate W with white incident light to obtain a magnified image of a unit area (e.g., a pixel area of ​​a TFT substrate) for identifying defects on the surface Wa. The microscope 10 includes an objective lens 11, a tube lens 17, and other optical systems, as well as a white light source 12 for irradiating the surface Wa with white incident light and its optical system (a reflecting mirror 13 and a half-mirror 14). Furthermore, the microscope 10 may also include, as needed, a monitor camera 15 for obtaining a monitor image of the magnified image of the surface Wa, and an optical system (a half-mirror 16) for the monitor camera 15.

[0025] The spectral spectrometer camera 20 is configured with a slit 23 and a grating element (diffraction grating) 21 on the optical axis 10P of the optical system of the microscope 10, and separates the wavelength of the light reflected from the surface Wa, and images the separated light onto the imaging surface 22a of the two-dimensional camera 22 via the relay lens system 24, and obtains the spectroscopic spectrum information of the magnified image of the surface Wa for each pixel of the imaging surface 22a by line spectrometry.

[0026] The information processing unit 30 processes the spectroscopic spectrum information output from the spectroscopic camera 20, and Figure 2 As shown, the apparatus includes a machine learning unit 31 that clusters the spectral information of each pixel, and a defect recognition unit 32 that recognizes defects based on the processing results of the machine learning unit 31. The recognition results of the defect recognition unit 32 are output to a control unit of a laser correction device (not shown) and are used to determine whether laser processing is required for the defect and to select a processing method when laser processing is performed.

[0027] like Figure 3 As shown, the spectral spectrum information output from the spectral spectroscopic camera 20 outputs a spectral spectrum distribution for each pixel P(Xn, Yn) on the imaging plane 22a of the two-dimensional camera 22. Each pixel P(Xn, Yn) has position information in the XY plane coordinates, and the position of the pixel P(Xn, Yn) corresponds to a specific position within the magnified image of the inspection object.

[0028] Furthermore, the spectral spectrum information for each pixel output by the spectral spectroscopic camera 20 differs from simple two-dimensional image information of the surface Wa. It includes information that can predict the layer structure of the position on the surface Wa corresponding to the pixel P(Xn, Yn) on the imaging surface 22a. This is because light reflected from the surface Wa of the multilayer film substrate W exhibits various spectral characteristics due to differences in the surface layers of the multilayer film, and exhibits a spectral distribution unique to the layer structure due to interference with light reflected at the interfaces between the layers of the multilayer film.

[0029] When clustering the spectral spectrum information of each pixel output by the spectral spectrophotometer 20, the machine learning unit 31 uses the number of layer structures existing in the unit area of ​​the multilayer film substrate W to be inspected as the number of clusters. The layer structure can be known in advance based on the design information of the multilayer film substrate W. Figure 3 In the example shown, there are 11 layer structures in the unit area of ​​the multilayer film substrate W, and 11 clusters No. 1 to 11 are set corresponding to this.

[0030] The machine learning unit 31 does not know what the layer structure corresponding to the cluster is, but gathers pixels having spectral spectrum information of similar spectral distribution into one cluster and assigns each pixel in the imaging surface 22a to a cluster. Figure 3 The example of 11 clusters is processed. Figure 3 As shown, the machine learning unit 31 generates a histogram in which the number of pixels clustered in each cluster is used as the frequency.

[0031] When there are no defects within the unit area under inspection, this histogram is identical to the designed histogram. However, when there are defects within the unit area, the frequencies in the histogram differ from the designed values. The defect identification unit 32 compares the frequency distribution of the histogram generated by the machine learning unit 31 with the frequency distribution of the histogram of the design values ​​without defects, and identifies the hierarchical structure of clusters with different frequencies as a hierarchical structure that may contain defects.

[0032] Figure 4 The process of the defect recognition unit 32 is schematically shown. The defect recognition unit 32 recognizes the relationship between each cluster and the layer structure set in the machine learning unit 31. Figure 4 As shown in the dashed circle in the figure, a cluster having a large difference between the frequency of the histogram of the design value and the frequency of the clustered histogram is determined, and the layer structure corresponding to the cluster is identified as a layer structure in which a defective portion may exist.

[0033] Then, if Figure 5As shown, the defect recognition unit 32 recognizes the position of the defective portion based on the difference between the coordinate position of the pixels clustered in the cluster with the frequency difference and the coordinate position of the normal pattern in the layer structure in the generated histogram. Figure 4 When there is a defective portion in the layer structure corresponding to cluster 7 in the graph, the coordinate position of the normal pattern of the layer structure is Figure 5 (a) shows the coordinate position distribution. In contrast, if the coordinate position distribution of the pixels actually clustered in cluster 7 is plotted on the XY coordinates, the following can be obtained: Figure 5 (b) shows the coordinate position distribution. In this case, the defect recognition unit 32 obtains Figure 5 (b) The distribution of Figure 5 (a) is the difference in the distribution of Figure 5 (c) shows the position of the defective portion (the coordinate position distribution of the pixels existing in the defective area).

[0034] In contrast, the machine learning unit 31 uses the defect position recognition result of the defect recognition unit 32 as shown in FIG. Figure 6 As shown, the spectral spectrum information of the pixels located in the defective area is re-clustered. At this time, the machine learning unit 31 additionally sets a plurality of defective portion clusters (Nos. 12 and 13) in addition to the clusters (Nos. 1 to 11) corresponding to the already set layer structure.

[0035] In this re-clustering, when the layer structure of the pixel position within the defect area is a known layer structure, it is clustered in the already set clusters (No. 1 to 11). However, when it is a foreign object other than the known layer structure, it is clustered in the added defect portion cluster (No. 12, 13). Moreover, it can be said that the pixels clustered in the added defect portion cluster (No. 12, 13) have a structure different from the designed layer structure, so it is possible to predict, for example, that the layer structure is caused by the adhesion of foreign matter. When the pixels are clustered in the defect portion cluster (No. 12, 13), the defect recognition unit 32 identifies the type of defect portion based on the spectroscopic spectrum information of the clustered pixels and the above prediction.

[0036] Figure 7 The following is an example of a defect recognition method using the defect recognition apparatus 1. When defect recognition begins, a workpiece, i.e., a multilayer film substrate W, is placed on a worktable S (step S01), and the inspection position of the surface Wa is determined, and the optical axis of the microscope 10 is moved to that position (step S02).

[0037] Then, white incident light is irradiated on the surface Wa of the inspection position, and a magnified image of the unit area to be inspected is obtained through the microscope 10, and the spectral spectrometer camera 20 having a camera surface 22a for imaging the magnified image is used to obtain the spectral spectrum information of each pixel of the camera surface 22a (step S03).

[0038] The acquired spectral spectrum information for each pixel is input to the machine learning unit 31 in the information processing unit 30, where the aforementioned clustering process is performed (step S04). Once the clustering process is completed, the number of layer structures present in the unit area to be inspected is used as the number of clusters, and a histogram is generated using the number of pixels clustered in each cluster as the frequency (step S05).

[0039] When the machine learning unit 31 generates a histogram, the defect recognition unit 32 compares the frequency distribution of the histogram generated by clustering with the frequency distribution of the histogram of the design value as described above, and confirms the presence of a defective portion based on the presence of clusters with frequency differences (step S06).

[0040] Then, if Figure 5 As shown, the defect recognition unit 32 determines the location of the defect (defective area) by comparing the coordinate position distribution of the pixels clustered in the cluster containing the defect with the coordinate position distribution of the normal pattern of the layer structure (step S07). Then, the machine learning unit 31 adds the cluster for the defect as described above and re-clusters the pixels within the determined defective area (step S08).

[0041] After the re-clustering, the defect identification unit 32 identifies the type of defect based on the distribution of pixels clustered in the defect cluster and the spectral spectrum information of the pixels clustered therein (step S09 ).

[0042] Afterwards, it is determined whether the inspection position is moved (step S10). If the inspection position is moved (step S10: yes), the processing after step S02 is performed on the moved inspection position. If the inspection position is not moved (step S10: no), the processing is terminated.

[0043] Figure 8 The following shows an example configuration of a laser correction device 2 including the defect recognition device 1. The laser correction device 2 corrects a defect recognized by the defect recognition unit 32 in the information processing unit 30 by irradiating it with laser light, and includes a laser irradiation unit 3 for irradiating laser light L coaxially with the optical axis of the microscope 10.

[0044] The laser irradiation unit 3 includes, for example, a laser light source 40, a laser scanner 42, etc. The laser light L emitted from the laser light source 40 is incident on the optical system of the microscope 10 via the reflector 41 and the Galvano mirrors 42A and 42B of the laser scanner 42, and is irradiated onto the surface Wa of the unit area from which a magnified image based on the microscope 10 can be obtained.

[0045] In the example shown in the figure, a switching mirror 18 is provided for entering / exiting the optical axis of the microscope 10. By making the switching mirror 18 enter the optical axis of the microscope 10, the reflected light from the surface Wa is incident on the spectral spectrometer 20, and by making the defective portion identification device 1 work, the switching mirror 18 is exited from the optical axis of the microscope 10, so that the laser correction device 2 that irradiates the laser L onto the surface Wa can be operated.

[0046] Regarding the laser correction device 2 equipped with such a defect recognition device 1, the defect recognition device 1 is first operated, and the defect recognition unit 32 transmits information such as the presence or absence of a defect, the position of the defect if a defect exists, the layer structure of the defect, and the type of the defect to the laser control unit 50. The laser control unit 50 determines whether to perform laser correction based on the above information transmitted from the defect recognition unit 32. When performing laser correction, the laser irradiation range is set based on the position information of the defect, and the processing method is set based on the layer structure and type information of the defect.

[0047] Furthermore, in the illustrated example, the magnified image of the microscope 10 is also formed on the monitor camera 15, so that laser correction can be performed while observing the image captured by the monitor camera 15 on the display device 52. At this time, the two-dimensional image captured by the monitor camera 15 is processed by the image processing unit 51 and sent to the laser control unit 50 and the information processing unit 30. The laser irradiation unit 3 can also be controlled based on this two-dimensional image.

[0048] According to the embodiments of the present invention described above, defects in a multilayer film substrate W can be mechanically identified in greater detail, and laser correction processing can be set based on this identification information. This enables high-quality correction processing that is not affected by the operator's skills, and the entire process from defect identification to processing can be automated, resulting in efficient and high-quality correction processing.

[0049] While the embodiments of the present invention have been described in detail above with reference to the accompanying drawings, the specific configuration is not limited to these embodiments. Design modifications, etc., that do not depart from the spirit of the present invention are also encompassed by the present invention. Furthermore, the aforementioned embodiments can be combined using their respective technologies as long as there are no particular contradictions or problems with their objectives and structures.

[0050] Explanation of symbols

[0051] 1-defect recognition device, 2-laser correction device, 3-laser irradiation unit, 10-microscope, 10P-optical axis, 11-objective lens, 12-white light source, 13-reflecting mirror, 14, 16-half-reflecting mirror, 15-monitor camera, 17-tube lens, 18-switching mirror, 20-spectral spectrometer camera, 21-grating element, 22-two-dimensional camera, 22a-imaging surface, 23-slit, 30-information processing unit, 31-machine learning unit, 32-defect recognition unit, 40-laser light source, 41-reflecting mirror, 42-laser scanner, 42A, 42B-Galvano mirror, 50-laser control unit, 51-image processing unit, 52-display device, S-workbench, W-multilayer film substrate, Wa-surface, L-laser

Claims

1. A defective part identification device, characterized in that: have: A microscope is used to irradiate the surface of the multilayer film substrate with white incident light to obtain a magnified image of a unit area of ​​the surface for identifying a defective portion; a spectral spectroscopic camera having an imaging surface for imaging the magnified image and outputting spectroscopic spectrum information of the magnified image for each pixel of the imaging surface; and An information processing unit processes the spectroscopic spectrum information output from the spectroscopic camera, The information processing unit includes: a machine learning unit that performs clustering processing on the spectral spectrum information of each pixel; and a defect recognition unit that recognizes a defective portion based on the processing result of the machine learning unit, The machine learning unit sets a cluster based on the layer structure existing in the unit area, and generates a histogram using the number of pixels clustered in the cluster as a frequency. The defect recognition section compares the frequency distribution of the generated histogram with the frequency distribution of a histogram in which no defects exist, and recognizes a defective portion by the presence of a cluster having a frequency difference. In the multilayer film substrate, the light reflected on the surface of the multilayer film substrate shows a spectral distribution unique to the layer structure due to the differences in the surface layers of the multilayer film substrate and due to interference with the light reflected at the interfaces of each layer of the multilayer film in the multilayer film substrate.

2. The defective portion identification device according to claim 1, characterized in that: The defect recognition unit recognizes the position of the defect based on a difference between a coordinate position of pixels clustered in the cluster having the frequency difference and a coordinate position of a normal pattern in the layer structure of the cluster.

3. The defective portion identification device according to claim 1 or 2, characterized in that: The machine learning unit sets a cluster for a defective portion in addition to a cluster corresponding to the layer structure existing in the unit area. The defect recognition section recognizes the type of the defect based on the pixels clustered in the defect cluster.

4. A defective part identification method, characterized in that: have: a step of irradiating the surface of the multilayer film substrate with white incident light to obtain a magnified image of a unit area on the surface for identifying a defective portion; a step of using a spectral spectroscopic camera having an imaging surface for imaging the magnified image, and acquiring spectroscopic spectrum information of the magnified image for each pixel of the imaging surface; a step of clustering the spectral spectrum information of each pixel acquired by the spectral spectroscopic camera into clusters set according to the layer structure existing in the unit area; a step of generating a histogram using the number of pixels clustered in the cluster as a frequency; and A process of comparing the frequency distribution of the generated histogram with the frequency distribution of a histogram without defects, and identifying a defective portion by the presence of a cluster having a frequency difference, In the multilayer film substrate, the light reflected on the surface of the multilayer film substrate shows a spectral distribution unique to the layer structure due to the differences in the surface layers of the multilayer film substrate and due to interference with the light reflected at the interfaces of each layer of the multilayer film in the multilayer film substrate.

5. The defective portion identification method according to claim 4, characterized in that: The position of the defective portion is identified based on the difference between the coordinate position of the pixels clustered in the cluster having the frequency difference and the coordinate position of the normal pattern in the layer structure of the cluster.

6. The defective portion identification method according to claim 5, characterized in that: The defect portion identifying method includes a step of re-clustering pixels identified as the defect portion, wherein a defect portion cluster is set in addition to the cluster, and the type of the defect portion is identified based on the pixels clustered in the defect portion cluster.

Images