Defect Judgment Method

Abstract

The object is to easily and accurately cut out an image of a region corresponding to a processed region after surface processing from an image of a base material before surface processing, thereby enabling highly accurate defect judgment to be easily performed. [Solution] The defect determination method includes a first image acquisition step of acquiring a first image of a substrate, a second image acquisition step of acquiring a second image of the substrate after the first image acquisition step, in which surface processing has been performed on each of a plurality of processing areas partitioned on the substrate, an image processing step of performing image processing on the first and second images, and a determination step of determining whether or not there are any defects in the plurality of processing areas based on the first and second images that have been image-processed. The image processing step generates a mask image that masks areas other than the plurality of processing areas in the second image, and cuts out at least partial images of areas corresponding to each of the plurality of processing areas from the first image using the generated mask image.

Description

[Technical Field]

[0001] The present invention relates to a defect determination method. [Background technology]

[0002] Patent document 1 discloses a method for inspecting a semiconductor device in which a plurality of semiconductor chips are formed on a first main surface of a semiconductor substrate, a metal layer is formed on a second main surface thereof, the surface quality of the metal layer is judged as good or bad, and the quality of the semiconductor chips on the first main surface is associated with the judgement result. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2009-231763 Summary of the Invention [Problem to be solved by the invention]

[0004] The surface of a substrate such as a semiconductor substrate may be partitioned into multiple processing regions where surface processing is performed, such as forming a coating of electrodes by vapor deposition. One possible method for determining whether defects exist in these processing regions is to cut out images of regions corresponding to each processing region from an image of the substrate before surface processing, convert them into feature quantities, and then determine the defective regions using a machine-learned model. In order to build a model that enables highly accurate defect detection, it is necessary to accurately cut out images of regions corresponding to each processing region. However, because each processing region is not clearly defined on the surface of the substrate before surface processing, it is not easy to accurately cut out images of regions corresponding to each processing region after surface processing from an image of the substrate before surface processing.

[0005] The present invention has been made in consideration of the above-mentioned circumstances, and aims to easily and accurately cut out an image of an area corresponding to the processed area after surface processing from an image of a base material before surface processing, thereby making it possible to easily perform highly accurate defect judgment. [Means for solving the problem]

[0006] In order to solve the above-mentioned problems, the defect judgment method of the present invention comprises a first image acquisition process for acquiring a first image of a substrate, a second image acquisition process for acquiring a second image of the substrate after the first image acquisition process, in which surface processing has been performed on each of a plurality of processing areas partitioned on the substrate, an image processing process for performing image processing on the first image and the second image, and a judgment process for judging whether or not there are any defective areas in the plurality of processing areas based on the first image and the second image that have been image-processed by the image processing process, wherein the image processing process generates a mask image that masks areas other than the plurality of processing areas in the second image, and uses the generated mask image to cut out partial images of areas corresponding to each of the plurality of processing areas from at least the first image. [Effects of the Invention]

[0007] According to the present invention, it is possible to easily and accurately cut out an image of an area corresponding to a processed area after surface processing from an image of a base material before surface processing, thereby making it possible to easily perform highly accurate defect judgment. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram showing the configuration of an information processing system including a defect determination device according to an embodiment of the present invention. [Figure 2] 3 is a flowchart showing a defect determination method according to the present embodiment. [Figure 3] 3 is a diagram for explaining steps S1 to S4, S13, and S14 shown in FIG. 2. [Figure 4] FIG. 3 is a diagram for explaining step S15 shown in FIG. 2. [Figure 5] FIG. 3 is a diagram for explaining step S16 shown in FIG. 2. [Figure 6] FIG. 3 is a diagram for explaining steps S17 and S18 shown in FIG. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Components with the same reference numerals in each embodiment have similar components in each embodiment unless otherwise specified, and description thereof will be omitted.

[0010] FIG. 1 is a diagram showing the configuration of an information processing system 1 including a defect determination device 4 of this embodiment.

[0011] The information processing system 1 is a system that uses a defect determination device 4 to determine whether or not there is a defect in a processing area of ​​a substrate to be surface-processed, based on an image of the surface of the substrate, and presents the determination result to a user. The information processing system 1 includes a user terminal 2 and a server 3 that is equipped with the defect determination device 4. The user terminal 2 and the server 3 are connected to each other via a network N so as to be able to communicate with each other.

[0012] The user terminal 2 includes a processing device 21 and a display device 22. The processing device 21 includes a processor and a memory, and the processor executes a program to realize the functions of the user terminal 2. The display device 22 includes a display and displays the processing results of the processing device 21.

[0013] The user terminal 2 acquires the captured image input by the user and transmits it to the defect determination device 4 of the server 3. The user terminal 2 receives the determination result of the defect determination device 4 from the server 3, displays it, and presents it to the user. The user terminal 2 may acquire the captured image by the user inputting the captured image captured by an external camera into the user terminal 2. Alternatively, the user terminal 2 may have a built-in camera and acquire the captured image by capturing the captured image through user operation.

[0014] The defect determination device 4 performs image processing on the captured image transmitted from the user terminal 2 to determine whether or not a defect exists in the processing area of ​​the substrate to be surface processed. The defect determination device 4 transmits the determination result to the user terminal 2. The substrate is, for example, a semiconductor substrate (wafer), a printed circuit board, or the like. The captured image is, for example, an image of the surface of the substrate. The surface processing is, for example, forming (depositing) a film such as an electrode layer or an insulating layer in multiple processing areas defined on the surface of the substrate by vapor deposition or the like. The specific method of forming the film is not particularly limited, and may be, for example, vapor deposition such as PVD or CVD, sputtering, thermal spraying, plating, or the like. However, the surface processing is not limited to forming a film, and may be processing performed on multiple processing areas defined on the surface of the substrate.

[0015] The defect determination device 4 includes a processing device 40 and a storage device 50. The storage device 50 includes storage for storing captured images received from the user terminal 2, as well as image processing results and determination results of the defect determination device 4. The processing device 40 includes a processor and a memory, and each function is realized by the processor executing a program. As each function, the processing device 40 includes an image acquisition unit 41, an image processing unit 42, a feature extraction unit 43, and a determination unit 44. Details of the image acquisition unit 41, the image processing unit 42, the feature extraction unit 43, and the determination unit 44 will be described later with reference to FIGS. 2 to 6.

[0016] Fig. 2 is a flowchart showing a defect determination method 100 of this embodiment. Fig. 3 is a diagram explaining steps S1 to S4, S13, and S14 shown in Fig. 2. Fig. 4 is a diagram explaining step S15 shown in Fig. 2. Fig. 5 is a diagram explaining step S16 shown in Fig. 2. Fig. 6 is a diagram explaining steps S17 and S18 shown in Fig. 2.

[0017] The defect determination method 100 of this embodiment is realized by cooperation between the user terminal 2 and the defect determination device 4 shown in Fig. 1. The defect determination method 100 of this embodiment includes steps S1 to S6 performed by the user terminal 2 and steps S11 to S19 performed by the defect determination device 4.

[0018] In step S1, the user terminal 2 acquires a captured image G0 of the substrate S before surface treatment (hereinafter also referred to as "pre-treatment"), which is a stage preceding the surface treatment, is performed (hereinafter also referred to as "before pre-treatment"). As shown in the captured image G2 of FIG. 3, the surface of the substrate S is subjected to surface treatment in which a coating C such as an electrode is formed by vapor deposition or the like. After the surface treatment, a plurality of processing areas PA, which are areas where the surface treatment has been performed, are defined on the surface of the substrate S. The pre-treatment is, for example, a surface treatment performed on the entire surface of the substrate S, and examples of the pre-treatment include polishing, cutting, cleaning, etching, and coating. Before the pre-treatment, the surface of the substrate S does not have a plurality of processing areas PA defined. Note that the symbol CA in FIG. 3 indicates an area corresponding to the plurality of processing areas PA defined on the substrate S after the surface treatment. The user terminal 2 acquires the captured image G0 of the substrate S before pre-treatment and stores the acquired captured image G0 by linking it to an identifier indicating that the acquired captured image G0 is an image before pre-treatment.

[0019] In step S2, the user terminal 2 acquires a captured image G1 of the substrate S after preprocessing. A plurality of processing areas PA are not partitioned on the surface of the substrate S after preprocessing. The user terminal 2 acquires the captured image G1 of the substrate S after preprocessing, and stores the acquired captured image G1 in association with an identifier indicating that the acquired captured image G1 is an image after preprocessing.

[0020] In step S3, the user terminal 2 acquires a captured image G2 of the substrate S after surface processing. A plurality of processing areas PA are partitioned on the surface of the substrate S after surface processing. The surface processing is, for example, forming (depositing) a film C in the processing area PA. In each processing area PA, for example, an electrode layer of a semiconductor chip is formed as the film C.

[0021] In step S4, the user terminal 2 transmits each of the captured images G0 to G2 acquired in steps S1 to S3 to the defect determination device 4 of the server 3. The defect determination device 4 performs image processing on each of the captured images G0 to G2 to determine whether or not there is a defective processing area PA among the multiple processing areas PA, and transmits the determination result to the user terminal 2.

[0022] In step S5, the user terminal 2 waits until it receives the judgment result from the defect judgment device 4, and when it receives the judgment result, it proceeds to step S6.

[0023] In step S6, the user terminal 2 displays the determination result of the defect determination device 4 on the display device 22.

[0024] In step S11, the defect determination device 4 waits until it receives the captured images G0 to G2 transmitted from the user terminal 2, and when it has received the captured images G0 to G2, it proceeds to step S12.

[0025] In step S12, the image acquisition unit 41 of the defect determination device 4 acquires each of the captured images G0 to G2 transmitted from the user terminal 2 and stores them in the storage device 50. The image acquisition unit 41 checks the identifiers associated with each of the acquired captured images G0 to G2, and identifies which manufacturing process (before pretreatment, after pretreatment, or after surface processing) the base material S has undergone for each of the acquired captured images G0 to G2.

[0026] In step S13, the image processing unit 42 of the defect determination device 4 generates a mask image G3 using the captured image G2 of the substrate S after surface processing acquired in step S12, and stores the mask image G3 in the storage device 50. Specifically, the image processing unit 42 creates a mask image G3 that masks areas NA other than the multiple processed areas PA (hereinafter also referred to as "non-processed areas NA") in the captured image G2 of the substrate S after surface processing, as shown in Fig. 3. For example, the image processing unit 42 generates the mask image G3 by binarizing the captured image G2 so that the pixel values ​​of the multiple processed areas PA and the pixel values ​​of the non-processed areas NA are different values ​​(e.g., processed area PA = 1, non-processed area NA = 0).

[0027] In step S14, the image processing unit 42 of the defect determination device 4 uses the mask image G3 generated in step S13 to cut out partial images G4 of the areas CA corresponding to the processing areas PA of each of the captured images G0 and G1 acquired in step S12 from the captured images G0 and G1. Additionally, the image processing unit 42 uses the mask image G3 generated in step S13 to cut out partial images G4 of the processing areas PA of the captured image G2 acquired in step S12 from the captured image G2. For example, the image processing unit 42 cuts out partial images G4 from each of the captured images G0 to G2 by performing a filter process on each of the captured images G0 to G2 using the mask image G3. This allows the image processing unit 42 to collectively cut out partial images G4 of the multiple processing areas PA or the multiple areas CA corresponding to these areas. The image processing unit 42 associates the partial images G4 cut out from the captured image G0 with an identifier indicating that the partial images G4 are pre-processed images, and stores the resulting partial images G4 in the storage device 50. The image processing unit 42 associates the partial image G4 cut out from the captured image G1 with an identifier indicating that it is an image after preprocessing and stores it in the storage device 50. The image processing unit 42 associates the partial image G4 cut out from the captured image G2 with an identifier indicating that it is an image after surface processing and stores it in the storage device 50.

[0028] In step S15, the feature extraction unit 43 of the defect determination device 4 acquires a one-dimensional power spectrum of each partial image G4 to extract features of each partial image G4 cut out in step S14. Specifically, as shown in FIG. 4, the feature extraction unit 43 performs a Fourier transform (two-dimensional Fourier transform) on each partial image G4 to acquire an amplitude spectrum and a phase spectrum of each partial image G4. The feature extraction unit 43 then acquires a two-dimensional power spectrum of each partial image G4 by squaring the acquired amplitude spectrum. The two-dimensional power spectrum is expressed in wavenumber space (or frequency space) such that the power spectrum spreads concentrically from the center, with the center as the origin. The feature extraction unit 43 integrates the two-dimensional power spectrum in the circumferential direction (azimuth direction) to convert it to one dimension and acquires the one-dimensional power spectrum. The one-dimensional power spectrum is expressed as a graph with the horizontal axis representing wavenumber (or frequency) and the vertical axis representing power (intensity). The one-dimensional power spectrum represents periodic structural information derived from the size, shape, and arrangement of the structures in the partial image G4. The feature extraction unit 43 acquires one-dimensional power spectra for each partial image G4, and stores each of the acquired one-dimensional power spectra in the storage device 50 while linking it to the partial image G4.

[0029] In step S16, the feature extraction unit 43 of the defect determination device 4 performs principal component analysis on the multiple one-dimensional power spectra acquired in step S15 to acquire principal component vectors and principal component scores. The principal components are expressed as vectors whose elements are, for example, a first principal component PC1, a second principal component PC2, a third principal component PC3, and so on. The principal component vectors indicate spectral components with large variance in a data group composed of multiple one-dimensional power spectra. The principal component scores indicate the proportion of each principal component vector included. The principal component scores reflect differences between the one-dimensional power spectra corresponding to differences between the partial images G4. That is, the feature extraction unit 43 extracts the principal component scores obtained by performing principal component analysis on the one-dimensional power spectra as features of the partial images G4.

[0030] In step S17, the feature extraction unit 43 of the defect determination device 4 uses the principal component vectors and principal component scores acquired in step S16 to generate scatter plots D0 to D2 showing the distribution of principal component scores for each of the captured images G0 to G2, as shown in Fig. 6. The scatter plot D0 is a scatter plot showing the distribution of principal component scores obtained by principal component analysis of the one-dimensional power spectrum of each partial image G4 cut out from the captured image G0. The scatter plot D1 is a scatter plot showing the distribution of principal component scores obtained by principal component analysis of the one-dimensional power spectrum of each partial image G4 cut out from the captured image G1. The scatter plot D2 is a scatter plot showing the distribution of principal component scores obtained by principal component analysis of the one-dimensional power spectrum of each partial image G4 cut out from the captured image G2.

[0031] In step S18, the determination unit 44 of the defect determination device 4 determines whether or not a defective portion exists in a plurality of processing areas PA partitioned on the substrate S after surface processing, based on the feature amounts extracted by the feature amount extraction unit 43. Specifically, the determination unit 44 determines whether or not a defective portion exists based on the distribution of the principal component scores obtained in step S17. For example, the determination unit 44 calculates outliers included in the scatter plots D0 to D2, as shown in FIG. 6. The outliers can be calculated by a known method. Then, the determination unit 44 identifies the partial image G4 corresponding to the calculated outlier, and identifies the captured images G0 to G2 and the processing area PA from which the partial image G4 was cut out, thereby being able to identify the processing area PA in which the defective portion exists.

[0032] The determination unit 44 may also calculate a boundary that distinguishes between good and bad in the distribution of principal component scores using a known method such as a support vector machine, logistic regression, or neural network.The determination unit 44 may then identify a partial image G4 corresponding to data classified as bad, and identify the captured images G0 to G2 from which the partial image G4 was cut and the processing area PA, thereby identifying the processing area PA in which the defective portion exists.The determination unit 44 stores the determination result in the storage device 50.

[0033] In step S19, the defect determination device 4 transmits the determination result of the determination unit 44 to the user terminal 2. Specifically, the defect determination device 4 transmits the position of the processing area PA in which the defective part identified by the determination unit 44 exists and the manufacturing process in which the defective part occurred to the user terminal 2 as the determination result of the determination unit 44. Furthermore, the defect determination device 4 transmits scatter diagrams D0 to D2 to the user terminal 2 together with the determination result of the determination unit 44.

[0034] In the example shown in FIG. 6, the vertical and horizontal axes of the scatter plots D0 to D2 represent the first principal component PC1 and the second principal component PC2, respectively. However, the principal components used for the vertical and horizontal axes of the scatter plots D0 to D2 are not limited to these. For example, the feature extraction unit 43 associates a teacher label (data in which pass / fail is replaced with a numerical value, e.g., pass = 0, fail = 1) with the partial image G4 on which each principal component score is derived, and colors each data in the scatter plots D0 to D2 according to the teacher label. The feature extraction unit 43 can then use the principal components that can distinguish pass / fail in the scatter plots D0 to D2 as the vertical and horizontal axes of the scatter plots D0 to D2. In this way, the defect determination device 4 can present to the user whether each data in the scatter plots D0 to D2 is classified as pass / fail in a visually easy-to-understand format. The scatter plots D0 to D2 may be three-dimensional scatter plots instead of two-dimensional.

[0035] 2, partial image G4 is cut out from captured images G0 to G2, but partial image G4 may be cut out from only captured images G0 and G1 before surface processing and used for defect determination. Also, in the defect determination method 100 shown in FIG. 2, Fourier transform and principal component analysis are performed to extract the feature quantities of partial image G4, but the feature quantities of partial image G4 may be extracted by performing deep learning using training data, for example. Also, in the defect determination method 100 shown in FIG. 2, defect determination is performed based on the distribution of principal component scores, but defect determination may also be performed using a machine-learned model that inputs partial image G4 and outputs a pass / fail determination result.

[0036] As described above, the defect judgment method 100 of this embodiment includes a first image acquisition process (steps S1, S2) for acquiring a first image (captured images G0, G1) of the substrate S, a second image acquisition process (step S3) for acquiring a second image (captured image G2) of the substrate S after the first image acquisition process, in which surface processing has been performed on each of the multiple processing areas PA partitioned on the substrate S, an image processing process (steps S13, S14) for performing image processing on the first image (captured images G0, G1) and the second image (captured image G2), and a judgment process (step S18) for judging whether or not there are any defective areas in the multiple processing areas PA based on the first image (captured images G0, G1) and the second image (captured image G2) that have been image processed by the image processing process. The image processing process generates a mask image G3 that masks areas NA other than the multiple processing areas PA in the second image (captured image G2), and uses the generated mask image G3 to cut out partial images G4 of areas CA corresponding to each of the multiple processing areas PA from at least the first image (captured images G0, G1).

[0037] As a result, the defect determination method 100 can automatically and collectively cut out partial images G4 of the areas CA corresponding to the multiple processing areas PA from the captured images G0 and G1 in which the multiple processing areas PA are not partitioned. Therefore, the defect determination method 100 can significantly reduce the man-hours required for the image processing process and easily perform defect determination. Additionally, even if the position of the substrate S in the captured images G0 to G2 is misaligned, the defect determination method 100 can accurately cut out the partial images G4 of the areas CA corresponding to the processing areas PA by performing a process to match the contours of the substrate S in the captured images G0 to G2 with the mask image G3. Furthermore, the defect determination method 100 can determine whether or not a defective portion exists in the processing area PA after surface processing from the partial images G4 cut out from the captured images G0 and G1 of the substrate S before surface processing, thereby identifying the manufacturing process in which the defective portion occurred. Therefore, the defect determination method 100 can easily and accurately cut out the image of the areas CA corresponding to the processing area PA after surface processing from the image of the substrate S before surface processing, thereby easily performing highly accurate defect determination.

[0038] Furthermore, the defect determination method 100 of this embodiment further includes a feature extraction step (steps S15 and S16) of extracting feature amounts of each partial image G4 cut out by the image processing step. The determination step determines whether or not a defect exists based on the feature amounts extracted by the feature extraction step.

[0039] As a result, the defect determination method 100 can detect even slight defects that would be overlooked by visual inspection, such as uneven surfaces. Moreover, the defect determination method 100 can identify in which manufacturing process such defects occurred. Therefore, the defect determination method 100 can easily and accurately extract an image of the area CA corresponding to the processed area PA after surface processing from the image of the base material S before surface processing, making it possible to easily perform highly accurate defect determination.

[0040] Furthermore, in the defect determination method 100 of this embodiment, the feature extraction step acquires a power spectrum obtained by performing a Fourier transform on the partial image G4, and acquires principal component scores as features obtained by performing principal component analysis on the multiple power spectra.The determination step determines whether or not a defect exists based on the distribution of the principal component scores of the multiple partial images G4.

[0041] As a result, the defect determination method 100 can detect defective areas even when surface processing is performed on defective areas that existed before surface processing and the defective areas are not visible in the partial image G4 after surface processing. In particular, assume that the surface of the substrate S before surface processing has slight irregularities that cannot be detected by edge detection. If a coating C is formed on such an irregular area (i.e., after surface processing), the coating C may have poor film formation, such as insufficient adhesion or film thickness, at the irregular area. However, the partial image G4 after the coating C is formed (after surface processing) may not show the defective film formation. Even in such cases, the defect determination method 100 can perform defect determination using the principal component scores obtained from the power spectrum of the partial image G4 before surface processing as features, thereby detecting defective areas that are not visible in the partial image G4 after surface processing. Furthermore, the defect determination method 100 can identify the manufacturing process in which such defective areas occurred. Therefore, the defect determination method 100 can easily and accurately cut out an image of the area CA corresponding to the processed area PA after surface processing from the image of the substrate S before surface processing, making it possible to easily perform highly accurate defect determination.

[0042] Furthermore, in the defect determination method 100 of this embodiment, the surface processing is to form a coating C on the processing area PA.

[0043] As a result, the defect determination method 100 can determine whether the coating C is defectively formed from the partial image G4 of the surface of the substrate S that serves as the base for the coating C.

[0044] Furthermore, in the defect determination method 100 of this embodiment, the base material S is a semiconductor substrate, and the processing area PA is an area where an electrode layer of a semiconductor chip is formed as a film C.

[0045] As a result, the defect determination method 100 can easily identify defective semiconductor chips from among a plurality of semiconductor chips diced from a semiconductor substrate.

[0046] Although the embodiments of the present invention have been described in detail above, the present invention is not limited to these embodiments and various modifications can be made without departing from the spirit of the present invention. In the present invention, elements of one embodiment can be added to elements of another embodiment, elements of one embodiment can be replaced with elements of another embodiment, or some of the elements of one embodiment can be deleted. [Explanation of symbols]

[0047] 1...information processing system, 2...user terminal, 21...processing device, 22...display device, 3...server, 4...defect judgment device, 40...processing device, 41...image acquisition unit, 42...image processing unit, 43...feature extraction unit, 44...judgment unit, 50...storage device, 100...defect judgment method, C...film, CA...area corresponding to processing area, D0 to D2...scatter plots, G0, G1...captured image (first image), G2...captured image (second image), G3...mask image, G4...partial image, N...network, NA...non-processing area, PA...processing area, S...substrate

Claims

1. a first image acquisition step of acquiring a first image of the substrate; a second image acquisition step of acquiring a second image of the substrate after the first image acquisition step, the second image being an image of the substrate on which surface processing has been performed in each of the plurality of processing regions defined on the substrate; an image processing step of performing image processing on the first image and the second image; a determining step of determining whether or not there are any defective parts in the plurality of processed regions based on the first image and the second image that have been subjected to the image processing in the image processing step, The image processing step generates a mask image that masks areas other than the plurality of processing areas in the second image, and cuts out partial images of areas corresponding to each of the plurality of processing areas from at least the first image using the generated mask image. A method for determining defects.

2. The method further includes a feature extraction step of extracting a feature of each partial image cut out by the image processing step, the determining step determines whether or not the defective portion exists based on the feature extracted in the feature extracting step; 2. The method for determining defects according to claim 1.

3. The feature extraction step includes: A power spectrum is obtained by performing a Fourier transform on the partial image; acquiring, as the feature, principal component scores obtained by performing principal component analysis on the plurality of power spectra; the determining step determines whether or not the defect portion exists based on a distribution of the principal component scores of the plurality of partial images.

3. The method for determining defects according to claim 2.

4. The surface processing is to form a coating on the processing area.

4. The method for determining defects according to claim 3.

5. the substrate is a semiconductor substrate, The processing region is a region where an electrode layer of a semiconductor chip is formed as the film.

5. The method for determining defects according to claim 4.

Images