WO2008068894A1 - Defect detecting device, defect detecting method, information processing device, information processing method and program - Google Patents

Abstract

[PROBLEMS TO BE SOLVED] A problem to be solved is to efficiently detect defects of an MEMS device without the necessity of an absolute model image while preventing mistakenly detecting with a high degree of accuracy. [MEANS FOR SOLVING THE PROBLEMS] A defect detecting device (100) picks up an image of a protein chip (35) formed in each die (30) of a wafer (1) for every first divided region (71), wherein each die is divided into a plurality of the first divided regions. The defect detecting device preserves the image together with an ID to identify each divided region (71) as an examination subject image, eliminates low frequency components from the examination subject image through a high-pass filter, then calculates average brightness for each pixel of each examination subject image having a corresponding ID, and forms a model image for each first divided region (71). The defect detecting device extracts a difference between the model image and each examination subject image as a difference image, carries out filtering of the difference image by the extraction of a Blob, extracts the Blob with not less than a predetermined area as a Blob defect, and classifies kinds of defects in accordance with a characteristic amount of the extracted Blob.

Description

 Specification

 Defect detection apparatus, defect detection method, information processing apparatus, information processing method, and program thereof

 Technical field

 The present invention relates to a defect detection apparatus capable of inspecting the appearance of a microstructure such as MEMS (Micro Electro Mechanical Systems) formed on a semiconductor wafer and detecting defects such as foreign matter and scratches. The present invention relates to a defect detection method, an information processing apparatus, an information processing method, and a program thereof.

 Background art

 [0002] In recent years, MEMS that integrate various functions in the fields of machinery, electronics, light, chemistry, etc., especially using semiconductor microfabrication technology, has attracted attention. Examples of MEMS devices that have been put into practical use include acceleration sensors, pressure sensors, and airflow sensors as sensors for automobiles and medical use. In particular, MEMS devices are also used in printer heads for ink jet printers, micromirror arrays for reflective projectors, and other actuators. Furthermore, MEMS devices are also applied in fields such as chemical synthesis and bioanalysis such as protein analysis chips (so-called protein chips) and DNA analysis chips.

 [0003] By the way, since a MEMS device is an extremely fine structure, inspection of defects such as foreign matters and scratches in its appearance is important in manufacturing. Conventionally, in the appearance inspection of MEMS devices, manual inspection using a microscope has been performed, but this inspection takes a lot of time and is judged to be performed visually by an inspector. Blur will occur.

[0004] Therefore, as a technique for automating such an appearance inspection, for example, in Patent Document 1 below, for example, a CCD (Charge Coupled Device) camera or the like images an arbitrary plurality of non-defective products. Are stored in the memory as a plurality of non-defective images, and after aligning the non-defective images, the same non-defective images are The average and standard deviation of the luminance values are calculated for each pixel at one position, and the average luminance value and standard deviation are compared with the luminance value of each pixel of the image obtained by imaging the inspected product. A technique for determining pass / fail of an inspection product is described.

 [0005] In addition, in Patent Document 2 below, when creating reference image data that serves as a reference for non-defective products in the inspection of patterns such as wiring boards and printed materials, each of the reference patterns is imaged. The reference pattern image is stored, the position of each reference pattern is aligned, the average value calculation or the intermediate value calculation is performed between each image data for each pixel, and data that has a large variation and abnormal values are avoided. It is described that reference image data serving as a proper reference is created, and pattern defects are detected by comparing the reference image data with inspection image data.

 Patent Document 1: Japanese Patent Laid-Open No. 2 065 _ 2 6 5 6 6 1

 Patent Document 2: Japanese Patent Laid-Open No. 11-7 3 5 1 3 (paragraph [0 0 8 0] etc.) Patent Document 3: Japanese Patent Laid-Open No. 2 0 0 1 _ 2 0 9 7 9 8 (Figure 1 etc.)

 Disclosure of the invention

 Problems to be solved by the invention

 [0006] However, even with the techniques described in Patent Document 1 and Patent Document 2, non-defective image data or reference image data (hereinafter referred to as model image data) serving as an inspection standard is Created based on images of multiple non-defective products prepared separately from the image to be inspected. Therefore, prior to the creation of model image data, it is necessary to determine whether or not it is a non-defective product and to select a non-defective product. It will be. And in the inspection of extremely small structures such as MEMS devices, where a few scratches or foreign objects become defects, preparing an absolute good product (model) itself is model image data. It is difficult in terms of maintenance and management.

[0007] In addition, in the techniques described in Patent Document 1 and Patent Document 2 described above, inspection objects such as wiring boards are placed on a table one by one and imaged, so that each inspection object is not defective. If there is an individual difference of May cause false detection, and inspection accuracy will be reduced.

 [0008] Furthermore, when the inspection object is a three-dimensional object with unevenness or curvature, uneven luminance may occur in the captured image due to illumination conditions, optical conditions, and the like. Sometimes noise may be mixed in, and noise may be misdetected as a defect.

 [0009] In relation to this problem, in Patent Document 3 above, the entire surface of a wafer having a matrix-like repetitive pattern is imaged as a single image, and a median filter or the like is applied to the captured image. Create a reference image (model image) that has been smoothed to remove unevenness due to light diffraction, and compare the reference image and the captured image for each region of interest to extract only the steep shade change. A method is described in which a processed image is obtained, and the brightness value of the extracted image is binarized to determine the presence or absence of a defective portion.

 [0010] In the technique described in Patent Document 3, an inspection is performed using a model image created based on an image obtained by capturing the entire wafer having a diameter of 20 Omm as a single image. In addition, an object having a repetitive pattern such as a semiconductor wafer is an inspection target. Therefore, it is difficult to apply to a visual inspection of a MEMS device having, for example, a few flaws in a few units, foreign objects and uneven brightness, and irregular shapes / turns.

 [001 1] In view of the circumstances as described above, the object of the present invention is to provide a MEMS device with high accuracy and efficiency while preventing false detection due to uneven brightness and noise, etc., without requiring an absolute model image. It is to provide a defect detection apparatus, a defect detection method, an information processing apparatus, an information processing method, and a program for the same.

 Means for solving the problem

[0012] In order to solve the above-described problem, a defect detection apparatus according to a main aspect of the present invention includes a structure in which a microstructure formed on each of a plurality of dies on a semiconductor wafer is divided into a plurality of areas of each of the dies. Imaging means for imaging each divided area, illumination means for illuminating the microstructure to be imaged, and each of the captured divided areas Storage means for storing each image as an inspection target image in association with identification information for identifying the position of each divided region in each die, and for each of the stored inspection target images, A first filtering means for performing filtering for removing low-frequency components in the inspection target image; and in each of the divided regions corresponding to the identification information between the dies of the filtered inspection target images. Model image creation means for creating an average image obtained by averaging the images to be inspected for each identification information as a model image, each created model image, and the filtering corresponding to the identification information for each model image And detecting means for detecting a defect of the microstructure by comparing each of the inspection target images.

 Here, the microstructure is a so-called M EMS (Micro Electro Mechanical System). The imaging means is, for example, a camera with a built-in imaging element such as a CCD (Charge Coupled Device) or a CMOS (Complementary Metal Oxide Semiconductor) sensor. The illumination means is, for example, a flash light emitting unit such as a high-intensity white LED (Light Emitting Diode) or a xenon lamp. The first filtering means is a high-pass filter in other words. Defects are, for example, scratches and foreign objects.

 [0014] With this configuration, a model image is created based on each image of each divided region to be inspected, so that it is difficult to obtain an absolute good product model (sample). Can be performed with high accuracy.

 [0015] Further, a plurality of the microstructures are formed on one semiconductor wafer, and each inspection object image is imaged under the same illumination condition by the imaging means and the light source. Each model image is created based on the image, and each model image is compared with each image to be inspected to detect defects in multiple microstructures, change illumination conditions such as shadows, and the lens of the imaging means. It can be detected accurately and stably without being affected by fixed noise such as dirt.

[0016] Then, the first filtering means can remove luminance unevenness (low frequency component) caused by the optical axis shift of the imaging means, the flatness of the microstructure, and the like. Since the model image is created and the defect is detected based on the filtered image, both the quality of the model image and the defect detection accuracy can be improved by this one-time filtering.

 [001 7] Furthermore, since the captured image to be inspected can be used for both model image creation processing and defect detection processing, it is more efficient than creating model images separately. Defects can be detected.

 [0018] In the defect detection apparatus, the detection unit includes a difference in which each model image and each inspection target image corresponding to the identification information are overlapped with each model image and a difference between both images is extracted as a difference image. Extraction means; and second filtering means for performing filtering for removing a pixel area having a predetermined area or less from a series of pixel areas having a luminance equal to or higher than a predetermined value in the extracted difference image. Also good.

 Here, the second filtering means performs filtering by analyzing, for example, a block of pixels having a predetermined (or a predetermined range) grayscale value in the difference image, that is, B I o b. Thus, further filtering by the second filtering means can prevent the noise component in the difference image from being erroneously detected as a defect, and the detection accuracy can be further improved.

 [0020] In the defect detection apparatus, the model image creating unit superimposes each inspection target image corresponding to the identification information, and calculates an average value of luminance values for each pixel constituting each inspection target image. You may have a means.

 [0021] Thereby, by calculating an average value for each pixel of each inspection target image, it is possible to effectively absorb the variation of each image and create a high-quality model image, and improve detection accuracy. be able to.

 [0022] In the defect detection apparatus, the imaging unit may continuously image the microstructure in each divided region having identification information corresponding to each die.

[0023] Thereby, the divided areas at the same position between the dies are collectively imaged continuously. Thus, it is possible to efficiently create a model image of each divided region and improve inspection efficiency.

 [0024] In addition, the imaging unit images the microstructures of all the divided regions in one die, and then the microstructures of the divided regions of the other die adjacent to the one die. You may make it image.

[0025] Thereby, by continuously capturing the divided regions at the same position between the dies together, it is possible to efficiently create a model image of each divided region and improve the inspection efficiency.

 [0026] In the defect detection apparatus, the microstructure discharges a reagent and a plurality of concave portions having a thin film bottom surface for introducing an antibody that cross-reacts with the reagent, and the reagent that does not react with the antibody. Therefore, it may be a screening examination container having a plurality of holes provided in the bottom surface of each of the recesses.

 [0027] Here, the container is a protein chip. As a result, for example, cracks and scratches on the thin film (membrane) of the protein chip, foreign matter adhering to the thin film, etc. can be detected with high accuracy.

 [0028] In this case, the model image creating means, prior to averaging each inspection object image corresponding to the identification information corresponding to each model image, the shape of each recess of the container in the inspection object image There may be a means for aligning each image to be inspected based on the above.

 [0029] Thus, by using the shape of each concave portion of the container, it is possible to accurately match the overlapping positions of the images to be inspected and to create a higher quality model image. Note that the alignment is performed by changing the relative position of each image by moving each image in the X and Y directions or rotating in the 0 direction.

[0030] Further, in this case, the difference extracting means prior to extracting the difference, the shape of each concave portion of the container in each model image, and each inspection corresponding to the identification information corresponding to each model image It may have means for aligning each model image and each inspection object image based on the shape of each recess in the object image. [0031] Thereby, by using the shape of each concave portion of the container, the overlapping position of the model image and the inspection target image can be accurately matched, and the defect can be detected with higher accuracy.

 [0032] In the defect detection device, the microstructure includes a plate member having a plurality of window holes for irradiating a plurality of electron beams, and a thin film provided to cover the window holes. It may be an electron beam irradiation plate.

 [0033] This makes it possible to detect, for example, cracks and scratches on the thin film (membrane) of the electron beam irradiation plate, foreign matter attached to the thin film, and the like with high accuracy.

 [0034] In this case, the model image creating means, prior to averaging each inspection object image corresponding to the identification information for each model image, the electron beam irradiation plate in each inspection object image There may be provided means for aligning each inspection object image based on the shape of each window hole.

 Thus, by using the shape of each window hole of the electron beam irradiation plate, it is possible to accurately match the overlapping positions of the images to be inspected and to create a higher quality model image.

 [0036] In this case, the difference extracting means prior to extracting the difference, the shape of each window hole of the electron beam irradiation plate in each model image, and the identification information in each model image. There may be provided means for aligning each model image and each inspection object image based on the shape of each window hole in each inspection object image corresponding to.

 [0037] Thereby, by utilizing the shape of each window hole of the electron beam irradiation plate, the overlapping position of the model image and the inspection target image can be accurately matched, and the defect can be detected with higher accuracy. .

[0038] A defect detection method according to another aspect of the present invention provides a microstructure formed on each of a plurality of dies on a semiconductor wafer for each divided region obtained by dividing the region of each die into a plurality of regions. An imaging step; an illumination of the microstructure to be imaged; and an image of each of the captured divided regions in each die. A step of storing as an inspection target image in association with identification information for identifying the position of each of the divided regions, and for each of the stored inspection target images, a low frequency component in each of the inspection target images is stored. A step of performing filtering for removal, and an average image obtained by averaging the inspection target images of the divided regions corresponding to the identification information between the dies among the filtered inspection target images. A step of creating each of the identification information as an image, comparing each of the created model images with the filtered image to be inspected corresponding to the identification information corresponding to each model image, Detecting a defect of the microstructure.

 [0039] In the defect detection method, the detecting step includes superimposing the model images and inspection target images corresponding to the identification information on the model images, and extracting a difference between the images as a difference image. And a step of performing filtering for removing a pixel area having a predetermined area or less from a series of pixel areas having a luminance equal to or higher than a predetermined value in the extracted difference image.

 [0040] An information processing apparatus according to still another aspect of the present invention is such that a microstructure formed in each of a plurality of dies on a semiconductor wafer is illuminated in each divided region in which each die is divided into a plurality. Storage means for storing each of the images picked up in (2) as an inspection target image in association with identification information for identifying the position of each of the divided regions in each die, and for each of the stored inspection target images A filtering unit that performs filtering to remove low-frequency components in each image to be inspected, and each division corresponding to the identification information between each die among the filtered images to be inspected A model image creating means for creating, for each identification information, an average image obtained by averaging the images to be inspected in each region as a model image, each created model image, and each model Detecting means for detecting a defect of the microstructure by comparing the filtered image to be inspected with the identification information corresponding to a Dell image.

[0041] Here, the information processing apparatus is a computer such as a PC (Personal Computer). So-called notebook type or desktop type.

 [0042] An information processing method according to still another aspect of the present invention is such that a microstructure formed in each of a plurality of dies on a semiconductor wafer is illuminated in each divided region in which each die is divided into a plurality. Storing each of the captured images as an inspection target image in association with identification information for identifying the position of each of the divided regions in each die, and for each of the stored inspection target images, Filtering for removing low-frequency components in each image to be inspected, and each inspection of each divided region corresponding to the identification information between the dies among the filtered image to be inspected An average image obtained by averaging the target images is created for each identification information as a model image, each created model image, and each identification image is associated with the identification information. And comparing each filtered image to be inspected corresponding to a report to detect a defect in the microstructure.

 [0043] A program according to still another aspect of the present invention provides a program in which a microstructure formed on each of a plurality of dies on a semiconductor wafer is divided into a plurality of divided regions obtained by dividing each of the dies into a plurality of dies. Storing each image captured under illumination for each of the images as an inspection object image in association with identification information for identifying the position of each of the divided regions in each die; and each of the stored inspection object images In contrast, a step for performing filtering for removing a low frequency component in each inspection target image, and each of the filtered inspection target images corresponding to the identification information between the dies. A step of creating an average image obtained by averaging the images to be inspected in the divided areas for each of the identification information as a model image, the created model images, and the models By comparing the respective inspection target image before Symbol filtering said identification information corresponding to the image, is intended to execute the steps of detecting a defect of the microstructure.

 The invention's effect

As described above, according to the present invention, an absolute model image is not required, and the ME can be performed with high accuracy and efficiency while preventing erroneous detection due to uneven brightness or noise. MS device defects can be detected.

 BEST MODE FOR CARRYING OUT THE INVENTION

 Hereinafter, embodiments of the present invention will be described with reference to the drawings.

 FIG. 1 is a diagram showing a configuration of a defect detection apparatus according to an embodiment of the present invention. As shown in FIG. 1, the defect detection apparatus 100 is a semiconductor wafer 1 made of, for example, silicon. (Hereinafter, also simply referred to as wafer 1), a wafer table 2, a XYZ stage 3 for moving the wafer table 2 in the X, Y and Z directions in the figure, and a CCD for imaging the wafer 1 from above The camera 6 includes a light source 7 that illuminates the wafer 1 during imaging by the CCD camera 6, and an image processing PC (Personal Computer) 10 that controls the operation of each unit and performs image processing to be described later.

 The woofer 1 is transported onto the woofer table 2 by a transport arm (not shown) and is adsorbed and fixed to the woofer table 2 by suction means such as a vacuum pump (not shown). Instead of directly adsorbing wafer 1 to wafer table 2, for example, a tray (not shown) that can hold wafer 1 is prepared separately, and the tray is adsorbed and fixed with wafer 1 held on the tray. You may do so. As will be described later, when a hole is formed in the wafer 1, it may be difficult to directly vacuum-suck the wafer, so the suction method using this tray is effective. On the wafer 1, a protein chip is formed as a MEMS device. The defect detection apparatus 100 is an apparatus for detecting defects such as foreign matters and scratches on the protein chip using the protein chip as an inspection object. Details of the protein chip will be described later.

The CCD camera 6 is fixed at a predetermined position above the wafer 1 and incorporates a lens, a shirter (not shown) and the like. Based on the trigger signal output from the image processing PC 10, the CCD camera 6 emits an image of a protein chip formed on a predetermined portion of the wafer 1, enlarged by a built-in lens, by a light source 7. The image is captured under the flashlight and the captured image is transferred to the image processing PC 10 To send. The XYZ stage 3 moves the wafer 1 in the vertical direction (Z direction), thereby changing the relative distance between the CCD camera 6 and the wafer 1, and the focal point when the CCD camera 6 captures the wafer 1 The position can be changed. The focal position may be varied by moving the CCD camera 6 in the Z direction instead of the XYZ stage 3.

[0049] Further, the lens of the CCD camera 6 is configured as a zoom lens, and the protein chip can be imaged at different magnifications by changing the focal length. In the present embodiment, it is assumed that the magnification of the CCD camera 6 can be varied in two stages of about 7 times (low magnification) and about 18 times (high magnification). The field size in the case of low magnification is, for example, 6 80 X 5 10 (m ² ), and the field size in the case of high magnification is, for example, 2 70 X 200 (U m ^). It is not a thing. Instead of the CCD camera 6, a camera incorporating another image sensor such as a CM OS sensor may be used.

 [0050] The light source 7 is fixed at a predetermined position above the wafer 1, and includes a flash lamp composed of, for example, a high-intensity white LED or a xenon lamp, and a flash lighting circuit that controls lighting of the flash lamp. Based on the flash signal output from the image processing PC 10, the light source 7 illuminates the predetermined portion of the wafer 1 by emitting a high-intensity flash for a predetermined time of, for example, several U seconds.

[0051] The XYZ stage 3 includes a motor 4 for moving the X stage 1 1 and the Y stage 1 2 along the movement axis 1 3 in the X direction, the Y direction, and the Z direction, respectively, and the X stage 1 1 and Y Encoder 5 is used to determine the moving distance of stage 1 2. The motor 4 is, for example, an AC servo motor, a DC servo motor, a stepping motor, a linear motor, etc., and the encoder 5 is, for example, various motor encoders, a linear scale, or the like. Each time the X stage 1 1 and the Y stage 1 2 move by a unit distance in the X, Y, and Z directions, the encoder 5 generates an encoder signal that is the movement information (coordinate information) and outputs the encoder signal. Output to PC 10 for image processing. [0052] The image processing PC 10 inputs the encoder signal from the encoder 5 and outputs a flash signal to the light source 7 based on the encoder signal. Is output. The image processing PC 10 outputs a motor control signal for controlling the driving of the motor 4 to the motor 4 based on the encoder signal input from the encoder 5.

 FIG. 2 is a block diagram showing a configuration of the image processing PC 10.

 As shown in the figure, the image processing PC 10 includes a CPU (Central Processing Unit) 21, ROM (Read Only Memory) 22, RAM (Random Access Memory) 23, I / O interface 24, HDD (Hard Disk Drive 25) A display unit 26 and an operation input unit 27, which are electrically connected to each other via an internal bus 28.

 [0054] The C P U 21 comprehensively controls each part of the image processing PC 10 and performs various calculations in image processing to be described later. The ROM 22 is a non-volatile memory that stores programs necessary for starting up the image processing PC 10 and other programs and data that do not require rewriting. The RAM 23 is used as a work area of the CPU 21 and is a volatile memory that reads various data and programs from the HDD 25 and ROM 22 and temporarily stores them.

 [0055] The input / output interface 24 is connected to the operation input unit 27, the motor 4, the encoder 5, the light source 7, and the CCD camera 6 with the internal bus 28, and the operation input unit 27 inputs the operation. This interface is used to input signals and exchange various signals with the motor 4, encoder 5, light source 7 and CCD camera 6.

[0056] The HDD 25 includes an OS (Operating System), various programs for performing imaging processing and image processing, which will be described later, various other applications, and a protein chip as an inspection target image captured by the CCD camera 6. Image data such as a model image (described later) created from the image and the image to be inspected, and various data for reference in imaging processing and image processing are stored in a built-in hard disk. [0057] The display unit 26 includes, for example, an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube), and the like. The image captured by the CCD camera 6 and various image processing images are displayed. Displays the operation screen. The operation input unit 27 includes, for example, a keyboard and a mouse, and inputs an operation from a user in image processing described later.

 Next, the protein chip formed on the wafer 1 will be described. FIG. 3 is a top view of the wafer 1. As shown in the figure, for example, 88 semiconductor chips 30 (hereinafter also simply referred to as chips 30 or dies 30) are formed on the wafer 1 in a grid. Of course, the number of dies 30 is not limited to 88.

 FIG. 4 is a top view showing one of the dies 30 on the wafer 1. As shown in the figure, each die 30 is formed with a protein chip 35 having a plurality of circular recesses 50 over its entire surface. Each die 30, that is, each protein chip 35, has a substantially square shape, and the length s of one side thereof is, for example, about several mm to several tens of mm, but is not limited to this dimension.

 FIG. 5 is an enlarged view showing one recess 50 in the protein chip 35. FIG. 4A is a top view of the recess 50, and FIG. 4B is a sectional view of the recess 50 in the Z direction.

 As shown in FIG. 4 and FIG. 5, a thin film (membrane) 53 having a plurality of holes 55 is formed on the bottom surface 52 of each recess 50 of the protein chip 35. The hole 55 is formed over the entire surface of the circular bottom surface 52 of each recess 50. The diameter d 1 of each recess 50 is, for example, several hundred; U m, and the diameter d 2 of each hole 55 is, for example, several m, and the depth of the recess 50 (from the top surface 51 to the bottom surface 52) The height h) is, for example, several hundreds; U m, but is not limited to these dimensions.

[0062] In this protein chip 35, for example, latex fine particles (latex beads) are placed on the bottom surface 52 of the recess 50 as a carrier, and an antibody (protein) is introduced as a reagent into the recess 50, Silicon for screening proteins with specific properties that adsorb to latex beads by antibody cross-reaction The container made The reagent (protein) that has not adsorbed to the latex beads is discharged from each hole 55 of the bottom surface 52, and only the protein having a specific property remains in the recess 50.

[0063] Here, a method for producing the protein chip 35 will be briefly described.

 First, a thin film 53 such as a silicon oxide film is formed on one surface of the wafer 1 by a C V D (Chemical Vapor Deposition) method. Next, a photoresist is applied to the other surface of the wafer 1, unnecessary portions are removed by a photolithography technique, and etching is performed using the resist pattern as a mask. As a result, a plurality of recesses 50 are formed on the wafer 1 while leaving the thin film 53. Then, a photoresist is applied to the thin film 53 of each recess 50, the hole 55 is removed by photolithography, and etching is performed using the resist pattern as a mask. As a result, the protein chip 35 composed of a plurality of recesses 50 having the thin film 53 formed with a large number of holes 55 as shown in FIG. 5 can be formed.

 Next, an operation in which the defect detection device 100 according to this embodiment detects a defect in the protein chip 35 will be described. FIG. 6 is a flowchart showing a rough flow of operations until the defect detection apparatus 100 detects a defect.

[0065] As shown in the figure, first, the CCD camera 6 captures an image of each die 30 on which the protein chip 35 is formed at the low magnification (step 1001). Specifically, as shown in FIG. 7, each die is divided into, for example, 1 8 rows X 1 3 columns in total 2 3 4 first divided areas 7 1, and a CCD camera 6 under the flash of the light source 7 An image for each first divided area 71 is acquired. The number and aspect ratio of the first divided areas 71 are not limited to this number. Each first divided area 71 is pre-assigned with an ID for identifying its position, and the HDD 25 of the image processing PC 10 stores the ID. With this ID, the image processing PC 10 can identify the first divided area 71 existing at the same position between different dies 30. Each die 30 is also assigned an ID, and the image processing PC 10 has each first divided region 7 1 force which die 30 has which first. It is possible to identify whether it is a divided area 71.

At this time, as described above, the image processing PC 10 outputs a motor drive signal to the motor 4 based on the encoder signal from the encoder 5 to move the XYZ stage 3, and the encoder signal The trigger signal and flash signal are generated based on the above, and the trigger signal is output to the CCD camera 6 and the flash signal is output to the light source 7, respectively.

[0067] Each time the XYZ stage 3 moves, the light source 7 emits a flash to the protein chip 35 every U seconds, based on the flash signal, and the CCD camera 6 Then, based on the trigger signal, for example, the first divided areas 71 of the protein chip 35 on the wafer 1 are continuously imaged at a speed of 50 sheets / second, for example.

 FIG. 8 is a diagram showing the locus of the imaging position when the CCD camera 6 images the protein chip 35 for each first divided area 71. In this embodiment, two imaging paths are conceivable, as shown in FIGS. (A) and (b).

 [0069] As shown in (a) of the figure, the CCD camera 6 starts from the leftmost die 30 of the die 30 having the maximum Y coordinate among the 88 dies 30 of the wafer 1, for example. For example, the first divided area 71 of each of the 18 rows X 1 3 columns of the die 30 is imaged continuously for every row, for example, and then moved to the next die 30 and all the first divided regions 71 for every row. 1 divided area 7 1 is imaged.

That is, the image processing PC 10 0 sets the imaging position in each first divided area 7 1 of one die 30, for example, the first divided area 7 1 belonging to the uppermost row and the leftmost column. Move to the right in the X direction, move to the right end, move one line in the Y direction, move to the left in the X direction, move to the left end, move one line in the Y direction, and move to the next line. When the image is moved to the right in the X direction, the operation is repeated.When imaging of all the first divided areas 71 of one die 30 is completed, the image is moved to the other adjacent die 30 and the same. A motor drive signal is output to the motor 4 so that the movement is repeated. Since the position of the CCD camera 6 itself is fixed, the XYZ stage 3 is actually It will move in the opposite direction to the trajectory shown in). The CCD camera 6 continuously images each first divided area 71 based on the trigger signal output from the image processing PC 10 in accordance with this movement.

 In addition, as shown in FIG. 7B, the CCD camera 6 continuously captures images of the first divided area 71 having an ID corresponding to each die 30 (existing at the same position). The imaging position may be moved to.

 That is, the image processing PC 10 sets the imaging position of the CCD camera 6, for example, the die 30 having the maximum Y coordinate from the leftmost die 30 as a starting point, It passes over each first divided area 7 1 (first divided area 7 1 indicated by a black circle) that has an ID corresponding to 30 and exists at the position where the X coordinate is the minimum and the Y coordinate is the maximum. In the second order, each of the first divided areas 7 1 (with white circles) adjacent to the first position in the X direction and having a corresponding ID is moved in the second order. The first divided area 71 1) shown above is moved so that the CCD camera 6 repeats the movement over each first divided area 71 located at the same position between the dies 30 in the same manner. To drive motor 4. Based on the trigger signal output from the image processing PC 10 in accordance with this movement, the CCD camera 6 performs all the operations for continuously imaging the plurality of first divided areas 71 having the corresponding ID. Repeat for Die 30.

The image processing PC 10 selects a path in which the imaging time is shortened from the two imaging paths and causes the CCD camera 6 to capture an image. When the imaging path shown in Fig. 5 (a) is taken, the imaging interval of each first divided area 71 at the time of each imaging, that is, the movement interval of the XYZ stage 3 is the interval of each first divided area 71. In the case where the imaging path shown in FIG. 5B is taken, the movement interval of the XYZ stage 3 is the same as the interval of each die 30. Therefore, the CPU 21 of the image processing PC 10 can calculate the driving speed of the motor 4 from these movement intervals and the imaging frequency of the CCD camera 6. By multiplying this driving speed by the entire imaging path shown in FIGS. 8 (a) and 8 (b) determined from the layout of the die 30 shown in FIG. 3, the first divided areas 71 of all the dies 30 are obtained. Same as when shooting Figures (a) and (b) The imaging time in each case is estimated. The image processing PC 10 compares the respective imaging times to determine which imaging route (a) and (b) is used, the imaging time is faster, and the imaging time is faster. Select the imaging path.

 [0074] Then, the image of each first divided region 71 captured by the CCD camera 6 is transmitted to the image processing PC 10 as an inspection target image together with the ID for identifying each first divided region 71, It is stored in the HDD 25 or RAM 23 via the input / output interface 24 of the image processing PC 10. In this embodiment, the size of the inspection target image captured by the CCD camera 6 is a so-called VGA (Video Graphics Array) size (640 × 480 pixels) image, but is not limited to this size.

 In the present embodiment, as described above, the CCD camera 6 moves the XYZ table 3 in the Z direction, thereby changing the distance between the wafer 1 and the protein chip 35 and having different focal positions. It is possible to take an image to be inspected. Figure 9 shows the situation.

 [0076] As shown in the figure, the XYZ stage 3 is based on the focus signal from the image processing PC 10 and is directed upward (Z1 direction in the figure) and downward (Z2 direction in the figure). The distance between the CCD camera 6 and the protein chip 35 is changed to, for example, the third stage (focal points F1 to F3). That is, the CCD camera 6 moves the XYZ stage 3 in the Z2 direction so that the focal point is aligned with the upper surface 51 of the protein chip 35 (focal point F 1), and then the XYZ stage 3 moves in the Z1 direction. By moving, the focus position is adjusted to the approximate middle position between the upper surface 51 and the bottom surface 52 of the protein chip 35 (focal point F2). It is possible to align (focal point F 3). Note that the variable focal position is not limited to three.

As described above, the defect detection apparatus 100 according to the present embodiment captures images at a plurality of different focal positions so that the inspection target is the protein chip 35 according to the present embodiment. Even in the case of a three-dimensional shape having a thickness (depth or height) in the z direction, it is possible to acquire an image of each position in the z direction and prevent defect detection omission. The CCD camera 6 sorts the image of each focal position imaged by the route shown in Fig. 8 (a) or (b) for each focal position and transmits it to the image processing PC 10 for image processing. The PC 10 identifies these images as inspection target images for each focal position and stores them in the HDD 25 or RAM 23. That is, as described above, when the focal points are F1 to F3, the CCD camera 6 repeats the movement along the imaging path shown in FIG. 8 (a) or (b) three times in total for each focal position. Imaging.

 Returning to the flowchart of FIG. 6, the CPU 21 of the image processing PC 10 obtains each inspection object image from the CCD force camera 6 in parallel with the imaging process by the CCD camera 6. A filtering process using a high-pass filter is performed on each acquired image to be inspected (step 102).

 [0079] The protein chip 35 in the present embodiment has a thin film 53 on the bottom surface 52, and uneven brightness may occur depending on the flatness of the thin film 53, for example, when the thin film 53 is bent. . In addition, luminance unevenness may also occur due to the deviation of the optical axis of the CCD camera 6 or the uniformity of how the light source 7 hits the flash. Such luminance unevenness is extracted as a difference in a difference extraction process with a model image described later, which leads to erroneous detection of a defect.

 This uneven brightness portion is a portion where the brightness changes gently in the inspection target image. That is, it can be said that the luminance unevenness component is a low frequency component. Therefore, in this embodiment, a high-pass filter is applied to each captured image to be inspected to remove this low frequency component.

 FIG. 10 is a flowchart showing the detailed flow of this high-pass filter process.

As shown in the figure, first, the CPU 21 of the image processing PC 10 reads a copy of the image to be inspected from the HDD 25 to the RAM 23 (step 61), and performs Gaussian blurring processing on the image to be inspected. Apply (step 62). In addition The blur setting value is set to a radius of about 15 to 16 pixels, for example, but is not limited to this setting value.

 [0082] In this Gaussian blurring process, pixels with high frequency components (for example, edge portions) in the original image to be inspected are blurred so as to capture pixels in the surrounding low frequency components, so a high blurring effect is obtained. Will be. On the other hand, the pixels of the low frequency component (for example, the uneven brightness portion) in the original inspection target image are low frequency components in the surrounding pixels, so the blurring effect is low and compared to the original inspection target image. Little change is seen. Therefore, the output image obtained by the Gaussian blurring process (hereinafter referred to as the Gaussian blurring image) is an image in which only the low-frequency components remain as a result of smoothing the high-frequency components in the original image to be inspected.

 [0083] Subsequently, the C P U 21 subtracts the Gaussian blurred image from the original image to be inspected (step 6 3). By this subtraction process, the low frequency component at the corresponding position in the Gaussian blurred image is bowed from the high frequency component in the original inspection target image, so that the original high frequency component remains and the original inspection target. The original low frequency component is removed by subtracting the low frequency component at the corresponding position in the Gaussian blurred image from the low frequency component in the image. In other words, the image obtained by this subtraction process is an image in which only the high-frequency component remains, with the low-frequency component removed from the original image to be inspected. C PU 21 updates the original image to be inspected with the image after the subtraction process and stores it in HD 25 (step 6 4).

Returning to the flowchart of FIG. 6, the CPU 21 determines whether or not each of the images to be inspected has been imaged for all of the first divided areas 71, and performs the filtering process by the high-pass filter for all the inspections. It is determined whether or not it has been performed on the target image (steps 10 3 and 10 4). If it is determined that all of the inspection target images have been imaged and filtered (Y es), this filter The process proceeds to a process of creating a model image of each divided region using the image to be inspected after the tulling (step 10 5). In the present embodiment, the imaging process of the inspection target image and the high-pass filter process are performed in parallel. The high-pass filter processing may be performed after the image capturing processing is completed for all the first divided areas 71 (that is, the processing in step 1002 and the processing in step 103 are performed). The reverse is also possible.)

 Here, the model image creation process will be described in detail. Fig. 11 is a flowchart showing the flow of processing until the image processing PC 10 creates a model image. Fig. 12 shows how the image processing PC 10 creates a model image. It is the figure shown conceptually.

 As shown in FIG. 11, first, the CPU 21 of the image processing PC 10 has an ID corresponding to each die 30 in the inspection target image after the high-pass filter processing. The inspection target image is read from the HDD 25 to the RAM 23 (Step 4 1), and the alignment of each of the read inspection target images is performed (Step 4 2). Specifically, the CPU 21 detects, for example, an edge portion of the recess 50 of the protein chip 35 from the images to be inspected, which are images of the first divided area 71 existing at the same position between the dies 30. , Etc., and align them while adjusting by shifting in the X and Y directions and rotating in the 0 direction so that the shapes overlap each other.

 For example, as shown in FIG. 12, the CPU 21 captures each inspection object image 4 having a corresponding ID obtained by imaging the first divided area 7 1 a existing at the same position between the dies 30. Read 0 a to 40 f, ■ ■ ■ ■. In this embodiment, since the number of dies 30 is 88, the total number of inspection target images 40 having corresponding IDs is 88. The C P U 21 overlaps all of the 88 images to be inspected 40 and aligns them based on the shape of the recess 50 and the like. Thus, by performing alignment based on the shape or the like of the recess 50, easy and accurate alignment is possible.

[0088] Subsequently, the CPU 21 calculates an average luminance value for each pixel (pixel) at the same position in each inspection object image 40 in a state where the above alignment is completed (step 4 3). . When the CPU 21 calculates the average luminance value for all the pixels in each inspection target image 40 in the first divided area 7 1 a (Yes in step 44), the CPU 21 calculates the result of the calculation. Based on this, the image composed of this average luminance value is model image 45 and Is generated and stored in the HDD 25 (step 45).

The CPU 21 repeats the above processing to determine whether or not the model images 45 have been created for all the first divided regions 71 corresponding to the dies 30 (step 46). If it is determined that the model image 45 is created (Y es), the process is terminated.

 [0090] Through the above process, an absolute good product sample cannot be obtained, such as protein chip 35. Even in MEMS device inspection, model image 45 is created based on actual inspection target image 40. Can be created. Each inspection image 40 may have defects such as foreign matter, scratches, and thin film cracks. Each die 30 is divided into a plurality of (in this embodiment, 234) first divided regions 71, and a plurality of (in this embodiment, 88 in this embodiment) average luminance values for 30 dies are obtained. By calculating, the defects in each inspection target image 40 are absorbed, and a model image 45 that is very close to the ideal shape can be created, and highly accurate defect detection becomes possible.

 [0091] As described above, since each inspection object image 40 in one first divided region 71 exists for each focus of F1 to F3, a model image 45 is also created for each focus. The Therefore, in the present embodiment, since the number of the first divided regions 71 is 234 in each die 30, 234 X 3 = 702 model images are created.

 [0092] Returning to the flowchart of FIG. 6, after creating the model image 45, the CPU 2

 1 performs a difference extraction process between the model image 45 and each inspection target image 40 after the high-pass filter for each first divided area 71 (step 106).

 Specifically, the CPU 21 performs X, Y based on the shape of the concave portion 50 existing in the model image 45 and each inspection object image 40 in the same manner as the alignment process at the time of creating the model image 45 described above. Align the image while adjusting in the 0 direction, extract the difference by subtraction processing of both images, perform binarization processing, and output as a difference image.

Then, the CPU 21 performs filtering by so-called BI ob extraction on the difference image (step 107). BI ob is in the difference image A block of pixels having a predetermined (or predetermined range) grayscale value.

The CPU 21 performs processing for extracting only B I ob having a predetermined area (for example, 3 pixels) or more from the B I o b from the difference image.

 FIG. 13 is a diagram showing a difference image before and after the B I ob extraction process.

 Figure (a) shows the difference image 60 before B I ob extraction, and (b) shows the difference image after B I ob extraction (hereinafter referred to as B I ob extraction image 65).

 In FIG. 9A, a white portion is a portion that appears as a difference between the model image 45 and the inspection target image 40. In the difference image 60, in order to emphasize the difference, a process for enhancing the luminance value, for example, about 40 times is performed on the original difference image. As shown in (a) of the figure, in the differential image 60 before BI ob extraction, in addition to defects such as foreign matter and scratches, for example, contamination of the lens 14 of the CCD camera 6 and illumination of the light source 7 Due to various factors such as uniformity, there is a small noise 84 shown in the part surrounded by a white broken line. If this noise 84 remains, it will lead to false detection of a defect, so it is necessary to remove this noise 84.

 [0097] The noise 84 has a smaller area than a defect such as a foreign object or a scratch. Therefore, as shown in FIG. 6B, the noise 60 is obtained by filtering the difference image 60 by removing BI ob below a predetermined area and extracting only BI ob larger than the predetermined area. 84 can be removed. By this B I ob extraction process, only the foreign matter 82 such as dust 81 attached to the protein chip 35 and the thin film crack 81 of the concave portion 50 of the protein chip 35 are extracted from the B l ob extraction image 65. At this time, the CPU 21 does not recognize the types of defects such as these foreign objects, cracks, and scratches, but merely recognizes them as defect candidates.

[0098] Subsequently, returning to the flowchart of FIG. 6, when a defect candidate is detected by the BI ob extraction process (Yes in step 108), the CPU 21 detects the protein chip in which this defect candidate is detected. Determine whether further images need to be taken at high magnification (with a narrow field of view) 35 (step 110). CPU 2 1 determines, for example, whether or not a user operation instructing that the first divided area 71 to which the inspection target image 40 where the defect candidate appears belongs belongs to be captured in more detail at a high magnification is input. If it is determined that high-magnification imaging is necessary (Y es), the first divided area 71 in which the defect candidate is detected and the other first divided area 71 having an ID corresponding to each die 30. Is imaged by the CCD camera 6 at a high magnification for each of the second divided areas 72 divided further finely (step 11 13).

 [0099] In the defect classification process described later, for example, based on the extracted BI ob area, it is determined whether or not it is a defect and the defect is classified, but the inspection object image captured at a low magnification is used. In the BI ob extracted image 65 created based on the above, the BI ob area may not be calculated accurately. In addition, when imaging at a low magnification, the correct shape of the defect cannot be recognized and the defect cannot be accurately classified. Therefore, by imaging the protein chip 35 at a higher magnification, it is possible to accurately determine whether or not it is a defect later and to perform defect classification processing.

 [0100] FIG. 14 is a diagram conceptually showing a state in which the first divided region 71 in which the defect candidate is detected is imaged at a high magnification for each second divided region 72. As shown in the figure, for example, when a defect candidate is detected from an inspection target image obtained by imaging a first divided region 71a having a certain die 30, this first divided region 71a is Furthermore, it is divided into a total of nine second divided areas 72 in 3 rows and 3 columns. Further, in the other dies 30, the first divided regions 71 having IDs corresponding to the first divided regions 71 a are also divided into the second divided regions 72, respectively. Each second divided region 72 is given an ID for identifying the position of each second divided region 72 in each die 30, similarly to each first divided region 71.

 [0101] The C C D camera 6 images each of the second divided areas 72 at the same size (VGA size) as the first divided area 71. That is, the C C D camera 6 images the second divided region 72 at a magnification three times that when the first divided region 71 is imaged. The captured image is stored in the HD 25 of the image processing PC 10 as the inspection target image together with the ID of each of the second divided regions 72.

[0102] Regarding the imaging path of each second divided area 7 2 of each die 30, CPU 2 1 selects the fastest path (a) or (b) in FIG. 8 as in the case of imaging of the first divided area 71. In other words, the CPU 21 captures all the second divided areas 72 in the divided area 71 of the one die 30 and then each second divided area 71 in the corresponding divided area 71 of the other die 30. Among the path to image the area 72 and the path to collectively image each second divided area 72 having the corresponding ID between the corresponding first divided areas 71 of each die 30. Determine which route is faster, and have the image taken faster.

 [0103] When the imaging of the first divided area 71 1 in which the defect candidate is detected and the second divided area 7 2 in the first divided area 71 corresponding thereto is finished (step 1 1 3), the CPU 2 1 Similar to the processing in steps 10 2 to 10 7 described above, filtering processing (step 1 1 4) and model image creation processing (step 1 1 7) using a high-pass filter are performed for each inspection target image, Difference extraction processing is performed on each inspection target image obtained by imaging each second divided region 72 in the first divided region 71 in which the defect candidate is detected (step 1 1 8), and further by BI ob extraction. Perform the filtering process (step 1 1 9).

 [0104] Since each inspection target image in the second divided area 72 is captured at a higher resolution than the inspection target image in the first divided area 71, the BI ob extraction process in step 1 1 8 The threshold value (pixel) of the BI ob region extracted in step B is set to a value larger than the threshold value of the BI ob region extracted in the BI ob extraction process for each first divided region 71 in step 107. Of course, there is no change in the actual area (; U m) of BIob on the protein chip 35 converted by the threshold value (pixel).

FIG. 15 is a diagram showing a comparison of each BI ob extracted image 65 extracted from each inspection target image in the first divided region 71 and the second divided region 72. BI _0! 3 extracts image 6 5 ₃ Fig. (A) is extracted from the first division region 7 1, Fig. (B) is a BI ob extracted image 6 5 b extracted from the second divided region 7 2 Show.

[0106] As shown in the figure, in the BI ob extracted image 65a of the first divided region 71a extracted at the above step 107, it is considered that the foreign object 82 is at the left end and the lower end portion. Although the area that appears is white, it is difficult to calculate an accurate area value because the area is very small. Therefore, as shown in FIG. 4B, the first divided area 71 is divided into nine second divided areas 72, and the second divided area 7 where the foreign matter 82 appears is shown. By imaging 2 at a high magnification, the foreign object 82 is displayed at a high resolution, and the area can be accurately calculated.

It should be noted that, when the defect candidate is extracted in step 10 08 without performing the above-described determination of necessity of high-magnification imaging in step 109, the high-magnification imaging is automatically performed. May be. In addition, when the performance of the image processing PC 10, the motor 4 and the encoder 5 is high and the processing time is within the allowable range, not only the first divided area 71 from which the defect candidate is extracted, All die 3 0 all first

The second divided area 7 2 in the one divided area 71 may be imaged and the model image 45 may be created for all the second divided areas 7 2. In this case, immediately after the BI ob extraction process for the first divided area 71 is completed, the imaging process for each of the second divided areas 7 2 is performed without determining whether or not high magnification imaging is required in the above steps 1 0 9. If the CPU 21 determines that there is a first divided area 7 1 in which a defect candidate is detected after performing high-pass filter processing and model image creation processing, the first

The B Iob extraction process may be performed for each second divided area 7 2 in the one divided area 71.

 [0108] Returning to the flowchart of FIG. 6, if it is determined in step 1009 that high-magnification imaging is not necessary (No), or the second divided region in steps 113 to 1119 above When the BI ob extraction process from 72 is completed, the CPU 21 performs the classification process of defect candidates appearing in the Biob extraction image 65 (step 110).

 That is, the CPU 21 is based on feature points such as area, perimeter, non-roundness, aspect ratio, etc., for each BI ob that appears white in the BI ob extracted image 65. Whether or not each BI ob is a defect is classified and whether the type of the defect is a foreign object, a flaw or a crack.

[01 10] Specifically, the image processing PC 10 is provided for each type of defect, such as a foreign object, a flaw, or a crack. These sample images are collected and their feature point data is saved as a feature point database in HDD 25, etc., and the saved feature point data and each BI ob extracted image 6 5 in the inspection target BI 5 Compare with each feature point detected from ob.

 [01 1 1] For example, the foreign matter in the present embodiment has a number of about several sides to about several tens; and the scratch has a length of several numbers; about U m to several hundreds. In addition, when comparing foreign objects with scratches, scratches have an aspect ratio that is extremely horizontal or vertical and has a longer perimeter than foreign objects. Furthermore, the thin film cracks appear in a curved line at the edge of the recess 50, but the roundness of the recess 50 is lower than that of a normal one. The image processing PC 10 stores these data as feature point data, and classifies the defects by comparison with each feature point of the detected BIo.

 [0112] Further, as described above, the protein chip 35 according to the present embodiment has the hole 55 having a diameter of, for example, a few; U m on the thin film 53 of the bottom surface 52 of the recess 50. The holes 5 5 serve to discharge the reagent. Therefore, even when foreign matter is adhered in the recess 50, if the diameter is smaller than the number of holes 55, the protein chip 35 is used because it is discharged from the holes 55 together with the reagent. There is no problem during screening. Therefore, for foreign objects, the diameter of the hole 55 is used as a threshold value, and foreign objects with a smaller diameter are not treated as defects. On the other hand, scratches and cracks are treated as defects unconditionally because normal screening cannot be performed due to the leakage of the reagent from there.

 [0113] As described above, in the BI ob extracted image 65 extracted from the first divided area 71, when the feature point cannot be measured accurately, the CPU 21 further increases the magnification. The feature points are measured using the BI ob extracted image 65 extracted from the captured second divided area 72, and various defects are classified. In this way, by performing high-magnification imaging as necessary, processing after defect detection can be performed smoothly.

[01 14] Then, the CPU 21 judges whether or not there is a defect for all defect candidates, and When a defect is classified (YES in Step 1 1 1), the information on the BI ob extracted image and the detected defect type is output to the display unit 26 as a detection result (Step 1 1 2), End the process. At this time, for example, the image processing PC 10 may display an image on the display unit 26 so that it can be recognized at a glance which type of defect exists at which position on the wafer 1.

[0115] On the basis of the output result, the user performs the removal work if there is a foreign object, and discards the protein chip 35 as a defective product if there is a scratch or crack. If no defect candidate is detected in step 108, the protein chip 35 to be inspected is processed as a non-defective product, and the defect detection process ends.

 [0116] Through the above operation, according to the present embodiment, even in the MEMS device such as the protein chip 35 where it is difficult to obtain a good sample, each first divided region 71 or each second divided region 72 Since a model image can be created based on the inspection target image 40, a highly accurate defect detection process can be performed. In addition, since the model image 45 is created based on each inspection object image 40 taken under the same optical conditions and illumination conditions, it is possible to prevent erroneous detection due to a difference in these conditions.

 [0117] The present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the scope of the present invention.

 [0118] In the above embodiment, the protein chip is applied as the MEMS device to be inspected, but the MEMS device is not limited to this. For example, an electron beam irradiation plate (EB window) can be applied as a MEMS device.

 FIG. 16 is a view showing the appearance of this electron beam irradiation plate. The figure (a) is a top view, and the figure (b) shows a cross-sectional view in the Z direction in the figure (a).

[0120] As shown in the figure, the electron beam irradiation plate 90 includes a plate 92 having a plurality of window holes 95 for irradiating an electron beam (EB), and each window hole 95. And a thin film 9 1 provided so as to cover.

[0121] The length w in the X direction and the length I in the Y direction of the plate 9 2 are each formed in a rectangular shape of, for example, several tens mm, and the length h in the Z direction is, for example, about several mm. It is not restricted to these lengths and shapes. Each window 95 is, for example, a square with a side s of several millimeters, but is not limited to this length and shape, and may be rectangular. The number of window holes 9 5 is 6 in 4 rows x 9 columns, but is not limited to this number.

 [0122] This electron beam irradiation plate 90 constitutes an electron beam irradiation apparatus by being connected to an end of a vacuum vessel (not shown). Electron beam (EB) force emitted from the electron beam generator provided inside the vacuum vessel is emitted into the atmosphere through the window hole 95 as indicated by the arrow in FIG. Is irradiated. This electron beam irradiation apparatus is used for various purposes such as sterilization, physical property modification, and chemical property modification of an object irradiated with an electron beam. By providing the thin film 91, it is possible to irradiate an electron beam while maintaining a vacuum state. Note that a plurality of thin films 91 may be stacked to form a multilayer film structure.

 [0123] This electron beam irradiation plate 90 is also formed on each die 30 on the wafer 1 by an etching process using a photolithography technique or the like, similar to the protein chip 35 in the above-described embodiment. In this case, the size of each die is the same as the size of the plate 92.

 [0124] The defect detection apparatus 10 0 0 also performs the same imaging process, high-pass filter process, model image creation process, BI ob extraction process as the above-described protein chip 35 for the electron beam irradiation plate 90. Etc., and detect defects such as foreign matter, scratches and cracks on the electron beam irradiation plate 90. It is also possible to take images at low and high magnifications and at multiple focal points in the Z direction. In the model image creation processing and BI ob extraction processing, each inspection target image is adjusted in the X, Y, and 0 directions so that the edge shapes of the window holes 95 appearing in each inspection target image overlap. While aligning.

[0125] In the inspection of the electron beam irradiation plate 90, for example, foreign matter Unlike the case of the above inspection of protein chip 35, the image processing PC 10 has a sample of the electron beam irradiation plate 90, etc. Based on this, original feature data is created to classify defects.

 [0126] Besides the protein chip 35 and electron beam irradiation plate 90 described above, various sensors such as an acceleration sensor, a pressure sensor, an air flow sensor, a printer head for an ink jet printer, and a reflective projector. It is possible to apply other MEMS devices such as micro-mirror arrays for use, other actuators, and various biochips as inspection objects.

 In the above-described embodiment, each image necessary for image processing such as the inspection target image 40 and the model image 45, the difference image 60 and the BI ob extracted image 65 is stored in the HDD 25. However, these images may be temporarily stored in the RAM 23, or temporarily stored in a buffer area provided separately from the RAM 23, and deleted when the defect classification process is completed. It doesn't matter if you do. In addition, among the images to be inspected above, it was found that the image in which the difference was not extracted by the difference extraction, that is, the image in which the defect was not detected was not detected because it is unnecessary in the subsequent processing. You may delete them one at a time. Furthermore, for the image to be inspected in the first divided region 71 captured at a low magnification, when the second divided region 72 is imaged at a higher magnification, after the second divided region 72 is imaged, the first divided region 71 Since the images to be inspected 71 are not necessary, these images may be deleted when the imaging of the second divided region 72 is completed. In the above embodiment, since the number of images to be captured is enormous, it is possible to reduce the load on the image processing PC by reducing the storage capacity of RAM 23 and HDD 25 by processing in this way. It becomes.

In the above-described embodiment, the filtering process using the high-pass filter is performed in consideration of the case where the thin film 53 of the bottom surface 52 of each of the recesses 50 of the protein chip 35 is bent. The flatness of the surface is high and brightness unevenness does not occur In such a MEMS device, this high-pass filter processing may be omitted. Further, the image processing PC 10 may measure the flatness of the imaging surface of the MEMS device to be inspected and determine whether or not to execute the high-pass filter processing according to the flatness.

[0129] In the filtering process using BI ob extraction in the above-described embodiment, for example, if a scratch or a crack of the thin film 53 appears continuously (in a line) in the differential image 60, it is normally recognized as a defect. Although it is possible, scratches and cracks may appear to fade as a continuum of dotted BI ob. In such a case, if each point-like BIob is smaller than the threshold value, the image processing PC 10 may not be recognized as a defect. Therefore, even if BI ob is smaller than the above threshold, BI ob that appears continuously in a predetermined direction (straight line or curved line) at predetermined intervals, such as scratches and cracks, is extracted in the Biob extraction process. It doesn't matter. Thereby, it is possible to detect a defect with higher accuracy.

 Brief Description of Drawings

FIG. 1 is a diagram showing a configuration of a defect detection apparatus according to an embodiment of the present invention.

 FIG. 2 is a block diagram showing a configuration of an image processing PC in an embodiment of the present invention.

 FIG. 3 is a top view of a wafer in one embodiment of the present invention.

 FIG. 4 is a top view showing one of the dies on the wafer in one embodiment of the present invention.

 FIG. 5 is an enlarged view showing one recess of a protein chip according to an embodiment of the present invention.

 FIG. 6 is a flowchart showing a general flow of operations until a defect detection apparatus detects a defect in an embodiment of the present invention.

 FIG. 7 is a diagram showing a state in which each die is divided into a plurality of divided regions in an embodiment of the present invention.

FIG. 8 is a diagram showing a locus of an imaging position when a CCD camera images a protein chip for each divided region in an embodiment of the present invention. FIG. 9 is a diagram showing a state in which the CCD camera captures an inspection object image at different focal positions in an embodiment of the present invention.

 FIG. 10 is a flowchart showing a detailed flow of high-pass filter processing in one embodiment of the present invention.

 FIG. 11 is a flowchart showing the flow of processing until the image processing PC creates a model image in an embodiment of the present invention.

 FIG. 12 is a diagram conceptually showing how an image processing PC creates a model image in an embodiment of the present invention.

 FIG. 13 is a view showing a difference image before and after the BIob extraction process in one embodiment of the present invention.

 FIG. 14 is a diagram conceptually showing a state in which a first divided area where a defect candidate is detected is imaged at a high magnification for each second divided area in one embodiment of the present invention.

 FIG. 15 is a diagram showing a comparison of each B Iob extracted image extracted from each inspection target image in the first divided region and the second divided region in the embodiment of the present invention.

 FIG. 16 is a view showing the appearance of an electron beam irradiation plate in another embodiment of the present invention.

Explanation of symbols

 1 ... Semiconductor wafer (wafer)

 3 ·· Χ Y Z stage

 4 ... Motor

 5 ... Encoder

 6- CCD camera

 7 ... Light source

 1 0… For image processing P C

 1 4 ... Lens

 2 1 ".C P U

 23 --- RAM

24 ... I / O interface -■ HDD

.. • Die (Semiconductor chip, --- Protein chip ..

-'Model image

.. • Recess

■■ • Top

■■ • Bottom

9 1… Thin film

-.Hole

.. • Difference image

-• B i ob extracted image ■■ • First divided area ■■ • Second divided area ■■ • Crack

■■ • Foreign matter

-- noise

.. • Electron beam irradiation play ■■ -Plate

-Trick

0 ... Defect detection device

Claims

The scope of the claims

 [1] An imaging means for imaging the microstructure formed on each of the plurality of dies on the semiconductor wafer for each divided region obtained by dividing the region of each die into a plurality of regions, and the microstructure to be imaged Illumination means for illuminating;

 Storage means for storing the imaged image of each divided area as an image to be inspected in association with identification information for identifying the position of each divided area in each die;

 A first filtering unit that performs filtering for removing the low-frequency component in each of the stored inspection target images, and each of the filtered inspection target images; Model image creating means for creating an average image obtained by averaging each inspection target image in each divided region corresponding to the identification information between dies for each identification information as a model image; and each created model image And detecting means for detecting defects in the microstructure by comparing the filtered image to be inspected with the identification information corresponding to the model images.

 A defect detection apparatus comprising:

 [2] The defect detection apparatus according to claim 1,

 The detection means includes

 Difference extracting means for extracting a difference between each model image and each inspection target image corresponding to the identification information corresponding to each model image as a difference image;

 A second filtering means for performing filtering for removing a pixel area having a luminance equal to or greater than a predetermined value in the extracted difference image from a series of pixel areas having a luminance equal to or higher than a predetermined value;

 A defect detection apparatus comprising:

[3] The defect detection apparatus according to claim 2,

The defect detection apparatus, wherein the model image creation means includes means for calculating an average value of luminance values for each pixel constituting each inspection object image corresponding to the identification information.

[4] The defect detection apparatus according to claim 2,

 The defect detection apparatus, wherein the imaging unit continuously images the microstructure in each divided region having identification information corresponding to each die.

[5] The defect detection apparatus according to claim 2,

 The imaging means captures the microstructures in each divided region of the other die adjacent to the one die after imaging the microstructures in all the divided regions in the one die. Feature defect detection device.

[6] The defect detection apparatus according to claim 2,

 The microstructure includes a plurality of recesses having a thin film bottom surface for introducing a reagent and an antibody that cross-reacts with the reagent, and a bottom surface of each recess for discharging the reagent that does not react with the antibody. A defect detection apparatus comprising a plurality of holes, and a screening inspection container.

[7] The defect detection apparatus according to claim 6,

 The model image creating means includes the identification information corresponding to each model image, prior to the averaging of each image to be inspected, based on the shape of each recess in the container in each image to be inspected. A defect detection device comprising means for aligning images.

[8] The defect detection apparatus according to claim 6,

 Prior to the extraction of the difference, the difference extraction means includes a shape of each concave portion of the container in each model image and each concave portion in each inspection target image corresponding to the identification information corresponding to each model image. A defect inspection apparatus comprising means for aligning each model image and each inspection object image based on a shape

[9] The defect detection apparatus according to claim 2,

 The microstructure is an electron beam irradiation plate having a plate member having a plurality of window holes for irradiating a plurality of electron beams, and a thin film provided so as to cover the window holes. Feature defect detection device.

[10] The defect detection device according to claim 9, The model image creating means is configured to determine the shape of each window hole of the electron beam irradiation plate in each inspection target image prior to averaging each inspection target image corresponding to the identification information corresponding to each model image. A defect detection apparatus comprising means for aligning each inspection object image on the basis thereof.

 [11] The defect inspection apparatus according to claim 9,

 Prior to the extraction of the difference, the difference extraction means includes a shape of each window hole of the electron beam irradiation plate in each model image, and each inspection object image corresponding to the identification information corresponding to the model image. A defect inspection apparatus comprising means for aligning each model image and each image to be inspected based on the shape of each window hole therein.

 [12] A step of imaging the microstructure formed on each of the plurality of dies on the semiconductor wafer for each divided region obtained by dividing the region of each die into a plurality of regions, and illuminating the microstructure to be imaged And steps to

 A step of storing the imaged image of each divided region in association with identification information for identifying the position of each divided region in each die as an inspection target image;

 Filtering each of the stored images to be inspected to remove a low frequency component in each image to be inspected;

 Among the filtered inspection target images, an average image obtained by averaging the inspection target images of the divided areas corresponding to the identification information between the dies is created as a model image for each identification information. Steps,

 Comparing the created model images with the filtered inspection object images corresponding to the identification information corresponding to the model images, and detecting defects in the microstructures;

 A defect detection method comprising:

[13] The defect detection method according to claim 12,

 The detecting step includes

Each model image, and each inspection corresponding to the identification information corresponding to each model image Extracting a difference from the target image as a difference image;

 A step of performing filtering for removing a pixel area of a predetermined area or less from a series of pixel areas having a luminance of a predetermined value or more in the extracted difference image;

 A defect detection method characterized by comprising:

 [14] The microstructure formed on each of the plurality of dies on the semiconductor wafer is an image obtained under illumination for each of the divided regions obtained by dividing each of the dies into the plurality of dies. Storage means for storing as an image to be inspected in association with identification information for identifying the position of the divided area;

 Filtering means for performing filtering for removing low frequency components in each inspection target image for each stored inspection target image, and among each filtered inspection target image, between each die Model image creation means for creating an average image obtained by averaging the images to be inspected in each divided region corresponding to the identification information as a model image for each of the identification information, each created model image, Detecting means for detecting defects of the microstructure by comparing the filtered image to be inspected with the identification information corresponding to each model image;

 An information processing apparatus comprising:

[15] Microstructures respectively formed on a plurality of dies on a semiconductor wafer, images obtained under illumination for each of the divided regions obtained by dividing each of the dies into a plurality of the respective dies in each of the dies. Storing as an image to be inspected in association with identification information for identifying the position of the divided region;

 Filtering each of the stored images to be inspected to remove a low frequency component in each image to be inspected;

 Among the filtered inspection target images, an average image obtained by averaging the inspection target images of the divided areas corresponding to the identification information between the dies is created as a model image for each identification information. Steps,

Each identification image corresponds to each created model image and each model image Comparing the filtered images to be inspected with each other, and detecting a defect in the microstructure.

 An information processing method comprising:

 In the information processing device,

 Microstructures respectively formed on a plurality of dies on a semiconductor wafer are images obtained by illuminating each divided region obtained by dividing each die into a plurality of divided regions. Storing as an image to be inspected in association with identification information for identifying a position;

 Filtering each of the stored images to be inspected to remove a low frequency component in each image to be inspected;

 Among the filtered inspection target images, an average image obtained by averaging the inspection target images of the divided areas corresponding to the identification information between the dies is created as a model image for each identification information. Steps,

 Comparing the created model images with the filtered inspection object images corresponding to the identification information corresponding to the model images, and detecting defects in the microstructures;

 A program for running