JP2004085503A - Classifying apparatus, yield control system, classifying method, substrate manufacturing method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a classifying apparatus capable of easily classifying specific types of defects. <P>SOLUTION: The classifying apparatus 1 comprises an imaging part 2 for imaging a substrate 9; an arithmetic part 4 for receiving the input of an image signal from the imaging part 2; and a computer 5. An arithmetic part 4 stores an image to be inspected acquired by the imaging part 2 and a reference image. The arithmetic part 4 creates a defect region image indicating a defect region from the image to be inspected and the reference image. The computer 5 generates a division region image indicating the region of a wiring pattern on the substrate 9 from the reference image. A defect feature amount extracting part 502 performs logical operation for each pixel of the defect region image and the division region image, to determine an evaluation value indicating the state of superposition between the defect region and a division region. A classifier 503 classifies defects, on the basis of the evaluation value. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a technique for classifying defects of a pattern to be inspected on an object.
[0002]
[Prior art]
2. Description of the Related Art In the field of inspecting a pattern formed on a semiconductor substrate, a color filter, a shadow mask, a printed wiring board, and the like (hereinafter, referred to as a “substrate”), a comparative inspection method using a multi-valued image has been mainly used. For example, a difference absolute value image indicating an absolute value of a pixel value difference between the inspection image and the reference image is obtained, and an area having a pixel value larger than a predetermined threshold value in the difference absolute value image is detected as a defect. .
[0003]
On the other hand, in order to correct a manufacturing process abnormality that causes a reduction in the yield in each step, the detected defects are sometimes classified (that is, the types and categories of the defects are determined). . As one method of the classification process, a feature amount derived from the entire defect such as the brightness near the defect in the image to be inspected used in the inspection and the area of the defect is calculated, and a neural network is used based on the feature amount. There is known a method of performing classification.
[0004]
[Problems to be solved by the invention]
By the way, when a defect detected by the comparative inspection method is classified using a feature amount derived from the entire defect, a specific type of defect cannot be classified. For example, when the inspection images 921, 922, and 923 shown in FIG. 1B are acquired with respect to the reference image 91 shown in FIG. 1A, the absolute differences corresponding to the inspection images 921 to 923 are obtained. The value images 931, 923, and 933 are acquired as shown in FIG. In FIGS. 1A and 1B, a region denoted by reference numeral 9a is a wiring region on the substrate, and a region denoted by reference numeral 9b is a background region.
[0005]
Then, in the difference absolute value images 932 and 933 of FIG. 1C, the areas denoted by reference numerals 932a and 933a are detected as areas indicating defects, respectively, and feature amounts (for example, defects) obtained from the entire areas 932a and 933a are obtained. Area and brightness) are used for defect classification. At this time, classification based on the area, brightness, and the like of the defect is possible, but it is difficult to classify the relative geometric relationship (including inclusion relationship, connection relationship, and the like) of the defect with respect to the wiring region 9a and the background region 9b. It is.
[0006]
For example, the defect area 922a of the image to be inspected 922 corresponding to the area 932a of the difference absolute value image 932 is an area (so-called isolated point (area)) included in the background area 9b, and the area of the difference absolute value image 933. The defect area 923a of the inspected image 923 corresponding to the area 933a is an area that divides the background area 9b (that is, an area that short-circuits the wiring areas 9a). Was difficult to do.
[0007]
The present invention has been made in view of the above problems, and has as its object to provide a technique for easily classifying a specific type of defect.
[0008]
[Means for Solving the Problems]
The invention according to claim 1 is a classification device for classifying defects of a pattern to be inspected on an object, comprising: data of a defect area image indicating a defect area of the pattern to be inspected on the object; A storage unit for storing data of a divided area image indicating an existing area as a divided area, and performing a logical operation for each pixel between an image based on the defective area image and an image based on the divided area image to overlap the defective area and the divided area. An evaluation value calculation unit that obtains an evaluation value indicating a state, and a classification unit that classifies defects on the pattern to be inspected based on the evaluation value are provided.
[0009]
According to a second aspect of the present invention, there is provided the classification device according to the first aspect, wherein the defect area generating the defect area image by comparing the inspection image indicating the inspection pattern with the reference image indicating the reference pattern. An image generation unit is further provided.
[0010]
According to a third aspect of the present invention, there is provided the classification device according to the second aspect, further including an imaging unit configured to capture an object and acquire data of an image to be inspected.
[0011]
According to a fourth aspect of the present invention, there is provided a yield management system for managing a production yield of an object on which a pattern is formed. And a storage unit for storing them in association with each other.
[0012]
The invention according to claim 5 is a classification method for classifying a defect of a pattern to be inspected on an object, a step of preparing data of a defect area image indicating a defect area of the pattern to be inspected on the object, A step of preparing data of a divided area image indicating a region where the reference pattern is present as a divided area; and performing a logical operation for each pixel of an image based on the defective area image and an image based on the divided area image to obtain the defective area. A step of obtaining an evaluation value indicating an overlapping state with the divided area; and a step of classifying defects on the pattern to be inspected based on the evaluation value.
[0013]
The invention according to claim 6 is a substrate manufacturing method for manufacturing a substrate on which a pattern is formed, wherein the step of taking out the substrate from a loading position, the step of forming a pattern on the substrate, and the step of imaging the substrate Obtaining the data of the image to be inspected indicating the pattern to be inspected, and generating a defect area image indicating the defect area of the pattern to be inspected by comparing the image to be inspected with the reference image indicating the reference pattern; An evaluation indicating the overlapping state of the defective area and the divided area by performing a logical operation for each pixel between the image based on the defective area image and the image based on the divided area image indicating the existence area of the reference pattern as the divided area. A step of obtaining a value, a step of classifying a defect with respect to the pattern to be inspected based on the evaluation value, and a step of transporting the substrate to an unloading position.
[0014]
According to a seventh aspect of the present invention, there is provided a program used for classifying a defect of a pattern, wherein execution of the program by a computer prepares data of a defect area image indicating a defect area of a pattern to be inspected in the computer. Performing the step of preparing data of a divided area image indicating the existence area of the reference pattern as the divided area, and performing a logical operation for each pixel between the image based on the defective area image and the image based on the divided area image A step of obtaining an evaluation value indicating an overlapping state between the defect area and the divided area; and a step of classifying defects on the pattern to be inspected based on the evaluation value.
[0015]
The invention according to claim 8 is the program according to claim 7, wherein the image based on the defect area image is an image obtained by expanding or contracting the defect area.
[0016]
A ninth aspect of the present invention is the program according to the eighth aspect, wherein the evaluation value is obtained according to a degree of expansion or contraction of the defect area.
[0017]
According to a tenth aspect of the present invention, there is provided the program according to the seventh or eighth aspect, wherein the evaluation value is obtained according to an area of a region where the defect region and the divided region overlap.
[0018]
According to an eleventh aspect of the present invention, there is provided the program according to any one of the seventh to tenth aspects, wherein the execution of the program by the computer prepares data of the image to be inspected indicating the pattern to be inspected in the computer. And preparing a reference image data indicating the reference pattern, and comparing the inspected image with the reference image to generate the defect area image.
[0019]
A twelfth aspect of the present invention is the program according to the eleventh aspect, wherein the defect area image is generated from a difference image between the inspection image and the reference image.
[0020]
The invention according to claim 13 is the program according to claim 11 or 12, wherein the execution of the program by the computer generates the divided area image by binarizing the reference image in the computer. The process is performed further.
[0021]
According to a fourteenth aspect of the present invention, there is provided the program according to any one of the eleventh to thirteenth aspects, wherein execution of the program by the computer causes the computer to execute the inspection image, the reference image, and the defect area. A step of obtaining an image feature based on at least one of the images is further executed, and the defect is classified based on the evaluation value and the image feature.
[0022]
The invention according to claim 15 is the program according to claim 14, wherein the step of classifying the defect includes the step of performing coarse classification based on the evaluation value, the result of the coarse classification, and the image. Performing a detailed classification based on the characteristic amount.
[0023]
The invention according to claim 16 is the program according to claim 14, wherein the step of classifying the defect includes a step of specifying a defect or a non-defect based on the image feature amount, and a step of specifying the defect. Performing a classification of the defect based on the evaluation value when the defect is detected.
[0024]
The invention according to claim 17 is the program according to claim 11 or 12, wherein the step of classifying the defect includes a step of performing a plurality of classifications to obtain a plurality of classification results; Acquiring a final classification result based on the classification result of the above.
[0025]
BEST MODE FOR CARRYING OUT THE INVENTION
FIG. 2 is a diagram showing a substrate manufacturing system 7 that plays a part in a process of manufacturing a semiconductor substrate (hereinafter, referred to as “substrate”). The substrate manufacturing system 7 processes the substrate to form a pattern on the substrate by forming a pattern on the substrate (the processing device 71 may be a plurality of processing devices), images the formed pattern on the substrate, and detects a pattern defect. And a loader 70a that includes a management device 72 connected to the classification device 1 and that transfers a substrate placed at a predetermined loading position to the processing device 71, and a classification device. The apparatus further includes an unloader 70b that receives the substrate on which the defect has been classified by the apparatus 1 and places the substrate at a predetermined unloading position.
[0026]
The management device 72 displays a management unit 73 including a CPU for performing various arithmetic processes and a memory for storing various information, a storage unit 74 for storing information transmitted from the classification device 1, and various types of information. It has a display 75. The management unit 73 has a monitoring unit 731 that manages the yield of the substrate, and the monitoring unit 731 receives the defect classification result transmitted from the classification device 1 and checks the inspection item, the type of the substrate, the inspection location, or the manufacturing process. The process is stored in the storage unit 74 in association with management items such as process conditions and a period from the start of manufacturing (used as information indicating whether or not the system is in a start-up period), and the defect classification database 741 is updated. I do.
[0027]
The monitoring unit 731 further monitors the statistical information derived from the defect classification database 741 and, for example, notifies the worker when the yield decreases after the start-up period ends. The operator refers to the defect classification database 741 via the display 75, grasps the type of the defect causing the yield reduction, and responds. As described above, the classification device 1 and the management device 72 constitute a yield management system.
[0028]
FIG. 3 is a diagram showing a flow of the substrate manufacturing process by the substrate manufacturing system 7. First, when the substrate is placed at the carry-in position (Step S11), the substrate is taken out of the carry-in position by the loader 70a and transported to the processing device 71 (Step S12). In the processing device 71, a predetermined process is performed on the substrate, and a wiring pattern is formed on the substrate (Step S13). After that, the substrate is transported to the classification device 1, where the wiring pattern on the substrate is inspected by the classification device 1, and the detected defects are classified (step S14). The details of the process of the classification device 1 will be described later. The defect classification result is transmitted to the management device 72, and the classification result is added to the defect classification database 741. The substrate is transported to the unloader 70b, and is placed at a predetermined unloading position (Step S15).
[0029]
As described above, in the substrate manufacturing system 7, the detection and classification of the defect of the pattern formed on the substrate are performed by the classification device 1, and the classification result is output to the management device 72. The management device 72 additionally stores the classification result in the defect classification database 741, and thereby easily and quickly outputs data related to the manufacturing yield, such as the type of defect for each item and the frequency of occurrence of various defects, to an operator or outside the system. The device can be acquired, and can be used, for example, for drafting a production plan in a post-process. In addition, quick and accurate diagnosis and correction of the manufacturing process can be performed, and the substrate manufacturing yield can be improved.
[0030]
FIG. 4 is a diagram illustrating a configuration of the classification device 1. The classifying apparatus 1 includes an imaging unit 2 that captures a predetermined area on the substrate 9 and obtains data of a multi-tone object image, a stage 3 that holds the substrate 9, and a stage 3 for the imaging unit 2. Is relatively provided.
[0031]
The image pickup unit 2 includes an illumination unit 21 that emits illumination light, an optical system 22 that guides the illumination light to the substrate 9 and receives light from the substrate 9, and an image of the substrate 9 formed by the optical system 22. It has an imaging device 23 for converting into a signal. The stage drive unit 31 has an X-direction moving mechanism 32 that moves the stage 3 in the X direction in FIG. 4 and a Y-direction moving mechanism 33 that moves the stage 3 in the Y direction. The X-direction moving mechanism 32 is connected to a motor 321 via a ball screw (not shown). When the motor 321 rotates, the Y-direction moving mechanism 33 moves along the guide rail 322 in the X direction in FIG. The Y direction moving mechanism 33 has the same configuration as the X direction moving mechanism 32. When the motor 331 rotates, the stage 3 moves in the Y direction along the guide rail 332 by a ball screw (not shown).
[0032]
The classification device 1 further includes an arithmetic unit 4 configured by an electric circuit, and a computer 5 configured by a CPU that performs various arithmetic processes, a memory that stores various information, and the like. The calculation unit 4 acquires an electric signal of the target object image from the imaging unit 2 to detect a defect and generates a defect area image used for defect classification. The computer 5 classifies the defect and performs classification. It plays a role as a control unit that controls each component of the device 1.
[0033]
4, a changeover switch 40, an image to be inspected memory 411, a reference image memory 412, a difference calculation circuit 42, a difference statistic calculation circuit 43, a normalization circuit 44, a threshold memory 413, a comparison circuit 45, and a defective area image memory Reference numeral 414 denotes the configuration of the arithmetic unit 4, and the divided region calculation unit 501, divided region image memory 531, defect feature amount extraction unit 502, and classifier 503 indicate the configuration or function of the computer 5.
[0034]
In the classification device 1, the computer 5 controls the stage driving unit 31 to relatively move the imaging position of the imaging unit 2 to a predetermined position on the substrate 9, and obtain a signal (of a signal) of the image acquired by the imaging unit 2. ) Is input to the arithmetic unit 4. In the arithmetic unit 4, first, the signal of the object image is output to the inspection image memory 411 or the reference image memory 412 via the changeover switch 40. The inspection image in the classification device 1 is, for example, an image of a pattern formed on a logic area of a die (that is, an area corresponding to one chip) arranged on the substrate 9.
[0035]
In the classifying apparatus 1, when an image to be inspected is acquired, the changeover switch 40 is connected to the image memory 411 to be inspected, and the stage driving unit 31 moves the stage 3 so that the imaging position of the imaging unit 2 is changed. Moves onto the logic area of the die on the substrate 9. Then, the data of the target image is stored in the target image memory 411, and the area of the target image in the target image is specified.
[0036]
When a reference image is obtained, the changeover switch 40 is connected to the reference image memory 412 side, and the imaging position of the imaging unit 2 is set to the same position of another die by the stage driving unit 31 (that is, the period of the cycle of the die pattern). (An imaging area separated by an integral multiple), and an image on a logic area of another die is acquired. Then, the data of the target object image acquired by the imaging unit 2 is stored in the reference image memory 412 in a state where the area of the reference image can be specified.
[0037]
The reference image may be acquired before the image to be inspected, or the reference image may be prepared in advance as an average image of a plurality of images already acquired (for example, a plurality of inspected images to be inspected). Good. Further, data of a golden template image generated based on CAD data (that is, an image in which no defect exists or is assumed to have no defect) may be stored in advance.
[0038]
The value of each pixel of the inspection image and the value of the corresponding pixel of the reference image are sequentially input from the inspection image memory 411 and the reference image memory 412 to the difference calculation circuit 42. The difference calculation circuit 42 calculates a signed difference and a difference absolute value obtained by subtracting the corresponding pixel value of the reference image from the pixel value of the image to be inspected, and outputs the signed difference to the difference statistic calculation circuit 43 and the difference absolute value to Each is output to the normalization circuit 44. Thereby, a signed difference image and a difference absolute value image are substantially generated.
[0039]
The difference statistic calculation circuit 43 accumulates signed differences of the number of pixels corresponding to one inspected image or reference image, and calculates a standard deviation of the accumulated signed differences. The standard deviation is output to a normalization circuit 44, and each of the absolute difference values input from the difference calculation circuit 42 is normalized by the standard deviation. That is, in the normalization circuit 44, each difference absolute value is divided by the standard deviation, and the result is multiplied by a predetermined coefficient to obtain a normalized difference absolute value. Note that a buffer for storing the absolute difference value may be provided in the normalization circuit 44, and the signed difference may be input from the difference statistic calculation circuit 43 to the normalization circuit 44 to obtain the absolute difference value.
[0040]
The normalized difference absolute values are sequentially output to the comparison circuit 45, and are compared with a predetermined threshold value stored in the threshold value memory 413 in advance. When the normalized absolute value of the difference is larger than a threshold value, a pixel value “1” indicating a defect is determined. When the normalized difference absolute value is smaller than a predetermined threshold value, a non-defective (ie, normal pixel) is determined. The indicated pixel value “0” is output to the defective area image memory 414. As a result, a binary image indicating a defect area of the wiring pattern in the inspection image (that is, a defect area of the inspection pattern) is stored in the defect area image memory 414 as a defect area image. As described above, the defect area image is generated and stored by comparing the inspected image with the reference image.
[0041]
FIG. 5A is a diagram illustrating the inspection image 611, and FIG. 5B is a diagram illustrating the reference image 62. The inspection image 611 and the reference image 62 are images in which the wiring pattern 6a is formed on the background area 6b. FIG. 6 is a diagram showing a defect area image 631 generated from the inspection image 611 in FIG. 5A and the reference image 62 in FIG. 5B. In FIG. 6, the area denoted by reference numeral 631a is specified as an area indicating a defect, and this area 631a corresponds to the defective area 611a of the inspection image 611.
[0042]
When the data of the defect area image 631 is obtained as described above, the computer 5 generates an evaluation value used for defect classification, and classifies the defect.
[0043]
As shown in FIG. 7, the computer 5 has a general computer system configuration in which a CPU 51 for performing various arithmetic processes, a ROM 52 for storing basic programs, and a RAM 53 for storing various information are connected to a bus line. The bus line further includes a fixed disk 54 for storing information, a display 55 for displaying various information such as images, a keyboard 56a and a mouse 56b for receiving input from an operator, and a computer such as an optical disk, a magnetic disk, and a magneto-optical disk. A reading device 57 that reads information from the readable recording medium 8 and a communication unit 58 that transmits and receives signals to and from other components of the classification device 1 may be appropriately connected via an interface (I / F). Connected.
[0044]
The computer 5 reads the program 80 from the recording medium 8 via the reading device 57 in advance and stores the program 80 on the fixed disk 54. Then, the program 80 is copied to the RAM 53 and the CPU 51 executes the arithmetic processing according to the program in the RAM 53 (that is, the computer executes the program), so that the computer 5 generates the evaluation value and deletes the defect. Perform classification processing.
[0045]
4, the divided area image memory 531 corresponds to a part of the RAM 53 of the computer 5, and the divided area calculation unit 501, the defect feature amount extraction unit 502, and the classifier 503 correspond to functions realized by the CPU 51 and the like. However, these functions may be realized by a dedicated electric circuit, or an electric circuit may be partially used.
[0046]
FIG. 8 is a diagram showing a flow of processing in which the computer 5 classifies a defect. Hereinafter, a description will be given with reference to the images illustrated in FIGS. 5A and 5B and FIG. First, the data of the defective area image 631 stored in the defective area image memory 414 is stored and prepared in the RAM 53 (step S21). Subsequently, the data of the reference image 62 stored in the reference image memory 412 is input to the divided region calculation unit 501, and the divided region calculation unit 501 indicates the region where the wiring pattern 6a (that is, the pattern to be referred to) exists. An image is generated (Steps S22 and S23).
[0047]
Specifically, first, a threshold value (hereinafter, referred to as “region determination threshold value”) for specifying the type of the region to which each pixel of the reference image 62 belongs (that is, the background region and the wiring region). Is obtained (step S22). Various methods may be used as a method for obtaining the region determination threshold. As shown in FIG. 9, when the histogram of the pixel values of the reference image 62 has a bimodal (For example, a pixel value that becomes a minimum between peaks) can be used as the region determination threshold.
[0048]
In addition, as shown in FIG. 10, when there is no bimodality, “Automatic threshold selection method based on discrimination and least square criterion” by Nobuyuki Otsu (Transactions of the Institute of Electronics, Information and Communication Engineers, Vol. J63-D, No. 4, April 1980, pp. 349-356). In this method, as a value for evaluating the appropriateness of the threshold value, a class separation degree based on the variance within a class (a group of pixel values separated by the threshold value) and the variance between the classes is adopted, and the class separation degree is maximized. The threshold is determined so that With this method, when an image is divided into two regions, a non-parametric optimal threshold value can be stably obtained even when the histogram of pixel values does not have bimodal characteristics.
[0049]
When the area determination threshold value is obtained as described above, the area type of each pixel value of the reference image 62 is specified according to the area determination threshold value, and the pixel value “0” indicating the background area or the wiring area is indicated. One of the pixel values “1” is output to the divided area image memory 531 (Step S23). As a result, data of a binary image (hereinafter, referred to as a “divided region image”) indicating the divided region indicating the region where the wiring pattern exists is stored and prepared in the divided region image memory 531 (step S24).
[0050]
FIG. 11 is a diagram illustrating a divided area image 64 generated from the reference image 62 of FIG. 5B. The area of the wiring pattern 6a of the reference image 62 has a pixel value “1” in the divided area image 64. The background area 6b of the reference image 62 is identified as the background area 64b having the pixel value “0” in the divided area image 64.
[0051]
Subsequently, the data of the divided area image 64 and the data of the defective area image 631 are input to the defect feature quantity extraction unit 502, and a logical operation for each pixel of the divided area image 64 and the defective area image 631 is performed, which will be described later. As described above, the geometric relative positional relationship between the wiring region and the background region (the relationship that divides or encompasses the other region) is obtained as an evaluation value (step S25). Details of the process relating to step S25 will be described later.
[0052]
When the evaluation value regarding the defect area 611a of the inspection image 611 is obtained, the evaluation value is input to the classifier 503, and the classification of the defect area 611a is performed in the classifier 503 (step S26). For example, the classifier 503 is an expert system constructed by an operator, and implements a defect classification based on an evaluation value by executing a program described as follows.
[0053]
if (background division probability> 0.5) then classification = short
else if (background inclusion probability> 0.5) then classification = isolated point
else then classification = other
The background division probability and the background inclusion probability in the above program are evaluation values obtained by a method described later in step S25.
[0054]
The classification result obtained by the classifier 503 is output as a signal R, and the signal R is input to the management device 72 shown in FIG. 1 and used for managing the yield of the substrate manufacturing system 7 as described above.
[0055]
Note that the classifier 503 may be a discriminator such as a discriminant function, a function tree, a decision tree, or a neural network. In this case, the computer 5 learns while creating teacher data, and responds to the discriminator. The generated defect determination condition may be automatically generated, and the generated defect determination condition may be input to the classifier 503. As a result, appropriate defect classification by an automatically constructed discriminant function, function tree, decision tree, neural network, or the like is realized.
[0056]
Next, a process of obtaining an evaluation value in the defect feature quantity extraction unit 502 will be described. FIG. 12 is a diagram showing a flow of a process for obtaining an evaluation value, which is a process performed in step S25 in FIG. In step S25, it is possible to obtain a plurality of types of evaluation values. The following description focuses on one of the evaluation values, the background inclusion probability. The background inclusion probability is a value indicating whether or not the defect region in the inspection image is included in the background region, or how much the defect region is included.
[0057]
First, a logical product of the value of each pixel of the divided area image and the value of the corresponding pixel of the defective area image is obtained by the defect feature quantity extraction unit 502 (step S31). Thereby, for example, in the defective area image 631 shown in FIG. 6, the area 631a indicating the defect whose pixel value is “1” overlaps with the wiring area 64a having the pixel value “1” in the divided area image 64 shown in FIG. An image in which only the region has the pixel value “1” (hereinafter, referred to as an “AND image”) is generated.
[0058]
FIG. 13 is a diagram illustrating an AND image 651 generated from the defect area image 631 and the divided area image 64. In the AND image 651, the values of all the pixels are “0”, indicating that the region 631a of the defective region image 631 and the wiring region 64a of the divided region image 64 do not overlap. If the pixel value “1” does not exist in the AND image, the defect area 611a is included in the background area 6b in the inspection image 611 shown in FIG. 6a). Therefore, the background inclusion probability is set to 1.0 (steps S32 and S35).
[0059]
On the other hand, if the area indicating a defect in the defective area image is specified to be slightly larger than the actual defective area in the image to be inspected depending on the detection condition, or if a corresponding pixel between the defective area image and the divided area image A minute shift may occur. For example, when the inspection image 612 shown in FIG. 14A is acquired, as shown in FIG. 14B, the defect area 632a indicating the defect is specified to be larger than the defect area 612a of the inspection image 612. The region image 632 may be generated. When the defect area image 632 is input to the defect feature extraction unit 502, an AND image 652 shown in FIG. 14C is generated from the defect area image 632 and the divided area image 64 (see FIG. 11) ( Step S31).
[0060]
When it is confirmed that an area 652a having a pixel value of “1” exists in the AND image 652 (step S32), the defect feature amount extraction unit 502 performs image processing on the defect area image 632 (step S34). As the image processing, a process of shrinking the area 632a indicating a defect is performed, and a new defective area image 633 shown in FIG. Then, an AND image 653 shown in FIG. 14E is generated from the new defective area image 633 and the divided area image 64 (Step S31).
[0061]
Since there is no area having a pixel value of “1” in the AND image 653 after the contraction processing has been performed once (step S32), processing for determining the background inclusion probability is performed (step S35). At this time, the background inclusion probability is determined according to the number of times of contraction processing based on the conversion curve shown in FIG. For example, since the area where the pixel value is “1” does not exist in the AND image 653 by one contraction process, the background inclusion probability is set to 0.8 from the conversion curve in FIG.
[0062]
Even when the contraction processing is repeated four times (that is, when the predetermined number of repetitions is reached), if there is an area with a pixel value of “1”, the fifth contraction processing is not performed (step S33), the background inclusion probability is set to 0.0 in step S35. That is, it is determined that the defect area is not clearly included in the background area in the inspection image. The conversion curve shown in FIG. 15 may be appropriately changed according to the degree of the process of shrinking the defective area in step S34.
[0063]
According to the above processing, when the area indicating the defect in the defect area image is clearly included in the background area in the divided area image, that is, when the defect area is clearly included in the background area in the inspected image, The inclusion probability is set to 1.0, the background inclusion probability is set to 0.0 when the defect area is not clearly included in the background area, and the background inclusion probability is set to 0.0 when the defect area is almost included in the background area. The inclusion probability is a value between 0.0 and 1.0.
[0064]
The fact that the area indicating the defect of the defective area image is included in the background area of the divided area image means that the area indicating the defect and the wiring area of the divided area image overlap when the defective area image and the reference image are overlapped. The background inclusion probability indicates whether the area indicating the defect overlaps the wiring area (that is, whether the defect area in the image to be inspected overlaps the wiring area), or how much substantially. It can be said that it is a value indicating an overlapping state of whether it can be regarded as overlapping.
[0065]
The processing for obtaining the background inclusion probability, which is one of the evaluation values, has been described above. However, in order to obtain the background inclusion probability at a higher speed, the area (or the pixel The evaluation value may be determined according to (number). FIG. 16 is a diagram illustrating an example of a conversion curve used for determining an evaluation value according to the number of pixels. With the conversion curve of FIG. 16, it is possible to quickly determine an evaluation value according to the number of pixels of the area where the defective area and the wiring area divided from the reference image overlap.
[0066]
Next, an example of obtaining the background division probability as another evaluation value will be described along the flow of the processing of obtaining the evaluation value in FIG. 12 (however, the processing in step S32 differs from the description in FIG. 12). Become.). The background division probability is a value indicating whether or not a defect area divides a background area in an image to be inspected, or how much it can be regarded as being substantially divided.
[0067]
First, assuming that the image to be inspected 614 shown in FIG. 17A is acquired by the imaging unit 2, the defect area image 634 shown in FIG. 17B is generated and prepared by the arithmetic unit 4. Then, the defect feature amount extraction unit 502 calculates the logical sum of the value of each pixel of the defective area image 634 and the value of the corresponding pixel of the divided area image 64 (see FIG. 11), and obtains a binary image (hereinafter, “ OR image 654 ") is generated as shown in FIG. 17C (step S31). Then, labeling is performed on an area having a pixel value of “1” in the OR image 654 and the divided area image 64, and the number of areas having a pixel value of “1” (hereinafter, referred to as “label number”) is obtained.
[0068]
The OR image 654 is an image showing an overlapping state of the area 634a indicating a defect of the defective area image 634 and the wiring area 64a of the divided area image 64, and the number of labels of the area 654a of the OR image 654 whose pixel value is “1”. Is smaller than the number of labels of the wiring area 64a of the divided area image 64, the state where the background area 64b is divided when the area 634a indicating the defect overlaps with the divided area image 64 (or the area 634a indicating the defect is It can be said that the defective area 614a also divides the background area 6b in the inspection image 614 (that is, the wiring pattern 6a is short-circuited).
[0069]
Therefore, in step S32, the number of labels of the OR image 654 and the number of labels of the divided area image 64 are compared (the processing is different from that described in FIG. 12), and as shown in FIG. 17C, the OR image 654 is obtained. Is smaller than the number of labels in the divided area image 64, the background division probability of the defective area 614a of the inspection image 614 is set to 1.0 (step S35).
[0070]
On the other hand, a region indicating a defect in the defect region image may be detected to be slightly smaller than the defect region in the image to be inspected, or a minute shift may occur between the defect region image and the divided region image. For example, when the inspection image 615 shown in FIG. 18A is obtained, and the defect area 635a in the defect area image 635 shown in FIG. 18B is detected to be smaller than the defect area 615a of the inspection image 615. The OR image 655 shown in FIG. 18C is generated from the defective area image 635 and the divided area image 64 (see FIG. 11) (step S31).
[0071]
When the number of labels in the OR image 655 in FIG. 18C is compared in step S32, the number of labels in the OR image 655 and the number of labels in the divided area image 64 become the same. In this case, image processing is performed on the defective area image 635 (step S34). As the image processing, a process of expanding a defect area 635a of the defect area image 635 is performed, and a new defect area image 636 having a defect area 636a is generated as shown in FIG.
[0072]
Then, an OR image 656 shown in FIG. 18E is generated from the new defective area image 636 and the divided area image 64 (Step S31). Since the number of labels of the OR image 656 after the expansion processing is performed once is smaller than the number of labels of the divided area image 64 (step S32), the background division probability is subsequently determined based on the conversion curve shown in FIG. Is done. For example, in the OR image 656 of FIG. 18E, the number of labels is smaller than the number of labels of the divided area image 64 by one expansion processing, and thus the background division probability of the defective area 615a of the inspected image 615 is 0.75. Is done.
[0073]
If the number of labels in the defective area image does not become smaller than the number of labels in the divided area image 64 even when the expansion processing is repeated three times, the fourth expansion processing is not performed (step S33), and the step S33 is performed. In S35, the background division probability is set to 0.0. That is, it is determined that the defect area does not clearly divide the background area in the inspection image. The conversion curve shown in FIG. 19 may be appropriately changed according to the degree of the processing for expanding the defective area.
[0074]
According to the above processing, when the area indicating a defect in the defect area image clearly divides the background area in the divided area image, that is, when the defect area clearly divides the background area in the inspected image, the background division probability Is set to 1.0. If the defective area does not clearly divide the background area, the background division probability is set to 0.0. If the defective area almost divides the background area, the background division probability is set according to the degree. Is a value between 0.0 and 1.0.
[0075]
In addition, dividing the background region of the divided region image by the region indicating the defect in the defective region image means that when the defective region image and the divided region image are overlapped, one region indicating the defect becomes the wiring region of the divided region image. The background division probability indicates whether the area indicating a defect overlaps the wiring area at a plurality of locations (that is, whether the defect area in the image to be inspected overlaps the wiring area at a plurality of locations). ) Or a value indicating an overlap state of how much the overlap can be considered.
[0076]
The example in which the background inclusion probability and the background division probability are obtained as the evaluation values has been described above. In the above-described divided region image, the pixel value of the background region is “1”, and the pixel value of the wiring region is “0”. In this way (in other words, by acquiring the background area as a divided area), the wiring inclusion probability indicating the degree to which the defective area is included in the wiring area (so-called a hole), and the defective area dividing the wiring area (That is, the wiring is disconnected (so-called open)), the wiring disconnection probability can be obtained as the evaluation value.
[0077]
As described above, in the classification device 1, a new defective area image obtained by expanding or contracting the defective area image or the area indicating the defect, that is, the logical operation for each pixel of the image based on the defective area image and the divided area image is performed. Then, an evaluation value indicating the overlapping state of the defect area and the divided area is obtained, and the defect is classified into the wiring pattern of the inspection image based on the evaluation value. As a result, it is possible to easily classify a specific type of defect in the wiring pattern, which cannot be easily classified by the conventional method. Further, by performing expansion or contraction processing on the defect area image, it is also possible to obtain the degree of the defect.
[0078]
Note that image processing such as expansion and contraction may be performed on the divided area image, and the evaluation value may be obtained by a logical operation of the processed divided area image and the defective area image. That is, the image used for the logical operation may be any image based on the divided region image (including the divided region image itself).
[0079]
FIG. 20 is a diagram illustrating another example of the configuration of the classification device 1. The classification device 1 shown in FIG. 20 includes a second defect feature amount extraction unit 502a, and the second defect feature amount extraction unit 502a is connected to the inspection image memory 411, the reference image memory 412, and the defect area image memory 414. . The first defect feature quantity extraction unit 502 in FIG. 20 is the same as the defect feature quantity extraction unit 502 in FIG. 4, and the other configuration is the same as that in FIG.
[0080]
FIG. 21 is a diagram showing a flow of the operation in which the classification device 1 classifies the defect, and shows only a portion corresponding to step S26 in FIG. When the evaluation value is obtained in step S25, the classifying apparatus 1 reads the data of the inspection image, the data of the reference image, and the data of the defect image from the inspection image memory 411, the reference image memory 412, and the defect area image memory 414. Are output to the second defect feature quantity extraction unit 502a.
[0081]
In the second defect feature extraction unit 502a, a feature which is a statistic of a pixel value such as an average value or a standard deviation of pixel values from an image to be inspected, a reference image, or a defect area image, or the area and height of a defect area , Width, image feature amount that is a geometric feature amount such as the angle of the center line of the defect area and the moment (step S41). The image feature amount of the signed difference image or the difference absolute value image may be obtained by the second defect feature amount extraction unit 502a. Further, the image feature amount may be obtained for the vicinity of the defect area in the inspection image, the absolute difference image, or the defect area image. That is, the image feature amount is obtained based on at least one of the inspected image, the reference image, and the defect area image.
[0082]
The evaluation value obtained by the first defect feature quantity extraction unit 502 is based on the state of overlap between the area indicating the defect in the defect area image and the divided area of the divided area image, that is, the defect area and the wiring pattern in the inspection image. Can be regarded as a kind of image feature amount based on the overlapping state of the first defect. However, since the image feature amount obtained by the second defect feature amount extracting unit 502a is not related to the above-mentioned overlapping state, the first defect It can be said that the image feature amount obtained by the feature amount extraction unit 502 is different. In the following description, “image feature amount” indicates an image feature amount obtained by the second defect feature amount extraction unit 502a.
[0083]
The evaluation value obtained by the first defect feature value extraction unit 502 and the image feature value obtained by the second defect feature value extraction unit 502a are input to the classifier 503, where the evaluation value and the image feature value are input. (Step S42).
[0084]
As described above, in the classification device 1 shown in FIG. 20, an image feature amount different from the evaluation value is obtained based on at least one of the inspection image, the reference image, and the defect area image, and the evaluation value and the image feature amount are obtained. Classification of defects is performed based on Therefore, a more appropriate classification result can be obtained as compared with the case where the classification is performed based on either the evaluation value or the image feature amount.
[0085]
FIG. 22 is a diagram illustrating an example in which the configuration related to classification is changed in the classification device 1 illustrated in FIG. 20. A first classifier 503 is connected to the first defect feature extraction unit 502, and a second defect feature extraction unit is connected. A second classifier 503a is connected to 502a, and the classification result of the first classifier 503 is input to the second classifier 503a. All the classifiers mentioned in the following description are constructed by an expert system, a discriminant function, a function tree, a decision tree, a neural network, and the like, like the classifier 503 shown in FIG.
[0086]
FIG. 23 is a diagram showing a flow of an operation in which the configuration shown in FIG. 22 performs defect classification, and corresponds to the processing performed in step S42 in FIG. When the image feature amount is obtained in step S41, the first classifier 503 performs defect classification based on the evaluation value (step S51). The classification performed in step S51 is a classification relating to the overlapping state between the defective area and the wiring pattern (hereinafter, referred to as “coarse classification”). Subsequently, in the second classifier 503a, the result of the coarse classification in the first classifier 503 and the image feature amount from the second defect feature amount extraction unit 502a are input, and the detailed classification of the defect is performed (step S52). ).
[0087]
As described above, the rough classification based on the evaluation value is performed by the configuration shown in FIG. 22, and the detailed classification is performed based on the result of the coarse classification and the image feature amount. , It is possible to appropriately and efficiently classify the defects. Further, the result of the rough classification can be used separately.
[0088]
FIG. 24 is a diagram showing a configuration in which a third classifier 503b is further added to the configuration shown in FIG. The third classifier 503b is connected to the second defect feature extracting unit 502a, and the first classifier 503 receives the classification result of the third classifier 503b.
[0089]
FIG. 25 is a diagram showing a flow of an operation in which the configuration shown in FIG. 24 performs defect classification, and shows a process between step S41 and step S51 in FIG. When the image feature amount is calculated in step S41, the image feature amount is input from the second defect feature amount extraction unit 502a to the third classifier 503b, and the third classifier 503b classifies the defect (non-defect) ( (It can be regarded as defect detection.) Is performed based on the image feature amount (step S61). Here, in the third classifier 503b, it is not always necessary to strictly specify a defect or a non-defect. For example, the third classifier 503b has a small defect area or a small difference absolute value in a difference absolute value image. Alternatively, an image whose feature is not clear from the image feature amount as to whether it is a defect or a non-defect may be specified as a non-defect.
[0090]
Subsequently, the specified result is input to the first classifier 503, and when the input result indicates a defect (step S62), the evaluation value from the first defect feature amount extraction unit 502 in the first classifier 503 is obtained. The defect is roughly classified based on (step S51), and further detailed classification is performed by the second classifier 503a based on the result of the coarse classification and the image feature amount (FIG. 23: step S52). When the result in the third classifier 503b indicates a non-defect, the defect classification processing ends (step S62).
[0091]
As described above, in the configuration shown in FIG. 24, a defect or a non-defect is specified based on the image feature amount, and when the defect is specified, the defect is classified. As a result, it is possible to determine whether or not a classification process is necessary before performing an advanced classification, so that the processing content of the downstream classifier is simplified and efficient classification is realized.
[0092]
FIG. 26 is a diagram showing a state in which three classifiers 503d, 503e, and 503f are provided in the classification device 1 shown in FIG. 4, and a majority decision unit 504 is added. FIG. 27 is a diagram showing a flow of an operation in which the configuration shown in FIG. 26 performs defect classification, and is a process corresponding to step S26 in FIG.
[0093]
When the evaluation value is obtained by the defect feature quantity extraction unit 502 in step S25, the evaluation value is input to the three classifiers 503d to 503f. Different classifiers are constructed in the classifiers 503d to 503f, respectively, and the classifiers 503d to 503f perform output according to the respective classification conditions (step S71). The classifiers 503d to 503f are, for example, neural networks each having different parameters, or different determiners such as a neural network, a decision tree, and a function tree. Then, the classification results of the respective classifiers 503d to 503f are input to the majority decision unit 504, and two or more of the three classification results are output as the signal R as the final classification result in the majority decision unit 504 (step S72). . When a plurality of classification results are obtained using a plurality of parameters, a plurality of parameters may be sequentially set in one classifier.
[0094]
As described above, by obtaining the final classification result based on multiple classification results obtained by performing multiple classifications, inappropriate classification parameters may be used or special classification values may be used. Even if one of the classifiers cannot perform appropriate classification, defect classification can be performed appropriately as long as the other two classifiers perform appropriate classification, and the defect classification accuracy can be improved. Is improved.
[0095]
Although the configuration and operation of the classification device 1 have been described above, the function of the calculation unit 4 of the classification device 1 may be realized by software by the computer 5. Hereinafter, the manner in which the computer 5 performs the same process of defect determination and defect area image generation as the arithmetic unit 4 will be described.
[0096]
FIG. 28 is a diagram illustrating a flow of a process in which the computer 5 performs the defect determination and the defect area image generation. The computer 5 first receives a signal from the imaging unit 2 while controlling the movement of the imaging unit 2 and the stage 3, and stores data of the target object image on the fixed disk 54 (may be stored in advance). The image to be inspected in the object image is specified by the CPU 51 and is prepared so as to be accessible (step S81). Further, reference image data is also prepared on the fixed disk 54 (step S82).
[0097]
For example, when the reference image is obtained directly from the imaging unit 2, the computer 5 appropriately moves the stage 3 by an integral multiple of the cycle of the die pattern to capture the inspection image and the reference image, or , A golden template image is prepared as a reference image. A part of the object image may be specified as the image to be inspected, and a region separated from the image to be inspected by an integral multiple of the period of the pattern may be specified as the reference image.
[0098]
When the image to be inspected and the reference image are prepared, a signed difference image between the image to be inspected and the reference image is obtained by the CPU 51 (step S83), and then normalized by the standard deviation of the pixel values of the signed difference image. The resulting difference absolute value image is generated (step S84). Then, one pixel of the normalized difference absolute value image is specified (step S85), and is compared with a predetermined threshold value to be binarized to a pixel value indicating a defect or a non-defect ( Step S86).
[0099]
By repeating steps S85 and S86 for each pixel of the normalized difference absolute value image, a defective area image is generated and stored in the fixed disk 54, and the processing of defect determination and defective area image generation is completed (step S85). S87). By the above processing, the processing of the arithmetic unit 4 is realized by the computer 5, and an appropriate defect area image is generated.
[0100]
Although the computer 5 has been described as generating a normalized difference absolute value image, similar to the calculation unit 4 shown in FIG. 4, each time one pixel is normalized, a defect or non-defect is obtained. May be determined. When the defect area image is generated by the computer 5, the processing steps can be flexibly changed. The operation of the arithmetic unit 4 is substantially the same as that of FIG. 28, and the functional configuration when the computer 5 performs the same operation as that of the arithmetic unit 4 is substantially the same as that of the arithmetic unit 4 shown in FIG.
[0101]
Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments, and various modifications are possible.
[0102]
The classification device 1 does not necessarily need to be incorporated in the substrate manufacturing system 7 in a so-called in-line state as shown in FIG. 2, and only acquires images in the manufacturing system, and is separately provided as a classification device having no imaging unit 2. You may.
[0103]
The defective area image may be obtained directly from the signed difference image (for example, directly using two threshold values). That is, the defect area image is generated based on the difference image between the inspection image and the reference image. Further, both the image to be inspected and the reference image may be binary images. In that case, the defect area image is obtained as a difference absolute value image (or a signed difference image) between the image to be inspected and the reference image.
[0104]
The divided area image does not necessarily need to be an image indicating two types of areas, and may be an image indicating three or more types of areas. In this case, the evaluation value is obtained based on the state of overlap between at least one type of region and the defective region. The classification area image may be provided from outside by an operator.
[0105]
In the above embodiment, the defect classification is performed on the semiconductor substrate 9. However, the classification device 1 is also used to classify the defect of the pattern formed on the color filter, the shadow mask, the printed wiring board, and the like. The method is particularly suitable for classifying defects in a fine pattern formed on a substrate.
[0106]
【The invention's effect】
According to the present invention, it is possible to easily classify a specific type of defect in a wiring pattern.
[0107]
Further, in claims 8 to 10, an evaluation value indicating the degree of defect can be obtained.
[0108]
Further, according to the fourteenth to seventeenth aspects, defects can be appropriately classified.
[Brief description of the drawings]
1A is a diagram illustrating an image to be inspected, FIG. 1B is a diagram illustrating a reference image, and FIG. 1C is a diagram illustrating an absolute difference image.
FIG. 2 is a diagram illustrating a substrate manufacturing system.
FIG. 3 is a diagram showing a flow of a substrate manufacturing process.
FIG. 4 is a diagram showing a schematic configuration of a classification device.
5A is a diagram illustrating an image to be inspected, and FIG. 5B is a diagram illustrating a reference image.
FIG. 6 is a diagram showing a defect area image.
FIG. 7 is a diagram illustrating a configuration of a computer.
FIG. 8 is a diagram showing a flow of processing for classifying defects.
FIG. 9 is a diagram illustrating an example of a histogram of a reference image.
FIG. 10 is a diagram showing another example of a histogram of a reference image.
FIG. 11 is a diagram showing a divided area image.
FIG. 12 is a diagram showing a flow of a process for obtaining an evaluation value.
FIG. 13 is a diagram showing an AND image.
14A is a diagram showing an image to be inspected, FIGS. 14B and 14D are diagrams showing defect area images, and FIGS. 14C and 14E are diagrams showing AND images.
FIG. 15 is a diagram illustrating an example of a conversion curve of a background inclusion probability.
FIG. 16 is a diagram illustrating another example of the conversion curve of the background inclusion probability.
17A is a diagram illustrating an image to be inspected, FIG. 17B is a diagram illustrating a defect region image, and FIG. 17C is a diagram illustrating an OR image.
18A is a diagram showing an image to be inspected, FIGS. 18B and 18D are diagrams showing defect area images, and FIGS. 18C and 18E are diagrams showing OR images.
FIG. 19 is a diagram illustrating an example of a conversion curve of a background division probability.
FIG. 20 is a diagram showing another example of the classification device.
FIG. 21 is a diagram showing a flow of processing for classifying defects.
FIG. 22 is a diagram showing a configuration relating to defect classification.
FIG. 23 is a diagram showing a flow of processing for classifying defects.
FIG. 24 is a diagram showing a configuration related to defect classification.
FIG. 25 is a diagram showing a flow of processing for classifying defects.
FIG. 26 is a diagram showing a configuration relating to defect classification.
FIG. 27 is a diagram showing a flow of processing for classifying defects.
FIG. 28 is a diagram showing a flow of processing of defect determination and defect area image generation.
[Explanation of symbols]
1 Classification device
2 Imaging unit
4 Arithmetic unit
5 Computer
6a Wiring pattern
6b Background area
9 Substrate
62 Reference Image
64 divided area images
64a wiring area
64b background area
72 Management device
74 storage unit
80 programs
414 Defective area image memory
502, 502a Defect feature quantity extraction unit
503, 503a-503f Classifier
531 divided area image memory
611,612,614,615 Inspection image
611a, 612a, 614a, 615a Defect area
631-636 defect area image
631a, 632a, 634a to 636a, 652a, 654a
S11 to S15, S21 to S26, S31 to S35, S41, S42, S51, S52, S61, S62, S71, S72, S81 to S87

Claims (17)

A classification device for classifying defects of a pattern to be inspected on an object,
A storage unit that stores data of a defect area image indicating a defect area of a pattern to be inspected on an object, and data of a divided area image indicating an existing area of a reference pattern as a divided area,
An evaluation value calculation unit that performs a logical operation for each pixel of the image based on the defective area image and the image based on the divided area image to obtain an evaluation value indicating an overlapping state between the defective area and the divided area;
A classifying unit that classifies defects for the pattern to be inspected based on the evaluation value;
A classification device comprising: The classification device according to claim 1,
A classification device further comprising a defect area image generation unit that generates a defect area image by comparing an inspection image indicating an inspection pattern with a reference image indicating a reference pattern. The classification device according to claim 2, wherein
A classification device, further comprising: an imaging unit that captures an object to acquire data of an image to be inspected. A yield management system that manages a manufacturing yield of an object on which a pattern is formed,
A classification device according to claim 3,
A storage unit that stores a classification result by the classification unit in association with a management item,
A yield management system comprising: A classification method for classifying defects of a pattern to be inspected on an object,
A step of preparing data of a defect area image indicating a defect area of the pattern to be inspected on the target,
A step of preparing data of a divided area image indicating an existing area of the reference pattern as a divided area;
Calculating an evaluation value indicating an overlapping state between the defective area and the divided area by performing a logical operation for each pixel of the image based on the defective area image and the image based on the divided area image,
A step of classifying defects on the pattern to be inspected based on the evaluation value. A substrate manufacturing method for manufacturing a substrate on which a pattern is formed,
Removing the substrate from the loading position;
Forming a pattern on the substrate,
A step of acquiring data of an image to be inspected showing the pattern to be inspected by imaging the substrate,
Generating a defect area image indicating a defect area of the inspected pattern by comparing the inspected image and a reference image indicating a reference pattern,
An evaluation indicating the overlapping state of the defective area and the divided area by performing a logical operation for each pixel between the image based on the defective area image and the image based on the divided area image indicating the existence area of the reference pattern as the divided area Determining the value;
Performing a classification of a defect for the pattern to be inspected based on the evaluation value;
Transporting the substrate to an unloading position,
A method of manufacturing a substrate, comprising: A program used for classification of pattern defects, wherein the execution of the program by a computer, the computer,
A step of preparing data of a defect area image indicating a defect area of the pattern to be inspected;
A step of preparing data of a divided area image indicating an existing area of the reference pattern as a divided area;
Calculating an evaluation value indicating an overlapping state between the defective area and the divided area by performing a logical operation for each pixel of the image based on the defective area image and the image based on the divided area image,
A step of classifying a defect with respect to the pattern to be inspected based on the evaluation value. The program according to claim 7, wherein
An image based on the defective area image is an image obtained by expanding or contracting the defective area. The program according to claim 8, wherein
The program is characterized in that the evaluation value is obtained according to a degree of expansion or contraction of the defect area. The program according to claim 7 or 8,
A program according to claim 1, wherein said evaluation value is obtained according to an area of a region where said defective region and said divided region overlap. The program according to any one of claims 7 to 10, wherein execution of the program by a computer includes:
A step of preparing data of an image to be inspected showing the pattern to be inspected,
A step of preparing reference image data indicating the reference pattern,
Generating the defect area image by comparing the inspected image and the reference image,
A program characterized by further executing The program according to claim 11, wherein
A program, wherein the defect area image is generated from a difference image between the image to be inspected and the reference image. The program according to claim 11, wherein execution of the program by the computer causes the computer to execute
A program that further executes a step of binarizing the reference image to generate the divided region image. 14. The program according to claim 11, wherein the execution of the program by the computer is performed by the computer based on at least one of the inspection image, the reference image, and the defect area image. Further executing a step of obtaining an image feature amount,
A program for performing defect classification based on the evaluation value and the image feature amount. The program according to claim 14, wherein
The step of classifying the defect,
Performing a rough classification based on the evaluation value;
Performing a detailed classification based on the result of the coarse classification and the image feature amount;
A program characterized by having: The program according to claim 14, wherein
The step of classifying the defect,
Identifying a defect or non-defect based on the image feature amount,
A step of classifying the defect based on the evaluation value when the defect is specified. The program according to claim 11 or 12,
The step of classifying the defect,
A step of performing a plurality of classifications to obtain a plurality of classification results;
Obtaining a final classification result based on the plurality of classification results,
A program characterized by having:

Images