JP2008258486A - Distribution analysis method and system, abnormality facility estimation method and system, program for causing computer to execute its distribution analysis method or its abnormality facility estimation method, and recording medium readable by computer having its program recorded therein - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a distribution analysis method and system for evaluating whether or not analysis target data having such a characteristic as distributed along a plane are similar to a specific distribution shape pattern, which can always provide a good accurate evaluation even when the analysis target data are varied by various factors. <P>SOLUTION: A distribution shape pattern is divided into a plurality of rectangular sections in the form of a grid for a plane of defined analysis target data, and a density value capable of taking a multiple value in a certain numeral range is defined for each of the rectangular sections so that the density value indicates a relative magnitude relationship to characteristic values of the analysis target data (step S101). Typical data similar to the distribution shape pattern and reflecting a variation caused between a plurality of the analysis target data are created (step S102). A similarity indicative of a degree between the analysis target data and the typical data is found (step S103). Evaluation is carried out according to the found similarity whether or not the analysis target data are similar to the distribution shape pattern. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a distribution analysis method and apparatus, and more particularly to a distribution analysis method and apparatus for evaluating whether or not analysis target data having characteristics distributed along a plane is similar to a specific distribution shape.

  The present invention also relates to an abnormal facility estimation method and apparatus for automatically estimating a manufacturing apparatus in which an abnormality has occurred from among a plurality of manufacturing apparatuses respectively used in a plurality of manufacturing processes sequentially performed on a substrate.

  The present invention also relates to a program for causing a computer to execute such a distribution analysis method or abnormal facility estimation method.

  The present invention also relates to a computer-readable recording medium recording such a program.

  In a production line for thin film devices such as semiconductor wafers, semiconductor displays, and hard disk magnetic heads, various inspections are performed for the purpose of improving yield and stabilizing. These inspections include, for example, a pattern inspection for detecting a defect in a circuit pattern caused by a foreign substance attached on a substrate, an electrical characteristic inspection for inspecting an electrical characteristic of a formed circuit, and the like. The production line monitors these inspection results on a daily basis.For example, abnormalities have occurred in the manufacturing process based on the increase in the number of defects detected by pattern inspection and fluctuations in electrical characteristics measured by electrical characteristic inspection. Check whether there is any. If an abnormality occurs in a certain manufacturing process among a plurality of manufacturing processes that are sequentially performed on the board, the inspection results should be investigated and analyzed to quickly identify the cause and take countermeasures. Thus, damage due to yield reduction can be minimized. Therefore, a system for collecting inspection information and processing history of each manufacturing apparatus is provided in the manufacturing line.

  When an abnormality occurs in the manufacturing process, the inspection result often appears as a characteristic distribution on the substrate surface. Since these distributions vary depending on the state of the abnormality and are closely related to the cause of the abnormality, examples of past characteristic distributions are compiled into a database together with the cause of the abnormality, and the cause of the abnormality is investigated and decision-making measures are taken. To be used. Therefore, it is important to detect the tendency of the characteristic distribution in the inspection process at an early stage, but in reality it is difficult to express the characteristic distribution numerically. It is difficult to detect and monitor the characteristic distribution.

  In order to overcome these difficulties, a method has been proposed in which a geometric characteristic of a pre-registered distribution is compared with a characteristic distribution of the substrate, and a substrate having a characteristic distribution similar to the registered distribution is detected. As such a technique, for example, in Patent Document 1 (Japanese Patent Laid-Open No. 11-45919), a characteristic distribution is treated as a binary image, a binary image having a distribution shape registered in advance and a binary image having a characteristic distribution of a substrate. Describes a template matching method used in image processing. Specifically, in the method of this document, the number of defects detected on the substrate in the inspection process is totaled for each grid-like pixel set with respect to the substrate surface. Next, threshold processing is performed with a predetermined threshold on the number of defects for each pixel, and the number of defects for each pixel is binarized. From this result, a binary image of the entire area of the substrate surface is created, and template matching in image recognition is performed on the template of the defect distribution illustrated in FIG. When the matching rate with is high, it is determined that it matches the defect distribution defined in the template.

As another method, for example, in Patent Document 2 (Japanese Patent Laid-Open No. 2003-100825), a region where defects are concentrated as shown in FIG. 22A is defined as a template. As shown in FIG. 22B, a method is described in which a substrate having a higher number of defects in the template region with respect to the total number of defects on the substrate is determined to have a higher degree of matching. In another method described in the same document, a template is defined that includes density values for lattice-shaped small areas set on the substrate surface, and the density value of the template and the analysis target substrate are defined for each small area. Describes a method of using a correlation coefficient with the number of defects counted as a degree of coincidence.
Japanese Patent Laid-Open No. 11-45919 JP 2003-100825 A

  However, the methods described in Patent Documents 1 and 2 have a problem that the accuracy is not good for the following two reasons.

  The first reason is that the methods described in these documents always compare with a distribution shape pattern defined by the same template, and therefore cannot cope with fluctuations in the distribution of defects that actually occur. Can be mentioned.

  The distribution of defects is closely related to the cause of abnormality, but the distribution of defects caused by the same abnormality cause does not always have the same shape. For example, as shown in FIGS. 4A to 4C, the defect distribution shape (refers to the position, shape, and area where the defect concentrates) varies depending on the timing, degree, and state of the cause of the abnormality. There are things to do. 4A shows an example in which the defect distribution shape is a circle having a relatively small area, FIG. 4B shows an example in which the defect distribution shape is an ellipse, and FIG. 4C shows the defect distribution shape in FIG. Each example is a circle having a larger area than that of (a). If the comparison is always made with the same template as in the methods described in Patent Documents 1 and 2, such a variation in distribution cannot be dealt with.

  The second reason is that, in the methods described in these documents, only the distribution shape of the defects is evaluated regardless of the number of defects, so that the number of defects is not reflected in the evaluation. It is done.

  Not only the defect distribution shape but also the number of defects fluctuates depending on the timing, degree, and state of the cause of the abnormality. For example, FIG. 5A shows the relationship between the number of defects caused by a certain cause of abnormality and the number of substrates affected by the cause of the abnormality (distribution of the number of applicable boards), and FIG. This shows a case where the distribution of the number of corresponding substrates that are affected by the shift. In the situation where the distribution of the number of substrates corresponding to the cause of abnormality is the distribution shown in FIG. 5A, the substrate 201A having the typical number of defects indicated by the point 501 is emphasized in order to identify the cause of the abnormality. It is desirable to do. On the other hand, even in the case of the same cause of abnormality, the substrate 201B having the number of defects indicated by the point 501 is shown in FIG. Compared to the case of 5 (a), it is desirable not to place importance on identifying the cause of the abnormality.

  However, in the method of Patent Document 1, since the number of defects for each pixel is binarized, all the cases where the number of defects in a pixel is equal to or greater than a set threshold value are treated as the same value. For this reason, the magnitude of the number of defects is not reflected in the evaluation. In the first method and the second method described in Patent Document 2, since the similarity is evaluated based on the ratio of the number of defects inside and outside the region defined by the template, the number of defects itself is not an evaluation target. The number of defects is not reflected in the evaluation.

  For these reasons, the methods described in Patent Documents 1 and 2 have a problem that the accuracy of evaluation is not good.

  Accordingly, an object of the present invention is a distribution analysis method and apparatus for evaluating whether or not analysis target data having characteristics distributed along a plane is similar to a specific distribution shape pattern. An object of the present invention is to provide a distribution analysis method and apparatus that can always evaluate with high accuracy even if data fluctuates.

  Another object of the present invention is to use a degree of similarity based on such a distribution analysis method on the inspection result of a substrate, thereby making it possible to use a plurality of manufacturing apparatuses sequentially used for a substrate. An object of the present invention is to provide an abnormal facility estimation method and apparatus for automatically estimating a manufacturing apparatus in which an abnormality has occurred.

  Another object of the present invention is to provide a program for causing a computer to execute such a distribution analysis method or abnormal facility estimation method.

  Another object of the present invention is to provide a computer-readable recording medium in which such a program is recorded.

In order to solve the above problems, a distribution analysis method of the present invention is a distribution analysis method for evaluating whether or not analysis target data having characteristics distributed along a plane is similar to a specific distribution shape pattern,
The distribution shape pattern is divided into a plurality of rectangular areas in a lattice shape on a plane corresponding to the plane defining the analysis target data, and the relative magnitude of the characteristic value of the analysis target data is determined for each rectangular area. In order to express the relationship, define a concentration value that can take multiple values within a certain numerical range,
Create representative data that is similar to the distribution shape pattern and reflects the variation that occurs between the multiple data to be analyzed,
The analysis target data is compared with the representative data to obtain a similarity indicating the degree to which the analysis target data is similar to the representative data, and the analysis target data corresponds to the distribution shape pattern according to the similarity. It is characterized by evaluating whether it is similar to.

  Here, the “analysis target data” may be any data as long as some characteristic is distributed along the plane.

  Further, “multi-value” refers to two or more values, preferably three or more values.

  If processing corresponding to the distribution analysis method of the present invention is executed by, for example, a computer, it is possible to quantitatively evaluate whether or not the data to be analyzed is similar to the distribution shape pattern. Moreover, the representative data reflects changes in the analysis target data. Therefore, in this distribution analysis method, even if the analysis target data fluctuates due to various factors, such as a change in the distribution shape of the characteristic or a change in the size of the characteristic value, the analysis target data is changed to the distribution shape. It can always be evaluated accurately whether it is similar to the pattern.

  Whether or not the representative data is similar to the distribution shape pattern is determined based on whether or not the distribution shape of the characteristic value of the representative data is similar to the distribution shape of the density value of the distribution shape pattern. It is desirable to do.

In the distribution analysis method of one embodiment, the creation of the representative data is as follows:
From the first data group formed by the plurality of analysis target data including the variation, the relative magnitude relation of the distribution of characteristic values seems to be similar to the relative magnitude relation of the distribution of density values of the distribution shape pattern. A first step of extracting a second data group,
And a second step of calculating the representative data from the distribution of characteristic values of the second data group obtained in the first step.

  According to the distribution analysis method of this embodiment, the representative data can be easily created.

  In the distribution analysis method of one embodiment, the first step includes a correlation coefficient between a characteristic value corresponding to each rectangular area of the first data group and a density value assigned to each rectangular area of the distribution shape pattern. Is extracted as data of the second data group such that is equal to or greater than a predetermined threshold value.

  In the distribution analysis method according to this embodiment, a useful second data group can be extracted regardless of the size (absolute value) of the characteristic value by using the correlation coefficient. The extracted second data group is similar to the above distribution shape pattern from the viewpoint of the relative magnitude relationship of the characteristic values regardless of the size (absolute value) of the characteristic values.

  In the distribution analysis method according to an embodiment, the second step averages characteristic values for each rectangular area for the second data group obtained in the first step, and calculates an average value for each rectangular area. The representative data is configured as a component.

  According to the distribution analysis method of this embodiment, the representative data is composed of the average value for each rectangular area as a component, so that the representative data is the data of the second data group, and thus the analysis target data. It reflects the fluctuations of

  In the distribution analysis method of an embodiment, when the characteristic value is averaged for each rectangular area for the second data group, the characteristic value corresponding to each rectangular area and each rectangular area of the distribution shape pattern are given. A weighted average weighted by a correlation coefficient with the density value is calculated.

  According to the distribution analysis method of this embodiment, the representative data is more similar to the distribution shape pattern as compared with the case of calculating a simple average. As a result, the evaluation of the similarity degree places more importance on the distribution shape of the distribution shape pattern.

In the distribution analysis method of one embodiment, the process for obtaining the similarity is as follows.
A third step of calculating a first difference between a distribution of characteristic values of predetermined reference data and a distribution of characteristic values of the analysis target data;
A fourth step of calculating a second difference between the distribution of characteristic values of the representative data and the distribution of characteristic values of the analysis target data;
And a fifth step of calculating the similarity according to both the first difference and the second difference.

  Here, it is sufficient that the “reference data” is fixed data that serves as a reference for calculating the similarity. For example, when the “characteristic value” is “number of defects” on the substrate, data indicating that “the number of defects is zero” can be set as “reference data” over the entire area of the substrate.

  In the distribution analysis method of this embodiment, according to both the first difference and the second difference, that is, the distribution of the characteristic value of the reference data and the distribution of the characteristic value of the analysis target data. On the other hand, the similarity is calculated according to the relative position of the characteristic value distribution of the analysis target data. In this case, even if the analysis target data fluctuates due to various factors, the analysis target data can be evaluated on a certain scale.

  In the distribution analysis method of one embodiment, the similarity calculated in the fifth step is obtained by dividing the reciprocal of the second difference by the sum of the reciprocal of the first difference and the reciprocal of the second difference. It is characterized by being obtained.

In the distribution analysis method of one embodiment,
The first difference calculated in the third step is a distance between the characteristic value distribution of the reference data and the characteristic value distribution of the analysis target data,
The second difference calculated in the fourth step is a distance between the characteristic value distribution of the representative data and the characteristic value distribution of the analysis target data.

  In the distribution analysis method of one embodiment, when obtaining the similarity, it is determined whether or not the distribution shape of the characteristic value of the analysis target data is similar to the distribution shape of the characteristic value of the representative data. The similarity is calculated according to the determination result.

  In the distribution analysis method according to the embodiment, in order to obtain the similarity, a distribution shape of characteristic values (represented by a relative magnitude relationship of characteristic values) is taken into consideration. As a result, the evaluation of the similarity degree places more importance on the distribution shape of the distribution shape pattern.

  In the distribution analysis method of one embodiment, when performing the determination, a correlation coefficient between a characteristic value corresponding to each rectangular area of the analysis target data and a characteristic value corresponding to each rectangular area of the representative data. And determining whether or not the correlation coefficient is equal to or greater than a predetermined threshold value.

  According to this embodiment, the determination can be easily performed.

  In the distribution analysis method of one embodiment, the characteristics of the analysis target data are distributed along the surface of the substrate.

  Here, the “substrate” widely refers to a glass substrate on which a thin film device is manufactured, a wafer on which a semiconductor device is manufactured, and the like.

  In the distribution analysis method of this embodiment, even if the analysis target data fluctuates due to various factors generated in the manufacturing process for processing the substrate, it is always determined whether or not the analysis target data is similar to the distribution shape pattern. Can be evaluated with high accuracy.

The abnormal equipment estimation method of the present invention is an abnormal equipment estimation method for estimating a manufacturing apparatus in which an abnormality has occurred, from among the manufacturing apparatuses used in each of a plurality of manufacturing steps sequentially performed on a substrate,
The distribution analysis method according to claim 11 is performed to obtain the similarity,
Based on a processing history representing a manufacturing apparatus that has processed each of the substrates in each of the manufacturing steps, a manufacturing apparatus commonly used for the substrates having the similarity equal to or higher than a predetermined threshold is extracted. And

  According to the abnormal equipment estimation method of the present invention, the manufacturing apparatus used in common for the substrates whose similarity is equal to or higher than a predetermined threshold is extracted, so that the manufacturing apparatus in which the abnormality has occurred is accurately estimated. be able to.

A distribution analysis apparatus according to the present invention is a distribution analysis apparatus that evaluates whether or not analysis target data having characteristics distributed along a plane is similar to a specific distribution shape pattern,
The distribution shape pattern is divided into a plurality of rectangular areas in a lattice shape on a plane corresponding to the plane defining the analysis target data, and the relative magnitude of the characteristic value of the analysis target data is determined for each rectangular area. A pattern registration unit that defines and assigns density values that can take multiple values within a certain numerical range to represent the relationship;
A representative data creating unit that creates representative data that is similar to the distribution shape pattern and reflects a variation that occurs between a plurality of the analysis target data;
The analysis target data is compared with the representative data to obtain a similarity indicating the degree to which the analysis target data is similar to the representative data, and the analysis target data corresponds to the distribution shape pattern according to the similarity. And a similarity evaluation unit that evaluates whether or not they are similar to each other.

  Here, the “analysis target data” may be any data as long as some characteristic is distributed along the plane.

  Further, “multi-value” refers to two or more values, preferably three or more values.

  According to the distribution analysis apparatus of the present invention, the distribution analysis method of the present invention can be implemented. That is, if the pattern registration unit, the representative data creation unit, and the similarity evaluation unit are configured by a computer program, for example, and the process is executed, the distribution shape of the characteristic changes or the size of the characteristic value changes. Even if the analysis target data fluctuates due to various factors such as, it can always be accurately evaluated whether the analysis target data is similar to the distribution shape pattern.

In the distribution analyzer of one embodiment, the representative data creation unit is
From the first data group formed by the plurality of analysis target data including the variation, the relative magnitude relation of the distribution of characteristic values seems to be similar to the relative magnitude relation of the distribution of density values of the distribution shape pattern. A first part for extracting a second data group;
And a second portion for calculating the representative data from the distribution of characteristic values of the second data group obtained in the first step.

  According to the distribution analyzer of this embodiment, the representative data can be easily created.

In the distribution analysis apparatus according to one embodiment, the similarity evaluation unit includes:
A third portion for calculating a first difference between the distribution of characteristic values of the predetermined reference data and the distribution of characteristic values of the analysis target data;
A fourth portion for calculating a difference between a second distribution of characteristic values of the representative data and a distribution of characteristic values of the analysis target data;
And a fifth portion for calculating the similarity according to both the first difference and the second difference.

  Here, it is sufficient that the “reference data” is fixed data that serves as a reference for calculating the similarity. For example, when the “characteristic value” is “number of defects” on the substrate, data indicating that “the number of defects is zero” can be set as “reference data” over the entire area of the substrate.

  In the distribution analysis apparatus according to this embodiment, according to both the first difference and the second difference, that is, the distribution of characteristic values of the reference data and the distribution of characteristic values of the analysis target data. On the other hand, the similarity is calculated according to the relative position of the characteristic value distribution of the analysis target data. In this case, even if the analysis target data fluctuates due to various factors, the analysis target data can be evaluated on a certain scale.

  In the distribution analysis apparatus according to one embodiment, the characteristics of the analysis target data are distributed along the surface of the substrate.

  Here, the “substrate” widely refers to a glass substrate on which a thin film device is manufactured, a wafer on which a semiconductor device is manufactured, and the like.

  In the distribution analysis apparatus of this embodiment, even if the analysis target data fluctuates due to various factors generated in the manufacturing process for processing the substrate, it is always determined whether the analysis target data is similar to the distribution shape pattern. Can be evaluated with high accuracy.

The abnormal equipment estimation apparatus of the present invention is an abnormal equipment estimation apparatus that estimates a manufacturing apparatus in which an abnormality has occurred from among the manufacturing apparatuses that are respectively used in a plurality of manufacturing steps that are sequentially performed on a substrate.
The distribution analysis device according to claim 13 is provided, and the similarity is obtained by the distribution analysis device,
Based on a processing history representing a manufacturing apparatus that has processed each of the substrates in each of the manufacturing steps, an abnormal facility estimation is performed to extract a manufacturing apparatus that is commonly used for the substrates whose similarity is equal to or greater than a predetermined threshold. It has the part.

  According to the abnormal equipment estimation apparatus of the present invention, since the manufacturing apparatus used in common for the substrates whose similarity is equal to or higher than a predetermined threshold is extracted, the manufacturing apparatus in which the abnormality has occurred is accurately estimated. be able to.

  The program of the present invention is a distribution analysis program for causing a computer to execute the distribution analysis method of the present invention.

  In another aspect, the program of the present invention is an abnormal equipment estimation program for causing a computer to execute the abnormal equipment estimation method of the above invention.

  The recording medium of the present invention is a computer-readable recording medium in which the distribution analysis program of the above invention is recorded.

  In another aspect, the recording medium of the present invention is a computer-readable recording medium in which the abnormal facility estimation program of the present invention is recorded.

  Hereinafter, the present invention will be described in detail with reference to the illustrated embodiments.

  FIG. 1 schematically shows the concept of a distribution analysis method according to an embodiment of the present invention, taking as an example the result of pattern inspection that detects, as a characteristic value, a circuit pattern defect caused by a foreign substance or the like in a substrate manufacturing process. . The pattern inspection is an inspection process for detecting the coordinates of the defect 202 of the circuit pattern formed on the substrate 201 as shown in FIG. When an abnormality occurs in a certain manufacturing apparatus, the defects 202 are often concentrated and distributed in a region unique to the abnormality within the surface of the substrate 201. This distribution analysis method is applied when detecting a substrate in which defects are concentrated in a specific region from a group of substrates that have undergone pattern inspection in the manufacturing process (as will be described later, based on the result of this distribution analysis, the distribution of the defect is distributed. The manufacturing equipment (abnormal equipment) that caused the problem can be identified early.)

  In this distribution analysis method, in step S101, the distribution shape of the characteristic value for comparison with the analysis target substrate is registered as a distribution shape pattern. An example of the distribution shape pattern is shown in FIG. As shown in FIG. 3A, the distribution shape pattern is obtained by dividing the surface of the substrate into a grid and setting a plurality of rectangular areas 301, 301,... And setting a density value for each rectangular area 301. Defined. The density value can take multiple values (in this example, three or more values) within a range of 0 to 1. A set of density values set for each rectangular area 301 represents a relative magnitude relationship of characteristic values on the substrate. In the pattern inspection, since the number of defects in each rectangular area corresponds to the “characteristic value”, the density value set for each rectangular area is obtained by converting the defect density into a relative value in the range of 0 to 1. Equivalent to. FIG. 3B and FIG. 3C show examples of distribution shape patterns. The color of the area in the figure represents the density value, and the darker the color, the higher the density value. Accordingly, the distribution shape pattern 302 in FIG. 3B represents a pattern in which defects are concentrated in the lower right area of the substrate, and the distribution shape pattern 303 in FIG. 3C has defects in the upper left area of the substrate. Represents a concentrated pattern. In general, in the manufacturing process, if the distribution shape of defects is different, the number of defects (raw value) generated in each region is different. Since the above density value represents only the relative magnitude relationship between the characteristic values on the substrate surface, defects are concentrated in any region on the substrate surface regardless of the number of defects (raw value). It can be easily set only from a relative point of view. The distribution shape pattern may be manually input by an analyst using knowledge about the manufacturing process, or may be automatically created using past cases.

  Next, in step S102 in FIG. 1, representative data that is similar to the distribution shape pattern and reflects a variation that occurs between a plurality of pieces of analysis target data from a distribution shape pattern and a representative data creation data group that will be described later. Create

  Here, the necessity of such representative data will be described with reference to FIGS. 4 (a) to 4 (c) mentioned above show examples of distribution of defects generated by a certain abnormal facility. Even if the same abnormal equipment is the cause, the defect distribution shape (that is, the position, shape, and area where the defects are concentrated) may differ depending on the timing, degree, and state of the cause of the abnormality. For example, in the defect distribution 401 in FIG. 4A, the distribution shape of the defects substantially matches that of the distribution shape pattern 303 in FIG. On the other hand, in the defect distribution 402 of FIG. 4B, the shape of the region where the defects are concentrated is an ellipse, and does not coincide with the distribution shape pattern 303 (circular shape) of FIG. In addition, the defect distribution 403 in FIG. 4C is wider than the distribution shape pattern 303 in FIG. In a situation where the defect distribution 401 of FIG. 4A is actually generated, if the substrate on which the defect distribution 401 of FIG. 4A coincides with the distribution shape pattern 303 of FIG. 3C is detected. Abnormal equipment can be estimated, and therefore the defect distribution 402 in FIG. 4B should not be emphasized. On the other hand, in the situation where more defect distributions in FIG. 4B are generated, it is more important to focus on the defect distribution 402 in FIG. 4B than in the above case in order to estimate abnormal facilities. It is considered appropriate.

  Further, not only the defect distribution shape but also the number of defects varies depending on the occurrence timing, degree, and state of the cause of the abnormality. As described above, FIG. 5A shows the relationship between the number of defects caused by a certain abnormality cause and the number of substrates affected by the abnormality cause (distribution of the number of corresponding substrates), and FIG. Shows a case where the distribution of the number of corresponding boards affected by the same cause of the abnormality is shifted. In the situation where the distribution of the number of substrates corresponding to the cause of abnormality is the distribution shown in FIG. 5A, the substrate 201A having the typical number of defects indicated by the point 501 is emphasized in order to identify the cause of the abnormality. It is desirable to do. On the other hand, even in the case of the same cause of abnormality, the substrate 201B having the number of defects indicated by the point 501 is shown in FIG. Compared to the case of 5 (a), it is desirable not to place importance on identifying the cause of the abnormality.

  In this way, in reality, the distribution shape of the characteristics changes depending on differences in the timing, degree, and state of the cause of the abnormality, and the size of the characteristic value changes. The characteristic data (this is appropriately referred to as “analysis target data”) fluctuates. Therefore, even if the analysis target data fluctuates due to various factors, the above-described representative data is required so that the substrate on which the defect has occurred can be appropriately detected. The representative data reflects the variation that occurs between the plurality of analysis target data, and the distribution shape of the characteristic value is similar to the distribution shape of the density value of the distribution shape pattern. The representative data is a target for comparison with the analysis target data.

  In step S102 in FIG. 1 described above, such representative data is converted into the first data group consisting of the distribution shape pattern data and the characteristic data of the substrate group manufactured in the same manufacturing process at the same time as the analysis target substrate. (Hereinafter referred to as “representative data creation data group”). The representative data creation data group may include characteristic data of the analysis target substrate. A specific method for creating representative data will be described in a later embodiment.

  Finally, in step S103 in FIG. 1, the characteristic data (analysis target data) of the analysis target board is compared with the representative data created in step S102, and the analysis target data is similar to the representative data. Find the degree of similarity that expresses the degree quantitatively (numerically). It is quantitatively evaluated whether or not the analysis object data is similar to the distribution shape pattern according to the similarity.

  In this case, since the representative data obtained from the representative data creation data group reflects changes in the analysis target data, even if the analysis target data changes due to various factors, the analysis is performed. Whether the target data is similar to the distribution shape pattern can always be evaluated with high accuracy.

  Note that the distribution analysis method according to the present invention is not applied only to the pattern inspection of a substrate, but can be applied to any data in which characteristics are distributed in a plane.

  FIG. 6 shows the configuration of a distribution analysis apparatus 600 according to an embodiment of the present invention suitable for implementing the above-described distribution analysis method (see FIG. 1). The distribution analysis apparatus 600 includes a pattern registration unit 602, a data collection unit 603 as an input unit, a representative data creation unit 606, a similarity evaluation unit 607, and a data output unit 608. An input device 601, an inspection information collection system 604, and an inspection device 605 are connected to the data collection unit 603 of the distribution analysis device 600, and an output device 609 is connected to the data output unit 608.

  The input device 601 is composed of a keyboard and a mouse, for example. In this example, the input device 601 provides the distribution analysis device 600 with the definition of the distribution shape pattern, the identification ID of the analysis target substrate, and the conditions forming the representative data creation data group (that is, the same condition as the analysis target substrate). This is used for inputting the designation of conditions) to be used as the representative data creation data group among the characteristic data of the substrate group manufactured in the manufacturing process.

  The pattern registration unit 602 receives the distribution shape pattern defined by the input device 601 and performs the process of step S101 for recording it in the database. The information received from the input device 601 is not only the distribution shape pattern, but also information related to the distribution shape pattern, for example, information about the device that causes the distribution shape pattern, is recorded in association with the distribution shape pattern. Also good.

  The data collection unit 603 receives the analysis information substrate identification data transmitted from the input device 601 to the distribution analysis device 600 and the characteristic data of the substrate that matches the conditions forming the representative data creation data group. Collect from 605. Then, the data collection unit 603 passes the obtained representative data creation data group to the representative data creation unit 606 and passes the characteristic data of the analysis target substrate to the similarity evaluation unit 607. At this time, if necessary, information such as the substrate identification ID and the inspection date may be collected together with the inspection data.

  The inspection information collection system 604 is an inspection information collection system that collects inspection information from inspection devices arranged in the manufacturing process. In this example, the inspection information collection system 604 is connected to the inspection device 605, and the inspection data and inspection date / time of the substrate processed by the inspection device 605, the identification ID of the substrate, and the like are accumulated. Further, the inspection information collection system 604 is connected to the distribution analysis apparatus 600, and the data collection unit 603 collects necessary inspection data from the inspection information collection system 604.

  The inspection device 605 is arranged in the manufacturing process and actually inspects the substrate. The inspection device 605 is connected to the distribution analysis device 600, and the data collection unit 603 collects necessary inspection data from the inspection device 605.

  The representative data creation unit 606 receives the representative data creation data group collected by the data collection unit 603 from the inspection information collection system 604 and the inspection apparatus 605 and the distribution shape data registered in the pattern registration unit 602, and the process of step S102 To create representative data. The created representative data is passed to the similarity evaluation unit 607.

  The similarity evaluation unit 607 receives the characteristic data of the analysis target board collected by the data collection unit 603 and the representative data created by the representative data creation unit 606, and performs the similarity evaluation process in step S103. The calculated similarity is passed to the data output unit 608. At this time, if necessary, conditions such as the identification ID of the substrate, the inspection date and time, representative data, etc. may be sent together with the similarity.

  The data output unit 608 receives the similarity calculated by the similarity evaluation unit 608, processes it into a data format used by the output device 609, and transmits it to the output device 609. At this time, if necessary, various types of data may be received from the pattern registration unit 602, the data collection unit 603, the representative data creation unit 606, and processed into a format used by the output device 608.

  The output device 609 outputs the distribution analysis result by the distribution analysis device 600 through a monitor, paper output, a magnetic disk, a portable semiconductor memory, or the like.

  A plurality of distribution shape patterns may be registered in the pattern registration unit 602. In this case, the representative data creation unit 606 creates representative data for each of the registered distribution shape patterns. Then, the similarity evaluation unit 607 evaluates the similarity between the characteristic data of the analysis target substrate and each representative data.

  Moreover, when all the information necessary for distribution analysis can be acquired from either one of the inspection information system 604 and the inspection apparatus 605, only one of the information that can be acquired may be connected.

  In addition, conditions necessary for the analysis are stored in the distribution analyzer 600, and when the inspection apparatus 605 inspects the substrate, the data collection unit 603 automatically causes the analysis target substrate characteristic data and the representative data creation data group. You may get

  In addition, the input device 601 and the output device 609 may be the same device, for example, a device in which a display unit serving as an output device is provided with a keyboard serving as an input device and integrated. Further, the input device 601 and the output device 609 may be included in the distribution analysis device 600.

  The output device 609 may receive and output information other than information necessary for distribution analysis from the input device 601, the inspection information collection system 604, or the inspection device 605 through the distribution analysis device 600.

  With the distribution analysis apparatus 600 having this configuration, the above-described distribution analysis method can be implemented.

  Note that the above-described distribution analysis method may be constructed as a program for causing a computer to execute the distribution analysis method.

  Further, such a program may be recorded on a computer-readable recording medium such as a CD-ROM and distributed. By installing the program in a general-purpose computer, the distribution analysis method can be executed by the general-purpose computer.

  The present invention is not limited to the distribution analysis apparatus 600 for pattern inspection, and can be applied to any data whose characteristics are planarly distributed. In that case, the data collection unit 603 is connected to an information collection system or characteristic measurement device in which target data is stored.

  Next, as Example 1, a method for realizing Step S102 and Step S103 in FIG. 1 will be specifically described.

  FIG. 7 shows steps S1021 and S1022 as a procedure for realizing step S102. In step S1021 in step S102, the relative magnitude relationship of the distribution of characteristic values on the substrate is similar to the relative magnitude relationship of the distribution of density values of the distribution shape pattern from the representative data creation data group. A second data group (hereinafter referred to as “similar data group”) is extracted. The similar data group needs to be selected so that the distribution shape of the characteristic data (characteristic value) included in the similar data group is similar to the distribution shape of the density value of the distribution shape pattern. However, the similar data group may include characteristic data whose distribution shapes are not similar, and need not include all characteristic data whose distribution shapes are similar. In step S1022 in step S102, characteristic data that best represents the distribution of the similar data group is calculated from the distribution of the similar data group extracted in step S1021, and the calculated characteristic data is used as the representative data.

  Next, a method for extracting a similar data group from the representative data creation data group in step S1021 will be specifically described.

  Each characteristic data included in the representative data creation data group is a coordinate representing the position of the defect in the circuit pattern, like the characteristic data of the analysis target substrate. Therefore, in order to compare each characteristic data included in the representative data creation data group and the distribution shape pattern, first, as illustrated in FIG. 8, the surface of each substrate 801 that provides the representative data creation data group, In the same manner as the distribution shape pattern is defined, a plurality of grid-like rectangular areas 301 are divided, and the number of defects in each rectangular area 301 is counted as a characteristic value of the rectangular area. Assuming that the number of rectangular regions 301 on each substrate is g, the characteristic data of the representative data creation data group is represented by a set of g numbers in the same manner as the distribution shape pattern. The numbers in each rectangular area 301 illustrated in FIG. 8 are obtained by counting the number of defects for each rectangular area 301 in the substrate 201 having the defect distribution shown in FIG. The characteristic value of a rectangular area where no number is described is zero. Hereinafter, in order to simplify the description, g = 2 unless otherwise specified.

  FIG. 9 is an example in which density values and characteristic values in the area 1 and area 2 as rectangular areas are plotted in the g-dimensional characteristic data space for the distribution shape pattern 901 and the representative data creation data group. When the distribution shape pattern 901 and the data 902 and 903 of each substrate forming the representative data creation data group are plotted, the distance between the origin and the characteristic data (the plotted point) indicates the density value or the size of the characteristic value. Therefore, data having a larger characteristic value is plotted at a position farther from the origin. Further, the characteristic data of the substrates having the same relative magnitude relationship (ratio in this example) between the characteristic values of the plurality of rectangular areas are plotted in the same straight line regardless of the characteristic value. Therefore, the angle formed by the origin and the characteristic data represents the distribution shape of the data. Considering them, for example, when a certain abnormal equipment has a defect with a distribution shape substantially equal to the distribution shape pattern 901, it is included in the representative data creation data group as shown in FIG. The data group 902 to be extracted as the similar data group from the characteristic data is densely distributed with a normal distribution spread around a certain point on the straight line 905 passing through the point representing the distribution shape pattern 901 and the origin. To do. In this case, data groups that should not be extracted as similar data groups are sparsely scattered outside the range where the data group 902 exists. In addition, when a certain abnormal equipment causes a defect with a distribution shape slightly deviated from the distribution shape pattern 901, as shown in FIG. 9B, the similarity among the characteristic data included in the representative data creation data group A data group 903 to be extracted as a data group is densely distributed with a normal distribution spread centered on a point slightly shifted from a straight line 905 passing through the point representing the distribution shape pattern 901 and the origin.

  Therefore, a certain angle range 904 expected from the origin is set as an allowable range in which the variation of the analysis target data is reflected on the distribution shape pattern 901 in a mode including a point representing the distribution shape pattern 901. Of the characteristic data included in the representative data creation data group, data whose points representing the data are included in the angle range 904 are data to be extracted as a similar data group. Therefore, a set of such characteristic data is extracted as a similar data group. On the other hand, data in which the point representing the data is not included in the angle range 904 is data that should not be extracted as a similar data group, and thus is not extracted.

この相関係数ｒは−１から１の範囲を取り、値が大きいほど分布形状パターンＴと特性データＭのなす角度は近い。そこで、相関係数ｒが予め定めた閾値以上であるような特性データＭを分布形状パターンＴと類似した類似データ群として抽出する。 Whether or not the characteristic data included in the representative data creation data group is included in the angle range 904 can be determined using, for example, a correlation coefficient that is a statistic for evaluating an angle formed by two sets of characteristic data. Certain characteristic data M included in the distribution shape pattern T and the representative data creation data group has g rectangular regions, and g-dimensional vectors T = (t ₁ ,..., T _g ), M = (m ₁ , ..., m _g ). At this time, the correlation coefficient r between the distribution shape pattern T and the characteristic data M is expressed by the following (Equation 1).

The correlation coefficient r ranges from −1 to 1, and the angle between the distribution shape pattern T and the characteristic data M is closer as the value is larger. Therefore, characteristic data M whose correlation coefficient r is equal to or greater than a predetermined threshold is extracted as a similar data group similar to the distribution shape pattern T.

Next, the method for calculating representative data in step S1022 will be specifically described. By the processing in step S1201, most of the characteristic data included in the similar data group is data similar to the distribution shape pattern 901. There is very little data that is not similar to the distribution shape pattern 901, and it is considered that the distribution is uniformly and sparsely distributed in the angle range 904. Therefore, by making the average value of the characteristic data included in the similar data group as representative data, the contribution of data not similar to the distribution shape pattern 901 is relatively reduced, and the representative data similar to the distribution shape pattern can be easily obtained. Can be sought. That is, when n characteristic data M ₁ ,..., M _n are included in the similar data group, the representative data P = (p ₁ ,..., P _g ) is expressed by the following (Equation 2).

  In this way, the representative data P that is similar to the distribution shape pattern 901 and reflects the variation of the analysis target data can be created.

  Next, the process for obtaining the similarity in step S103 in FIG. 1 will be specifically described.

  First, the concept of similarity will be described with reference to FIG. In the g-dimensional characteristic data space of FIG. 10, points representing characteristic data 1001 of the analysis target substrate, points representing representative data 1002 created in step S102 in FIG. 1, and points representing predetermined reference data 1003 are plotted. Has been. Here, the reference data is fixed data that serves as a reference for calculating similarity, and in this example, characteristic data of a substrate manufactured in a manufacturing process in a state where no abnormal equipment exists is used. When the “characteristic value” is “the number of defects” on the substrate as in the example, data indicating that “the number of defects is zero” can be set as the “reference data” over the entire area of the substrate. Here, the similarity indicates the degree to which the analysis target data 1001 is similar to the representative data 1002. Specifically, the first difference 1004 between the characteristic value distribution of the reference data 1003 and the characteristic value distribution of the analysis target data 1001, the characteristic value distribution of the representative data 1002, and the analysis target data 1001. Is compared with the second difference 1005 between the distribution of the characteristic values of the first and second values 1005 in comparison with the first difference 1004. As the second difference 1005 becomes smaller, the degree of similarity increases quantitatively. It is a defined value.

  FIG. 11 shows a procedure for calculating such similarity in step S103 in FIG. As shown in FIG. 11, in step S1031, a first difference 1004 between the reference data 1003 and the analysis target substrate characteristic data 1001 is calculated. Next, in step S1032, a second difference 1005 between the representative data 1002 created in step S102 of FIG. 1 and the analysis target substrate characteristic data 1001 is calculated. In step S1033, the similarity is calculated from the two differences 1004 and 1005.

More specifically, in order to obtain the first difference 1004 between the reference data 1003 and the analysis target board characteristic data 1001 in step S1031, the distance between the reference data 1003 and the analysis target board characteristic data 1001. determine the D _0. When the characteristic data X of the analysis target board is g-dimensional vector X = (x ₁ ,..., X _g ) and the reference data B is g-dimensional vector B = (b ₁ ,..., B _g ), the above distance D ₀ is It is calculated by the following formula.

  In the pattern inspection, it is desirable to use a state in which no defect exists on the substrate as a reference, and therefore, B = (0,..., 0) as reference data B. (Equation 3) is a distance scale called Euclidean distance.

  In order to obtain the second difference 1005 between the representative data 1002 and the analysis target substrate characteristic data 1001 in step S1032, the distance D between the representative data P and the analysis target substrate characteristic data X may be obtained. The distance D can be obtained by replacing the reference data B with the representative data P in (Equation 3) in step S1031.

Then calculates the quantitative similarity S is compared with the distance D ₀ and the distance D in step S 1033. In this example, the similarity S is larger as the distance D is smaller than the distance D ₀ , and the evaluation scale is constant regardless of the analysis target substrate data 1001 or representative data 1002. Thus, it is defined by the following (Equation 4).

  Since the similarity S defined in (Equation 4) is always a value in the range of 0 to 1, any distribution shape pattern or analysis target data can be evaluated on a certain scale. .

  In this way, the process of FIG. 1 can be specifically realized. That is, representative data is created by the processing of steps S1021 to S1022 in FIG. 7 for the predefined distribution shape pattern, and the similarity is calculated by the processing of steps S1031 to S1033 in FIG. Depending on the degree of similarity, it is possible to quantitatively evaluate whether or not the data to be analyzed is similar to the distribution shape pattern.

  In the above description, the first difference 1004 between the characteristic data 1001 of the analysis target board and the reference data 1003 and the second difference 1005 between the characteristic data 1001 of the analysis target board and the representative data 1002 are Euclidean. Although the calculation is performed using the distance (Equation 3), the difference may be calculated using a distance scale other than the Euclidean distance such as the Mahalanobis distance.

  FIG. 12 shows a detailed configuration of a distribution analysis apparatus 600 according to an embodiment of the present invention suitable for carrying out the above-described distribution analysis method (see FIGS. 1, 7, and 11).

  The lattice data conversion unit 1201, which is an internal component of the data collection unit 603, coordinates the data group for representative data creation and the pattern inspection data of the analysis target substrate collected by the data collection unit 603 from the inspection information collection system 604 and the inspection apparatus 605. The unit information is converted into the number of defects for each rectangular area 301 as shown in FIG. The converted characteristic value (number of defects in each rectangular area) is sent to the representative data creation unit 606 and the similarity evaluation unit 607 by the data collection unit 603 as the characteristic data of each substrate.

  The similar data group extraction unit 1202 which is an internal component of the representative data creation unit 606 receives the representative data creation data group sent from the data collection unit 603 and the distribution shape data registered in the pattern registration unit 602, and FIG. The similar data group is extracted by performing the process of step S1021. The extracted similar data group is transferred to a representative data calculation unit 1203 that is also an internal component of the representative data creation unit 606.

  The representative data calculation unit 1203, which is an internal component of the representative data creation unit 606, receives a similar data group from the similar data group extraction unit 1202, which is also an internal component of the representative data creation unit 606, and performs the process of step S1022 in FIG. To calculate the characteristic value of the representative data from the distribution of characteristic values of the similar data group, that is, create representative data.

The first distance calculation unit 1204, which is an internal component of the similarity evaluation unit 607, includes reference data 1003 recorded as a constant on a memory (not shown) and characteristic data 1001 of the analysis target substrate sent from the representative data creation unit 603. 11 is performed, and the process of step S1031 in FIG. 11 is performed to calculate the distance D ₀ between the reference data 1003 and the characteristic data 1001 of the analysis target substrate. The calculated distance D ₀ is sent to a similarity calculation unit 1206 that is also an internal component of the similarity evaluation unit 607.

  The second distance calculation unit 1205 that is an internal component of the similarity evaluation unit 607 receives the representative data 1002 created by the representative data creation unit 606 and the analysis target substrate characteristic data 1001 sent from the data collection unit. 11, the distance D between the representative data 1002 and the analysis target substrate characteristic data 1001 is calculated. The calculated distance D is sent to a similarity calculation unit 1206 which is also an internal component of the similarity evaluation unit 607.

The similarity calculation unit 1206, which is an internal component of the similarity evaluation unit 607, has the distance D calculated by the first distance calculation unit 1204 and the second distance calculation unit 1205, which are also internal components of the similarity evaluation unit 607. ₀ and D are received and the processing of step S1033 in FIG. 11 is performed to calculate the similarity S of the characteristic data of the analysis target substrate with respect to the distribution shape pattern. The calculated similarity S is sent to the data output unit 608.

  Other configurations of the distribution analysis apparatus 600 are the same as those in FIG.

  With the distribution analysis apparatus 600 having this configuration, the above-described distribution analysis method can be realized.

  Next, as Example 2, another method for calculating representative data in step S1022 in FIG. 7 will be described.

  In Example 1 described above, the average value of characteristic data included in the similar data group, more specifically, a value obtained by simple averaging is used as representative data. On the other hand, in the second embodiment, when the characteristic values are averaged for each rectangular area with respect to the similar data group instead of the simple average, the characteristic value corresponding to each rectangular area and the distribution shape pattern are assigned to each rectangular area. A weighted average weighted by a correlation coefficient with the density value is calculated. A value obtained by weighted averaging is used as representative data.

Specifically, in step S1022 of FIG. 7, the correlation coefficient between the similar data group M ₁ ,..., M _n composed of n characteristic data and the distribution shape pattern T is set to r _i (i = 1,..., N ). The correlation coefficient r _i is a value already calculated by (Equation 1) in step S1021 of FIG. At this time, as the representative data P, a weighted average obtained by assigning the weight of the correlation coefficient r to each data included in the similar data group is calculated by the following (Equation 5).

  The difference between the representative data calculated in the first embodiment and the representative data calculated in the second embodiment will be described with reference to FIG. FIG. 13A shows a mode in which representative data 1303 is calculated by the method of the first embodiment when a similar data group 1302 is extracted for a certain distribution shape pattern 1301. On the other hand, FIG. 13B shows a mode in which the representative data 1303 is calculated by the method of the second embodiment when a similar data group 1302 is extracted for a certain distribution shape pattern 1301. In FIG. 13, a straight line 1305 that passes through the point representing the distribution shape pattern 1301 and the origin, and an angle range 1304 that is set to include the straight line 1305 and that is an allowable range that reflects fluctuations in the analysis target data. And are shown together.

  In the first embodiment, the representative data 1303 is calculated as a simple average value of the similar data group. Therefore, as shown in FIG. 13A, the point representing the representative data 1303 in the g-dimensional characteristic data space is the similar data group 1302. Is located approximately in the middle of the point (a set of points). On the other hand, in the second embodiment, the representative data 1303 is calculated by the weighted average according to (Equation 5). Therefore, as shown in FIG. 13B, the point representing the representative data 1303 in the g-dimensional characteristic data space is The representative data 1303 is plotted at a position closer to a straight line 1305 that passes through the point representing the distribution shape pattern 1301 and the origin, as compared with the data created in 1. As a result, when the representative data 1303 is calculated by the method of the second embodiment, the evaluation result in step S103 in FIG. 1 indicates the characteristics of the analysis target substrate as compared to the case where the representative data is calculated by the method of the first embodiment. The closer the data is to the distribution shape of the distribution shape pattern 1301, the higher the similarity S is evaluated.

  In this case, the evaluation of the similarity degree is more focused on the distribution shape of the distribution shape pattern 1301 than in the case of the first embodiment.

  Next, as Example 3, another method for calculating the similarity in step S103 in FIG. 1 will be described.

  As shown in FIG. 14, it is assumed that representative data 1401 is obtained as a point in the g-dimensional characteristic data space. In FIG. 13, a straight line 1405 passing through the point representing the representative data 1401 and the origin is also shown. As described above, in the g-dimensional characteristic data space, the distance between the origin and the characteristic data (plotted points) represents the magnitude of the characteristic value. The angle formed by the origin and the characteristic data represents the distribution shape of the data. Generally speaking, compared to the difference 1402 in the characteristic value with respect to the representative data 1401, analysis is often performed with an emphasis on the difference 1403 in the distribution shape. However, in such an analysis, if the similarity is calculated by (Equation 4) in step S1033 in FIG. 11, the difference 1402 in the characteristic value and the difference 1403 in the shape are handled in the same way. The result may not be obtained. Therefore, the degree of similarity may be evaluated with more emphasis on the difference 1403 in the distribution shape.

  As described above, as a specific method for evaluating the similarity with an emphasis on the distribution shape difference 1403, for example, in step S103 in FIG. 1, there is an angle formed between the representative data 1401 and the characteristic data of the analysis target substrate. It is conceivable to calculate the degree of similarity only in the case of being within the range by (Equation 4). FIG. 15 shows the procedure.

  That is, first, in step S1030, the correlation coefficient between the representative data and the characteristic data of the analysis target substrate is obtained by the above-described (Equation 1). When the obtained correlation coefficient value is equal to or greater than a predetermined threshold value, it is determined that the difference in distribution shape between the representative data and the characteristic data of the analysis target substrate is within an allowable range, and steps S1031 to S103 in the first embodiment The similarity is obtained in the process of S1033. On the other hand, if the value of the correlation coefficient obtained in step S1030 is less than a predetermined threshold value, it is determined that the difference in distribution shape between the representative data and the characteristic data of the analysis target substrate is outside the allowable range, and step In S1034, the similarity is set to 0 (zero). As a result, the similarity obtained in step S1033 or step S1034 becomes the similarity of the characteristic data of the analysis target substrate.

  In such a case, the evaluation of the similarity degree is more focused on the distribution shape of the distribution shape pattern.

  Next, as Example 4, using the substrate analysis result obtained by the distribution analysis method described above, an abnormality that estimates a manufacturing facility (that is, an apparatus in which an abnormality has occurred) that caused the generation of a characteristic distribution in the manufacturing process A facility estimation method will be described.

  FIG. 16 schematically shows an example of a manufacturing process of a substrate to which this abnormal facility estimation method is applied. This manufacturing process includes a plurality of manufacturing facilities a, b, c, and d that are sequentially executed on the substrate. The pattern inspection e is performed after the completion of these manufacturing processes.

  In a certain process included in the manufacturing process, in this example, the processes b and d, a plurality of facilities each capable of executing the manufacturing process are arranged in order to perform the manufacturing process in parallel. In this example, three facilities, Unit 1, Unit 2, and Unit 3, are arranged in step b, and two units, Unit 1, and Unit 2, are arranged in step d. Each substrate that has flowed into the manufacturing process as the manufacturing process progresses is processed at any time by any equipment arranged in the manufacturing process in order to increase the production capacity. It is not determined which equipment is used to process each substrate in the process. In this arrangement, when the equipment inspected by the pattern inspection in FIG. 16 is processed in each process, as shown in FIG. The ratio of the substrates processed in step 1 is substantially equal in each process, such as 1/3 each, and the ratio of substrates processed in each facility in process d is 1/2. Hereinafter, information such as which equipment has processed each substrate in each process is referred to as a processing history.

  On the other hand, if abnormalities occur in any of the equipment in the process and defects are concentrated in a specific area on the board, extracting the board with a similar characteristic distribution will cause the equipment that caused the problem to exist Then, the board | substrate processed by the causal equipment will be extracted unevenly. However, since the equipment for processing these substrates in processes other than the causal process is not constant, the above-described bias does not occur in other processes. For example, out of the substrates that have undergone the manufacturing process of FIG. 16, when a substrate group in which defects are concentrated in the upper right part in the first machine in step b is extracted, as shown in FIG. Then, the ratio of the substrates processed in Unit 1 (4/5 in this example) was high, and the ratio of substrates processed in Units 2 and 3 (1/10 in this example) was reduced. There is a bias in the number of processed substrates. There is no deviation between the facilities in the ratio of the substrates processed in step d for the same substrate group. From this, by extracting a group of substrates with similar characteristic distribution, collecting the processing history of the substrate, and analyzing the process where the number of processed substrates among the facilities allocated to the same process is most uneven, It can be seen that the equipment can be estimated.

  FIG. 18 shows a method for estimating the abnormal equipment that is the cause by using the distribution analysis result of the substrate characteristic value by the above-described distribution analysis method on the assumption of such a situation.

  First, in step S1801 including a series of steps S101 to S103 in FIG. 1, the degree of similarity with respect to the distribution shape pattern to be estimated is evaluated for each substrate included in the substrate group to be an abnormal facility estimation target. . However, in step S102, a representative data creation data group is obtained from the characteristic data of all the boards included in the estimation target board group, and representative data is created using the representative data creation data group. In step S103, representative data is created. The similarity to the distribution shape pattern is evaluated for all the substrates included in the estimation target substrate group.

  In step S1802, boards whose similarity obtained in step S1801 is equal to or greater than a predetermined threshold are extracted, and those boards are set as board groups corresponding to the distribution shape pattern to be estimated. As the threshold value, for example, a value that allows a human to visually determine that the substrate corresponds to the distribution shape pattern to be estimated is employed.

  In step S1803, a processing history in each manufacturing process is acquired for the substrates included in the substrate group extracted in step S1802. Since this information is usually collected and managed by a process information collection system installed in the manufacturing process, it can be easily obtained.

  In step S1804, the processing history acquired in step S1803 is tabulated for each process of the manufacturing process, and the number of processed sheets for each facility is calculated.

  Finally, in step S1805, the number of processed sheets for each facility in each process is compared, and the facility having the largest deviation in the number of processed sheets is statistically examined to estimate that it is an abnormal facility. If there is no significant difference in the number of processed sheets, it is determined that the characteristic distribution is not caused by a specific facility.

  According to this abnormal equipment estimation, since the deviations in the processing history are compared from substrates having similar characteristic distributions, it is possible to perform abnormal equipment estimation with high accuracy.

  FIG. 19 shows the configuration of an abnormal equipment estimation device 1900 according to an embodiment of the present invention suitable for implementing the above-described abnormal equipment estimation method. This abnormal facility estimation apparatus 1900 includes a pattern registration unit 1902, a characteristic data collection unit 1903 as an input unit, a distribution analysis unit 1904, a board group extraction unit 1905, a history data collection unit 1906 as an input unit, a processing frequency calculation unit 1908, An abnormal equipment estimation unit 1909 and a data output unit 1910 are included. The characteristic data collection unit 1903 is connected to an input device 1901, an inspection information collection system 604 and an inspection device 605, the history data 1906 is connected to a process information collection system 1907, and the data output unit 1910 is connected to an output device 1911. Yes.

  The input device 1901 is composed of, for example, a keyboard and a mouse. In this example, the input device 1901 is an abnormal facility estimation device 1900, the range of the inspection date and time of the substrate that is the target of abnormal facility estimation, the range of the number of defects detected by the inspection device, or the target of abnormal facility estimation. It is used to input conditions for abnormal equipment estimation targets such as the scope of the manufacturing process. In addition, the distribution shape pattern used for abnormal equipment estimation is also registered in the abnormal equipment estimation apparatus 1900 using the input device 1901.

  The pattern registration unit 1902 has the same function as the pattern registration unit 602 of the distribution analysis apparatus 600 shown in FIG. 6, receives the distribution shape pattern defined by the input device 1901, and records it in the database. The information received from the input device 1901 is not only the distribution shape pattern but also information related to the distribution shape pattern, for example, information related to the device that causes the distribution shape pattern, and is recorded in association with the distribution shape pattern. Also good.

  The characteristic data collection unit 1903 collects, from the inspection information collection system 604 and the inspection apparatus 605, the inspection data of the board group and the identification ID of the board that match the conditions of the estimation target board transmitted from the input device 1901 to the abnormal equipment estimation apparatus 1900. Then, it is passed to the distribution analysis unit 1904. At this time, information such as the inspection date and time may be collected together with the inspection data.

  The inspection information collection system 604 is the same as the inspection information collection system 604 connected to the distribution analyzer 600. The inspection information collection system 604 is connected to the abnormal equipment estimation apparatus 1900, and the characteristic data collection unit 1903 collects necessary inspection data from the inspection information collection system 604.

  The inspection device 605 is the same as the inspection device 605 connected to the distribution analysis device 600. The inspection device 605 is connected to the abnormal equipment estimation device 1900, and the data collection unit 1903 collects necessary inspection data from the inspection device 605.

  The distribution analysis unit 1904 obtains the estimation target board group characteristic data and the distribution shape pattern registered in the pattern registration unit 1902 sent from the characteristic data collection unit 1903, and performs the processing of steps S102 and S103 in FIG. Do. Then, the obtained similarity degree of each estimation target board group with respect to the distribution shape pattern is associated with the board ID and passed to the board group extraction unit 1905. The internal configuration of the distribution analysis unit 1904 includes a representative data creation unit 606 and a similarity evaluation unit 607 as in the distribution analysis device 600 (see FIG. 6).

  The board group extraction unit 1905 receives the similarity of the estimation target board group from the distribution analysis unit 1904, and extracts a board whose similarity is equal to or higher than a predetermined threshold. The extracted board identification ID is passed to the history data collection unit 1906 and the processing frequency calculation unit 1908.

  The history data collection unit 1906 receives the substrate group identification ID extracted from the estimation target substrate group by the substrate group extraction unit 1905, and searches the process information collection system 1907 for processing history information of the substrate included in the substrate group. The search result is passed to the processing frequency calculation unit 1908.

  The processing frequency calculation unit 1908 uses the identification ID of the substrate received from the substrate group extraction unit 1905 and the processing history information of each substrate received from the history data collection unit 1906 to calculate the number of processings for each facility arranged in each process. Calculate and pass to the abnormal equipment estimation unit 1909.

  The abnormal equipment estimation unit 1909 performs statistical processing on the number of processed sheets received from the processing frequency calculation unit 1909, and the equipment having the largest number of processed sheets assigned to the process having the largest deviation in the number of processed sheets between facilities. Is detected and passed to the data output unit 1910 as an abnormal equipment estimation result.

  The data output unit 1910 receives the abnormal equipment estimation result from the abnormal equipment estimation unit 1908, processes the data into a data format used by the output device 1911, and transmits it to the output device 1911. At this time, if necessary, various types of data may be received from the pattern registration unit 1902, the characteristic data collection unit 1903, the distribution analysis unit 1904, the history data collection unit 1906, and processed into a format used by the output device 1911.

  The output device 1911 outputs the abnormal equipment estimation result by the abnormal equipment estimation device 1900 through a monitor, paper output, a magnetic disk, a portable semiconductor memory, or the like.

  A plurality of distribution shape patterns may be registered in the pattern registration unit 1902, and abnormal facility estimation may be performed for each distribution shape pattern.

  Moreover, when all the information necessary for abnormal equipment estimation can be acquired from any of the inspection information collection system 604, the process information collection system 1907, and the inspection apparatus 605, only the apparatus or system necessary for acquisition may be connected.

  Further, the input device 1901 and the output device 1911 may be the same device, for example, a device in which a display unit serving as an output device is provided with a keyboard serving as an input device and configured integrally. Further, the input device 1901 and the output device 1911 may be included in the abnormal equipment estimation device 1900.

  Further, the output device 1911 receives and outputs information other than information necessary for abnormal facility estimation from the input device 1901, the inspection process information collection system 604, the process information collection system 1907, and the inspection device 605 through the abnormal facility estimation device 1900. Also good.

  As described above, the pattern registration unit 1902 and the distribution analysis unit 1904 have the same functions as the distribution analysis device 600. Therefore, as shown in FIG. 20, these components can be configured separately from the abnormal facility estimation apparatus 1900 as the distribution analysis apparatus 600. In the configuration shown in FIG. 20, the input device 1901 receives the similarity and identification ID of each substrate from the distribution analysis device 600 and transmits them to the substrate group extraction unit 1905. The subsequent processing is performed in the same manner as the processing by the abnormal facility estimation apparatus 1900 in FIG.

  According to this abnormal equipment estimation apparatus 1900, the above-mentioned abnormal equipment estimation method can be implemented.

  In addition, you may construct | assemble as a program for making a computer perform the above-mentioned abnormal equipment estimation method.

  Further, such a program may be recorded and distributed on a computer-readable recording medium such as a CD-ROM. By installing the program in a general-purpose computer, the abnormal equipment estimation method can be executed by the general-purpose computer.

It is a figure which shows typically the concept of the distribution analysis method of one Embodiment of this invention. It is a figure which shows the example of the test result of a pattern test | inspection. It is a figure which shows typically the concept of a rectangular area and a distribution shape pattern. It is a figure which shows typically the fluctuation | variation of the distribution shape of the defect by various factors. It is a figure which shows typically the fluctuation | variation of the number of defects by various factors. It is a figure which shows the structure of the distribution analyzer of one Embodiment of this invention. It is a figure which shows the procedure which produces representative data in the distribution analysis method of the said one Embodiment. It is a figure which shows typically the concept of the characteristic data in one Embodiment of this invention. It is a figure which shows typically the method of extracting a similar data group from the data group for representative data creation in order to create the said representative data. It is a figure which shows typically the concept of the similarity degree in one Embodiment of this invention. It is a figure which shows the procedure which evaluates the said similarity. It is a figure which shows concretely the internal structure of the distribution analyzer of the said one Embodiment. It is a figure which shows the two aspects which define the said representative data. It is a figure explaining the shift | offset | difference (difference) from the distribution shape which the said representative data has, and the magnitude | size of a characteristic value. It is a figure which shows the procedure which calculates the similarity which has a definition different from the said similarity. It is a figure which shows typically the example of the manufacturing process of the board | substrate used as the application object of the abnormal facility estimation method of one Embodiment of this invention. It is a figure which shows the example which compared the number of sheets processed with the manufacturing equipment at the time of the normal in the said manufacturing process, and the time of abnormality. It is a figure which shows typically the concept of abnormal equipment estimation in one Embodiment of this invention. It is a figure which shows the structure of the abnormal equipment estimation apparatus of one Embodiment of this invention. It is a figure which shows the modification of the abnormal equipment estimation apparatus of FIG. It is a figure explaining a certain prior art. It is a figure explaining another prior art.

Explanation of symbols

600 Distribution analyzer 1900 Abnormal equipment estimation device

Claims (21)

A distribution analysis method for evaluating whether analysis target data having characteristics distributed along a plane is similar to a specific distribution shape pattern,
The distribution shape pattern is divided into a plurality of rectangular areas in a lattice shape on a plane corresponding to the plane defining the analysis target data, and the relative magnitude of the characteristic value of the analysis target data is determined for each rectangular area. In order to express the relationship, define a concentration value that can take multiple values within a certain numerical range,
Create representative data that is similar to the distribution shape pattern and reflects the variation that occurs between the multiple data to be analyzed,
The analysis target data is compared with the representative data to obtain a similarity indicating the degree to which the analysis target data is similar to the representative data, and the analysis target data corresponds to the distribution shape pattern according to the similarity. A distribution analysis method characterized by evaluating whether it is similar to. The distribution analysis method according to claim 1,
The above representative data is created
From the first data group formed by the plurality of analysis target data including the variation, the relative magnitude relation of the distribution of characteristic values seems to be similar to the relative magnitude relation of the distribution of density values of the distribution shape pattern. A first step of extracting a second data group,
And a second step of calculating the representative data from the distribution of characteristic values of the second data group obtained in the first step. The distribution analysis method according to claim 2,
In the first step, the correlation coefficient between the characteristic value corresponding to each rectangular area of the first data group and the density value assigned to each rectangular area of the distribution shape pattern is greater than or equal to a predetermined threshold value. A distribution analysis method characterized by extracting simple data as the second data group. The distribution analysis method according to claim 2,
In the second step, characteristic values are averaged for each rectangular area for the second data group obtained in the first step, and the representative data is configured using the average value for each rectangular area as a component. A distribution analysis method characterized by The distribution analysis method according to claim 4,
When the characteristic value is averaged for each rectangular area in the second data group, the correlation coefficient between the characteristic value corresponding to each rectangular area and the density value assigned to each rectangular area of the distribution shape pattern A distribution analysis method characterized by calculating a weighted weighted average. The distribution analysis method according to claim 1,
The process for obtaining the similarity is as follows:
A third step of calculating a first difference between a distribution of characteristic values of predetermined reference data and a distribution of characteristic values of the analysis target data;
A fourth step of calculating a second difference between the distribution of characteristic values of the representative data and the distribution of characteristic values of the analysis target data;
A distribution analysis method comprising: a fifth step of calculating the degree of similarity according to both the first difference and the second difference. The distribution analysis method according to claim 6,
The similarity calculated in the fifth step is obtained by dividing the reciprocal of the second difference by the sum of the reciprocal of the first difference and the reciprocal of the second difference. Method. The distribution analysis method according to claim 6,
The first difference calculated in the third step is a distance between the characteristic value distribution of the reference data and the characteristic value distribution of the analysis target data,
The distribution analysis method characterized in that the second difference calculated in the fourth step is a distance between the characteristic value distribution of the representative data and the characteristic value distribution of the analysis target data. The distribution analysis method according to claim 1,
When obtaining the similarity, it is determined whether or not the distribution shape of the characteristic value of the analysis target data is similar to the distribution shape of the characteristic value of the representative data, and the similarity is determined according to the determination result. A distribution analysis method characterized by calculating. The distribution analysis method according to claim 9,
When performing the determination, a correlation coefficient between a characteristic value corresponding to each rectangular area of the analysis target data and a characteristic value corresponding to each rectangular area of the representative data is calculated, and the correlation coefficient is A distribution analysis method characterized by determining whether or not a predetermined threshold value or more. In the distribution analysis method according to any one of claims 1 to 10,
A distribution analysis method characterized in that the characteristics of the data to be analyzed are distributed along the surface of the substrate. An abnormal facility estimation method for estimating a manufacturing apparatus in which an abnormality has occurred, from among the manufacturing apparatuses used in a plurality of manufacturing steps sequentially performed on a substrate,
The distribution analysis method according to claim 11 is performed to obtain the similarity,
Based on a processing history representing a manufacturing apparatus that has processed each of the substrates in each of the manufacturing steps, a manufacturing apparatus commonly used for the substrates having the similarity equal to or higher than a predetermined threshold is extracted. An abnormal equipment estimation method. A distribution analysis device that evaluates whether analysis target data having characteristics distributed along a plane is similar to a specific distribution shape pattern,
The distribution shape pattern is divided into a plurality of rectangular areas in a lattice shape on a plane corresponding to the plane defining the analysis target data, and the relative magnitude of the characteristic value of the analysis target data is determined for each rectangular area. A pattern registration unit that defines and assigns density values that can take multiple values within a certain numerical range to represent the relationship;
A representative data creating unit that creates representative data that is similar to the distribution shape pattern and reflects a variation that occurs between a plurality of the analysis target data;
The analysis target data is compared with the representative data to obtain a similarity indicating the degree to which the analysis target data is similar to the representative data, and the analysis target data corresponds to the distribution shape pattern according to the similarity. A distribution analysis apparatus comprising: a similarity evaluation unit that evaluates whether or not they are similar to each other. The distribution analyzer according to claim 13.
The representative data creation unit
From the first data group formed by the plurality of analysis target data including the variation, the relative magnitude relation of the distribution of characteristic values seems to be similar to the relative magnitude relation of the distribution of density values of the distribution shape pattern. A first part for extracting a second data group;
And a second part for calculating the representative data from the distribution of characteristic values of the second data group obtained in the first step. The distribution analyzer according to claim 13.
The similarity evaluation unit
A third portion for calculating a first difference between the distribution of characteristic values of the predetermined reference data and the distribution of characteristic values of the analysis target data;
A fourth portion for calculating a difference between a second distribution of characteristic values of the representative data and a distribution of characteristic values of the analysis target data;
A distribution analysis apparatus comprising: a fifth portion that calculates the degree of similarity according to both the first difference and the second difference. The distribution analyzer according to claim 13.
A distribution analysis apparatus characterized in that the characteristics of the analysis target data are distributed along the surface of the substrate. An abnormal facility estimation device that estimates a manufacturing device in which an abnormality has occurred, from among the manufacturing devices used in each of a plurality of manufacturing steps that are sequentially performed on a substrate,
The distribution analysis device according to claim 13 is provided, and the similarity is obtained by the distribution analysis device,
Based on a processing history representing a manufacturing apparatus that has processed each of the substrates in each of the manufacturing steps, an abnormal facility estimation is performed to extract a manufacturing apparatus that is commonly used for the substrates whose similarity is equal to or greater than a predetermined threshold. An abnormal equipment estimation device characterized by comprising a section.   A distribution analysis program for causing a computer to execute the distribution analysis method according to claim 1.   An abnormal equipment estimation program for causing a computer to execute the abnormal equipment estimation method according to claim 12.   A computer-readable recording medium on which the distribution analysis program according to claim 18 is recorded.   A computer-readable recording medium on which the abnormal equipment estimation program according to claim 19 is recorded.

Images