JP2007049020A - Method and device for sorting defect distribution - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method for sorting a defect distribution which can accurately sort defect distributions on substrates, even when the positional displacement or size of a defect on a substrate to be inspected is changed with respect to a known defect distribution pattern as a sorting reference. <P>SOLUTION: A plurality of closed regions Di are set to be partially overlapped with on another on the surface of a substrate (S301). Defect information including at least information on each defect position is acquired by detecting each defect on the substrate with the use of an inspection device (S302). A feature amount that indicates a feature of each defect belonging to the closed region is calculated for each of the closed regions Di on the basis of the defect information (S303). A defect distribution on substrate is sorted by comparing the calculated feature amount with a set of features indicating the known defect distribution pattern (S304 to S306). <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a defect distribution classification method and a defect distribution classification apparatus for detecting defects generated on a substrate and classifying the distribution thereof. An example of a substrate to be inspected is a wafer on which a thin film device is formed.

  In the manufacturing process of 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. When the yield decreases, the results of these inspections are investigated and analyzed to quickly identify the cause and take countermeasures, thereby minimizing the damage caused by the decrease in yield.

  In general, there are two methods for analyzing the cause of the decrease in yield: a physical analysis in which an analyst observes the state of a defect using an optical microscope, a scanning electron microscope, etc., and statistics on inspection information collected from the process. There is a numerical analysis that performs manual or empirical processing. In order to efficiently perform numerical analysis, a mechanism for collecting inspection information and processing history is often installed in the manufacturing process, and analysts browse inspection results of substrates on which defects have occurred through this mechanism. Or, the defect positions on a plurality of substrates can be displayed in an overlapping manner. In particular, in an abnormality caused by a manufacturing process or a manufacturing apparatus, defects often appear on the substrate in a distribution inherent to the abnormality, and the latter function is frequently used for the investigation.

  Here, when the results (defect positions) obtained by inspecting a large number of substrates are simply overlapped, inspection of substrates that are not related to the cause of yield reduction or substrates that are abnormal due to another cause The results are mixed and it is difficult to extract a distribution specific to the abnormality in question. Therefore, if a decrease in yield is detected, the analyst will extract the substrates that are assumed to be related to the abnormality from the manufacturing lot based on manufacturing knowledge and experience, and superimpose the defect positions on those substrates. Check the defect distribution. Therefore, it takes a lot of time until the cause is specified after appropriate extraction is performed, and the analysis work depends on the skill of the person in charge of analysis.

  In order to solve this problem, several regions are set on the surface of the substrate, and the defect distribution on the substrate is automatically determined based on the similarity between the occurrence of defects in each region and the known defect distribution pattern. A classification method has been proposed (see, for example, Patent Document 1). In patent document 1, in the manufacturing process of a semiconductor device, an inspection target wafer is divided into a plurality of concentric regions, and a histogram in the r direction (radial direction) in which the number of defects in those regions is arranged in order is created. Create a histogram in the θ direction (angular direction) by dividing the wafer into multiple fan-shaped areas and arranging the number of defects in those areas in order, and based on the positions of the change points of these two histograms, Classify the defect distribution.

Further, a method has been proposed in which a pattern of a combination necessary for determining the similarity is automatically generated with respect to a defect distributed in a ring shape and a defect distributed in a block shape (see, for example, Patent Document 2).
Japanese Patent Laid-Open No. 9-191032 JP 2003-59984 A

  By the way, in the actual manufacturing process, a certain amount of misalignment or change in the size of defects occurs. For this reason, the method of Patent Document 1 has the following problems.

  For example, as shown in FIG. 20A, a certain range on the inspection target wafer 2010 is divided into a region 2001 and a region 2002, and a defect 2003 is located in the center of the region 2001 in a known defect distribution pattern 2000. Shall. In this case, in the inspection target wafer 2010, the defect 2004 corresponding to the defect 2003 is correctly determined to belong to the region 2001 like the defect 2003 even if the defect 2004 is slightly displaced from the central portion corresponding to the defect 2003. The On the other hand, as shown in FIG. 20B, in the known defect distribution pattern 2000, when the defect 2005 is located at the end of the region 2001, the defect 2006 corresponding to the defect 2005 is detected in the inspection target wafer 2010. Is erroneously determined to belong to another region 2002 even though the positional deviation is similar to that of the defect 2004 in FIG. 20A. For this reason, the method of Patent Document 1 has a problem that even if the inspection target wafer 2010 actually has a defect distribution pattern similar to the known defect distribution pattern 2000, it is not correctly classified.

  Although the method of Patent Document 2 is intended to solve this problem, there are problems that the number of combination patterns is enormous and that the method cannot be applied to classification of defect distributions other than circular and massive.

  Accordingly, an object of the present invention is to provide a defect on a substrate even when a positional deviation or a change in size of the defect on the substrate to be inspected occurs with respect to a known defect distribution pattern which is a reference for classification. An object of the present invention is to provide a defect distribution classification method and a defect distribution classification apparatus that can classify distributions with high accuracy.

In order to solve the above problems, the defect distribution classification method of the present invention is:
Set multiple closed areas on the surface of the substrate so that they partially overlap each other,
Each of the defects on the substrate is detected by an inspection apparatus, and defect information including at least information indicating the position of each defect is obtained.
Based on the defect information, for each of the closed regions, to calculate a feature amount representing the characteristics of the defects belonging to the closed region,
The defect distribution on the substrate is classified by comparing the set of the feature quantity with the set of the feature quantity representing a known defect distribution pattern.

  Here, the “known defect distribution pattern” means information representing the distribution of the defect on the substrate when a known type of defect occurs.

  In the defect distribution classification method of the present invention, a plurality of closed regions are set so as to partially overlap each other on the surface of the substrate. Therefore, when a positional deviation or a change in the size of a defect on the substrate to be inspected is observed, a defect belonging to a certain closed region does not completely deviate from the closed region, but is separated from the closed region. There is a probability of entering a portion where this closed region overlaps (this is referred to as an “overlapping portion”). Thereby, the defect contributes to the calculation of the feature amount for each of the closed region and another closed region. Therefore, the defect distribution on the substrate is classified into the conventional example by classifying the defect distribution on the substrate by comparing the group formed by the feature amount with the group formed by the feature amount representing a known defect distribution pattern. Compared with higher accuracy. As a result, the occurrence of an abnormality classified into a known defect distribution pattern can be detected at an early stage, and a quick countermeasure can be taken.

  It is preferable that the set formed by the feature amounts is a vector having components corresponding to the closed regions, and each component represents a feature amount of a defect belonging to the corresponding closed region. In such a case, the calculation for the comparison can be performed by vector calculation. Therefore, the calculation for the comparison can be easily performed.

  In the defect distribution classification method according to an embodiment, the plurality of closed regions are rectangular regions having the same dimensions as each other and arranged in a matrix on the substrate surface.

  Here, the rectangular area includes not only a rectangular area but also a square area.

  In the defect distribution classification method according to this embodiment, the plurality of closed regions can be easily stacked on each other without any gaps on the substrate surface.

In one embodiment of the defect distribution classification method,
A plurality of the closed regions are set corresponding to regions having different defect densities in the known defect distribution pattern,
A portion where the pair of closed regions overlap each other corresponds to a transition region between the regions having different defect densities.

  Here, the “transition region” means a region where the density of defects corresponds to between the regions having different defect densities.

  In the defect distribution classification method according to this embodiment, the feature amount for each closed region corresponding to a certain known defect distribution pattern is determined based on whether the defect distribution on the wafer should be classified into the known defect distribution pattern. Express exactly. Therefore, even if the known defect distribution pattern has a complicated shape, the defect distribution on the wafer can be classified with high accuracy. In addition, if the types of defect distribution patterns are limited to some extent and most of them are known in advance, the number of wafer feature values can be limited. Therefore, the calculation is simplified.

  In the defect distribution classification method according to an embodiment, when the feature amount is obtained for a certain closed region, the defect information is weighted according to the degree of overlap for a portion where the closed region overlaps another closed region. It is characterized by performing.

  Here, the “degree of overlap” means the number of closed regions that overlap a certain region.

  In the defect distribution classification method according to the embodiment, when the feature amount is obtained for a certain closed region, a portion (overlapping portion) of the closed region that overlaps with another closed region depends on the degree of overlap. Since the defect information is weighted, the defect at the overlapping portion accurately contributes to the calculation of the feature amount for each of the closed regions overlapping therewith. Therefore, the defect distribution on the substrate can be classified with higher accuracy.

In one embodiment of the defect distribution classification method,
Classification by the above comparison is
A feature vector having a component corresponding to each of the closed regions, each representing a feature amount of a defect belonging to the corresponding closed region, and a plurality of feature vectors each representing a known defect distribution pattern are compared. Then, the defect distribution on the substrate is classified into the closest one of the known defect distribution patterns.

  In the defect distribution classification method of this embodiment, the calculation for the comparison can be performed by vector calculation. Therefore, the calculation for the comparison can be easily performed.

  The defect distribution classification method according to an embodiment is characterized in that a sum of weights for the defect information is set to 1.

  In the defect distribution classification method according to this embodiment, the defect in the overlapping portion accurately contributes to the calculation of the feature amount for each of the closed regions overlapping therewith. Therefore, the defect distribution on the substrate can be classified with higher accuracy.

  In the defect distribution classification method according to an embodiment, the feature amount for each closed region represents a defect density in the closed region.

  In the defect distribution classification method according to this embodiment, the feature for each closed region can be accurately represented by the feature amount.

The defect distribution classification apparatus of this invention is
A closed region setting section for setting a plurality of closed regions on the surface of the substrate so as to partially overlap each other;
An inspection apparatus for detecting defects on the substrate and acquiring defect information including at least information indicating the position of each defect; and
Based on the defect information, for each closed region, a feature amount calculation unit that calculates a feature amount that represents the feature of the defect belonging to the closed region;
A classification unit that classifies the defect distribution on the substrate by comparing the group formed by the feature amount with the group formed by the feature amount representing a known defect distribution pattern.

  According to the defect distribution classification apparatus of this invention, the defect distribution classification method of the said invention can be implemented and there can exist the above-mentioned effect. That is, in the defect distribution classification apparatus according to the present invention, a plurality of closed regions are set so as to partially overlap each other on the surface of the substrate. Therefore, when a positional deviation or a change in the size of a defect on the substrate to be inspected is observed, a defect belonging to a certain closed region does not completely deviate from the closed region, but is separated from the closed region. There is a probability of entering a portion where this closed region overlaps (this is referred to as an “overlapping portion”). Thereby, the defect contributes to the calculation of the feature amount for each of the closed region and another closed region. Therefore, the defect distribution on the substrate is classified into the conventional example by classifying the defect distribution on the substrate by comparing the group formed by the feature amount with the group formed by the feature amount representing a known defect distribution pattern. Compared with higher accuracy. As a result, the occurrence of an abnormality classified into a known defect distribution pattern can be detected at an early stage, and a quick countermeasure can be taken.

  A defect distribution classification apparatus according to an embodiment includes an occurrence frequency display unit that displays on a screen an occurrence frequency of a substrate in which a defect distribution is classified into a certain defect distribution pattern based on a result classified by the classification unit. It is characterized by that.

  In the defect distribution classification apparatus according to the embodiment, the occurrence frequency display unit displays the occurrence frequency of the substrate in which the defect distribution is classified into a certain defect distribution pattern on the screen based on the result classified by the classification unit. To do. Therefore, the person in charge of analysis can easily recognize the abnormality occurring in the manufacturing process of the substrate by looking at the display on the screen.

  A defect distribution classification apparatus according to an embodiment includes a warning unit that issues a warning when a substrate whose defect distribution is classified into a certain defect distribution pattern is generated based on the result of classification by the classification unit. And

  In the defect distribution classification apparatus of this embodiment, the warning unit issues a warning when a substrate in which the defect distribution is classified into a certain defect distribution pattern is generated based on the result classified by the classification unit. Therefore, the person in charge of analysis can easily recognize an abnormality occurring in the manufacturing process of the substrate by the warning and can take a countermeasure quickly.

  Further, it is desirable to provide a database unit that stores a plurality of sets of feature quantities representing the known defect distribution pattern. In this case, the defect distribution on the substrate can be easily classified into any of the known defect distribution patterns.

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

(First embodiment)
FIG. 1 shows the configuration of a defect distribution classification system 100 which is an embodiment of the defect distribution classification apparatus of the present invention. The semiconductor wafer 101 as the substrate is processed as a thin film device as a result of being processed in the manufacturing process 102 (for simplicity, this is referred to as a “wafer” including during and after the manufacturing process). Here, the manufacturing process 102 is to confirm a plurality of processes divided for each manufacturing process unit for the wafer 101 such as a cleaning process, a film forming process, an exposure process, and an etching process, and whether these manufacturing processes are performed correctly. And an inspection process for inspecting the wafer 101. The inspection process includes an appearance inspection for inspecting a defect of a circuit pattern formed on the wafer 101 and a foreign object inspection for inspecting a foreign substance existing on the wafer 101. The defect distribution classification system 100 is connected. In the present invention, an abnormality on the wafer 101 detected by the inspection apparatus 103 regardless of the inspection type is referred to as “defect”. After the inspection is completed, the inspection apparatus 103 transmits information for identifying the detected wafer 101 and defect information 105 describing the positions of all the defects on the wafer 101 to the defect distribution classification system 100.

  The defect distribution classification system 100 includes a data input / output unit 106 that performs data input / output processing, a calculation unit 107 that performs classification processing on defect information 105 and learning processing of known defect distribution information, a warning generation unit 108 as a warning unit, And an inspection DB (database) 109 for recording defect information and classification results, and a dictionary DB 110 for recording information representing the distribution of known types of defects on the wafer (hereinafter referred to as “defect distribution pattern”). .

  An information display client 112 is connected to the defect distribution classification system 100 via a network 111. The display client 112 can output a warning when an abnormality occurs by communicating with the data input / output unit 106, and can display the inspection result and the classification result recorded in the inspection DB 109 in a graph or a table. A dictionary registration client 113 is also connected to the network 111, and performs a known defect distribution pattern registration process for the dictionary DB 110 by communicating with the data input / output unit 106. Note that the network 104 and the network 111 may be the same.

  Next, details of processing by the defect distribution classification system 100 will be described.

  FIG. 2 is an example of defect information transmitted from the inspection apparatus 103 to the defect distribution classification system 100. The inspection apparatus 103 acquires detailed information such as coordinates X and Y and defect size of all the defects 201 detected on the wafer 101. The detailed information here is a measurement item of the inspection apparatus, such as the size of the foreign matter in the case of foreign matter inspection. All the detected defect information is transmitted to the defect distribution classification system 100 together with information such as the identification ID assigned to each wafer 101, the identification ID of the lot including the wafer 101, and the time when the inspection was performed.

The defect information 105 transmitted to the defect distribution classification system 100 is received by the data input / output unit 106 and sent to the calculation unit 107. FIG. 3 shows a schematic flow of processing in the calculation unit 107. As shown in this schematic flow, the arithmetic unit 107 first acts as a closed region setting unit, and a plurality of closed regions on the wafer surface according to a definition set in advance in a memory region (not shown) existing in the defect distribution classification system 100. setting the D _i so as to partially overlap each other (step S301). Here, D _i is an i-th (i is a natural number) closed region, and a set of fully closed regions is represented by D = {D ₁ , D ₂ ,..., D _i ,. Next, the coordinate information included in the defect information 105 is compared with the range occupied by each closed region D _i created in step S301, and a defect belonging to each closed region D _i is extracted (step S302). From the defect information then represents the position of the extracted defect, it calculates the characteristic amounts x _i representing the feature of the defect in each closed region D _i (step S303). Later for the contents of the feature x _i. Next, a set formed by all the feature values x _i calculated for the wafer 101 is set as a feature vector X = (x ₁ , x ₂ ,..., X _i ,...) (Step S304). Next, the similarity between the feature vector X of the wafer 101 obtained in step S304 and the feature vector (representative vector described later) representing each known defect distribution pattern recorded in the dictionary DB 110 is calculated (step S305). Finally, the defect distribution on the wafer 101 is classified into the defect distribution pattern having the highest similarity (step S306).

  Next, the processes of steps S301 to S306 will be described in detail.

4 shows an example of a closed region _{D i} set in step S301. In this example, a single closed region D _i is a rectangular area defined by a portion with the wafer 101 plane (in this example, square regions) is 401. The end of this rectangular area 401 is set to overlap with the end of another rectangular area. In the example of FIG. 5A, the two rectangular regions 501 and 502 are set to have an overlapping portion 503 between them. Such an overlapping portion is not limited to between two rectangular areas, and may be set between more rectangular areas. For example, in the example of FIG. 5B, four rectangular regions 504 to 507 are set to overlap each other. In this case, two adjacent rectangular areas are overlapped by a portion 508 along the side, and a portion (indicated by cross-hatching) 509 corresponding to a corner of the rectangular area has four rectangular areas 504 to 507 around it. Are overlapping. FIG. 6 shows an example in which a plurality of rectangular regions are set in a matrix so as to partially overlap each other on the surface of the wafer 101 in this way. Since the rectangular areas are used as the closed areas, the rectangular areas can be easily overlapped with no gap therebetween.

In step S302, defects existing in the closed region D _i set in step S301 are extracted as defects belonging to the closed region D _i . FIG. 7A separately shows rectangular regions 501 and 502 of FIG. 5A. Here, a white area 601 in the rectangular area 501 and a white area 602 in the rectangular area 502 are non-overlapping parts (referred to as “non-overlapping parts”). A range 603 indicated by stippling in the rectangular area 501 and a range 603 indicated by stippling in the rectangular area 502 correspond to the overlapping portion 503 shown in FIG. 5A. A defect existing in the range 601 belongs to the rectangular area 501, and a defect existing in the range 602 belongs to the rectangular area 502. A defect existing in the range 603 belongs to both the rectangular area 501 and the rectangular area 502.

In the present embodiment, the closed region D _i is set to have a partial overlap between the other closed regions. For this reason, if it is determined that a defect existing in each closed region D _i simply belongs to the closed region D _i , it is assumed that the defect existing in the overlapping portion belongs to all the closed regions corresponding to the position. is counted, the number of the sum of the entire wafer defect of the defect number belonging to each closed region D _i becomes not match. In addition, since the defects present in the overlapping portion are used in the calculation of the feature amount in each closed region D _i , the defects existing in the non-overlapping portion with respect to the wafer feature vector X obtained in step S304 described later. It will have a greater influence than. Therefore, the concept of affiliation u _ij is introduced as the degree to which the j th defect p _j belongs to the i th closed region D _i . At this time, the degree of membership u _ij of the defect p _j existing in the non-overlapping portion in the closed region D _i is set to 1, the degree of membership u _ik of the defect p _k existing outside the closed region D _{i is set} to 0, and The degree of membership u _ij of the defect p _j existing in the overlapping portion in the region D _i is set to a value between 1 and 0. Note that when all the regions D _i do not have an overlapping portion with other closed regions, the value of the degree of membership u _ik of the defect p _j is expressed by the following equation.

また、領域Ｄ_ｉに属する欠陥数Ｎ_ｉは次式で表される。

The number of defects N _i belonging to the region D _i is expressed by the following equation.

また、ウェハに存在する欠陥の合計数Ｎは次式で表される。

The total number N of defects present on the wafer is expressed by the following equation.

本実施形態では、閉領域Ｄ_ｉが互いに一部重なっている場合であっても式（２）および（３）の関係が成り立つように、上述のように所属度ｕ_ｉｊの概念を拡張して、式（１）に代えて次の式（４）を用いる。

In the present embodiment, the concept of the degree of membership u _ij is extended as described above so that the relationship of the expressions (2) and (3) is established even when the closed regions D _i partially overlap each other. The following formula (4) is used instead of the formula (1).

この式（４）を実現するには、各閉領域Ｄ_ｉ内の各部分（重なり部分と非重なり部分）について重なりの度合いに応じた重みｗを定義して合計が１になるように設定し、各部分についての重みｗがその部分に存在する欠陥の所属度に等しいものとすればよい。このようにした場合、各閉領域Ｄ_ｉに属する欠陥の数Ｎ_ｉおよびウェハに存在する欠陥の合計数Ｎはそれぞれ式（２）および式（３）をそのまま用いて求めることができる。

To achieve this equation (4), the total is set to be 1 defines the weight w corresponding to the degree of overlap for each portion (overlapped portion and the non-overlapping portions) in each closed region D _i The weight w for each part may be equal to the degree of membership of the defect existing in that part. In this case, the number N _{i of} defects belonging to each closed region D _i and the total number N of defects existing on the wafer can be obtained by using the equations (2) and (3) as they are, respectively.

The degree of affiliation u and the weight w will be described using the example of FIG. 7A. For the non-overlapping portion 601 in the rectangular area 501 and the non-overlapping portion 602 in the rectangular area 502, the weight w is set to 1, respectively. For the overlapping portion 603, the weight w is determined so that the sum is 1 between the rectangular area 501 and the rectangular area 502. In general, the weight w may be a value that is biased between the closed regions D _{i due} to, for example, a predetermined priority, or may be an equal value between the closed regions D _i . When an equal value is set between the closed regions D _i , the weight w is a value obtained by dividing 1 by the number of overlaps. That is, the weight w of the overlapping portion 603 of the two rectangular areas 501 and 502 is 0.5. When a defect exists in the non-overlapping portion 601, the degree of membership u of the defect with respect to the rectangular area 501 is 1, and the degree of belonging u with respect to the rectangular area 502 is 0. On the other hand, when a defect exists in the non-overlapping portion 603, the defect belongs to both the rectangular area 501 and the rectangular area 502 with the degree of belonging u = 0.5.

  The example of FIG. 7B shows each rectangular area 504 to 507 in FIG. 5B separately. For the non-overlapping portions 604 to 607, the weight w = 1 is set. In addition, since the two rectangular areas overlap each other in the portions 608 to 611 along the side, the weight w is set to 0.5 so that the sum of the weights w becomes 1. Since the four rectangular regions 504 to 507 overlap each other in the portion 612 corresponding to the corner, the weight w is set to 0.25 so that the sum is 1.

  Also, as shown in FIG. 7C, when the overlapping portion 615 between the rectangular region 613 and the rectangular region 614 is relatively wide, the overlapping portion 615 is divided into several segments, and weights w are respectively assigned to these segments. It is also possible to set. FIG. 7D shows the rectangular area 613 and the rectangular area 614 in FIG. 7C separately. In FIG. 7D, three segments 616 to 618 are set by dividing the overlapping portion 615 in FIG. 7C. For example, the segment 616 is located on the innermost side (the side closer to the non-overlapping part 613) in the overlapping part of the rectangular area 613 and on the outermost side (the side far from the non-overlapping part 614) in the overlapping part of the rectangular area 614. positioned. The segment 617 is located in the middle in the overlapping part of both rectangular areas. The segment 618 is located on the outermost side (the side far from the non-overlapping part 613) in the overlapping part of the rectangular area 613 and is located on the innermost side (the side closer to the non-overlapping part 614) in the overlapping part of the rectangular area 614. ing. From this positional relationship, the weight w for the overlapping portion of the rectangular area 613 is, for example, from the side closer to the non-overlapping portion 613 in order from the side closer to the non-overlapping portion 613 so that the sum of the weights w becomes 1 for each overlapping segment. On the other hand, the weight w for the overlapping portion of the rectangular area 614 is set to 0.3, 0.5, and 0.7 sequentially from the side far from the non-overlapping portion 614.

  By performing weighting in this way, the defect in the overlapping portion accurately contributes to the calculation of the feature amount for each of the rectangular regions overlapping therewith. Therefore, the defect distribution on the wafer can be classified with high accuracy.

  As for the feature amount used in the processing after step S303 in FIG. 3, a learning method and a classification method, which will be described later, various methods have been proposed not only in the field of semiconductor manufacturing but also in the field of pattern recognition. Although any method can be used in this method, an example of these methods will be described.

The feature quantity calculation processing in step S303, will be described using the defect density as a feature amount x _i. In general, the defect density C _i of a certain region is defined by the following equation (5), where N _{i is} the number of defects included in the region and A _i is the area of the region.

C _i = N _i / A _i (5)
Now described with respect to closed region D _i, the area A _i of the closed region D _i can be easily calculated from the size of the area to be set. The number N _i of defects belonging to closed region D _i is defined by equation (2) and (4), is a value obtained by adding the appertaining to that closed region D _i for all the defects belonging to the closed region D _i . In this example, the defect density obtained by Equation (5) for each closed region D _i is a feature amount x _i representing the feature of the defect in the closed region D _i .

In step S304, a set formed by all the feature amounts x _i calculated for the wafer 101 is set as a feature vector X represented by the following equation (6).

X = (x ₁ , x ₂ ,..., X _i ,...) (6)
Specifically, the feature vector X of the wafer 101 has the number of dimensions corresponding to the number of all the closed regions D _i , and each component x _i (i = 1, 2,...) Corresponds to each closed region D _i. Represents the defect density C _i .

  In the next step S305, as described above, the similarity between the feature vector X of the wafer to be inspected and the feature vector (representative vector described later) representing each defect distribution pattern registered in the dictionary DB 110 is calculated. Therefore, before describing the details of the processing in step S305, the registration processing in the dictionary DB 110 will be described.

  The dictionary DB 110 records distribution information based on the defect information of the wafer when the defect distribution occurs for a known defect distribution pattern. It is assumed that the defect information of the wafer when each known defect distribution occurs is prepared in advance by the person in charge of analysis. The target wafers do not have to be all wafers in which the respective defect distributions are generated, and some wafers in which typical defects are generated in relation to the defect distributions may be extracted from the production lot. When the person in charge of analysis inputs the defect information of the wafer group to the dictionary registration client 113, the dictionary registration client 113 transmits the defect information to the calculation unit 107 through the data input / output unit 106 of the defect distribution classification system 100. The unit 107 starts a dictionary registration process.

FIG. 8 shows the flow of the dictionary registration process. The processing in steps S801 to S804 is the same as the processing in steps S301 to S304 in FIG. That is, in steps S801 to S804, the arithmetic unit 107 sets the closed region D _i for the defect information of each wafer transmitted from the dictionary registration client 113 and sets the wafer feature vector X as in the process of FIG. Ask for. Next, in this dictionary registration process, after obtaining feature vectors for all transmitted wafers (step S805), the learning process described below is started (step S806), and the learning result is recorded in the dictionary DB 110 (step S806). Step S807). The dictionary DB 110 may record not only the learning result but also defect information and other information of the wafer group that is the learning source.

Any conventional method can be applied to the learning processing in step S806 as in the feature amount calculation processing. However, in the present embodiment, the learning processing is performed as follows. That is, as shown in FIG. 9, a wafer in a production lot in which a known defect distribution pattern (represented as “pattern 1”, “pattern 2”,... In FIG. 9) is generated for each defect distribution pattern. Grouped as groups 901, 902,. For each wafer group 901, 902,..., An average feature vector is obtained between the wafers belonging to that wafer group. For example, the feature vectors of the wafers belonging to the wafer group 901 are X ₁₁ = (x ₁ ¹¹ , x ₂ ¹¹ ,..., X _i ¹¹ ,...), X ₁₂ = (x ₁ ¹² , x ₂ ¹² ,..., X _i ¹² , ...), ... Then, an average feature vector X ₁ = (x ₁ ¹ , x ₂ ¹ ,..., X _i ¹ ,...) Is obtained between the wafers belonging to the wafer group 901 of the defect distribution pattern 1. Here, x _i ¹ is an average value of the i-th component among the wafers belonging to the wafer group 901, and when the number of wafers belonging to the wafer group 901 is m ₁ , x _i ¹ = (x ₁ ^{11 _{+ x 2 12 + ... + x}} i 1i + ...) it is represented as / _{m 1.} Further, the feature vectors of the wafers belonging to the wafer group 902 are X ₂₁ = (x ₁ ²¹ , x ₂ ²¹ ,..., X _i ²¹ ,...), X ₂₂ = (x ₁ ²² , x ₂ ²² ,..., X _i ²² , ...), ... Then, an average feature vector X ₂ = (x ₁ ² , x ₂ ² ,..., X _i ² ,...) Is obtained between the wafers belonging to the wafer group 902 of the defect distribution pattern 2. Here, x _i ² is the average value of the i-th component among the wafers belonging to the wafer group 902, and when the number of wafers belonging to the wafer group 902 is m ₂ , x _i ² = (x ₁ ²¹ + X ₂ ²² + ... + x _i ²ⁱ + ...) / m ₂ . In this way, average feature vectors X ₁ , X ₂ ,... Are obtained for each wafer group 901, 902,. These average feature vectors X ₁ , X ₂ ,... Are representative vectors representing known defect distribution patterns 1, 2,. These representative vectors are recorded in the dictionary DB 110 as learning results (step S807).

FIG. 10 shows the wafer to be inspected when only two closed regions D ₁ and D ₂ are set on the wafer (that is, when each vector is represented by a two-dimensional component (x ₁ , x ₂ )). A feature vector X and representative vectors X ₁ , X ₂ , X ₃ of known defect distribution patterns ₁ , ₂ , ₃ are schematically shown. Symbols ○, □, and Δ in FIG. 10 correspond to the wafers from which the defect distribution patterns 1, 2, and 3 are learned, respectively.

In step S305 of FIG. 3 (classification process), the representative vector of each known defect distribution pattern is compared with the feature vector of the inspection wafer, and the similarity r between them is calculated. As is clear from FIG. 10 shown above, the distances ν ₁ , ν ₂ between the representative vectors X ₁ , X ₂ , X _{3 of the} defect distribution patterns ₁ , ₂ , ₃ and the feature vector X of the wafer to be inspected. , Ν ₃ is closer, it can be judged that the similarity is higher. The similarity r _i between the representative vector of each defect distribution pattern and the feature vector of the wafer to be inspected can be defined by the following equation (7) from the distance ν _i between them.

ν _i = | X _i -X |
= {(X ₁ ⁱ -x ₁ ) ² + (x ₂ ⁱ -x ₂ ) ² + ... + (x _i ⁱ -x _i ) ² + ...} ^1/2
r _i = (ν _i ) ⁻¹ (7)
For all defect distribution patterns, the similarity r is calculated using equation (7). Then, the defect distribution on the inspection target wafer is classified into a defect distribution pattern that maximizes the similarity r (step S306). Such a calculation method of similarity is called a minimum distance discrimination method. Note that a threshold is set for the similarity r, and if r does not exceed the threshold for any defect distribution pattern, it may be determined that the defect distribution on the wafer does not apply to any defect distribution pattern. . The results classified by the processes in steps S301 to S306 are recorded in the inspection DB 109 together with wafer defect information. An example of the data structure of the inspection DB 109 is shown in FIG. The inspection DB includes items such as wafer ID, lot ID, processing date and time, classification result, detailed item 1, and detailed item 2. In the “classification result” column, it is recorded which defect distribution pattern the wafer (upper defect distribution) is classified into. In the example of FIG. 11, the wafer AAA-123 in the production lot AAA is classified as “pattern 1”.

  FIG. 12 explains the degree of defect affiliation with respect to each area when the rectangular areas 501 and 502 shown in FIG. 5A are set on the wafer. Since a defect 1201 of a certain “known pattern” registered in the dictionary DB 110 is located in the non-overlapping portion 601 in the rectangular area 501, the degree of belonging to the rectangular area 501 is 1. Since the defect 1202 on the “inspection wafer 1” inspected by the inspection apparatus 103 is slightly shifted from the defect 1201 and exists in the overlapping portion 603 of the rectangular area 501 and the rectangular area 502, the defect 1202 is present in both the rectangular area 501 and the rectangular area 502. Each belongs with a certain degree of membership (a value between 0 and 1). Further, the defect 1203 on the “inspection wafer 2” inspected by the inspection apparatus 103 is further shifted from the defect 1202 and exists in the non-overlapping portion 602 of the rectangular area 502, so that it does not belong to the rectangular area 501 and does not belong to the rectangular area 502. The degree of affiliation with respect to becomes 1. FIG. 13 shows the degree of influence that the defects 1201, 1202, and 1203 shown in FIG. 12 have on the feature amounts of the respective rectangular areas 501 and 502. As can be seen from FIG. 13, as the defect moves away from the rectangular area 501 in order (in the order of defects 1201, 1202, and 1203), the feature quantity of the rectangular area 501 decreases stepwise. In other words, as the defects approach the rectangular area 502 in order (in the order of defects 1201, 1202, and 1203), the feature amount of the rectangular area 502 increases stepwise. As described above, according to this defect distribution classification method, even when a positional deviation or a change in size of a defect on the inspection target wafer occurs with respect to a known defect distribution pattern that is a reference for classification, A positional shift or the like can be reflected in the feature value step by step. Therefore, the defect distribution on the wafer can be classified with high accuracy.

  In the present embodiment, the rectangular area having the same size as the closed area is set on the wafer. However, the present invention is not limited to this, and areas having different sizes may be set. Also, the shape of the closed region can be any figure such as a concentric circle, a sector, or a combination of these.

  The above is the classification process by the calculation unit 107.

  Next, a description will be given of how to display the occurrence frequency of the wafer classified into each defect distribution pattern using the result of the classification process. As shown in FIG. 1, a display client 112 serving as an occurrence frequency display unit is connected to the defect distribution classification system 100 via a network 111. The analysis person operates the display client 112 and designates the defect distribution pattern whose occurrence frequency is to be investigated and the period during which the inspection is performed. Other information such as the model name of the semiconductor device fabricated on the wafer may be designated. The condition designated by the person in charge is transmitted to the data input / output unit 106. The data input / output unit 106 that has received the condition searches the inspection DB 109 for a wafer that matches the condition, and transmits defect information of the wafer to the display client 112. As shown in FIG. 14, the display client 112 rearranges the received wafer defect information in the order of inspection date and displays the result on the display screen 1400 as a graph. In the example of FIG. 14, the frequency of occurrence of the inspection target wafer classified into “pattern 1” and “pattern 2” defined in the dictionary DB 110 is shown by a graph 1401 and a graph 1402. The horizontal axis of each graph is the inspection date and time, and the received wafers are classified in the order of inspection date and time. For example, a vertical line 1403 in the “pattern 1” graph 1401 indicates that a certain wafer is classified into the “pattern 1”. As can be seen from the graph 1401, vertical lines are densely displayed in the period 1404, and defects of “pattern 1” are concentrated in this period. Further, when viewing the graph 1402, it can be seen that the density of the straight line increases toward the right side of the graph, and the occurrence frequency tends to increase. Since the inspection DB 109 records in which lot the wafer is included, as shown in FIG. 15, the ratio of the occurrence of wafers classified into each defect distribution pattern in the lot is shown in lot units. It can also be displayed on the display screen 1400 as a graph. In the example of FIG. 15, graphs 1501 and 1502 in lot units for wafers to be inspected classified into “pattern 1” and “pattern 2” defined in the dictionary DB 110 are shown. The horizontal axis of each graph is the inspection date and time as in FIG. The vertical axis indicates how many wafers classified in the defect distribution pattern are generated in the lot. Of the 10 wafers included in a certain lot, for example, if there are 2 wafers classified as “pattern 1”, they are plotted at a position of 0.2 on the vertical axis. Therefore, the person in charge of analysis can easily recognize the abnormality occurring in the manufacturing process by looking at the display on the screen 1400.

  Since the display client 112 has defect information as numerical information or character information, it is possible to select arbitrary data from the graphs of FIGS. 14 and 15 and display it in a table format. It is also possible to draw a schematic diagram of the wafer and mount a screen showing the defect position on the schematic diagram from the position information of each defect. It is also possible to provide defect information received from the defect distribution classification system 100 to another system (not shown).

  Next, a function for warning when a specific defect distribution pattern occurs will be described. The defect distribution classification system 100 shown in FIG. 1 records identification numbers of defect distribution patterns to be monitored in a storage unit (not shown). When the classification result is calculated by the calculation unit 107, the classification result is immediately sent to the warning generation unit. The warning generation unit 108 compares the defect distribution pattern identification number recorded in the storage unit (not shown) with the defect distribution pattern identification number obtained from the classification result, and displays a match through the data input / output unit 106 if they match. A warning is sent to the client 112. When the display client 112 receives the warning, the display client 112 displays a warning message on the screen and prompts the person in charge of analysis to take a response. Therefore, the person in charge of analysis can take measures quickly.

  As described above, in the present embodiment, a plurality of rectangular areas are set so as to partially overlap each other on the wafer surface, and a weight w considering the degree of overlap is set for overlapping portions of the rectangular areas. The defect information 105 transmitted from the inspection apparatus 103 is extracted with the above weighting for each area. As a result, even if there is a defect position shift or size change with respect to each defect distribution pattern registered in the dictionary DB 110, it can be classified with high accuracy.

(Second Embodiment)
As described above, the shape of the closed region set on the wafer is not limited to the rectangular region. In the present embodiment, a plurality of the closed regions are set corresponding to the difference in defect density for each region of the known defect distribution pattern.

  For example, in the defect distribution pattern shown in FIG. 16A, it is assumed that a defect dense region 1601 exists in a curved arc shape in the lower right portion of the wafer, and the other portion is a region 1602 having a rough defect. Further, in the defect distribution pattern shown in FIG. 16B, it is assumed that a region 1603 having a dense defect in a three-pointed shape exists in the upper left portion of the wafer, and the other portion is a region 1604 having a rough defect. The regions 1601 and 1603 where the defects are dense are approximately in the range, and there are transition regions (not shown) around which the defect density gradually decreases as the defects move away from the dense regions 1601 and 1603.

  In such a case, for example, as shown in FIG. 17B, a closed region (shown by stippling) 1702 having substantially the same shape as the dense region 1601 and slightly larger is set. As shown in FIG. 17A, a closed region (shown by stippling) 1701 having substantially the same shape and size as the region 1602 having a rough defect is set so as to make a pair with the closed region 1702. As shown in FIG. 17C, a portion 1703 where the closed regions 1701 and 1702 overlap each other corresponds to a transition region between a dense defect region 1601 and a coarse defect region 1602. Further, for example, as shown in FIG. 18B, a closed region (shown by stippling) 1802 that is substantially the same shape as the dense region 1603 and slightly larger is set. As shown in FIG. 18A, a closed region (shown by stippling) 1801 having substantially the same shape and size as the region 1604 having a rough defect is set so as to form a pair with the closed region 1802. As shown in FIG. 18C, a portion 1803 where the closed regions 1801 and 1802 overlap each other corresponds to a transition region between a dense defect region 1603 and a coarse defect region 1604.

  At this time, the weight w is set according to the overlapping degree of each part so that Formula (4) may be satisfy | filled similarly to 1st Embodiment. For example, for the closed region 1701 shown in FIG. 17C, the weight w = 1 of the non-overlapping portion 1701c shown in white, the weight w = 0.5 of the overlapping portion 1703, and the weight of the non-overlapping portion 1702c in the closed region 1702 shown by cross-hatching. Set w = 0. For the closed region 1702, the weight w = 1 of the non-overlapping portion 1702c indicated by cross-hatching, the weight w = 0.5 of the overlapping portion 1703, and the weight w = 0 of the non-overlapping portion 1701c in the closed region 1701 indicated by white background Set. Similarly, for the closed region 1801 shown in FIG. 18C, the weight w = 1 of the non-overlapping portion 1801c shown in white, the weight w = 0.5 of the overlapping portion 1803, and the non-overlapping portion 1802c in the closed region 1802 shown by cross-hatching. Set the weight w = 0. For the closed region 1802, the weight w = 1 of the non-overlapping portion 1802c indicated by cross-hatching, the weight w = 0.5 of the overlapping portion 1803, and the weight w = 0 of the non-overlapping portion 1801c in the closed region 1801 indicated by white background Set.

Thus, the defect density of each region when the weight w is set can be easily obtained using the equation (5). For example, assuming that the defect densities of the closed regions 1701 and 1702 are C ₁₁ and C ₁₂ , respectively, the feature quantity x ₁ of the defect distribution pattern shown in FIG. 16A can be expressed by x ₁ = (C ₁₁ , C ₁₂ ). Similarly, if the defect densities of the closed regions 1801 and 1802 are C ₂₁ and C ₂₂ , respectively, the feature quantity x ₂ of the defect distribution pattern shown in FIG. 16B can be expressed by x ₂ = (C ₂₁ , C ₂₂ ). .

  In this way, feature amounts may be obtained for all known defect patterns, and a set formed by these feature amounts may be used as the wafer feature vector X. That is, in this case, the feature vector X of the wafer is defined by the following equation (8).

X = (C ₁₁ , C ₁₂ , C ₂₁ , C ₂₂ ,..., C _i1 , C _i2 ,...) (8)
By defining the feature vector X of the wafer with Expression (8), it can be handled in the same manner as the feature amount of the first embodiment. That is, the classification process can be performed by applying the processes of FIGS. 3 and 8 to the equation (8).

In the present embodiment, the number of regions having different defect densities is not limited to two. For example, as shown in FIG. 21A, it is assumed that there are regions 2101 and 2102 with dense defects and a region 2103 with sparse defects. At this time, the closed regions 2104, 2105, and 2106 shown in FIG. 21B are set by the method described above. These three closed regions have transition regions 2107, 2108, and 2109 shown in FIG. 21C. In FIG. 21C, a transition region 2107 is a transition region between the closed region 2104 and the closed region 2106, a transition region 2108 is a transition region between the closed region 2105 and the closed region 2106, and a transition region 2109 is a transition region between the closed regions 2104 to 2106. It is. Since the transition region 2109 is an overlapping portion of the three closed regions, w = 0.33 is set if weights are equally given to the respective closed regions.
In this defect distribution classification method, the feature quantities for a plurality of closed regions corresponding to a certain known defect distribution pattern accurately determine whether or not the defect distribution on the wafer should be classified into the known defect distribution pattern. To express. Therefore, even if the known defect distribution pattern has a complicated shape, the defect distribution on the wafer can be classified with high accuracy. Further, if the types of defect distribution patterns are limited to some extent and most of them are known in advance, the number of dimensions of the feature vector X of the wafer can be reduced. Therefore, the calculation is simplified.

(Third embodiment)
A three-layer neural network as shown in FIG. 19 can be used for the classification processing in step S305 in FIG. 3 and step S805 in FIG.

  In the first embodiment, if the number of closed regions set on the wafer is m and the number of defect distribution patterns is n, a neural network in which the number of neuron elements in the input layer = m and the number of neuron elements in the output layer = n is used. The above classification process can be performed. The number of neuron elements in the intermediate layer is arbitrary.

In step S305 and step S805, the i-th component value x _i of the feature vector X is input to the i-th neuron element of the input layer. The j-th neuron element in the output layer corresponds to the j-th defect distribution pattern. In the learning process in step S805, for all wafers registered in the dictionary DB 110, the neuron corresponding to the defect distribution pattern of that wafer. Learning is performed so that the output value of the element is 1 and the output values of the other neuron elements are 0. When such a neural network is used, parameters such as the connection weight of each neuron element are recorded in the dictionary DB 110 as a learning result of a known defect distribution pattern. In the classification process in step S305, the wafer is classified into a defect distribution pattern corresponding to a neuron element in the output layer having the largest output value for the input wafer feature value. At this time, if any output value of the output layer neuron does not reach a certain value, it may be classified as not matching any pattern.

  As described above, the processes in steps S305 and S805 can be performed using a method other than the minimum distance identification method used in the first embodiment.

となる。 When using a neural network for the second embodiment, when the number of m _i of the closed region is set on a wafer in a pattern i, the number neuron element of the input layer,

It becomes.

  When the feature values of each closed region set on the wafer are expressed using k values, the number of neuron elements in the input layer is m × k. In this case, the configuration of other parts of the neural network is the same as described above.

It is a figure which shows the structure of the defect distribution classification | category system of one Embodiment of this invention. It is a figure explaining the defect information acquired with the inspection apparatus contained in the said defect distribution classification system. It is a figure which shows the flow of the classification process by the calculating part of a defect distribution classification system. It is a figure explaining the concept of the closed area | region set on a wafer. It is a figure which shows the example which sets so that two rectangular area | regions as a closed area may mutually overlap. It is a figure which shows the example which sets so that four rectangular area | regions as a closed area may mutually overlap. It is a figure which shows the example which sets a some rectangular area so that it may mutually overlap in the substantially whole area on a wafer. It is a figure explaining the weighting of the defect in each area | region on a wafer. It is a figure explaining the weighting of the defect in each area | region on a wafer. It is a figure explaining the weighting of the defect in each area | region on a wafer. It is a figure explaining the weighting of the defect in each area | region on a wafer. It is a figure which shows the flow of the dictionary DB registration process of the said defect distribution classification system. It is a figure explaining the learning process which learns the representative vector showing a known defect pattern. It is a figure which illustrates typically the distance of the representative vector showing a known defect pattern, and the feature vector of an inspection object wafer. It is a figure which shows the data structure of test | inspection DB. It is a figure explaining the change of the feature-value by the change of a defect position. It is a figure which shows the influence degree with respect to the feature-value by the change of a defect position as a table | surface. It is a figure which shows the example of a display which monitors the generation frequency of the defect distribution classified for every known defect distribution pattern per wafer. It is a figure which shows the example of a display which monitors the generation frequency of the defect distribution classified for every known defect distribution pattern within a manufacturing lot. It is a figure which shows an example of a known defect distribution pattern. It is a figure which shows another example of a known defect distribution pattern. It is a figure which shows one closed area | region defined according to the defect distribution pattern of FIG. 16A. It is a figure which shows the closed region which makes a pair with the closed region of FIG. 17A. It is a figure which shows the state with which the closed area | region of FIG. 17A and the closed area | region of FIG. 17B have overlapped. It is a figure which shows one closed area | region defined according to the defect distribution pattern of FIG. 16B. It is a figure which shows the closed region which makes a pair with the closed region of FIG. 18A. It is a figure which shows the state with which the closed area | region of FIG. 18A and the closed area | region of FIG. 18B have overlapped. It is a figure explaining the concept of the learning process and classification process by a neural network. It is a figure explaining the problem of the prior art which classifies a defect. It is a figure explaining the problem of the prior art which classifies a defect. It is an example of a known defect distribution pattern in which a plurality of defects have a dense region. It is a figure which shows the closed area | region defined according to the defect distribution pattern of FIG. 21A. It is a figure which shows the state which the closed area | region of FIG. 21B has overlapped.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Defect distribution classification system 101 Wafer 103 Inspection apparatus 104 Network of manufacturing process 105 Defect information of inspection object wafer 106 Data input / output part 107 Operation part 108 Warning generation part 109 Inspection DB
110 Dictionary DB
111 Network between terminals 112 Information display client 113 Dictionary DB registration client

Claims (10)

Set multiple closed areas on the surface of the substrate so that they partially overlap each other,
Each of the defects on the substrate is detected by an inspection apparatus, and defect information including at least information indicating the position of each defect is obtained.
Based on the defect information, for each of the closed regions, to calculate a feature amount representing the characteristics of the defects belonging to the closed region,
A defect distribution classification method for classifying the defect distribution on the substrate by comparing a set formed by the feature amount with a set formed by a feature amount representing a known defect distribution pattern. In the defect distribution classification method according to claim 1,
The defect distribution classification method, wherein the plurality of closed regions are rectangular regions having the same dimensions as each other and arranged in a matrix on the substrate surface. In the defect distribution classification method according to claim 1,
A plurality of the closed regions are set corresponding to regions having different defect densities in the known defect distribution pattern,
The defect distribution classification method, wherein a portion where the closed region overlaps corresponds to a transition region between regions having different defect densities. In the defect distribution classification method according to claim 1,
Defect distribution classification characterized by weighting the defect information according to the degree of overlap for a portion where the closed region overlaps another closed region when obtaining the feature value for a certain closed region Method. In the defect distribution classification method according to claim 1,
Classification by the above comparison is
A feature vector having a component corresponding to each of the closed regions, each representing a feature amount of a defect belonging to the corresponding closed region, and a plurality of feature vectors each representing a known defect distribution pattern are compared. And classifying the defect distribution on the substrate into the closest one of the known defect distribution patterns. The defect distribution classification method according to claim 4,
A defect distribution classification method, wherein a sum of weights for the defect information is set to 1. In the defect distribution classification method according to claim 1,
The defect distribution classification method, wherein the feature amount for each closed region represents a defect density in the closed region. A closed region setting section for setting a plurality of closed regions on the surface of the substrate so as to partially overlap each other;
An inspection apparatus for detecting defects on the substrate and acquiring defect information including at least information indicating the position of each defect; and
Based on the defect information, for each closed region, a feature amount calculation unit that calculates a feature amount that represents the feature of the defect belonging to the closed region;
A defect distribution classification apparatus comprising: a classifying unit that classifies the defect distribution on the substrate by comparing a group formed by the feature quantity with a group formed by a feature quantity representing a known defect distribution pattern. In the defect distribution classification apparatus according to claim 8,
A defect distribution classification apparatus comprising: an occurrence frequency display unit configured to display on a screen an occurrence frequency of a substrate whose defect distribution is classified into a certain defect distribution pattern based on a result classified by the classification unit. In the defect distribution classification apparatus according to claim 8,
A defect distribution classification apparatus comprising: a warning unit that issues a warning when a substrate whose defect distribution is classified into a certain defect distribution pattern is generated based on the result of classification by the classification unit.

Images