JP2005197629A - Method and apparatus for specifying problematic process - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technology capable of easily specifying a process/apparatus in which a problem happened among the processes of manufacturing a thin-film device. <P>SOLUTION: In the apparatus 110 for specifying a problematic process among the processes of manufacturing a thin-film device, inspection information is inputted from in-line inspection apparatuses 102a, 102b, and an anomaly is automatically detected based on defect position information or visual information. If an anomaly exists, necessary information such as apparatus evaluation information 106, product inspection information 107, manufacturing history information 108 or the like is downloaded from upper-level database 105 to make an analysis for specifying the problematic process/apparatus. As such, since the anomaly is automatically detected by being triggered by the input of the in-line inspection information and the analysis is made by downloading the necessary information from the upper-level database, it is not necessary for an operator to select analysis objects, and it is possible to easily specify the problematic process/apparatus. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention is applied to a method for identifying a process (problem process) and / or an apparatus (problem apparatus) in which a problem causing a defect in a thin film device manufacturing process occurs based on inspection information obtained by the inspection apparatus. Related to effective technology.

  Thin film devices such as semiconductor wafers, liquid crystal displays, and hard disk magnetic heads are manufactured through a number of processing steps. In the manufacture of such a thin film device, pattern defect inspection or foreign matter inspection is performed every several series of steps for the purpose of improving yield and stabilization. Further, detailed observation / analysis may be performed by a review device. Based on such inspection information, an abnormality of a process or a process or an apparatus is discovered and a problem apparatus causing a defect is specified.

As a problem process identification method based on inspection information, for example, Patent Document 1 discloses a method of identifying a problem process by machine difference analysis using the number of defects. Attempts have also been made to estimate problem processes by identifying defect distribution patterns specific to equipment or process anomalies. For example, Patent Document 2 discloses a method for identifying a problem process by collating and analyzing defect distribution image data with a case database capable of estimating the problem process. Patent Document 3 describes a defect as a problem process based on a distribution state. A method for classifying the associated user-defined events is described.
JP 2000-12640 A Japanese Patent Laid-Open No. 11-45919 US Pat. No. 5,982,920

  By the way, the method according to Patent Document 1 is based on the premise that there are a large number of manufacturing apparatuses in each process. Since statistical processing is performed, inspection information on a large number of product substrates is necessary to perform a highly accurate analysis. It is. In addition, since the analysis target is selected by the worker, the effect of the analysis depends on the experience and knowledge of the worker.

  In addition, in order to identify a problem process by defect distribution pattern identification in Patent Document 2 and Patent Document 3, a library of distribution pattern information associated with the problem process is necessary. Generally, such a library is constructed. Is not easy.

  Therefore, the object of the present invention can be obtained in an inspection process with a single product substrate without depending on the experience and knowledge of the operator even when there is no accumulation of past inspection information associated with the problem process. The object is to provide a technique capable of identifying a problem process based on inspection information. Another object of the present invention is to provide a technique capable of easily accumulating inspection information associated with a problem process.

  In order to achieve the above object, the problem process identification method and apparatus of the present invention first inputs inspection information of a certain process of one product substrate, and automatically detects an abnormality based on the inspection information. When there is an abnormality, predetermined information is downloaded from the upper database, and the problem process is specified based on the product inspection information and the information loaded from the upper database.

  In particular, in the common path analysis method, product inspection information for a predetermined period of the same inspection process as the product substrate is loaded from the upper database, and the product substrate and the defect distribution state or the defect appearance similarity are high and the similarity is low. Search multiple items, load product route information corresponding to each product board and searched product inspection information from the upper database, start product boards and those with high similarity in common, and start those with low similarity A device that has not been used is extracted as a candidate device for a problem.

  According to the present invention, the abnormality detection is automatically performed in response to the input of the product inspection information, and when there is an abnormality, the information necessary for the analysis is downloaded from the upper database and the analysis of the problem process is performed. It is not necessary for the operator to select data to be analyzed, and it is not necessary to prepare a user-defined library, and it is possible to easily identify the problem process.

  In addition, since the information of the inline inspection is used, the period from the problem occurrence to the countermeasure can be shortened.

  Furthermore, according to the result output method of the present invention, product inspection information, problem device name, person in charge and status are displayed as a list, so that follow-up of investigation and countermeasures can be facilitated. From the report screen, product inspection information for which the problem device has already been identified and problem process countermeasure information can be registered, so that it is possible to easily accumulate information effective for specifying the problem process and investigation countermeasures for the next and subsequent times. .

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 1 is a diagram showing the concept of a problem process identification system according to an embodiment to which a problem process identification method of the present invention is applied, as an example of problem process identification in a semiconductor wafer manufacturing process. Reference numeral 101 denotes a semiconductor wafer manufacturing process. As described above, a semiconductor wafer is manufactured through a plurality of processing steps, and product inspection and review are performed every several series of steps. In the figure, intermediate steps are extracted and shown.

  The product wafer is inspected by the inline inspection apparatus 102a, and then inspected again by the inline inspection apparatus 102b after the processing steps by the manufacturing apparatuses 103a, 103b, and 103c. The manufacturing apparatuses 103a, 103b, and 103c are periodically evaluated for each manufacturing apparatus using the face plate inspection apparatus 104. The device evaluation information 106 thus obtained and the product inspection information 107 obtained by the in-line inspection are transferred to the upper database 105 and stored for a certain period. In addition to this, the high-order database 105 also records manufacturing route information 108 and apparatus log / maintenance information 109.

  The problem process identification device 110 extracts problem candidate devices by any method of device evaluation information collation analysis, common path analysis, machine difference analysis, device log correlation analysis, or any combination thereof. First, product inspection information output by the inline inspection apparatus 102 (102b) is input by the information acquisition unit 111. Next, the presence or absence of abnormality is automatically detected by the abnormality detection means 112 based on the inputted product inspection information. If there is an abnormality, the information acquisition unit 111 downloads information necessary for the above analysis from the upper database 105.

  Next, the similar information search unit 113 collates and searches for information similar to the abnormal examination information. Next, the problem device extraction unit 114 extracts the problem candidate process and the device based on the inspection information having the abnormality and the manufacturing route information 108 related to the searched information. Finally, the result display unit 115 displays the problem candidate device extraction result, and the result storage unit 116 creates and stores a report.

  Next, information stored in the upper database 105 will be described.

  The apparatus evaluation information 106 is information obtained by evaluation using the face plate inspection apparatus 104, and is acquired periodically and for each manufacturing apparatus. When the manufacturing apparatus has a plurality of chambers, the evaluation is performed for each chamber. Specifically, after starting the face plate wafer with the manufacturing apparatus, the face plate inspection apparatus 104 detects defects such as foreign matter and scratches, and acquires defect position information.

  In addition, when a review is performed using a review apparatus and defect classification is performed automatically or manually, defect appearance information such as a defect class can be acquired. In addition, defect analysis information may be obtained by performing detailed analysis using a detailed analysis device such as an SEM type review device or an Auger analysis device equipped with an EDS function. Therefore, a review device and a detailed analysis device may be included in the face plate inspection device 104.

  The defect position information, appearance information, composition information, and the like obtained by such inspection, review, and detailed analysis are referred to as apparatus evaluation information 106. The apparatus evaluation information 106 is stored in association with the manufacturing apparatus ID, the chamber ID, and the evaluation date / time.

  The product inspection information 107 is information obtained by inspecting by the in-line inspection apparatus 102 every several series of processing steps of the product. The in-line inspection device 102 is an optical visual inspection device, an SEM visual inspection device, a foreign matter inspection device, and the like, detects “defects” such as pattern defects, scratches, and foreign matters, and outputs position information thereof. Similar to the apparatus evaluation, a review apparatus and a detailed analysis apparatus may be included in the in-line inspection apparatus 102, and thereby, it is possible to acquire defect appearance information and composition information. The product inspection information 107 is stored in association with the ID of the wafer to be inspected and the inspection process name.

  The manufacturing route information 108 is information recorded for each product wafer or lot, and includes a manufacturing apparatus ID, a chamber ID, and a start date / time that are started in each processing step. In addition, the inspection apparatus ID and the inspection date and time inspected in the inspection process are also included.

  The apparatus log / maintenance information 109 includes time fluctuations of various processing conditions of the manufacturing apparatus, deviation amounts from the set values of the processing conditions, parts replacement of the manufacturing apparatus, and information on disassembly and cleaning. For example, in the etching apparatus, the temperature, pressure, gas flow rate, applied high frequency power and the like in the chamber are processing conditions.

  Next, the operation of the problem process identification device 110 will be described in detail.

  The problem process identification apparatus 110 first inputs the product inspection information 107 output by the inline inspection apparatus 102 by the information acquisition unit 111. This input may be performed directly from the inline inspection apparatus 102 or may be performed via the host database 105. When inputting directly from the inline inspection device 102, it is automatically transferred to a predetermined storage area. If there is unprocessed information, the next processing, that is, automatic abnormality detection processing for that information is performed. I do. In the case of passing through the upper database 105, after the inspection is completed, the product inspection information is transferred to the upper database 105, and at the same time, the wafer ID and the inspection process name are transferred to the problem process identification device 110. The problem process identification device 110 adds this to the inspection completed wafer list. One set of wafer ID and inspection process name is read from the inspection-completed wafer list. If unprocessed, product inspection information corresponding to the wafer ID and inspection process name is downloaded from the upper database 105, and automatic abnormality detection processing is performed. After that, “processed” information is added to the list.

  Next, the problem process identification device 110 automatically detects an abnormality based on the input product inspection information by the abnormality detection unit 112. Anomaly detection is performed based on defect distribution information or appearance information. First, a method for detecting an abnormality based on defect distribution information will be described. The defect distribution state is analyzed based on the defect position information in the product inspection information. The analysis of the defect distribution state is performed by a method disclosed in, for example, “Practical Pattern Detection from Distributed Defect Points on a Semiconductor Wafer”, Proceedings of MVA2002-IAPR Workshop on Machine Vision Applications, pp.10-13, Dec. 2002. Do. According to this method, defects are classified into random defects and regional defects depending on the distribution state. There are four significant shape pattern classes in the area defect, which are shown in FIG. If any of these is detected, it is determined that there is an abnormality.

  Next, a method for detecting an abnormality based on defect appearance information will be described. The defect appearance information is defect class information added by classification based on the defect image and appearance obtained by review. The classification is automatically performed using, for example, a method described in JP-A-7-201946. The presence / absence of abnormality is determined by the number of defects of a class designated in advance. The number of defects of a specified class may be obtained by a total number review, but it is generally difficult to perform a total number review when the number of defects is large. In that case, defect sampling is performed by the method described in “Appearance Inspection Method Using Defect Point Sampling Technology”, 13th Visual Inspection Automation Workshop, pp. 99-104 (December 2001), After review and defect classification, the number of defects in each defect class may be estimated. The target defect class is specified by recipe for each product type / process. It may be specified regardless of the type and process. Also, the number of defect classes that can be specified is not limited to one and may be plural.

  If it is determined by the above method that there is an abnormality, the information acquisition unit 111, the similar information search unit 113, and the problem device extraction unit 114 are used to identify the problem process using the same information as when the abnormality was detected. Hereinafter, the problem process identification method in the present invention will be described with reference to FIGS. 3 to 7 and FIGS.

  FIG. 3 is a diagram for explaining a problem process identification method based on apparatus evaluation information collation analysis. The information used is defect position information. After inputting the product inspection information, the defect distribution is analyzed by the method described above, and an abnormality is detected by detecting a significant shape pattern from the defect position information 311 (step 301). When there is a significant shape pattern, the ID of the wafer to be inspected and the inspection process name associated with the target product inspection information are read.

  Next, necessary information is downloaded from the upper database 105 (step 302). First, the manufacturing route information 108 associated with the ID of the wafer to be inspected is downloaded. The manufacturing route information 108 includes a device ID corresponding to each processing step, a chamber ID and start date / time, and a device ID and inspection date / time corresponding to each inspection step. After sorting in the order of the start date and time and the inspection date, the inspection process (102b) corresponding to the target product inspection information is searched. The previous inspection process (102a) is searched, and information on the manufacturing apparatuses 103a to 103c in the meantime is acquired.

  Next, in the case of each manufacturing apparatus and a plurality of chambers, several sets of apparatus evaluation information 106 evaluated before and after the start date and time of the wafer to be inspected are downloaded. The number of sets of device evaluation information 106 per manufacturing device is determined in advance, and is downloaded until it reaches a predetermined number or information is not found in order from the start date. Reference numerals 312a to 312c denote defect position information obtained by the face plate inspection in the apparatus evaluation information, and correspond to the manufacturing apparatuses 103a to 103c, respectively. Actually, a plurality of sets of information is downloaded to each manufacturing apparatus, but here, one set of information is shown.

  Next, all the downloaded defect position information 312 and product inspection defect position information 311 are collated (step 303), and defect position information 312b having the highest degree of similarity is selected (step 304). The matching method may be any method in which pattern matching is performed by imaging, or a feature amount representing each distribution state is calculated and a distance in the feature amount space is examined. The manufacturing apparatus 103b corresponding to the defect position information 312b having the highest similarity is set as the problem apparatus (step 305).

  FIG. 4 is a diagram for explaining a method of identifying a problem process by the same apparatus evaluation information collation analysis as described above using appearance information. This method is effective in the case where defects are randomly distributed, in which no abnormality is detected by the above method. After inputting the product inspection information, the number of defects of the designated class is acquired by the above-described method, and when the number exceeds the reference value, it is determined that “abnormality exists” (step 401). The same defect composition information 413 as the defect appearance information 411 of the representative defect image of the designated class is used in a later step.

  Next, the apparatus evaluation information 106 is downloaded by the same method as described above (step 402). However, defect appearance information 412 is used instead of the defect position information. In addition to this, the defect composition information 414 may be used, but is not essential. The defect appearance information 412a to 412c and the defect composition information 414a to 414c correspond to the manufacturing apparatuses 103a to 103c, respectively. Although not shown in the drawing, a plurality of sets of information are downloaded to each manufacturing apparatus. Since a plurality of defect images are obtained by one device evaluation, one set of information includes appearance information and composition information of a plurality of defect images.

  Next, the defect appearance information 412 and defect composition information 414 of the defect image are compared with the defect appearance information 411 and defect composition information 413 of the product inspection (step 403), and a set of information of 412b and 414b having the highest similarity is selected. (Step 404). As a matching method, there is a method based on a distance in a feature amount space by calculating a feature amount of a defect from an image. Similarly, the feature amount may be calculated from the composition information, and the similarity may be calculated based on the distance in the combined feature amount space. Alternatively, first, those containing the same element may be extracted from the composition information, and the distances in the feature amount space of the images may be compared. The manufacturing apparatus 103b corresponding to the set of the defect appearance information 412b and the defect composition information 414b having the highest similarity is set as the problem apparatus (step 405).

  Even if an abnormality occurs in the product inspection, there is a case where no abnormality appears in the apparatus evaluation information. Possible reasons are that the starting conditions at the time of device evaluation are different from the conditions for starting the product, the timing of device evaluation is poor, and the defect is not likely to occur on the face plate. In such cases, common path analysis is often effective.

  FIG. 5 is a diagram for explaining a problem process identification method by common path analysis. In the drawing, a method of identifying a problem process based on defect distribution using defect position information is shown, but defect appearance information and composition information may be used as in the apparatus evaluation information collation analysis. The abnormality detection method is the same as that in the case of the apparatus evaluation information collation analysis (step 501). When there is an abnormality, all the past product inspection information 107 for a predetermined period of the same kind and the same inspection process as the wafer to be inspected is downloaded from the upper database 105 (step 502).

  Next, the wafer having the same mode defect as the new product inspection information is searched (step 503). That is, the degree of similarity of all downloaded past product inspection information with new product inspection information is calculated. The method for calculating the similarity is the same as in the case of the apparatus evaluation information collation analysis. Then, past product inspection information whose similarity is higher than a reference value designated in advance is selected. In this example, data surrounded by a broken line and a dotted line is selected. Next, the product route information 108 corresponding to the new product inspection information and the selected past product inspection information is downloaded (step 504).

  Next, from the manufacturing route information corresponding to each product inspection information, the manufacturing apparatus started in the processing process between the previous inspection process is examined, and a common apparatus is extracted (step 505). In the drawing, an inspection wafer starter corresponding to each of product inspection information surrounded by a solid line, a broken line, and a dotted line is surrounded by a solid line, a broken line, and a dotted line. In this example, since the processing is performed by the paths of the devices A2, B2, and C2, the routes of the devices A1, B2, and C2, and the routes of the devices A3, B2, and C1, the devices that the device B2 has in common It is.

  Finally, the common device B2 is specified as the problem device (step 506). In the case where the common device is a device having a plurality of chambers, it is possible to identify the occurrence location of the problem in more detail by examining whether the chambers are also common. That is, if the chamber is common, it is generated in the chamber, and if it is not common, it is generated outside the chamber such as during transportation.

  Alternatively, past product inspection information whose similarity is lower than a predesignated reference value may be simultaneously selected and used for common path analysis. The problem device identification method in this case will be described with reference to FIG.

  FIG. 16 shows a product in which an abnormality is detected in the product inspection, surrounded by a solid line, a product having a similarity higher than the reference value by a broken line, and a product having a similarity lower than the reference value by a dotted line. Moreover, the construction apparatus corresponding to each is shown with a continuous line, a broken line, and a dotted line. Construction devices that are common to products whose similarity is higher than the reference value and products whose abnormality is detected are the devices B2 and C2, but the device C2 that has not started the product whose similarity is lower than the reference value is identified as the problem device. . That is, a device that is common to products with high similarity and is not common to products with low similarity is set as a problem candidate device.

  Another processing flow example of common path analysis will be described with reference to FIG. FIG. 11 is a diagram for explaining another problem process identification method based on the common path analysis. First, the product inspection information of the same inspection process immediately before the product inspection information in which abnormality is detected is downloaded (step 1101). Next, the similarity is calculated by the method described above (step 1102), and similar or dissimilarity is determined by comparing with a predetermined threshold value (step 1103). Next, it is checked whether or not the cumulative similarity determination number exceeds a predetermined threshold (step 1104). If it exceeds, the manufacturing route information is downloaded by the above-described method (step 1105), and the common device is extracted (step 1106). ). If it does not exceed, the product inspection information of the same inspection process one time earlier is downloaded, and the same processing is repeated.

  Still another processing flow example of the common path analysis will be described with reference to FIG. FIG. 12 is a diagram for explaining still another problem process identification method based on the common path analysis. First, the product inspection information of the same inspection process immediately before the product inspection information in which abnormality is detected is downloaded (step 1201). Next, the similarity is calculated by the method described above (step 1202), and similar or dissimilarity is determined by comparing with a predetermined threshold value. From the manufacturing route information of the products downloaded so far, devices that are common to the products determined to be similar and have not started the products determined to be dissimilar are extracted as problem candidate devices (steps 1203 and 1204). If there is one problem candidate device (step 1205), the common path analysis is terminated. If there are a plurality of products, the product inspection information of the same inspection process one time earlier is downloaded, and the same processing is repeated.

  When the common path analysis is performed using the defect appearance information, it is necessary to consider that there are a plurality of defect data for one wafer. Therefore, after searching for a defect having an appearance similar to the defect on the wafer detected as abnormal, it is determined whether the wafer has a similar defect occurrence status, that is, a wafer having the same problem. An apparatus that is common to the wafers in which the same problem has occurred and not common to the wafers in which the same problem has not occurred is extracted as a problem candidate apparatus. Hereinafter, a common path analysis method using defect appearance information will be described in detail.

  First, a method for calculating the similarity of defect appearance will be described. As a method for calculating the similarity, it is conceivable that the category of the automatic defect classification at the time of review is set to 1 regardless of the distance, and that of the different category is set to 0. In the case of different categories, it is conceivable to apply a coefficient less than 1. However, the category of automatic defect classification at the time of review does not always carry out classification suitable for problem process identification. By performing more detailed classification based on this, it can be expected to improve the accuracy of the problem device identification.

  For this purpose, an optimal feature amount space for further classifying defects in the same category is defined, and the similarity is defined so that the similarity becomes higher as the Euclidean distance in the feature amount space is closer. Keep it. Hereinafter, a method for calculating the optimum feature amount space will be described with reference to FIGS.

  FIG. 15 shows an example of a data structure suitable for performing detailed classification based on the image feature amount. The inspection target inspection process information 1501 includes a unique number, an inspection process name, the number of wafers that are reviewed and stored in the corresponding inspection process, the number of defects, and the number of defect classification categories at the time of review of the corresponding inspection process. Composed. At the same time as new wafer review information is added, the number of review wafers and the number of review defects in the corresponding review target inspection process information are updated. The category information 1502 includes a unique number, an index (inspection process number) to the review target inspection process information 1501, a category name, the number of defects reviewed and information stored in the corresponding inspection process / category, and a detailed classification recipe. Consists of the file name. The number of defects is updated as new wafer review information is added. The detailed classification recipe file is created after teaching detailed classification, which will be described later, and includes teaching data and detailed classification parameters calculated based on the teaching data. The review wafer information 1503 includes a unique number, an index to the corresponding review target inspection process information 1501 (inspection process number), a lot ID, a wafer ID, the total number of defects, and the number of review defects. One is created per process. The review defect information 1504 is information on each defect that has been reviewed, and includes a unique number, an index (review wafer number) to the review wafer information 1503, a review image name, a category name, a detailed category name, and an image feature amount. The The review image name is an image file name that is referred to when teaching detailed classification, which will be described later, and the category name indicates a defect classification category name at the time of review, and follows information output from the review apparatus. The detailed category name is information added as a result of the detailed classification teaching described later. The feature amount follows information output from the review device.

  FIG. 13 is a diagram illustrating a flow for calculating an optimum feature amount space under the above conditions. First, an inspection process for calculating the optimum feature amount space is designated by the operator (step 1301). As the designation method, it is conceivable to display a list of inspection processes based on the inspection target inspection process information 1501 and select from the list. Information on the number of review wafers, the number of review defects, and the number of categories may be displayed in the list.

  Next, the number of defects for each category is displayed based on the category information 1502 corresponding to the inspected inspection process (step 1302).

  Next, a category for calculating the optimum feature amount space is designated by the operator (step 1303).

  Next, images, image feature amounts, and other additional information are downloaded for all accumulated defect images in the inspected inspection process / category (step 1304). That is, the review wafer information 1503 of the designated inspection process, the review wafer number corresponding thereto, the review defect information 1504 of the designated category, and the image referred to by the review image name are downloaded.

  Next, a list of images is displayed for the category specified by the operator (step 1305).

  FIG. 14 shows an example of the defect image display and operation screen. There is a defect image list display window 1402 in the screen 1401, and defect images of a specified category are displayed. As for “A-01” shown in the figure, the device code of the problem process identified may be displayed together, or the distribution feature identification result may be displayed together although not shown. The status window 1409 displays the status. At this point, for example, “Teach me” is displayed.

  Next, detailed classification is taught by the operator (step 1306). The screen 1401 includes a detailed classification category display window 1403 and a defect image display window 1405 that is not a detailed classification target. The detailed classification category display window 1403 further includes defect image display windows 1404a to 1404d for each detailed classification category. In the teaching operation, for example, by drag and drop, the defect image is moved from the defect image list display window 1402 to the defect image display window 1404 for each detailed classification category or the defect image display window 1405 not for detailed classification. Detailed defect category-specific defect image display windows 1404a to 1404d represent different categories and are displayed one defect at a time. The window 1404d is in a state where no defect image is taught. The numerical values at the bottom of the window indicate the number of defects taught by the denominator and the number of defects displayed by the numerator. You can change the displayed defect by clicking the arrow at the bottom of the window.

  The window of the detailed classification category in which defects exceeding the threshold value are taught is displayed separately from other categories such as changing the background color. As the threshold value, a number necessary for calculating the optimum feature amount space is set in advance. For example, when the threshold value is 10, only the window 1404c is displayed with a different background color. When the number of detailed classification categories is insufficient, an empty window 1404 is displayed when an add button 1410 is pressed. A defect image that is determined to be excluded in the classification is moved to the detailed classification non-target defect image display window 1405. These defect images are not used for subsequent processing.

  During the teaching operation, “Teach me (changed)” is displayed in the status window 1409. If there are two or more categories and more defects than the above threshold value are taught, "Yes" is added and displayed. Whenever the state of “optimization process is possible” is selected, the optimum feature amount space for identifying the defect taught in the detailed classification category can be calculated by clicking the optimization button 1406 (step 1307). The calculation uses defects in the detailed classification category in which a number of defects equal to or greater than a threshold value is taught. As a method, any of general pattern recognition methods such as a linear discriminant method may be applied. At this time, defect images may remain in the defect image list display window 1402, and these defects are regarded as unclassified and are not used for calculation. Also, the defect of the detailed classification category in which the number of defects equal to or greater than the threshold is not taught is not used.

  Subsequently, automatic classification is performed for all defects except those in the detailed classification category for which the number of defects equal to or greater than the threshold is not taught and defects that are not classified, and the list display is changed according to the classification result (step 1308). Any known identification method may be used for automatic classification of each defect. The method of classifying into the category of taught defects with the shortest distance in the calculated feature amount space or the feature amount distribution of each defect category based on the teaching data, and the feature amount of the defect to be classified A method of classifying the category into the category having the highest occurrence probability is conceivable. Reflecting the detailed classification result, there is no defect image in the defect image list display window 1402, and the defect image in the detailed classification category in which the number of defects equal to or greater than the threshold value is taught is updated. The defect images in the other windows do not change. During the optimization calculation and automatic classification, the window 1409 displays “feature space optimization in progress”, and after completion of the automatic classification, “please confirm feature space optimization completion (first time)” is displayed. .

  Next, the mistaken defect is taught in the correct category, or re-teaching is performed by moving it out of the target (step 1309). If even one of them has been moved, the window 1409 displays “Please confirm feature space optimization completion (first time) (being changed)”. After the re-teaching is completed, the optimization calculation similar to the above is performed by clicking the optimization button 1406 (step 1310). Defects that have not been moved during re-teaching are considered to have been taught in categories classified by automatic classification.

  Subsequently, automatic classification is performed for all defects by the same method as described above, and the list display is changed according to the classification result (step 1311). After the automatic classification is completed, the window 1409 displays “Please confirm feature space optimization completion (second time)”.

  Next, confirmation by the operator is performed (step 1312). If a registration button 1407 is clicked here, a detailed classification recipe file is created, the corresponding category information 1502 is updated, and the process ends (step 1313). The detailed classification recipe file includes the number of detailed classification categories, teaching defect data for each detailed category, and a conversion vector from the original feature amount space to the optimized feature amount space. If there is a parameter required for the classifier to be used, this information is also included. Teaching may be performed again by the operator, and the same optimization process may be repeated. If it is determined that the optimization is impossible and the user wants to quit during the process, for example, when the stop button 1408 is clicked, the process ends without registering anything (step 1314). After completion, the screen 1401 is erased and the screen returns to the defect category designation.

  The optimal feature amount space calculation method has been described above. However, the data structure and the operation screen are not limited to the present embodiment, and any form that can display a list for teaching operation and confirmation may be used. . As described above, by defining the optimum feature amount space, it is possible to accurately perform detailed classification strongly related to the problem apparatus. However, there is no restriction that the optimization must be performed for the common path analysis. When the optimization is not performed, the similarity can be determined based on the distance in the normal feature amount space.

  When calculating the optimum feature amount space again for the defect image of the inspection process / defect category in which the optimum feature amount space has been registered by the above method, the feature that has been defined without performing the first teaching. Automatic classification is performed based on the distance in the quantity space, and the result is displayed. Confirmation is performed, and when the optimization button 1406 is clicked, the second optimization calculation and automatic classification are performed, and the result is displayed. Registration is performed by clicking a registration button 1407, and the process ends. Alternatively, teaching and optimization are performed again.

  In addition, the first teaching may be performed by default. For example, on the assumption that there is information on defects generated from different problem devices in the same inspection process and the same defect category, the result of performing the classification for each problem device by default is displayed in the detailed classification category display window 1403. Alternatively, the result of extracting clusters in the unoptimized feature amount space is set as the default classification. Alternatively, in the inspection process, a feature amount space defined in different same defect categories may be selected, and a cluster extraction result on the space may be used. Thereafter, teaching of addition and correction is performed by a user operation, and the first optimization calculation and automatic classification are performed. Thereafter, the procedure is the same as above.

  The above-described method can also be applied to a case where an inspection image, a defect automatic classification result, and a feature amount obtained by appearance inspection are used without performing a review. Furthermore, if the defect area information is acquired together with the image feature amount and the feature amount is recalculated, it is possible to cope with the case where the feature amount is added or the definition is changed.

  Next, a method for discriminating between a wafer in which a problem has occurred and a wafer in which no problem has occurred based on the similarity defined in the optimum feature amount space obtained above will be described with reference to FIGS. .

  FIG. 17 is a diagram illustrating a flow for identifying a problem process by common path analysis using defect appearance information. First, after inspection / review (step 1701), an abnormality is detected based on the review information using a method of detecting an abnormality based on the defect appearance information described above (step 1702).

  Next, review defects in the defect category that became the basis for determining an abnormality on the detected wafer is classified according to the detailed classification recipe obtained above, and defects in the detailed classification category with the largest number of defects are abnormal. Are selected as defects to be characterized (step 1703).

  Next, review wafer information 1503 of a product wafer in a predetermined period of the same inspection process as the detected wafer, review wafer number corresponding to the wafer, and review defect information 1504 of the above category are downloaded (step 1704).

  Next, a defect similar to the defect characterizing the abnormality selected in step 1703 is searched based on the similarity defined in the optimum feature amount space obtained above (step 1705). The past product wafer downloaded in step 1704 is divided into a group in which a problem has occurred and a group in which a problem has not occurred (step 1706) depending on the number of similar defects. An apparatus that is not common to unprocessed wafers is extracted as a problem candidate apparatus (step 1707).

  FIG. 18 is a diagram for explaining an example of a method for discriminating between a wafer in which a problem has occurred and a wafer that has not occurred on the basis of defect appearance information. According to this method, when the detailed classification category of the defect that characterizes the abnormality of the detected wafer is A, the review defect of the past product is classified in detail according to the detailed classification recipe obtained above. Count the number of defects with A for each wafer. A wafer having a large number of defects in the detailed classification category A is determined as a wafer having a problem, and a wafer having a small number of defects in the detailed classification category A is determined as a wafer having no problem. The determination of the number of defects may be in accordance with a threshold value designated in advance, or may be determined by a relative order in the downloaded wafer.

  FIG. 19 is a diagram for explaining another example of a method for discriminating a wafer having a problem from a wafer having no problem based on defect appearance information. According to this method, when the detailed classification category of the defect characterizing the abnormality of the detected wafer is A, first, the review defect of the past product is projected on the feature amount space optimum for the detailed classification obtained above. To do.

  Next, all defects in the detailed classification category A of the detected wafer are projected onto the same feature amount space, and the past product review defects projected within a predetermined distance from the projection position are counted for each wafer. To do. Alternatively, review defects of a predetermined number of past products are counted for each wafer in order from the closest one. A wafer having a large number of counted defects is identified as a wafer having a problem, and a wafer having a small number of counted defects is determined as a wafer having no problem. The determination of the number of defects may be in accordance with a threshold value specified in advance as in the above-described method, or may be determined based on the relative rank in the downloaded wafer.

  As described above, by searching for similar defects based on image feature quantities that are strongly related to the problem device, it is possible to accurately determine the wafer that has a problem and the wafer that has not occurred, and there is little data This makes it possible to identify problem devices with high reliability.

  There is also a method of performing machine difference analysis using past product inspection information that is similar to common path analysis but not similar to new product inspection information.

  FIG. 6 is a diagram for explaining a problem process identification method based on the machine difference analysis. Abnormality detection (step 601) and past product inspection information download (step 602) are the same as in the case of common path analysis. Next, the new product inspection information and the downloaded past product inspection information are digitized (step 603). When the defect position information is used, the similarity of the distribution pattern of the past product inspection information with the new product inspection information as a reference is calculated by the method described above. When the defect appearance information is used, the number of defects of the designated class is calculated by the method described above.

  Next, the manufacturing route information 108 of the wafer to be inspected is downloaded (step 604), and machine difference analysis is performed by a one-way analysis of variance (step 605). That is, the average and variance for each device are calculated for each process, and it is checked whether there is a significant difference between the devices using the overall average and variance values. Here, an example in which two apparatuses exist in each of three processing steps is shown. A1 shows the apparatus A1 of the process A, and below it shows the distribution similarity with the standard of the product started by the apparatus A1 or the average and variance of the number of defects of the designated class. In this example, there is a significant difference between the devices in the process B, and since the device B2 has a higher distribution similarity, the device B2 is identified as a problem device (step 606). Here, the average and variance are calculated for each device. However, the average and variance are calculated for each set of a plurality of devices, for example, A1 and B1, A1 and B2, A2 and B1, and A2 and B2. It is also possible to specify the combination in which.

  Common path analysis and machine difference analysis are based on the premise that there are multiple manufacturing devices in one process. However, if there is only one device or all the wafers to be inspected for downloaded information are the same route. Cases are also conceivable.

  FIG. 7 is a diagram for explaining a problem process identification method based on device log correlation analysis that is effective in such a case. Here, the operation after abnormality detection is shown. First, the past product inspection information 107 on the same path as the wafer to be inspected of the product inspection information whose abnormality is detected is downloaded and digitized in the same manner as in the case of machine difference analysis. The digitization method may be the same as in the case of machine difference analysis. However, when defect position information is used, the pattern intensity of the detected distribution pattern may be calculated. The pattern intensity is obtained by calculating the defect density inside the pattern by superimposing the defect position information and the pattern, or calculating the defect density ratio inside and outside the pattern.

  Next, the apparatus log information 109 of the apparatus that started the wafer to be inspected is downloaded. The apparatus log information is information representing a change in processing conditions during wafer processing. Here, the processing condition parameters of each apparatus are illustrated as one, but there are actually a plurality of parameters.

  Next, based on the start date in each process of the wafer to be inspected, a set of parameter values of processing conditions corresponding to the distribution similarity or pattern strength or the number of defects of the designated class is obtained and the correlation is calculated. In this example, since the correlation between the processing condition parameter of the apparatus B and the pattern intensity is strong, the apparatus B is identified as the problem apparatus.

  As described above, one problem device is specified by one type of analysis. However, even if a plurality of devices are listed in descending order of similarity as problem candidate devices, they are out of the gist of the present invention. There is no. In addition, a device identified by a plurality of methods may be used as a problem device by freely combining the above methods. Alternatively, the above methods may be performed in the specified order, and may be terminated when a problem device is found. The combination method and the comprehensive determination method can be set in advance by a recipe for each product type process.

  Next, in the problem process identification system according to the present invention, a result output method will be described with reference to FIGS.

  Since the problem process identification system of the present invention automatically performs analysis in response to the input of product inspection information, the operator does not need to be always on. However, since investigation and countermeasures are performed manually, it is necessary to notify the manager of the line QC and the person in charge of the apparatus.

  FIG. 8 shows an example of the result display of the analysis performed by the problem process identification system of the present invention. This is a list of wafers in which at least one problem candidate device is specified, and includes at least information on the wafer to be inspected and the name of the problem device. The information on the wafer to be inspected includes information that can specify the wafer to be inspected, such as the product type, lot number, inspection process name, and inspection date.

  In addition, the importance level, the person in charge, and the status information may be displayed. The importance is automatically added by the system based on product inspection information when an abnormality is detected. In other words, when anomaly is detected by significant shape distribution pattern detection, importance is added based on the number of defects in the pattern, defect density, defect density ratio inside and outside the pattern, etc., and the number of defects in the specified class exceeds the reference value. If an abnormality is detected, the importance is added based on the number of defects. However, it may be changed by an authorized user. The person in charge is the name of the person in charge of the apparatus, and in order to display this, it is necessary to have data associated with the name of the apparatus and the person in charge in advance in the upper database 105 or the system. The status indicates the status of the investigation / measure, and is either uninvestigated, investigated, no problem, completed, or registered. As described above, the number of problem candidate devices is not limited to one, and is displayed according to the priority order.

  The latest information is added to the top of this list, but it is good to be able to display it by sorting or filtering according to each item. Further, the display color may be changed depending on the status. At the same time as adding to the list, the same contents as the list item may be sent to the line QC manager and the person in charge of the apparatus by mail.

  The problem process identification system of the present invention simultaneously creates a problem process identification analysis report. The user can select from the list and display the report screen.

  FIG. 9 is an example of a report display screen. The report describes at least information 903 of the wafer to be inspected in which an abnormality has been detected and the problem candidate device name. In addition, product inspection information 901 of a defect map is created and described from the defect position information. Furthermore, if an abnormality is detected by detecting the number of defects or significant shape distribution pattern, the pattern strength, a keyword indicating the pattern shape, such as “peripheral ring”, or the specified class name if an abnormality is detected by the number of defects in the specified class Also, the number of defects in the designated class, the same defect composition information as the representative defect image in the designated class, and the like may be described.

  The importance level 902 is automatically added by the system, but may be changed by an authorized user. In that case, a change screen is displayed when the importance of the screen is clicked. The information on the problem candidate device is indispensable for the device name, but in addition, the name of the person in charge of the device, the analysis method, the information 905 from which the device is specified, and the investigation / measurement status 906 may be described. In this example, the case where the apparatus is specified by the apparatus evaluation information collation using the defect position information is shown. In this case, the defect map of the apparatus evaluation information having the highest similarity is described.

  When the defect appearance information is used, the same defect composition information as the defect image of the apparatus evaluation information having the highest similarity is described. When the apparatus is specified by the common path analysis, a defect map of the wafer used for the common path extraction or a set of defect image and composition information is described. In the case of machine difference analysis, a graph representing the average and variance of each distribution similarity or the number of defects of a specified class shown in FIG. 6 is described. In the case of the apparatus log correlation analysis, the variation graph of the processing pattern as shown in FIG. 7 is overlaid with the variation graph of the pattern intensity of the wafer used for the correlation analysis or the number of defects of the specified class.

  The investigation / countermeasure status 906 displays a survey countermeasure detail display / input screen when the detail button is clicked.

  FIG. 10 shows an example of the investigation countermeasure detail display / input screen. The survey results and countermeasures are input only by authorized users, usually device staff. When the person in charge finds out that there is a problem with the device after the investigation is completed, the person in charge marks “problem” and enters detailed information. If it turns out that there is no problem with the device, mark "no problem". By clicking the update button, the input information is confirmed and the entry date is changed. Clicking the cancel button returns to the original state. Click the back button to return to the report display screen. The status information reflects the input. At the same time, the status information of the list is updated.

  After the countermeasure is completed, click the details button in the same way to display the investigation countermeasure details display / input screen, mark the countermeasure completed, enter the countermeasure details, and click the update button. The status of the report display screen has already been corrected. When there are a plurality of problem candidate devices, information on the investigation / measure status 906 is described for each device.

  When the status is “measured”, two types of information can be registered by the problem process identification system of the present invention. One is product inspection information for which a problem device has been identified. When the registration button under the problem device name shown in FIG. 9 is clicked, the product inspection information used for abnormality detection is associated with the problem device name and registered in the user library 907. This library may be in the system or in the upper database 105. The registered information is handled in the same manner as the device evaluation information. That is, in the apparatus evaluation information collation analysis, the apparatus evaluation information is downloaded from the upper database 105, and at the same time, the user library is searched for information associated with the wafer inspecting apparatus, and if there is information, it is added to the collation analysis. .

  The other is problem process countermeasure information. This is information including the product inspection information, the information on the wafer to be inspected, the name of the problem device, the information used to identify the problem device, the details of the investigation result, and the details of the countermeasures shown in FIG. The storage location may be either in the system or in the upper database 105. The registered data can be referenced by inputting the device name and performing a search. In addition, when the problem device name shown in FIG. 9 is clicked, problem process specifying information related to the device may be referred to. By this method, it is possible to shorten the time for investigation and countermeasures.

  In the above embodiment, the problem process identification system is independent from the upper database, and the embodiment in which the data is downloaded and analyzed has been described. However, the function of this system is incorporated in the upper database. Is included in the scope of the present invention.

It is a figure showing the concept of the problem process identification system of embodiment which applied the problem process identification method of this invention. In embodiment of this invention, it is a figure showing the example of a significant shape distribution pattern. In embodiment of this invention, it is a figure for demonstrating the problem process identification method by the apparatus evaluation information collation analysis based on defect distribution. In embodiment of this invention, it is a figure for demonstrating the problem process identification method by the apparatus evaluation information collation analysis based on a defect external appearance. In embodiment of this invention, it is a figure for demonstrating the problem process identification method by a common path | route analysis. In embodiment of this invention, it is a figure for demonstrating the problem process identification method by machine difference analysis. In embodiment of this invention, it is a figure for demonstrating the problem process identification method by an apparatus log correlation analysis. In embodiment of this invention, it is a figure showing the example of a problem apparatus specific information list. In embodiment of this invention, it is a figure showing the example of a problem apparatus specific report screen. In embodiment of this invention, it is a figure showing the example of a survey countermeasure detail display and input screen. In embodiment of this invention, it is a figure for demonstrating another problem process identification method by a common path | route analysis. In embodiment of this invention, it is a figure for demonstrating another problem process identification method by a common path | route analysis. In an embodiment of the invention, it is a figure showing the processing flow of the optimal feature-value space calculation for carrying out detailed classification of a defective image. In an embodiment of the invention, it is a figure showing an example of an operation screen for teaching detailed classification of a defective image. In an embodiment of the present invention, it is a figure showing a suitable data structure for carrying out detailed classification of a defect image. In embodiment of this invention, it is a figure showing the product by which abnormality was detected by product inspection, the product whose similarity is higher than a reference value, and the product whose similarity is lower than a reference value. In an embodiment of the invention, it is a figure showing a processing flow of problem process specification by common path analysis using appearance information of a defect. In an embodiment of the present invention, it is a figure explaining a method of discriminating a wafer with which a problem has occurred, and a wafer which has not occurred based on appearance information of a defect. In an embodiment of the present invention, it is a figure explaining another method of discriminating a wafer with which a problem has occurred, and a wafer which has not occurred based on appearance information of a defect.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 101 ... Semiconductor wafer manufacturing process, 102 (102a, 102b) ... In-line inspection apparatus, 103 (103a, 103b, 103c) ... Manufacturing apparatus, 104 ... Face plate inspection apparatus, 105 ... Host database, 106 ... Apparatus evaluation information, 107 ... Product Inspection information 108 ... Manufacturing route information 109 ... Equipment log / maintenance information 110 ... Problem process identification device 111 ... Information acquisition means 112 ... Abnormality detection means 113 ... Similar information search means 114 ... Problem device extraction means, 115 ... Result display means, 116 ... Result storage means, 311 ... Defect position information of product inspection information, 312 (312a, 312b, 312c) ... Defect position information of apparatus evaluation information, 411 ... Defect appearance information of product inspection information, 412 (412a, 412b, 412c) ... defect appearance information of apparatus evaluation information, 413 ... product Defect composition information in inspection information, 414 (414a, 414b, 414c) ... Defect composition information in apparatus evaluation information, 901 ... Product inspection information in which abnormality is detected, 902 ... Importance, 903 ... Information on inspected wafer, 904 ... Problem candidate Device information, 905... Information based on device identification, 906... Investigation / measurement status, 907... User library, 908. Problem process countermeasure information, 1102 (1102 a, 1102 b) ... Workmanship evaluation device, 1107. , 1401 ... screen, 1402 ... defect image list display window, 1403 ... detailed classification category display window, 1404 (1404a, 1404b, 1404c, 1404d) ... defect image display window by detailed classification category, 1405 ... non-detail classification target defect image display Window, 1406 ... optimization button, 407 ... registration button, 1408 ... stop button, 1409 ... status window, 1410 ... Add button, 1501 ... This review is the inspection process information, 1502 ... category information, 1503 ... review wafer information, 1504 ... review defect information.

Claims (15)

In a manufacturing process of a thin film device manufactured through a plurality of processes, a method for identifying a problem process causing a defect based on a product inspection result performed in the manufacturing process,
Load product inspection information including defect distribution information or appearance information obtained in the inspection process of one product substrate,
Anomaly detection is automatically performed based on the defect distribution information or appearance information,
If there is an abnormality, load the specified information from the upper database,
Based on the product inspection information and the predetermined information, the problem process is identified by common path analysis,
In the common path analysis method, product inspection information for a predetermined period of the same inspection process as the product substrate is loaded from the upper database, and a defect distribution state with the product substrate from the product inspection information for the predetermined period Alternatively, based on the similarity of the defect appearance, a plurality of substrates that have the same problem as the product substrate and a substrate that does not have the same problem as the product substrate are respectively searched, and the product substrate and the retrieved product inspection The manufacturing route information corresponding to each information is loaded from the upper database, the product substrate and the substrate in which the problem has occurred are started in common, and the device that has not started the substrate in which the problem has not occurred is a problem A method for identifying a problem process, which is a method for extracting as a device candidate. In the problem process identification method of Claim 1,
The method of searching a plurality of substrates having the same problem as the product substrate and a substrate not having the same problem as the product substrate from the product inspection information of the predetermined period is the predetermined period. A problem process specifying method, wherein the product inspection information is sequentially searched one by one in time until the cumulative number of substrates having the problem exceeds a predetermined threshold value. In the problem process identification method of Claim 1,
The method of searching a plurality of substrates having the same problem as the product substrate and a substrate not having the same problem as the product substrate from the product inspection information of the predetermined period is the predetermined period. The problem inspection method is characterized in that the product inspection information is sequentially retrieved one by one in time, and the problem device candidates to be extracted are searched until they can be narrowed down to one. In the problem process identification method of Claim 1,
The abnormality detection method uses the appearance information of the defect, and determines that there is an abnormality when the number of defects in any defect category exceeds a predetermined reference value,
The common path analysis is performed based on appearance information including a defect category and an image feature amount obtained by inspection of the product substrate,
In order to perform the detailed classification, the similarity of the appearance of the defect is calculated using the image feature amount based on teaching to the detailed classification category of defects in the same category and in the same inspection process and the same defect category in advance. The higher the distance between data in the optimal feature space, the higher it is calculated,
The substrate having the same problem as the product substrate and the substrate not having the same problem as the product substrate are determined based on the number of defects having a high appearance similarity with the product substrate. Characteristic process identification method characterized. The problem process identification method according to claim 4,
The method for teaching the detailed classification category of the defect includes displaying a list of defects in the same inspection process and the same defect category designated by the operator, teaching the detailed classification category by the operator, and performing the detailed classification based on the teaching. An optimum feature amount space for identifying the category is calculated, and the defects in the same inspection process and the same defect category are classified based on the distance between the data in the feature amount space, and the results are displayed in a list. A problem process identification method characterized in that teaching to a detailed classification category is performed again. The problem process identification method according to claim 4,
Based on the number of defects having a high degree of similarity in appearance with the product substrate, a method of discriminating a substrate that has the same problem as the product substrate and a substrate that does not have the same problem as the product substrate,
In the abnormality detection, the defect of the product substrate classified into the defect category in which the number of defects exceeds a predetermined reference value is classified in detail, and the detailed classification category having the largest number of defects is selected as the detailed classification category characterizing the abnormality.
Count the number of defects classified into the detailed classification category for each substrate,
A substrate having the same number of counted defects as a substrate having the same problem as the product substrate, and a substrate having a small number of counted defects having the same problem as the product substrate. A problem process identification method characterized by the above. The problem process identification method according to claim 4,
Based on the number of defects having a high degree of similarity in appearance with the product substrate, a method of discriminating a substrate that has the same problem as the product substrate and a substrate that does not have the same problem as the product substrate,
In the abnormality detection, the defect of the product substrate classified into the defect category in which the number of defects exceeds a predetermined reference value is classified in detail, and the defect classified in the detailed classification category having the largest number of defects is defined as a defect characterizing the abnormality. Selected,
Counting the selected defects and defects within a predetermined number within the predetermined distance or the closest distance in the optimum feature amount space for performing the detailed classification for each substrate,
A substrate having the same number of counted defects as a substrate having the same problem as the product substrate, and a substrate having a small number of counted defects having the same problem as the product substrate. A problem process identification method characterized by the above. In the problem process identification method of Claim 1,
The abnormality detection method uses the defect distribution information, analyzes a defect distribution state to detect a significant shape pattern, and determines that there is an abnormality when a significant shape pattern exists, and identifies a problem process Method. In the problem process identification method of Claim 1,
The common path analysis method, device evaluation information collation analysis, machine difference analysis, device log correlation analysis, or any combination of them is used to identify the problem process,
The apparatus evaluation information collation analysis method loads apparatus evaluation information related to the product board start apparatus from the higher-level database, collates the product inspection information with each apparatus evaluation information, and sets an apparatus having a high degree of similarity as a problem apparatus. A problem process identification method characterized by being a method. The problem process identification method according to claim 9,
The machine difference analysis method loads a plurality of past product inspection information of the same type and the same inspection process as the product substrate from the upper database, and defect distribution state or defect appearance information of the plurality of past product inspection information. Process / apparatus in which a significant difference appears by loading each manufacturing route information of the plurality of past product inspection information from the upper database and performing analysis of variance for each process based on the digitized information. A method for identifying a problem process, characterized in that the method is a problem device. The problem process identification method according to claim 9,
The apparatus log correlation analysis method loads product inspection information of a product substrate processed through a common path with the product substrate from the upper database and device log information of the product substrate starter, and each product inspection information Identifies the problem process characterized by digitizing the defect distribution status or defect appearance information based on the above, calculating the correlation with the processing condition data of each starter, and using the starter with a strong correlation as the target device Method. The problem process identification method according to any one of claims 1 to 11,
In addition, a list including at least information on the inspected product in which an abnormality has been detected and one or more problem candidate device names is displayed.
A problem process identification method comprising creating and storing a report including at least information on a product to be inspected in which an abnormality has been detected and one or more problem candidate device names. In the problem process identification method of Claim 12,
In addition, register the product inspection information used for abnormality detection in association with the problem device name,
A problem process identification method comprising: registering the product inspection information, the information on the inspected product, the problem device name, the information used to identify the problem device, the details of the investigation result, and the details of the countermeasure contents . It is a problem process identification device for a manufacturing process of a thin film device manufactured through a plurality of processes,
Information acquisition means for acquiring information directly or indirectly from the host database and the inspection device, the review device, and the manufacturing device;
An anomaly detection means for automatically detecting an anomaly based on inspection information;
Similar information search means for collating and searching for information similar to the inspection information detected abnormally,
Problem device extraction means for extracting problem candidate processes and devices based on manufacturing path information related to the detected inspection information and the searched inspection information;
A result display means for displaying at least the inspection information in which an abnormality is detected and the problem candidate device name;
A problem process specifying apparatus comprising: a result storing means for generating and storing a report including at least an abnormality detected inspection information and a problem candidate apparatus name. The problem process identification device according to claim 14,
Furthermore, similar information teaching means for teaching detailed classification into inspection information groups similar to each other by an operator,
A problem process specifying apparatus, comprising: a feature quantity optimizing unit for calculating a feature quantity space optimal for identifying the detailed classification taught above.

Images