TWI230349B - Method and apparatus for analyzing manufacturing data - Google Patents

Abstract

A method for data mining information obtained in an integrated circuit fabrication factory (""fab"") that includes steps of: (a) gathering data from the fab from one or more of systems, tools, and databases that produce data in the fab or collect data from the fab; (b) formatting the data and storing the formatted data in a source database; (c) extracting portions of the data for use in data mining in accordance with a user specified configuration file; (d) data mining the extracted portions of data in response to a user specified analysis configuration file; (e) storing results of data mining in a results database; and (f) providing access to the results.

Description

1230349 A7 _ B7 V. Description of the Invention () One or more specific embodiments of the present invention relate to a method and equipment for analyzing information generated by a factory, which may refer to (but not limited to) integrated circuits (&Quot; 1C ") manufacturing or assembly (,, semiconductor assembly, or, fab ") factory, etc. Figure 1 shows the production or manufacturing (" semiconductor ") of the integrated circuit (,, ic) in accordance with the prior art The basic structure of the yield analysis tool of the manufacturing plant, or, "manufacturing plant" factory. As shown in Figure 1, the photomask factory 1000 generates a photomask 1010. As further shown in Figure 1, the manufacturing process works Tracking system 1020 ("WIP tracking system 1020 ") When it is used in a manufacturing plant, it is used to trace wafers on wafers or substrates (the terms of wafers and substrates are used interchangeably to represent various semiconductor crystals (Including but not limited to glass substrates) when manufacturing (and testing) integrated circuits. WIP tracking system 1020 wafer tracking system uses (but is not limited to) implant tool 1030, diffusion, oxidation, deposition tool 1040, chemical mechanical leveling tool 1050 (, Cmp tool 105). ,,), resistance slope coating tool 1060 (such as, but not limited to, covering photoresist tool), step feed tool 1070, configuration tool 108, etching / cleaning tool 1090, laser test tool 1 100 'Parameter test tool i (1), wafer sorting tool 112, and final test tool 1 130. Most of these tools are used in manufacturing plants to produce integrated circuits, but the above list is only an example (others can be added.) As shown further in Figure 1, the manufacturing plant includes many tools for obtaining tools. Page 4 This paper size applies Chinese National Standard (CNS) A4 specification (210X297 mm) (please read the precautions on the back before filling this page). Thread 1 Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs Consumer Cooperatives 1230349 A7 B7. Printed by the Consumers' Cooperative of the Property Bureau V. Description of the invention () Layer measurement values and systems used to automate various processes. For example (but not limited to) shown in Figure 1, the tool layer measurement and automation system includes the tool layer measurement and automation system. Tool database 1210 for automatic operations, such as process tool management (such as process method management), and tool sensor measurement data collection and analysis. For example (but not limited to) a PC server 123 used as a demonstration to download process methods Data to the tool (through the process method module 1233), and the tool sensor (via the sensor module 1235) receives the tool sensor measurement data, the system Method data and tool sensor measurement data are stored in, for example, tool database 1210. As further shown in Figure 1, the manufacturing plant includes many process measurement tools, such as defect measurement tools 1260 and 1261, and mask defect measurement tools 1265. Defect Measurement Tool 1267, Defect Inspection Tool 1270 (" DRT 1 2 7 0), CD 'Zeri Tool 1 2 8 0 (" Key Dimension Measurement Tool 1 2 8',), and Voltage Contrast Measurement Tool 129. Among them, the process measurement tool is driven by the process evaluation tool 1300. As further shown in Figure 1, application-specific analysis tools drive certain process measurement tools. For example, the defect management tool 1310 analyzes the data generated by the defect measurement tools 1260 and 1261. Mask analysis tool 1 320 analyzes the data generated by the mask defect measurement tool 1265; coverage analysis tool 1330 analyzes the data generated by the coverage defect measurement tool 1267; cd analysis tool 1340 analyzes the data generated by the CD measurement tool 1280, and test Body tool 1350 analysis by laser test tool "100", parametric variable test tool 1 1 1 0, Ba circle classification tool 1 1 2 〇 and the final test tool i 丨 3 〇. Page 5 This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm) (Please read the precautions on the back before filling this page)

A7 B7 Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 1230349 V. Description of Invention () The figure further shows that the database tracking / correlation tool obtains data from one or more application-specific analysis tools via a communication network. For example, the statistical analysis tool 1400 obtains data from, for example, the defect management tool 1310, the CD analysis tool 1340, and the test body tool 1350, and then stores them in the relational database 1410. Finally, the yield management method should be stored on the data in the data extraction database 1420, which is extracted from the communication network routing wip tracking system 1020 and the tool database 1210. Yield management systems used in manufacturing plants in previous techniques have encountered many problems. Figure 2 shows a prior art process used in a manufacturing plant, which was monitored using the End-Of-line referred to herein. The finish line monitoring system uses a "tracking indicator" feedback loop. In Figure 2, tracking indicators such as (but not limited to) low yield, poor quality, and slow device speed can be identified at block 2000. Then, in box 2010, the "bad product" metric (that is, the measurement value associated with the wafer product that generated the tracking index) will be compared with the original specification of the metric. If the metric is "out of specification", The process continues until 2 30 'where an action will be taken against " out of specification " and then feedback will be provided to the process control engineer to correct the " out of specification ". On the other hand, if the metric The value system is "in the specifications", and the process continues to 2020, where the factory's past failures will be analyzed. If it is determined that the problem has occurred, the process continues to 2040, otherwise (ie, no previous experience) the process continues to 2050. At block 2040! , Acting on a product or tool recommendation about previous experience, and then providing feedback to the process control engineer to take the same action as before. As shown in box 2050, page 6 This paper size applies the Chinese National Standard (CNS) A4 specification (210 × 297 mm) (Please read the precautions on the back before filling this page)

1230349 A7 B7 V. Description of the invention () Perform a failure correlation analysis on the old data of the tool or device. If correlation is obtained, the process continues to 2060, otherwise (no correlation) the process continues to 2070. At block 2600, the "bad" tool or device is "corrected" and then provided to the process control engineer. At block 2070, a maintenance job will be performed. There are several related issues with the aforementioned finish line monitoring process. For example: (a) low yield is usually caused by several problems; (b) the "specification" limit is usually set based on the results of undefined theory; (c) the knowledge of previous production failure history is usually not written 'even if written It has not been widely distributed; (d) the data and the right to use the data are scattered; and (e) a working hypothesis must be generated before performing a correlation analysis, and the number of correlations is very large, and the resources used for correlation analysis can be used However, it is very limited. For example, the typical data feedback and problem correction engineering process usually requires the following steps: (a) define the problem (usually takes about one day); (b) select key analysis variables such as percentage of yield, percentage of defect And so on (usually takes about one day); (c) forming a hypothesis about the selected key analysis variable anomaly (usually takes about one day); (d) using various "instinct sense" methods to Assumed classification (usually takes about i days); (e) Develop experimental strategies and experimental test plans (usually takes about one day); (f) Perform testing and collection Data (usually takes about 5 days); (g) Adapt the model (usually takes about 1 day); (h) Diagnose the model (usually take about 1 day) 丨; ( 〇 Interpret the model (usually takes about 1 day); and ⑴ Perform recognition tests to prove that it has improved (usually takes about 20 days), and if there is no improvement, start from page (c) The paper size applies the Chinese National Standard (CNS) A4 specification (210X 297 mm) (Please read the precautions on the back before filling out this page). Order printed by the Intellectual Property Bureau Staff Consumer Cooperative of the Ministry of Economic Affairs A7 B7 1230349 V. Description of the invention ( The next experiment usually involves five repetitions. As a result, the typical problem renewal time is about 7 months. As the line width shrinks and new technologies and materials are used to make integrated circuits (such as steel metallization) , And new low-dielectric-constant films), it is important to reduce the increase in defect rate (due to process or ingenious dyeing). The time required to find the root cause is the key to overcome the defects. These problems are transferred to 300 mm (8-leaf) crystals. It is more difficult to overcome when round Therefore, inability to maintain yield becomes a major difficulty when doing many things at the same time. In addition to the above problems, the problems also occur in semiconductor manufacturing plants that have to spend a lot of money on defect detection equipment and defect data management software to try their best to monitor defects And continuously reduce the defect density. Current prior art technologies in defect data management software require the development of multiple solutions: defect trends (eg, paretos analysis by defect type and size); (b ) Wafer layer defect to yield diagram; and (C) Destruction ratios by type and size on a special and artificial basis. For each of these solutions, the main disadvantage is that users must have prior knowledge of what they want to draw . However, due to the wide range of data, the likelihood of users moving to the root cause is low. In addition, even if a graph can be generated for each variable, for such a large number of graphs, it is practically impossible for the user to analyze all of them. In addition to the above problems, a further problem is that the data used in semiconductor manufacturers is mostly "indirect measurement data". The term "indirect measurement data" in this article refers to the data collected by indirect measurement! The indirect measurement system is assumed to be related to the manufacturing process of the manufacturer in a predictive manner. For example, the metal wire is patterned in the product. After the circuit is installed, the critical dimensions may be used. Page 8 This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X 297 mm) (Please read the precautions on the back before filling out this page). Printed by the Intellectual Property Bureau Staff Consumer Cooperative 1230349 A7

Order

1230349 A7 B7 V. Description of the invention () Incorporated into the analysis table of a specific integrated circuit data measurement value, because the integrated circuit data measurement value is a single discontinuous measurement value. It is not easy to combine " (Read the notes on the back before filling this page.) Processing tool time basic data and discontinuous data measurements limit the use of processing tool time basic data as a method to optimize plant efficiency. In addition to the problems mentioned above, the use of relational databases to store data generated in the manufacturing plant is also a source of problems. For example, relational databases such as (but not limited to) RACLE and SQL feeders are organizations and It is generated by referring to the needs of defined or specified relationships between data components. In use, users of relational data technology (such as programmers) provide a summary of how one data component is related to another 70 pieces of data. Once the database is Implantation, the program user will query the information contained in the database according to the pre-built relationship. This ^ 'as this relationship data When the library was trapped in the manufacturing plant, the relationship between the previous technology and the shell database had two inherent problems. The first problem was that the user (such as a programmer) was generating a clear summary of the data that needed to be modeled (ie, the relationship and data). Database table) must have an in-depth knowledge of the data. This profile is particularly safe to maintain the control of the relationship of data components. The software that puts data in the database and the program that takes data from the database_. The software must use the database The second problem is that although the relational database is used to retrieve small data transactions (such as financial transactions, air reservations and others) with excellent TPS (ie transaction processing specifications) evaluation, Its performance in generating i data and supporting decision-making support systems (such as data warehousing and data mining) for improving the production yield of manufacturing plants is inadequate. &Amp; With the problems identified above, further problems should be used Previous page 10 This paper size applies to the Chinese country i ^ s) A4 ^ 2_i〇x297; -------- 1230349 、 Invention description () Data analysis calculations in technology The results of this method are used in the semiconductor process industry to quantify production yield problems. This algorithm includes linear regression analysis and human: mining using decision tree data. These algorithms encounter two basic problems in a given data set almost often There are more than one problem that affects yield; however, the best use of these algorithms is to find a "solution", a non-specific system & factory quantification-group shadow # yield dispersion problem = These algorithms cannot be fully automated " delivery, analysis; that is, linear regression points) need to manually prepare and define variable categories before analysis, and the decision tree mining requirements-"human users" are defined in the analysis — The target variable / analysis itself defines various parameters. In addition to the problems mentioned above, the problems arising from further development are data mining and other large data groups. For example, according to previous techniques, it is obvious that data mining only uses a certain degree of domain knowledge (that is, about one: those domain representatives " interested " information) to filter the data set and analyze the size of the variables in the data. And the number ... reducing the data set that will reduce Fenshikui ~~ M in one room, collect the surgery / model through the specialized value system (ie, the definition of importance), and then guess will drive the analysis Systematic ”To make this methodology effective, tools are usually manually configured. The evaluation result is operated by the person who will: evaluate-result. These people are usually the bearers of the process being evaluated, because this is the data needed to collect the data and form the appropriate questions, Jt I λ, and gather the expertise of the data set (more precisely, for special processes Data collection and establishment of satellite associations to annoy this 5 = need to lead to 1 T >-experts in the cluster, lack of working time, and obtain results between different processes, due to the inconsistency of the helmet and the data mining Most of the procedures require human intervention to drive the specifications on page 11 ;: public love) 6 Please read the notes on the back before filling out this page)-Order · Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 1230349 V. Description of the invention () In the end, even if it succeeds, most of the "gains" have been lost or reduced. For example, the time-consuming manual processing of data and analysis is very expensive in terms of labor and equipment, and if results are not obtained early, there will not be enough time to make changes. In addition to the issues identified above, further issues arise as follows. An important part of the yield improvement and factory efficiency improvement monitoring efforts focuses on the correlation between the finish line functional test data, online parameter data, online measurement data, and specific process tools used to manufacture integrated circuits. In realizing this correlation, it is necessary to determine the relationship between a specific "data number block" and the block associated with all factory processing tool data (where the processing tool data table is a category attribute). The definition of good correlation is processing A specific niche of the instrument (ie category) data has a category within the block that is associated with the value of a selected number block (called the strain number or "DV"). This analysis is aimed at Identifies a category (such as a factory process tool) that may cause undesired Dv readings and removes it from the manufacturing process until the engineer can be sure that the process tool is operating correctly. Given in the semiconductor manufacturing database In a large number of tools and "types of tools" category materials, it is difficult to use manual spreadsheet search techniques (known as "common research") to separate processing tools such as one-selling. Despite this limitation, there are techniques in the semiconductor industry to detect poor processing tools or types of process data. This can be done, for example, by performing a product commonality analysis. Of course, this technique f seeks a specific place, two sleeves ^-said < knowledge beforehand ' and it is quite time consuming if the user is not aware of the nature of the failure. Another technique uses advanced ‘data mining algorithms, such as gods: the web or decision trees. These techniques may be effective, but on page 12 of the data mining needs, the paper size is suitable for the national standard (CNS) A4 specification of the rich country (2 297 public reply (please read the precautions on the back before filling this page)-mouth line! Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs

V. Description of the invention (1230349 Extensive field expertise makes it difficult to establish. In addition, these data mining algorithms are given slowly because this general data analysis technique requires a lot of calculation costs. Using the above analysis techniques, users are trying to It takes more time for basic or complex analysis to identify a problem than to actually correct it after discovering the bad processing tool. Finally, in addition to the above problems, further problems arise as follows. Data mining algorithms such as neural networks, rules Inductive search and decision trees usually have a more satisfactory method than general linear statistics to find relevance in large data sets. However, when using these algorithms to analyze large-scale data in low-cost hardware (such as Windows 2000 servers) Data sets will have some restrictions. The main concerns in these restrictions are the random use right memory (RAM) and the expansion of the central processing unit (CPU) load required for this technology. Usually a large semiconductor manufacturing data set ( For example, neural network analysis such as greater than 40 million characters) will continue to exceed After several hours, it may even violate the Windows 2000 operating system's 2G ram limit. In addition, the rule summary or decision tree analysis of this large data set does not necessarily violate the RAM limit of a single window program, but It may still take several hours before the analysis is completed. It is required to be able to solve one or more of the above problems in this technique. Bright and clear purpose and summary: One or more specific embodiments of the present invention are conducive to meeting the requirements of this technique. In particular, a specific implementation of the present invention is a method for information obtained in a data mining integrated circuit manufacturing plant (" manufacturing plant "), including the Chinese standard (CNS) A4 on page 13 of this paper. Specifications (21〇χ 297 mm) (Please read the notes on the back before filling in this page) Order · Line 1 Member of the Intellectual Property Bureau of the Ministry of Economy Η Printed by Consumer Cooperatives 1230349 A7 B7 5. Description of the invention () The following steps: ⑷ gather Data from one or more systems, tools, and databases in the manufacturing facility that generate data or ... materials within the manufacturing facility; ㈨ format the data and format it The data is stored in the source database; ⑷ extracting a part of the data according to the user-specific profile to extract the data; 资料 extracting the data from the extracted part of the data in response to the user-specific analysis A configuration file; ⑷ storing the results of this data mining in a results database; and (f) providing a simple description of the use of the results: Figure 1 shows the integrated circuit production or manufacturing plant (a " "Semiconductor assembly plant" or "assembly plant") the basic structure of the yield analysis tool; Figure 2 shows the 'a previous technology process used in the assembly plant, the better part of this previous process refers to the finish line monitoring; Figure 3 shows a manufacturing data analysis system 'manufactured according to one or more specific embodiments of the present invention, and an automatic data flow from raw unformatted input data to data mining results, the results of which are applied to use one or more of the present invention Integrated circuit manufacturing process of a specific embodiment; FIG. 4 shows one or more specific embodiments of the present invention for forming an unstructured data item to enter the intelligent The logic data flow of the intelligence base method; Figure 5 shows the original time base data (example, and especially the graph of the beam current versus time of the process tool); Figure 6 shows how to convert the 5th The original time basic information in the picture is divided into pieces. Page 14 This paper size applies Chinese National Standard (CNS) A4 specification (21 × 297 mm) (Please read the precautions on the back before filling this page) Order. Department of Economics Printed by the Intellectual Property Bureau Employee Cooperative Cooperative 1230349 A7 B7 V. Description of the invention () paragraph; Figure 7 shows how the original time basic data is related to fragment 1 in Figure 6; Figure 8 shows within fragment 7 on BIN-S Reliability of γ range; Figure 9 shows the data mining operation of a 3-layer branch; Figure 10 shows the distribution queue implemented by a DataBrainCmdCeter according to one or more specific embodiments of the present invention; Figure 11 shows according to this The user edits and configures the analysis template user interface part of the slot interface unit set manufactured by one or more embodiments of the invention; FIG. 12 shows one or more specific embodiments according to the present invention. The analysis template part of the manufacturing profile is shown in the example; Figure 13 shows a super pyramid box structure '; Figure 14 shows a super pyramid box with emphasis on one of the layers; Figure 15 shows the extraction from the second layer of the super pyramid box The superblock layer (self-organizing map) of the superblock; Figure 16 shows a self-organizing chart with high, low, and middle regions, where each n cluster and low cluster are labeled for subsequent automated map matching analysis; Figure 17 shows a unit highlighting a super square; Figure 18 shows the number defined by a numerical distribution, virtual, and category, Figure 19 shows the calculation of the number of falling differences (falling differences (number = all falling differences (not in any circle)) Sum) and diameter score ^ diameter score = average diameter of Dv of the three circles), where Dv category is divided according to the Dv number page 15 (please read the precautions on the back before filling this page), tr. Line one Printed by A7 B7 1230349, Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the Invention () Distribution; Figure 20 shows the calculation of the main scores for a given IV, considering three factors: the size of the difference, the diameter score The size and number of times the IV appears on the DV scoring list for the series; Figure 21 shows an example of a subset of the data matrix entered into the DataBrain module; Figure 22 shows a number (BIN) versus category ( Example of tool ID); Figure 23 shows a scoring threshold for three tools; Figure 24 shows a defect data file from a defect inspection tool or defect review tool in the manufacturing plant; Figure 25 shows a data compilation algorithm An example of the resulting data matrix; and Figure 26 shows a typical output from the DefectBrain module. (Please read the notes on the back before filling this page) Printed on page 16 by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs, drawing number comparison instructions: 1000 photomask factory 1020 WIP tracking 1 040 diffusion, oxidation, deposition tools 1060 resistance coating Tool 1080 Deployment tool 110 0 Laser test tool 1120 Wafer sorting tool 1210 Tool library 1 2 3 3 Process recipe module 1010 Mask 1030 Implant tool 1050 * CMP tool 1070 Step feed tool 1090 Etching / cleaning tool 11 1 0 Variable test tool 11 3 0 I Final test tool 1230 PC server 1235 Sensor module This paper size is applicable to China National Standard (CNS) A4 specifications (21 × 297 mm) 1230349 A7 B7 Employees of Intellectual Property Bureau, Ministry of Economic Affairs Printed by a consumer cooperative (please read the notes on the back before filling this page) Detailed description of the invention of the 4050 file system: One or more specific implementations of the present invention can be used to improve yield, by providing one or more of the following : (A) The data input of the body circuit (" 1C ") manufacturing plant ("semiconductor manufacturing plant" or "· manufacturing plant"), that is, by establishing multiple formats Description (1260) Defect measurement tool 1265 Mask defect measurement tool 1270 Defect inspection tool 1290 Voltage contrast measurement tool 1 3 1 0 Defect management tool 1330 Coverage analysis tool 1 3 5 0 Test body tool 1 4 1 0 Relation database 3000 data Analysis system 3020 data conversion module 3040 main loader module 3 05 5 ^ and touch interface module 3070 network manager module 3090 results database 3110 graphics and analysis engine 4000 manufacturer data warehouse 4020 data parser 4040 Data Loader 4060 Relation Database 1261 Defect Measurement Tool 1267 Coverage Defect Measurement Tool 1280 CD Measurement Tool 13 00 Process Evaluation Tool 1320 Mask Analysis Tool 1340 CD Analysis Tool 1 4 0 0 Statistical Analysis Tool 1 4 2 0 Extract Database 3010 ASP data conversion module 3030 adaptive database 3050 wisdom library 3060 master builder module 3080 data think tank engine module 3100 network visualization module 3 1 20 web server database 4010 unformatted data stream 4 0 3 0 Formatted data stream page 17 This paper size applies to China National Standard (CNS) A4 specifications (210X297 mm) 1230349

V. Description of the invention () Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs and the Consumer Cooperatives; It is stored in, for example (but not limited to) an Oracle system; (c) decision analysis data input and rapid output of multiple data sets for analysis; (d) independent analysis automation and use "data value system π automatically issues; e) multiple data mining techniques, such as, but not limited to, neural network-like, rule induction, and multivariate statistics; (f) visualization tools, with the diversity of f0 statistics to quantify the results obtained; and (g) an application The " ASP " is used by end-to-end network delivery systems to provide rapid configuration. Using one or more specific embodiments of the present invention, a typical data feedback and problem remediation engineering process usually requires the following steps: (a) define the problem (usually occurs in about 0 days); monitor all key analysis variables, Such as the percentage of yield, the percentage of defects, etc. (usually occurs in about 0 days), (c) forming a hypothesis about the abnormality of all key analysis variables (usually occurs in about 0 days); (d) using statistics Reliability level and key instruction provided by hypothetical rating (ie, based on experience in a profile), which indicates how to assign false designs or evaluate hypotheses, including, for example, but not limited to, the weighting of certain artificial intelligence rules-please pay attention to The correction key for category data, such as tool data, is different from the correction key used for digital data, such as detection data, which usually takes about i days); (Develop experimental strategies and experimental test days (usually occurs about i day); 时间 the time it takes to perform the test is about ... ⑷ adapt the model (usually distributed = p: about 1 day); (h) diagnose the model (usually occurs Time is about working days); (i) money: explain the model (generally occurs "time is j1 large), and (j) perform recognition tests to prove improvement (generally occurs about 20 days) without the need for Repeat. Page 18 Please read the intentions on the reverse page and fill in this page) # · · 线 一 1230349 A7 V. Description of the invention (), Ό Fruit, the typical time to correct the problem is about one and a half ( 1.5) month. Figure 3 shows a factory data analysis system 3ηΛΛ that is not manufactured according to one or more specific embodiments of the present invention (please read the precautions on the back before filling this page) • 3000 and the original unformatted The data automatic process of inputting data to the result of data mining is applied to the integrated circuit manufacturing process using the present invention or one or more specific embodiments. According to the second specific embodiment of the present invention, the tortoise is yours. By automating the steps of the knife analysis process and the process from one stage of the analysis to the other, it will enable manual data mining to a process and data mining. Disadvantages in the process are greatly reduced In addition, according to one or more specific embodiments of the present invention, it can provide users or customers with the right to use when setting up data analysis, and can use commercially available and installed interfaces such as the Internet. Browser Lookup:--Application Service Provisioning (" ASP, :) System Distribution Method (ie, a method known to those skilled in the art-item-based data conversion method) is to do this- The preferred method of the network browser interface. According to one or more specific embodiments of the manufacturer data analysis system 3000 shown in FIG. 3, the data used for collection and analysis can be from one or more manufacturers. A company located in one location, or data that can be used for collection and analysis is one from each company's one or more manufacturing plant locations. In addition, one or more employees ’cooperatives of the Intellectual Property Bureau of the Ministry of Economy printed this In a specific embodiment, the user who sets and / or checks the results may be different users or customers from different departments of the same company, or different users or customers from different departments of different companies or Households, according to the data with the data preservation requirements and methods to account management! To isolate. According to one or more embodiments of the present invention: (a) Sister Sister You a tribute is automatically retrieved, processed, and formatted so that the data mining tool can be used < use, (b)-price 19 The paper size of this page applies to the Chinese National Standard (CNS) A4 specification (210X297 public love) A7 B7 1230349 V. Description of the invention () The value system has been used and the problem has automatically occurred, so that the data mining tool will respond to the relevant results; and (c) The results will be delivered automatically and can be used remotely so that relevant actions can be taken quickly based on the results. As shown in Figure 3, the ASP data conversion module 3010 is a data aggregation process or module that obtains different types of data from at least one of many different types of data sources in the manufacturing plant, such as (but not limited to): (A) product equipment historical data from MES ("Management Execution System"); (b) data from a device interface data source; (c) processing tool recipes and processing tool test programs from the data source provided by the manufacturer; and (D) Original equipment data from sources provided by Manufacturing Wei, such as (but not limited to) probe test data, E-test (electrical test) data, defect measurement data, remote diagnostic data collection, and post-processing data. According to one or more specific embodiments of the present invention, the Asp data conversion module 3010 accepts and / or collects data transmitted in the format specified by the customer- and / or tool, such as (but not limited to) storing the original data from the tool The customer data collection database (centralized or otherwise) that outputs data, and / or directly from the data source. Further, this information can be received or collected on a scheduled basis or on demand. Furthermore, data can be encrypted, and FTR files can be transmitted over a security network (such as security e-mail) like a customer's intranet. According to one or more specific embodiments of the present invention, the A ”data conversion module 3010 is a software application program executed on a PC server, and is based on various methods known to those skilled in the art, C + + , Perl and Visual Basic. [For a practical example] X cases are generally available, including: (a) WIP (process work) resources, U negative information, usually including about each batch (a batch of wafers usually Refers to 25 pieces that are moved in a crystal box during processing. Page 20 (please read the precautions on the back before filling this page). Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 1230349

V. Description of the invention () (Please read the notes on the back before filling this page) Yangyuan) 12,000 items (WIp information is usually used by process engineers), (b) Equipment interface information, such as Original processing tool data for batches of 12,000 items. Please note that in the past equipment, the interface information is usually not used by anyone; (c) system scale information, which usually includes 1,000 items per batch (system level scales) Information is usually used by process engineers); (defect information, usually including 1,000 items per batch (defect information is usually used by yield engineers); (e) E_test (electrical test) information, usually including 10,000 items per batch (E-test information is usually used by the device engineer); and (f) classification (with / data records and bitmap) information, which usually includes 2,000 items per batch (classification information is usually The right to use by the production engineer). The average person should understand that these data can reach about 136,000 unique measurements per wafer. · 'The Intellectual Property Bureau of the Ministry of Economic Affairs' employee consumption cooperation Printed as shown in Figure 3, the data conversion module 3002 converts and / or compiles the original data received by the ASP data conversion module 3010 into any of the various methods known to those skilled in the art. The data format of key / stop / data, and then the conversion data is stored in the adaptive database 3 030. The data conversion process implemented by the data conversion module 3020 needs to classify the data; merge processing such as (but not limited to) ) Manufacturing-test batch ID conversion (for example, this is used for boundary), wafer ID conversion (such as detection and cut ID) and wafer / photomask / mould coordinate normalization and conversion (such as but not limited to In the normalization of the coordinates, a notch or a wafer reference measurement is used; and the data specifications such as (but not limited to) tests and Bin detection data (for example, some end-line detection tests may have 10 to 10) 〇 failure modes), the specification limits of metrological data, etc., and the calculated data types, page 21 This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm) 1230349 A7 B7 V. Description Ming () such as (but not - limited to) the limit, the batch, the wafer level and regional information according to the present invention.

One or more specific embodiments, the data conversion module · 3 020 is a software application program executed on a PC + g + + device, and according to the person skilled in the art,

Any one of the methods is compiled with Oracle Dynamic PL-SQL & Perl. Extracting Secret Horses According to one or more specific embodiments of the present invention, the data conversion processing by the data conversion module 3 009 杳 杳 Gong Shi needs to insert more f7 according to the knowledge of the person skilled in the art. A general-purpose compiler that converts the original data file into an "Industrial file format of" Gu Ba M 1 and Brother Knife Format "(that is, the data format is ,, generalized ,, to Qin Ai @ Α 致 不 & How many formats of the input data are to be converted, only a few recorded k-V ,,, ° According to one or more specific embodiments of the present invention, converted $ 妙 · # 田 案 maintains the "layer" in the original data Information (which enables subsequent processes to reach from low: data " arrival, high-g data) and not include industry-specific information. Once the raw material is entered in this format, it will be entered into the adaptive database 3300 for storage. According to According to one or more specific embodiments of the present invention, a general file format for inputting data is defined by using the following hierarchical structure: Aid, Where? When? What? And numerical values. For example, for a semiconductor Manufacturing plant, which is clearly defined as follows: " Product ID "is identified by one or more of: batch size, wafer size, trench ①, mask ID, mold 10, and sub-mold Kelvin coordinates, where is it?" Is derived from: process flow / assembly One or more of the line manufacturing steps and sub-steps are identified; " when? &Quot; is identified by date, one or more of the measurement times; ,, what? ", Such as by the name of the quantity (such as but not limited to Yield), measured value type / category, and one or more of the discriminated edges in the wafer classification. The value? Is defined as, for example (but not limited to) the yield: 5 i 4%. Use this (please read the precautions on the back first) (Fill in this page again) Order 2 Printed by the Consumers Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs

1230349

V. Description of the invention (specific embodiment 'any plant data can be expressed. According to the invention + JL> ^,, β one or more specific embodiments, the data conversion module 3 02 0 is generally compelling (please (Please read the notes on the back before filling this page) X. Snail translation of the new type of data collected by the ASP data conversion module 3 〇 丨 〇 4 # FREE 丨 β points, ⑺ 疋 ': the shell material conversion module 3020 will Generate an "over-the-air" database "hands-on" so that this new type of data is stored in the adaptive database 3030 by, for example, generating a miscellaneous method for data usage rights ^ § Miscellaneous programs. Finally, according to this Invention $ _々 #,, or multiple specific embodiments, when the data reaches the manufacturer's data analysis system 3GG (), the data will be stored in the adaptive database 3030. According to one or more specific embodiments of the present invention, The ASP data conversion module, and 301 includes a module that aggregates processing tool sensor data from the SmartSysTM database (The SmartSysTM program is programmed by Applied Materials and can be collected, analyzed, and stored, for example, from Processing by the manufacturer Tool sensor data). In addition, the data conversion module 3020 includes a module that converts SmartSysTM processing tool sensor data into data prepared by the main loader module 3040 and the main builder module 3060. The data group used for data mining. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. According to one or more specific embodiments of the present invention, a data compilation algorithm can use the time basis of the individual processing (ie factory or assembly line) tools. Data to establish a "direct" link between the measurement data and existing non-optimized manufacturing plants (ie, waste conditions). An important part of this data compilation algorithm is compiled in the (factory) Or assembly line) The time-based operating condition data generated in the processing tool becomes the key (integrated circuit specific statistics') and can then be analyzed by the DataBrain data think tank engine module 3 080 in accordance with the following detailed description to Provide automatic data mining error Page 23 &quot; — ....... —......... This paper size applies to China National Standard (CNS) A4 specification (210X297 (%) ------- 1230349 Δ7 Α7 ___ Β7 V. Description of the invention () Analysis of false debt measurement. According to one or more specific embodiments of the present invention, the following steps will be implemented to compile the time-based processing tool data: {Please read the notes on the back before filling this page) a. Generate a configuration file (using the user interface detailed below), the specific digitized fineness is used for general time basic data format; and b · Compile the time-based processing tool data retrieved by the ASP data conversion module 3010, and change from various (such as, but not limited to) ASCII data and file formats to a general time-based data slot format for the profile. A specific example of the definition of the format of a general time base data file is shown below. According to these specific embodiments, it is advantageous for a file to be considered &quot; valuable &quot; without the need for all data ranges to be complete. Conversely, as described below, certain data ranges can be filled with "post-processing" data communication routes with the semiconductor, MES host, and later implanted. <Beginning of Header> [Product ID Code] [ Batch ID code] [Parent batch ID code] [Wafer ID code] [Groove ID code] Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs [WIP module] [WIP submodule] [WIP submodule Steps: ί [Tracking input date] [Tracking output date] Page 24 This paper size applies Chinese National Standard (CNS) A4 specifications (210X297 public copy) 1230349 A7 B7 V. Description of the invention () [Processing tool ID] [Using the Process tool recipe] <End of header> <Data start> (Parameter start) [English name of parameter] [Parameter ID number] [Data collection start time] [Data collection end time] Time increment, data value 1 time increment, Data value 2 Time increment, data value 3 (Please read the precautions on the back before filling this page) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs <Parameter end> <Parameter start> [Parameter English name] [Parameter ID number [Data collection start time] [Data End time of the set] Time increment, data value 1 time increment, data value 2 time increment, data value 3 page 25 This paper size applies the Chinese National Standard (CNS) A4 specification (210X 297 mm) 1230349 A7 B7 5 Explanation of the invention () <End of data> According to this specific embodiment, the items indicated in italics are needed to properly incorporate the file content into the integrated circuit data metrics. As described above, according to one or more of the present invention In a specific embodiment, a configuration file for compiling time-based data clearly states the range and which set of time-based negative materials will represent wafer statistics. According to one such specific embodiment, a setting slot may also contain some information about the setting The file will process information about those time-based raw data formats, as well as options for compressing one or more of the original buildings. An example of a specific embodiment of a profile. <Header Start> [Extensions Applied to This Profile File name] [Original data compressed file &lt; Y or N>] [Generate image compressed file &lt; number of files / parameters &gt;] [Image compressed file resolution] <End of header> &lt; Analysis The beginning of the head> [Overall graphic statistics &lt; On / Off>, N segment] [X-axis time statistics &lt; On / Off &gt;, N segment] [Z axis parameter statistics &lt; On / Off &gt;, .N [Snippet] The following explains the above-mentioned settings &quot; file parameters. Extension file name: In the configuration file, the bank lists the extension file name and / or the name of the page. This paper size applies the Chinese National Standard (CNS) A4 specification (210X297). Dong) (Please read the notes on the back before filling out this page) Order · Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs

1230349 A7 — _ B7 V. Description of the invention () Key words that indicate that a given original, general time basic data file will be compiled using the parameters defined in a given configuration file β Original data compression file: the original in the configuration file Row 'indicates that if a compressed copy of the original negative material must be maintained-using this option will cause the trough case to be compressed and stored in the reach directory structure. Generate image compression file: This line in the configuration file indicates that if the data in the original time base data file must be drawn in the standard x_y format, so that the "raw" view of the data can be stored and quickly retrieved without compression, and interactive Map raw data files (these files can be large and can add up to 10-20 billion bytes per month for a single processing tool). The number of image options allows most snapshots of different key areas of the x-y data map to be stored, so that a "zoomed in" view of the data is also available. Image Compression Slot Resolution: This line in the setting slot defines the standard image compression level that will be applied to any X-y graphics captured by the Generate Image Compression option. Overall chart statistics: This line in the settings slot indicates that the system will generate overall statistics for all file formats processed by the profile in the problem. The following describes how these statistics are generated. X-axis time chart statistics: This line in the configuration file indicates that the system will generate statistical data defined by the X axis as the time range for all file formats processed by the setting slot in the problem. The following describes how these statistics are generated. ! Percentage data graph statistics: This line in setting k indicates that the system will generate percentage statistics for all file formats processed by the profile in the question. Page 27 This paper size applies Chinese National Standard (CNS) A4 specifications (210X297 mm) ) (Please read the precautions on the back before filling out this page} -Order · Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs

V. Description of the invention (each piece printed by the Consumer Property Agency of the Intellectual Property Bureau of the Ministry of Economic Affairs printed 1230349 materials' The following will describe how these statistics are generated. Ten "4 how are generated. The following will describe these according to one or more of the present invention. , (Also known as X-axis time graph system 1 watts of statistical data helmet 1 plus) is a parameter-based parameter for the mother-time-based data map and the product is oblique ^ ^ for each given time base Group and a given parameter, the data is divided into many pieces according to U v ^, the meaning of the slot 1. The X-axis time graphic piece ^ The edge Λ means to take the entire width of the X-axis (from the small X value to the largest The vdu, v phase η, and 4 will divide the X-axis range into N identical increments 1. Generate statistics for each segment and understand how it works' First refer to Figure 5, which shows, Examples of materials, especially process tools ...: Shows the original time base information, he has a chart of the function of the beam current as a function of time. Figure 6 shows how to put the fragments, and Figure 7: 'Time The basic information is divided into 6 pieces of pictures 丨. Chu Bei枓 How to relate to the following typical segment statistics (example is N segment 10 groups of statistics) ·· 1 · within the segment · area 2 · the average value of the gamma axis of the data in the segment · 3 · the data in the segment Standard deviation of the γ-axis value 4 · Slope of the fragment 5 · Minimum value of the Y-axis of the fragment 6 · Maximum value of the Y-axis of the month (7 · Percentage change between the average value of the Y-axis and the previous fragment 8 · The average value of the Y-axis and the Percentage change of one snippet page 28 (please read the precautions on the back before filling in the clothing page)

1230349 at Β7 V. Explanation of the invention (9 · Y-axis standard deviation value and the percentage change of the previous segment 10 · Y-axis standard deviation value and the percentage change of the next segment Figure 8 shows the trust of the Y range in the segment 7 BIN-S Degree &amp; Example. Using the above information, the process engineer can adjust the processing tool configuration, summer mode (processing tool setting) to obtain a lower range of BIN-S failures. Statistics are always based on one of the inventions or In specific embodiments, the following 2 data are volume statistics calculated from data that has not been cleaned up by Tukey's data: 1. The total area under the curve 2. The number of changes in the Y-axis slope by 10% or more toward 3 · 95% data width of the X axis (ie, 95% of the data is retrieved from the center of the data to the right and left) printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 4. 95% of the X axis data width is the average value of the γ axis 5_ 95% X-axis data width γ-axis standard deviation 6.95 ° / 〇 X-axis data width γ-axis range 7 · Area under the X-axis 95% curve 8 · X-axis leftmost 2.5% data width 9 · X-axis maximum Area under the curve of 2.5% on the left Data width 11 · The area under the 2.5% curve on the far right of the X axis 1 2 · 90% data width of the X axis (ie, the data center is opened to retrieve 90% of the data to the left) ί 13 · 90% of the X axis data Γ-axis average value of the width 14 · 90% γ-axis standard deviation of the width of the X-axis data Page 29 This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 Gongchu) 1230349 Α7 Β7 V. Description of the invention () 15 · 90 / 〇X-axis data width of the γ-axis range (please read the notes on the back before filling this page) 16 · Area under the 90% curve of the X-axis 1 7 · The data width of the leftmost 5% of the X-axis 1 8 · Area under the curve of the leftmost 5% of the X-axis 19 · Data width of the rightmost 5% of the X-axis 20 · Area of the data under the rightmost 5% of the X-axis 21 · Area 75% of the data of the X-axis (ie, starting from the center of the data) Get 75% of the data to the right and left) 22 · 75% average value of the γ axis of the X-axis data width 23 · 75% standard deviation of the γ-axis data width γ-axis 24 · 75% γ-axis data width of the γ axis Scope The area under the curve of 75% of the X-axis printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 26. The leftmost side of the X-axis is 12_5 ° /. Data width 27 · X Area under the 12.5% curve on the far left side of the axis 28. Data width of I2 · 5% on the far right side of the X axis 29. Area under 12.5% curve on the far right side of the X axis Although the percentages used in the above specific embodiments are general percentages such as 90 , 95, 75, etc., other percentage embodiments can also be used. Among them, the percentage can be modified to a value located therein, for example, the ‘Essence’ part of the data is intentionally adjusted to make it somewhat wider. In a further specific embodiment, the overall statistical data similar to the above is calculated by clearing 5000% Duke ’s data, and in another specific embodiment, the overall statistical data similar to the above is 500% Duke ’s Calculated by clearing data. Page 30 This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm) 1230349 A7 A7 ___ _ B7 V. Description of the invention () According to one or more specific embodiments of the present invention, the percentage data statistics 2 are as above The 10 groups of statistical materials listed in the x-axis time graphic statistics. The difference between percentage data and x-axis time statistics lies in the way the film is rich and meaningful. In the x-axis time statistics, the clip is based on N identical parts on the x-axis. However, for percentage statistics, the width of the segment changes on the X axis, because the moment is defined by the percentage of data contained in the segment. For example, if the percentage data is divided by 10 segments &quot;, the first segment will be the first 10% of the data (the leftmost 10% of data points with reference to the x-axis). As further shown in Figure 3, the main loader module 3040 (which can be triggered by time-generated events or negative material arrival events) retrieves formatted data (such as data file 3035) from the adaptive database 30; then It is converted into a wisdom bank of 3050. According to one or more specific embodiments of the present invention, the wisdom database 3050 includes a database of Oracle relationships known to those skilled in the art. According to one or more specific embodiments of the present invention, when data is gradually imported from the manufacturer, when entering, the main loader module 3040 tests the directory of the adaptive database 3030 and determines that a sufficient amount of data has arrived It can be retrieved and transferred to the wisdom library 305 (). According to one or more specific embodiments of the present invention, the main loader module 3040 and the wisdom library 3050 include at least for management, reference and extraction of a large number of Method and equipment for unstructured and related data. According to one or more specific embodiments of the present invention, the wisdom database 3050 is a hybrid database, which includes at least a wisdom database relation database component and a wisdom database file system. The component is based on a specific embodiment of this type. 'The relational database component uses a split-index algorithm to generate a right-of-use key for separation and storage in a distribution file. Page 31 This paper applies Chinese National Standard (CNS) A4 specifications ( 210X 297 mm) (Please read the notes on the back before filling out this page)-Order · Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 1230349 V. Description of the invention () Information in the case library Advantageously, this can quickly convert unstructured raw data into a formal structure, so it can bypass the limitations of commercially available database products, and has the advantage of quickly storing structured files on disks and arrays. According to the invention In one or more specific embodiments, the first step of setting the wisdom bank 3050 involves defining an applicable layer of discontinuous measurement data that may be present. However, according to one or more specific embodiments of the present invention, it will not In order to start building a wisdom database 3050 for the discontinuous data, the number of layers of a given discontinuous data needs to be considered in advance. In contrast, only a new layer and a certain number of points in the wisdom database 3 05 need to be defined The relationship between the previous layers (sub-layers or super-layers to understand this relationship 'will consider the following example. In a manufacturing plant, a common layer may be a collection of wafers; this can be compiled as the i-th in the wisdom library 3050 Second, each specific wafer in the wafer set can be compiled into layer 2. Second, any subgroup of any wafer on the wafer can be programmed into layer 3 (or a majority layer, depending on Consistency of this subgroup category Advantageously, the flexibility of the wisdom database 3050 enables any given data type to be stored in the wisdom database 3050, as long as its nature can be indexed to the lowest level of the range that can be applied to that data type. Intellectual Property of the Ministry of Economic Affairs Printed by the bureau / consumer cooperative (please read the notes on the back before filling this page). Advantageously, according to one or more specific embodiments of the present invention, the loading process of the data of the wisdom bank 3050 is more than that of the traditional correlation database. The data loading process should be easy. As for the wisdom database 3050, each new data type must be targeted to the special layer ID to indicate the discontinuity between the manufacturing layer and the measured data (or the old data). The format of the relationship is heavy. For example, in the manufacturing industry, the information given to the industry must be rewritten to include lines such as "Layer 丨 ID", "Layer 2" quotation ID, etc., and then the measured values are for the set of wafer combinations. Be documented. This Wisdom Page 32

1230349 A7 _ B7 V. Description of the invention (The nature of the printed wisdom library 3050 printed by the Intellectual Property Bureau's Consumer Cooperatives of the Ministry of Economics can be loaded with any applicable information without the need to define a summary of a specific correlation database. Advantageously, according to this One or more of the inventions, in specific embodiments, the wisdom library 3050 is designed to output large data sets to support automated data analysis tasks (described below), using a hash-link algorithm to quickly accumulate and link large amounts of data. The design of outputting large data sets of related data sets usually requires a large number of "table_links" in the database, (accumulation and output data). As is known, using the related data table forms and links will cause this kind of The program of the large data set is highly concentrated on the CPU, and it is advantageous that it is different from the case of using a "hash-link" algorithm to output a large data set from the wisdom bank 3050. Figure 4 illustrates one or more of the present invention. Various embodiments are used to form an unstructured data event into a logical data streamline of the method of entering the wisdom bank 3050. As shown in FIG. 4, manufacturing The factory data is retrieved in box 40. Data Warehouse 4000. This manufacturer data can be any of many different shapes, or originate from any of many different sources, including (but not limited to) old data in the database There are data and real-time data from process equipment of monitoring equipment (such as unlimited sensors)] Secondly, the unstructured material will be input into the data parser 4020. It should be understood that the data obtained by the manufacturer's warehouse from 4000 The method and frequency do not affect the performance of the data profiler 4020, the data carrier 4040, or the wisdom library 3050. Second, the data profiler 40 outputs a formatted data stream 4030. The format acceptable by the data loader 4040 (this is only a format issue, and does not include any &quot; knowledge &quot; about the data, i.e., only how the data appears). Page 33 This paper applies Chinese national standards ( CNS) A4 specification (210X297 mm) Data loss (please read the precautions on the back before filling this page) Order · Line-1230349 A7 B7 V. Description of the invention () (Please read the precautions on the back before filling this page)Secondly, the database loader 4040 reads the formatted data stream 4030. The database loader 4040 uses a hash-index algorithm to generate index keys between the data components and their positions in the slot system 4050 (eg, as in Not limited to, the hash-index algorithm uses the data layer ID of the data element to generate the index key.) The data is then stored in the file system 4050 for subsequent reference and use, while referring to the hash-index key of the file system 4050 It is stored in the relational database 4 0 60. According to one or more specific embodiments of the present invention, the wisdom database 3050 is generated by loading data into tables divided and indexed by each layer in the Oracle 9i data center. -, Order · Line · The Intellectual Property Bureau of the Ministry of Economic Affairs' Employee Consumption Cooperative printed a review of Figure 3, the main builder module 3060 entered the wisdom library 3050 and used a profile (using user edits and profiles Interface module 305 $ is generated) to establish the data structure, which is used as the input of the data mining process (explained below). The user edit and profile interface module 3055 enables the user to generate a profile of the configuration data used by the master builder 3060. For example, the master builder 3060 obtains the data specified by the configuration file from the wisdom database 3050 (such as, but not limited to, a special type of data within a special range of parameter values), and then combines another from the wisdom database 3〇. Data specific to the profile obtained in 50 (for example, without limitation, another special type of data within another special range of parameter values). To achieve this, the wisdom database system components of the wisdom database 3050 are deduced from the wisdom database relationship database components of the wisdom database 3050, so that different levels of data can be quickly merged into one. Information vector cache records & amp vector cache of informati〇n) &quot;, which is converted into data for data mining. Profiles allow users to use hash-index to define new relationships, thus generating new "information vector cache records" information, which will be converted to information. Page 34

Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs

1230349 f The method described below is used for data mining (the data is referred to herein as "superblock"). According to one or more specific embodiments of the present invention, the main builder module 3 060 is software executed on the pC servo server. Application written in Oracle Dynamic PL-SQL and peri based on any one of many methods known to those skilled in the art. In operation, the main builder module 3 06 () uses the configuration file to retrieve and / or extract the superblock definition. Second, the master builder module 3060 uses a hyperblock definition to generate the -information vector cache record definition. Secondly, the master builder module, and 3060 generates the information vector cache a recorded according to the definition of the information vector cache record, by (a) using the hash_index key from the wisdom library relationship of the wisdom library 3050 The database component retrieves a list of files or data components identified or specified by the information vector cache record definition; (b) retrieves the archive file from the Qinhuiku file system component of the smart library 3050; and (c) uses the The data components identified within the information vector cache record definition are populated with the information vector cache record. Secondly, the main builder module 3600 uses the definition of the superblock to generate the superblock from the information vector cache record in the following description manner, where the superblock is designated when it is passed by a manufacturer data analysis system 3000. An ID is used to identify the analysis result and can be used by the customer to review the analysis result. The main builder module 3060 includes sub-modules that create superblocks, clean up the superblocks &quot; shell materials to remove data that will adversely affect data mining results according to any of a number of methods known to those skilled in the art, Link hyperblocks to make the analysis of many different variables feasible, and edit (translated Bin and parameter data into a method used for data mining (such as, but not limited to, conversion events, driving data into stored data). Page 35 This paper Standards are applicable to China National Standard (CNS) A4 specifications (210 &gt; &lt; 297) (please read the precautions on the back before filling this page). Order and line_ 1230349 A7 B7 V. Description of the invention () According to one or more of the invention In a specific embodiment, the main builder module 3 060 includes a shell material, a month processor, or a server (for example, 0 + +, etc.), which is manufactured according to any one of a variety of methods known to those skilled in the art: ), Wherein the data cleaning can be performed according to the standard proposed in the setting slot, or according to a special benchmark when receiving user input. According to one or more specific embodiments of the present invention, the master builder module 3060 Including outputting the spreadsheet to one of the modules in various file formats, such as (but not limited to) SAS (a database tool known to those skilled in the art), .jmP (JUMP diagrams are based on those skilled in the art Known to visualize and analyze xy data), χΐ3 (a Microsoft Excel spreadsheet known to one skilled in the art), and txt (a text file format known to one skilled in the art). According to the present invention One or more productive embodiments, the main builder 3 060 includes a module that receives a user-generated hyperblock as input, and transfers the information vector cache record to the data think tank engine module 3808 for use. According to one or more specific embodiments of the present invention, the data conversion module 3 020, the main loader module 3040, and the main builder 3060 are executed to provide continuous update adaptive database 3030, wisdom library 3050, and The output data is used for data mining. As further shown in Figure 3, the network management module 3070 transfers the output data from the master builder module 3060 to the data think tank engine module 3080 for analysis. Once established by the master 3060. Normalized data sets can be used for> data mining, an automated data mining process analyzes the data set according to the analysis template-it is expected to maximize or minimize the variables according to the relevant analysis template, and the paper size on page 36 applies China National Standard (CNS) A4 Specification (210X 297 mm) (Please read the precautions on the back before filling this page)-Order · Line · Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs Consumer Cooperatives 1230349 A7 B7 V. Description of the invention ( ) (Please read the precautions on the back before filling this page) The relative importance of these variables can still be considered. The data think tank engine module 3 080 includes user edit and profile interface module 3055, which contains an analysis configuration The setting and template builder module provides a user interface to create user definitions, configuration parameter values, and data. An automated mining case is used for the data think tank engine drive module 3080. Thus, the data think tank engine module 3008 implements the automated data mining process by using a combination of the statistical properties of the variables and the relative attributes of the variables in a self-learning neural network. The statistical distribution and the size of an "important" variable (by an analysis template) or, for example (but not limited to) the effect of a variable on the structure of a self-organizing neural network diagram ("SOM"), automatically generates A criterion on which a related problem is formed, and then presented to a data mining algorithm that is best suited to handle the given particular type of data set. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. According to one or more specific embodiments of the present invention, the data think tank engine module 3 0 0 0 provides flexible, automated, and iterative data mining of large unknown data sets. The statistical comparison in the data set provides the &quot; delivery &quot; operation. The flexibility of this algorithm is particularly useful in the exploration process, where the data consists of numeric and category attributes. Examples of algorithms that need to be fully explored for this data include, but are not limited to, special variation analysis (ANOVA) techniques that can correlate categories and digital data. In addition, more than one algorithm is usually required to fully explore the statistical comparisons in this data. This information can be obtained in modern discrete manufacturing processes such as semiconductor manufacturing, circuit board assembly or flat panel display manufacturing. . (Data think tank engine module 3080 ^ includes a data mining software program (herein referred to as the DataBrainCmdCenter program), which uses a paper included in the configuration file on page 37. This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm) A7 B7 1230349 V. Analysis template and data set of the invention description () for data mining analysis β According to one or more specific embodiments of the present invention, DataBrainCmdCenter shoes One of the following data mining algorithms &lt; Please read the notes on the back before filling out this page) One or more: SOM (a data mining algorithm known to those skilled in the art), · Rule method (&quot; RI ", A data mining algorithm known to those skilled in the art; MahaCU, a data mining algorithm to be explained below: Kazakhstan associates digital data with category or attribute data (such as, but not limited to, its ID)); Reverse Mahaku (a data mining algorithm to be described below, its associated category or attribute data (such as, but not limited to, the processor I ID) digital data); Multi-level analysis automation, where SOM is used for data mining, and where the output from SOM is used for data mining, using (a) RI, and (b) Mahaku; pigin (to be described below) An invented algorithm for data mining is described); DefectBrain (a data mining algorithm that will be detailed below); and Selden (a predictive model data mining algorithm known to those skilled in the art). Ministry of Economic Affairs Printed by the Intellectual Property Bureau employee consumer cooperative. In one or more embodiments of the present invention, the DataBrain Cm Center program uses a central control program to enable the use of multiple data mining algorithms and statistical methods. In particular, according to one or more In this specific embodiment, the central control program enables the results from a data mining analysis to be input or executed for subsequent branch analysis. As a result, the DataBrainCmdCenter program provides an automated and flexible mechanism for exploring data and configures the system by the user Set the depth and logic of the groove-dominated analysis (logical, capable of unrestricted data exploration without restrictions on the number and type of analyses In its most common form, the manufacturer's data analysis system has 3000 points. Page 38 This paper money is the Chinese national fresh (CN ^^ 2ig); ^ 1230349

V. Description of the invention ((Please read the notes on the back before filling out this page) The Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs prints and analyzes the information received by most manufacturing companies, not all of which is owned or controlled by the same legal entity As a result, different data sets can be analyzed simultaneously in parallel shellfish mining analysis and then reported to different users. In addition, even when the received data comes from a single manufacturing plant (that is, owned or controlled by a single legal entity) ), Different data sets can be analyzed by different clusters in the legal entity at the same time in the data mining analysis performed in parallel. In this case 'such data mining execution can be effectively implemented in the farm. According to According to one or more specific embodiments of the present invention, the data think tank engine module 3080 serves as an automated command center and includes the following components: (a) — Quite useful &gt; DataBrainCmdCenter program (analysis of decision and control program) Branch); and further includes: (i) a DataBmnCmdCenter queue management φ (according to the person skilled in the art Any one of a variety of methods), which includes at least a set of distributed slave server arrays in a server farm, one of which is set as a master queue; (11) a DataBrainCmdCeiUer load balancer program (according to those skilled in the art Any one of a variety of methods known), which balances the distribution and workload in the server farm; (out) a DataBrainCmdCenter account management program (manufactured according to any of the methods known to those skilled in the art) ' It can create, manage and monitor the status of customer accounts and combine analysis results; and (b) user edit and profile interface module 3055 (manufactured according to any of a variety of methods known to those skilled in the art) ' A user can use the analysis template information in the configuration file for data mining. According to this specific embodiment, the DataB rain Cm dC enter program is mainly negative, and the third requirement is that the paper size applies the Chinese National Standard (CNS) A4 specification ( 210X 297 mm) A7 B7 1230349 V. Description of the invention () Responsible for managing a data mining work queue and automatic distribution of work To an Internet window server server or an array of server servers. The DataBrainCmdCenter program and the user edit and configure the file interface module 3 0 5 5 into an interface to receive turns for system configuration parameters. According to one or more specific embodiments, data mining is defined as a set of analysis operations composed of multiple data sets and analysis algorithms. The work is managed by the DataBrainCmdCenter queue manager program, which is a program that resides in an individual server. The master queue manager on the slave slave queue. The master queue manager logically distributes the data mining work to the available servers (implemented by the data think tank module) so that the work can be performed simultaneously. The results of the branch execution analysis are collected by the DataBrainCmdCenter program, and if required, they will be entered into subsequent executions as required by the job profile. In addition, the DataBrainCmdCenter program controls the server field load balance. Balancing helps to gain efficiency and control the server resources available in the server farm. Appropriate load balancing can be achieved by real-time monitoring of individual server farm server queues' and other information on execution time status based on any of a number of methods known to those skilled in the art. In one or more embodiments of the present invention, the DataBrainCmdCenter account management program can create, manage, and monitor the status of a customer's account related to ongoing automated analysis based on any of a variety of methods known to those skilled in the art. Management and status communications provide control feedback to the DataBrain CmdCenter queue manager program and the DatdraincmdCenter load balancer program. &quot; According to one or more specific embodiments of the present invention, the paper size of the data mining analysis is applicable to the Chinese National Standard (CNS) A4 specification (21 × 297 mm) (Please read the precautions on the back before filling in this Page)-Order · Line · Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic AffairsΪ 230349

2. Description of the invention (m is to analyze the data f to find clusters of materials that can provide correlations (this step may require the use of several data mining steps to use various types of data that may provide ::: linkage to analyze the data). This step is ㈣ 疋: driven by the data type identified by Tanaka. Then in the following steps, the associated data can be analyzed to determine the parameter data that may be related to the cluster (this step may need to be used in several data mining steps It is possible to provide various types of data for this correlation to analyze the data.) This step is also driven by the type of data set in the configuration file and the type of data mining analysis to be performed. In the following steps after iw, the parameter data may Data analysis to determine processing tools that may be associated with related parameter data (this step may require the use of various types of processing tools to analyze the gas data in several data mining steps). Then in the following steps, Processing tool sensor data may analyze category data to determine opinions about processing tools that may fail This step may require several data mining steps to use various types of sensors that may provide this correlation to analyze the data.) According to one such specific embodiment, a system of data mining analysis techniques is the use of SOM, Next is the rule induction method, then ANOVA, and then the statistical method. Figure 9 shows a 3-layer, branched data mining analysis execution example. As shown in Figure 9, the DataBrainCmdCenter program (in the user-created profile) Guided by the analysis template) Perform an SOM data mining analysis, in which the cluster digital data is related to (for example, but not limited to) the yield, and the definition of the & rate is related to the manufacturing speed of integrated circuits in the manufacturing plant. Second, as described in Figure 9 shows the progress of the DataBrainCmdCenter program (on a user's page 41, the paper size applies the Chinese National Standard (CNS) A4 specification (210X297)) (please read the precautions on the back before filling this page)-Order · Line Economy Printed by the Consumer Cooperatives of the Ministry of Intellectual Property Bureau A7 B7 1230349 V. Description of the invention () Guided by the analysis template generated by s): ( Perform a map matching analysis on the output of the som data mining analysis (explained below) for cluster matching when it is related to, for example, but not limited to, electrical test results; and (b) the SOM data mining analysis The output is subjected to rule induction data mining analysis to provide a rule explanation of the cluster, when it is related to, for example, but not limited to, electrical test results. Second, as shown in Figure 9, the DaUBirainCmdCenter program (generated by a user Guided by the profile analysis template): (a) Perform a reverse mahaku and / or ANOVA data mining analysis on the output of rule induction data mining analysis to correlate category data to digital data. Process tool settings such as (for example, but not limited to) the weights and measures measurements made at the processing tool; and (b) performing a map of the map matching analysis output) and anova data mining analysis to cross-link digital data to The category information is related to process tools such as (for example, but not limited to) to sensor measurements. Figure 10 shows a distribution queue implemented by the DataBrainCmdCenter program according to one or more specific embodiments of the present invention. FIG. I i shows a user interface portion of an analysis template of a user editing and setting slot interface module 3055 manufactured according to one or more specific embodiments of the present invention. FIG. 12 shows a part of a profile analysis template manufactured according to one or more embodiments of the present invention. According to one or more specific embodiments of the present invention, a calculation algorithm (hereinafter referred to as "map matching,") uses SOM to achieve dynamization and focus analysis (ie, to provide &gt; an automated definition of a problem statement). Especially according to this One or more specific embodiments of the invention, SOM provides a place for wafer clusters with the same parameters. The 4th paper size is applicable to China National Standards (CNs) A4 specifications (210x297 mm) (Please read the precautions on the back before (Fill this page)-Order · Line · Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 1230349 A7 --------- V. Description of the invention () (Please read the precautions on the back before filling this page) Figure For example, if a map is generated for all parameters in the data set, it can be used to determine how many unique yield problems exist for a given product within a given day (for a given product, t). Maps can be used to define well-defined "questions" to require further data mining and analysis. Because the nature of self-organizing maps can automate the analysis, users of S0M map matching technology only need to keep the "attention" changes in the manufacturing plant. Number the name list to achieve a fully automated "delivery". .SOM automatically organizes the data, and identifies the separation and domination (ie impact) of data in a data set representing different "manufacturer problems". Cluster. This SOC cluster (combined with the map matching algorithm described below) enables each, and interested, variable to be affected by any known data in any cluster that is interested in the behavior of the variable. Description. In this way, the use of SOM in combination with map matching algorithms enables manufacturers to fully automate the "delivery" of analytical techniques to cope with multiple shock yield issues (or other important issues). Employees' Cooperatives, Bureau of Intellectual Property, Ministry of Economic Affairs Before printing the SOM analysis of a data set, a self-organizing map must be generated for each block in the data set. In order to generate these maps, a super pyramid box as shown in Figure 13 will be created. The super pyramid shown in Figure 13 The cube has 4 layers. According to one or more specific embodiments of the present invention, all super pyramid cubes grow to layers of 2MX2An, where n is the number of layers based on 0. In addition, Each layer of the pyramid represents a hyperblock, that is, each layer of the pyramid represents a field in the data set. Figure 14 shows the second layer of a 16 block data set (based on 0). According to a Or more [such specific embodiments, the more the depth of the super pyramid, the greater the width of the super cube (2Ληχ2Λη), and the depth of the super pyramid is maintained as a fixed number of data group fields. The paper size applies the Chinese National Standard (CNS) A4 specification (2ΐ〇χ297mm) ~ * ------ A7 1230349 _______B7 _ V. Description of the invention () (Please read the notes on the back before filling this page) Figure 15 shows a super cube layer extracted from the second layer of a super-pyramid cube and from a super cube layer of a super cube (self-organizing map). As shown in Figure 15, the neurons (ie, cells) in each layer represent the approximate values of the real and global records of the block. When the depth of the pyramid goes down, the superblock becomes larger, and the neurons in the block increase to converge to the true value recorded in the actual block represented by each layer of the data block. Because of memory limitations and the computational time involved, it is neither practical nor feasible to grow the pyramid until the neuron converges to the true value it represents. In contrast, according to one or more specific embodiments of the present invention, the pyramid will grow to meet a certain critical value, or reach a predetermined maximum depth. Subsequently, according to one or more specific embodiments of the present invention, the SOM analysis will be performed on the last layer of cubes generated by the pyramid. Once a SOM is generated for each column of the data set, the following steps will be taken to achieve an automated map matching data analysis. L ·-^ Generate a snapshot (anti-frost) Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs Given a digital strain number ("DV") (data block), locate a type of neural map in the data box about the DV. This type of neural map produces a combination of all possible colored regions detailing the three regions. These three areas are: high (mound), low (pool), and middle area, and any given cell on the neural-like map will fall into one of this area. For easy understanding of this specific embodiment, »4 colors are designated as two regions, blue is the middle region and red is the low region. Then in the first step, a delta is determined, which is the interval at which the poisoning time needs to be moved to generate a snapshot of the colored area, which is used as the basis for the automated map matching analysis. Please note that two critical points are required when moving to obtain all snapshot combinations. Page 44 This paper size applies the Chinese National Standard (CNS) A4 specification (21〇χ 297 public love) 1230349 A7 B7 V. Description of the invention (marked 'i.e. There needs to be a mark representing the critical point of a low area and another mark representing a high area. By changing these two marks and using the difference, a snapshot combination of all requirements can be generated. The difference 1 is calculated as follows: difference = (Percentage of data distribution—this is the value set by the user) * 2. Secondly, the high and low marks move to the average value of the data in this block. In this initial state, the cells in all neural maps will fall in ocher red The area of the towel—Next, the low mark moves to the left by a difference. Then all cells are scanned, and then the appropriate color will be specified according to the following steps. If the relevant cell value is: (average "· 25} &lt; cell value &lt; low If it is marked, it will be assigned a red color. If the relevant cell value line is: (Highly marked cell value &lt; (average value +1.25)), it will be assigned a green color. If the relevant cell value line is · (low mark) &lt; The cell value &lt; (Xin mark) will be assigned a blue color. At each snapshot (repeated), all the high and low areas are marked, and then SOM automatic analysis is performed (detailed later). Then, the low mark is moved to the left by a difference to generate another snapshot. Then all the high and low areas are marked and automatically analyzed by SOM. This process will continue until the low mark is less than (average -1.25 σ). The low mark will be reset to the initial state ', and the high mark will advance to the right by a difference, and then repeat the process. This will continue until the high mark is greater than (average value + 1 · 25σ), which is represented by the following virtual program ( In Chinese). Set High—Marker = Shelter data ¥ Average setting Low—Marker = Average data of stop data Set Delta = (Data distribution percentage, this is set by the user. Page 45 This paper scale applies to China Standard (CNS) A4 specification (210X297 mm) {Please read the notes on the back before filling out this page}-Ordered · Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 1230349 V. Description of the invention () Fixed value) * 2 σ Set Low_Iterator = Low_Marker and set High_Interator = High—Marker to keep the loop, when (out §11_116: &amp; 1: 〇1 * &lt; (average value +1.25 (7)) start the loop to keep the loop, When (High-Iter at or &gt; (average-1.25 σ)) starts to cycle through the cells and use the High-Iterator and Low-Iterator as the threshold to program the cells to color according to the above steps. Take this snapshot by marking the high and low group exhaustion. Perform this snapshot automatic map matching analysis (see the next section for details). Set Low—Iterator = Low-Iterator-Difference. End of Loop Set High-Iterator = High_Iterator + Differential End of Loop. Figure 16 shows a self-organized map with high, low, and middle regions, and labeled high and low cluster regions for subsequent automatic map matching analysis. II. Automatic map matching analysis-a snapshot (repeatedly) Each 3-color area snapshot generated in the first step will be analyzed as shown on page 46. This paper size applies the Chinese National Standard (CNS) A4 specification (210X297 mm) (please first Read the notes on the back and fill out this page)-Order · Line: · Printed by the Consumers 'Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 B7 1230349 Five Printed and Invented by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economy () Area (the user clearly indicates that they are interested in the pool (low) or ridge (high) area in a selected DV (block) neural map). One area of interest will be called the source area and the other opposite area will be called the target area. It is used to obtain the automatic SOM rating of other (independent variables &quot; IV ", that is, not a DV block in the data block) the map level, and the row (record) based on the same block is highlighted by the data block. Therefore, if a Line 22 of the data set is located at line 10 and line 40 of a given DV neural map. For all other maps such as IV, the cell position (22, 40) will also include line 22 of the data set. In particular, Figure 17 shows a cell that highlights a super cube. As seen in Figure 17, a "best fit" record has been established so that when it protrudes through the layers of the super cube, it best matches the layers of each layer. estimated value. In short, the goal is to analyze at least the source and target area records and determine the degree of difference between the two. Because the records of each group in this type of neural map are the same, various types of neural maps can be rated according to how strongly the source group and the target group are different. This score is then used to rank neuron-like maps from highest to lowest. A higher score indicates that the two groups in the neural-like map are different from each other, and a relatively lower score indicates that the two groups are very similar to each other. Therefore, the goal is to find a class IV neural map, where the differences between the two groups are greatest. The steps shown below are used to achieve this. a. The source cluster is rated from highest to lowest according to the impact score. The impact score of each cluster is calculated as follows: Impact score = (average of the actual field-average value of the neural map) the number of unique records in the cluster / all records in the block. 'b • Start with the highest-rated cluster of sources and group them according to the following criteria. Page 47

A7 B7 1230349 V. Description of the invention () Labeled by the neighbors of each cluster. · Each standard is weighted according to the weight, and the actual designation result is the weighted average. · (Please read the notes on the back before filling this page) 1 · Close to the source cluster. This system calculates the central distance from the target cluster to the source cluster, where the central cell line occupies the cells in the center of the cluster. After deciding the two cells, the center distance can be calculated using Bishop's theorem. 2. The number of unique records in the cluster. 3. Comparison of the average of surrounding cells to the average of surrounding cells. This gives a one-to-man relationship, i.e., a source cluster is associated with many of its adjacent target clusters. c. Mark &amp; records in all source clusters as Group 1 and records in all target clusters as Group 2. This will be used to determine the differences between the two groups based on the following. d. Use a scoring function with ethnic group 1 and ethnic group 2 as inputs to calculate the π score of IV ". This scoring function includes (for example, but is not limited to) an improved T-test scoring function; a color contrast scoring function An IV impact scoring function, etc. The improved T-test scoring function is implemented as follows: The Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs prints maps of each IV (Nerve-like) map, and calculates the improved T-test for group 1 to group 2. The improved-T-test is based on a comparison of the general T-test of the two annihilates. The difference is that after the T-test score is calculated, the final score is calculated by multiplying the T-test score by a reduction ratio. Page 48 This paper size applies to China National Standard (CNS) A4 (210X297 mm) 1230349

〇dified—T — TEST = (reduction ratio) * τ-TEST The reduction ratio is calculated from the number of records in the target population of points that exceeds the average of the source population. Then subtract this number from the number of records in the target population that are lower than the average of the source population. Finally, the reduction ratio is calculated by dividing by the number of all records in the target group. Reduction ratio = absolute value of (number of target records below source average-number of target records exceeding source average) / (number of records in target area). Keep this score for later grade IV neuromap evaluation. The color contrast scoring function is implemented as follows: Compare the color contrast between groups 1 and 2 on a class IV neural map. The IV impact scoring function is implemented as follows: Multiply the color contrast score determined above by an impact based on the DV type neural map. fraction. e · Repeat steps d · f · for each type IV neural map of the hyperblock, and grade the type IV neural map according to the modified-T-test score. If the modified-T-test score approaches zero before all IVs are used or before a user-specific critical point is reached, the remaining iv-type neural maps will be rated using the general τ-test score. g. Store a certain percentage of the previous type IV neuromap according to the user profile settings. ί π · Production results and rotation results to shout out the other methods Page 49 This paper size applies the Chinese National Standard (CNS) A4 specification (210 × 297 public love) (Please read the precautions on the back before filling this page) Order · Line-Printed by A7 B7 1230349 of the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. Description of the invention () Select a certain percent x (according to the user profile) type IV neural map before the highest overall score. According to two or more specific embodiments of the present invention, automated results will be generated for the user to review for each selected snapshot. a · Display a neural map selected for IV. The som map of the independent variable will be the background map with the strain number ridges and pool clusters on it, with significantly different external colors and obvious cluster marks. The description of the map is based on the actual value of the critical point of the color boundary in two different colors (such as green, red, and blue). b. Perform this particular selected DV to get actual results. For a given selected DV, this is the actual result of how the IV differs from the others. c. A smaller data set will be written, which contains only records of the source and target area of the composition. This smaller data set will serve as a benchmark for further analysis by other data analysis methods. For example, to get the "problem of automation", this smaller data set is fed back to a regular induction data analysis engine that has the appropriate shape of the region from which map matching is performed. These regions will form a regular induction data analysis that will be explained ' 'Problems'. Rule induction uses statistical validity to generate rules for the interaction of explanatory variables. It searches the database to find the hypotheses that best fit the problems that have arisen.

Ιν · Repeat the above steps for all the right DVs; MII repeats steps I to III for all DVs specified in the profile by the user. Carry out all management work and prepare to produce automatic map matching results Page 50 This paper size applies the Chinese National Standard (CNS) A4 specification (21〇χ 297 mm) (Please read the precautions on the back before filling this page)

Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 1230349 Α7 Β7 V. Description of the invention (

Report 'and input these implementation solutions F solution valley into other data analysis methods. According to one or more specific embodiments of the present invention, the data think-tank module I includes-referred to herein as &quot; Pigin'f material mining algorithm program. This is an invented data mining performance scapular method method, which decides (for a target number variable) that other variables in a data set contribute to (that is, are associated with) the deduction of the target variable. Although Pigin does not analyze category data (and accordingly, it is narrower than other data mining algorithms), the analysis is faster and more efficient than using other standard data mining algorithms. This algorithm handles the target variable ("DV &quot;) of the target variable (the variable to be explained by the data mining operation). The algorithm operates according to the following steps. Step 1 · Treat a number distribution of DV as a series of The category is based on a user-configurable parameter that puts little data into each category based on the decision. Steps are not shown in Figure 18, which are defined by a digital distribution as a "virtual" category. Step 2: Once the step丨 Define the dV group (or faction). For each DV category, a series of trustworthy circles will be calculated for each dV category based on the category that is consistent with other numerical variables in the data group (hereinafter referred to as the independent variable or nIV &quot;) Step 3: According to all the iv trust circles that are fully expanded, a diameter fraction and a drop difference number are assigned to the variable for subsequent use to determine which IV is most closely related to an analyst's view as a "target", DV❹ The diameter or the number of high dropouts usually indicates that there is a correlation from DV to IV, and it is better. "Steps 2 and 3 are shown in Figure 19, which shows the calculation of the number of dropouts (one dropout = all dropouts). The sum of the difference (not in any circle) and the diameter fraction (a diameter fraction == the average DV diameter of three circles), where the DV category is based on the digital distribution of DV. In essence, Figure 19 is a trust graph, Among them, page 51 of this paper applies the Chinese National Standard (CNS) A4 specification (210X297 mm) (Please read the precautions on the back before filling out this page)-Order · Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs and Consumer Cooperatives 1230349 A7 B7 V. Description of the invention () (Please read the notes on the back before filling out this page} Each diamond shape represents a group, and the end point of the diamond produces a circle, and the day is on the right side of the chart (these circles represent "95% trust Circle "). Step 4 is repeated. Once all 1Vs are assigned a score based on the DV definition of step 1, then DV is redefined to slightly change the definition of distribution. Once redefined, 'All IVs will be recalculated according to the new DV category definition. Points. The process of resetting the definition of the DV category will continue until it meets the number of users specified in the analysis template. Step 5: Overall score. When all the responses are completed, follow the steps described in steps 1 and 4. There will be a series of IV ratings for the definition of various dVs. These lists will be incorporated to form a list of "major scores," which are most relevant to the target DV, when listing the major scores for a given IV We will consider three factors: the size of the difference, the diameter score, and the number of times the IV appears on the DV score list series. This three factor combined with some basic exclusion criteria for "useless results" forms the most suitable DV for a given target. A list of correlation IVs. This is shown in Figure 20. It should be noted that although one or more of these specific embodiments use a fall difference number and diameter fraction for each of the IVs in contact, specific embodiments of the invention are not limited to This type of score, and in fact there are further specific examples of using other scoring functions to calculate the IV score. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. According to one or more specific embodiments of the present invention, the data think-tank module includes a correlation program (Mahaku), which associates digital data with category or attribute data (such as but not (Limited to the processing tool ID), its program provides: (a) fast statistical output of grading according to qualitative rules; (b) scoring based on [diameter score and / or fall difference score; (c) a critical point for scoring In order to eliminate the poorly performing tool ID '(d) —the ability to choose to display the numbers in front of the “find”; and (e) the 5th paper size applies the Chinese National Standard (CNS) A4 specification (210X297) (%) 1230349 A7 B7 V. Description of the invention (1) Ability to perform a reverse execution, in which the result of "finding" (tool ID) can make the number of strains affected by these "finding" (tool ID) and The parameters (numbers) are displayed. (Please read the precautions on the back before filling out this page} Figure 21 shows an example of a subset of the data matrix of the smart module module associated with the input data. This example shows batches Benchmark Point-line survey data (BIN), process tool ID (Eq-Id) and process time (Trackin). Similar data measurements are also generated based on a wafer, location (mask) or mold. Figure 22 shows a number (BIN ) Example of category (tool m). Using BIN (number) as the strain number, the above data think tank module association program produces similar graphics for Eq-Id (category) in all data matrices. The width of the diamond shape on the left side of the figure Represents the number of batches executed by the tool, and the circle diameter on the right side of the drawing represents 95% confidence level. To classify many graphics, the sum of the space between the circles (that is, the area not surrounded by the circle), and the top and top of the top circle The total distance between the bottom of the circle below will be used as part of the formula to calculate which is better. "Flatness number, or ,, diameter fraction,". The above information think tank module association program classifies the graph to provide What type of score the user chooses is an important basis when it is better. The Intellectual Property Bureau, Ministry of Economic Affairs, Employee Consumption Cooperative, prints another view based on this specific embodiment of the present invention, above The data think tank module related program sets a scoring threshold. Although there are usually many processing tools that can be used to process a specific layer of the integrated circuit, only a part is normal use. There are many other processing tools. It is abnormal use that makes the data wrong, and may cause unnecessary interference when processing the data. The above data think tank module related program can apply a user-defined scoring value, so that the paper size on page 53 applies the Chinese national standard ( CNS) A4 specification (210X297 mm) 1230349 A7 B7 V. Description of the invention (Poorly performing guards can be transitioned before analysis. For example, if the critical point of the score is set at 90, Figure 23 shows that XTOOL3 will be filtered out in the round two tools, and ㈣㈣⑷ and coffee ⑽ contain more than ㈣ products. According to one or more specific embodiments of the present invention, the above-mentioned data think tank module association program provides the option of "scoring the previous number". Using this feature, the user can determine the maximum number of results that can be displayed for each strain. Therefore, the above-mentioned data think tank module correlation program performs analysis on all independent variables. 'Only when the figure number enters, the number before the scoring number enters the range' will be displayed. According to the present invention-or more specific In the embodiment, the above-mentioned data think tank module: the link program also performs a reverse execution (reverse maha library), in which the category (such as, but not limited to, industrial ID) is a strain number, and the numerical parameters affected by the category (such as, but not limited to, BIN, electrical test: weights and measures, etc.) are displayed according to the importance of the basin. The importance (scoring) is the same as when the number is executed on the tool ID. This execution can be the "Daisy" key, where The work ID detected during normal execution can automatically become the strain number during reverse execution. According to one or more specific embodiments of the present invention, the data think-tank module includes a program called DefeetBrain module. Defective skills will be equal to the defect title. However, for this analysis, the defect information is changed to one. Format it (detailed below). Round 24: Defect inspection tools or defect inspection tools in the second manufacturing plant Examples of generated defect data files. In particular, this file usually includes data about x and y coordinates, X and y mold coordinates, size, defect type classification code, and image information of each defect on the wafer. According to this One or more of the inventions are implemented on page 54. The paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) (Please read the precautions on the back before filling this page) Staff of the Intellectual Property Bureau of the Ministry of Economics Printed by the consumer cooperative A7 B7 1230349 V. Description of Invention () For example, the data conversion module 3020 compiles the defect data file into a matrix containing at least size, classification (such as defect type), and defect density on the mold layer. Figure 25 shows an example of a data matrix generated by a data conversion algorithm. Secondly, according to a specific embodiment of the present invention, the DefectBrain module includes at least an automated data mining error detection program that grades defect issues based on scoring techniques. .According to this program, special size picking or the impact of a defect type can be used here as the "Kill ratio", the parameter plus Quantify. The damage ratio is defined as follows: Damage ratio = the age of a mold with a defective pattern, the number of all molds with a defective pattern. Another parameter that can be used is the percentage of loss, which is defined as follows: percentage of loss ~ In the above definition, a bad mold refers to a mold that cannot function. Figure 26 shows a typical output of the data think-tank module program. In Fig. 26, a digital pattern of a mold containing a special defect pattern (in this case, fine marks) is drawn with respect to the number of defects of the pattern on a mold. Since the active (ie good) and non-active (ie damaged) mold information exists in the data matrix ', it can directly determine which of the molds with special defect types is good / damaged. Therefore, in Fig. 26, the frequencies of both the good and damaged molds are plotted, and the ratio of the damaged molds to all the defective molds (that is, the failure ratio) is plotted graphically. In this figure, the slope of the graph segment has been compared with the slope of all other graphic segments generated by the data think-tank module program, and then it is evaluated from the highest slope to the lowest slope. With the highest slope, page 55. This paper size is applicable to the Chinese National Standard (CNS) A4 specification (21〇χ 297). (Please read the precautions on the back before filling out this page} 4 ·· -Order · Line-Ministry of Economy Wisdom Printed by the Consumer Cooperative of the Property Bureau 1230349 A7 Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 5. The description of the invention will affect the most important part of the yield, and it has value for the yield improvement project. 〃 This chart is important The feature is the data intelligence module, the program can adjust the number of defects on the X axis, and the maximum value of # 丄 默. If not; with this ability, in the event of an abnormal number of defects on the mold (such as In the absence of importance or artifacts, the slope rating will be wrong. According to one or more embodiments of the present invention, the data think tank module is designed to be used as a data cleaner (for example, according to the knowledge of those skilled in the art). Any of a variety of methods-Perl and C + + software programs made), data compilation = (for example, Perl and C + + software programs made according to any of the methods known to those skilled in the art ) And data filters (such as Perl and C ++ software programs made in accordance with any of the various methods known to those skilled in the art), where data cleaning, data compilation, and / or filtering According to one or more specific embodiments of the present invention, the data think tank module is a software application program running on a PC server, and Any one of the various types known to those skilled in the art is written in C + 10 and SOm. According to one or more specific embodiments of the present invention, the output of the data think tank engine module 3 080 is Microsoft FoxPrTM data Database format result database 3090. Furthermore, according to one or more specific embodiments of the present invention, the network manager module 3070 manufactured according to any one of the methods known to those skilled in the art includes The security FTR transmission software can be used by users or customers to send data to the data think tank engine module 3080 for analysis. Page 56 This paper standard applies to China National Standard (CNS) A4 specification (2 10X 297 public love) (Please read the precautions on the back before filling out this page)-Order · Line: ··

Su I I 1230349

The rules are based on the η life machine and the τ fetch (1) and the forest (B〇〇lea, Zhu M, reply to the door of data mining, 4 different methods (in this case, the rules are summarized, or provide some information about the points). The 模板 ping template determines Α $ _ Α φ 4 or, the calculation effect is determined by the algorithm variables in the configuration file, depending on what special data is used U ^ ^ ± 1. The nasal method k includes at least one, and the result " (Keyo data or categorical variable types) Information box ^ shell type, the system meter output map is pre-integrated, which can be classified by users to accompany each automatic data acquisition in multiple specific embodiments, This kind of data from the kite &quot; ... wide-range data mining &quot; is the same as ^ ^ ^ „, the 枓 果 身 枓 group, which includes only the constellation 禾 之 field. After the automatic data mining analysis is completed, I hope that such resources are available. The cheeks are all stored in the result database 3900. Distribution: Printed by the consumer cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs as shown in Figure 3, according to one or more of the present invention In a specific embodiment, the network visualization module 3100 executes a graphics and analysis engine 3ιιο A database of results 390 generated by the data think tank engine module 3080 is provided to provide (for example, without limitation) an HTML report stored in a web server database 3120. According to one or more of the present invention In a specific embodiment, the web server database 3 120 may be used by a user to transmit a report on a PC, such as, but not limited to, a PC in accordance with any one of a variety of methods known to those skilled in the art. One or more specific embodiments, the network visualization module 3 1 00 can interactively report the results, and the network browser can be used to generate charts, reports, and Pοwerp〇int files for output, generation, and repair. The paper size of this page applies the Chinese National Standard (CNS) A4 specification (210X297mm) 1230349 A7 B7 V. Description of the invention () Printing and modification of the profile, account management, notification of the results by email and Multiple users can use the right to share information. In addition, according to one or more specific embodiments of the present invention, the network visualization module 3100 allows users to generate Microsoft Powe rPoint (and / or Word) online cooperation documents, enabling multiple users (under appropriate security access) to view and modify. According to one or more specific embodiments of the present invention, the network visualization module 3 丨〇〇 is a software application running on a pc server, and is written using Java Applets, Microsoft Active Server Messaging (ASP) programs, and XML. For example, the network visualization module 3 1 00 includes a management module ( For example, a software application running on a pc server and written in online Microsoft ASP according to any of a number of methods known to those skilled in the art, which can be used for new user settings (such as but not limited to including Enter various system functions, security specifications), and can be used for user rights (including use rights (such as, but not limited to) data analysis results, profile settings, and the like). The network visualization module 3100 also includes a job monitoring module (for example, a software application running on a PC server), and it is based on any one of a number of methods known to those skilled in the art. (Written in ASP), which allows users to view analysis results and produce reports. The network visualization module 3100 also includes a charting module (for example, a software application running on a pc server, and is written according to any one of the many methods known to those skilled in the art, the network Microsoft ASp), It allows users to generate special charts using their web browser. The network visualization module 3 100 also includes a link-block module (eg, a software application running on a pc server and is based on any one of a variety of methods known to those skilled in the art) Written by Microsoft ASP), which allows users to read the information on page 58 (please read the precautions on the back before filling in this page). Install · -Order · Line 1230349

V. Description of the invention () Mining and / or forming a pre-block combination data set. Network Visualization 3 1 00 also includes-a device module (for example, running on a pc feeder: a physical application, and based on the various methods known to those skilled in the art, who is the Internet-ASP (Compiled), which allows the user to filter the data collected in the superblock before mining such data, and the filtering step is performed according to user-specific criteria. The network visualization module 31 steps include-an online data tool module (for example, running on a pc server: a physical application, and according to a variety of methods known to those skilled in the art = any one of the network Microsoft (Written in ASP), which allows users to use their network = browser for data mining on a special benchmark. According to various embodiments of the present invention, a user may set up a profile to enable the network visualization group 3100 to prepare a statistical process control (&sp; r) chart of information, which enables the messenger to use a web browser Track predetermined data metrics. Those skilled in the art should understand that the above disclosure is for demonstration and explicit use only. Accordingly, it is not intended to be exhaustive or to cover or embody the invention in the particular form disclosed. For example, notwithstanding the above It was discussed that some dimensions are for demonstration only, because the specific embodiments described above can be made according to various designs, and the actual size of such designs will depend on the circuit requirements. Or molding f Please read the notes on the back before filling This page}-Order · Line-Printed by the Employees' Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 9 5 The size of the dance paper is applicable to China National Standard (CNS) A4 (210x297)

Claims (1)

^ 30349 _ Qing patent scope is used to extract the information of an accumulated product. ^ Obtaining assets from a manufacturing plant ("manufacturing plant"). This method includes the following steps: gathering from the manufacturing plant · within the manufacturing plant Production + ten &amp; multiple systems, tools, and data bases in the method t Λ the manufacturing data collected; medium; shell materials and storage of these data to a source data and according to the user-specific settings Part of the data is used for mining ~ shell material mining, to return to the application-specific analysis profile; = save the results of the data mining to-the results database; and ^ the right to use the results. The library is exercised (please read the notes on the back before filling out this page) — — 11 2. = Please refer to the method described in item i of the patent scope, where the aggregation is available on a network or on a scheduled basis. The process of instant material gathering. -Order. Step 3. As described in item i of the scope of patent application, the step of the μ right includes the step of providing the right of use through a network. Online_ The method described in item 3 of the patent application, wherein the step of providing the right of use through a network includes the step of providing the right of use by using a browser. ; Lubu page 60 This paper size applies Chinese National Standard (CNS) A4 specification (210X297 mm) 1230349 as Bo C8 -------- E8 _ 一 ____ 六 、 Scope of patent application The method described in item 1 of the patent scope, wherein the step of gathering is performed in real time, and the steps of formatting the data and storing the formatted data to a source database are also performed in real time. 6. The method as described in item 丨 of the scope of patent application, wherein the step of aggregating includes the step of aggregating data transmitted in a format specified by the customer and / or a tool. 7. The method as described in item 丨 of the patent application scope, wherein the step of aggregating includes the step of aggregating encrypted data. 8 · The method as described in item No. of the Patent Scope, wherein the step of formatting the data includes formatting the data into a product identification, a location (Where?), A time (When?), A subject (What?), And Steps in the hierarchy of value. 9. The method described in the scope of patent application (item 8), wherein the product identification is based on the batch 1D, wafer ID, groove ID, mask id, mold ID, and sub-mold x, y Karst coordinates. Defined by one or more. (Please read the notes on the back before filling this page) Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. The method described in item 8 of the scope of patent application, where the location (Where?) Is manufactured by the manufacturing streamline / combination line. Or more defined. / Page 61 This paper size applies to China National Standard (CNS) A4 specifications (210X 297 public love) VI. Patent application perimeter I 1. It is defined by one or more of the date / time of the measured value 12. == The method described in item 1, wherein the formatted v'g knitting and processing time data of the processing tools in the manufacturing process during wafer processing is used to become a key integrated circuit based on the-profile Steps for specific statistics. Π. The method as described in the item "Patent Patent", the storage step further includes extracting data from the source database and storing the extracted data into at least an inclusive-relationship database component and a slot system. Steps of mixing the components of the database. 14. The method as described in item 13 of the scope of the patent application, wherein the relational database uses a hashing and indexing algorithm to resolve the differences stored in the paging system components. The continuous data generates the right-of-use key. 15. The method as described in item 14 of the scope of the patent application, wherein the step of extracting is printed by the consumer property cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs, including using a hash-link algorithm to extract from the hybrid Steps to accumulate data in the database. 16. The method as described in item 14 of the scope of the patent application, wherein the step of extracting includes obtaining the i-block definition using the profile and using the super-block definition to generate an information vector cache Record definition, and generating a page of page 1230349 A8 B8 C8 D8 6. The steps of patent cache and vector cache record. 17 · The method according to item 16, wherein the step of generating an information vector cache record includes the following steps: (a) using a hash cable bow key to retrieve the information vector cache record definition obtained from the relational database component A list of files and data elements; (b) retrieving the file from the file system component; and (c) inserting the identified data elements defined in the vector fast record definition into the information vector cache record. 18. If applying for a patent The method described in the first item of the scope, wherein the step of extracting further includes the step of using a superblock definition &lt; the negative vector cache record to generate a superblock. 19. The method described in the first item of the patent application scope, The data mining step includes the steps of using one or more self-organizing map data mining, rule induction data mining, data mining to associate digital data to category or attribute data, and data mining to associate category or attribute data to digital data. Printed by the Consumer Cooperatives of the Ministry of Intellectual Property Bureau 20 · The method described in item 丨 of the scope of patent application, in which the steps of data extraction include Self-organizing map data mining to form clusters, map matching on the output of the SOM data mining, analysis to perform cluster matching, and the SOM data mining output to summarize data mining to analyze the rules of a cluster, On the output of data extraction based on the rules on page 63, this paper size applies the Chinese National Standard (CNS) A4 specification (210X297 public love) '^-1230349 ABCD VI. Patent application related category data to digital data, and in the map Steps for matching data mining and associating digital data to category data. 21. The method as described in item 20 of the scope of patent application, wherein the SOM data mining step automatically organizes the database and identifies a data group that represents a different "manufacturer problem" Separation and domination of data clusters, and map matching analysis describes any interested data within a cluster that is known to have an effect on the behavior of the "variable" variable. (Please read the precautions on the back before filling out this page.) · -Order · Line: · Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs, page 64. This paper size is applicable to the Chinese National Standard (CNS) A4 (210 × 297) (Centimeter)