JP2008171090A - Failure analyzing device, failure analyzing method, and failure analyzing program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a failure analyzing device for detecting any failure, and for performing correlation analysis in a short time. <P>SOLUTION: Inspection results 4, 6, 8, 12, and 14 for every product and every inspection are stored as pre-conversion data so as to be associated with the product lot number of each product and the inspection item of each inspection, and the set of a plurality of inspection items is generated from the combination of a pair of inspection items, and a data conversion formula for converting a pair of pre-conversion data corresponding to inspection items composing the set of the inspection items into converted data capable of calculating correlation strength is stored so as to be associated with each of inspection items composing the set of the corresponding inspection items, and the pre-conversion data are converted into the converted data by the data conversion formula for each of the inspection items composing the set of the corresponding inspection items, and the correlation strength is calculated between the converted data for each set of the inspection items, and a sequence is given to the set of the inspection items in the order of the stronger correlation strength, and the inspection items composing the set of the inspection items are read as defective factor items in the order of the earlier sequence. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a defect analysis apparatus, a defect analysis method, and a defect analysis program for analyzing defects generated in a product based on inspection results of inspections performed during and after the product manufacturing process.

  Products such as semiconductor devices and personal computers are manufactured through hundreds of manufacturing processes. Before and after these manufacturing processes, inspections are performed and defective products are excluded each time. However, a defect may be found by inspection after the product is completed. Such a defect causes a minor abnormality in a certain manufacturing process, and since the product is not a defective product, the product was subjected to the next manufacturing process as a non-defective product with minor abnormalities. It is thought that the abnormality developed into a defect. At this time, the inspection results of the plurality of inspections performed before and after the plurality of manufacturing processes should capture a minute change caused by a slight abnormality. The presence or absence of this minute change can be detected by correlation analysis of a plurality of inspection results. Then, if this correlation analysis is repeated and the inspection of the analysis target is traced back to the start history of the manufacturing process, the manufacturing process and the manufacturing apparatus that have caused the minor abnormality that caused the failure are finally identified. be able to.

  However, since there are hundreds of inspections corresponding to hundreds of manufacturing processes, if correlation analysis is performed on all pairs of inspections, the number of correlation analyzes exceeds 10,000 and manufacturing that has caused abnormalities. It took time to identify the process.

As a technique for supporting such correlation analysis, for example, a technique for performing analysis based on past analysis patterns (for example, see Patent Document 1) and a technique for applying regression tree analysis to yield analysis (for example, see Patent Document 2). ) Has been proposed.
Japanese Patent Laid-Open No. 2003-379882 JP 2000-284578 A

  The time taken for the correlation analysis is that the number of times is large, but it has been found that it takes time for each correlation analysis. If the inspection results of the individual inspections are left as they are, they cannot be subjected to correlation analysis, and it is necessary to process the inspection results for each correlation analysis.

  An object of the present invention is to provide a failure analysis apparatus, a failure analysis method, and a failure analysis program capable of performing correlation analysis in a short time after detection of a failure.

In order to achieve the above object, the defect analysis apparatus of the present invention stores the inspection result for each product as the pre-conversion data in relation to the product lot number for each product and the inspection item for each inspection. A pre-conversion data storage unit;
A set generation unit that generates a set of a plurality of inspection items from a combination of a pair of inspection items;
A data conversion formula for converting a pair of pre-conversion data corresponding to each of the inspection items constituting the set of inspection items into converted data capable of calculating the correlation strength, for each inspection item constituting the corresponding set of inspection items A conversion-type storage unit for storing in relation to
A data conversion unit that converts pre-conversion data into converted data by the data conversion formula for each inspection item that constitutes a set of corresponding inspection items;
A correlation strength calculator for calculating a correlation strength between the converted data for each set of inspection items;
An order assigning unit for assigning an order to the set of inspection items in order of strong correlation strength;
And a failure cause item reading unit that reads out the inspection items constituting the set of inspection items in the order from the earliest as a failure cause item.

Further, the defect analysis method of the present invention is
In relation to the product lot number for each product and the inspection item for each inspection, the inspection result for each inspection for each product is stored as pre-conversion data,
Generate multiple sets from a combination of a pair of inspection items,
A data conversion formula for converting a pair of pre-conversion data corresponding to each of the inspection items constituting the set of inspection items into converted data capable of calculating the correlation strength, for each inspection item constituting the corresponding set of inspection items Relate to, remember,
For each inspection item constituting the set of corresponding inspection items, the pre-conversion data is converted into converted data by the data conversion formula,
Calculate the correlation strength between the converted data for each set of inspection items,
Give order to the set of inspection items in order of strong correlation strength,
Inspection items constituting a set of inspection items are read out as failure cause items in order from the earlier order.

The defect analysis program of the present invention is
On the computer,
In relation to the product lot number for each product and the inspection item for each inspection, the pre-conversion data storage procedure for storing the inspection results for each inspection as the pre-conversion data for each product,
A set generation procedure for generating a set of a plurality of inspection items from a combination of a pair of inspection items;
A data conversion formula for converting a pair of pre-conversion data corresponding to each of the inspection items constituting the set of inspection items into converted data capable of calculating the correlation strength, for each inspection item constituting the corresponding set of inspection items A conversion formula storage procedure for storing in relation to
A data conversion procedure for converting the pre-conversion data into converted data by the data conversion formula for each inspection item constituting a set of corresponding inspection items,
Correlation strength calculation procedure for calculating correlation strength between the converted data for each set of inspection items;
An order assignment procedure for assigning an order to the set of inspection items in order of strong correlation strength;
A failure cause item reading procedure for reading the inspection items constituting the set of inspection items as the failure cause item in the order from the earliest is performed.

  According to the present invention as described above, it is possible to provide a failure analysis apparatus, a failure analysis method, and a failure analysis program capable of performing correlation analysis in a short time after detection of a failure.

  Next, embodiments of the present invention will be described in detail with reference to the drawings as appropriate. In each figure, common portions are denoted by the same reference numerals, and redundant description is omitted.

  As shown in FIG. 1, the failure analysis apparatus 30 includes a receiving unit 15, an inspection result storage unit 31, an inspection target storage unit 29, a conversion formula creation support unit 17, a conversion formula storage unit 18, a data conversion unit 19, a conversion Completed data storage unit 20, correlation analysis processing unit 21, correlation analysis result storage unit 22, output unit 25, mask database storage unit 24, mask database creation support unit 23, defect cause item extraction support unit 27, defect cause manufacturing apparatus extraction support A portion 28 is provided. The defect analysis device 30 may include a start history creation device 9 that outputs the start history 10 or may connect the start history creation device 9 externally. The inspection result storage unit 31 includes a pre-conversion data storage unit 32 and a normal time data storage unit 33.

  Outside the defect analysis apparatus 30, as an in-process inspection apparatus group 1, a device dimension inspection apparatus 3 that inspects a device that is a part of a product and outputs a device dimension inspection result 4; A device electrical characteristic inspection device 5 that outputs the inspection result 6 and a product in-process product inspection device 7 that inspects the product and outputs the product inspection result 8 in the manufacturing process are arranged. Further, outside the defect analysis apparatus 30, as a post-completion inspection apparatus group 2, a post-completion device inspection apparatus 11 that inspects a device and outputs a post-completion device inspection result 12, a product inspection and a post-completion product inspection result A post-completion product inspection device 13 that outputs 14 is arranged.

  As shown in FIG. 2, the conversion formula creation support unit 17 includes a set generation unit 41, a normal data reading unit 42, a normal item scatter diagram creation unit 43, a normal item scatter diagram display unit 44, and an error determination criterion. A value setting support unit 45, an error conversion formula setting support unit 46, a character data conversion formula setting support unit 47, a numerical data conversion formula setting support unit 48, a normal data conversion unit 49, and a normal inter-item scatter diagram storage unit 50 are provided. is doing.

  As shown in FIG. 3, the mask database creation support unit 23 includes a normal correlation strength calculation unit 51, a correlation strength determination unit 52, and a correlation label assignment unit 53.

  As shown in FIG. 4, the correlation analysis processing unit 21 includes a set extraction unit 55, an inspection item reading unit 56, a converted data reading unit 57, a correlation strength calculation unit 58, and an order assignment unit 59.

  As shown in FIG. 5, the failure cause item extraction support unit 27 includes a failure cause item reading unit 61, a converted data reading unit 62, a cause / appearance item scatter diagram creation unit 63, and a cause / appearance item scatter diagram display unit. 64, a normal item scatter diagram extraction unit 65, a normal item scatter diagram overlap display unit 66, and a defect cause item designation support unit 67.

  As shown in FIG. 6, the failure cause manufacturing apparatus extraction support unit 28 includes a start history extraction unit 71, a start history list display unit 72, a marker display unit 73, an inspection target display unit 74, and a failure cause manufacturing apparatus designation support unit 75. have.

  Next, a failure analysis method using the failure analysis apparatus 30 described above will be described. Assume that a product flows through a manufacturing process, and a plurality of products are manufactured sequentially.

  As shown in FIG. 7, in the failure analysis method according to the embodiment of the present invention, first, a start history 10 for each product is created in a start history creating apparatus 9 (see FIG. 1) in step S1. In step S2, the construction history creation device 9 stores the construction history 10 for each product in relation to the product lot number that can identify the product.

  Next, in step S3, if it is determined as defective in each inspection, for example, the inspection performed by the device dimension inspection apparatus 3 in FIG. The name of the part, the name of the manufacturing process that directly manufactures the part, and the manufacturing apparatus are stored in relation to the inspection item of the inspection as an inspection target.

  In step S4, the device size inspection device 3, the device electrical property inspection device 5, the product inspection device 7 in the manufacturing process, the device inspection device 11 after completion of the inspection device group 2 after completion, and the product after completion The inspection device 13 performs inspection for each inspection item for each product.

  In step S5, the device size inspection device 3, the device electrical property inspection device 5, the product inspection device 7 in the manufacturing process, the device inspection device 11 after completion of the inspection device group 2 after completion, and the product after completion In each of the inspection apparatuses 13, a device lot inspection result 4, a device electrical property inspection result 6, a product inspection result 8 in the manufacturing process, a device inspection result 12 after completion, and a product inspection result 14 after completion are inspected. And stored in relation to the inspected inspection item.

  In step S6, a strong correlation strength pair detection support method (pre-processing) is performed. Here, pre-processing is performed in preparation for the occurrence of a product defect. By this preprocessing, a pair of inspection items can be set as a set (a set of inspection items), and the correlation strength of a plurality of sets can be quickly calculated, and a set of strong correlation strengths can be detected from the plurality of sets.

  For example, as an example of a strong correlation strength set, the inter-item scatter diagram of FIG. 8 can be raised. Such a strong correlation strength is considered to be caused by the cause of the occurrence of defects. It is considered that the inspection item A and the inspection item B are both affected by the cause of the occurrence of the defect. On the other hand, as an example of a set of weak correlation strengths, the inter-item scatter diagram of FIG. 9 can be raised. Such a weak correlation strength is considered to be an indication that at least one of the inspection item A and the inspection item B is not affected by the cause of the failure.

  As a more specific example, consider a case where the semiconductor device is a product and the circuit pattern of the semiconductor device is an object to be inspected. As shown in FIG. 10, it is assumed that inspections such as “wiring 1 dimension”, “wiring 1 foreign object number”, “wiring 1 appearance”, and “wiring 1 resistance” are performed as inspection items in the manufacturing process. As a pair of inspection items in a manufacturing process to be subjected to correlation analysis, a set of “wiring 1 dimension” and “wiring 1 foreign matter”, a set of “wiring 1 dimension” and “wiring 1 appearance”, and “wiring 1 “Appearance” and “Wiring 1 foreign matter” set, “Wiring 1 resistance” and “Wiring 1 dimension” set, “Wiring 1 resistance” and “Wiring 1 foreign matter count” set, “Wiring 1 resistance” and “Wiring 1” Six sets of “appearance” sets can be generated. Note that the same set of inspection items such as “wiring 1 dimension” and “wiring 1 dimension” can be combined, but they are excluded because they have no physical significance in obtaining the correlation strength. In addition, the combination of “wiring 1 dimension” and “wiring 1 foreign substance” and the combination of “wiring 1 foreign substance count” and “wiring 1 dimension” in which the order of the inspection items is reversed are counted as two sets in combination. Although the correlation intensities are equal to each other and it is sufficient to obtain only one of them, the other is excluded.

  FIG. 11 is an inter-item scatter diagram of a set of “wiring 1 resistance” and “wiring 1 dimensions”. The “wiring 1 dimension” on the horizontal axis and the “wiring 1 resistance” on the vertical axis are logarithmically expressed. Even when the wiring is normal, the wiring 1 dimension and the wiring 1 resistance have a proportional relationship. However, if an abnormality occurs, the wiring dimension becomes narrower and further disconnected (open), the wiring resistance increases from the high resistance to the upper limit exceeding the measurement range. Indicates resistance. That is, it can be seen that the correlation strength increases as the dimensions become smaller than normal. In addition to the correlation analysis by the combination including “product pass / fail” such as the “wiring 1 resistance” and “wiring 1 dimension”, the correlation analysis between inspection items in each manufacturing process is also a cause of failure. It can be seen that it is effective for extraction.

  As another specific example, a case is considered in which a personal computer having a semiconductor device is a product, and inspection targets are inspection items after completion of the personal computer and inspection items in the manufacturing process of the semiconductor device. .

  As shown in FIG. 12, the inspection items in the manufacturing process include “wiring 1 dimension” “wiring 1 foreign object number” “wiring 1 appearance” “wiring 1 resistance” “wiring 2 dimension” “wiring 2 foreign object number” “wiring 2 external appearance” It is assumed that an inspection of “Wiring 2 resistance” is performed. “Power-off short check”, “Power-on short check”, “OS activation check”, “CPU operation check”, “Memory operation check”, “Margin” by binary evaluation of pass or fail as inspection items after completion It is assumed that a check and a “product pass / fail” inspection are performed. When the component semiconductor device has a two-layer structure of the wiring 1 and the wiring 2, for the sake of simplicity, for example, if it can be assumed that the wiring 2 has a structure that does not cause a manufacturing defect, the inspection item “ A set including “wiring 2 dimension”, “wiring 2 foreign substance count”, “wiring 2 appearance”, and “wiring 2 resistance” can be excluded from the correlation analysis target.

  FIG. 13 is an inter-item scatter diagram of a set of “wiring 1 dimension” and “power supply OFF short check (power ON short check)”. Since the “power OFF short check (power ON short check)” on the vertical axis is evaluated with two values of pass (pass) or fail (fail), the pass is converted to the numerical value 1 and the fail is the numerical value 0. Has been converted. As shown in FIG. 13, if the “wiring 1 dimension” is large, a short circuit occurs in the wiring 1 and an increase in standby current or an abnormal current occurs. Therefore, “Power OFF short check (power ON short check)” "Fail. Therefore, the short circuit of the wiring 1 can be determined by the “wiring 1 dimension” and the “power OFF short check (power ON short check)”. On the contrary, if the correlation strength of the set of “wiring 1 dimension” and “power supply short-circuit check (power-on short circuit check)” is strong, it can be estimated that the short circuit of the first wiring, which is a component, is the cause of the failure. Then, it can be considered that the manufacturing process and the manufacturing apparatus of the first wiring are the cause of the failure. Therefore, in step S3, if the correlation strength between the “wiring 1 dimension” and the “power supply OFF short check (power ON short check)” is strong in the inspected storage unit 29, the cause of the failure In order to be able to extract the first wiring and the manufacturing process of the first wiring as the part name, the “wiring 1 dimension” and the “power-off short check” (power-on short check), It is convenient to store one wiring and the manufacturing process of the first wiring in association with each other.

  FIG. 14 is a scatter diagram between items of a set of “wiring 1 dimension” and “OS activation check”. Since the “OS startup check” on the vertical axis is evaluated with a binary value of pass or fail, the path is converted to a numerical value 1 and fail is converted to a numerical value 0. As shown in FIG. 14, if the “wiring 1 dimension” is small, an open occurs in the wiring 1 and the OS does not start, so that “OS start check” fails. Therefore, the opening of the wiring 1 can be determined by the “wiring 1 dimension” and the “OS activation check”. On the contrary, if the correlation strength of the set of “wiring 1 dimension” and “OS activation check” is strong, it can be estimated that the opening of the first wiring as a component is the cause of the failure. Then, it can be considered that the manufacturing process and the manufacturing apparatus of the first wiring are the cause of the failure. In step S3, if the correlation strength of the set of “wiring 1 dimension” and “OS activation check” is strong in the inspected object storage unit 29, the first wiring and the first wiring are used as the part names that cause the failure. It is convenient if a set of “wiring 1 dimension” and “OS activation check” and the first wiring of the part name and the manufacturing process of the first wiring are stored in association with each other so that the manufacturing process can be extracted. is there.

  FIG. 15 is a scatter diagram between items of a set of “wiring 1 dimension” and “product pass / fail”. Since the “product pass / fail” on the vertical axis is evaluated by a binary value of pass or fail, the pass is converted to a numerical value of 1 and the fail is converted to a numerical value of 0. As shown in FIG. 15, if the “wiring 1 dimension” is small, the wiring 1 is open, and if the “wiring 1 dimension” is large, a short circuit occurs in the wiring 1 and a “product pass / fail” failure occurs. Become. Therefore, whether the wiring 1 is open or short can be determined by “wiring 1 dimension” and “product pass / fail”. On the contrary, if the correlation strength of the set of “wiring 1 dimension” and “product pass / fail” is strong, it can be inferred that the opening and short-circuiting of the first wiring as a component is the cause of the failure. Then, it can be considered that the manufacturing process and the manufacturing apparatus of the first wiring are the cause of the failure. In step S3, if the correlation strength of the set of “wiring 1 dimension” and “product pass / fail” is strong in the inspected object storage unit 29, the names of the first wiring and the first wiring as the part names causing the failure are displayed. It is convenient to store a set of “wiring 1 dimension” and “product pass / fail”, and the first wiring of the part name and the manufacturing process of the first wiring so that the manufacturing process can be extracted.

  FIG. 16 is a scatter diagram between items of a set of “number of foreign matters for wiring 1” and “OS start check (product pass / fail)”. As shown in FIG. 16, when the “number of foreign objects in the wiring 1” increases, the wiring 1 is opened, and the OS and the product do not start. Therefore, “OS startup check (product pass / fail)” fails. Therefore, the opening of the wiring 1 can be determined by “the number of foreign objects in the wiring 1” and “OS activation check (product pass / fail)”. On the other hand, if the correlation strength of the set of “number of foreign objects per wiring” and “OS startup check (product pass / fail)” is strong, it can be estimated that the opening of the first wiring as a component is the cause of the failure. Then, it can be considered that the manufacturing process and the manufacturing apparatus of the first wiring are the cause of the failure. In step S3, if the correlation strength of the combination of “number of foreign objects in wiring 1” and “OS activation check (product acceptance / rejection)” is strong in the inspected storage unit 29, the name of the part causing the failure is the first. In order to be able to extract the manufacturing process of the wiring and the first wiring, a set of “the number of foreign objects in wiring 1” and “OS activation check (product pass / fail)”, and the manufacturing process of the first wiring and the first wiring of the part name It is convenient if you associate and remember.

  FIG. 17 is an inter-item scatter diagram of a set of “Wiring 1 resistance” and “OS activation check (product pass / fail)”. As shown in FIG. 17, when “Wiring 1 resistance” increases, a delay occurs in the operation timing, so that “OS startup check (product pass / fail)” fails. Therefore, the delay of the operation timing of the wiring 1 can be determined by “wiring 1 resistance” and “OS activation check (product pass / fail)”. On the contrary, if the correlation strength of the set of “wiring 1 resistance” and “OS start check (product pass / fail)” is strong, it can be estimated that the delay of the operation timing of the first wiring as a component is the cause of the failure. Then, it can be considered that the manufacturing process and the manufacturing apparatus of the first wiring are the cause of the failure. In step S3, if the correlation strength of the set of “wiring 1 resistance” and “OS activation check (product pass / fail)” is strong in the inspected storage unit 29, the first wiring is identified as the name of the component causing the failure. And the first wiring manufacturing process, the set of “wiring 1 resistance” and “OS activation check (product pass / fail)” and the first wiring of the part name and the manufacturing process of the first wiring are related to each other. It is convenient to memorize it.

  Next, returning to FIG. 7, it is assumed that a product defect is detected after step S6. When the occurrence of a product defect is detected by, for example, the completed product inspection device 13, a failure occurrence signal is transmitted to the completed product inspection device 13 in step S7, and the failure occurrence signal is received by the receiving unit 15. Is done. Note that step S7 is not necessarily required.

  In step S8, the receiving unit 15 acquires the product lot number of the defective product from the post-completion product inspection apparatus 13 that has detected the defect. The product lot number need not be acquired by an inspection device such as the product inspection device 13 after completion, but may be input by an operator to the receiving unit 15 (input device).

  In step S <b> 9, the receiving unit 15 acquires the inspection item of the inspection in which the defect is detected, that is, the defect appearance item from the post-completion product inspection apparatus 13 that has detected the defect. Although it is not always necessary to acquire defect appearance items, it is sufficient to select a group including a defect appearance item as a group for performing correlation analysis by acquiring defect appearance items. The number can be reduced.

  In step S10, a strong correlation strength pair detection support method (main process) is performed. Here, a post-main processing is performed for a defective product that has already occurred. By this main processing, it becomes possible to quickly calculate a plurality of sets of correlation strengths using a pair of inspection items as a set, and a set of strong correlation strengths can be detected from the plurality of sets. Details will be described later.

  In step S11, the defect cause item extraction support unit 27 implements a defect cause item extraction support method. Specifically, it assists in extracting defect cause items from inspection items constituting a set of strong correlation strengths.

  In step S12, the defect cause manufacturing apparatus extraction support unit 28 implements a defect cause manufacturing apparatus extraction support method. Specifically, the failure cause manufacturing apparatus in which a minor abnormality detected in the failure cause item is formed is supported using the start history 10.

  Next, details of the strong correlation strength pair detection support method (pre-processing) in step S6 of FIG. 7 will be described. This pre-processing is processing before defect detection. Inspection results are treated as normal data.

  As shown in FIG. 18, first, in step S21, in the receiving unit 15, for each product lot number and for each inspection item, as a test result, a device dimension inspection result 4 (see FIG. 1), a device electrical characteristic test result. 6. Receive product inspection result 8 in manufacturing process, device inspection result 12 after completion, and product inspection result 14 after completion.

  Next, in step S22, the normal data storage unit 33 stores the inspection results 4, 6, 8, 12, and 14 for each inspection item in association with the inspection item as normal data.

  In step S23, the conversion formula creation support unit 17 assists the operator in creating a plurality of data conversion formulas that can convert normal data into data that can be subjected to correlation analysis. Step S23 includes steps S24 to S33, which will be described in order below.

  In step S24, the pair generation unit 41 (see FIG. 2) sequentially reads a pair of inspection items and generates a pair.

  In step S25, the normal data reading unit 42 reads normal data corresponding to each inspection item constituting the set.

  In step S26, the normal inter-item scatter diagram creation unit 43 creates a normal inter-item scatter diagram for each set by a pair of inspection items.

  In step S27, the normal inter-item scatter diagram display unit 44 displays the normal inter-item scatter diagram on the display unit to prompt the operator to convert it.

  In step S28, the error determination reference value setting support unit 45 supports the setting of a determination reference value as to whether or not it is error data as a stage before deleting the error data from the normal data or converting it to a specific numerical value. To do. Then, an error discriminant is set.

  Specifically, the normal inter-item scatter diagram of FIG. 19 is displayed on the screen. It is assumed that the operator determines that data a is data to be subjected to correlation analysis, and data b is error data that should not be subjected to correlation analysis. That is, the correct correlation function d can be obtained only with the data a, but if the data b is also included, an incorrect correlation function c is obtained. The error determination reference value setting support unit 45 prompts the specification of a determination reference value 35 that can determine error data. When the determination reference value 35 is input by the operator through the GUI, the setting of the determination reference value 35 is completed.

  In step S29, the error conversion formula setting support unit 46 supports setting of a conversion formula (conversion method) for error data. Specifically, the error data determined by the determination reference value 35 is deleted or converted into a specific numerical value as shown in FIG.

  In step S30, the character data conversion equation setting support unit 47 supports setting of a conversion equation (conversion method) for converting character data in normal data into numerical data. Specifically, as shown in FIG. 23, when the inspection item B is evaluated with a binary value of pass or fail as in “product pass / fail”, the pass is set to a numerical value of 1. The operator can set the conversion formula so that the value can be converted to zero.

  In step S31, the numerical data conversion equation setting support unit 48 supports setting of a conversion equation (conversion method) for performing code conversion and logarithmic conversion of numerical data of normal data. Specifically, in code conversion, positive and negative signs are exchanged or unified into one of the signs. In the logarithmic conversion, even when the strong correlation strength cannot be obtained as in the natural number notation in FIG. 21, the strong correlation strength may be obtained in the logarithmic notation in FIG. This is because the data group a1 is converted into the data group a2 and the data group b1 is converted into the data group b2.

  In step S32, the normal data converter 49 converts the normal data based on the set data conversion formula, and corrects the normal inter-item scatter diagram. Note that steps S28 to S32 may be repeatedly executed.

  In step S33, the normal inter-item scatter diagram storage unit 50 stores the corrected normal inter-item scatter diagram in association with the set.

  Next, in step S34, the conversion formula storage unit 18 (see FIG. 1) stores the conversion formula in association with the inspection item and the set. Note that one conversion formula is not determined for one inspection item, and different conversion formulas may be set for one inspection item as long as the pairs are different.

  In step S35, the mask database creation support unit 23 (see FIG. 1) supports creation of a set of mask data that is not subject to correlation analysis. Step S35 includes steps S36 to S38.

  In step S36, the normal correlation strength calculator 51 (see FIG. 3) calculates the correlation strength between the inspection items of the normal data converted using the conversion formula.

  In step S37, the correlation strength determination unit 52 determines whether the correlation strength is greater than or less than a threshold value.

  In step S38, the correlation label assigning unit 53 attaches a label of “strong correlation” to a group having a correlation strength equal to or higher than the threshold, attaches a label of “weak correlation” to a group having a correlation strength less than the threshold, and masks Create a database. Specifically, as shown in FIG. 24, the mask database can be expressed as a matrix table in which rows are composed of n rows of inspection items 1 to n and columns are also composed of n columns of inspection items 1 to n. The region where the rows and columns intersect represents a set. The number of sets is n × (n−1) / 2. That is, the region 40 is excluded from the target because the same inspection item set such as the combination of the inspection item n and the inspection item n can be combined, but there is no physical significance in obtaining the correlation strength. In the region 39, the combination of the inspection item n and the inspection item 1 and the combination of the inspection item 1 and the inspection item n in which the order of the inspection items is reversed can be counted as two sets in combination, but the correlation strength is Since it is sufficient to obtain only one of them because they are equal to each other, the other is excluded. Further, since the corresponding region 38 of the mask set whose correlation strength is equal to or greater than the threshold value in step S37 has a strong correlation before the occurrence of the defect, it cannot be used for detection of the defect. Not subject to analysis. Correlation analysis is performed on the mask set corresponding to the finally remaining region 37.

  In step S39, the mask database storage unit 24 (see FIG. 1) stores a mask database in which the labels “not implemented” and “implemented” are associated with the set.

  This completes the preprocessing that should be performed before the failure is detected. Then, as shown in FIG. 25A, it is assumed that a defect is detected such that the personal computer of the product does not operate when the power is turned on. Specifically, in step S7 in FIG. 7, the completed product inspection apparatus 13 transmits a defect occurrence signal, and the receiving unit 15 receives the defect occurrence signal. Note that step S7 is not necessarily required by the post-completion product inspection apparatus 13. Further, as shown in FIG. 25B, it is clarified by the post-completion device inspection apparatus 11 (see FIG. 1) that the defect has occurred in the circuit board in the personal computer.

  In step S8, the receiving unit 15 acquires the product lot number of the defective product from the post-completion product inspection apparatus 13 that has detected the defect. In step S <b> 9, the receiving unit 15 acquires the inspection item of the inspection in which the defect is detected, that is, the defect appearance item from the post-completion product inspection apparatus 13 that has detected the defect.

  Next, details of the strong correlation strength pair detection support method (main process) in step S10 of FIG. 7 will be described. This main process is a process after defect detection. Inspection results 4, 6, 8, 12, and 14 are treated as data including abnormality.

  As shown in FIG. 26, first, in step S41, the receiving unit 15 (see FIG. 1) receives the inspection results 4, 6, 8, 12, and 14 of the inspection items related to the acquired product lot number. To do. Further, the production lot number of the product manufactured before and after the product of this production lot number is also acquired from the start history 10.

  Next, in step S42, the pre-conversion data storage unit 32 stores the inspection results 4, 6, 8, 12, and 14 for each inspection item in association with the inspection item as pre-conversion data.

  In step S43, the data conversion unit 19 (see FIG. 1) reads the pre-conversion data and the data conversion formula corresponding to the inspection item for each set, and converts the pre-conversion data into converted data by the data conversion formula.

  In step S44, the converted data storage unit 20 stores the converted data in association with the inspection item.

  In step S45, the set extraction unit 55 (see FIG. 4) extracts the mask set to be analyzed corresponding to the regions 37 and 38 in FIG. Only the mask set corresponding to the region 37 may be extracted.

  In step S46, the inspection item reading unit 56 reads the inspection items constituting the extracted set.

  In step S47, the converted data reading unit 57 reads the converted data corresponding to the extracted set and the inspection items constituting the set.

  In step S48, the correlation strength calculation unit 58 calculates the correlation strength between the converted data corresponding to the pair of inspection items constituting the extracted set.

  In step S49, the correlation analysis result storage unit 22 (see FIG. 1) stores the calculated correlation strength in association with the corresponding group.

  In step S50, the order assigning unit 59 assigns the order to the set in the order of strong correlation strength.

  In step S51, the output unit 25 outputs a set to a screen display or a form in the order in which the order is given.

  In step S52, if the label “strong correlation” is attached to the group displayed on the screen using the mask database in the output unit 25, the “strong correlation” is also distinguished from the other groups to match the display of the group. indicate. Here, it is considered that the group displayed without the display of “strong correlation” is that a strong correlation has occurred due to the occurrence of a defect. Note that if only the mask set corresponding to the region 37 is extracted in step S45, the strong correlation set can be excluded at that stage.

  Next, details of the defect cause item extraction support method in step S11 of FIG. 7 will be described.

  As shown in FIG. 27, first, in step S61, the defect cause item reading unit 61 reads a pair of inspection items constituting a set in order of the assigned order. The read inspection item becomes a candidate for a defect cause item.

  Next, in step S62, the converted data reading unit 62 reads the converted data corresponding to the set and the read inspection item.

  In step S63, the cause / appearance item scatter diagram creation unit 63 creates a cause / appearance item scatter diagram from the pair of read converted data.

  Of the pair of converted data, one relates to the inspection item in which the defect is detected (appears), and the other is the inspection item that detects the abnormality in the manufacturing process that caused the minor abnormality that caused the defect. It is thought that there is also a case. For example, as shown in FIG. 28, in order to simplify the explanation, the correlation analysis is not performed between all the inspection items, but the inspection items after completion and the inspection items in the manufacturing process, for example, “OS startup check”, “product” Assume that a correlation analysis with “pass / fail” is performed. The inspection items in which defects are detected are, for example, “OS startup check” and “product pass / fail”. Correlation analysis is performed between “OS startup check”, “product pass / fail”, and all inspection items in the manufacturing process. The pairs subjected to correlation analysis are sorted in the order of strong correlation strength, and a scatter diagram between cause and appearance items is generated in descending order of correlation.

  As a result, for example, as shown in FIG. 29, if a strong correlation is obtained between “Wiring 1 resistance” and “OS startup check (product pass / fail)”, a partial wiring 1 pattern 36 as shown in FIG. It is considered that an open occurred due to the high resistance g. On the contrary, as shown in FIG. 30, since a strong correlation is not obtained between the “wiring 1 dimension” and the “OS activation check (product pass / fail)”, the disconnection f of the wiring 1 pattern 36 as shown in FIG. Therefore, it is considered that the “wiring 1 dimension” does not contribute to the defect.

  In step S64, the cause / appearance item scatter diagram display unit 64 displays the cause / appearance item scatter diagram on the screen.

  In step S65, the normal-time item scatter diagram extraction unit 65 extracts the normal-time item scatter diagram corresponding to the set from the normal-time item scatter diagram storage unit 50 of FIG.

  In step S66, the normal scatter diagram between items is displayed in the normal scatter diagram overlap display unit 66 in a superimposed manner on the cause / appearance item scatter diagram. The operator compares the normal item scatter diagram with the cause / appearance item scatter diagram to determine whether the reason why the correlation strength is strong in the cause / appearance item scatter diagram is due to defects. to decide.

  In step S67, the cause-of-failure item designation support unit 67 assists the operator to designate the cause of the cause of failure item as the cause of failure item using a GUI on the screen.

  Finally, the details of the defect cause manufacturing apparatus extraction support method in step S12 of FIG. 7 will be described.

  As shown in FIG. 32, first, in step S71, a start history corresponding to the product lot number of the product in which a defect is detected is indicated in the start history extracting unit 71 (see FIG. 6). Extract from (see).

  Next, in step S72, the start history list display unit 72 displays a list of the extracted start history on the display unit.

  In step S <b> 73, the marker display unit 73 displays a “defect cause item” marker on the inspection history inspection item corresponding to the defect cause item.

  In step S74, in the defect cause manufacturing apparatus designation support unit 75, based on the defect cause item (inspection item) from the inspected object storage unit 29 (see FIG. 1), it is the object to be inspected, that is, the cause of the defect. The part name, the name of the manufacturing process that directly manufactures the part, and the manufacturing apparatus are displayed. The failure cause manufacturing apparatus designation support unit 75 supports the designation of the manufacturing apparatus displayed by the operator as the defect cause manufacturing apparatus using a GUI on the screen.

  According to the embodiment of the present invention, since a conversion equation or the like is prepared before the failure is detected, the correlation analysis can be performed in a short time after the failure is detected. For this reason, a product defect can be reduced by discovering a defect cause item and a defect cause manufacturing apparatus at an early stage. Note that the failure analysis apparatus 30 of the embodiment may be realized by causing a computer to execute a failure analysis program.

It is a lineblock diagram of a failure analysis device concerning an embodiment of the present invention. It is a block diagram of a conversion type | formula creation assistance part. It is a block diagram of a mask database creation assistance part. It is a block diagram of a correlation analysis processing unit. It is a block diagram of a defect cause item extraction assistance part. It is a block diagram of a defect cause manufacturing apparatus extraction assistance part. 3 is a flowchart of a failure analysis method according to an embodiment of the present invention. It is an example of the dispersion | distribution chart between items which shows strong correlation strength. It is an example of the dispersion | distribution chart between items which shows weak correlation strength. It is the combination table | surface (1) of the inspection item for drawing the dispersion | distribution chart between items. It is an inter-item dispersion | distribution figure of the inspection item "wiring 1 dimension" and "wiring 1 resistance." It is the combination table | surface (2) of the inspection item for drawing the dispersion | distribution chart between items. It is a dispersion | distribution diagram between items of the inspection item "wiring 1 dimension" and "a power supply OFF short check (power ON short check)." It is an inter-item dispersion | distribution figure of inspection item "wiring 1 dimension" and "power supply start check". It is an inter-item dispersion | distribution figure of inspection item "wiring 1 dimension" and "product pass / fail". It is an inter-item dispersion diagram of inspection items “number of foreign objects in wiring 1” and “OS startup check (product pass / fail)”. It is an inter-item dispersion diagram of inspection items “Wiring 1 resistance” and “OS startup check (product pass / fail)”. It is a flowchart of the detection assistance method (preprocessing) of the group of strong correlation strength. It is an example of the dispersion | distribution chart between items containing error data. FIG. 20 is an example of an inter-item scatter diagram that is redrawn from the inter-item scatter diagram of FIG. 19 by removing error data. It is an example of the dispersion | distribution chart between items by the numerical data which are not logarithmically described. FIG. 22 is an example of an inter-item dispersion diagram in which numerical data of the inter-item dispersion diagram of FIG. It is an example of an inter-item scatter diagram in which character data (pass / fail) is converted into numerical data (1/0) and redrawn. It is a table | surface (the 1) which shows the structure of a mask database. (A) is a figure which shows the condition of the defect generation | occurrence | production in a product, (b) is a figure which shows that the defect generate | occur | produced in the circuit board. It is a flowchart of the detection assistance method (this process) of the group of strong correlation strength. It is a flowchart of the extraction assistance method of a defect cause item. It is a table | surface (the 2) which shows the structure of a mask database. It is an inter-item dispersion diagram of inspection items “Wiring 1 resistance” and “OS startup check (product pass / fail)”. It is an inter-item dispersion diagram of inspection items “wiring 1 dimension” and “OS startup check (product pass / fail)”. It is a figure which shows that the defect generate | occur | produced with the wiring 1 pattern which comprises a circuit board. It is a flowchart of the extraction assistance method of a defect cause manufacturing apparatus.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Inspection device group in manufacturing process 2 Inspection device group after completion 3 Device dimension inspection apparatus 4 Device dimension inspection result 5 Device electrical property inspection apparatus 6 Device electrical property inspection result 7 Product inspection device in manufacturing process 8 Product inspection result in manufacturing process 9 Construction history creation device 10 Construction history 11 Device inspection device after completion 12 Device inspection result after completion 13 Product inspection device after completion 14 Product inspection result after completion 15 Receiving unit 17 Conversion formula creation support unit 18 Conversion formula storage unit 19 Data conversion unit 20 Converted data storage unit 21 Correlation analysis processing unit 22 Correlation analysis result storage unit 23 Mask database creation support unit 24 Mask database storage unit 25 Output unit 26 Various forms 27 Defect cause item extraction support unit 28 Defect cause manufacturing device extraction support unit 29 Subject Inspection object storage unit 30 Defect analysis device 31 Inspection result storage unit 2 Pre-conversion data storage unit 33 Normal data storage unit 35 Judgment reference value 36 Wiring 1 pattern 41 Pair generation unit 42 Normal data reading unit 43 Normal item scatter diagram creation unit 44 Normal item scatter diagram display unit 45 Error Judgment reference value setting support unit 46 Error conversion formula setting support unit 47 Character data conversion formula setting support unit 48 Numerical data conversion formula setting support unit 49 Normal data conversion unit 50 Normal item scatter diagram storage unit 51 Normal correlation strength calculation Unit 52 Correlation strength determination unit 53 Correlation label assignment unit 55 Set extraction unit 56 Inspection item read unit 57 Converted data read unit 58 Correlation strength calculation unit 59 Order assignment unit 61 Defect cause item read unit 62 Converted data read unit 63 Cause / Scatter chart creation part between appearance items 64 Cause / appearance item scatter chart display part 65 Normal item scatter chart extraction Unit 66 Normal item scatter diagram overlap display unit 67 Failure cause item designation support unit 71 Start history extraction unit 72 Start history list display unit 73 Marker display unit 74 Inspected object display unit 75 Failure cause manufacturing device designation support unit

Claims (8)

In relation to the product lot number for each product and the inspection item for each inspection, the pre-conversion data storage unit that stores the inspection result for each inspection for each product as pre-conversion data,
A set generation unit that generates a set of a plurality of inspection items from a combination of a pair of inspection items;
A data conversion formula for converting a pair of pre-conversion data corresponding to each of the inspection items constituting the set of inspection items into converted data capable of calculating the correlation strength, for each inspection item constituting the corresponding set of inspection items A conversion-type storage unit for storing in relation to
A data conversion unit that converts pre-conversion data into converted data by the data conversion formula for each inspection item that constitutes a set of corresponding inspection items;
A correlation strength calculator for calculating a correlation strength between the converted data for each set of inspection items;
An order assigning unit for assigning an order to the set of inspection items in order of strong correlation strength;
A failure analysis apparatus comprising: a failure cause item reading unit that reads out the inspection items that constitute a set of inspection items in the order from the earlier order as a failure cause item. A mask database creation support unit that supports creation of a mask database that records a set that is not a target of the order as a mask set;
In the order giving unit, do not give the order to the mask set, or
The defect analysis apparatus according to claim 1, wherein the defect cause item reading unit does not read inspection items constituting the mask set as defect cause items.   3. The failure analysis apparatus according to claim 2, wherein the mask database creation support unit sets the set having a correlation strength equal to or greater than a predetermined value as the mask set before the occurrence of a defect in the product. . The data conversion formula includes
An error discriminant for discriminating error data included in the pre-conversion data;
The failure analysis apparatus according to any one of claims 1 to 3, further comprising: an error conversion formula for converting the error data into data capable of calculating a correlation strength. The data conversion formula includes
5. The defect according to claim 1, further comprising: a character data conversion formula that converts character data included in the pre-conversion data into numerical data capable of calculating a correlation strength. Analysis device. The data conversion formula includes
The numerical data conversion formula for converting numerical data included in the pre-conversion data into numerical data that includes at least one of code conversion and logarithmic conversion and that can calculate the correlation strength is included. The defect analysis apparatus according to any one of 5. In relation to the product lot number for each product and the inspection item for each inspection, the inspection result for each inspection for each product is stored as pre-conversion data,
Generate a set of multiple inspection items from a pair of inspection items,
A data conversion formula for converting a pair of pre-conversion data corresponding to each of the inspection items constituting the set of inspection items into converted data capable of calculating the correlation strength, for each inspection item constituting the corresponding set of inspection items Relate to, remember,
For each inspection item constituting the set of corresponding inspection items, the pre-conversion data is converted into converted data by the data conversion formula,
Calculate the correlation strength between the converted data for each set of inspection items,
Give order to the set of inspection items in order of strong correlation strength,
A failure analysis method, wherein inspection items constituting a set of inspection items are read out as failure cause items in the order from the earlier order. A pre-conversion data storage procedure for storing in the computer the product lot number for each product and the inspection item for each inspection, and storing the inspection result for each inspection as the pre-conversion data for each product;
A set generation procedure for generating a set of a plurality of inspection items from a combination of a pair of inspection items;
A data conversion formula for converting a pair of pre-conversion data corresponding to each of the inspection items constituting the set of inspection items into converted data capable of calculating the correlation strength, for each inspection item constituting the corresponding set of inspection items A conversion formula storage procedure for storing in relation to
A data conversion procedure for converting the pre-conversion data into converted data by the data conversion formula for each inspection item constituting a set of corresponding inspection items,
Correlation strength calculation procedure for calculating correlation strength between the converted data for each set of inspection items;
An order assignment procedure for assigning an order to the set of inspection items in order of strong correlation strength;
A failure analysis program for executing a failure cause item reading procedure for reading out inspection items constituting a set of inspection items as a failure cause item in order from the earlier order.

Images