JP5166108B2 - Abnormal factor analysis program, recording medium therefor, and abnormal factor analyzer - Google Patents

Description

  The present invention relates to an abnormality factor analysis program for analyzing, for example, an abnormality factor such as a decrease in yield occurring in a manufacturing process of a semiconductor or the like with high accuracy, a recording medium recording the abnormality factor analysis program, and an abnormality based on the abnormality factor analysis program. The present invention relates to an abnormality factor analyzer that performs factor analysis processing.

  For example, as shown in Patent Documents 1 and 2, etc., in the manufacturing process of a semiconductor or the like, for the purpose of improving the quality of the product or improving the production efficiency, a defect factor such as a defect, a chip, a crack or the like Analyzing abnormal factors as quickly as possible based on the history of devices used in the manufacturing process, test results, design information, various measurement data, etc.

Conventionally, in a semiconductor manufacturing process or the like, a regression tree analysis or the like is performed as a method of anomaly factor analysis in quality control or process control as disclosed in Patent Document 1. In this type of quality control or process control, various management data are collected.

For example, the main data collected in the semiconductor manufacturing process is roughly classified as follows.
・ Manufacturing conditions (manufacturing process conditions such as pressure, temperature, gas flow rate)
・ Characteristic values (Product characteristic values related to semiconductor device characteristics and quality characteristics such as resistance, capacitance, and conductivity)
・ Inspection items (product characteristic values, defect contents, number of defective products, yield, etc. quality control or process control item data)

  As a result of collecting these data, for example, when the yield is significantly reduced, or there are many non-standard values in certain characteristic values, the above quality control or process control method is used to determine the cause of the abnormality. Anomaly factor analysis is performed.

A conventional general anomaly factor analysis procedure will be described. First, when an abnormality occurs in a product, the inspection item data (hereinafter referred to as inspection data) indicating the abnormality of the product group manufactured in the same period as the product (the range of the period is, for example, several weeks to several months). 1) and the characteristic data of the characteristic values of these product groups.
FIGS. 10A to 10C are graphs showing correlation examples of the characteristic data 1 to m with respect to the inspection data 1. Based on these correlation graphs, a correlation coefficient between the inspection data 1 and each characteristic data is calculated, and characteristic data having a stronger correlation with respect to the inspection data 1 than the other is determined for the abnormality. Here, it is assumed that the characteristic data calculated in this way is characteristic data 1.

FIG. 10D to FIG. 10F are graphs showing correlation examples of manufacturing conditions (processing conditions) 1 to k with respect to the characteristic data 1. Based on these correlation graphs, a correlation coefficient between the characteristic data 1 and each processing condition is calculated, and a processing condition having a stronger correlation with the inspection data 1 than the other is determined for the abnormality. The processing conditions thus determined are estimated as the abnormal factor in the conventional abnormal factor analysis.
JP 2006-318263 A JP 2001-306999 A

  In the conventional abnormality factor analysis method described above, the correlation between each characteristic data and the inspection data for each processing condition is examined, and the characteristic data is narrowed down to the processing conditions in order. In order to estimate, the problem that the estimation precision was low had arisen. For example, a specific example of such a problem is as follows by a semiconductor process management example in which “resistance value” is set as characteristic data and “temperature” and “gas flow rate” are set as processing conditions 1 and 2, respectively. As is generally known, due to the physical properties of semiconductors, the resistance value of a device decreases when it is processed at a high temperature, and conversely, the resistance value of a device increases when it is processed at a low temperature. In general, when the treatment is performed in an environment where the gas flow rate is high, the resistance value becomes low, and the resistance value becomes high when the gas flow rate is low. Thus, even if one resistance value is taken, it fluctuates depending on the processing temperature and the used gas flow rate. Therefore, if the cause of the failure is temperature, the resistance value is affected according to the gas flow rate value, so that the correlation between the resistance value and the temperature itself may be unclear or inaccurate. Therefore, the abnormality factor analysis by the conventional estimation method has a low accuracy because it does not consider the influence of each processing condition.

  Moreover, in a general semiconductor manufacturing process, the manufacturing condition data is generally composed of 100 to 200 parameter groups, and the characteristic value data is composed of 20 to 40 parameter groups. There are 20 to 80 inspection data. Therefore, in the situation where there are many processing condition parameters that affect one characteristic data, unless the conventional abnormality factor analysis method is able to grasp in advance how they relate to each other, It was practically impossible to identify. Even if a data map is created three-dimensionally, correlation analysis or identification analysis work by humans has been impossible.

  Therefore, in view of the problem of the conventional abnormality factor analysis technique, the object of the present invention is to extract anomaly factors easily and accurately even under an inspection situation that correlates as a cause parameter such as a large number of processing parameters. The present invention provides an abnormality factor analysis program that can perform the above, an abnormality factor analysis program recording medium that records the abnormality factor analysis program, and an abnormality factor analysis device that can perform highly accurate abnormality factor analysis based on the abnormality factor analysis program.

In the first aspect of the present invention, a product group manufactured with at least n cause parameters P _k (1 ≦ k ≦ n) is inspected with a specific analysis parameter D, and the analysis parameter D is abnormal. Is selected from the cause parameter P _k (1 ≦ k ≦ n) in an abnormality factor analysis program for extracting the cause of the abnormality from the cause parameter P _k (1 ≦ k ≦ n). Forming two three-dimensional data groups with two different cause parameters P _i and P _j as the X-axis and Y-axis, and the analysis parameter D as the Z-axis, and the N three-dimensional data groups a step of configuring said analysis parameter D is N ₁ pieces of normal two-dimensional data set projected onto the X-Y plane by extracting normal three-dimensional data set in the normal range, from the N 3D data set The analysis parameter Extracting an abnormal three-dimensional data group in which the parameter D is in an abnormal range and projecting it onto the XY plane to form N _two abnormal two-dimensional data groups; the normal two-dimensional data group and the abnormal two-dimensional data A step of deriving an overlapping degree S _ij of the data group, a step of creating a network diagram in which the cause parameters P _i and P _j are connected using a plurality of the overlapping degrees S _ij , and concentration from the network diagram An abnormality factor analysis program including a step of deriving a cause hub parameter having a high degree and extracting the cause parameter _Pk causing the abnormality.

According to a second aspect of the present invention, in the first aspect, n (n−1) / 2 overlapping degrees S _ij over a range of 1 ≦ i ≦ n−1 and i <j ≦ n are set. A step of deriving, a step of classifying the overlapping degree S _ij in ascending or descending order, and a cause parameter P _i corresponding to the overlapping degree S _ij by using the overlapping degree S _ij from a minimum value to a desired rank value. by connecting the P _j is abnormal factor analysis program including a step of creating a network diagram.

According to a third aspect of the present invention, in the first or second aspect, the X-axis and the Y-axis are equally divided into integers obtained by converting N1 ^{/ 2} into integers, and the XY plane is divided. Forming a lattice, projecting a three-dimensional data group into the lattice mass to form the normal two-dimensional data group and the abnormal two-dimensional data group, and the normal two-dimensional data group and the abnormal two-dimensional data group are the same common number proportion of deriving a number of overlapping N ₁ ∩N ₂ that the sum for all the grid squares, the number of overlapping N ₁ ∩N to the total number N ₂ of the data if present in the lattice squares (N ₁ ∩N ₂ ) An abnormality factor analysis program including a step of deriving / N (hereinafter referred to as cosine measure) as the overlap degree _Sij .

According to a fourth aspect of the present invention, in the first, second, or third aspect, the analysis parameter D is an inspection parameter, and the cause parameter _Pk evaluates a processing condition parameter of the product or the processing condition. This is an abnormality factor analysis program that is a quality evaluation parameter for the purpose of

According to a fifth aspect of the present invention, in the first, second or third aspect, the analysis parameter D is a quality evaluation parameter for evaluating a processing condition of a product, and the cause parameter P _k is the processing This is an abnormal factor analysis program that is a condition parameter.

According to the sixth aspect of the present invention, a product group manufactured with at least n cause parameters P _k (1 ≦ k ≦ n) is inspected with a specific analysis parameter D, and the analysis parameter D is abnormal. , The cause parameter P _k (1 ≦ k ≦ n) is stored in a recording medium storing an abnormality factor analysis program for extracting the cause of the abnormality from the cause parameter P _k (1 ≦ k ≦ n). 2) forming two three-dimensional data groups with two different cause parameter parameters P _i and P _j selected from X and Y as the analysis parameter D and the Z axis, respectively, a step of configuring said analysis parameter D is N ₁ pieces of normal two-dimensional data set projected onto the X-Y plane by extracting normal three-dimensional data set in the normal range of pieces of three-dimensional data set, the N 3D Extracting an abnormal three-dimensional data group in which the analysis parameter D is in an abnormal range from the data group and forming N ₂ abnormal two-dimensional data groups projected on the XY plane; and the normal two-dimensional data group _Deriving an overlapping degree S _ij of the abnormal two-dimensional data group, and creating a network diagram connecting the cause parameters P _i , P _j using a plurality of the overlapping degrees S _ij ; An abnormality factor analysis program recording medium storing an abnormality factor analysis program having a step of deriving a cause hub parameter having a high degree of concentration from the network diagram and extracting the cause parameter _Pk causing the abnormality.

According to a seventh aspect of the present invention, in the sixth aspect, n (n−1) / 2 overlapping degrees S _ij over a range of 1 ≦ i ≦ n−1 and i <j ≦ n are set. A step of deriving, a step of classifying the overlapping degree S _ij in ascending or descending order, and a cause parameter P _i corresponding to the overlapping degree S _ij by using the overlapping degree S _ij from a minimum value to a desired rank value. by connecting the P _j is abnormal factor analysis program storage medium comprising the step of creating a network diagram.

According to an eighth aspect of the present invention, in the sixth or seventh aspect, the X axis and the Y axis are equally divided into an integer number obtained by converting N ^1/2 into an integer, and the XY plane is divided. Forming a lattice, projecting a three-dimensional data group into the lattice mass to form the normal two-dimensional data group and the abnormal two-dimensional data group, and the normal two-dimensional data group and the abnormal two-dimensional data group are the same common number proportion of deriving a number of overlapping N ₁ ∩N ₂ that the sum for all the grid squares, the number of overlapping N ₁ ∩N to the total number N ₂ of the data if present in the lattice squares (N ₁ ∩N ₂ ) An abnormality factor analysis program recording medium including a step of deriving / N (hereinafter referred to as cosine measure) as the overlapping degree _Sij .

According to a ninth aspect of the present invention, in the sixth, seventh, or eighth aspect, the analysis parameter D is an inspection parameter, and the cause parameter _Pk evaluates a processing condition parameter of the product or the processing condition. It is an abnormality factor analysis program recording medium that is a quality evaluation parameter for the purpose.

According to a tenth aspect of the present invention, in the sixth, seventh or eighth aspect, the analysis parameter D is a quality evaluation parameter for evaluating a processing condition of a product, and the cause parameter P _k is the processing parameter. It is an abnormality factor analysis program recording medium that is a condition parameter.

In an eleventh aspect of the present invention, a product group manufactured with at least n cause parameters P _k (1 ≦ k ≦ n) is inspected with a specific analysis parameter D, and the analysis parameter D is abnormal. In the abnormality factor analyzer that extracts an abnormality factor causing the abnormality from the cause parameter P _k (1 ≦ k ≦ n) and outputs an analysis result, the cause P _k (1 ≦ k ≦ n) 2) means for configuring N three-dimensional data groups with two different cause parameters P _i and P _j selected from X and Y as the analysis parameters D and Z as the axes, respectively, from the three-dimensional data set and means for configuring said analysis parameter D is N ₁ pieces of normal two-dimensional data set projected onto the X-Y plane by extracting normal three-dimensional data set in the normal range, the N of From the 3D data group, Means for extracting an abnormal three-dimensional data group in which the meter D is in an abnormal range and projecting it onto the XY plane to form N _two abnormal two-dimensional data groups, the normal two-dimensional data group and the abnormal two-dimensional data Means for deriving the degree of overlap S _ij of the data group, means for creating a network diagram connecting the cause parameters P _i , P _j using a plurality of the degree of overlap S _ij , and concentration from the network diagram It is an abnormality factor analyzer having means for deriving a cause hub parameter having a high degree and extracting the cause parameter _Pk causing the abnormality.

In a twelfth aspect of the present invention, in the eleventh aspect, n (n−1) / 2 overlapping degrees S _ij over a range of 1 ≦ i ≦ n−1 and i <j ≦ n are set. Means for deriving; means for classifying the overlapping degree S _ij in ascending or descending order; and using the overlapping degree S _ij from a minimum value to a desired rank value, cause parameters P _i corresponding to the overlapping degree S _ij , This is an abnormality factor analyzing apparatus including means for connecting P _j and creating a network diagram.

According to a thirteenth aspect of the present invention, in the eleventh or twelfth aspect, the X-Y plane is divided by dividing the X axis and the Y axis into integer numbers obtained by integerizing N ^1/2. Means for forming a normal two-dimensional data group and the abnormal two-dimensional data group by projecting a three-dimensional data group into the lattice mass, and the normal two-dimensional data group and the abnormal two-dimensional data group are the same wherein the number of overlapping N ₁ ratio of ∩N ₂ common number and means for deriving the overlap number N ₁ ∩N ₂ summation on all the grid squares, to the total number N of data when present in the lattice squares (N ₁ ∩N ₂ ) An abnormality factor analyzer including means for deriving / N (hereinafter referred to as a cosine measure) as the degree of overlap _Sij .

According to a fourteenth aspect of the present invention, in the eleventh, twelfth or thirteenth aspect, the analysis parameter D is an inspection parameter, and the cause parameter _Pk is an evaluation of a processing condition parameter of the product or the processing condition. This is an abnormality factor analysis device that is a quality evaluation parameter for the purpose.

In a fifteenth aspect of the present invention, in the eleventh, twelfth or thirteenth aspect, the analysis parameter D is a quality evaluation parameter for evaluating a processing condition of a product, and the cause parameter _Pk is the process It is an abnormality factor analyzer which is a condition parameter.

According to the processing procedure of the abnormality factor analysis program according to the first aspect of the present invention, a normal three-dimensional data group in which the analysis parameter D is in a normal range is extracted from the N three-dimensional data groups, and XY N ₁ normal two-dimensional data groups projected onto a plane are formed, and an abnormal three-dimensional data group in which the inspection parameter D is in an abnormal range is extracted from the N three-dimensional data groups, and the XY plane is extracted. A network in which N _two abnormal two-dimensional data groups projected on the network are formed, the degree of overlap S _ij between the normal two-dimensional data group and the abnormal two-dimensional data group is derived, and the cause parameters P _i and P _j are connected. creating a figure so derive the higher cause hub parameters of concentration level from said network diagram extracting the cause parameter P _k to rise to the abnormal, correlations between the plurality of cause parameter P _k Exhaustive, it is possible to perform certain of the cause parameter relating to the abnormal factor. Therefore, in this embodiment, the abnormal factor is extracted with high accuracy without eliminating the influence of the composite parameter that is selected by the conventional factor analysis method, and the accuracy of the abnormal factor analysis in quality control or process management is improved. Can be planned.

As the analysis parameter D, it is possible to use inspection parameters relating to the group characteristics and the product characteristics of the final product. Group characteristics include yield, number of products, product outline, number of defective categories, and the like. Product characteristics include, for example, diode characteristics (current and voltage) if the final product is a diode, resistance values if a resistor, transistor characteristics if a transistor, and IC characteristics if an IC. The cause parameter P _k is the above-described manufacturing condition (1), characteristic value (2), or the like in the case of a semiconductor manufacturing process.
Further, as the analysis parameter D and the cause parameter P _k , a quality evaluation parameter for evaluating a processing condition of a product can be used. A typical quality evaluation parameter is a TEG parameter (Test Element Group Data) used in a semiconductor or the like. TEG parameters are determined by placing evaluation elements such as resistors (R), capacitors (C), transistors (Tr), wiring patterns, etc. in the device during the manufacturing process. It is used to measure current, electric capacity, transistor characteristics (voltage amplification factor, current amplification factor, etc.) or disconnection state. The measured values of these evaluation elements are not product characteristics, but if the product characteristics are abnormal, they indicate an abnormal or defective state, and serve as an index for evaluation of processes and the like. In addition to using the TEG parameter as the analysis parameter D, when the inspection parameter is used as the analysis parameter D, the TEG parameter can be used as the cause parameter P _k to determine an abnormal factor.

The degree of overlap S _ij can be derived by a common area discrimination process for discriminating from the number of the normal two-dimensional data group and the abnormal two-dimensional data group existing in the common area (section) on the XY plane. . Alternatively, the derivation is performed by converting each data of the normal two-dimensional data group and the abnormal two-dimensional data group from a predetermined base point on the XY plane into vector data in which the inclination and size are determined, and the vector inclination And vector discrimination processing for discriminating vector data having the same size.

According to the second aspect of the present invention, n (n−1) / 2 overlapping degrees S _ij are derived over a range of 1 ≦ i ≦ n−1 and i <j ≦ n, and the overlapping The degree S _ij is classified in ascending order or descending order, the overlapping degree S _ij is used from the minimum value to the desired rank value, and the cause parameters P _i and P _j corresponding to the overlapping degree S _ij are connected to form a network diagram. Since the network diagram is created from the combination of all the two cause parameters P _i and P _j , the cause hub parameters having a high degree of concentration can be derived with high accuracy, thereby improving the accuracy of the abnormal cause analysis. Can do.

According to the third aspect, the X-axis and the Y-axis are equally divided into integers obtained by converting N1 ^{/ 2} into integers, and the XY plane is latticed, and the three-dimensional pattern is formed in the lattice mass. When the normal two-dimensional data group and the abnormal two-dimensional data group are configured in a matrix data array by projecting a data group, and the normal two-dimensional data group and the abnormal two-dimensional data group exist in the same lattice mass the common number of data to derive a number of overlapping _{N 1} ∩N ₂ that the sum for all the grid squares in the overlap ratio of the number _{N 1} ∩N ₂ to the total number _{N (N 1 ∩N 2) /} N ( cosine measure) _Is derived as the degree of overlap S _ij , it is possible to easily realize an abnormality factor analysis program capable of extracting an abnormality factor with high accuracy only by matrix arrangement processing of data groups and cosine measure calculation processing. The advantage of using the cosine measure is that the cosine measure is smaller as the overlap between the distribution of normal data and abnormal data is smaller among all the cause parameters, so the combination of the cause parameters P _i and P _j having a smaller cosine measure is The difference between normal and abnormal distribution is large, and the ability to discriminate between normal and abnormal is increased. From this advantage, by arranging the cosine measures obtained for the two-dimensional data P _ij corresponding to the combination of the cause parameters P _i and P _j in ascending order, the cause parameter that causes the abnormality can be easily and reliably determined. .

According to the fourth aspect of the present invention, the analysis parameter D is the inspection parameter, and the cause parameter _Pk is a product processing condition parameter or the quality evaluation parameter. The processing condition parameter of the factor can be directly derived, or abnormal processing conditions can be indirectly derived from the quality evaluation parameter.
In order to obtain the processing condition parameters as gas pressure (P), gas flow rate (Q), temperature (T), for example, in a semiconductor factory, a pressure sensor, a mass flow meter, a thermometer, etc. are installed in the piping of a manufacturing facility. However, the equipment cost increases to detect these parameters. According to the present embodiment, by using the TEG parameter as the quality evaluation parameter as the cause parameter P _k , an abnormal processing condition parameter can be extracted from the relationship between the TEG parameter and the processing condition parameter. By grasping the relationship between parameters and processing condition parameters in advance, it is possible to extract abnormal processing condition parameters without always performing measurement using the above sensors, etc., which contributes to reducing the cost of equipment such as sensors. .

In the fifth aspect of the present invention, since the analysis parameter D is the quality evaluation parameter and the cause parameter P _k is the processing condition parameter, the TEG parameter as the quality evaluation parameter is used for the analysis parameter D. Accordingly, an abnormal processing condition parameter can be extracted from the relationship between the TEG parameter and the processing condition parameter.

According to the sixth aspect of the present invention, there is provided a recording medium storing the abnormality factor analysis program according to the first aspect. Therefore, the recording medium according to the present embodiment has all the program effects described in the first embodiment.
As a recording medium in the present invention, a computer-readable recording medium such as a flexible disk, a magnetic disk, an optical disk, a CD, an MO, a DVD, and a hard disk is selected. By installing the recording medium according to this embodiment in a computer, the computer can be operated to analyze an abnormal factor, and an abnormal factor analyzing program can be executed in many individuals, homes, and businesses.

  According to the seventh aspect of the present invention, there is provided a recording medium storing the abnormality factor analysis program according to the second aspect. Therefore, the recording medium according to the present embodiment has all the program effects described in the second embodiment. By installing the recording medium according to this embodiment in a computer, the computer can be operated to analyze an abnormal factor, and an abnormal factor analyzing program can be executed in many individuals, homes, and businesses.

  According to the eighth aspect of the present invention, there is provided a recording medium storing the abnormality factor analysis program according to the third aspect. Therefore, the recording medium according to the present embodiment has all the program effects described in the third embodiment. By installing the recording medium according to this embodiment in a computer, the computer can be operated to analyze an abnormal factor, and an abnormal factor analyzing program can be executed in many individuals, homes, and businesses.

  According to the ninth aspect of the present invention, there is provided a recording medium storing the abnormality factor analysis program according to the fourth aspect. Therefore, the recording medium according to the present embodiment has all the program effects described in the fourth embodiment.

  According to the tenth aspect of the present invention, there is provided a recording medium storing the abnormality factor analysis program according to the fifth aspect. Therefore, the recording medium according to the present embodiment has all the program effects described in the fifth embodiment.

  The abnormality factor analysis device according to the eleventh aspect of the present invention includes not only a computer in which the abnormality factor analysis program according to the first aspect is installed from the outside, but also all electronic devices incorporating this computer function. The Therefore, the abnormality factor analysis device according to the present embodiment has all the program effects described in the first embodiment.

  The abnormality factor analysis device according to the twelfth aspect of the present invention includes not only a computer in which the abnormality factor analysis program according to the second aspect is installed from the outside, but also all electronic devices incorporating this computer function. The Therefore, the abnormality factor analysis apparatus according to the present embodiment has all the program effects described in the second embodiment.

  The abnormality factor analysis apparatus according to the thirteenth aspect of the present invention includes not only a computer in which the abnormality factor analysis program according to the third aspect is installed from the outside, but also all electronic devices incorporating this computer function. The Therefore, the abnormality factor analysis device according to the present embodiment has all the program effects described in the third embodiment.

  The abnormality factor analysis device according to the fourteenth aspect of the present invention includes not only a computer in which the abnormality factor analysis program according to the fourth aspect is installed from the outside, but also all electronic devices incorporating this computer function. The Therefore, the abnormality factor analysis device according to the present embodiment has all the program effects described in the fourth embodiment.

  The abnormality factor analysis apparatus according to the fifteenth aspect of the present invention includes not only a computer in which the abnormality factor analysis program according to the fifth aspect is installed from the outside, but also all electronic devices incorporating this computer function. The Therefore, the abnormality factor analyzer according to this embodiment has all the program effects described in the fifth embodiment.

  An abnormality factor analyzer according to an embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 shows a schematic configuration of an abnormality factor analyzer of the present embodiment. This abnormality factor analysis apparatus inspects a product group (semiconductor product group) manufactured by at least n cause parameters P _k (1 ≦ k ≦ n) with a specific analysis parameter D, and the analysis parameter D is abnormal. The indicated abnormal factor is calculated and extracted from the cause parameter _Pk , and the analysis result is output to the printing device (printer 14) and the display device (display 12).

  The abnormality factor analysis apparatus according to the present embodiment includes a control unit 1 that performs an abnormality factor analysis based on the abnormality factor analysis program according to the present invention. The control unit 1 includes a CPU 2 composed of a microprocessor, and a program execution unit that executes a built-in abnormality factor analysis program under the control of the CPU 2. The control unit 1 is connected to external management devices 15A to 15N such as a production management system management device, various semiconductor manufacturing devices, and product inspection devices so that data communication is possible. The control unit 1 includes an interface unit 10 for registering and storing data transferred from the external management devices 15A to 15N as a file or a database, and a file storage unit for storing registration data of the interface unit 10 in the program search file. 11 is connected. The control unit 1 includes a search condition input unit 13 including a key input device for inputting data search conditions for factor analysis, such as analysis parameters D and inspection target parameters, a printer 14 for printing analysis results, and analysis information. Is connected to the display 12.

The program execution unit includes a data search unit 3 that retrieves data matching the search condition from the file storage unit 11 based on the search condition input by the search condition input unit 13, and every combination of parameters from the searched data. The data is divided into N ^1/2 integers of the total number of data N and the number of data for each cell is calculated, and the data processed in the data processing unit 4 is used for each combination. The cosine measure calculation unit 5 is arranged to calculate the cosine measure and arrange them in ascending order of the values. The control unit 1 is connected to an external output unit 6 that outputs data to the outside such as the printer 14 and the display 12. The external output unit 6 includes a screen setting search condition setting unit 7 of the display 12 that is used when data search conditions are input by the search condition input unit 13, and the search result of the factor analysis process in a spreadsheet format or a graph format. The search result processing unit 8 for display and the cosine measure calculation unit 5 calculate the arrangement position by using each parameter arranged in ascending order, that is, in ascending order, as a network diagram, and create a network diagram with the desired number of parameters A network processing unit 9 is included.

The processing procedure of the abnormality factor analysis program of this embodiment is performed by analyzing two different cause parameters P _i and P _j selected from the cause parameters P _k (1 ≦ k ≦ n), respectively, as the X axis and the Y axis. Step A forming N three-dimensional data groups with the parameter D as the Z axis, and extracting a normal three-dimensional data group in which the analysis parameter D is in the normal range from the N three-dimensional data groups on the XY plane Step B constituting the projected N ₁ normal 2D data group, and an abnormal 3D data group in which the analysis parameter D is in the abnormal range are extracted from the N 3D data groups and projected onto the XY plane. step D to derive a step C that constitutes the N ₂ pieces of abnormality two-dimensional data set, the overlapping degree S _ij of the normal two dimensional data set and abnormal 2D data set, by using a plurality of overlapping degree S _ij , Processing condition parameters _i, comprises the step F of extracting the steps E to create a network diagram and connect the P _j, the cause parameter P _k which derives high causes hub parameter of the degree of concentration from a network diagram occurring abnormality.

2 and 3 show details of the factor analysis processing based on the abnormal factor analysis program built in the control unit 1. FIG.
First, as an initial setting operation, the setting of the processing condition parameter P _k and the inspection parameter D is key-input by the search condition input unit 13 (step S1).
FIG. 11 shows details of the search condition input process. In the present embodiment, an inspection parameter or a TEG parameter can be used as the analysis parameter D, and any input analysis parameter D is set and stored (steps S40 to S43). Further, as the cause parameter P _k , a product processing condition parameter or a TEG parameter can be used, and any one of the input cause parameters P _k is set and stored (steps S44 to S47). TEG parameters are determined by placing evaluation elements such as resistors (R), capacitors (C), transistors (Tr), wiring patterns, etc. in the device during the manufacturing process. It is a parameter that is used to measure current, electric capacity, transistor characteristics (voltage amplification factor, current amplification factor, etc.), disconnection state, etc. in combination.

In the present embodiment, a description will be given of an example in which the inspection parameter is selected as the analysis parameter D and the processing condition parameter is selected as the cause parameter _Pk .
The processing condition parameter _Pk and the inspection parameter D are the same as those in the data group described above. For example, the processing condition parameter P _k includes manufacturing condition (manufacturing process conditions such as pressure, temperature, gas flow rate, etc.) data and characteristic values (resistance, capacitance, product characteristics related to quality characteristics such as semiconductor device characteristics such as conductivity). ) Consists of data. The inspection parameter D includes inspection item (product characteristic value, defect content, number of defective products, yield management and other quality control or process management item data) data.

  When the start of factor analysis is keyed by a key operation of a start key (not shown) of the search condition input unit 13, it is confirmed whether or not the data set by the search condition input unit 13 is stored (step) S2, S3). When the setting data is not stored, data registration is performed (step S5). The data registration is performed by communicating with the external management devices 15A to 15N, receiving data transfer from the external management devices 15A to 15N, and storing the data in the interface unit 10.

If the search data has already been stored, the registration data of the interface unit 10 is created according to the set condition, and a search data set file Dt is created and stored in the file storage unit 11 (step S4). Next, a three-dimensional plot process of data (step S6) is performed on the file F _D (P _i , P _j ). In the three-dimensional plot processing, two different processing condition parameters P _i and P _j selected from the processing condition parameters (cause parameters) P _k (1 ≦ k ≦ n) are set as the X axis and the Y axis, respectively. File data composed of N three-dimensional data groups having the parameter (analysis parameter) D as the Z axis is created. FIG. 4 schematically shows an example of a three-dimensional data group. In FIG. 4, the data group corresponding to the two processing condition parameters P _i and P _j with respect to the inspection parameter D is indicated by a circle.

  After the creation of the three-dimensional data group, the normal or abnormal data group is separated and extracted based on the inspection parameter D (step S7). Separation of the normal or abnormal data group is performed, for example, by extracting an abnormal three-dimensional data group having the inspection parameter D in the abnormal range Ds from N three-dimensional data groups as shown in FIG. In FIG. 4, the abnormal three-dimensional data group is indicated by black circles, and the normal three-dimensional data group is indicated by white circles.

The normal / abnormal three-dimensional data group extracted in step S7 is subjected to projection processing that constitutes N ₁ normal two-dimensional data groups and N ₂ abnormal two-dimensional data groups projected onto the XY plane, respectively. (Step S8). FIG. 5 shows combinations of processing condition parameters that are converted into two-dimensional data. P _ij represents a combination of the processing condition parameters P _i and P _j . For the purpose of analyzing abnormal factors, it is not necessary to convert all the two combinations of processing conditions P _k (1 ≦ k ≦ n) into two-dimensional data, and P _ij = P _ji , and processing condition parameters P _i , P _j among the combinations of _j, excluding the combination of i = j such P _11, P _22, 2-dimensional data of the half of the total combination of the processing condition parameters is performed. In other words, n (n−1) / 2 two-dimensional data groups are created over the range of 1 ≦ i ≦ n−1 and i <j ≦ n.

FIG. 6 schematically shows the two-dimensional data subjected to the projection processing. In FIG. 6, as in FIG. 4, the abnormal data group is indicated by a black circle, and the normal data group is indicated by a white circle. For example, 2-dimensional data group P ₁₃ of the processing conditions P ₁ and the processing conditions P ₃ are abnormal data in the central portion as compared to the 2-dimensional data group P _1n treatment conditions P ₁ and the processing condition P _n is gathered Distribution state.

When the two-dimensional data processing is completed for all search data, data group processing for normalizing the two-dimensional data group is performed (steps S9 and S10). In this data group processing, the X-axis and the Y-axis are equally divided into integers obtained by converting N1 ^{/ 2} into integers, and the XY plane is gridded to normalize the total number N of data. Is called.

FIG. 7 shows an example of a matrix arrangement in which data is rearranged with the unit partition widths of the X-axis and Y-axis subjected to the integer processing. (7A) in FIG. 7 shows an example of an array of abnormal two-dimensional data groups, and (7B) shows an example of an array of normal two-dimensional data groups. In FIG. 7, the size of each partition width (R1, R2,..., Rα-1, Rα) on the horizontal axis or the vertical axis is an integer equal division value obtained by rounding off N ^1/2 .

With respect to the normal two-dimensional data group and the abnormal two-dimensional data group arranged in the matrix, an overlapping number N ₁ ∩N ₂ is derived by summing up the common number of data for all lattice masses when they exist in the same lattice mass (step S11). In FIG. 7, for example, one square (R2, R3) has one normal data (see 7B) and five abnormal data (see 7A). The number N ₁ ∩N ₂ is 6. Using the number of overlaps N ₁ ∩N ₂ , the degree of overlap S _ij between the normal two-dimensional data group and the abnormal two-dimensional data group is derived for all data (steps S12 and S13).

FIG. 8 shows a process for deriving the overlapping degree _Sij . (8A) of FIG. 8 shows a case where (N ₁ ∩N ₂ ) / N (cosine measure) is obtained from the ratio of the number of overlaps N ₁ ∩N ₂ . In this case, the ratio (N ₁ ∩N ₂ ) / N (cosine measure) of the overlap number N ₁ １N ₂ with respect to the total number N is derived as the overlap degree S _ij (step S20). Anomalous factors can be extracted with high accuracy only by the cosine measure calculation process.
After calculating the cosine measure, the measures are classified in ascending order (step S21), and the processing conditions P _i and P _j corresponding to the cosine measure are connected using the minimum value to the desired rank value, and the network diagram is obtained. A process condition parameter that is created and becomes an abnormal factor is evaluated (step S14).

Of all the parameters, the smaller the overlap between the distribution of normal data and abnormal data, the smaller the cosine measure. In other words, the combination of the processing condition parameters P _i and P _{j having} a smaller cosine measure increases the difference between normal and abnormal distributions, and increases the ability to discriminate between normal and abnormal. Therefore, by arranging the cosine measures obtained for the two-dimensional data P _ij corresponding to the combination of the processing condition parameters P _i and P _j in ascending order, it is possible to determine the processing condition parameter that causes an abnormality. For this discrimination process, it is preferable to create a network diagram, and a processing condition parameter having a high degree of concentration can be clearly determined from the number of connections.

Derivation of the overlapping degree S _ij is performed by a common area discriminating process that discriminates from the number of normal two-dimensional data group and abnormal two-dimensional data group existing in the common area (section) in the XY plane as in the cosine measure. However, it is also possible to perform vector discrimination processing by determining vector data having the same inclination and size from the predetermined base point on the XY plane, and determining vector data having the same vector inclination and size. .
(8B) of FIG. 8 shows the derivation process of the overlapping degree _Sij by the vector discrimination process. After data processing (step S10) for the normal two-dimensional data group and the abnormal two-dimensional data group, or at that time, vector data is formed by determining the inclination and size of each data from a predetermined base point on the XY plane. Is performed (step S30). This vector value (vector inclination and magnitude) discriminates the vector data existing in the same lattice of the matrix lattice and the lattice within the allowable neighborhood range, and derives the sum Nv of the common number (step S30). In order to create a network diagram, the total Nv data is arranged in ascending order (step S30), and the process proceeds to a network diagram creation process (step S14).

  FIG. 9 shows an example of a procedure for creating a network diagram when arranged in ascending order of cosine measure. (9A) of FIG. 9 shows a connection state of the processing condition F1 (gas flow rate condition 1) and the processing condition T2 (temperature condition 2), which is a combination of the processing condition parameters having the smallest cosine measure. Next, when the combination of the processing condition parameters with the smallest cosine measure is the processing condition F1 (gas flow condition 1) and the processing condition W1 (weight condition 1), as shown in (9B), the connection state connected to the processing condition F1 Migrate to Hereinafter, the connection is continued by the parameter correlation that appears in ascending order of the cosine measure. (9C) shows a network diagram obtained by continuing the connection up to an arbitrary rank. The processing condition parameter F1 and the processing condition T1 (temperature condition 1) having a high degree of concentration can be clearly searched and determined as processing condition hub parameters from the number of connections in the network diagram of (9C). Whether to create a network diagram up to which stage may be arbitrarily determined in ascending order in advance, but the number of connections with a high degree of concentration is determined to a predetermined value, and the network diagram is created until it appears You may do it.

In the abnormality factor analyzer according to the present embodiment, the degree of overlap S _ij (cosine measure) between N ₁ normal 2D data groups and N ₂ abnormal 2D data groups is derived, and the processing condition parameters P _i , A network diagram in which P _j is connected is created, and a processing condition hub parameter having a high degree of concentration is derived from the network diagram. Thus, the correlation among a plurality of processing condition parameters P _k is covered in the derivation process, and the abnormality Process condition parameters related to factors can be specified, and abnormal factors can be extracted with high accuracy without excluding the influence of complex parameters that would be selected by conventional factor analysis methods. The accuracy of abnormality factor analysis can be improved.

  When the processing condition hub parameter is retrieved, data is output to the printer 14 and the display 12. The network diagram is also output to the display 12 as appropriate, and the monitor diagram is output so that it can be visually judged. The network diagram can be formed in a form that is freely diffused in a plane, a multi-linearity placed at the top and connected in a central spiral, a connection on a spherical surface, a multi-linearity Can be used for the convenience of visual recognition by the operator, using a form in which the vertices are arranged as vertices and connected in a central circular shape, a form in which multiplicity is arranged as the vertices and connected in a central grid.

  It should be noted that the present invention is not limited to the above-described embodiment, and various modifications, design changes and the like within the scope not departing from the technical idea of the present invention are included in the technical scope. Nor.

  According to the present invention, in a wide range of industrial fields such as semiconductors, electronic devices, foodstuffs, etc., it is possible to improve the analysis accuracy of abnormal factors in manufactured products, etc., and to improve production efficiency, production management capability, product quality, etc. Can contribute.

It is a schematic block diagram of the abnormality factor analyzer which is one Embodiment of this invention. It is a flowchart of the abnormal factor analysis process of the embodiment. 3 is a flowchart subsequent to the abnormality factor analysis processing of FIG. 2. It is a figure which shows typically an example of the three-dimensional data group used for the said abnormality factor analysis process. It is a figure which shows the combination table | surface of the process condition parameter converted into two-dimensional data from the said three-dimensional data group. It is a figure which shows typically the two-dimensional data by which the projection process was carried out. It is a figure which shows the example of a matrix arrangement | sequence which rearranged data by the unit division width | variety of the X-axis and Y-axis which performed the integer-ized process. It is a figure which shows the derivation _| leading-out process of overlap degree _Sij . It is a figure for demonstrating the creation process of a network diagram. It is a graph which shows the example of a correlation of each characteristic data 1-m with respect to the test | inspection data 1, or manufacturing conditions 1-k. It is a flowchart which shows the detail of a search condition input process.

Explanation of symbols

1 Control unit 2 CPU
3 Data search unit 4 Data processing unit 5 Cosine measure calculation unit 6 External output unit 7 Search condition setting unit 8 Search result processing unit 9 Network processing unit 10 Interface unit 11 File storage unit 12 Display 13 Search condition input unit 14 Printers 15A to 15N External management device

Claims (15)

When a group of products manufactured with at least n cause parameters P _k (1 ≦ k ≦ n) is inspected with a specific analysis parameter D, the abnormality occurs when the analysis parameter D indicates an abnormality. In an abnormality factor analysis program for causing a computer to execute an abnormality factor analysis for extracting an abnormality factor to be extracted from the cause parameter P _k (1 ≦ k ≦ n), the computer stores the cause parameter P _k and the analysis parameter D. Two different cause parameters P _i and P _j selected from the file storage means and the cause parameters P _k (1 ≦ k ≦ n) extracted from the file storage means are set as the X axis and the Y axis, respectively. the analysis parameter D retrieved from said file storage means and means for configuring the N pieces of three-dimensional data set as a Z axis, the N 3D Means for configuring the analysis parameter D is N ₁ pieces of normal two-dimensional data group to extract the normal three-dimensional data set projected onto the X-Y plane in the normal range of the data group, the N 3D data means for configuring the analysis parameter D is abnormal three-dimensional data set extracted by the _two N projected onto the X-Y plane abnormal two-dimensional data set in the abnormal range of the group, and the normal two-dimensional data set Means for deriving the degree of overlap S _ij of the abnormal two-dimensional data group, means for creating a network diagram connecting the cause parameters P _i , P _j using a plurality of the degree of overlap S _ij , and means for deriving a high cause hub parameter of the degree of concentration from a network diagram extracting the cause parameter P _k to rise to the abnormality, and means for outputting the cause parameter P _k to rise to the abnormal Fault factor analysis program for causing a function Te. Means for deriving the overlapping degree _{S ij} is ranging 1 ≦ i ≦ n-1 and i <j ≦ n, n ( n-1) / 2 pieces of it from the means for deriving the overlapping degree _{S ij} It said means for creating a network diagram, by utilizing a means for classifying the degree of overlap S _ij in ascending or descending order, the classified the overlapping degree S _ij from the minimum value to the desired rank values, the overlapping degree S _ij the cause parameter P _i, fault factor analysis program according to claim 1 comprising a means for creating the network view by connecting a P _j corresponding to. The means for configuring the normal two-dimensional data group and the means for configuring the abnormal two-dimensional data group are divided into an integer number obtained by dividing the X axis and the Y axis into integers by ^{N1 / 2,} and the X-Y plane lattice reduction, by projecting the three-dimensional data set into the lattice mass consists means constituting the abnormality two-dimensional data group and the normal two-dimensional data set to derive the overlapping degree S _ij means, means for deriving a number of overlapping N ₁ ∩N ₂ to a common number and total for all grid squares of the data when the abnormality two-dimensional data group and the normal two-dimensional data set are in the same grid in the mass, The ratio (N ₁ ∩N ₂ ) / N (hereinafter referred to as cosine measure) of the number of overlaps N ₁ ∩N ₂ with respect to the total number N is derived from means for deriving the degree of overlap S _ij . Abnormal cause analysis program. The abnormality factor according to claim 1, 2 or 3, wherein the analysis parameter D is an inspection parameter, and the cause parameter _Pk is a product processing condition parameter or a quality evaluation parameter for evaluating the processing condition of the product. Analysis program. 4. The abnormality factor analysis program according to claim 1, wherein the analysis parameter D is a quality evaluation parameter for evaluating a processing condition of a product, and the cause parameter _Pk is a processing condition parameter of the product . When a group of products manufactured with at least n cause parameters P _k (1 ≦ k ≦ n) is inspected with a specific analysis parameter D, the abnormality occurs when the analysis parameter D indicates an abnormality. In an abnormality factor analysis program for causing a computer to execute an abnormality factor analysis for extracting an abnormality factor to be extracted from the cause parameter P _k (1 ≦ k ≦ n) , the computer stores the cause parameter P _k and the analysis parameter D. Two different cause parameters P _i and P _j selected from the file storage means and the cause parameters P _k (1 ≦ k ≦ n) extracted from the file storage means are set as the X axis and the Y axis, respectively. the analysis parameter D retrieved from said file storage means and means for configuring the N pieces of three-dimensional data set as a Z axis, the N 3D Means for configuring the analysis parameter D is N ₁ pieces of normal two-dimensional data group to extract the normal three-dimensional data set projected onto the X-Y plane in the normal range of the data group, the N 3D data means for configuring the test parameter D is abnormal three-dimensional data set extracted by the _two N projected onto the X-Y plane abnormal two-dimensional data set in the abnormal range of the group, and the normal two-dimensional data set Means for deriving the degree of overlap S _ij of the abnormal two-dimensional data group, means for creating a network diagram connecting the cause parameters P _i , P _j using a plurality of the degree of overlap S _ij , and means for deriving a high cause hub parameter of the degree of concentration from a network diagram extracting the cause parameter P _k to rise to the abnormality, and means for outputting the cause parameter P _k to rise to the abnormal A fault factor analysis program to function Te by the abnormal factor analysis program stored in the recording medium, fault factor analysis program recording medium which is a readable by the computer. Means for deriving the overlapping degree _{S ij} is ranging 1 ≦ i ≦ n-1 and i <j ≦ n, n ( n-1) / 2 pieces of it from the means for deriving the overlapping degree _{S ij} It said means for creating a network diagram, by utilizing a means for classifying the degree of overlap S _ij in ascending or descending order, the classified the overlapping degree S _ij from the minimum value to the desired rank values, the overlapping degree S _ij the cause parameter P _i, fault factor analysis program recording medium of claim 6 comprising a means for creating the network view by connecting a P _j corresponding to. The means for configuring the normal two-dimensional data group and the means for configuring the abnormal two-dimensional data group are divided into an integer number obtained by dividing the X axis and the Y axis into integers by ^{N1 / 2,} and the X-Y plane lattice reduction, by projecting the three-dimensional data set into the lattice mass consists means constituting the abnormality two-dimensional data group and the normal two-dimensional data set to derive the overlapping degree S _ij means, means for deriving a number of overlapping N ₁ ∩N ₂ to a common number and total for all grid squares of the data when the abnormality two-dimensional data group and the normal two-dimensional data set are in the same grid in the mass, 8. The method according to claim 6, further comprising means for deriving a ratio (N ₁ ∩N ₂ ) / N (hereinafter referred to as a cosine measure) of the overlap number N ₁ ∩N ₂ with respect to the total number N as the overlap degree S _ij . Abnormal cause analysis program record Medium. The abnormal factor according to claim 6, 7 or 8, wherein the analysis parameter D is an inspection parameter and the cause parameter _Pk is a product processing condition parameter or a quality evaluation parameter for evaluating the product processing condition. Analysis program recording medium. 9. The abnormality factor analysis program recording medium according to claim 6, 7 or 8, wherein the analysis parameter D is a quality evaluation parameter for evaluating a processing condition of a product, and the cause parameter _Pk is a processing condition parameter of the product. . When a group of products manufactured with at least n cause parameters P _k (1 ≦ k ≦ n) is inspected with a specific analysis parameter D, the abnormality occurs when the analysis parameter D indicates an abnormality. In the abnormality factor analyzer that extracts an abnormal factor to be extracted from the cause parameter P _k (1 ≦ k ≦ n) and outputs an analysis result, file storage means for storing the cause parameter P _k and the analysis parameter D; Two different processing condition parameters P _i and P _j selected from the cause parameter P _k (1 ≦ k ≦ n) taken out from the file storage means are set as the X axis and the Y axis, respectively, and from the file storage means. Means for configuring N three-dimensional data groups with the extracted inspection parameter D as a Z axis, and the analysis parameter D is within a normal range from the N three-dimensional data groups The means and the inspection parameter D is abnormal range of the N three-dimensional data group constituting the N ₁ pieces of normal two-dimensional data set projected onto the X-Y plane by extracting normal three-dimensional data set in A means for extracting N ₂ abnormal 2D data groups extracted from an abnormal 3D data group and projected onto the XY plane, and a degree of overlap S between the normal 2D data group and the abnormal 2D data group means for deriving _ij , means for creating a network diagram in which the cause parameters P _i and P _j are connected using a plurality of the overlapping degrees S _ij , and a processing condition hub having a high degree of concentration from the network diagram means for extracting the cause parameter P _k which derive the parameters arising the abnormal, the occurring abnormality the cause parameter fault factor analysis apparatus characterized by comprising means for outputting the P _k Means for deriving the overlapping degree _{S ij} is ranging 1 ≦ i ≦ n-1 and i <j ≦ n, n ( n-1) / 2 pieces of it from the means for deriving the overlapping degree _{S ij} The means for creating the network diagram classifies the overlapping degree S _ij in ascending or descending order, and uses the classified overlapping degree S _ij from a minimum value to a desired rank value, and the overlapping degree S _ij corresponding to the cause parameter P _i, fault factor analysis apparatus according to claim 11 comprising a means for creating the network view by connecting a P _j. The means for configuring the normal two-dimensional data group and the means for configuring the abnormal two-dimensional data group are divided into an integer number obtained by dividing the X axis and the Y axis into integers by ^{N1 / 2,} and the X-Y plane lattice reduction, by projecting the three-dimensional data set into the lattice mass consists means constituting the abnormality two-dimensional data group and the normal two-dimensional data set to derive the overlapping degree S _ij Means for deriving an overlapping number N ₁ ∩N _{2 obtained} by summing up the common number of data for all lattice masses when the normal two-dimensional data group and the abnormal two-dimensional data group exist in the same lattice mass; 13. The method according to claim 11, comprising means for deriving a ratio (N ₁ ∩N ₂ ) / N (hereinafter referred to as a cosine measure) of the number of overlaps N ₁ ∩N ₂ with respect to the total number N as the degree of overlap S _ij . Anomaly factor analyzer. 14. The abnormality factor according to claim 11, wherein the analysis parameter D is an inspection parameter, and the cause parameter _Pk is a product processing condition parameter or a quality evaluation parameter for evaluating the product processing condition. Analysis equipment. The abnormality factor analysis device according to claim 11, wherein the analysis parameter D is a quality evaluation parameter for evaluating a processing condition of a product, and the cause parameter P _k is a processing condition parameter of the product .

Images