Data processing: artificial intelligence – Knowledge processing system – Knowledge representation and reasoning technique
Reexamination Certificate
2001-01-26
2003-09-30
Davis, George B. (Department: 2121)
Data processing: artificial intelligence
Knowledge processing system
Knowledge representation and reasoning technique
Reexamination Certificate
active
06629090
ABSTRACT:
FIELD OF THE INVENTION
The present invention in general relates to a technology for obtaining the relation between data values widely used in industrial fields and extracting a significant result for producing an industrially predominant result.
BACKGROUND OF THE INVENTION
In a semiconductor fabrication process, operations for finding out a yield-deterioration factor as quick as possible are performed in accordance with the history of a device used in a fabrication stage, test results, design information, and various measured data values in order to improve the yield.
Particularly, in the case of development of a new product or review of an existing fabrication process, it is important to extract variable information or regularity hidden in data from the above various original data groups in order to make a data analysis efficient and high reliable. Moreover, by integrating and studying the extracted information and regularity, it is possible to find a knowledge not easily found by an engineer and effectively use the knowledge to find a yield-deterioration factor. Data mining is a data analyzing method for realizing the above mentioned, particularly utilize in finance and circulation fields. Because these industrial fields use a large quantity of data, it is suitable to apply the data mining to the industrial fields.
FIG. 1
is a conceptual illustration showing a general data analyzing method to which data mining is not applied. In the case of the general data analyzing method, individual original data values extracted from databases
2
a
,
2
b
, . . . of an original data group
1
are directly analyzed by an analyzing tool group
3
. According to the analysis result, decision making is performed. The analyzing tool group
3
includes a statistical analysis component
4
a
and a chart drawing component
4
b.
FIG. 2
is a conceptual illustration showing a conventional data analyzing method executed in a semiconductor fabrication process. A conventional semiconductor fabrication process uses a general data analyzing method to which data mining is not applied. An original data group
5
is provided with a data base
6
a
for design data, a data base
6
b
for process data, a data base
6
c
for device data, and a data base
6
d
for test results and the like. Analysis-object data
7
is constituted of the original data extracted from the databases
6
a
,
6
b
,
6
c
, and
6
d
. The analysis-object data
7
is processed by a data processor
8
and data
9
for low-yield factors is obtained.
The analysis-object data
7
is extracted in accordance with an analysis procedure or layer. Data to be extracted or data to be used for an analysis is decided in accordance with the past know-how, experience, and skill of each engineer. That is, a decision data is left to the discretion of an engineer who performs an analysis. Moreover, the analysis result is shown in the form of a correlation diagram, trend graph, or histogram.
In general, a device-difference analysis is performed in order to clarify a yield-deterioration factor in a semiconductor fabrication process.
FIG. 3
is a conceptual illustration showing the flow of a lot in the device-difference analysis and
FIG. 4
is a box and whisker chart showing the yield value of a lot in the device-difference analysis every device used. The box and whisker chart is drawn for each fabrication process. In the case of the device-difference analysis, it is extracted which device most influences the yield in each fabrication process from the data for a device used for the process of each lot. Then, a process in which a yield difference is most remarkable and a device used are identified in accordance with an obtained box and whisker chart.
However, the above device-difference analysis has a problem that extremely large man-hour is required for the analysis because the number of fabrication processes is several hundreds at present. Moreover, when a difference between devices is not clearly obtained or conditions are complexly combined, there is a disadvantage that it is sometimes difficult to determine.
Furthermore, because an analysis is progressed in accordance with the past know-how, experience, or skill of each engineer in the case of a conventional analyzing art, it cannot be avoided that the efficiency or reliability of an analysis is unavoidable. Therefore, a data analysis art is desired which makes it possible to decrease the rate depending on the know-how, experience, or skill of an engineer, effectively use study of the knowledge for efficiently executing the analysis by each analyzing tool so that there is no leak and the study result, and even evaluate the accuracy of the study result.
FIG. 5
is a schematic view showing a configuration of records used for the data analysis of the classification analysis which is one of data mining methods. Generally, in the case of a data analysis, a variable purposing searching a fluctuation cause or fluctuation pattern is referred to as a purpose variable and a variable for explaining the fluctuation of purpose variables is referred to as an explanation variable. Records
10
a
,
10
b
, . . . ,
10
i
are divided into purpose-variable data values
11
a
,
11
b
, . . . ,
11
i
and explanation-variable data values
12
a
,
12
b
, . . . ,
12
i.
The efficiency or reliability of an analysis is changed depending on an object used as purpose variables or explanation variables or a type of analysis to be performed. Therefore, it is necessary to evaluate the reliability or accuracy of an analysis.
To analyze the data for the yield of a semiconductor fabrication process in many cases, a purpose variable uses a yield and a explanation variable uses the history of a device used, test results, design information, and various measured data values. To more efficiently perform an analysis and improve the reliability of the analysis, it is necessary to perform the processing for clarifying the relation between purpose variables and an explanation variables in an original data group and an existing analyzing-tool group as shown in FIG.
1
.
Data mining is effective as a method for clarifying the relation between purpose variables and explanation variables.
FIG. 6
is a conceptual illustration for showing a general data analyzing method to which data mining is applied. In the case of the data analyzing method to which data mining is applied, a rule file
16
is generated by an device
15
for extracting processing of features and regularities hidden in data (data mining) in accordance with individual original data value extracted from data bases
14
a
,
14
b
, . . . of an original data group
13
.
Then, individual original data value extracted from the data bases
14
a
,
14
b
, . . . is analyzed by an analyzing tool group
17
in accordance with the rule file
16
. Decision-making is performed in accordance with the analysis result. The analyzing tool group
17
includes, for example, a statistical analysis component
18
a
and a chart drawing component
18
b.
When applying data mining to a yield data analysis, an action for improving a yield is decided in accordance with a data mining result, it is determined whether to take action for that or not, or an action effect is estimated. For this, quantitative evaluation or accuracy of a data mining result is necessary.
A regression tree analysis is particularly effective among classification analyses of the data mining method. One of advantages of the regression tree analysis is that results are output as an comprehensible rule and expressed by a general language or a database language such as an SQL language. Therefore, by effectively using the reliability or accuracy of these results, it is possible to perform effective decision making or take actions in accordance with the result.
The regression tree analysis is described below. The regression tree analysis is applied to a set constituted of records comprising explanation variables showing a plurality of attributes and a purpose variable to be influenced by the explanation variable, which identifies an attri
Shirai Hidehiro
Tsuda Hidetaka
Davis George B.
Fujitsu Limited
LandOfFree
method and device for analyzing data does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with method and device for analyzing data, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and method and device for analyzing data will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3029780