JP2010092260A - Thinking quality improvement system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To achieve a workflow pattern along a thinking pattern of an analyst, in a thinking quality improvement system. <P>SOLUTION: Functions of a data processing engine including at least tabulation, narrowing, and classification in a workflow function are made versatile, and they are combined to regularly process modification or processing of data of quality analysis. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a thought quality improvement system, for example, realizing a workflow pattern according to an analyst's thought pattern, functionalizing the know-how of analysis work by an engineer, and improving the accuracy and speed of analysis, In addition, the present invention relates to a configuration for automating analysis as a management task for a problem that becomes steady based on a result obtained by analysis.

  Conventionally, for quality improvement of semiconductor manufacturing, engineers defined target data for each process and department, and systematized analysis methods. As a result, quality improvement of the total product requires analysis methods, system usage, and combinations for each process, and this is done with the thought and judgment of engineers (people).

  The present inventor made a practical quality improvement, engineer's thinking, analysis work into a workflow that complements the difference between skills, and using repository technology that inputs and manages intentional comments of multiple engineers, It has been proposed to grow a highly important workflow into a knowledge system by self-learning (see, for example, Patent Document 1). In this proposal, the engineer navigates through the analysis and ultimately realizes automation of the work itself.

  However, in this proposal, drill-down automation has not been achieved because, for example, a fixed condition is taken over when different engines are linked, or a part of the flow operation by an engineer is required again based on the analysis result. In addition, there is a flow design that cannot be combined depending on conditions because there is a filter function between individual engines in the workflow.

  Therefore, as a further improvement, the present inventor linked the extraction conditions and analysis results between the engines (processing components) in the workflow design, so that the extraction specified in the previous engine, the filter conditions, and the analysis Providing an independent conditional branch and classification and filter engine separated from the data type and analysis engine so as to automate drilldown by passing the result to the next engine and to cope with various flow branches (For example, refer to Patent Document 2). This proposal makes it possible to design workflows that are more in line with engineer thinking.

  In this proposal, workflow design is performed on a matrix, and one matrix is created for one business flow. A plurality of workflows can be set for one business flow using the workflow engine listed in the general engine, and one row of the matrix is assigned to one workflow. A plurality of workflows are created in each row of the matrix in order from the top, whereby the basic order of the workflows is determined.

  The user clicks various engines listed in groups in the engine selection area for each category, creates each workflow on the matrix, and arranges a predetermined engine in each workflow area. For example, the various engines are grouped into general engines, data extraction engines, data processing engines, result engines, and the like.

  Specifically, the workflow engines listed in the general engine are clicked to create each workflow on the matrix. Next, the engines necessary for the business flow are sequentially clicked from the engines listed in the extraction engine or the processing engine, etc., and arranged in the area of each workflow. In this case, an execution order (seq number) is issued corresponding to the arrangement position of the engine arranged in each workflow, and processing by each engine is performed according to this execution order.

If necessary, a data connection line listed in the general engine is clicked to connect an engine arranged in a certain workflow area with an engine arranged in another workflow area. The data flow can be freely set between the engines by the data connection line.
JP 2005-182635 A JP 2007-188481 A

  However, in the proposal of the above-mentioned Patent Document 2, as a technical problem, various groupings and filters are generated according to the engineer's analytical thinking when performing filtering and branching in the workflow, regardless of the data type, It is necessary to normalize the engine function that ranks data regularly and counts the frequency.

  In addition, as an issue of detailed analysis by engineers and analysis of the flow design process, it is necessary to integrate drill-down analysis of object format that performs the next analysis from one result in an interactive format into the workflow frame. ing.

  Therefore, an object of the present invention is to realize a workflow pattern in accordance with an analyst's thought pattern in a thought quality improvement system.

  From one aspect of the present invention, the data processing engine function including at least totaling, narrowing down, and classification in the workflow function is generalized, and the processing and processing of quality analysis data are normally processed from the combination thereof. A thought quality improvement system is provided.

  According to the disclosed thought quality analysis system, it is possible to increase the degree of freedom of the combination of workflow data processing engines, and to create an analysis flow in accordance with the engineer's thoughts more flexibly. This makes it possible to systemize (functionalize) a variety of quality analysis methods for complex and sophisticated manufacturing failure problems, thereby shortening and automating the improvement.

  In addition, it is possible to significantly reduce the manufacturing cost by reducing the time required for improvement and stable maintenance of the manufacturing quality and by reducing the number of engineers.

  Here, an embodiment of the present invention will be described. The present invention is a technology related to basic design of a basic system, and is a basic specification in system development. The internal processing of software in the workflow is disclosed in the above-mentioned Patent Document 2, and the present invention solves the technical problem of Patent Document 2, and as an improvement technique for realizing a system that is more in line with the engineer's thinking, The basic specifications will be described below. In the thought quality improvement system based on the improvement of the present invention, the engine functions such as data grouping, tabulation, narrowing down, and classification in the workflow function are normalized, and a combination of the combination results in a workflow pattern that matches the analyst's thought pattern. Realize. Therefore, each data processing engine is generalized so that each engineer can arbitrarily set conditions in grouping, tabulation, narrowing down, classification, and the like.

  In addition, it is possible to normalize the function for each engine for ranking the data, filter the count, count for the count, etc., process it freely regardless of the data type, and combine the processing engines It is possible to construct various judgment and branch flow patterns.

  As a result, engineer-thought numerical determination and numerical behavior determination can be realized by a combination of normalized engines, and multiple determinations and branching workflows can be created by combining a plurality of engine processes. For example, the number of LOT processes in quality analysis, the relief rate due to standard relaxation, and indexing of in-plane distribution can be realized by a combination of engines without specially functionalizing.

In this case, each data processing engine is configured so that the processing functions such as date data narrowing and classification are shared in advance in each data processing engine, and the date processing function is normalized.
There are various notations for year / month / day / hour / minute / second indicating the update date / time, etc. For example, the date processing is unified in the [yyyymmdd hhmms] format to enable common handling of data with various date / time notations. To do.

  In addition, the routine (automatic) processing by the workflow and the drill-down analysis of the object from the result screen can be processed in parallel. For example, as a means of realizing drill-down and workflow integration from one result to the next result, the flow control XML (extensible Markup Language) process is started from the result screen, so that it is actually unified with control in the workflow. Meanwhile, the analyst fuses interactive processing from the result screen and automatic processing from the flow screen to make analysis and management seamless.

  In this way, by performing the flow control XML processing from the result screen in the back, the analyst can drill down from one result screen to the next analysis, drilling down one data in detail It is possible to combine interactive drill-down analysis for identifying factors and routine automatic monitoring by workflow, and to set the monitoring flow from the factor identification.

  In addition, it is desirable that the data totaling engine in the workflow function has a function of automatically acquiring grouping totaling items with designated items and holding them as internal group keys.

  Based on the above, next, engine normalization according to the first embodiment of the present invention will be described with reference to FIG. Here, as an example, a flow for automatically creating a relief LIST and its data trend when the monitor characteristic value is relaxed from the old standard to the new standard will be described.

  First, a workflow engine having a matrix of 3 rows and 9 rows is created by clicking a workflow engine grouped in the general engine. Next, the “LOT extraction” engine and the “monitor extraction” engine grouped in the data extraction engine are sequentially clicked, and the “filter” engine, the “aggregation” engine, and the “filter” engine grouped in the data processing engine are clicked. Click sequentially to place in the upper row (first line in the workflow). Next, the “filter” engine, the “aggregation” engine, the “filter” engine, and the “merge” engine grouped in the data processing engine are sequentially clicked and arranged in the middle stage (second line in the workflow). Next, the “merge” engine and the “filter” engine grouped in the data processing engine and the “trend” engine grouped in the result engine are sequentially clicked and arranged in the lower stage (third line in the workflow). Next, each engine is connected by a data connection line grouped into general engines.

  An execution order (seq number) is issued so as to flow from the left to the right in order from the top, corresponding to the placement position of the engine placed in each workflow, and processing by each engine is performed according to this execution order. .

Here, the specific operation will be described. First, the third “filter” engine from the left in the upper stage (first row) is out of the condition (0.35 m>) that the threshold voltage V _th is 0.35 mV or more of the old standard. Then, the next “aggregation” engine aggregates the number of non-standard points in the lot (LOT) and wafer (WF) units, and further the non-standard score is converted to wafers (WF) via the “filter” engine. ), For example, lots (LOT) and wafers (WF) having three or more points are cut out, and a LIST of lots (LOT) and wafers (WF) determined to be shipped NG in the old standard is created.

Next, the third “filter” engine from the left in the middle row (second row) cuts out data that deviates from the condition (0.345 m>) that the threshold voltage V _th is 0.345 mV or higher of the new standard, and the next “aggregation” Lots (LOT) and wafers (WF) are counted by the engine, and the non-standard points are counted via the “filter” engine, and the lots (LOT) whose non-standard points are within 3 points in the wafer (WF). ), A wafer (WF) is cut out, and a LIST of lots (LOT) and wafers (WF) determined to be shipped OK according to the new standard is created.

  Next, the LIST of the judgment NG in the upper standard (first line) of the old standard and the judgment OK in the middle standard (second line) of the new standard is completely performed on the lot (LOT) and wafer (WF) by the “merge” engine. By performing the merge processing based on the coincidence (and condition), it is possible to perform the LIST of the old standard that is NG and the new standard that is OK, that is, the LIST of the lot (LOT) and the wafer (WF) relieved by the new standard.

  Next, the middle (second row) relief lot (LOT), wafer (WF) LIST, and the original “monitor data extraction” engine are completely matched by the lower (third row) “merge” engine (and conditions) ), The original measurement data can be combined with the rescued lot (LOT) and wafer (WF), and as a result, the trend of the rescue lot (LOT) and wafer (WF) monitor data is automatically created. can do.

  In this way, by combining the normalized “filter” engine, “aggregation” engine, and “merge” engine in a hierarchical manner, for example, what was shipped NG in the old standard after the standard change is changed to the new standard. The function of creating the trend of the monitor trend on the rescued lot (LOT) and wafer (WF) is functionalized by combining the three engines without coding directly in the program source.

  Next, with reference to FIG. 2 and FIG. 3, the internal processing by the combination of normalized engines according to the second embodiment of the present invention will be described. This is a combination of the normalized “filter” engine, “aggregation” engine, and “merge” engine, which reduces data processing loops that were previously coded directly into the program source, such as narrowing down and summarizing specific data. On the GUI (Graphical User Interface), engineers design and functionalize according to the thinking of analysis themselves.

FIG. 2 shows a workflow that is arranged and designed with a GUI using the present technology. Here, as an example, the average of measurement points in a specific wafer (WF) in a lot (LOT) and a wafer (WF) determined to be shipment OK by combination determination of two measurement items in shipment determination of monitor characteristic values A flow for counting values will be described. The content of the flow is the standard (pass / fail) judgment in the wafer (WF) plane. Item A is 9 points of measurement, all points (9 points) are within standard, and item B is 9 points of measurement, 7 points are within standard The combination lot (LOT) and wafer (WF) are determined as shipment OK, and using the data of the determination OK, specific immediate fixed points in the wafer (WF) are upper (TOP) and lower (BOTTOM) Position data is extracted, and the group average is calculated for each lot (LOT) and wafer (WF) for items A and B, respectively. Here, the threshold voltage _Vth of the MOSFET is measured at item A at nine measurement positions in the center, upper, lower, right, left, upper right, upper left, lower right, and lower left in the wafer (WF), and item B The case where the drain current I _on when the MOSFET is _on is measured will be described.

  As shown in FIG. 2, two “filter” engines and one “aggregation” engine are sequentially arranged in the first row (first row) in the workflow area of 5 rows and 6 columns. In the second row (second row), a “filter” engine, a “total” engine, a “filter” engine, and a “merge” engine are sequentially arranged. In the third row (third row), a “filter” engine, a “total” engine, and a “filter” engine are sequentially arranged. In the fourth stage (fourth row), two “filter” engines, one “aggregation” engine, and two “merge” engines are sequentially arranged. Also in this case, an execution order (seq number) is issued so as to flow from the left to the right in order from the upper stage, corresponding to the arrangement position of the engine arranged in each workflow. Processing is performed.

  FIG. 3 is a flowchart of the processing and calculation flow processed in the background of the workflow shown in FIG. First, processing by the engines arranged in the first row (first row) shown in FIG. 2 in the flow in the right end column will be described first. First, the data of items A and B are extracted by the data extraction at the upper center of the flowchart, and then the measurement position information (X, Y) items in the wafer (WF) plane are combined using the classification function of the “filter” engine. Create a label for the measurement position. Next, the flow moves to the rightmost column, and then the upper position (TOP) and lower position (BOTTOM) for item A are extracted from the previous measurement position label by the “filter” engine, and the item is counted by the “aggregation” engine. A group average is calculated for each lot (LOT) and wafer (WF) from the data of the upper position (TOP) and the lower position (BOTTOM) of A.

Next, in the flow in the second column from the right, processing by each engine arranged in the second stage (second line) shown in FIG. 2 will be described. First, for the item A, those that satisfy the scrap standard condition 122054> are extracted by the “filter” engine. Next, the data count under the shipping standard conditions of, for example, −0.006431 ≦ and 0.16468 ≧ is calculated by the group total for each lot (LOT) and wafer (WF) by the “total” engine. Next, the data frequency items tabulated by the “filter” engine are extracted with all points (9 points), and the LIST of the lot (LOT) and the wafer (WF) of the shipment judgment OK of item A are created. This is because strict shipping management is required for the threshold voltage _Vth .

  Next, the lot (LOT) of shipment judgment OK of item A, the LIST of wafer (WF) by the “merge” engine, and the upper position (TOP) and the lower position (BOTTOM) counted by the “total” engine of the workflow in the rightmost column ) And the group average value of the lot (LOT) and the wafer (WF) are logically combined with each other (and condition).

Next, in the flow in the third column from the right, processing by each engine arranged in the third row (third row) shown in FIG. 2 will be described. First, for the item B, for example, those satisfying the shipping standard conditions of −0.006431 ≦, 0.16468 ≧ are extracted by the “filter” engine. Next, the data count is calculated for the lot (LOT) and wafer (WF) groups by the “aggregation” engine. Next, 7 or more data frequency items tabulated by the “filter” engine are extracted, and a shipment LIST lot (LOT) and wafer (WF) LIST are created. This is because the condition for the drain current I _on may be relaxed from the threshold voltage V _th .

  Next, processing by each engine arranged in the fourth stage (fourth row) shown in FIG. 2 in the flow of the leftmost column will be described. First, for the item B by the “filter” engine, abnormal values due to measurement trouble are excluded under the conditions of −1 ≦, 1 ≧. This does not occur due to an abnormality in the manufacturing process, but is accident measurement data that occurs due to poor contact of the probe during measurement, and is not original characteristic data. Since this abnormal value remains mixed, it becomes a factor for making an erroneous determination in quality determination, so it is excluded by the “filter” engine.

  Next, data of the upper position (TOP) and lower position (BOTTOM) is extracted from the measurement position label for the item B by the next “filter” engine, and the upper position (TOP) and lower position of the item B by the “aggregation” engine. The group average of the (BOTTOM) data is calculated for each lot (LOT) and wafer (WF). Next, by the “merge” engine, the shipment OK lot (LOT) and wafer (WF) LIST of the item B created in the flow in the third column from the right, and the upper position (TOP) and the lower position of the item B of the left end flow The group average value of (BOTTOM) is logically connected by a perfect match (and condition) of lot (LOT) and wafer (WF).

  Finally, the upper position (TOP) and the lower position (BOTTOM) in the wafer (WF) of the item B, which is the result of the two “merge” engines in the previous stage, in the seven-point shipment OK as a result of the two “merge” engines in the previous stage. And the group average value of the top position (TOP) and the bottom position (BOTTOM) in the wafer (WF) in all items (9 points) shipment OK of Item A, lot (LOT), wafer Logical conjunction is performed with a perfect match (and condition) of (WF).

  Thereafter, using a “graph” engine or the like, the LIST of the lot (LOT) and wafer (WF) that have passed the shipment judgment by a combination of two items and the data of the specific location are displayed, or “CSV output” Data is provided using an engine or the like, and the flow is completed.

  As described above, in the second embodiment of the present invention, by combining a normalized engine, loops such as data processing and tabulation that have been directly coded in the program source in the past can be considered on the GUI and by the engineer himself. Accordingly, it can be arbitrarily designed and functionalized individually, so that it can contribute to quality control of a semiconductor manufacturing process that meets the needs of actual users.

  Next, an interactive drill-down analysis function on the result screen according to the third embodiment of the present invention will be described with reference to FIGS. 4 and 5. In the third embodiment, as in the back flow design screen shown as an example in FIG. 4, the arrangement and setting of each engine according to the drill-down operation are performed, and the histogram analysis is performed from the trend analysis screen in FIG. When the screen transitions to the screen by interactive drill-down, in fact, the flow design on the back of FIG. 4, the condition is dynamically moved from the trend graph engine to the histogram engine, Realize interactive drill-down analysis.

  In other words, it is possible to reverse start the condition linkage and processing from the flow order result screen to other analysis engines, the actual operation is the flow editing screen, but the visual UI drills down from the result screen to the result screen To do. Here, an example is shown in which the histogram engine next to the trend engine is activated and executed on the flow screen.

  FIG. 5 is a diagram illustrating a transition from the trend analysis result screen to the histogram analysis result screen. In this way, the flow editing engine is actually operating, but the analyst realizes a drill-down UI over the next histogram analysis from one graph result indicating the trend.

  Next, an algorithm (software design document) of the data processing engine according to the fourth embodiment of the present invention will be described with reference to FIGS. In the fourth embodiment, as processing of data, the three processing functions of cutout, classification, and combination are normalized and integrated into one GUI.

  FIGS. 6A to 6C are image diagrams of the GUI of the data processing engine. In the GUI, the condition tag is switched according to the check (CHK) of the radio button (exclusive button) that is cut out, classified, and combined. FIG. 6A shows a case where the normal filter (cutout) of the radio button is checked. Here, “RADIUS 4” is selected from the filter candidate list, and “03-04” and “equal” are completely set as conditions. An example in which the logical product of coincidence (and condition) is set is shown.

  FIG. 6B shows a case where the radio button classification filter is checked, and “BIN-RATE.PASS-0” is selected from the filter candidate list, and “0.5 or more and less than 0.8” is selected as a condition. An example of selection and setting is shown. FIG. 6C shows a case where the radio button column combination filter is checked, and shows an example in which “LOT” and “WAFER” are selected and combined from the filter candidate list.

  In the conventional technique, processing functions such as cutout, classification, and combination as processing are individually designed according to data types, and actual program sources are also coded in individual classes. On the other hand, in the present invention, as shown in FIG. 6, the three software specifications of cutout, classification, and combination are standardized, the program source classes are aligned and shared, the user's UI usability (usability) is improved, and The degree of freedom of the combination has been dramatically improved. In addition, the sharing of program source classes can reflect the revision of one basic function to the three functions, which contributes to the development efficiency and the flexibility of the function revision and the shortening of the version. At that time, it is desirable to improve development efficiency by normalizing and sharing conditions such as date, characters, and numerical values as much as possible and classifying software coding.

  The basic algorithm of the processing engine is as shown in FIGS. 7 to 10, and here, each algorithm is shown as a basic concept. FIG. 7 shows a basic algorithm of the processing engine, which reads the item of the engine to be passed, further checks the item data (CHK), and receives the condition setting of date, character, and numerical value. While inputting, classifying, combining, and sharing a class, the processing input and the result are normally held. Thus, the processing engine itself can be repeatedly connected to process the composite condition. Subclasses of classification, classification, and combination are added to this basic algorithm, and each is made a normal function.

  FIG. 8 shows an example of processing engine segmentation: character item algorithm, FIG. 9 shows an example of processing engine classification: character item algorithm, and FIG. 10 shows an example of processing engine combination algorithm. It is.

  The embodiments of the present invention have been described above, but the present invention is not limited to the configurations and conditions shown in the embodiments. For example, the first embodiment or the second embodiment described above is shown as an example applied to the quality control of the semiconductor integrated circuit device, but is not limited to the quality control of the semiconductor integrated circuit device. It is applied to quality control in the manufacturing process.

In the second embodiment, the number of measurement points is 9 points, the item A (V _th ) is within the standard for all points, and the item B (I _on ) is 7 points or more within the standard. Although it is OK, the score within the standard is arbitrary in each item. Further, the number of measurement points is not limited to nine, but may be other measurement points such as five points measurement of the center, upper, lower, right, and left, and the number of points within the standard is changed according to the number of measurement points.

FIG. 10 is a flowchart for automatically creating a relief LIST by standard relaxation and its data trend. This is a workflow designed and arranged using a GUI. It is a flowchart of the flow of the process and calculation processed with the background of the workflow shown in FIG. It is a figure which shows an example of the back flow design screen. It is explanatory drawing which showed a mode that it moved to the result screen of a histogram analysis from the result screen of a trend analysis. It is an image figure of GUI of a data processing engine. It is explanatory drawing of the basic algorithm of a process engine. Cutting out a processing engine: It is explanatory drawing of an example of the algorithm of a character item. Classification of processing engine: It is explanatory drawing of an example of the algorithm of a character item. It is explanatory drawing of an example of the algorithm of a process engine coupling | bonding.

Claims (4)

The data processing engine function including at least totaling, narrowing down, and classification in the workflow function is generalized, and processing or processing of quality analysis data is normally processed by a combination of the data processing engines. Thinking quality improvement system. 2. The thought quality improvement system according to claim 1, wherein an interactive drill-down process for starting a plurality of engines in parallel from one of the data processing engines is performed. The thought quality improvement system according to claim 1 or 2, wherein the data processing engine has a common processing function including at least date data narrowing and classification, and normalizes the date processing function. 4. The data collection engine includes a totaling engine that automatically acquires a grouping totaling item for data totaling using a specified item and stores the data as an internal group key. 5. The thought quality improvement system described in the paragraph.

Images