ABNORMALITY MEASURING METHOD AND ABNORMALITY MEASURING APPARATUS

• US 20170276720A1
• Filed: 10/05/2016
• Published: 09/28/2017
• Est. Priority Date: 03/23/2016
• Status: Active Application

• Current Assignee: Industrial Technology Research Institute
• Original Assignee: Industrial Technology Research Institute
• Inventors: HSU, Yao-Chung, CHANG, Sen-Chia, HSIEH, Tsung-Jung
• Granted Patent: US 10,281,519 B2
• Time in Patent Office: Days
• CPC Class Codes: G01R 31/2601 Apparatus or methods therefor G01R31/2607, G01R31/2642 take precedence
• CPC Class Codes: G01R 31/2642 Testing semiconductor operation lifetime or reliability, e.g. by accelerated life tests
• CPC Class Codes: G08B 21/18 Status alarms G08B21/02 takes precedence
• CPC Class Codes: Y02P 90/30 Computing systems specially adapted for manufacturing

Abstract

An abnormality measuring method and an abnormality measuring apparatus of equipment are provided. The abnormality measuring method includes the following steps: acquiring a feature sequence corresponding to a life cycle according to recipe information and sensing information, wherein the feature sequence includes a plurality of feature subset sequences, and the life cycle is relative to a plurality of process runs; performing repeatedly a life segment analyzing process to acquire a plurality of life segments of the life cycle and each of the plurality of the feature subset sequences corresponding to one of the plurality of life segments; building a corresponding trending distribution of each of the plurality of life segments according to a corresponding feature subset sequence of the life segment; and determining whether to send an alarm message according to a plurality of trending distributions.

Claims

1. An abnormality measuring method, comprising:

acquiring a feature sequence corresponding to a life cycle according to a recipe information and a sensing information, wherein the feature sequence includes a plurality of feature subset sequences, and the life cycle is relative to a plurality of process runs;

performing repeatedly a life segment analyzing process to acquire a plurality of life segments of the life cycle and each of the plurality of feature subset sequences corresponding to one of the plurality of life segments;

building a corresponding trending distribution of each of the plurality of life segments according to a corresponding feature subset sequence of the life segment; and

determining whether to send an alarm message according to a plurality of trending distributions.

2. The method of claim 1, wherein performing repeatedly the life segment analyzing process to acquire the plurality of life segments of the life cycle and the each of the plurality of feature subset sequences corresponding to the one of the plurality of life segments further includes:

performing the life segment analyzing process according to an analyzing start point to acquire a segment transformation point corresponding to the one of the plurality of life segments; and

redefining repeatedly the analyzing start point according to the segment transformation point until no more segment transformation point being acquired;

wherein an initial analyzing start point is defined according to a start point of the life cycle.

3. The method of claim 2, wherein the life segment analyzing process includes:

selecting randomly one set of the feature subset sequence from the plurality of feature subset sequences;

analyzing each of the plurality of feature subset sequences for a corresponding distribution status starting from the analyzing start point to acquire a candidate transformation point and a feature distinguishing capability index of the candidate transformation point;

repeating the above steps until reaching an iterative critical value;

defining the candidate transformation point having a highest feature distinguishing capability index to be the segment transformation point; and

defining a feature subset sequence corresponding to the segment transformation point having the highest feature distinguishing capability index to be the feature subset sequence of the life segment.

4. The method of claim 3, wherein analyzing the each of the plurality of feature subset sequences for the corresponding distribution status starting from the analyzing start point to acquire the candidate transformation point and the feature distinguishing capability index of the candidate transformation point further includes:

getting a first window according to the start point of the life cycle in a selected feature subset sequence;

getting a shifting reference point and a second window corresponding to the shifting reference point in the selected feature subset sequence,

according to the analyzing start point and a window shifting amount;

comparing the first window with the second window;

identifying whether a distribution difference of the second window with respect to the first window is greater than a critical value;

repeating the above steps while the distribution difference of the second window with respect to the first window is less than a critical value until the shifting reference point reaches an end point of the life cycle; and

defining the shifting point to be the candidate transformation point while the distribution difference of the second window with respect to the first window is greater than the critical value.

5. The method of claim 3, wherein analyzing the each of the plurality of feature subset sequences for the distribution status starting from the analyzing start point further includes:

determining window lengths of both the first window and the second window according to the analyzing start point.

6. The method of claim 3, wherein analyzing the each of the plurality of feature subset sequences for a distribution status starting from the analyzing start point further includes:

determining a window shifting amount according to the analyzing start point.

7. The method of claim 1, wherein the feature sequence relates to a degree of deviation between the recipe information and the sensing information.

8. The method of claim 1, wherein the recipe information includes a plurality of recipe values and the sensing information includes a plurality of sensing values, and each of the plurality of sensing values corresponds to one of the plurality of recipe values, and acquiring the feature sequence corresponding to the life cycle according to the recipe information and the sensing information further includes:

generating the feature sequence according to a difference value between each of the plurality of recipe values and a corresponding sensing value thereof and a timing sequence of the life cycle.

9. The method of claim 1, wherein building the corresponding trending distribution of the each of the plurality of life segments according to the corresponding feature subset sequence of the life segment further includes:

training a corresponding health model for the each of the plurality of life segments, according to the plurality of feature subset sequences corresponding to the plurality of the life segments, respectively; and

acquiring the corresponding trending distribution for each of the plurality of life segments, according to the corresponding health model for the each of the plurality of life segments.

10. The method of claim 1, wherein each of a plurality of tending distributions relates to a Mahalanobis distance between data in the life segment of the corresponding feature subset sequence and data in a first window of a start point of the life cycle.

11. The method of claim 1, wherein determining whether to send the alarm message according to the plurality of trending distributions further includes:

generating a warning threshold correspondingly according to the plurality of trending distributions; and

comparing the plurality of trending distributions with the warning threshold to determine whether to announce the alarm message or not.

12. An apparatus for measuring abnormality, comprising:

a feature extraction module, acquiring a feature sequence corresponding to a life cycle according to a recipe information and a sensing information, wherein the feature sequence includes a plurality of feature subset sequences, and the life cycle is related to a plurality of process runs;

a life segment analyzing module coupled to the feature extraction module, performing repeatedly a life segment analyzing process to acquire a plurality of life segments of the life cycle and each of the plurality of feature subset sequences corresponding to one of the plurality of life segments;

a trending analyzing module coupled to the feature extraction module and the life segment analyzing module, building a corresponding trending distribution of each of the plurality of life segments according to a corresponding feature subset sequence of the life segment; and

a warning module coupled to the tending analyzing module, determining whether to send an alarm message according to a plurality of trending distributions.

13. The apparatus of claim 12, wherein the life segment analyzing module further performs the life segment analyzing process according to a analyzing start point to acquire a segment transformation point corresponding to one of the plurality of life segments, and redefines repeatedly the analyzing start point according to the segment transformation point until no more segment transformation point is acquired;

wherein an initial analyzing start point is defined according to a start point of the life cycle.

14. The apparatus of claim 13, wherein the life segment analyzing process includes:

selecting randomly one feature subset sequence from the plurality of feature subset sequences;

analyzing each of the plurality of feature subset sequences for a corresponding distribution status starting from the analyzing start point to acquire a candidate transformation point and a feature distinguishing capability index of the candidate transformation point;

repeating the above steps until reaching an iterative critical value;

defining the candidate transformation point having a highest feature distinguishing capability index to be the segment transformation point; and

defining a feature subset sequence corresponding to the segment transformation point having the highest feature distinguishing capability index to be the feature subset sequence of the life segment.

15. The apparatus of claim 14, the life segment analyzing module further gets a first window according to the start point of the life cycle in a selected feature subset sequence, and gets a shifting reference point and a second window corresponding to the shifting reference point in the selected feature subset sequence, then analyzes the start point according to the analyzing start point and a window shifting amount, and compares the first window with the second window to identify whether a distribution difference of the second window corresponding to the first window is greater than a critical value, and repeats the above steps while the distribution difference of the second window with respect to the first window is less than a critical value until the shifting reference point reaches an end point of the life cycle, and defines the shifting point to be the candidate transformation point while the distribution difference of the second window with respect to the first window is greater than the critical value.

16. The apparatus of claim 14, wherein the life segment analyzing module further determines window lengths of both the first window and the second window according to the analyzing start point.

17. The apparatus of claim 14, wherein the life segment analyzing module further determines a window shifting amount according to the analyzing start point.

18. The apparatus of claim 12, wherein the feature sequence relates to a degree of deviation between the recipe information and the sensing information.

19. The apparatus of claim 12, wherein the recipe information includes a plurality of recipe values and the sensing information includes a plurality of sensing values, and each of the plurality of sensing values corresponds to one of the plurality of recipe values, and the feature extraction module acquires the feature sequence corresponding to the life cycle according to the recipe information and the sensing information, and generates the feature sequence according to a difference value between each of the plurality of recipe values and a corresponding sensing value thereof and a timing sequence of the life cycle.

20. The apparatus of claim 12, wherein the trending analyzing module further builds a corresponding trending distribution of each of the plurality of life segments according to a corresponding feature subset sequence of the life segment, trains a corresponding health model for the each of the plurality of life segments, according to the plurality of feature subset sequences corresponding to the plurality of the life segments, respectively, and acquires the corresponding trending distribution for each of the plurality of life segments, according to the corresponding health model for the each of the plurality of life segments.

21. The apparatus of claim 20, wherein each of the plurality of tending distributions relates to a Mahalanobis distance between data in the life segment of the corresponding feature subset sequence and data in a first window of a start point of the life cycle.

22. The apparatus of claim 12, the warning module further generates a warning threshold correspondingly according to the plurality of trending distributions, and compares the plurality of trending distributions with the warning threshold to determine whether to announce the alarm message or not.

23. The apparatus of claim 12, wherein the feature extraction module, the life segment analyzing module, the trending analyzing module and the warning module are realized by a first processor, a second processor, a third processor and a forth processor, respectively.