Detecting degradation of components during reliability-evaluation studies

• US 7,162,393 B1
• Filed: 09/01/2005
• Issued: 01/09/2007
• Est. Priority Date: 09/01/2005
• Status: Active Grant

• Current Assignee: Oracle America, Inc. (Oracle Corporation)
• Original Assignee: Sun Microsystems Incorporated (Oracle Corporation)
• Inventors: Gross, Kenny C., Lopez, Leoncio D., Vacar, Dan, McElfresh, David K.
• Primary Examiner(s): Barlow; John
• Assistant Examiner(s): Walling; Meagan S
• Application Number:
  US11/219,091
• Time in Patent Office: 495 Days
• Field of Search: 702/183, 702/185, 702/182, 702/33, 702/34, 702/35, 702/36
• US Class Current: 702/183
• CPC Class Codes: G01R 31/2849 Environmental or reliability testing, e.g. burn-in or validation tests of individual semiconductors G01R31/2642; of printed circuits boards G01R31/2817; of IC's G01R31/2855
• CPC Class Codes: G06F 11/008 Reliability or availability analysis
• CPC Class Codes: G06F 11/24 Marginal checking or other specified testing methods not covered by G06F11/26, e.g. race tests

Abstract

One embodiment of the present invention provides a system that determines the reliability of a component in a system. During operation, the system monitors inferential variables associated with a number of specimens of the component. The system then collects degradation data by first computing a likelihood value that indicates whether an inferential variable associated with a specimen of the component is behaving normally or abnormally. Next, the system determines whether the specimen of the component has degraded based on the likelihood value. If the specimen of the component is determined to have degraded, the system records the time when the specimen of the component was determined to have degraded. The system also uses the degradation data to determine the reliability of the component in the system.

Claims

1. A method for determining the reliability of a component in a system, comprising:

monitoring inferential variables associated with a number of specimens of the component during a reliability-evaluation study;

collecting degradation data by;

computing a likelihood value that indicates whether an inferential variable associated with a specimen of the component is behaving normally or abnormally;

determining whether the specimen of the component has degraded based on the likelihood value; and

if the specimen of the component is determined to have degraded, noting the time when the specimen of the component was determined to have degraded; and

generating a population-failure distribution for the number of specimens from the degradation data; and

using the population-failure distribution to generate a component-reliability model, wherein the component-reliability model is used to determine the reliability of the component past the point in time when the reliability-evaluation study was terminated for the number of specimens of the component.

2. The method of claim 1, wherein using the likelihood value to determine whether the specimen of the component has degraded involves:

comparing the likelihood value with an upper threshold;

comparing the likelihood value with a lower threshold;

if the likelihood value is greater than or equal to the upper threshold, determining that the specimen of the component has degraded;

if the likelihood value is less than or equal to the lower threshold, determining that the specimen of the component is behaving normally; and

if the likelihood value is between the upper threshold and the lower threshold, determining that an inadequate amount of information is available to determine whether the specimen of the component is behaving normally or whether the specimen of the component has degraded.

3. The method of claim 1, wherein computing the likelihood value involves:

computing the probability that a null hypothesis is true using a current value of the inferential variable and historical values of the inferential variable, wherein the null hypothesis is true if the specimen of the component is behaving normally and has not degraded;

computing the probability that an alternative hypothesis is true using the current value of the inferential variable and historical values of the inferential variable, wherein the alternative hypothesis is true if the specimen of the component has degraded; and

computing the likelihood value by dividing the probability that the alternative hypothesis is true by the probability that the null hypothesis is true.

4. The method of claim 3, wherein the alternative hypothesis test can include:

a positive-mean test, which tests whether the mean of a time-series for the inferential variable is above a reference level;

a negative-mean test, which tests whether the mean of the time-series for the inferential variable is below a reference level;

a nominal-variance test, which tests whether the variance of the time-series for the inferential variable is proportional to a scale factor;

a inverse-variance test, which tests whether the variance of the time-series for the inferential variable is proportional to the inverse of the scale factor;

a positive first-difference test, which tests whether the time-series for the inferential variable is increasing;

or

a negative first-difference test, which tests whether the time-series for the inferential variable is decreasing.

5. The method of claim 1,

wherein prior to monitoring inferential variables, the method comprises computing the mean and the variance of the time-series for the inferential variable by monitoring the inferential variable during a training phase;

and

wherein the mean and variance are used to normalize the mean and the variance of the time-series for the inferential variable while monitoring the inferential variable during the monitoring phase.

6. The method of claim 1, wherein the inferential variable can represent a current passing through the component or a voltage applied to the component.

7. The method of claim 1,

wherein the reliability-evaluation study is an accelerated-life study of the component;

and

wherein the monitored data can be processed in real-time or can be post-processed.

8. An apparatus for determining the reliability of a component in a system, comprising:

a stress-test chamber configured to accommodate specimens of a component;

a degradation-detection mechanism;

wherein the degradation-detection mechanism is configured to;

monitor inferential variables associated with a number of specimens of the component during a reliability-evaluation study;

collect degradation data, wherein to collect degradation data, the degradation-detection mechanism is configured to;

compute a likelihood value that indicates whether an inferential variable associated with a specimen of the component is behaving normally or abnormally;

determine whether the specimen of the component has degraded based on the likelihood value; and

if the specimen of the component is determined to have degraded, to note the time when the specimen of the component was determined to have degraded; and

to

generate a population-failure distribution for the number of specimens from the degradation data; and

to

use the population-failure distribution to generate a component-reliability model, wherein the component-reliability model is used to determine the reliability of the component past the point in time when the reliability-evaluation study was terminated for the number of specimens of the component.

9. The apparatus of claim 8, wherein while using the likelihood value to determine whether the specimen of the component has degraded, the degradation-detection mechanism is configured to:

compare the likelihood value with an upper threshold;

compare the likelihood value with a lower threshold;

if the likelihood value is greater than or equal to the upper threshold, to determine that the specimen of the component has degraded;

if the likelihood value is less than or equal to the lower threshold, to determine that the specimen of the component is behaving normally; and

if the likelihood value is between the upper threshold and the lower threshold, to determine that an inadequate amount of information is available to determine whether the specimen of the component is behaving normally or whether the specimen of the component has degraded.

10. The apparatus of claim 8, wherein while computing the likelihood value, the degradation-detection mechanism is configured to:

compute the probability that a null hypothesis is true using a current value of the inferential variable and historical values of the inferential variable, wherein the null hypothesis is true if the specimen of the component is behaving normally and has not degraded;

compute the probability that an alternative hypothesis is true using the current value of the inferential variable and historical values of the inferential variable, wherein the alternative hypothesis is true if the specimen of the component has degraded; and

to

compute the likelihood value by dividing the probability that the alternative hypothesis is true by the probability that the null hypothesis is true.

11. The apparatus of claim 10, wherein the alternative hypothesis test can include:

a positive-mean test, which tests whether the mean of a time-series for the inferential variable is above a reference level;

a negative-mean test, which tests whether the mean of the time-series for the inferential variable is below a reference level;

a nominal-variance test;

which tests whether the variance of the time-series for the inferential variable is proportional to a scale factor;

a inverse-variance test, which tests whether the variance of the time-series for the inferential variable is proportional to the inverse of the scale factor;

a positive first-difference test, which tests whether the time-series for the inferential variable is increasing;

or

a negative first-difference test, which tests whether the time-series for the inferential variable is decreasing.

12. The apparatus of claim 8,

wherein prior to monitoring inferential variables, the degradation-detection mechanism is configured to compute the mean and the variance of the time-series for the inferential variable by monitoring the inferential variable during a training phase;

and

wherein the mean and variance are used to normalize the mean and the variance of the time-series for the inferential variable while monitoring the inferential variable during the monitoring phase.

13. The apparatus of claim 8, wherein the inferential variable can represent a current passing through the component or a voltage applied to the component.

14. The apparatus of claim 8,

wherein the reliability-evaluation study is an accelerated-life study of the component;

and

wherein the monitored data can be processed in real-time or can be post-processed.

15. A computer system for determining the reliability of a component in a system, comprising:

a stress-test chamber configured to accommodate specimens of a component;

a degradation-detection mechanism;

wherein the degradation-detection mechanism is configured to;

monitor inferential variables associated with a number of specimens of the component during a reliability-evaluation study;

collect degradation data, wherein to collect degradation data, the degradation-detection mechanism is configured to;

compute a likelihood value that indicates whether an inferential variable associated with a specimen of the component is behaving normally or abnormally;

determine whether the specimen of the component has degraded based on the likelihood value; and

if the specimen of the component is determined to have degraded, to note the time when the specimen of the component was determined to have degraded; and

to

generate a population-failure distribution for the number of specimens from the degradation data; and

to

use the population-failure distribution to generate a component-reliability model, wherein the component-reliability model is used to determine the reliability of the component past the point in time when the reliability-evaluation study was terminated for the number of specimens of the component.

16. The computer system of claim 15, wherein while using the likelihood value to determine whether the specimen of the component has degraded, the degradation-detection mechanism is configured to:

compare the likelihood value with an upper threshold;

compare the likelihood value with a lower threshold;

if the likelihood value is greater than or equal to the upper threshold, to determine that the specimen of the component has degraded;

if the likelihood value is less than or equal to the lower threshold, to determine that the specimen of the component is behaving normally; and

if the likelihood value is between the upper threshold and the lower threshold, to determine that an inadequate amount of information is available to determine whether the specimen of the component is behaving normally or whether the specimen of the component has degraded.

17. The computer system of claim 15, wherein while computing the likelihood value, the degradation-detection mechanism is configured to:

compute the probability that a null hypothesis is true using a current value of the inferential variable and historical values of the inferential variable, wherein the null hypothesis is true if the specimen of the component is behaving normally and has not degraded;

compute the probability that an alternative hypothesis is true using the current value of the inferential variable and historical values of the inferential variable, wherein the alternative hypothesis is true if the specimen of the component has degraded; and

to

compute the likelihood value by dividing the probability that the alternative hypothesis is true by the probability that the null hypothesis is true.

18. The computer system of claim 17, wherein the alternative hypothesis test can include:

a positive-mean test, which tests whether the mean of a time-series for the inferential variable is above a reference level;

a negative-mean test, which tests whether the mean of the time-series for the inferential variable is below a reference level;

a nominal-variance test;

which tests whether the variance of the time-series for the inferential variable is proportional to a scale factor;

a inverse-variance test, which tests whether the variance of the time-series for the inferential variable is proportional to the inverse of the scale factor;

a positive first-difference test, which tests whether the time-series for the inferential variable is increasing;

or

a negative first-difference test, which tests whether the time-series for the inferential variable is decreasing.

19. The computer system of claim 15,

wherein prior to monitoring inferential variables, the degradation-detection mechanism is configured to compute the mean and the variance of the time-series for the inferential variable by monitoring the inferential variable during a training phase;

and

wherein the mean and variance are used to normalize the mean and the variance of the time-series for the inferential variable while monitoring the inferential variable during the monitoring phase.

20. The computer system of claim 15, wherein the inferential variable can represent a current passing through the component or a voltage applied to the component.

21. The computer system of claim 15,

wherein the reliability-evaluation study is an accelerated-life study of the component;

and

wherein the monitored data can be processed in real-time or can be post-processed.