SRC Forum - Message Replies

Forum: Reliability & Maintainability Questions and Answers

Topic: Reliability & Maintainability Questions and Answers

Topic Posted by: Reliability & Maintainability Forum ( )
Organization: System Reliability Center
Date Posted: Mon Aug 31 12:47:36 US/Eastern 1998

Back to message list Show all replies Topics List About this forum
Original Message:

Posted by: Mok Yin Liong ( )
Organization:Defence Science Organization National Laboratories
Date posted: Mon Feb 22 21:03:50 US/Eastern 1999
Subject: PEM Reliability Modelling
I have some questions regarding the PEM reliability model given in the RAC publication "Reliable Application of PEM" : i. When we divide the failure rate obtained from this model (in failures per million calendar hrs) by the duty cycle (power on time over total time), we get the equivalent operating failure rate (in failures per million operating hrs). This PEM operating failure can then be combined (=added?) with that obtained through Mil-Hbk-217 for other components. By doing so, are we assuming that PEM has a much higher non-operating failure rate than other components (since we are ignoring the non-operating failure rate of the other components)? ii. If we assume a duty cycle of 100% (i.e. the PEM is powered up all the time), can we treat the failure rate obtained from the RAC reliability model as the operating failure rate? Does this violate any underlying assumptions of the RAC PEM reliability model? iii. Based on the limited information given for the RAC PEM model, it appeared that ESD/EOS failures are omitted. Should an additional term be added to this model to take care of ESD/EOS failures? iv. Should there be a different base failure rate for microprocessors and memory devices? Data from some publications indicate that microprocessors have a much higher failure rate than memory devices (up to ten times). Thank you.


Subject: RAC PEM Reliability Model
Reply Posted by: Bill Denson ( )
Organization: RAC
Date Posted: Mon Apr 19 11:22:43 US/Eastern 1999
Question i: By dividing the failure rate obtained from the RAC PEM model (in units of failures per million calendar hours) by the duty cycle, the equivalent operating failure rate is obtained (in failures per million operating hours). The implicit assumption in making this conversion is that all failures are attributable to operating stresses and that the failure rate for nonoperating periods is zero. It is recommended to do this only in the event that the nonoperating failure rate for the other components in the system is assumed to be zero, and the failure rate prediction from the PEM model has to be incorporated into that prediction. If the nonoperating failure rate for other components are determined, then the operating, nonoperating, and cycling failure rates can be separated by using the PEM model. However, this is not necessarily a simple process because the failure rate due to cycling stresses would need to be apportioned into the operating and nonoperating (environmental) portions since it is predicted explicitly in the PEM model. Querstion ii: Using a 100% duty cycle in the PEM model has the effect of making the operating failure rate the dominant term and makes the nonoperating (environmental) and cycling terms go to zero. In this case, it is the appropriate way to use the PEM model if indeed the duty cycle is 100%. However, if the duty cycle is not 100%, the underlying fundamental assumptions of the model are violated because the failures due to environmental and cycling stresses will not be accounted for. Question iii: The PEM model is currently being updated, and a separate EOS term is being added. These models will be incorporated into RACs Prism reliability assessment software which will be a merging of 1. New electronic component prediction models (similar in form to the PEM model), 2. The system level reliability assessment methodology RAC has recently developed, and 3. RAC databases for electronic and nonelectronic field failure rates. Question iv: The field failure rate data on which the PEM model is based did not indicate a statistically significant difference between microprocessors and memory devices. If data collected in the future indicates such a difference, future versions of the model will reflect that difference.

Reply to this message