By Professor Joseph Bernstein
To this day, the users of our most sophisticated electronic systems that include opto-electronic, photonic, MEMS device, etc. are expected to rely on a simple reliability value (FIT) published by the supplier. FIT is incorrectly determined today due to the product qualification use of HTOL (High Temperature Operating Life) and JEDEC or other standards. Manufacturers reports ‘zero-failure’ data from single-point tests using a single-mechanism model to fit an expected MTTF at the operator’s nominal expected ‘use’ conditions, giving erroneous and misleading results.
The zero-failure qualification is well known as a very expensive exercise providing nearly no useful information to the user. As a result, designers often rely on HALT testing and on handbooks such as Fides, Telecordia or Mil Handbook 217 to estimate the failure rate of their products, knowing full well that these approaches act as guidelines rather than as a reliable prediction tool.
Furthermore, with zero failure required for the “pass” criterion as well as the poor correlation of expensive HTOL data to test and field failures, there is no communication for the designers to utilize this knowledge in order to build in reliability or to trade it off with performance. Prediction is not really the goal of these tests; however, current practice is to assign an expected failure rate, FIT, based only on this test even if the presumed acceleration factor is not correct.
We expose, in this tutorial, the actual lies that are propagated today based on incorrect use of statistics by JEDEC and other standards organizations. We then demonstrate a simple way to achieve accurate predictive reliability assessment by way of “Failure In Time” (FIT). We will evaluate the goal of finding MTBF and evaluate the wisdom of various approaches to reliability prediction. Our goal is to predict reliability based on the system environment including space, military and commercial. It is our intent to show that the era of confidence in reliability prediction has arrived and that we can make reasonable reliability predictions from qualification testing at the system level. I will demonstrate how physics of failure models in conjunction with qualification testing through a Multiple Test Operational Life (MTOL) matrix solution makes cost-effective reliability predictions that are predictive and based on the system operating conditions. Furthermore, we will show experimental evidence that the thermal activation energy is non-constant over the operational temperatures as well as a non-constant voltage acceleration factor in standard devices.