Previous	Next

This is the second key question that leads to understanding reliability. Just as with the question about failure mechanisms, the root requirement for answering this question is ...failures. Without causing representative failures, the very best that can be done is determine when failures WON'T occur. Unfortunately, this is also the goal of the tests commonly performed at 125°C for 1,000 hours, or at 150°C for 512 hours. These tests almost certainly will not generate failures. With this result, you can't say anything about the device reliability except that failures probably won't occur under those same conditions. Without failures, there's no information to predict what might happen after longer test durations or higher temperatures. This puts the user at high risk since he or she normally expects devices to have a useful life of roughly 20 years at his or her operating temperature, say 85°C.

Analyses of these failures not only answers the question about when failures will occur, but also enables predictions to be made for a circuit based upon a knowledge of its operating conditions. These failures can be used to judge all future process changes and evaluate particular design parameters.

In answering the previous question about failure mechanisms, it was stated that failures do not occur at a constant rate. This is known by characterizing failure distributions under highly accelerated conditions. As with other semiconductor companies who've performed this analysis, TriQuint has found GaAs devices fail in "lognormal" failure distributions. Times to failure for each of the GaAs elements manufactured at TriQuint are similar. For FETs at 260°C, it was shown above that half the failures occur in about 800 hours. The other elements can be as easily defined, but each one was tested under the optimum conditions to produce representative failures, and none of the results have a common base. It's also probably not very useful to know how long it takes MesFETs to fail at 260°C if the maximum specified temperature is 150°C. To present the results in a common format you must know how temperature affects the failure mechanisms.