Some of the common methods of reliability prediction utilizing
calendar or lot size dependence and various "k" factors
for complexity, end use, product family, and environment are
briefly reviewed. These techniques, although sometimes accurate and
simple, do not provide adequate reliability tradeoff information
and do not fully treat the effects of vendor variability and
improvements in the state-of-the-art. From these techniques it is
frequently difficult to understand causes of observed failure rates
and to determine what can be done to achieve cost-effective
reliability.
One method of alleviating these objections is to derive a model
based on microcircuit failure mode and mechanism knowledge. The
current status of microcircuit failure knowledge is reviewed.
Considerable use is made of data which has been acquired at very
high stress levels and the relationship between this data and end
use data is discussed.
Microcircuit failures can be grouped into four essentially
independent categories: failures due to time degradation, failures
due to mechanical stress, serial failures due to two or more events
occurring in succession, and parts which never worked. A failure
prediction model based on failure modes and mechanism knowledge
would contain each of these elements. The resulting model might
prove unwieldy and of dubious value because of its size. However, a
simplified model can be employed. If the user knows how it was
derived, it can be a useful and flexible tool. One form that such a
model might take is presented and discussed.