A New Approach to the Extrapolation of Accelerated Life Test Data.

Acceleration factors are commonly used to extrapolate accelerated life data to operating conditions. This method assumes that the failure time distributions belong to the same scale family under both the accelerated and operating conditions. In our studies of the complicated material systems used in interconnection technology, we find this assumption is generally not valid. Consequently, we have developed an alternate approach based on simple rate kinetic approximations to the degradation process. These models lead to acceleration transforms which map the failure distribution at one set of stress conditions to the failure distribution at another set of stress conditions. When the kinetic process is governed by a single step, the mapping involves simple multiplication by a scalar factor so the acceleration transform reduces to the usual acceleration factor. Acceleration transform models have been developed for six types of degradation reaction. We have used this approach in combination with a designed, statistical, temperature-humidity- bias accelerated life test to study failure in a printed circuit material caused by the growth of conductive filaments along the glass reinforcing fibers. The kinetic rate model which best described the failure time distribution changed from a simple single-step reaction model at high relative humidities to a competing reaction model at lower humidities. This change clearly identified a humidity threshold for the degradation process. As a result, the acceleration transform estimate of the printed circuit failure rate at operating conditions was considerably lower than that obtained from a simple acceleration factor estimate. When the acceleration transform method is combined with efficient experimental design techniques, the result is a powerful and accurate method for interpreting accelerated life test data. A major advantage arises from the fact that extrapolation can be interpreted in terms of physical processes and thus physically unrealistic models can be rejected.