Effect of the product term on reliability prediction of integrated circuits

Abstract

Data for reliability prediction of integrated circuits is obtained either from well designed accelerated life tests or from MIL‐HDBK‐217E. Although a number of researchers have questioned the validity of prediction based on the data derived from this standard, it continues to remain the main source of failure data. Manufacturers normally specify chip reliability by a single failure rate, which is known to be incorrect. In a chosen reliability test, devices fail with different failure mechanisms characterized by their associated Arrhenius parameters. An overall failure rate for an integrated circuit is obtained by adding the individual failure rates. This amounts to ignoring the terms containing the product of probability of failure for different failure mechanisms. These terms can be ignored only if different failure mechanisms are mutually exclusive. Highlights this point, by using the published data and shows that a significant improvement is obtained by including the product terms. Demonstrates the idea of dropping the product term is sterile from theoretical as well as practical points of view and as such must be included. The chosen model and the example considered are for illustration only and the conclusion arrived at is not affected by the inherent uncertainties buried under the prediction process and associated tools. In this case it is equivalent to an improvement by 6 FITS. This approach is likely to have a meaningful impact on the economics of reliability prediction.