The determination of component part failure rates is required for most military contracts normally during the early portion of the program definition phase. For new technology devices whose failure rates have not been established nor available in recognized military documents, it has been impossible to accurately predict the reliability of the system using these new devices. A technique developed by authors enables a prediction to be made for new state-of-the-art devices using a mathematical transfer model.
This paper describes a mathematical transfer model using established failure rate data from bipolar integrated circuits (IC). These bipolar devices contain similar characteristics, basic materials, processes, testing, inspection, environmental, physics of failure, and application, to the new device. Failure rates for the new device, large scale array (LSA ) metal-oxide-semiconductor (MOS) are obtained through use of this model.
For the example described in this paper, extensive empirical bipolar IC postmortem analysis was analyzed, normalized and related to the commonalities of MOS technology. The failure mechanisms were grouped into the following major defect categories: (1) passivation, (2) metallization, (3) leads, (4) diffusion, (5) die bonding, (6) foreign material, (7) packaging, (8) contamination, and (9) material discrepancies. Each of these defects was then normalized to diffusion area, metal area, number of leads, average lead pad size, thin oxide area, gate area, crossover area and internal package area. A transfer model was then established to estimate the percent of the failure mode applicable to the MOS device along with the justification. The result of the completed transfer model enables a reliability prediction to be made by summing the contributions of each major failure mechanism based on the applicable areas and number of leads. A ten percent contingency factor is added to this total to account for failure mechanisms unique to the MOS device.
Since all of the empirical data was derived from a high reliability program, a table was developed to adapt the transfer model for programs with lower reliability control levels.
This model is readily adaptable for other devices new to the state-of-the-art where reliability has not been established.