A methodology to assess microprocessor fan reliability

Microprocessor fan reliability is a critical part of the overall reliability of the thermal solution in microelectronics products. It is difficult to compare various fan life data due to differences in definition of failure, acceleration factor, sample size, failure distribution and resulting calculation of MTTF (Mean Time To Failure) or L₁₀ (time at which 10% of the population has failed). This paper addresses these issues in detail. Fan failure data is analyzed and an acceleration factor model (for temperature acceleration of lubricant degradation) based on Arrhenius functional form is recommended. Various acceleration factor models being currently used are compared to the chosen model. If failures exist, Weibull distribution should be the distribution of choice. For test planning, however, a Weibull distribution requires an assumption for Weibull slope, β. A default β value of 3.0 is recommended. Starting from a target design fan life, the paper outlines a step-by-step approach to accelerated test design or interpreting/validating existing reliability data. Since unaccelerated life tests to failure are prohibitively long, the paper provides a criterion to evaluate/compare zero fails data based on Weibull distribution and sample size. A sensitivity study on impact of sample size and using a different Weibull slope assumption is included. Both Weibull slope and L₁₀ together are important for reliability characterization of microprocessor fans.