The major influence on the reliability of electronics is temperature, yet thermal/fluid modeling is plagued with uncertainties and unknowns. Nonetheless, if appropriate values of these unknown parameters are available for any specific electronics package, then its temperature response can be accurately predicted using modern thermal/fluid analysis tools.
Traditionally, uncertainties are dealt with by a combination of testing, safety factors or margins, and worst-case design scenarios. Analyses are performed iteratively in a repetitive “point design evaluation” mode. Computer-based design simulation tools have emphasized increasing detail and fidelity to physical phenomena, seemingly ignoring the fact that the inputs to these simulations are highly uncertain.
This paper describes both current and future methods of dealing with uncertainties in thermal engineering. It introduces advanced tools and alternative methodologies that can automate not only the quantification of reliability, but can also help synthesize designs on the basis of reliability. It advocates using rapid gains in computer speed not to increase the degree of detail in a model, but to help the engineer find a robust design by automating high-level design tasks.