RMQSI Answers ForumCategory: Electrical Component ReliabilityCalculations for "Hybrid Microcircuits"
rowand asked 12 years ago

I’m trying to work out the reliability of a circuit, and I’m down to the final component, but I’m unsure about how to perform the reliability calculations on it.

The component I’m stuck on is a “Power Integrations LinkSwitch-TN Off-Line Switcher IC”. The device combines many subcomponents (MOSFET, logic circuitry, regulator, etc) into one IC.

Using the MIL-HDBK-217F I would assume that I should analyse it as a “Hybrid Microcircuit” and break it down into its subcomponents and calculate the individual failure rates before summing them together to get the total failure rate, but without knowing the exact internal circuitry, I’m not sure how best to analyse it.

Any suggestions would be appreciated.

1 Answers
smorris answered 12 years ago

Generally speaking, the MIL-HDBK-217F, Notice 2 is not that accurate. This because it has not been updated in over 13 years. Even when it was more up to date, there was always raging debate regarding its accuracy. Preforming predictions using MIL-HDBK-217, and similar sources, is often half “art” and half “science” (some would argue 1% science, 99% art). Therefore, while you can apply the hybrid model for something like this, my approach is most often to use the a conservative part category/failure rate estimate (art). For example, in general, linear mocrocircuits have a higher failure rate than digital devices. Since the part in question contains a regulator, this would push me toward selecting a conservative failure rate from the linear microcircuit category, typically by selecting a failure rate representing a more complex linear device (i.e., a category with a greater transistor count than the regulator portion of the circuit itself), to account for the other portions of the circuit I am ignoring.
You should view MIL-HDBK-217 results as a “ball park” estimate and document the assumptions behind your analysis. If you need more accuracy on a specific device, I would recommend trying to get actual vendor test data. I am responding from the perspective of someone who most often needs a “quick and dirty” failure rate estimate to feed into a higher level system level model that usually contains lots of redundancy that we are trying to model. The low level inputs, such as what part category to use, become much less important at this level. However, for situations where they do become important (e.g., the device is used in something that is non-redundant), then we go back and take a closer look at the lower level modeling and/or try to further confirm our failure rate estimates from other sources (actual vendor data).