Electronics Assembly Failure Modes Past the Rated Temperature Extremes??.....

[Smokey]

What are the actual failure modes at the extremes of the temperature ratings for PCB Assemblies, both hot end and cold end?

Here is a list of the typical ratings:
Commercial = 0C to 70C      (32F to 140F)
E1 Extended = -25C to 75C (-13F to 167F)
Industrial = -20C to 85C     (-4F to 185F)

Then the Automotive stuff:
Grade 0 = -40°C to +150°C (-40F to 302F)
Grade 1 = -40°C to +125°C (-40F to 257F)
Grade 2 = -40°C to +105°C (-40F to 221F)
Grade 3 = -40°C to +85°C   (-40F to 185F)

Crystals not starting, caps drying out, thermal cycle breaking solder joints, etc are obvious ones. 
But what goes wrong in something like a processor?

[TheMG]

Processors especially ones with high clock frequencies can become unstable at high temperatures (lockups, crashing, program corruption).

[TheUnnamedNewbie]

A lot of the failures might not be obvious failures that are catastrophic, but rather something like failiure to meet spec. For examples, analog circuits that have lower SNR, less gain, lower linearity, etc. Also, lower lifetime, which is something you will only notice after a long time (A cap that is rated to 85C won't suddenly become a short at 86C, or even at 90 or 100C, but it won't live that long...)

[Berni]

It just means how far the manufacturer grantees the device to operate within spec.

If you actually try to run the components outside the range you will find they still work until you get to some very extreme temperatures. Last time i tested my board at high temperature that had components rated for up to 105°C i went past that and ended getting the board to 150°C and it kept running perfectly fine. I stopped there, since it was already more than enough, but it is possible it would have kept running until the solder would melt and components fall off.

In general what tends to fail are semiconductors. At the extremes of temperature the characteristics of transistors inside chips could drift far off enough for the chip to start misbehaving in weird ways. As said above particularly sensitive are high speed digital things that are already running on the edge of how fast the transistors can go, so things like fast SoCs, RAM etc... where they might start making mistakes and crash. Since they are running to close to the limits means that small changes in the transistor characteristics can make enough of a timing difference to break the logic.

[CaptDon]

We did loads of 'HALT' testing at GE and I always disagreed with the results. They would push stuff to extreme limits and try to interpolate life expectancy. Totally bogus results and field failures not matching predicted results was the proof of the bogus theories!! (But you got to keep those rocket scientist test guys employed) How does one of those 'wet bag' super caps exploding at high temperature predict how long it will live when it will never see a temperature over 80C in real life? Anyway, aside from parts simply exploding we did find that magnetics tend to saturate and simply not support their intended job above around 110C. Some magnetics seem to undergo a permanent change at high test temperatures. Most of the extremely hard to cool logic blocks like ZYNC or ARTIX would stop working until returning back to in-spec temperatures and most of the ethernet stuff involving magnetics would fail at high temps. We test our locomotive stuff to -40C and most everything seemed to work o.k. as long as there was no condensate. Even the coated boards were affected by condensate!

[Bassman59]

What are the actual failure modes at the extremes of the temperature ratings for PCB Assemblies, both hot end and cold end?

Crystals not starting, caps drying out, thermal cycle breaking solder joints, etc are obvious ones. 
But what goes wrong in something like a processor?

Operation outside of specified parameters is "not defined." It means anything can happen.

Maybe the processor will lock up. Maybe a memory location will get flipped. Maybe a sinkhole will open under a highway in the desert. Maybe my cat will start speaking English. No one can say.

[Smokey]

Are there any research papers that investigated this type of stuff?  It seems like there was some government/academic research paper studying everything else related to reliability.  Like I can point to a few papers on tin whiskers with great plots and graphics and all that stuff. 
Where is the hard research results on failures at temperature extremes?  You would think NASA would have a ton of this considering space is always at one of the extremes.  I can't imagine the JPL guys building mars rovers are saying "eh, this should probably be fine, but who knows?!"

[exmadscientist]

I used to work with a guy who had done this stuff in the past. It's reasonably common for military applications as almost nothing off-the-shelf will meet military temperature ranges. As such there are people and places that specialize in it, though it's more departments than standalone shops (or at least used to be). These places don't talk too much about what they do because it's how they make money.

Your best bet is probably to buy one of these guys dinner and have a chat.

[jonpaul]

This is a long conversation and a hugh field trip study.

There are no fast and simplicity answers.

Please see numerous books and refs on reliability of electronic devices, also,  MILSTD-217

Find the " bathtub curve" of failures vs time, over life,
infant mortality, middle age and old,age failures rates.

Use law of chemical réactivité vs température, Arrhenius_equation
https://en.m.wikipedia.org/wiki/Arrhenius_equation
Réactivité doubles with every 10 dec C increase in temperature.

Bon courage

Jobs

[Someone]

Are there any research papers that investigated this type of stuff?

FPGA's have had timing models over temperature, so you could predict/estimate what timing would fail and how it would fail. That was alongside lifetime de-ratings with temperature. Its rarely just one thing, too many parameters.

[srb1954]

What are the actual failure modes at the extremes of the temperature ratings for PCB Assemblies, both hot end and cold end?

Here is a list of the typical ratings:
Commercial = 0C to 70C      (32F to 140F)
E1 Extended = -25C to 75C (-13F to 167F)
Industrial = -20C to 85C     (-4F to 185F)

Then the Automotive stuff:
Grade 0 = -40°C to +150°C (-40F to 302F)
Grade 1 = -40°C to +125°C (-40F to 257F)
Grade 2 = -40°C to +105°C (-40F to 221F)
Grade 3 = -40°C to +85°C   (-40F to 185F)

Crystals not starting, caps drying out, thermal cycle breaking solder joints, etc are obvious ones. 
But what goes wrong in something like a processor?

I have seen microprocessor systems fail to start up at very low temperatures.

The supply current of CMOS chips tends to increase as the temperature decreases and, if you are starting from a cold start, the extra current drain can overload the PSU causing the supply voltages to sag and the processor never starts up. You need to give the system a really good cold soak before testing startup performance because if the chip is slightly warmer from earlier operation it can often start up OK and maintain its temperature at a reasonable level through its own power dissipation.

[Smokey]

...

Your best bet is probably to buy one of these guys dinner and have a chat.

I'm not that kind of engineer!!!!

[Vovk_Z]

  You would think NASA would have a ton of this considering space is always at one of the extremes.  I can't imagine the JPL guys building mars rovers are saying "eh, this should probably be fine, but who knows?!"

)) Several Mars expeditions were failed (at least two Soviet ones), so possibly sometimes those kind of design is present too.