There's actually not much of a difference at all.  Generally, speed grades all come from the same wafer.  They grade the parts, much like the tolerances on resistors and mark accordingly.

So long as you do not voltage overstress the parts, they should be OK.  There are some potential hot-electron effects that could degrade performance over time, particularly if the parts are used at the cold extreme continually, although, unless you heatsink the parts to the ambient, once powered up, they're probably running 10º to 15º degrees warmer anyway.

TTFN

I agree with IRstuff's assessments.

I would probably restrict the CPU or any other part to clock speeds that aren't near limits.

Generally low temperatures won't shorten part life but will extend it.  So if the part works today it will work in two years.  The problem is in speed reduction.  Also a part taken below its rating can have structural problems due to expansion/ contraction aspects.

I would talk to the manufacturer, they can suggest specific concerns or lack there of.

Keith Cress
Flamin Systems, Inc.- http://www.flaminsystems.com

From my own practical experience with this sort of thing, the biggest problem may be the internal clock oscillator circuit. It may be better to drive the processor from an external commercial clock oscillator module qualified to start and run at the desired low temperature.

Designing crystal oscillators to start and run over extremely wide temperature ranges is far from trivial. Best buy a module rather than use the processors own internal feedback loop with an unknown low cost crystal.

That is probably the only analog part of the whole system, and unless it has an internal analog to digital converter or some other analog circuitry, the digital part of the processor should work fine.

I saw one unit that was supposed to operate at lower temperature.  They bridged a 330 ohm (5V) resistor over the top of micro and dabbed a bit of epoxy on it.  Either they know something or are just hopeful.  Maybe they expect the system starts in good weather and that just keeps it up to temp, or they don't care if it takes 30 minutes for the clock to start.  I suspect the percentage of units that won't work at low temperature is very small.

Zenisthemon: Warpspeed is definitely right about the processor clock, however you may need to pay just as much attention to an external packaged oscillator module as using the on on-chip one, particularly if you are running your processor at its maximum clock speed.

You don't say what the processor is, but we have found that even some "military" grade processors, which are qualified to run at -55C, either won't start or crash at -10C unless the clock waveform is as near-perfect to that stated in the processor data sheet. This is particulary true of meeting the processor's clock minimum width requirement: many packaged oscillator output waveforms don't have 50:50 mark-space duty cycle, and can also have slow rise/fall times which effectively shorten the clock minimum width even more.

MOS transistor mobility increases as temperature decreases, so generally, CMOS and MOS processors run faster at cold, hence the various geeks who try to run their micros in LN₂ environments.

However, quartz oscillators probably go the other way, so it may be possible that the slowness of the edges of the clock causes race and metastability problems in the processor at cold.

TTFN

As Warpspeed pointed out, digital clock oscillators are the issues with some lower temperature applications. The oscillators may not start.

As IRStuff point out, With semiconductor parts where there is a commercial version and an industrial version with different temperature specs, frequently the die is the same and the difference is that one was tested for the lower temperature, and the other was not.

However, you indicate the issue is a CPU and a DSP. Are either in a BGA-type package? Some manufacturers limit their BGA-type packages at the lower temperature - there is a concern with the different expansion rates of the package and the PCB. This could lead to failures after a number of temperature cycles. I don't have data at hand, but I believe there is additional concern if the solder is lead-free.

It depends on the level of testing.  Unless you can thoroughly test EVERY operation and EVERY condition at -25ºC, you cannot be sure that the device can be guaranteed to work at that temperature.

Our processor was supposed to operate at -55ºC and was tested on our own tester.  It was more than a year into production that we found out that we missed a worst case carry-ripple that only operated to -35ºC.

TTFN

We made a similar discovery with some Australian designed telecom equipment we export to China. It was tested to -20C in our own environmental chamber, but the northern winters in China can sometimes reach -45C in a very bad year.

Some very fortunate connections gave us acces to a massive defense department environmental chamber that allowed testing down to -80C, and a whole lot of unexpected new problems became evident. The clock oscillator was the biggie, but also some of the larger bypass capacitors lost so much capacitance that the power supply regulator could sometimes burst into oscillation.

Nothing else just plain "burst"?

Keith Cress
Flamin Systems, Inc.- http://www.flaminsystems.com

No, no physical damage at all that we could see, and it continued working fine with some further testing.

The instant frost, condensation and resulting mini flood when it was finally removed from the chamber was fairly spectacular though.

Generally, there's minimal trapped moisture in parts during manufacture.  Ceramic packaged parts are usually sealed under a nitrogen blanket, and similar conditions exist for most other components.  The reason, of course, is that trapped moisture will cause corrosion and/or shorts, so just to maintain product life requires elimination of moisture.

TTFN