The first thing to mention is that these grades are all manufactured identically, and their performance is measured so that they can be put into one of these three bins. There might only be two real bins, but at this point it's speculation.
The temperature plays a part in the speed as well - the higher the temperature, the lower the maximum clock frequency. At cold temperatures, the chip will run faster. Cold temperature failures are usually hard failures, in the sense that reducing clock frequency won't fix things. "Worst case" depends on your application. Here are a few scenarios that could happen.
- The PLL could fail, and the chip would not operate.
- The memory could enter a failure mode (can't write or can't read reliably)
- Data corruption due to datapath hold-time violations
- Improper behavior due to hold-time violations
- Excessive EMI due to uncontrolled transitions in the output busses
There is a distinct possibility that they only qualified the industrial temperature range part because a wider specification means more time and money. In that case, all grades will probably work down to -40C. There may also be more durable packaging with the industrial range part.
If you are using this part for a hobby project, you may be comfortable "risking it". You may also be able to qualify individual parts, but any manufactured device will be a tough sell without the wide temperature range chips.
2nd Edit! Modified my answer about semi-conductors based on jk's answer below, read the history if you want to see the wrong bits I modified!
Everything gets weird within certain limits. I mean, sure, the resistance improves in conductors but it increases in semi-conductors, and that change effects how the IC works. Remember that the way that transistors work on the basis that you can modify their resistance, and if the temperature drops so low that you can no longer decrease their resistance, you've got an issue! Imagine that suddenly your semi-conductor essentially became a resistor... how do you control it? It no longer behaves the same way! Now I'm a bit confused at where you're getting the -25°C, as the industrial/military spec should put it at -40°C for the minimum operating temp.
But for the space question, I can answer that as I work in a space lab! In general you have three thermal concerns in space:
1) In space, you only radiate heat. Radiation is a terrible way to get rid of heat. In the atmosphere, you conduct heat into the air around you which makes cooling a lot easier. So in space, you have to put big heatsinks on to get the heat into larger radiative surfaces.
2) If you have a component which doesn't generate heat, then space is happy to let you get really friggin' cold! In general, what you do is you have active heating elements to keep components which don't generate more heat than they radiate but have thermal limits.
3) Heat swings are common because you will exit and re-enter the sun's rays. Thus you need to have active thermal management where you have a big heatsink which can radiate heat when it's hot, and a heater for when it's not.
You can also get extended temperature range devices which go lower and higher, but there's pretty much always a limit. Some of them are for where the cold temperature will crack the die because the metal will shrink more than the plastic (or vice versa) which is why they list limits for storage as well!
The limit is mostly in materials. You also tend to get space-rated chips made out of ceramic for the packaging, which can also raise or lower the thermal limits.
Anyway, I hope that explains it for you. I can try and answer any other questions, but I'll admit the physics of low-temperature semiconductors is not my forte!
1st Edit:
Here's a link to a wikipedia entry about the idea that at lower temperatures there are fewer electrons which are excited enough to generate a current flow through a semiconductor lattice.
This should give you a good idea of why the resistance becomes higher, and why 0 Kelvin would have never been an option.
Best Answer
The highest CASE temperature mentioned is in a diagram on page 9 - and max is about 110 C.
The highest device temperature quoted, apart from soldering temperaures is 150 C at the top of page 2 - and this is JUNCTION temperature.
Ambient temperatures are relevant but not under designer's control.
The JUNCTION is the actual IC core, inside the chip.
Junction temperature relative to case is determined by power input and Rjc = thermal resistance junction to case. BUT the datasheet seems to be bad in this respect and give only Rja (shown as Theta_ja) = resistance junction to air. You can see by using Rja whether you need a heatsink and whether that the datasheet should give Rjc. Looking at the figures:
Rja with 2 layer board = 27 C/W.
Device is said to be 93% efficient.
At 20W in 7% is dissipated at heat in IC = 1.4W.
At 27 C/w IC junction will rise above air by 27 C/W x 1.4 W ~+38 C rise.
Tj max = 150 C, so allowable ambient / air / board temperature is 150C - 38 = 112C.
If we use the 37 C/W Rja for single layer board we get differential rise junction to air of
37 x 1.4W =~ 52C rise over Tair, and max air/board temp of 150 - 52 = 98C.
Gven the various assumptions made, 98C is close to the specified 85C max - so we can see why they specify it. At 85C Tboard you can r un the IC at 20W power in and somewhat worse than 93% efficient and not (quite) exceed the 150C allowed max junction temperature.