Originally Posted by overclockerfx
As a chemical engineering student I might say that processors, in terms of voltage and temperature, are limited by physical properties of the chemical compounds they are composed of. NOTE: I'm not any kind of expert on processor technology, but just based on what I know about chemistry and the thin-films used in processor manufacture I can say this.
The latest processors are getting to so small transistor sizes (22nm) that there aren't too many atoms in the transistor (just 200 silicon atoms - and even less with all the other larger atoms they tend to incorporate in transistors), which means there's a greater chance of leakage as there's less of a "buffer" between the transistors and also less room for electrons (-) and holes (+) (charge conductors in a semiconducting material). So if you pump more voltage across a circuit it also means there will be more charge flowing through it (higher current) if the ratio of voltage and current remains unchanged. What this means is that there will be a lot more electrons trying to get into the circuit which might eventually cause short-circuits as they don't all have room to flow through and it's energetically more viable to jump to another transistor i.e. short-circuit. This obviously can cause significant damage and is often the extreme case, which would probably require a LOT of voltage. Anyways the probability of leakage increase with increased voltage and therefore reduces the life-expectancy of your processor.
Temperature on the other hand relates to the material properties of the die. These depend on the chemical compounds. Obviously silicon dioxide. which is essentially sand doesn't melt very easily, however nowadays there's plenty of other materials in the transistors that can start to thermally degrade to some extent already at the temperatures that we usually talk about here i.e. +100C.
With regard to this, I'd say that temperature is less of an issue of "thermal degradation" of compounds as it is an issue of diffusion. As you mentioned, these transistors are on very small scales, (hundreds of atoms), so I think the issue with temperature is that the doped semiconductor materials usually used in the source/drain within the transistor will slowly degrade over time due to thermal migration of the doped atoms between the p-type and n-type regions. I.e. looking at this picture:
At any temperature (above 0 K), you will get diffusion of the p-type dopant into the n-type channel and vice versa. Over time (a very significant time at 25 C), you will reach a state where the distribution of dopants is exactly the same over the source, drain, and channel, and at that point the transistor will not work at all (and probably will stop working long before that due to leakage current as the semiconductors lose their doping). Since this is due to diffusion, and typically diffusion is proportional in increase to T^(3/2), you're seeing an exponential increase in degradation due to dopant migration at higher temperatures. It's time based, so it's unlikely you'll kill your chip immediately in virtually any case where you go to a high temperature because of this, you'd have to be more worried about something melting/burning at that point. But over time, higher temperatures will lead to this process happen significantly faster as many of you have already conjectured.
It's probably possible to calculate exactly how long this process would take at various temperatures if the dimensions and dopant levels of the current Intel finfet transistors are known. I also think that this is probably where companies like AMD/Intel get their lifespans from, and target operating temperatures. For example, AMD has said that their new R9 290 card is supposed to operate at 95 C for it's lifetime, and that it's not an issue. I've personally thought that it's silly that people complain about the temperature on the cards after this, because really what AMD is telling us is that at 95 C, the card's structure will not change due to diffusion to make it unusable during it's lifetime. They probably have done this calculation on rates of dopant migration. Of course this principal can be applied to other parts of the card as well, such as the cache etc... and I don't really have knowledge about how those structures look, but in theory they should suffer from the same issue.
With regard to voltage, I was always under the impression that higher voltage just means that you can have significantly higher levels of quantum tunneling through regions where otherwise you wouldn't have it. I can't really think of a reason why increased electron movement would actually cause damage. Perhaps I'm thinking about this wrong, but it's not like the electron is crashing through the channel or the gate dielectric and forming a short circuit path in it's wake. It's really just jumping from one side to another via the quantum tunneling phenomenon. My guess would be that higher voltages lead to high local temperatures due to increased resistance, and this could possibly be causing localized defects, particularly on the interfaces of different materials, such as the metal in the source/drain and the semiconductor material next to it. Possibly why you don't have chips dying at 2V when they're cooled by LN2. I can image that your CPU probably would instantly die on air at that voltage.
This is all speculation though, similar to the above poster I am a Chemical Engineer, so most of this is based off of how materials tend to act under various physical conditions.