> This only works efficiently, though, if a very thin layer of thermal paste is applied between CPU and heatsink in such a way that minimises the chance of creating “air bubbles” (air has a bad thermal conductivity).
The issue here isn't really air bubbles. It's true that air has very poor thermal conductivity, but the real issue is that the thermal paste itself has very poor thermal conductivity compared to the solid copper heatsink. The paste is really just supposed to fill in any microscopic pits in the surface to get the copper in good contact. People go to great lengths to make the surfaces as smooth as possible, going as far as to hand lap their heatsinks to a mirror polished finish.
Indeed, TIM is a poor heat conductor compared to metals. If surfaces were ideally polished, TIM would be absolutely not necessary. We’ve done this and saw significant improvement in thermals. Next best thing is liquid metals that have an order of magnitude higher thermal conductivity than anything like arctic silver. Big problem though: they are also electrically conductive so if a drop gets under CPU pins that’s the end of it.
Without going these exocitic routes, the goal of using TIM is to fill up gaps (due to surfaces not being ideal) to replace no thermal conductivity in those places with some thermal conductivity. What you do not want to do, however, is put TIM where there was already good metal-on-metal contact. That’s why too much TIM is definitely bad.
The maximum thickness of the paste layer is determined wholly by the clamping pressure and coplanarity of the thermal interface. You can only end up with an excess of paste between the CPU and heatsink if a) the heatsink contact plate or CPU package aren't flat or b) the clamping mechanism isn't properly pressing them together.
Apply too much thermal paste and it'll just squeeze out of the edges, which is merely inconvenient as long as the paste is reasonably non-conductive. Apply too little paste and you'll get air gaps, which have terrible thermal conductivity. If in doubt, just spread the paste to ensure that it fully covers the contact area.
Yes, you always want the minimum amount which will ensure good contact across the whole surface. You should be pressing (hard) to mate the surfaces and squeeze out excess thermal transfer medium anyway. Desktop/server CPU sockets typically have a mechanical lever action to do that for you and secure the package in place.
> This only works efficiently, though, if a very thin layer of thermal paste is applied between CPU and heatsink in such a way that minimises the chance of creating “air bubbles” (air has a bad thermal conductivity).
The issue here isn't really air bubbles. It's true that air has very poor thermal conductivity, but the real issue is that the thermal paste itself has very poor thermal conductivity compared to the solid copper heatsink. The paste is really just supposed to fill in any microscopic pits in the surface to get the copper in good contact. People go to great lengths to make the surfaces as smooth as possible, going as far as to hand lap their heatsinks to a mirror polished finish.