silicone grease

The big problem is likely that one must design for it to be dry, so when the grease drools away, the product won't melt.

Joe Gwinn

I've used graphite cloth as gap filler. It seems to work and it isn't messy.

Meas. Sci. Technol. 7 (1996) 1653–1664.

Silpads. I used silpads since the 80's. I don't know if they are better or worse than grease, but they are good enough and most importantly they are consistant. What happens with testing on the prototypes, happens with production units. Shorts were always with the little vinyl grommets, you use to keep the screws from shorting to the part. You can squeeze those down to tight and cause a short.

ISTR on one of the overclocking hacker CPU cooling sites someone tried everything from dry to cooking oil and engine oil. The marginal best was some exotic "liquid metal" silver loaded brand I have never heard of and the worst by a long way was dry.

The biggest change was from dry to some sort of heat exchange medium is by preventing an air gap. It was a significant difference too.

The problem is that your flat surfaces are not exactly flat so that the direct metal contact area can actually be quite small if there is any surface roughness. Air is a rather good insulator and metals don't radiate well at all. Silicon grease prevents air gaps and anything similar will do the same job. It is just that silicon oils and greases are less inclined to evaporate or go rancid and corrode your parts.

There is no such thing as flat. There is only different levels of rough. Extremely flat is extremely expensive.

In theory, when you bolt two metal object together, they will only contact at three microscopic points.

We always put some solder onto the tip of the soldering iron before soldering. Otherwise the heat will simply not transfer to the workpiece, even if both the soldering iron and the workpice are flat.

Diamond doesn't melt at room temperature. There is a diamond/graphite/liquid carbon triple point at 12 GPa at 5000 K, but your lump of metal wouldn't get hot enough to get anywhere near that.

The three microscopic points will distort so that that the compressive force is spread over three finite - but small - areas.

With very clean surfaces under vacuum you get microwelding as the metal at the microscopic contact points diffuses into the surface it is squeezed up against. This can be a problem if you wan to dismantle your gear later.

Most manuals on basic electronics will include tables of typical Rth for various mounting media. This is drawn from empirical data that you can duplicate on your own bench.

If isolation is not required, I've had excellent results introducing soft copper wafers into the junction. These seem to deform to reduce issues with surface flatness. Also useful when efficient high-current conduction is required between the two surfaces.

Properly-mixed silicon oil won't make a mess if properly applied. If you recheck older junctions, you'll see that they are fairly hard to break, due to their air-free composition.

This is old tech.

RL

Numbers can be found. For instance see page 8 of Philips AN10384

piglet

I expect they will stretch when heated, so you might see increasing failures over time. Not an easy problem...

It was quite a simple setup.

Same heatsink, same stress test and note down the CPU core temperature at equilibrium. CPUs are convenient in already being well instrumented - the biggest difference was nothing vs anything else.

There is an 80:20 rule at work here - you get 80% of the improvement by eliminating the tiny air gap by wetting it out with a heat transfer medium and the rest is incremental using ever more exotic materials.

In the extreme they still use the near lethal BeO ceramic material in some high power RF transistors since it is second only to diamond for thermal conductivity whilst being an electrical insulator.

You say that there are no numbers. Where have you been looking?

If you are prepared to de-rate accordingly then there isn't really a problem but if you want to run them at full power then they need to be in intimate contact with their heat sink and that means wetted by some sort of heat transfer medium. I was quite impressed with the bluetack like stuff that came with my Raspberry Pi passive aluminium heatsink.

I found the pads more annoying to handle than silicone grease YMMV. Getting them on square was much harder than just adding a dab of goo.

It's not stretching that's the problem, but cold flow or creep "the tendency of any solid material to move or deform over a period of time under the influence of persistent mechanical stress, with no recovery of shape when the stress is removed."

The recommended mounting torque is about 6lb-in and the plastic screws I am familiar with have max rated torques much less than half that.

Even if you find a plastic screw reliable for that torque including a conical belleville compression washer might be good to accommodate slippage and creep.

I suspect a good spring clip may be more consistent and quicker in production.

piglet

Designing audio amplifiers, it often worked out more economical to use two output devices run gently, rather than a single one pushed to its limits. The two thermal paths between die and heatsink are effectively in parallel and carrying half the power each, so you can run the die cooler [preferred] or the heatsink hotter [if you really must].

You also keep the device well below its maximum current rating where the current gain is higher and it needs less drive power. Economies in the drive stage and current protection may well compensate for the cost of an extra lump of silicon and a couple of resistors.

The Bergquist pads I was looking at are theortetically 1.8 W/mK material, but the actual TO-220 theta suggests about half that in real life.

T-global makes pads that are rated 5 W/mK.