What's also important is to get it flat, as opposed to smooth.
Absolutely crucial.
The ideal is a machining process that yields as close to a fine
quality surface as possible, then a surface grind or even follow that
by a finishing with one or more steps of abrasive polishing.
The idea is to have any polishing be as minimal as possible so as
not to undulate the surface flatness notably.
Sanding
may make it shiny, but it's also liable to grind a curve into a
surface,
Homogenized (read across the whole surface) light "sanding" on a
surface removes media more evenly than you might initially think, if
done using the right methods.
and that will wreck the thermal contact and in the extreme
result in insulator damage.
Cupped or undulated surfaces indeed are not good heat sink
interfaces. I am quite sure, however, that John meant a more
regimented polishing technique.
In the extreme, two mirror finish surfaces that DO have good flatness
characteristics will be the absolute best thermal interface.
Extrusions tend to be too curvy/wavy for
really good thermal contact.
Most are extruded thick, stress relieved, faces get machined flat
(like a jointer in woods), then it gets stress relieved again, and
hard anodized likely after it gets cut to the customer specified
lengths. There are excellent extruded sinks made though.
I recently did a 1.5 kilowatt design with about a 20" inch long
2.5 inch wide by 5.5 inch tall sink was used for some 24 HUGE IGBTs.
They were ALL soldered in on exactly the same plane. All mechanical
clamping (no screws) and mounting of the sink and devices were done
before any solder joints were formed. I taught the girls how to
assemble these right to keep us from getting blow ups in the test lab
after assembly.
When one of these big tab devices runs away, it ain't pretty.
Letting the smoke out stinks too. :-]
For really good thermal performance,
extrusions should be machined flat, like with a slow pass with an end
mill or a fly cutter.
Any machining ops can be done if the primary interface was extruded
with a thick enough cross section (and it should have been). One must
just use coolants and watch cut rates.
Grease will fill surface roughness pretty well, but curvature can open
up a gap that will seriously increase theta.
Absolutely true. Ideal is two flat surfaces (sounds like a
consumer electronics store chain).
Minor things like surface
waviness can throw away a good fraction of a transistor's dissipation
capability. A reasonable goal is 100 micro-inches total roughness plus
flatness across the transistor footprint. You can buy BeO/AlO2/AlN
insulators that are smooth and flat to a few micro-inches.
Yes, and as we knoe <sic>, the device are good and flat, so if the
sink is off badly those insulators will fracture from the pressure
gradients on the high spots.
Power transistors themselves seem to be remarkably flat.
They got standards to keep. And face too! :-]
Hahahah... in more ways than one.
This sort of stuff starts to matter when you want to push transistors
up against their thermal limits, like in fast/RF stuff where you can't
just add more parts in parallel.
You Da Man!
I seem to remember a 50 Watt single device.
Actually, if one looks REAL close, that IS about 8 FETs in parallel
on one substrate element in one package. What was it? 50W at 10GHz.
Can't parallel up at high freq on a big breadboard. Too many
parasitics and such, eh?