[Avro_Arrow]

I've seen many creative methods of mounting this chip properly.
 
Here is my suggestion: Thermally conductive adhesive.
 
In my previous employment as an Engineering Technician it was
my responsibility to cobble together test equipment.
One of my pieces for testing laptop computer boards was a CPU
with an integrated heat sink and fan. With nothing to mount the
heat sink to, it was just glued to the CPU. The adhesive I used
was a 3M product but I have been unable to remember or discover
what it was. It was a single part product that came in a tube
like tooth paste. Once this stuff was cured, it would rip the die
right of the base of the CPU rather than let go.
If you have ever wondered what the connection between the die
and the base was, it is a super small BGA.
 
Here is another product I found that may give the same result:
http://www.mgchemicals.com/products/8331.html
 
Since it is a "cold" solution, they is no risk of frying your
chip with too much heat trying to solder the thermal pad underneath.
 
I'd be glad to hear other peoples ideas and solutions.

 

[wakibaki]

Here's what I did:-
 

 

 

 

 
I tacked the chips to the regular pads, soldering all of them to reduce the likelihood of the chip detaching while I was doing the large pad.
 
In all cases I used solder paste to make the connection, filling the hole and then melting the paste to solid solder with a fine tipped iron. In the case of the homebrew board (not the one shown here), I rough drilled the hole leaving some swarf flakes attached to the chip side of the board. There was some overlap of the pad and hole edges, so I trusted that the rough edge and the (small) thickness of copper and some wicking into the gap would make a good-enough connection on the chip side, and of course there was no problem overfilling the hole on the back side. I did 2 on the homemade boards which showed no sign of serious overheating in use (+/-12V). They got warm, but not finger-burning warm.
 
The thru-plated holes on the manufactured boards meant there was no problem with the 2 of those I've built. It's an M3 hole IIRC. The layout software respected the clearance between it and the regular pads.
 
In no case was there a problem with a short occurring between the pad and the regular pins.
 
None of the chips show any sign of (audible) damage from the process, but I still haven't got round to doing any instrumented test. My impression is that they're pretty robust.
 
The assembled amplifier:-
 

 
w