We experimented with various resistors, 0603 through 1206, with a
thermal imager that has plenty of resolution to find the hot-spot temp
on all the parts.

Turns out that the actual part thetas are astronomical, and what cools
resistors is the copper pads and traces, more than the bodies
themselves. An 0603 soldered to big copper pours can dissipate as much
as a 1206 ditto; the hot spot temps are the same. That makes sense
physically, as theta is determined by the L/W ratio and the substrate
thickness, and all the resistors we checked were all 20 mil thick
alumina.

A better substrate, BeO or AlN, would help a bit, at great expense.

If there's a ground plane close, that helps a lot.

The easiest thing to do is fire up the board and scope it with a Flir,
and see how hot stuff gets. You can borrow mine, or come here and use
it.

ftp://66.117.156.8/IR_0026.jpg

ftp://66.117.156.8/IR_0032.jpg

John