Re: Gradient Effects
There will be some stress / strain on any mounted die but
whether there is a topside gradient, I doubt.
Even without "power devices" you can still see significant
localization of power dissipation. Consider an op amp with a
10mA output rating and a 3V supply; maybe 25mW or so, seems
insignificant but with modern geometries this might be in 100x100um
or less area and on a par with the power densities seen in "real"
power devices. Plus, with power concentrated at the topside,
you will see more lateral thermal conduction as a percentage
of total power, than in a larger area (or, larger lateral extent
than wafer thickness) power device. Lateral thermal conduction
is directly going to make a lateral thermal gradient. The more
geometries shrink, the more power densities rise and the more
variation in gradient will be localized.
You can address this somewhat by adding "cooling zone" extent to
your layouts, preventing too-tight packing of heater against "receiver".
Centroid layout is good for when you do not know a priori the
final orientation of your sensitive pair against the thermocline
(as in a standard op amp cell for general use). Back in the olden
times when an op amp was the whole product, we would just
"point the bow into the waves" and call it good. Hard to do a good
centroided layout in single layer metal, with minimum transistors
almost as big as bond pads....