for opamps, you can estimate the thermal-time-constant, the long-tail of settling, by extracting the spacing between the Input Diffpair and the Output heat-generating large transistors.
1 millimeter silicon cube has thermal Time Constant of 11.4 milliSeconds.
100 micron silicon cube is 100X faster, or 114 microSeconds.
10 micron silicon cube is another 100X faster, at 1.14 microSeconds.
1 micron silicon cube is another 100X faster, at 11.4 nanoSeconds.
For ease of thinking, copper conveniently is about 20% faster in these numbers. A 1 centimeter cube of copper is (about) 100X slower than a 1 milliMeter cube of silicon, at 900 milliSeconds. And that time constant remains the same as you pump heat laterally thru thin copper foil, thus you can begin to predict the thermal settling of PCB structures.
A (10 centimeter)^2 layer of copper foil has thermal Tau another 100X slower, at 90 seconds.
Thus a 4 inch PCB region needs 1.5 minutes to reach just ONE TAU of thermal accuracy. This assumes heat only flows along the 4" direction, thru a contiguous region of copper.
By the way, you can compute the thermal Tau (time constant) using specific-heat of silicon, and the thermal resistance (inverse of thermal conductivity) of silicon.
The books provide the thermal_diffusivity of silicon as 8.8e-5 meter^2/second
I computed the inverse:
thermal time_constant of silicon = 11,400 seconds for a cubic meter
thermal time_constant of copper = 9,000 seconds for a cubic meter
If you draw a cube, and draw the heat flows from one face to the opposite face, then chop the cube into 1,000 smaller cubes (10*10*10), you will see the thermal capacity of each (smaller) cube drops by 1,000X but the thermal resistance rises by 10X.
The result is: for 10X small cube, the thermal time_constant (TAU) reduces by 100X.