I did extensive numerical studies and found I could invert the 1D heat equation for multiple flows such that the prediction accuracy was 1-2% error for a period equal to the recorded data. That assertion generated all the controversy I was involved in. I just wanted to demonstrate that I reliably forecast the drift. For that I need a signficiant amount of multichannel data. No one would provide any so I'm building my own system to compare 40 new and 40 old LM317s voltage references over time in immediate proximity in an environmentally controlled chamber.
The dominant source of initial drift is the thermo-viscous relaxation of the wirebond strains. Diffusion is probably next, but you can only peel the onion one layer at a time.
But I finished lunch and need to get back to clearing the clutter out my shop s1500 sq ft tp get it properly lighted and wired and my long delayed screen room built.
Have Fun!
Reg
