"could"?
Dude, LARGE form factor, higher power, older legacy dies were more stable from die element to die element.
The tiny stuff now though? Too many minute variables.
It is like a high tension line. The voltage drop at the other end is the same, but the voltage carried determines how much of an effect that loss has on the overall transmission.
The bigger devices... Lets use non-real numbers to illustrate.
The big guys take one mole of material per layer to build, so a whole die of them is X number of moles of material.
Losing a few million or billion molecules of that count will be insignificant.
The small guys though... only take up one ten thousandth of a mole per element layer the number of molecules 'lost' in that die form factor per device has a much more significant effect.
Big or small, using a diode junction for sensing temperature may be very accurate in tracking linearly with temperature change. But there still needs to be a reference point calibrated against so that those numbers shown, tracking that temp change actually come close to real values.
IOW, they all need individual calibration to give accurate data.
They may well all work and track EXACTLY the same, but their starting value will vary.
non-cooled IR Focal Plane Arrays have the same problem, and they need cal from frame to frame to frame to be right.
They use an instrument contained ambient shutter, and look at it, then the scene, then the shutter, then the scene, etc., and they can track accurately better, since the bolometers on the CMOS heat up with the imagery they view as well., and the instrument itself has different "chassis" temperatures from usage to usage. They have to constantly check a reference to offset any shift.
These diodes... think of them as "pixels". Every one of them needs calibration. The FPA has software for it. We have to hand match and cull. That is why there are different accuracy classes for the devices too. Some are just fails from higher tolerance batches.