volkan,

What are you using the ADC for? Perhaps there is a better way to get precision results without "tempting the devil".

Steve,

All of my experience with 16 bit ADC's has been as non-integrated devices.

I never had the opportunity to play with an integrated hi-resolution convertor (10 bits has been it).

But if we are allowed to be purely theoretical;

It seems to me that you could compensate for Drift if the characteristic is known, and if the driving mechanism was detectable (i.e. temp).

Perhaps you could compensate for DNL and INL using the same technique folks have used for resolution improvement, that of modulating the ADC input with a known, self-generated tri-angular waveform "sized" to traverse at least 1 lsb boundry.

If you designed the modulation to traverse several "steps" you could move the ADC "non-conformity" to different locations of the incoming signal. Once you subtract your tri-angle modulation, average the results, you might have pretty good accuracy.

This technique has the advantage of diverting major design and test resources to a minor design component and well away from the major design goal thus ensuring that engineering's reputation for "wheel re-invention", "scope creep", and "windmill tilting" stays intact.

regards,
p