Hello!
While designing a SAR ADC i found out that Rail-to-Rail OpAmp that i used, have non-linear offset in relation to common mode. That has caused in high INL(though DNL is good).
I think that this happened because input make up from 2 differential pairs nfet and pfet. When common mode near gnd or vdd is working only one differential pair and offset only from this pair affecting ADC. When common mode in the middle of range, offsets from both pair adding to each other and overall offset through all common-mode range became non-linear. Could this be the cause of the bad INL?
If this the case, did somebody know what can i do about this? Except using dynamic offset compensation like auto-zero or chopper.
Maybe not so much that it's one or two pairs, but
moving from one to the other you transition from one
Vio to another (over some range of common mode).
The "both active" might be the best (and normal) as
the two Vios will be averaged somewhat.
Since this is a sampled data system, why are you
opposed to auto-zero? You surely have phases that
you can "hide" that activity behind. Ping-pong, even,
if one has to be active for a full cycle.
Many SAR ADCs work simply on a comparison to GND
(if you have split supplies this is likely). If so then you
may find a lot easier task, maybe just a clocked
comparator whose common mode range includes GND.
Much simpler than a RRIO op amp to design, and the
"latched" phase can hide the zero action behind it.