Vladislav D wrote on Feb 23rd, 2012, 1:02am:mixed_signal wrote on Feb 22nd, 2012, 4:16pm:Vladislav D wrote
"Also, if these transistors in weak inversion, you have exponential current dependence on Vt... this is really bad for the offset. "
I think exponential relation hurts current and hence leads to current mismatch since I is exponential function of Vgs-Vth. But Vgs is Logarithmic function of I. So, offset should not be an issue in weak inversion.
Thanks
For differential pair this is not an issue but for a current mirror it is
This is a bit confusing, so I'll restate what I think Vladislav meant.
Offset of the opamp will be ΔV
T(diff) + ΔV
T(mirrorp)*g
m(mirrorp)/g
m(diff) + ΔV
T(mirrorn)*g
m(mirrorn)/g
m(diff), where mirrorp and mirrorn indicate the pmos and nmos mirrors of the folded cascode opamp.
ΔV
T(diff) decreases as the square root of the area and has nothing to do with operating point. It could be significant if the sizes are small.
Note that the other terms are multiplied by the ratio of g
ms. If the diff pair gm is less than the mirror gm the V
T mismatch from the mirror is actually amplified. This is why you want to operate the mirror in strong inversion - i.e make gm g
m(mirror) small compared to g
m(diff). (This results in vdsat as large as possible, which unfortunately means an accurate mirror requires a lot of headroom.)
Noise follows the same principles.
The net result of all this is that you want to make your diff pair W/L ratio large, and the length of the mirror devices as large as you can. (It isn't obvious at first, but it turns out making the mirror devices wider does not help.) This definitely means the mirror should be operated in strong inversion and the diff pair in weak inversion if noise noise or offset is important.
Oh crap, I wrote all this and then remembered the OP is using CDS to null offset. Never mind >:(
However,
doing all this will minimize thermal noise which might be an issue, and it really should be how you design any opamp even if you *think* you don't care about these things.
I'm afraid the folded cascode topology might not give you the "near rail-to-rail" performance you require. If that is the case I would do a simple two stage amplifier with
Ahuja-type compensation. Standard Miller compensation could also be used. The noise and offset minimization principles I just explained apply to any opamp topology, and IMO are about the most important design consideration in opamp design.
rg