Anyone knows what is the "Compensation tab" here?

Hi. Note that comment saying "area depends on mask alignment". It's also true for the compensation tab, but with the opposite sign. Imagine the top plate is shifted down slightly in the y direction. The tab is added so that +deltaC on top is compensated by a -deltaC on the bottom.

Compensation tab can be used also as a secondary Poly2 contact for better frequency response, making it a win-win.

An interesting small little trick to keep cap parasitics insensitive to mask shift. To make the intention more clear, I would move the compensation tab left to align with the “real” connection line.

Does anyone have any numbers on how much mask shift one can expect? Kind of curious how big this issue is in comparison to other mismatch sources

The overlap design rule of the vias should give an idea, it will be less than that.

I suppose how much it matters is also a function of W and L. For some choices of W and L it could matter quite a lot, for others basically none.

Curiosity got the better of me and I ran the numbers. For sky130 MiM the offset is worse case 0.065 um which is the metal4/via3 enclosure as Harald suggested and also is the value for

tol_m3

in the spice models. Assuming a connecting metal of width 1 um and 0.065 um offset there's a 0.065 um^2 area error. Assuming a comically small MiM cap of 2x2 um that's an error of 1.6%, that's obviously nothing compared to the cap area tolerance which is listed as 20% for sky130 MiM. Take a 20x20 um cap and it's a 0.02% error. To me the change in absolute capacitance is completely dwarfed by the normal tolerance of the capacitor. (note: fringe capacitance is not taken into account here) The bigger question then is about mismatch between devices. I would guess that any mask misalignment would match between devices and this wouldn't have any (maybe too strong a word) improvement. But there may be some faulty assumptions there.