Are the MIM capacitors any good for very small capacitance values? Say, in the range of 20fF? I'm considering using them to do impedance matching but I'm afraid that the process tolerances might kill my matching.

I think the main problem is not only process tolerance but also parasitic capacitances. 20fF is a very small value and a short metal interconnect line will easily add some fF of capacitance. So you should do a layout parasitic extraction and examine all the parasitic contributions. If you need matched pairs of capacitances you should draw a matched layout structure, keeping same orientation etc. That said I think MIMcaps are ok for small values. They have good linearity and low parasitic (fringe, bulk) capacitances, for sure better than MOScaps.

If you run into too many issues with MiM parasitics you could also attempt to generate capacitors in fractal structures as a comparison, which are also very linear if done correctly. But I would personally stick with the more standard MiM caps, and do whatever you can to be able to use larger cap values in the design if you expect a realistic chance of good matching. https://ieeexplore.ieee.org/document/735545

I was thinking about using mesh MOM capacitors in M1-M2 as they are proven to be more stable regarding fabrication variations. And, yes, MOS capacitors are way worse. Especially at 60GHz :)

@Leonardo Gomes For high frequencies (GHz and up), usually the best is finger-caps (MOM) in higher metal layers, say

m3

and/or

m4

in SKY130. When you push the caps up in the metal stack, the parasitic cap to the substrate is lower, plus the Q is decent as the sheet resistance is a bit lower. You can tune Q by selecting finger length and number of fingers, and finger width and spacing you might want to keep at a minimum. Parasitic inductance is also OK if finger

L

is not too long. We have been using finger caps up to 300GHz, with no issues. MIM you can use if the finger caps get too large.

There are a bunch of predefined MOM layouts in the SkyWater process in the

sky130_fd_pr

library, with various configurations of fingered layers and shielding layers. They are characterized for RF and have specific SPICE models associated with them.

The issue might be the specific capacitance value I need... I'll dig around. Thanks for the information!