Pelgrom's law that matching is proportional to 1/sqrt(area) no longer seems to apply, at least with MOM finger (fringe) capacitors.

I have noticed in two different processes (65nm and 22nm) that, for example, a 1000fF cap made of 100 10fF caps matches much better than a single 1000fF capacitor. In fact, the standard deviation was 12x better when 10fF unit caps were used. That's huge.

This oddity is confirmed in the design manual's data, but they don't provide an explanation.

Does anyone know what's going on here? It completely violates Pelgrom's findings.