How does 1/f noise scale in smaller-geometry processes? Assume that I have a current-mirror that has constant area for 0.35um, 0.18um processes, etc. How would you expect the 1/f componet of the output current to scale?

I would think that the effect of the (smaller number of) traps would be larger in a smaller-geometry process.  Also, 1/f noise can average out for large devices: one trap is occupied, but another isn't, and the effects cancel.

Probably also the trap density is a function of the maturity of the process.  In fact, the developers of the HiSIM mosfet model claim that 1/f noise needs no fitting parameters for a mature process!

S. Matsumoto et al,  "1/f noise characteristics in 100nm-MOSFETs and its modeling for circuit simulation,"  IEICE Trans. Electron., vol. E88-C, pp. 247 254, 2005.