How Fast is a Wilson Current Mirror?

Assuming you want a reasonable phase margin (meaning, if you have two poles, keep the 2nd one at 2x UGB).

You'll see why your PMOS mirror usually gets away without any explicit compensation cap..

If we assume,

Then, assuming the source follower works good (i.e., its Cgs doesn't come into play because its V(S) = V(S) due to unity gain.. (nice, in reality, you don't get that because of the gmbs - so, straightaway, some of the CGS *has* to come into play)

Also, assuming that 

We get, if you go around the loop,

and 

So, if one substitutes the expression for gm, one gets, again, using 

What we are interested in is, how fast will the cap that the input current sees take to charge to the final voltage. Or, how does this Cc (which is what the I_in now sees) compare with the effective input cap for a simple current mirror? If you call that cap C_simple, it has to be 2 * W1 * L1 *Cox

Giving

Moral? If you want to see a benefit, keep the mirror devices' aspect ratio as low as possible and the aspect ratio of the cascode (which does the unity gain buffer role) devices as large as possible.