There some design techniques out there which can improve the gain.linearity product, but best way to  improve linearity with out losing gain is to increase power consumption (and area).
Given I don't know the freq/topology of your proposed mixer its difficult to comment further, but your specs don't seem unachieveable given a realistic power budget.

You should really be using NMOS gm stage, to get best gain to power consumption ratio. There should be no reason that you should use PMOS over NMOS. Assuming you are doing down conversion in a receiver, the input should be ac coupled from the previous stage, so DC bias level should be setup within the mixer itself.
Using NMOS will definately help you get more gain.linearity product for the same power.
If you have headroom issues (i.e < 1V) then you can have a NMOS input gm stage with current mirrored (with ratio) into an NMOS LO switching stage.

sivacha wrote on Nov 16^th, 2006, 6:52am:


This does not make sense. here are a few point which might help you.
1) The LNA output signals are always small, ac coupling at 2.4GHz with MIM capacitors should introduce little loss as you couple with a high pass filter (with cut off less than 2.4GHz). I have done this on many occasions.
2) nmos mobility is normally about 3 times greater than in pmos, as a consequence you get much more gm for the same current and device geqometry. this is fact.
3) The linearity of the gm stage if it is just a differential pair is dependant on the overdrive (vgs-vt) you are operating the device at. If you have excess gain in your nmos implemenation, you can decrease gain and increase linearity by reducing W/L to increase overdrive given the same bias current. Other standard linearing techniques like degeneration or offset coupled pairs (like bjt multitanh) exist.
4) You should confirm that your linearity is truly limited by your input stage (by simulating it standalone) and not due to output headroom or LO switching issues.

cheers
aw

loose-electron wrote on Nov 18^th, 2006, 11:44am:

Yes, the classic Gilbert paper "the Multi-tanh principle" explains it all very well...
Anyway, i have used them in MOS and bjt forms in gilbert cell style cells for modulator and mixers... because of the increased input capacitance & fast gm roll off outside of the flat (linearised) gm section, perhaps not the best method for linearising high frequency first downconversion mixer in a wireless receiver....
More suited for well defined signal excursions in a modulator of a wireless transmitter... extremely flat gm characteristic gives for very low distortion products ... pretty robust to corners, mismatch etc.
Used them extensively in gmC filters (triplet forms) for receiver channel filtering.
All in all I like the technique...

Used them extensively in gmC continuous-time filters (in triplet form) for receiver channel filtering.