NF of a LNA in AWR EM simulation

Hi. I am designing an LNA on AWR. When I run the simulation without EM_Extract I get 0.47 dB NF. When I run it with EM on and X,Y cell size= 1 mm I get around 0.53 dB NF. When I decrease the X,Y cell size to 0.1mm it gives 0.61 dB NF.
My question is which one I should consider as more accurate? As the cell size goes down NF keeps increasing. My goal is to have NF<0.47 dB(Since it is the first stage of a 4 transistor cascade).

But those differences are so small, what you expected ?
If you use nonlinear model, its noise characteristic has already-mostly-an error factor and I guarantee you that will be much worse than this difference.

Small cell size captures skin effect more accurately, so some slight increase in loss is typical.