Thanks for your feedback! I am using the 32-bits free version.
Regarding your answer, the point is that the nano structures are smaller than the wavelength, so there should not be any divergence or convergence. Indeed, looking at Ez in the reflected field, I see that is close to 0 across the spectrum, meaning that there is not diffraction due to the nano structures.
I tried to move both observation area and observation points further to meet the far field approximation and still results are very different at the target wavelength. I am clueless 🙂