I checked your last simulation file and the paper you want to reproduce [Z. Fang, et al. “Plasmon-Induced Doping of Graphene”, ACS Nano 6, 10222 (2012)]. Besides the corrections that @vivek mentioned already regarding the size of the circles and the boundary conditions, there is an important consideration for the z span of the TFSF source and the cross-section analysis groups (scattering and absorption). It seems like Fig. 1 in the paper is meant to describe the cross section of the nonamer on top of the oxide layer; therefore, the TFSF and the cross-section analysis groups should only enclose the nonamer and part of the oxide layer, not the silicon underneath. After these changes I got a curve that looks much closer to the result in the paper:
I should also mention that I used two mesh override regions to make sure the mesh is uniform in the x an y directions inside the TFSF (as suggested here) and to get enough mesh points along the z direction inside the nonamer. To speed up the simulation I reduced the size of the simulation window, making sure there is a distance of at least half the maximum wavelength between the structure and the PML boundaries.
The scattering peak at ~1000nm and the absorption peak at ~800nm are smaller than in the paper but I think this is an issue related to the mesh. You should do some convergence tests refining the mesh in the mesh override regions. Also, I suggest checking the effect of increasing the number of PML layers to make sure the PML is not affecting the results.
Take a look at the modified simulation file: modified with TFSF source - FG.fsp (320.7 KB).
Hope this helps!