I have made a modified version of your simulation file: modified_front.fsp (311.0 KB)
Using this file, I was able to get the following plot of the transmission spectrum where there are peaks near 8.5 and 12 um which is similar to where there are peaks in the experimental data:
I made the assumption that the metal can be modelled as PEC (perfect electrical conductors) which behave ideally with 100% reflection. This assumption would be valid if the skin depth of the metal is much smaller than the wavelength of light, and it allows you to use a much courser mesh to model the metal which will require substantially less memory and simulation time.
I also made the following changes:
- I increased the simulation time to allow enough time for the fields to propagate through the simulation region and fully decay - I think that having too short of a simulation time is likely a main cause for the results not matching the experimental results. The effect of having too short of a simulation time is discussed in more detail here: https://kb.lumerical.com/en/index.html?ref_sim_obj_frequency_monitors_simulation_time.html
- I changed the PML profile from “Standard” to “Steep angle”, and I increased the number of PML layers to 64 to help absorb light which may be diffracted to grating orders which travel at steep angles away from normal.
- I added a mesh override region over the metal layers to set the dz mesh step size to make sure the thickness of the metal is properly resolved
- I changed the mesh refinement method from conformal variant 2 to conformal variant 0 since conformal variant 2 can sometimes lead to numerical artifacts if the mesh step size is not fine enough
I think it would be a good idea to do some further convergence testing of the settings, like described on the simulation methodology page here:
You may also want to check the assumption that the metal can be represented as PEC by simulating the Au and Ti materials instead of PEC. Since this would require a finer mesh, and increased simulation time and memory requirements, you may consider running 2 separate simulations with 1 source in each and summing the results to get the result for circular polarization as shown in the example here:
The advantage of this method is that you can use symmetry in the boundary conditions when you have only 1 source in the simulation which will allow you to reduce the memory and simulation time required for the simulation.
Hopefully this helps!