Waveguide Bend loss - Choosing the simulation size, mesh, BC's




I am having trouble with calculating the bend loss for a waveguide; mostly, my troubles surround the convergence of the loss to a reasonable answer. In the waveguide_bend.lms file (directly downloaded from the “Bent Waveguide Analysis” page) I click “calculate modes” and observe that my BENT (r=1.5 microns) propagation loss 8.24 dB/cm.

However, if I double the number of FDE mesh cells for the x and y directions, then rerun the simulation I get a loss propagation loss of 2.72 dB/cm (a significant difference). What is going on here? Are the default settings of this example not setup to be accurate?

I’ve been adjusting the simulation region area and the number of mesh cells for this example, and it seems I can attain just about any loss imaginable (large, small, negative, positive). I hear the PML boundary conditions are necessary to absorb the radiative losses. But is there a methodical way to set the mesh and simulation area settings to ensure accurate loss calculations? Generally, it seems I’ll simply calculate less loss the further away my PML boundaries are.

I understand for the true bend loss we need to add the contributions from the overlap mismatch, which I haven’t had any problems with.

Bending loss decrease with larger bending?
Bending loss decrease with larger bending?

Yes, I tried what you say and it produces this error. However, if you increase the number mesh cells using an over-grid mesh of size (1.8 x 1.8 um ) and mesh size (5 nm in X and 10 nm in Y)

In this case, the loss is slightly above the 8.24 dB/cm. So, the number of mesh cells doesn’t significantly change the loss values.

It seems that as you change the FDE mesh cells, this affects the PML thickness which produces false results. To check this, I decreased the number of the PML layers from 12 to 6 for the same meshing and here is what I get: a loss of 2.13 dB.

the question about the bent waveguide example
FDE waveguide bends radiative loss convergence

So, the conclusion is that the thickness of PML affects the bend loss calculation until it is thick enough and the loss converges.
I tried to change the number of X mesh cells in the FDE and see what happens to the loss:
200 --> loss = 8.6 dB/cm
160 --> loss = 10.04 dB/cm
120 --> loss = 10.84 dB/cm
110 --> loss = 10.84 dB/cm
100 --> no mode found.
Regarding the simulation region, the restrictions on its size are described here : https://kb.lumerical.com/en/solvers_finite_difference_eigenmode_bend.html


Thanks for your answer. Does this mean that I need to increase the number of PML layers until the loss converges? Or is there a better way to choose the simulation parameters?


When I tried to increase the number of PML layers above 12, finding the mode took forever on my PC. It could work on yours, though. I amn’t really sure if there is a rule of thumb for this. But I would choose the largest PML thickness possible. I will tell you if I reach any new conclusion.