Waveguide bending loss calculation


Dear Lumerical staff,

I have a question on waveguide bending loss calculation.
First, I have a deep etched InP waveguide cross-section with bend mode profile calculated in MODE solution.

mode.lms (308.6 KB)

It is recommended such waveguide to use > 100 um bending radius to reduce bending loss and with 50um radius mode solution show me a ~ 80 dB/cm TE loss which is non-trivial. However slight changes of PML or simulation domain size may leads to “no mode can be found”. In addition it seems when reducing bending radius (<30um for example), at some point no TE bend mode can be found again. However as I check with 3D FDTD, such mode can still exist (but with higher loss). In that case I usually increase mode solver mesh accuracy to find the more lossy mode but this is not always helpful. It seems bending loss calculated from mode solver also fluctuate when changing domain size and accuracy as well and I do wish to know how to solve those problems.

  1. I also run 3D simulation with 30 nm mesh size at mesh override region and run this huge simulation on cluster.

However even if I subtract two mode mismatch loss (~0.26 dB) at straight/bend junction, 3D FDTD simulation still shows like ~ 280 dB/cm loss at 1550 nm wavelength. I am not sure if addition loss is due to mesh staitcasing and scattering but normally even a tilted waveguide with decent accuracy should not have a propagation loss > 200 dB/cm.

Since 3D FDTD simulation shows much higher bending loss than Mode solution evaluation, I would like to know which result is more reliable? As for me I tend to trust 3D FDTD but I cannot explain the big deviation between two methods.

Arthur Teng


HI Arthur,
I am working on a similar problem using the conventional silicon waveguide. However, I get slightly higher losses in MODE solution. I think this could be due to my FDTD simulation setup. I didn’t use any override meshes. I wonder if you have done the FDTD using mesh accuracy = 8 without any additional meshes, have you? For the sake of comparison, I am interested to know the losses of the FDTD in this case (I can’t apply a meshing similar to yours to my case because I work on my laptop :wink: )
I actually doubt it is the discontinuity of the meshes that introduces extra losses. Also, the huge difference in the losses can be attributed to using different material in MODE and FDTD.