Photonics

Photonics

Difference in results between FDTD and MODE (FDE) for waveguide?

    • aberg
      Subscriber

      I created a simple straight Si waveguide surrounded by SiO2 in both FDTD and MODE. I would like to find the mode at the tip of the waveguide (and later compute the overlap between the waveguide mode and a fibre mode).

      In FDTD, I added a mode source (forward direction) at the left edge of the waveguide and a monitor at the right edge to obtain the field profile.

      Separately In MODE, I added an FDE solver (2-dimensional) at the right edge of the waveguide, at the same place where I had the FDTD monitor in the other simulation.

      I expected both of these methods to produce the same output mode since they correspond to the same structure. I ensured that both solvers extend beyond the boundaries of the structure. However, FDTD and FDE do not produce the same mode, because when I calculate the overlap between the mode from FDTD and the mode from FDE, the overlap is < 0.5 when it should be close to 1. They also give very different optimized overlaps with a fiber mode. (Because we are interested in the overlap/power coupling and not the absolute power, we are not normalizing the modes, which should be ok, I think.)

      How can I ensure that FDTD and FDE will give the same result? I am using the exact same waveguide and oxide geometry for both. Because this is a straight waveguide, the length of the waveguide should not matter, right? The mode profile at the tip should be the same regardless of the waveguide length. Thank you.

    • Guilin Sun
      Ansys Employee
      I guess it is most likely due to mode degeneracy, or mesh size. Is this a square waveguide cross section? if so, please try to use symmetry BC: https://support.lumerical.com/hc/en-us/search?utf8=%E2%9C%93&query=symmetrical
      In addition, please make sure both software uses the same BCs and the same mesh size, as well as region. Actually inside FDTD it uses the FDE solver. Finally, please make sure the wavelength is the same, as sometimes it can be different if not set properly.
    • aberg
      Subscriber
      It is not a square waveguide, unfortunately. It is a standard 500 nm x 220 nm SOI waveguide. I am using PML boundary condition for both with the same mesh size and wavelength. Should the 3D FDTD solver region have the same size 2D cross section as the FDE solver region? So if the waveguide (5 um long) is along the x axis, and the FDE region is at x = 0 with Y x Z dimensions = 4 um x 4um, then the FDTD region should also have dimensions 5 um x 4 um x 4um ? In general it is better to have a big simulation region where the boundaries are far away from the region of interest (waveguide), right? (To avoid reflections for metal BC and extra loss/gain for PML BC)
      Also how large does the mode source in FDTD need to be for the fundamental TE mode? Should the mode source be only as big as the waveguide (500 nm x 220 nm), or must it be almost as big as the entire FDTD simulation region? Do the size of the mode source region and the size of the FDTD solver region have any relation? Thanks very much!
    • Guilin Sun
      Ansys Employee
      Q: Should the 3D FDTD solver region have the same size 2D cross section as the FDE solver region?
      A: yes, to compare results from different tools, please make sure everything is the same.
      Q: So if the waveguide (5 um long) is along the x axis, and the FDE region is at x = 0 with Y x Z dimensions = 4 um x 4um, then the FDTD region should also have dimensions 5 um x 4 um x 4um ?
      A: no, both do not need very long length. Make sure their cross section is the same.
      Q: In general it is better to have a big simulation region where the boundaries are far away from the region of interest (waveguide), right?
      A: generally it is good. But make sure the mesh is not too coarse. I would suggest to use mesh size instead of mesh cell number.
      Q:Also how large does the mode source in FDTD need to be for the fundamental TE mode?
      A: usually we suggest to not truncate the mode profile. You can visualize its intensity, probably using log scale is better. As how how small the intensity at the boundary, it is not quantified. But you can test different sizes and see how the size affects the result and what accuracy do you need. intensity smaller than 1e-6 should be ok in most cases.
      Q:must it be almost as big as the entire FDTD simulation region?
      A: I would suggest so if possible. As long as correct mode can be found, larger size of the mode source size is better. It has minimal effect on the memory requirement.
      Since this topic is to compare the results, please make sure the mode solver region has the same settings (including the material, where FDTD uses fitted result but FDE may use linear interpolation). As mentioned previously, FDTD uses FDE solver for mode/port source. If everything is the same, they should give the same result.


Viewing 3 reply threads
  • You must be logged in to reply to this topic.