Why am I seeing a significant difference in coupling efficiency from EME to FDTD?

drewskidrewski Member Posts: 2

I am simulating the evanescent coupling between two waveguides after the waveguides overlap for some distance X (i.e. how much light is transmitted from waveguide A to waveguide B when the waveguides overlap for some distance X). I have identical simulations setup in EME and FDTD (they have identical mesh sizes over the waveguides of 10 nm , very fine non uniform conformal variant 1 background meshes, PML boundary conditions, same simulation region size in Y/Z, etc.). The structures of the two waveguide are also the same (identical dimensions between simulations with the smallest feature size being 50 nm). When simulating in FDTD, the transmission for a 5 um by 5 um output monitor located 15 um after the waveguides stop overlapping is around 67%. When the monitor is located 100 um from when the waveguides stop overlapping, this is still 67%. A 20 um by 20 um monitor located at 1um from when overlapping finishes shows 98% transmission, while the same size monitor at the 15 um position shows around 87%, and again the same monitor at the 100 um position shows around 72%. The EME simulation always shows 87% for the abs(S_21)^2 and abs(S_12)^2 matrix value for a 5um by 5um port size. When I run a propagation sweep in EME to determine if the distance between the end of the overlapping region and the final port makes a difference, I find the value is 87% regardless of how long this distance is.

This seems like a significant discrepancy between EME and FDTD for the same simulation, and I'm not sure which simulation to trust before I can continue with design development. Please see my attached screenshots of the settings I'm using below. I would like to determine why the difference is occurring and make sure EME and FDTD match one another to within 2%. Thanks so much for any assistance (is there a way to upload my simulation file?).


  • drewskidrewski Posts: 5Member

    Correction: in FDTD when the 20 um by 20 um monitor is 15 um away from the point that the overlapping region stops, the transmission is 94% and not 87%. Apologies.

  • gsungsun Posts: 1,654Ansys Employee

    If you have premium support, you can send us email with your simulation files. However, in the Forum no file is allowed to download by Ansys employee.

    So you want to match EME and FDTD within 2%?

    Those are two different methods in computational electromagnetism and each has their own sources of errors. You mentioned that you set up the simulation identical but actually it may not be the same: the material fitting and the PML.

    Individually speaking, for FDTD, when the monitors are located in the same uniform waveguide mimic very long size, their transmission should be very similar if not the same, provided that the simulation time is long enough. The small error may come from discretization, causing numerical dispersion and thus the final transmission. I personally trust the monitor closer to the cavity, which has small numerical dispersion error.

    Since you use mesh accuracy 4, the mesh is finer and it is likely the PML may cause some reflection. I would suggest to use more number of layers.

    Since this is a cavity, smaller autoshutoff level will be needed, depending on the resonance property of this device. The higher the resonance (larger Q), the smaller the autoshutoff level.

    Please use long enough simulation time and let the autoshutoff min to terminate the simulation.

    For EME, it will use the raw data of the material refractive index if you do not fit it. I would suggest that you export the FDTD refractive index,

    getfdtdindex - Script command

    and then import it into EME material database: Creating new sampled data materials in FDTD you can use more frequency points.

    Then the EME result will depend on how many modes are actually excited and how many modes you used in simulation. Even if the device, mesh, and material are exactly the same as FDTD, its result can be different from FDTD. You can also compare the EME results with different number of modes. Please note that in such device, only 3 groups are good, and each group can have only one cell, since inside each group the waveguide is uniform.

    In summary, since different algorithm has different sources of errors, and their effects to the final result are not the same. So setting them in the same mesh size and the same configuration do not guarantee the same results, except you have done considerably converging test. Please refer those links:

    Convergence testing process for FDTD simulations

    Convergence testing process for EME simulations

Sign In or Register to comment.