Why does the mode loss fluctuate dramatically when reducing the mesh step?

xiaotuxiaotu Member Posts: 1


I am trying to obtain the loss level introduced by a gold(Au) strip in a waveguide. The schemetic diagram is shown in the following.

However, when I reduced the mesh step(from 50nm to 1nm) near the gold(Au) region ('mesh Au'), the simualted loss flucturate dramatically, shown in the above figure. Could you please help to find the causes for loss fluctuation and how to modify my setting to obtain a more accurate loss level? My FDE setting is shown in the following. Thank you so much.


  • kjohnsonkjohnson Posts: 204Ansys Employee

    Hello @xiaotu ,

    A very fine mesh is often required at metal/dielectric interfaces where the field amplitudes rapidly change over a short distance, so it's expected that the results won't converge until the mesh is on the order of nanometers. Have you tried reducing the mesh even further than 1 nm to see if the loss converges? It might help your testing to use metal BCs, so the only loss in the system is due to absorption in the Au and not radiation losses through the GaAs layer into the PML boundaries. You can also speed up your testing by using symmetry boundary conditions on the x min boundary, either anti-symmetric or symmetric, depending on the polarization of the mode you are interested in. Please try that and let me know if it helps.

  • xiaotuxiaotu Posts: 2Member

    Hi @kjohnson

    Thank you so much for your reply.

    The loss does tend to converge when I try a finer mesh less than 1nm. But I cannot reduce the mesh step less than 0.5nm, since it would report error during symbolic and numerical factorization. I think this error is caused by too many mesh points, so I tried a smaller simulation region to reduce the mesh point. However, since the simulated waveguide is a weakly guide waveguide, I don’t know how to determine the minimum simulation area that can be set to ensure reliable simulation results. Currently I reduce region to 25um*20um, and the simulated electric distribution in log scale is shown in the following figure (metal boundary is applied). Could I continue reduce the simulation region?

    In addition, I have another question about ‘optimization and sweeps’ with metal boundary. Metal boundary will introduce a large number of nonphysical modes. So how should I set ‘Sweep’ up to obtain the dispersion of a certain mode? Thank you again.

  • kjohnsonkjohnson Posts: 204Ansys Employee

    Hello @xiaotu ,

    Sorry for the late reply. If you're running into memory problems due to the mesh, I would change the span of the mesh override regions so they are only placed where the fine mesh is required. This would be where the field is changing sharply, at the interface between the metal and the semiconductor materials. A fine mesh is not really required in the middle of the metal. Also, the mesh should be fine in the direction normal to the metal surface. The mesh does not need to be fine in the direction parallel to the interface. If you make these adjustments you should be able to make the mesh finer without requiring too much memory.

    Regarding to how large the simulation region needs to be, similar to the mesh this ultimately must be determined with convergence testing. First you might want to check to see if the mode is bound, as described on this page:

    If it's unbound, you will have to use PML boundary conditions for this convergence test (once you're sure the mesh issue is resolved).

    To get the dispersion of a mode you can use the FDE solver's built-in frequency sweep tool rather than a parameter sweep . This tool has an option for tracking the selected mode:

    Let me know if you have any questions.

Sign In or Register to comment.