Lumerical Question about Computation Time in FDTD study - grating coupler at 4200nm wavelength

tgartnertgartner Member Posts: 5


I have built a model to run the particle swarm optimization for a basic 2-D grating coupler with a gaussian input in Lumerical. The original model was built to optimize a SOI grating coupler at a wavelength of 1550nm with a scalar gaussian source (the typical case study for these couplers). Running the particle swarm on this model with 4 different variables (grating period, fill factor, source x-position, angle of source input) to maximize the transmission into the waveguide takes <20 mins with nice results at a mesh setting of 3 by default.

I have changed this same model to optimize an air clad, silicon-on-sapphire (SOS) grating coupler at 4275nm wavelength with a gaussian source (thin lens). I added the sapphire material into the library via its Sellmeier formula. With the same mesh settings as the previous model this one takes hours upon hours to finish (I can't say how long as it has not finished).

Any ideas as to why there is such a HUGE difference in computation time? Is it the wavelength being much longer (if I run the SOS model at a wavelength of 1550nm it goes much faster)? Perhaps I need to change the mesh?

Or am I missing something/doing something wrong?

I can provide my file if that would help?



  • gsungsun Posts: 1,152Ansys Employee
    edited March 22

    This could be due to mainly two reasons:

    1: memory issue. Please check the memory requirement and the physical memory that is available in your computer.

    The meh size is actually dependent on the shortest wavelength inside the material. Mesh accuracy 3 is just a relative number, meaning 14 meshes per smallest wavelength. FOr the same simulation region, if the material has higher refractive index, it will use more memory.

    2: resonance. This could be physical resonance, or due to improper PML.You can check the spectrum from a point time monitor and see if there is strong resonance. and such strong resonance can be out side of the interested spectrum range. If it is the later, you can use longer source power by setting the pulse length longer in time domain at source. If you change the PML location along propagation direction and the resonance peak changes, it is likely this is caused by PML reflection.

    In addition, please check the material fitting, if you are using broadband source. Occasionally the material fitting may have artificial "peak" such as this one

    When you change scalar Gaussian to vectorial Gaussian source, how many plane waves you used? please use the default number, do not use too many. and if the beam has higher NA than the grating coupler needs, it can cause strong diffraction, which in turn causing PML reflection mentioned in point 2. I would suggest that you use a scalar Gaussian for test, or an import source from a similar fiber or other waveguide.

  • tgartnertgartner Posts: 10Member

    Thank you for your comments.

    So to be clear:

    -there will be more mesh elements studying a source with wavelength 4275nm than a source with wavelength 1550nm (as you said its 14 mesh per smallest wavelength)?

    -I think there is an error with my added Sapphire material. I entered it via its Sellmeier coefficients as follows:

    However when i go the material explorer and select fit and plot there is only this, which does not show any change in index:

    Is there anything else I needed to do when adding the material other than enter its Sellmeier coefficients?

    Lastly, as per your remarks on using a scalar gaussian. I am trying to model this as it will be set up in my lab, which is a 4275nm laser source, focused through a lens (which I am assuming NA=0.5 for now) onto the grating coupler. I can then manipulate the spot diameter by varying where in the focal length of the lens the grating plane is. I used the default 200 plane waves in the model settings. Should I not be using the thin-lens option for modelling this type of experimental setup?

    Thank you!

  • gsungsun Posts: 1,152Ansys Employee

    about the mesh dependence to wavelength: the mesh size depends on min(lambda)/n where n is the refractive index for dielectric materials where loss is not significant.

    The Sellmeier material model can only give you one refractive index without dispersion, since it is mostly used in frequency-domain method, not in time domain. In time domain the material property needs to meet kk relationship, which requires data in imaginary part. In addition, Although this model can be dispersive, FDTD does not treat it as dispersive!

    If you want a dispersive material, you will need create data and then import the data, as a function of wavelength or frequency. please refer

    let's focus on the huge long simulation time issue: check the memory; test to see if there is strong resonance inside or outside of the interested spectrum from the time monitor. As for the Gaussian beam, after using the default of 200 plane waves, you can use the vectorial one. but before the memory issue is resolved, I suggest to use scalar one in order to isolate the cause of the issues.

Sign In or Register to comment.