FDTD simulation for ultra-wide bandwidth

I am using FDTD v2017a to run simulations over a wide bandwidth from 1200nm to 1700nm. I simulated a regular waveguide tapers using the EME monitor to extract the fundamental TE mode transmission efficiency. I first ran the FDTD simulation over the entire 500nm wavelength range with the linear wavelength spacing and 501 data points. I then ran the simulation on the same structure with 5 sweeps of 100nm wavelength ranges. Each sweep covers 100nm wavelength range and have 101 data points. Both simulations have the same mesh setting and conformal mesh accuracy.

I suppose that the two simulations should give me similar results. However, the results are very different. I wonder if there is an efficient and accurate way to simulate a structure over such a wide bandwidth.

I attached with the FDTD file and the results of the two simulations.

Thank you.


WG_StripTaper_WG350650L3_Model.fsp (3.7 MB)

Dear @hany

Thank you for providing a detailed question and supporting documents.

In your simulations, you are using a single mode for light injection:

This means that for the wavelength span of 1.2-1.7 um, software calculates mode at the center of pulse (this is the center frequency which is not necessarily the center wavelength and seems to be at lambda=1.4um) and assumes that this is the shape of the mode at other wavelengths. While single mode injection works fine when light is confined (shape of the mode does not change with wavelength), it is not precise for sub-wavelength designs where you have an evanescent field. If you look at your second plot, you can see that for wavelength 1.2-1.3um, transmission is almost 1 (because mode shape is similar from 1.2-1.3) for these confined modes but drops quickly at higher wavelength (1.6-1.7um) where the shape of the mode changes drastically from one wavelength to another. Please note that mode expansion monitor calculates mode overlap for transmission plots.

I tried to run simulations with multifrequency mode calculations with 505 frequency points (I used a course mesh and disabled override mesh to decrease the simulation time) and here is the result:

You can see that the results are around 1 (with some fluctuations). I could spend more time to remove some of the ripples and improve the problem of transmission above 1 (by using finer mesh and better PML BCs), but I guess the idea is clear that you expect a transmission around 1 for all the wavelength if you use a multimode frequency injection.

If you had success on removing ripples and plot with transmission around 1, please update it here for future references.
Let me know if you had further questions.

1 Like

I assume that this “multifrequency mode calculation” is the new feature in the 2017 version.

I will redo the simulations and let you know the results.


Thats true, its the new feature added in 2017 release. You can read more about it here.

That would be great. Thanks

I ran the simulation using the multifrequency mode calculation in the source setting with 501 frequency points. The result is shown below. The mesh setting is conformal mesh level 6 with override mesh 10nmx5nmx10nm (x,y,z). The ripples still exist.

Hi @hany

Thanks for the update. Can you run simulations with higher frequency points for both mode calculations and monitors? The current plots are reasonable (have only 1% fluctuations which is not too crazy for numerical simulations) but I expect you can improve it by improving the resolution of injected and collected light.


Unfortunately, simulation with higher frequency points cannot be performed due to the computer RAM limits.