Learning Forum

[Henrick Höllerer]

Hello,

I am using Lumerical MODE with an FDE solver to find guided modes in a bent waveguide.

After upgrading to Lumerical Version 2024 R1.1, I encounter a strange issue. Sometimes findmodes gives me an output like this: 

In this case, findmodes returns the maximum number of modes it is set to find with all modes looking like this.

That happens however inconsistently. If I run my script again without changing anything (e.g. setup, solver settings, …), I might find the correct modes or obtain the same strange result again. The same error appears also in older Lumerical MODE scripts that run reliable on previous versions before and on different machines.

Does anyone know why this error occurs and how it can be fixed?

Thank you for your help. 

[Amrita Pati]

Hi Henrick,

It looks like some strange modal interaction is going on with the simulation boundaries. Would you be able to share a screenshot of the simulation geometry? Sometimes if there is a gap between the boundaries and geometry (by geometry I mean substrate/superstarte here, I can't tell if there is one or if this is just a waveguide surrounded by the background material), it creates an artificial cavity which can result in these unphysical modes. I would also be interested to know what BCs are you using? Metal or PML? If you are using metal, try switching to PML and rerun the simulation.

Regards,
Amrita

[Henrick Höllerer]

Hi Amrita,

thank you for your fast response. I am already using PML boundaries. Here is a screenshot of my geometry. 

I am considering a fully-etched Lithiumniobate waveguide on a SO2 substrate and with a SO2 top layer. On the right part of my screenshot you can see the setup script for this geometry. On the left part the geometry settings of the FDE solver are shown. For the FDE solver I chose the “2D X normal” setting with yspan = 4 * waveguide top width and zspan = 3 * waveguide height. The mesh is defined by a maximum mesh step of 20nm in both y- and z-direction.

When enlarging the YZ view, sometimes a small gap appears and disappears that shouldn't be there. Is that what you mean? I attributed this just to the discrete display. Since when meshing the structure before running the simulation, no such gap appears:

The strange thing is, that this simulation setup without changing anything runs sometimes also correctly.

Best regards,
Henrick

[Amrita Pati]

Hi Henrick,

Thanks for sharing these details! 

When enlarging the YZ view, sometimes a small gap appears and disappears that shouldn't be there. Is that what you mean?

I believe that might just be an artifact of the visualization. I was referring to an actual physical gap between the substrate and boundaries. Or in other words, the substrate terminating before the PML boundaries. But that is not the case here.

From the screenshot, it looks like you are simulating a bent waveguide. I missed that detail from your first post. There are a few important considerations while simulating them, the most important one I would say is the simulation span. In straight waveguides, the modal energy is more or less localized in the core region. In a bent waveguide, however, there is coupling into the radiation modes that can interact with the PML creating artifical losses/gain. Typically, it is suggested that the fields at the PML should decay to about 1e-05 to 1e-04 of the max fields. I will recommend running a few simulations with increasing simulation span, especially in the direction opposite to the bend as this where most of the radiation takes place. You can monitor how the different properties like losses, neff, etc. change over these spans. They should converge if the simulation is large enough and that would be the right choice of simulation span. You can consider using a mesh override region for just the waveguide, and keep the mesh relatively coarse (default mesh) in the background.

Let me know if you also see this strange behavior for larger simulation spans.

Regards,
Amrita

 

[Henrick Höllerer]

 

Hello Amrita,

thank you for this suggestion. I tried it and unfortunately this strange behaviour occurs also for large simulation spans randomly, i.e. sometimes I get a physical result sometimes not. Independent of the parameter settings.

For instance, here I used a quite large solver area with yspan = 9.6 microns and zspan = 4.8 microns for a (bent) waveguide of width = 800nm, height = 600nm and bent radius of 100 micron. The default mesh has a max. stepsize of 20nm. Around the waveguide I used an override mesh with a max. step size of 5nm.

Without changing anything – just re-running the exact same simulation as above – I can also obtain this (correct) result:

So, doing a parameter sweep is possible, but not helpful as these unphysical results happen randomly and are just outliers of an otherwise converging curve.
I am running this simulation on a computational server of my university. I was wondering if this behaviour could be caused by the computational resource management of Lumerical and the server. 

Best regards,
Henrick

 

[Amrita Pati]

Hi Henrick,

Thanks for confirming! 

I am running this simulation on a computational server of my university. I was wondering if this behaviour could be caused by the computational resource management of Lumerical and the server. 

I don't think that should happen. 

It could be a bug. I will check internally to see if there have been any bug fixes since 2024 R1.1 and let you know. I will recommend you to uncheck the "use max index" (just below "search near n"). Then in the next field "n", try setting it's value to 2.18 (close to the correct mode's). Then run the simulation multiple times to see if you still see these strange modes in some of the runs.

Secondly, for the background mesh, try using fewer mesh cells, you may try the following:

Increasing the width of the background mesh makes the PML thicker and can increase it's performance.

It still puzzling to me why this strange behavior would appear only for some of your runs. I will look at this in a bit more detail and come back.

Regards,
Amrita

[Henrick Höllerer]

 

 

Hello Amrita,

thank you very much for your help.
I now tried both of your suggestions, setting n=2.18 and using fewer mesh cells for the background mesh. 
Unfortunately, the error still occurs but interestingly only when I execute the code for the first time (e.g. after restarting MODE):

The effective indices of the found “false modes” are then equal to the effective index I set in the modal analysis tab. I re-run the (exactly same) simulation 20 times and only observed the error for the first run - which is at least significantly less than before. Moreover, I repeated the sweep of the waveguide top width with the new settings and run the code on a different computational server of my university, too. In both cases I made again the observation: for the first time running the simulation the strange mode result occurred.

Best regards,
Henrick

 

 

[Amrita Pati]

Hi Henrick,

Thanks for testing! It most likely is a bug. Unfortunately, I will have to reproduce it to submit a bug report. Since, I do not have the file myself I will see if I can create something similar. I will keep you posted.

Regards,
Amrita