Bandstructure simulation runtime behaving strangely when changing materials


#1

Hi everyone,

I’ve recently observed some strange behavior re: sporadic runtimes for bandstructure simulations. I am calculating the bandstructure for a standard photonic crystal. I have been using the same set of parameters for some time (fairly reasonable, mesh accuracy at 2, 12 PML layers etc, nothing crazy).

A standard run at a single point in k-space previously would take less than one minute, very consistently. Today I attempted to change the material to compare results (previously SiN with user input refractive indez, changing to GaAs which already had a material profile in the drop down.)

When I did this, a single k space point simulation jumped to ~7 minutes. Also, when I run my optimization code, the progress behaves strangely, with some simulations stalling and others progressing until 0-5 seconds remaining and then stalling, as shown below.

Any idea what might be going on? Could a simple change of material cause such a drastic change in runtime?

Thanks,
Chris


#2

Dear @cflower

Thank you for detailed post and sorry for a late reply.

This is strange and I don’t have a clear answer for you. Can you please upload your simulation file for a review?

Thanks


#3

Not a problem, thanks for replying! I am not sure I can post the code (I’ll have to talk to my collaborators, there may be some restrictions on what can be uploaded…)

However I did fiddle around with it and now the simulation time mismatch is manageable. 1 minute vs ~2.5 minutes.

What I discovered was that with the SiN sample, we simply used “user defined dielectric” and input the refractive index explicitly. However for GaAs, we used the material entry that was already in the library “GaAs - Palik”

I tried using user defined dielectric and inputting the GaAs refractive index and it reduced the runtime significantly.

This is curious to me, I’m not sure what other information is encoded in the material… maybe I will make a separate post and ask about that.


#4

Dear @cflower

Thanks for additional information.

I was slightly confused that material fitting causing this problem but here is why changing material results in increased simulation time: When you change the material, you should expect to see a completely different behavior. The resonance modes of photonic crystal cavity might be improved in the case of GaAs material, resulting in longer time to trigger the auto shutoff level. This can be true when you change the material from a fixed dielectric constant to dispersive material.

To summarize, material fitting takes a very small time to be done (say micro seconds), however, the physics and behavior of simulation might change drastically resulting in completely different electromagnetic field behavior which can increase simulation time.

I hope this answered your question.


#5

I see, thanks for your reply. That explains it. Unfortunate that I probably can’t do much about it!