Learning Forum

[Ruzan]

Hi everyone,

We are interested in simulating high quality factor metasurface structures with quality factors of 50,000. When using conventional Lumerical engine licenses, the simulation lasts ~12 hours even if I use 120 cores. I am wondering if someone has benchmarked Lumerical FDTD vs. Comsol. Does anyone know if it is more optimal to use Lumerical or Comsol when considering high quality factor structures?

Also, has anyone studied which software (Lumerical or Comsol) provides more accurate result within a reasonable time frame?

Thanks!

[Guilin Sun]

Comparing performance is very challenge and depends several factors. Lumerical's FDTD is the benchmark in the maret for a long time and with proper settings it is the fastest of its kind, in particular in 3D simulations.

Are you only interested in the quality factor and resonance frequency, or total transmission? if it is only qualifactor and resonance, it should be much faster as it does not need to simulate to fulfill the autoshutoff min, with either Fourier transform method or findresonance. There are several examples online.  As early as in 2008 Harvard is the first to demonstrate Q ~1 million from a phtonic crystal cavity.

If it is for transmission, autoshutoff min should be fulfilled. However it may not need about 12 hours. FDTD, in particular Lumerical FDTD is well designed for parallel distributed computing. However limited by the hardware configration and specs, the "scale" may not be linear as the number of cores. that said, more cores do not necessarily mean faster simulation. Too many cores for small jobs can actually hinder the simulation speed. You will need to test the Resource configuration and see how many cores/processes/threads are the most efficient, by checking the estimated simulation time.

I have a post for slow convergence:  https://forum.ansys.com/forums/topic/ansys-insight-slow-convergence-in-fdtd/#post-205484 

Please have a look.

If it is still a problem please give some screenshots for the device and FDTD settings. and if you have the premium support you can also email us.

[Ruzan]

Thank you for your response.

 

I am interested in the transmission spectrum of the resonator array. I need to set autoshutoff to 10^-9 so that the transmittance spectrum looks reasonable. I use a point monitor to extract transmittance and phase, which is located 7 um away from the structure to ensure that I have a plane wave at the location of the point monitor. My operating wavelength is 1550 nm. 

I will try the suggestions in the link to see whether I can reduce the simulation time.