Learning Forum

[azarulsani]

I am currently trying to simulate light propagation thru a Mach Zehnder Interferometer (MZI) polymer based waveguide for heavy metal sensing.

Right now, i am facing problem when trying to simulate a 5000 um x 5000 um and bigger design sizes. My workstation laptop became so laggy or at certain time, it ran out of memory and unable to complete the simulation.

My workstation laptop spec is as follows:

Lenovo P14S Gen 2

Intel i5 11th gen 1135G7 CPU (4 cores 8 threads) | 40GB DDR4 RAM | Nvidia T500 4GB DDR5 discrete graphics card 

Here are some parameters that i used for the simulation:

Substrate Silicon (Si) = 5.5 um thick | Silicon Dioxide (SiO2) = 4.5 um thick 

core - SU8 = 4 um width, 4 um height

cladding - air

i've tried changing the settings in varFDTD - Mesh step from 1e-6 right up to 0.1.

is there any other parameters that i should look into? or varFDTD has device size limitation for the simulation to run successfully? or i need to get a better workstation spec?

Appreciate your help. Thanks. 

[Guilin Sun]

It is not the absolute size that matters, it is the optical size (in microwave people call electric size) that matters. For dielectric materials with refractive index n, the optical size is geometrical length divided by min(lambda0/n) where lambda0 is the wavelength. For heavy lossy materials such as some metals, it is the skip depth.

FDTD is a discrete method. therefore the simulation volume has to be discretized into small "bricks", called Yee cells. The default meshing size is min(lambda0/n)/N where N is called mesh density. By default our FDTD uses N=10, which is a coarse mesh. For more accurate result, N can be larger.

In addition, the simulaiton needs to store the geometry buildings, the material property, source, monitor and boundary conditions. therefore FDTD has an option to let you check the memory requirement before running. If

the running memory requires more, then you will need to upgrade your computer to cloud or HPC.