Learning Forum

[marius.crouzier]

Hello,

I am simulating an optical resonator on a susbstrate and looking the reflected field.

The substrate is made of Si02 with a 220nm layer of Si and 3 nm of native SiO2.

The nanoparticle is a 150*100nm rectangle with 3 layers : 2 of gold seperate by on of SiO2 the height of the gold rectangles is 30 nm and 50 nm for the SiO2 one.

The plane wave source is above the nanoparticle, I use a power monitor to calculate the reflexion coefficient. For the boundary conditions, I have PML for zmin and zmax, antisymetric for xmin and xmax, and symmetric for ymin and ymax.

I also use a mesh refinement for the nanoparticle and the area of native SiO2 just below the nanoparticle with a mesh step of 3 nm.

I have run this simulation and calculate the reflexion coefficient for different mesh accuracy ( 2-7) as you can see in the following figure.

R_meshAcc.png

[Guilin Sun]

Thank you for such efforts doing the converge.
However, converge not only depends in the mesh accuracy, it also depends on the PML performance, the autoshutoff min and the simulation time. When you change the mesh accuracy, make sure the PML thickness does not change much, better to keep about the same. Take the example of mesh accuracy 2, suppose 8 layers are sufficient, then mesh accuracy 4 may need 16 layers etc. In addition, please make sure that the period is meshed in an integer number of meshes. I would suggest that you override mesh in the periodic direction. You can change the mesh size according to the mesh accuracy. for example, you use override mesh close to mesh accuracy 2, then you will need 1.8 ~2 time smaller for mesh accuracy 4. The mesh accuracy 2,3,4,5,6,7,8 correspond to N=10,14,18,22,26,28,and 32 where the mesh size is min(wavelength)/N .
Once the mesh is set properly, please modify PML, and autoshutoff min, and use long simulation time to let the autoshutoff min to terminate the simulation.
One more thing: the conformal mesh technique also plays role in the accuracy, in particular when metal material is involved. In this case, If you want to do converge test, it is recommended to use "conformal variant 1":
Selecting the best mesh refinement option in the FDTD ...
Understanding Mesh Refinement and Conformal Mesh in FDTD
Please refer this article for FDTD convergence testing: Convergence testing process for FDTD simulations

[marius.crouzier]

Hello,
Thank you very much for your answer, it helps.
However, I need more explanations :
What is the period which need to be meshed in an interger number of meshes ?
For the autoshutoff min and simulation time, I need to decrease the autoshutoff min and increase the simulation time. What are the typical values for these parameters ?
Thanks,

[Guilin Sun]

It is the xy period. As you know, the results depend on the periodic nature in XY: antisymetric for xmin and xmax, and symmetric for ymin and ymax.
The typical values are the default ones. They work most of the time but not for every simulation.