Learning Forum

[guodong.zhu]

Hi expert,

I tried to simulate a gold structure using RCWA and FDTD solver from Lumerical, but they gave quite different results.

For RCWA, I have tried to increase the layers of gold up to 50 or increase the number of k vector up to 3000.

Here are the results:

RCWA:

Refection:

 

FDTD:

Refection:

Can anyone give me any suggestions?

Regards,

Guodong

[Amrita Pati]

Hi Guodong,

You have run the the RCWA and FDTD for different ranges of wavelengths. I think it will be easier to compare them for the same range.

Since you are using gold, I would expect that there is a large index contrast between the gold and it's surrounding. I will recommend using the tangent vector field method as shown below:

Do your FDTD results look accurate to you? Did you perform any convergence testing on the FDTD simulation? Are these results for normal or angled incidence? Thank you!

Regards,
Amrita

[Amrita Pati]

Hi Guodong,

I looked into it a little bit internally. It seems that we have seen convergence issues in RCWA while simulating metallic structures, when there is a large refractive index contrast. This has also been documented: https://web.stanford.edu/group/fan/publication/Liu_ComputerPhysicsCommunications_183_2233_2012.pdf.

It is common to use a large number of k-vectors in plamsonic structures (like you are using) to gain convergence. And as I previously mentioned, using the tangent vector field method will help. Let me know if you are able to see improved results.

Regards,
Amrita

[guodong.zhu]

Hi Amrita,

Thank you very much for your insightful comments. Following your suggestions, I have obtained some new results, which I have shared below.

 

Regards, 

Guodong

 

[guodong.zhu]

Hi Amrita,

Thank you for the comments.

I tried the tangent vector field with 3000 k vector (it ran almost 24 hours). The results improved but still displayed significant differences when compared to the FDTD simulation results

Below is the comparison:

I also compare the results with or without checking tangent field vector for 1000 k vectors.

Here is the field distribution (xy cut) at the center of my structure for the wavelength at the peak. Gold structure is at the center.

RCWA:

FDTD:

 

Here is my RCWA solver settings:

Do you have any furthure suggestions? I am still runing simulation with 4000 k vectors. However, it might take a very long  time.

 

Regards,

Guodong

[Amrita Pati]

Hi Guodong,

I don't really think I have any other suggestions. I was expecting the fields to look more refined for such a high k-vector value. This might mean that we might require more k-vectors. Let's wait for the 4000 k-vector result. I can reach out to the R&D team to see if anything else can be done.

Regards,
Amrita

[guodong.zhu]

Thank you!  please reach out the R&D to see what else we can do.

 4000 k-vector might requaire  a very long time run.

Regards,
Guodong

[Amrita Pati]

Hi Guodong, 

I spoke to the R&D team on this. Unfortunately, they also agreed that the large index contrast can make it very challenging for RCWA to attain convergence. The last thing that they recommend is changing the number of mesh cells. I believe the "mesh type" of your simulation is "auto", which automatically refines the mesh based on the number of k-vectors. You can change the mesh type to "custom":

Then fix the k-vector at one value (let's say 1000), and increase the mesh cells, for example: 100x100, then 200x200, and so on. Let me know if you see any improvements.

Regards,
Amrita