Learning Forum

[Handon Um]

The RCWA and FDTD show different reflectance results in the same regular pyramid structure.

I know the RCWA has a relatively lower accuracy than FDTD, so I tried to improve the accuracy of RCWA results.

 

I have changed the number of relative interfaces k vectors, the number of relative interfaces, and the k domain.

RCWA results show something changed as k vector increases and the number of relative interfaces, but the results are still different from the FDTD result.

(k domain didn't affect the accuracy of results.)

 

I don't know the reason why the RCWA shows a big difference from FDTD.

The value of k has been increased to the level that the computer's memory can run programs.

Is there a way to improve the accuracy of RCWA to a level similar to that of FDTD?

 

The simulation results and my program files are attached below.

If you can, please check my setting of FDTD and RCWA in fsp files.

 

https://drive.google.com/file/d/18zMcwKmJRkb7Cq-v0TNkE-jDUx5WI2dm/view?usp=drive_link

https://drive.google.com/file/d/17Z0xKsg4beiLttKTG4WpC0HpMSM9u95Q/view?usp=drive_link

[Amrita Pati]

Hi Handon, 

Thanks for sharing detailed results of your convergence tests! There could be a number of reasons why the FDTD and RCWA results don't match. 

1. Material Fitting: Both RCWA and FDTD use the multi-coefficient model for material fitting. I would recommend you to check the material fitting in both the cases to see how they compare.
2. Mesh/k-vectors: You mentioned that the k-vectors do not affect the results, it can be because the structure does not need a large number of k-vectors and attains convergence with a small number. But, did you do convergence testing in FDTD wrt the mesh refinement? I am just curious about how confident you are about the FDTD results.
3. Can you also check the index result in both FDTD and RCWA to see how the structure is resolved?

Sorry, I am not allowed to download any files, but would you be able to share screenshots of the simulation region in FDTD and the screenshots of Mesh Settings and Boundary Conditions Tab of FDTD?

Regards,
Amrita

[Handon Um]

Thanks for your kind reply.

 

First, I set a result from PVlighthouse's calculator (OPAL2) as a reference. And the trend is similar between the reference and my FDTD results.

So I can be confident with my FDTD results.

The problem is the RCWA results. It's quite different from the reference and FDTD results.

I found the index results and material fitting are okay.

Why do the current RCWA results differ from FDTD?

Also, I checked the link so that you can download my files.

RCWA: https://drive.google.com/file/d/1Y6KB5FzyHkrw02NlFsnB8AUykna4fW3q/view?usp=sharing

FDTD: https://drive.google.com/file/d/17Z0xKsg4beiLttKTG4WpC0HpMSM9u95Q/view?usp=sharing

 

[Amrita Pati]

Hi Handon,

Thanks for sharing additional details. I apologize for the late reply, I was off for a day and we also had a long weekend here. Let me think about this in more detail. I will get back to you soon.

Regards,
Amrita

[Handon Um]

Hello, Amrita

Can I ask you if there is something you figured out for my queries?

I'm still waiting for your reply.

[Amrita Pati]

Hi Handon,

Sorry I completely missed out on this post. So, the thing is that we are not allowed to download files from ALF as per policy. From the screenshot of the geometry tab I do not see any issue with the setup of the structure. In cases like this we can only advise user to perform convergence testing with larger number of interfaces and k-vectors until they see the expected behavior/results. I would like to point out that PVlighthouse's calculator result doesn't consider any material absorption which is not the case here in FDTD or RCWA. But the main problem is that the results of FDTD and RCWA are not converging. I would ask you to a run the simulation with an arbitrarily large number of k-vectors and interfaces. Check if you see any resemblance to the expected values. Check the "use tangent vector field" (typically used for metal-dielectric intefaces or high index-contrast) in this calculation as well. If you still do not see any improvement, let me know. I will try to get more details from you and run some tests.

Regards,
Amrita

[Handon Um]

 

Hello, Amrita.

In the case of 0.5 um pyramid,

When the pyramid structure is made of several layers by a script, the results of FDTD and RCWA converge well.

However, the results start to show a difference again when the pyramid size increases.

It seems that more memory would be needed to run the RCWA accurately.

 

 

[Amrita Pati]

Hi Handon,

Thanks for letting me know! It is nice to see that you have found an approach to solve the problem that I myself wouldn't have thought of. It is very interesting that you are attaining convergence when you create this pyramid using 10/100/200 rectangles but at the same do not see that when using similar number of interfaces in the RCWA algorithm. The algorithm should automatically divide the pyramid into this layered structure, the user shouldn't have to worry about this step. One question for you, I see that you have compared the reflection from the layered structure in FDTD and RCWA. But how does the reflectance of the layered structure in RCWA compare to the reflectance of the smooth structure in FDTD?

I still think that this is worth more investigation on our end. I will try to bring this up in one of our internal meetings and let you know if we gain some insights into it.

Regards,
Amrita

[Handon Um]

Hello, Amrita.

I missed one thing in my test.

The size of the pyramid structure in my latest reply is 0.5 um while the size of the first pyramid that occurs a large gap in the result of RCWA is 5 um.

When the pyramid size increases, it starts to show a difference between the FDTD and RCWA.

 

I think the thing is that an increase in the pyramid size requires a lot of resources to get accurate results.

 

I sincerely apologize for my mistake.

[Amrita Pati]

Hi Handon,

No worries at all. That is correct, you will require more resources in FDTD and RCWA. That's why we ask customers to do convergence testing. At least, with regard to the mesh accuracy. 

Regards,
Amrita