View groups and posts below.
This post is from a suggested group
I tried to run a straightforward geometry of layered Gold(20nm)/Alumina(140nm)/Gold(20nm)/Glass(500nm) structures. I want to study the resonance of this cavity structure. I sweep over the wavelengths(400nm to 1000nm). But I got the following message: The requested 128x128x76000 (18.6GB) array exceeds the maximum array size preference (13.8GB). This might cause MATLAB to become unresponsive. Do you have some suggestions for me?
This post is from a suggested group
Hi everyone, I'm running a MATLAB simulation using the WaveSim function. However, I'm having some issues using Mex. I run a small version of my simulation on my laptop, where I can use the Mex option. For my full-scale simulation, I run the code on a Linux-based cluster. Is there a Linux version of the Mex file I think it is mexa64 or is there a source file I could use to make this mexa64 file?
Hi Marco,
Thanks a lot for pointing out the issue(s) with Mex! You mentioned that setting `opt.usemex=True` now works for you. Is that in MATLAB 2024 or an earlier version? Due to some changes in MATLAB 2024, Mex is currently not working, but we have already starting working on fixing that. We will soon push an update for MATLAB 2024.
As for Linux, we currently do not have a version that works with it, but we will keep you updated on developments!
This post is from a suggested group
Hi everyone, Is there a benchmark/comparison for the performance of the package for higher index contrast configurations like a waveguide (silicon on insulator or other)? The FAQ mentions that the convergence will be slower but it would be interesting to know how much slower and how it compares to other methods. Does anyone have information about this?
Thanks for the response. I checked the Brueker and Osnabrugge papers. But was curious cause those compare for low index contrast. Particularly the Brueker case fig 6 has a 3500x difference in time. @Swapnil you mention a factor 53 vs 3500. Both are solid improvement. Can you comment on why sometimes the speed increase vs FDTD is larger than in other cases? @Evangelos Marakis: From a user perspective I would say I care about the total volume I can simulate in reasonable times with reasonable accuracy. If FDTD needs finer meshing for that, it just factors into that total volume. More or less all methods struggle with high index contrast so it's nice to see you a comparative advantage also in those cases. The other tricky case is metals, where FDTD and others need even finer grid steps.
This post is from a suggested group
Hi all,
I have a question regarding Wavesim capabilities. I am a beginner to EM simualtions.
Can it be used for periodic structures (varying refractive index as cosine function in 16micron depth). I am using RCWA (Maxwells equations) for this which has to descretize a 16 micron thickness into hundreds of layers to give good result. I was wondering if wavesim will provide a faster solution for this. I am interested in polarisation, diffraction, reflection and transmission of my structure.
Thank you.
Hi, Imaad. Welcome to the Wavesim forum! If my understanding of RCWA is correct, it requires a constant refractive index in the z-direction (for each layer). Is that right? Assuming yes, Wavesim does not have this limitation, and you can assign a different refractive index for every pixel/voxel. I think there's some potential for improvement in speed or accuracy, or maybe even both! We can help you set up a simulation to test this.
This post is from a suggested group
When the domain decomposition will be out??
The Python code with domain decomposition of the MBS for the Helmholtz equation is now out on GitHub! The arXiv preprint that goes along with it: https://arxiv.org/abs/2410.02395
This post is from a suggested group
Welcome to our group Wavesim Forum Group! A space for us to connect and share with each other. Start by posting your thoughts, sharing media, or creating a poll.
Hi Sudhir,
Thanks for your question. There could be two ways to go about solving this:
Assuming you have enough RAM (>> the 18.6 GB array you are trying to create), the issue could be the maximum array size limitations in MATLAB. You could try increasing this limit following the solution in one of these links: 1, 2, 3
If you do not have enough RAM, defining such a large problem won’t be possible without some tricks. In this case, could you confirm a few things for us? a. Whether setting the flag opt.usemex=True solves your problem (or if there is any difference at all), since it handles the memory a bit better b. What boundary widths do you use? Setting boundary_widths=2*wavelength could save some memory c. The array size is considerably large, and understandably so, since you need to sample at a very high rate to resolve the thin gold layer. Could you confirm what PPW value you use? If the above approaches don’t work, it seems the only way to fit the problem on your system would be by reducing its size (either by reducing the PPW as far as possible or even shrinking some of the layers if that is feasible)
We look forward to hearing from you. Hopefully, one of these suggestions helps, but let us know either way!