How to Choose Between FDTD, FEM and MoM | Open-Source Codes

[martindmaas]

TL;DR Summary

How to choose between FDTD, FEM and MoM for Electromagnetic Simulations, discussion about Open Source Software.

Hi,

I've just written an article about how to choose between FDTD, FEM and MoM for Electromagnetic Simulations, and also with brief reviews of some of the best Open Source alternatives out there.

https://www.matecdev.com/posts/differences-fdtd-fem-mom.html

I've seen that in this forum many people seem to be using commercial simulation software. Just wondering... Are you using this commercial software because open-source codes fall short in some regards?

Thanks for sharing your view!

 

[berkeman]

Welcome to PF.

Interesting article and blog site. I'll have to spend some more time reading through it.

 

[Joshy]

I think you covered choosing the preferred tool properly. To summarize it I would use time domain simulations on electrically large models; FEM frequency domain for 3D models meaning z-direction has a bigger impact on the results, and MoM for 2D models like planar structures.

I’m not familiar with the open source for these as I only use what my employer provides or authorizes if whatever they have isn’t helping me meet the requirements. I haven’t ran into any major issues with the industry tools and results often align with measured results. I can speak about my experience with open source tools. They are usually incomplete or the models are too basic; they also aren’t friendly with other tools including the industry standard say for instance if I were to need to share my setup or results with a counterpart or different type of engineer using different tools. I would suspect but do not know that the adaptive process (especially meshing) has fast algorithms to help the simulations converge more quickly.

 

[martindmaas]

I think you covered choosing the preferred tool properly. To summarize it I would use time domain simulations on electrically large models; FEM frequency domain for 3D models meaning z-direction has a bigger impact on the results, and MoM for 2D models like planar structures.

I’m not familiar with the open source for these as I only use what my employer provides or authorizes if whatever they have isn’t helping me meet the requirements. I haven’t ran into any major issues with the industry tools and results often align with measured results. I can speak about my experience with open source tools. They are usually incomplete or the models are too basic; they also aren’t friendly with other tools including the industry standard say for instance if I were to need to share my setup or results with a counterpart or different type of engineer using different tools. I would suspect but do not know that the adaptive process (especially meshing) has fast algorithms to help the simulations converge more quickly.

Thank you Joshy.

I would be aware of using FDTD in electrically large problems... it could be really computationally expensive. If MoM can also be applied to that problem, it should be a lot faster, if the underlying MoM implementation works correctly.

As for interoperability with industry tools, that's a good point. I believe that could be a weak spot of most open-source tools, indeed. I guess the main issues would be multiphysics integration, and the possibility of writing CAD files as output of the models, am I right?

 

[Joshy]

I opt for 2.5D MoM simulations whenever I can or if I’m only looking for quick qualitative insight, but it becomes less accurate when the geometry is not planar say for instance a connector, via transition, or cavity; also when I’m simulating something more assembled say for instance mounting on a tall pole or on an aircraft. Some tools have some great secret sauce features such has a hybrid type of simulation where it meshes differently in different regions. Try a simulation with two really long lines with side by side via transitions say for instance underneath a BGA and let me know if MoM gives you accurate results on the coupling. You’re going to want the 3D meshing on those via transitions for sure. All of the MoM I’ve worked with are also really difficult to include a lot of higher order effects say for instance something above 10 GHz I’ll probably really want to include skin effect, but have often have to kind of cheat and just manually say the metal is a lower conductivity. I think it’s because when the 2.5D simulation performs mathematical operations there are some assumptions that the 3D transform will be more uniform than it is in real life. Another thing I forgot to mention is sometimes I’m working with really sensitive stuff that I don’t want to share in detail if I don’t have too, and so I wouldn’t want to work with something open source if I want to send an encrypted model to my counterparts.

There’s obviously a tradeoff between computational resources (or time) and accuracy depending on what you’re simulating. If you blindly opt for MoM trying to cut corners for a faster simulation in my BGA example above or a flexible substrate with different zones such as coverlay layer…. sure you’ll save time, which will give you plenty of time to clear your office or cube before your stakeholders come to you. It’s not to say that you cannot achieve correct results with MoM, but it requires a lot more thought and more detailed setup.

 

[martindmaas]

I opt for 2.5D MoM simulations whenever I can or if I’m only looking for quick qualitative insight, but it becomes less accurate when the geometry is not planar say for instance a connector, via transition, or cavity; also when I’m simulating something more assembled say for instance mounting on a tall pole or on an aircraft. Some tools have some great secret sauce features such has a hybrid type of simulation where it meshes differently in different regions. Try a simulation with two really long lines with side by side via transitions say for instance underneath a BGA and let me know if MoM gives you accurate results on the coupling. You’re going to want the 3D meshing on those via transitions for sure. All of the MoM I’ve worked with are also really difficult to include a lot of higher order effects say for instance something above 10 GHz I’ll probably really want to include skin effect, but have often have to kind of cheat and just manually say the metal is a lower conductivity. I think it’s because when the 2.5D simulation performs mathematical operations there are some assumptions that the 3D transform will be more uniform than it is in real life. Another thing I forgot to mention is sometimes I’m working with really sensitive stuff that I don’t want to share in detail if I don’t have too, and so I wouldn’t want to work with something open source if I want to send an encrypted model to my counterparts.

There’s obviously a tradeoff between computational resources (or time) and accuracy depending on what you’re simulating. If you blindly opt for MoM in my BGA example above or a flexible substrate with different zones such as coverlay layer…. sure you’ll save time, which will give you plenty of time to clear your office or cube before your stakeholders come to you.

Yes, some or even most of the MoM implementations, can definitively have meshing/convergence issues.

In my research group, we once run a detailed comparison of our research software vs a commercial MoM and found that (probably) the secret sauce in the MoM was leading to wrong solutions and non-convergence, even for a relatively simple geometry. Things like connectors or small gaps in high-frequency problems are major offenders I guess. I guess that open-source software doesn't include the secret sauce so at least it is easier to tell when something isn't working.

Regards!