Improved automatic/default meshing for RF

[yuanshen-flexcompute]

Current issue:

The default mesh (or lack thereof) for RF often does not give good results. From my experience so far, here are some common causes:

• Metallic structures are not adequately resolved. The solution is to manually add layer refinement. It would be good if this could be automated in some way.

  • A sub-issue: layer refinement does not currently work for 2D metallic structures (zero thickness). It would be great if we allow layer refinement spec to work with zero thickness in such scenarios (corner/bounds snapping/refinement)

• For subwavelength simulations, the min_steps_per_sim_size setting is more appropriate than the default min_steps_per_wvl. The best number for that setting is not 100% clear to me yet -- whether it's case-by-case dependent on geometry, or whether we can find a happy number that works for most cases.

• A longer term goal is automatic mesh refinement (in the style of HFSS). This could be a custom study type, e.g. modeler.run_mesh_refinement(). The idea is to iteratively refine the mesh until some figure(s) of merit (usually S-parameter) converges.

Discussion:

The goal of creating this issue is to start the conversation on how we can better set up meshing for RF, in order to lessen the burden on users. Some questions on my mind:

• Where are we implementing this? Front end? API level? Applets?
• What does the implementation look like? Default settings for API functions? Custom auto-called "mesh setup" functions in the RF module? Others?
• Transparency and versatility are important -- the user should be able to see what is added by default, and easily modify it if they wish.

Happy to hear everyone's opinions.

[yuanshen-flexcompute]

This is not one issue with one fix, but I figure it would be good to start an issue for tracking. Improvements on default meshing has been requested by @Frank-DingYC , especially to make the front-end GUI more user-friendly.

Tagging RF team @weiliangjin2021 @dmarek-flex @dbochkov-flexcompute . Feel free to discuss. Thanks!

[yuanshen-flexcompute]

Here's my idea for automatic layer refinement:

• Include it as a parameter option in TerminalComponentModeler or some other class
• Automatically scans the structure list for medium = PEC or LossyMetal. Alternatively, the user can provide their own list.
• Automatically creates LayerRefinementSpec for each structure. If structures are in the same z-plane, consolidate them into one LayerRefinementSpec.
• The user can control the settings for LayerRefinementSpec, but we could choose intelligent defaults. E.g. for corner refinement, we can set dl to be a fraction of the smallest structure geometry length (hard) or a fraction of the shortest side length of the bounding box (easy).
• Warnings/checks to make sure that min step size is not too small on the generated mesh. The most probable cause is if the metal layer is too thin, without being 2D.

This is just a rough idea with lots of implementation details missing. I'll do some additional testing on my own, to see if it's useful.