Min size autodetect

[dbochkov-flexcompute]

I am not sure about naming SmallestFeatureSpec and also how deep it is tucked away:

td.Simulation(
    ...
    grid_spec=td.GridSpec.auto(
        layer_refinement_specs=[
            td.LayerRefinementSpec(
                corner_finder=td.CornerFinderSpec(
                    dl_min_spec=td.SmallestFeatureSpec(
                        concave_resolution=1,
                        convex_resolution=2,
                        mixed_resolution=3,
                    )
                )
            )
        ]
    )
)

would appreciate any suggestions

[weiliangjin2021]

Regarding deeply nested field, maybe concave_resolution etc. can be part of CornerFinderSpec fields directly? We have a field distance_threshold already, that might also have some impact on the minimal feature.

Another questions is what would be good default values for those resolutions?

[weiliangjin2021]

Went through it again, and now makes more sense to me: your default value in the code is probably good enough.

A high-level question: this PR looks also partially resolve the short-issue. I guess they will work together?

[weiliangjin2021]

convex

[weiliangjin2021]

Not sure if we should use 0 or None when not applied

[weiliangjin2021]

np.cross seems to have issues in photonforge: #2317

[dbochkov-flexcompute]

Went through it again, and now makes more sense to me: your default value in the code is probably good enough.

A high-level question: this PR looks also partially resolve the short-issue. I guess they will work together?

yeah, it does partially resolve it, but not as robustly as the other one. For example, if structure has rounded corners everywhere this autodetection feature would not see anything. But given that the other approach is more computationally intensive, might still be good to have a rough estimation like this one

[weiliangjin2021]

Just tested the microstrip: I removed lumped elements and substrate layer refinement so that they don't set the minimal grid size. Before this PR, I get

And with the default parameters in this PR, I get

Quite nice meshing!