[jeffhsu3]

Adds two optional layer-stack fields. Both are motivated by transistor layers, but neither is specific to a PDK.

cut_by

A list of layers subtracted from this layer before it is extruded. The layer is built from region - cut, so the removed area is genuinely absent from the mesh rather than covered up by something drawn on top of it.

wrap_around

Names a reference layer that this layer wraps down around instead of being extruded flat. The layer still caps the reference over its own z range, and a second slab carries the walls down beside the reference to z_extend below it:

    z+height ┌──────────────┐  ← cap (the layer's own polygons)
             │              │
         z   ├──┬────────┬──┤
             │  │        │  │  ← walls (layer NOT reference)
   fin top   │  │  fin   │  │
   fin bot   │  └────────┘  │
  wrap_z_bot └──┘        └──┘

That is what makes a FinFET gate read correctly. Without it the gate floats above the fins with them poking up through it.

You mentioned that transistor layers currently carry a very small offset to avoid render artifacts — ihp-sg13g2.yaml has Activ at z: -0.3995 and GatPoly at z: -0.399, half a nanometre apart. wrap_around is meant to be a version of that for the gate/reference case, since the two layers then no longer share a z range at all. I have not touched any existing config here; if it is useful I can look at whether any of them could drop their offset. This is one possible way to implement the fin/gates for ASAP7.

Implementation

_build_mesh now takes a list of z-slabs and welds them into a single mesh, and _slab_geometry holds what used to be its body. That is what keeps wrap_around cheap: the walls reuse the same hole triangulation and boundary-edge walls as the cap, from import_gdsii: Build the extruded layers with numpy. A layer with neither field is one slab and follows exactly the path it did before.

Testing

I exercised both fields on sky130 through a custom config rather than only on ASAP7 — poly wrapping diff and cut by licon. The wrap produces the expected three z levels, a layer with neither field still produces two, and no poly vertex lands inside a licon shape. Meshes come out with no degenerate faces and no loose vertices. tox passes. Need to double check gate cut on ASAP7 though as its still listed in the layer window.

🤖 Generated with Claude Code

[dnltz]

@jeffhsu3 - please remove Claude-Session and your Signed-off-by should have your full name, not a nickname. I will review later :)

[jeffhsu3]

Thanks sorry!

[dnltz]

I locally tested it and it looks great. Thanks!

Claude found one bug and has one suggestion. Can you take a look at them?

[dnltz]

LGTM! Can you please delete the Claude-Session line from your commit? It's not visible to anyone except you and shouldn't be committed.

[dnltz]

@jeffhsu3 thank you!