MIM cap layer lumps dielectric + top plate into one SiO₂ level — doesn't work for sims

[vvahidd]

Summary

The mim layer in the LayerStack (tech.py) lumps the MIM dielectric and the MIM top-plate metal into a single 0.19 µm LayerLevel made of sio2. This is fine for layout/DRC/3D-visualization, but it makes the MIM capacitor unusable for EM/electrostatic capacitance extraction — it under-predicts the cap density by ~7–8×.

Current definition

ihp/tech.py (v0.2.8):

# MIM capacitor - dielectric 40nm (TISMIM) + top plate 150nm (TMIMTOP)
# Sits on top of Metal5
mim=LayerLevel(
    layer=LAYER.MIMdrawing,
    thickness=0.04 + 0.15,   # = 0.19 µm
    material="sio2",
    info={"mesh_order": 23},
),

The comment itself documents that this single block is really two physically distinct things:

1. a 40 nm MIM dielectric (TISMIM) — the actual capacitor dielectric, a high-k nitride (≈ SiN, εᵣ ≈ 6.8–7.5), not SiO₂, and
2. a 150 nm MIM top-plate metal (TMIMTOP) — a conductor, here modeled as dielectric.

What the official IHP-Open-PDK does

The authoritative EM cross-section sg13g2_for_EM.xs models these as two separate layers grown from the same MIM mask:

t_mimdiel = 0.04                          # 40 nm MIM dielectric
mimdiel   = mask(mask_MIM).grow(t_mimdiel)
t_mim     = 0.15                          # 150 nm MIM top-plate metal
mim       = mask(mask_MIM).grow(t_mim)
t_vmim    = t_topvia1 - t_mim - t_mimdiel
vmim      = mask(mask_VMIM).grow(t_vmim)

So the canonical stack is: Metal5 → 40 nm dielectric → 150 nm top-plate metal → Vmim → TopMetal1. The single lumped sio2 block is what diverges from IHP's own EM reference.

Suggested fix

Split the single mim level into two LayerLevels (both can reference LAYER.MIMdrawing):

mim_diel=LayerLevel(
    layer=LAYER.MIMdrawing,
    thickness=0.04,           # 40 nm MIM dielectric
    zmin=z_m5_top,
    material="sin",           # high-k nitride, εᵣ ≈ 7.5
    info={"mesh_order": 23},
),
mim_top=LayerLevel(
    layer=LAYER.MIMdrawing,
    thickness=0.15,           # 150 nm MIM top-plate metal
    zmin=z_m5_top + 0.04,
    material="aluminum",      # conductor (top electrode)
    info={"mesh_order": 23},
),

This matches sg13g2_for_EM.xs, leaves the GDS/DRC footprint unchanged, and reproduces the 1.5 fF/µm² spec capacitance. The Vmim zmin/thickness already account for the 0.19 µm offset, so they remain consistent.

[vvahidd]

@das-dias

[vvahidd]

i think we also have abstract layers in gf, where we define a layer that is not for physical meaning but for either simulation or marker purposes