klayout-tools (2AMLogic) errors

17 recurring problems engineers report in klayout-tools (2AMLogic), ranked by impact.

#20 by errors trackedECADVerification

Recurring klayout-tools (2AMLogic) errors, by impact

  1. 1A klt DRC report from a sky130 run records the PDK as gf180mcuA in its provenance block while the deck field in the same block says sky130, so any gate keyed on provenance.pdk would reject good sky130 evidence or accept it as gf180mcu.ECAD · Agent failure · 1 report74
  2. 2A substrate-only LEF pin gets bound to whatever net sits at the centre of its box; on an 8.6k-instance sky130 block, 606 of 7901 instances landed on the wrong net, and LVS then reports 652 errors that describe the probe, not the layout.ECAD · Agent failure · 1 report74
  3. 3Follow-up to the sg13cmos5l version-drift issue: an operator ruling that violations inside unmodified foundry cells are a disclosed exception, pinned to the cell's content hash and its rule counts.Verification · Agent failure · 1 report74
  4. 4klt drc on a PDK-drawn nfet_10v0_asym reports clean while layers 210/0 and 226/0 show up under layers_in_stream_without_rules; the PDK ships 35 ldnmos and 38 ldpmos rule outputs and the curated deck has none of them, so a clean report says nothing about coverage.ECAD · Agent failure · 1 report74
  5. 5The synthesis check re-hashes RTL and liberty but records no hash of the netlist it produced, so it cannot catch a committed gate-level netlist that was hand-edited or went stale after resynthesis. It also needs the PDK on the checking host, so signoff CI skips it.ECAD · Agent failure · 1 report74
  6. 6Upstream klayout-tools issue behind sky130-trng #386, covering the extractor writing a device outside the model bins without warning.ECAD · Agent failure · 1 report74
  7. 7For chip layout extraction, the --pins option folds label names but ignores their order, so the subcircuit header follows layout label order. A layout labeled out, en, vss, vdd against a schematic header of en, out, vdd, vss trips pin_order_mismatch and the extracted leg never runs. The workaround is a hand-written wrapper subcircuit in the testbench.ECAD · Agent failure · 1 report66
  8. 8klt extract on a 25k-device block ran past a 3600 s timeout twice with no output at all; a caller can't tell slow from hung or pick a sensible time bound, and the issue asks for progress on stderr or a hash-keyed unchanged shortcut.Verification · Agent failure · 1 report66
  9. 9No built-in device voltage-rating check across corners: users hand-write a B-source per terminal pair, a .meas per pair, keep the limit table outside the tool, and re-aggregate per corner in a wrapper script, even though the limits already sit in the PDK model cards.Verification · Agent failure · 1 report66
  10. 10Signoff only accepts a SPICE run as evidence. A spec row checked by an analytical model in Python "reads as uncovered although its evidence exists, and nothing in the manifest ties the record to the row." The bad workaround would be wrapping the model in a dummy SPICE netlist "which would pass without testing anything."Verification · Agent failure · 1 report66
  11. 11test_ldmos_extracts_with_stable_source_and_drain asserts unsupported_devices == [] but gets a non-empty ldmos entry on current main; the device class is present but the extractor still marks it unsupported, in the same gf180 LDMOS thread as the missing-rules story.ECAD · Agent failure · 1 report66
  12. 12The ASAP7 LVS deck only knows FinFET devices, so any schematic with a resistor or capacitor mismatches its layout every time, and analog and I/O cells on ASAP7 cannot reach a full LVS match.ECAD · Agent failure · 1 report66
  13. 13The curated sg13g2 rule nwell.enclosing.pactiv.1 applies to every P+ Activ shape, while IHP's own deck first removes shapes in a recognized SVaricap device. On a layout with only SVaricap instances, the curated deck reports 64 violations, all in SVaricap, while IHP's primary runset and its maximal deck report 0. This answers the sg13g2-vco question, where an integrator saw 64 violations in an unmodified block and could not tell if they were real.ECAD · Agent failure · 1 report66
  14. 14The klt extract and klt pex commands write diodes with A= and P=, which ngspice's D element does not accept, so every delta row with a diode dies with a fatal error and no extracted vs schematic comparison runs for any diode. Reproduced on two klt versions against the gf180mcu PDK.ECAD · Agent failure · 1 report66
  15. 15The seal ring check re-implements the IHP deck's rules and is labelled published_rule, but "no run of the real deck was compared against klt ring-check." The test kit only asserts "broken", so rule geometry isn't cross-checked.Verification · Agent failure · 1 report66
  16. 16The sg13cmos5l deck ignores the PDK runset's SRAM marker scoping, so it reports thousands of NWell findings, 100% of them inside the SRAM layer, where the foundry runset reports 0.ECAD · Agent failure · 1 report66
  17. 17OpenROAD runs through Docker and fleet workers get permission denied, while the tool only says the toolchain is missing, and the reference netlist exists only as a place-and-route output, so the LVS report cannot be regenerated.ECAD · Agent failure · 1 report58

How it fails

  1. Agent failure17

Engineering tools that address these errors

Sourced from public community reports. Costs and delays on individual entries are modeled estimates, not measurements.