Netlist Semantic Cleanup
Use this skill when converting generated netlists into readable, reviewable
Spectre/SPICE reference decks.
Principle
The primary cleanup engine is semantic understanding by the model and engineer.
Do not treat netlist cleanup as a blind text rewrite. A checker can report
leftover tool artifacts, but it cannot decide circuit boundaries, node meaning,
or instance names.
Workflow
- Preserve the raw generated artifact. Never edit ADE/PEX output in place.
- Read enough context to understand the circuit: ports, hierarchy, stimulus,
measurements, clocking, biasing, and intended DUT boundary.
- Read
references/cleaning.md before curating a real design.
- Draft the semantic cleanup plan:
- DUT subckt or wrapper interface.
- Testbench-owned supplies, clocks, stimulus, loads, probes, and analyses.
- Run-deck-owned parameters, model includes, corners, sweeps, and saves.
- Rename map for random nodes/instances, plus unresolved review notes.
- Create the clean netlist as a curated source artifact. Use model reasoning to
split DUT/testbench/run decks and to choose semantic names.
- Run the checker only after the semantic pass:
python skills/netlist/scripts/check_spectre_netlist.py netlist/dut/block.scs --mode dut
python skills/netlist/scripts/check_spectre_netlist.py netlist/tb/tb_block.scs --mode tb
The checker reports suspicious MOS tail parameters, random-looking node names,
generic instance names, and DUT/testbench mixing. It does not rewrite files and
should not be used as the main cleanup mechanism.
Rules
- Preserve MOS instance parameters by default when numerical behavior matters.
Removing layout side-effect tails is an explicit stripped-reference choice,
not the default.
- If a stripped reference is requested, it may remove
ad/as, pd/ps,
nrd/nrs, sa/sb/sca/scb, sp*, DFM, stress, proximity, and extraction
parameters, but expect performance differences and validate against the
full-parameter or raw deck.
- Split reusable artifacts into DUT, testbench, and run decks. A clean DUT file
should not contain supplies, clocks, sweeps, saves, probes, or analysis
statements.
- At the top of every curated netlist file or section containing MOS devices,
add an explicit terminal-order comment, for example:
// MOS terminal order: D G S B (drain gate source bulk/body).
- Replace random names (
net1, _net23, N_*, numeric mesh nodes, I42,
M0) with semantic names where the circuit meaning is known. If meaning is
unknown, leave a short review note instead of guessing.
- Validate after cleanup with a parser or Spectre syntax/smoke run when
Cadence is available.
1---2name: netlist3description: Semantic cleanup and curation of Spectre/SPICE transistor netlists for analog review or reusable reference circuits. Use when Codex needs to understand generated ADE/PEX netlists, split DUT/testbench/run decks, optionally strip layout-derived MOS tail parameters by reasoning, rename random nodes/instances into semantic names, or prepare clean comparator/opamp/SAR reference netlists. Scripts in this skill are checkers only, not primary cleanup engines.4---56# Netlist Semantic Cleanup78Use this skill when converting generated netlists into readable, reviewable9Spectre/SPICE reference decks.1011## Principle1213The primary cleanup engine is semantic understanding by the model and engineer.14Do not treat netlist cleanup as a blind text rewrite. A checker can report15leftover tool artifacts, but it cannot decide circuit boundaries, node meaning,16or instance names.1718## Workflow19201. Preserve the raw generated artifact. Never edit ADE/PEX output in place.212. Read enough context to understand the circuit: ports, hierarchy, stimulus,22 measurements, clocking, biasing, and intended DUT boundary.233. Read `references/cleaning.md` before curating a real design.244. Draft the semantic cleanup plan:25 - DUT subckt or wrapper interface.26 - Testbench-owned supplies, clocks, stimulus, loads, probes, and analyses.27 - Run-deck-owned parameters, model includes, corners, sweeps, and saves.28 - Rename map for random nodes/instances, plus unresolved review notes.295. Create the clean netlist as a curated source artifact. Use model reasoning to30 split DUT/testbench/run decks and to choose semantic names.316. Run the checker only after the semantic pass:3233```bash34python skills/netlist/scripts/check_spectre_netlist.py netlist/dut/block.scs --mode dut35python skills/netlist/scripts/check_spectre_netlist.py netlist/tb/tb_block.scs --mode tb36```3738The checker reports suspicious MOS tail parameters, random-looking node names,39generic instance names, and DUT/testbench mixing. It does not rewrite files and40should not be used as the main cleanup mechanism.4142## Rules4344- Preserve MOS instance parameters by default when numerical behavior matters.45 Removing layout side-effect tails is an explicit stripped-reference choice,46 not the default.47- If a stripped reference is requested, it may remove `ad/as`, `pd/ps`,48 `nrd/nrs`, `sa/sb/sca/scb`, `sp*`, DFM, stress, proximity, and extraction49 parameters, but expect performance differences and validate against the50 full-parameter or raw deck.51- Split reusable artifacts into DUT, testbench, and run decks. A clean DUT file52 should not contain supplies, clocks, sweeps, saves, probes, or analysis53 statements.54- At the top of every curated netlist file or section containing MOS devices,55 add an explicit terminal-order comment, for example:56 `// MOS terminal order: D G S B (drain gate source bulk/body)`.57- Replace random names (`net1`, `_net23`, `N_*`, numeric mesh nodes, `I42`,58 `M0`) with semantic names where the circuit meaning is known. If meaning is59 unknown, leave a short review note instead of guessing.60- Validate after cleanup with a parser or Spectre syntax/smoke run when61 Cadence is available.
arcadia-1