signoff-loop

signoff-loop Skill

Execute a staged, artifact-first open-source EDA flow from specification to GDSII with full signoff checks using OpenROAD-flow-scripts (ORFS), and close the loop by learning from every run (the two memory DBs + engineer_loop) so DRC/LVS violations are eliminated and timing closes at the best Fmax. Prefer deterministic scripts for execution, keeping the agent focused on planning, generation, diagnosis, and iteration.

Dataset construction lives in the companion def-graph skill. This skill produces the clean, signed-off 6_final.def/.odb/.spef; converting those into PyG graph datasets (the five views b–f, tech-lib/LEF parser, feature + label extraction) is def-graph's job — see its SKILL.md.

Environment Setup

Every flow script sources scripts/flow/_env.sh on entry, which autodetects ORFS + tool paths and lets the user override any single value. You do not need to source anything manually.

Resolution order (first hit wins, per value)

1. Variable already set in the caller's environment — ORFS_ROOT=... run_orfs.sh ... wins unconditionally.
2. User env file — path in $R2G_ENV_FILE (if set).
3. In-skill override file — references/env.local.sh (copy from references/env.local.sh.template).
4. ORFS-provided env — $ORFS_ROOT/env.sh (once ORFS_ROOT is known).
5. System-wide env — /opt/openroad_tools_env.sh (if present).
6. Autodetect — command -v <tool> on $PATH, then a list of well-known install paths (e.g. $ORFS_ROOT/tools/install/OpenROAD/bin/openroad, $HOME/oss-cad-suite/bin/yosys, /usr/local/bin/klayout).

Checking what the skill found

bash scripts/flow/check_env.sh

Prints the resolved ORFS_ROOT, every tool binary it picked, and the platforms it can see. Exits non-zero if a required tool is missing.

Overriding just a few values

# One-off override for a single run
ORFS_ROOT=/opt/ORFS OPENROAD_EXE=/opt/openroad/bin/openroad \
  bash scripts/flow/run_orfs.sh design_cases/my_design nangate45

# Or persist overrides in a file
cp references/env.local.sh.template references/env.local.sh
# ...then edit the exports you care about; every subsequent flow picks them up.

Available platforms

nangate45, sky130hd, sky130hs, asap7, gf180, ihp-sg13g2 (default: asap7).

Workflow

1. Normalize the Specification First

• Convert free-form requirements into a structured specification before writing RTL.
• Read references/spec-template.md and produce input/normalized-spec.yaml.
• If clock/reset, IO, target flow, or timing targets are missing, stop and ask the user or record explicit assumptions.

2. Initialize a Project Directory

• Create a run folder under design_cases/<design-name>/ using scripts/project/init_project.py.
• The layout follows references/workflow.md.
• Directories created: input/, rtl/, tb/, constraints/, lint/, sim/, synth/, backend/, drc/, lvs/, rcx/, reports/.

3. Generate RTL and Testbench Separately

• Write RTL to rtl/design.v.
• Write testbench to tb/testbench.v.
• Keep assumptions and design notes in reports/rtl-notes.md.

4. Run Validation in Strict Order

1. Run scripts/project/validate_config.py <project-dir> before ORFS backend to catch config/RTL issues early.
2. Run lint/syntax checks before simulation.
3. Run simulation before synthesis.
4. Run synthesis before backend (ORFS).
5. Do not skip failed stages unless the user explicitly requests it.

5. Run Backend with ORFS

• Prepare constraints/config.mk and constraints/constraint.sdc.
• Use scripts/flow/run_orfs.sh to invoke the ORFS Makefile.
• ORFS runs place-and-route natively (no Docker required).
• Collect results from the ORFS results directory.

5b. Check Timing Before Signoff (Tiered WNS + TNS)

After ORFS completes, extract PPA and run the timing gate:

1. Run scripts/extract/extract_ppa.py <project-dir> reports/ppa.json to extract timing metrics.
2. Run scripts/reports/check_timing.py <project-dir> to classify WNS and TNS and write reports/timing_check.json.
3. The script independently classifies WNS and TNS, then takes the worse of the two as the combined tier. A design with small WNS but large TNS (many slightly-violating paths) is caught.
4. Read reports/timing_check.json and act on the tier:

Tier	Criteria	Agent Action
clean	WNS >= 0, TNS >= 0	Proceed to signoff.
minor	WNS >= -2.0 AND TNS >= -10.0	Auto-fix: update clk_period in constraint.sdc to suggested_clock_period from the JSON, then re-run backend. Report the fix to the user after the fact.
moderate	WNS >= -5.0 AND TNS >= -100.0 (but not clean/minor)	Stop. Present the numbered options from the JSON to the user. Wait for their choice.
severe	WNS < -5.0 OR TNS < -100.0	Stop. Present options with strong warning.
unconstrained	WNS > 1e+30	Stop. SDC clock port mismatch. Present options. Do NOT proceed.

5. The JSON includes wns_tier and tns_tier fields so the agent can explain which metric triggered the tier (e.g., "TNS escalated this from minor to moderate").
6. Only proceed to signoff checks (step 6) after timing is resolved.

5a. (Optional) Fmax search — find the fastest closing period

Before committing to a clock period, you can characterize the design's Fmax:

python3 scripts/reports/fmax_search.py <project-dir> [platform] [--verify]

Loose-first search using cheap placement-stage timing (each probe runs only ORFS_STAGES="synth floorplan place"). It reports a predicted-signoff Fmax (reports/fmax_search.json), corrected by a learned per-family slack-deterioration model. The number is a proxy (UNVERIFIED) — post-place timing is optimistic vs signoff. Pass --verify to confirm the winner with one full flow (and feed the result back to tighten the model). This does NOT replace the step-8 check_timing gate, which still runs on the final backend.

Knobs: --probe-timeout, --place-fast (whole-search conservative lower bound for hang-prone designs), --keep-variants. The search is sequential; cross-design parallelism is achieved by running multiple invocations concurrently.

6. Run Signoff Checks (DRC, LVS, RCX)

After a successful backend run, run signoff checks in order:

DRC (Design Rule Check)

Two tool options are available:

1. KLayout DRC (default) — scripts/flow/run_drc.sh <project-dir> [platform]

   • Uses ORFS make drc target with platform .lydrc rules
   • Outputs: drc/6_drc.lyrdb, drc/6_drc_count.rpt, drc/6_drc.log

2. Magic DRC (sky130 only) — scripts/flow/run_magic_drc.sh <project-dir> [platform]

   • Uses Magic's built-in DRC engine with sky130A tech file
   • Requires the sky130A PDK; the script reads $PDK_ROOT/sky130A/libs.tech/magic/sky130A.tech (set PDK_ROOT via references/env.local.sh — /opt/pdks is only the fallback default).
   • Outputs: drc/magic_drc.rpt, drc/magic_drc_count.rpt, drc/magic_drc_result.json
   • Supported platforms: sky130hd, sky130hs

LVS (Layout vs Schematic)

Two tool options are available:

1. KLayout LVS (default) — scripts/flow/run_lvs.sh <project-dir> [platform]

   • Uses ORFS make lvs target with platform .lylvs rules + CDL netlist
   • Gracefully skips platforms without LVS rules (produces lvs/lvs_result.json with status "skipped")
   • Outputs: lvs/6_lvs.lvsdb, lvs/6_lvs.log, lvs/6_final.cdl
   • nangate45: uses adapted FreePDK45 rules with connect_implicit("VDD"/"VSS") for bulk merging and schematic.purge for unused cell pins (e.g., QN on DFFR_X1)
   • Large design warning: KLayout LVS on designs >100K cells (black_parrot, swerv) takes >60 minutes. Use LVS_TIMEOUT=7200 for these designs. The default 3600s may not be enough.

2. Netgen LVS (sky130 only) — scripts/flow/run_netgen_lvs.sh <project-dir> [platform]

   • Two-step flow: Magic extracts SPICE from GDS, then Netgen compares against Verilog netlist
   • Requires the sky130A PDK (Magic tech + $PDK_ROOT/sky130A/libs.tech/netgen/sky130A_setup.tcl). Set PDK_ROOT via references/env.local.sh; /opt/pdks is only the fallback default.
   • Outputs: lvs/extracted.spice, lvs/netgen_lvs.rpt, lvs/netgen_lvs_result.json
   • Supported platforms: sky130hd, sky130hs
   • This is the production sky130 LVS path — prefer it over KLayout LVS on sky130 (the ORFS KLayout sky130 rule deck is not production-grade; see references/failure-patterns.md, "sky130 LVS").
   • Antenna-diode designs are handled automatically: the script normalizes Magic's diode X-subcircuit instances to D devices (perim=→pj=) and runs netgen with MAGIC_EXT_USE_GDS=1, so sky130_fd_sc_hd__diode_2 matches instead of flattening.
   • Designs with port-to-port feedthroughs (assign out_port = in_port) need export POST_GLOBAL_PLACE_TCL = <skill>/scripts/flow/orfs_hooks/buffer_port_feedthroughs.tcl in config.mk before the backend run — SPICE cannot express two top-level ports on one net, so without the hook LVS fails "Top level cell failed pin matching". The hook is a no-op for designs without feedthroughs (safe to set everywhere); a backend re-run is required when adding it. See references/failure-patterns.md, "sky130 LVS" cause 5.

RCX (Parasitic Extraction)

3. RCX — scripts/flow/run_rcx.sh <project-dir> [platform]
   • OpenRCX parasitic extraction via OpenROAD
   • Generates Tcl script (rcx/run_rcx.tcl) with define_process_corner, extract_parasitics, write_spef
   • Reads 6_final.odb from ORFS results, writes SPEF output
   • Outputs: rcx/6_final.spef, rcx/rcx.log, rcx/run_rcx.tcl

Extract results into JSON for reporting and dashboard:

• scripts/extract/extract_drc.py <project-root> reports/drc.json
• scripts/extract/extract_lvs.py <project-root> reports/lvs.json
• scripts/extract/extract_rcx.py <project-root> reports/rcx.json

The complete strict evidence bundle is mandatory (round-2 pilot P0-2, failure-patterns #53): every signoff pass must also leave reports/route.json (extract_route.py), reports/ppa.json (extract_ppa.py), reports/timing_check.json (check_timing.py — the FINAL timing confirmation that upgrades an Fmax winner from proxy to qualified), and the binding reports/signoff_manifest.json (scripts/reports/build_signoff_manifest.py: per-report sha256s, SDC digest + stamped period vs Fmax winner, confirming run + DEF/GDS digests, strict_clean verdict with an ENUMERATED strict_missing). fix_signoff.sh and tools/run_signoff.sh emit all of this automatically at the end of every run; strict callers use build_signoff_manifest.py --strict, R2G_SIGNOFF_STRICT_EXIT=1 (run_signoff), or R2G_BATCH_STRICT_EXIT=1 (batch_flow) to turn a non-clean aggregate into a nonzero exit. Before a strict campaign, verify the platform can even satisfy it: scripts/flow/platform_capability.py --platform <p> --strict (also printed by check_env.sh; R2G_STRICT_PLATFORMS makes it required there).

• If DRC/LVS is fail, attempt automated real-layout fixes: scripts/flow/fix_signoff.sh <project-dir> [platform] [--check drc|lvs|both] (See references/signoff-fixing.md.)
• If the backend aborted at route (congestion / DRT timeout, exit 124 — orfs_status=fail, orfs_fail_stage=route), relieve it BEFORE signoff: scripts/flow/fix_signoff.sh <project-dir> sky130hd --check route (lowers CORE_UTILIZATION so DRT converges; learnable + A/B-validated route_relief). See references/failure-patterns.md "Routing Congestion".

Fix-Learning Loop

The skill learns from every fix attempt so candidate strategies are proposed in evidence-ranked order on the next similar violation.

• Record. fix_signoff.sh and check_timing.py --journal append lossless, session-keyed rows to reports/fix_log.jsonl (one per iteration: strategy, before/after counts, pre-fix violation class, verdict). fix_signoff.sh uses an adaptive budget (base 3 iters, hard cap 8, early-stop after 2 non-improving iters past the base).
• Ingest. Step-10 ingest (knowledge/ingest_run.py) reads fix_log.jsonl into the Tier-1 fix_events table and writes a run_violations snapshot for every run — clean or not (the full violation landscape). It then auto-runs fix_log_manager.manage() (toggle R2G_FIX_AUTOLEARN, default on).
• Learn. learn_heuristics.py derives Tier-2 fix_trajectories (per-episode path, including abandoned episodes and failed strategies — negative learning) and folds them into Tier-3 fix_recipes inside heuristics.json.
• Apply. When a recipe exists for the design's family/platform/violation class, diagnose_signoff_fix.py reorders the strategy list by empirical clearance — there is no hard gate, all real-fix strategies are always proposed, priority-ordered. diagnose_signoff_fix.py <proj> --check drc --list prints the evidence-ranked candidate set as JSON. Hard safety clamps are unchanged.
• Symptom index. Learned repair experience is keyed by a symptom signature (knowledge/symptom.py: {check, class, predicates} → a stable symptom_id), NOT the design-family name. learn_heuristics.py emits a top-level symptoms[symptom_id] projection in heuristics.json (pooled across families/platforms, with by_platform + evidence_designs provenance); diagnose_signoff_fix.py looks recipes up by symptom and seeds an informed cross-platform prior for untried strategies (so a fix learned on nangate45 transfers to e.g. sky130hd). It also surfaces the matching active prose lesson (via search_failures.lessons_for_symptom) at the fix-decision point. observe.py health (degradation alerts) and analyze_execution.py (fix-proposal triage) are operator-invoked CLIs over the same store.

See references/signoff-fixing.md ("Fix-Learning Loop") and knowledge/README.md.

Engineer Loop (campaign mode)

Use campaign mode when you need to run the full flow unattended across many designs — or when you want the A/B-gated recipe-learning cycle to run autonomously. The campaign orchestrator (scripts/loop/engineer_loop.py) drives the flow scripts, ingests results, triggers learning, and manages A/B trials without human gates.

# Add a project to the campaign ledger
python3 scripts/loop/engineer_loop.py add \
    --ledger design_cases/_batch/campaign.jsonl \
    --project design_cases/my_design [--platform nangate45]

# Run the campaign (optionally limit to N designs)
python3 scripts/loop/engineer_loop.py run \
    --ledger design_cases/_batch/campaign.jsonl [--max N]

# Inspect per-design state
python3 scripts/loop/engineer_loop.py status \
    --ledger design_cases/_batch/campaign.jsonl

The ledger is JSONL (last-state-wins); kill/restart is safe — the campaign resumes where it left off. States: pending → flow → signoff → fixing → clean | escalated | abandoned.

Hard rules for campaign mode:

• Phase-1 runs workers=1 (single-process); do not run two campaigns sharing a DESIGN_NAME concurrently.
• Never run two configs with the same DESIGN_NAME + FLOW_VARIANT concurrently.
• Never run more than one LVS job concurrently for designs > 100 K cells.
• Only promoted recipes affect live strategy ranking; shadow and candidate recipes are inert until their A/B trial completes.

When the loop opens an escalation (unknown symptom, exhausted catalog, unseen crash, or repeated regression), drain it following the agent runbook in references/engineer-loop.md ("Escalation Drain"). That document also covers provenance queries (observe.py trace) and the full safety-invariant list.

Platform Support Matrix

Platform	KLayout DRC	KLayout LVS	Magic DRC	Netgen LVS	RCX
nangate45	Yes	Yes	No	No	Yes
sky130hd	Yes	Yes	Yes	Yes	Yes
sky130hs	Yes¹	Yes	Yes	Yes²	Yes
asap7	Yes	No	No	No	Yes
gf180	Yes	Yes	No	No	Yes
ihp-sg13g2	Yes	Yes	No	No	Yes

¹ sky130hs has no ORFS-shipped DRC deck; run_drc.sh deliberately reuses the sibling sky130hd.lydrc (pure sky130A tech-layer rules, no hd-specific content) via KLAYOUT_DRC_FILE= on the make command line (failure-patterns.md #32). ² Requires the sky130hs.lyt lefdef repair (tools/patch_sky130hs_lyt.py, applied by eda-install's platform-rules step) — the stock file makes def2stream drop ALL DEF geometry, turning every Netgen LVS into a false top-pin mismatch; run_netgen_lvs.sh guards portless extractions as infra errors (failure-patterns.md #33).

7. Treat Artifacts as Source of Truth

• Save logs, reports, VCD waveforms, netlists, SPEF, configurations, and summary files.
• Prefer file outputs over GUI tools. GUI viewers like GTKWave/KLayout are optional helpers.

8. Diagnose Before Editing

• For failures, read references/failure-patterns.md.
• Classify the failure: specification gap, RTL bug, testbench bug, synthesis issue, backend/configuration issue, DRC violation, LVS mismatch, or RCX extraction error.
• Fix the smallest plausible cause first.

9. Summarize Each Stage Clearly

• State pass/fail status.
• List key artifact paths.
• Record assumptions, blockers, and next recommended actions.
• For signoff: report DRC violation count, LVS match/skip status, RCX net count and total capacitance.

10. Ingest the Run into the Knowledge Store

After every flow — successful, failed, or partial — run:

python3 knowledge/ingest_run.py design_cases/<project>

This reads the structured JSON artifacts produced by the extraction scripts and appends one row to knowledge/knowledge.sqlite. It never parses raw ORFS logs.

Then rebuild derived artifacts:

python3 knowledge/learn_heuristics.py
python3 knowledge/mine_rules.py

• knowledge/heuristics.json is consumed automatically by suggest_config.py on the next project — no CLI changes required.
• knowledge/failure_candidates.json is a review queue, not a rule source. Surface new signatures to the user and, if confirmed, edit references/failure-patterns.md by hand.

A family/platform pair appears in heuristics.json only after at least 3 successful runs under that configuration.

10b. Share / Transfer the Knowledge Store Across Users (git-friendly)

knowledge.sqlite is the tracked, committed store — a fresh clone is pre-trained immediately. A binary SQLite blob cannot be combined across operators (git only 3-way-merges text; two operators' campaigns would conflict and one would clobber the other), so when you need to share or merge experience across operators, knowledge/knowledge_sync.py is an on-demand tool: it exports a deterministic, git-friendly text bundle (knowledge/store/, one NDJSON file per table — regenerable, not committed by default) and performs a real honesty-gated cross-operator union:

# After learning, re-export the committed text bundle (keeps it in sync with the DB):
python3 knowledge/knowledge_sync.py export

# A NEW user folds another operator's experience into their local store. The merge is
# ADDITIVE (dedups by natural content key — run_id/symptom_id are portable; surrogate
# ids are re-assigned) and is REFUSED+rolled back if it would break an honesty gate:
python3 knowledge/knowledge_sync.py merge --bundle path/to/their/store
python3 knowledge/knowledge_sync.py merge --from-db path/to/their/knowledge.sqlite

# Bootstrap a fresh store from a bundle only (rebuilds knowledge.sqlite):
python3 knowledge/knowledge_sync.py import --bundle knowledge/store --db knowledge/knowledge.sqlite

# Honesty CI gate (the real gate — runs the 5 honesty gates over the committed store):
python3 knowledge/honesty.py --db knowledge/knowledge.sqlite
# On-demand drift check for an EXPORTED bundle you intend to share (reports
# "no committed bundle" by design when none exists — NOT a CI gate post-revert):
python3 knowledge/knowledge_sync.py status

Commit workflow: knowledge.sqlite is the committed store — commit it (and heuristics.json) after ingest/learn, as before. The knowledge/store/ bundle is gitignored and only produced on demand (export) when you want to hand experience to another operator or review a diff; status confirms an exported bundle matches the DB. After any merge, run learn() + engineer_loop ab-drain so imported recipes re-validate locally. See knowledge/README.md ("Sharing the store across users").

Hard Rules

• Do not start backend if simulation is failing.
• Do not start ORFS if synthesis failed or the top module is unclear.
• Do not start signoff checks (DRC/LVS/RCX) if backend did not produce a GDS/ODB.
• Run check_timing.py after every backend run. It checks both WNS and TNS. For minor violations (WNS >= -2.0 AND TNS >= -10.0), auto-fix by increasing clock period and re-running. For moderate/severe/unconstrained, stop and present numbered fix options — do not proceed without the user's decision.
• Do not silently invent missing interfaces, clocks, resets, or timing targets without documenting assumptions.
• Prefer single-clock MVP flows. Macro designs (fakeram45) are supported with proper config (see "Macro / Hard Memory Designs"). Escalate to the user before attempting CDC, multi-clock, or DFT.
• Use the scripts in scripts/ for repeatable operations instead of re-inventing shell commands each time.
• Do not hand-source any system env file before running EDA tools — every flow script sources scripts/flow/_env.sh, which autodetects ORFS and tool paths (see "Environment Setup"). /opt/openroad_tools_env.sh is only one optional source in that chain and may be absent.
• When a batch produces a mix of pass/fail, diagnose with references/failure-patterns.md (see "Batch-Campaign Failure Patterns") and apply the repo-level batch fixer before any code changes. That fixer (tools/fix_orfs_failures.py in the agent-r2g repository — not shipped with the installed skill) handles the six dominant failure modes (memory inference, IO-pin perimeter overflow, place density >1, PDN straps, missing include dirs, stage timeouts) by rewriting config.mk. When running standalone, apply the same patterns by hand using references/failure-patterns.md. Do not hand-edit configs case-by-case in batch — extend the fix tool so future batches self-heal.
• Floorplan sizing policy (validated on 495-design batch):
  • Explicit DIE_AREA is only safe when pin count ≤ ~200 and RTL fits in the area. Prefer CORE_UTILIZATION when in doubt.
  • When PPL-0024 reports a required perimeter, derive DIE_AREA = 0 0 S S with S = ceil((required_perim / 4) * 1.3) rounded up to 10um.
  • For designs with memory inference, set SYNTH_MEMORY_MAX_BITS = 131072 (default 4096 is too tight for register files and FIFOs).

Default Project Layout

design_cases/<design-name>/
├── input/
│   ├── raw-spec.md
│   └── normalized-spec.yaml
├── rtl/
│   └── design.v
├── tb/
│   └── testbench.v
├── constraints/
│   ├── config.mk
│   └── constraint.sdc
├── lint/
│   └── lint.log
├── sim/
│   ├── sim.log
│   └── output.vcd
├── synth/
│   ├── synth.ys
│   ├── synth.log
│   └── synth_output.v
├── backend/
│   └── RUN_<timestamp>/
│       ├── final/              # GDS, DEF, ODB
│       ├── logs/               # Per-stage logs
│       ├── reports/            # Timing, area, power
│       ├── drc/                # DRC results (copied)
│       ├── lvs/                # LVS results (copied)
│       └── rcx/                # RCX results (copied)
├── drc/
│   ├── 6_drc.lyrdb            # KLayout DRC violation database (XML)
│   ├── 6_drc_count.rpt        # Violation count
│   ├── 6_drc.log              # DRC log
│   └── drc_run.log            # Full make output
├── lvs/
│   ├── 6_lvs.lvsdb            # KLayout LVS comparison database (XML)
│   ├── 6_lvs.log              # LVS log
│   ├── 6_final.cdl            # CDL netlist
│   ├── lvs_run.log            # Full make output
│   └── lvs_result.json        # Only if skipped (no rules)
├── rcx/
│   ├── 6_final.spef           # SPEF parasitic data
│   ├── rcx.log                # OpenRCX extraction log
│   └── run_rcx.tcl            # Generated Tcl extraction script
├── reports/
│   ├── ppa.json               # PPA metrics + geometry
│   ├── progress.json          # ORFS stage completion
│   ├── run-history.json       # Multi-run comparison
│   ├── run-compare.json       # Baseline vs current delta
│   ├── diagnosis.json         # Issue detection & suggestions
│   ├── drc.json               # DRC summary (violations, categories)
│   ├── lvs.json               # LVS summary (match/mismatch/skipped)
│   ├── rcx.json               # RCX summary (net count, cap, res)
│   └── demo-summary.md        # Human-readable summary
└── metadata.json

Resource Map

• Read references/spec-template.md when the specification is incomplete or ambiguous.
• Read references/workflow.md when you need the phase-by-phase execution order.
• Read references/orfs-playbook.md before setting up or debugging the ORFS backend.
• Read references/failure-patterns.md when a run fails and you need a triage path.
• Read references/ppa-report-guide.md when summarizing synthesis/backend reports.
• To build the physical-design graph dataset (per-node/edge features, per-cell/net labels, the five PyG views b–f, and the tech-lib/LEF parser) from the signed-off DEF/LEF/SPEF this flow produces, switch to the companion def-graph skill.
• Use scripts in scripts/ for initialization, spec normalization, environment checks, lint, simulation, synthesis, ORFS backend, DRC, LVS, RCX extraction, result collection, GDS preview rendering, dashboard generation, and run summaries.
• Use scripts/flow/orfs_hooks/ for Tcl sourced into ORFS stages via the generic PRE_<STAGE>_TCL / POST_<STAGE>_TCL env hooks (set them in config.mk, not the shell — ORFS scrubs exported variables). Currently: buffer_port_feedthroughs.tcl (POST_GLOBAL_PLACE_TCL) splits port-to-port feedthrough nets behind real buffers so Netgen LVS top-level pins match; a no-op for designs without feedthroughs.
• Use assets/examples/simple-arbiter/ as the first smoke-test case.
• Use assets/config-template.mk and assets/constraint-template.sdc as default backend configuration templates.

Quick Start

Prerequisites

Required Tools (all installed on this machine):

• python3 (3.13+)
• yosys (synthesis)
• iverilog + vvp (simulation)
• openroad (place & route, OpenRCX)

Optional Tools (also available):

• verilator (faster lint/simulation)
• klayout (GDS visualization, DRC, LVS)
• magic (DRC, SPICE extraction for sky130)
• netgen-lvs (LVS comparison for sky130)
• gtkwave (waveform viewing)
• sta / opensta (static timing analysis)

Running a Full Flow

1. No manual env setup needed — flow scripts autodetect ORFS/tools via scripts/flow/_env.sh. Run scripts/flow/check_env.sh to confirm what was found.

2. Initialize a run directory with scripts/project/init_project.py <design-name>.

3. Save user requirements to input/raw-spec.md.

4. Normalize them into input/normalized-spec.yaml using scripts/project/normalize_spec.py.

5. Write or copy rtl/design.v and tb/testbench.v.

6. Run scripts/flow/check_env.sh to verify tool availability.

7. Run scripts/flow/run_lint.sh, then scripts/flow/run_sim.sh, then scripts/flow/run_synth.sh.

8. After those pass, prepare constraints/config.mk and constraints/constraint.sdc.

9. Run scripts/flow/run_orfs.sh <project-dir> for the backend.

10. Extract PPA: scripts/extract/extract_ppa.py <project-dir> reports/ppa.json

11. Run timing gate: scripts/reports/check_timing.py <project-dir> — reads reports/timing_check.json:

    • tier=clean: proceed to step 12.
    • tier=minor: auto-fix clock period per suggested_clock_period, re-run step 9, then re-check.
    • tier=moderate/severe/unconstrained: stop, present options to user, wait for decision.
    • Check wns_tier and tns_tier to explain which metric drove the tier.

12. Run signoff checks (only after timing gate passes or user approves):

    • scripts/flow/run_drc.sh <project-dir> [platform] (KLayout DRC)
    • scripts/flow/run_magic_drc.sh <project-dir> [platform] (Magic DRC, sky130 only)
    • scripts/flow/run_lvs.sh <project-dir> [platform] (KLayout LVS)
    • scripts/flow/run_netgen_lvs.sh <project-dir> [platform] (Netgen LVS, sky130 only)
    • scripts/flow/run_rcx.sh <project-dir> [platform]

13. Extract remaining results:

    • scripts/extract/extract_drc.py <project-root> reports/drc.json
    • scripts/extract/extract_lvs.py <project-root> reports/lvs.json
    • scripts/extract/extract_rcx.py <project-root> reports/rcx.json
    • If DRC/LVS status is fail, attempt real layout fixes (antenna diode insertion + repair iters, route effort, density relief; LVS triage): scripts/flow/fix_signoff.sh <project-dir> [platform] [--check drc|lvs|both] [--max-iters 3] Real-fixes-only — never relaxes the rule deck. Residual stuck/timeout/KLayout-crash cases are reported honestly. See references/signoff-fixing.md. 13b. (Dataset building — optional) To convert the signed-off 6_final.def/.odb/.spef into PyG graph datasets — per-cell/per-net labels (congestion, wirelength, timing, IR drop), per-node/edge features, and the five graph views b–f — switch to the companion def-graph skill (scripts/flow/run_labels.sh / run_features.sh / run_graphs.sh there). It reads the same backend artifacts this flow produces.

14. Diagnose issues: scripts/reports/build_diagnosis.py <project-root> reports/diagnosis.json

15. Get config suggestions: knowledge/suggest_config.py <project-dir> (optional, useful for tuning)

16. Generate the dashboard with scripts/dashboard/generate_multi_project_dashboard.py. The index page carries two read-only knowledge-store panels at the top (above the project grid), computed by scripts/reports/build_lineage_view.py:

    • Knowledge Store Health — total runs, ORFS status distribution, % partial/unknown, learnable family/platform pairs (≥3 successes via knowledge_db.is_success), signoff positives, and whether heuristics.json is populated (the EMPTY case renders red — the screaming diagnostic that learning is inert).
    • Config Tuning Provenance — the config_lineage diff chain per design/platform (changed/added/removed config keys + orfs/drc/lvs outcome delta prev→cur).

    INVARIANT: these panels are a strictly descriptive, READ-ONLY projection over knowledge.sqlite / config_lineage / heuristics.json (opened mode=ro; the projection writes only JSON). They are NEVER wired into suggest_config as an auto-tuner. The config-variant lineage is a loose single-parent diff chain, not a true DAG.

17. Serve it with scripts/dashboard/serve_multi_project_dashboard.py 8765.

MVP Scope

Default to an MVP flow that supports:

• Single module or small design
• Single clock domain
• Simple reset behavior
• Generated or hand-authored Verilog RTL
• Testbench-driven simulation
• Yosys synthesis
• ORFS backend run with nangate45 platform
• DRC signoff check
• LVS signoff check (where platform supports it)
• OpenRCX parasitic extraction (SPEF)
• Report collection, signoff summary, and dashboard

Escalate to the user before attempting CDC, multi-clock constraints, DFT, or signoff-quality closure.

Macro / Hard Memory Designs (Validated)

Macro designs using fakeram45 on nangate45 are supported and validated (riscv32i, tinyRocket, swerv, bp_multi_top — all produce GDS + pass RCX). LVS passes for designs <150K cells; designs >150K cells may need extended LVS timeout.

Some designs instantiate hard memory macros (fakeram45 on nangate45, SRAM on sky130). These require extra config:

1. Verilog blackbox stubs — Provide module definitions for all macros referenced in the RTL. For designs using SYNTH_HIERARCHICAL=1, stubs must be actual module implementations (not (* blackbox *) attributes), because Yosys CELLMATCH pass needs cost info. Wrap the fakeram macros inside the BSG-style wrapper modules with real port connections.

2. ADDITIONAL_LEFS / ADDITIONAL_LIBS — Point to the platform's LEF and LIB files for each macro type.

3. CDL_FILE for LVS — The platform config.mk sets a default CDL_FILE that only includes standard cells. Macro designs need a combined CDL with both standard cells and fakeram subcircuit definitions. Use override export CDL_FILE (the override keyword is critical — without it, the platform config.mk, which is included after the design config, will silently overwrite your CDL_FILE).

4. MACRO_PLACEMENT_TCL — Provide a Tcl script for macro placement. Use find_macros (not all_macros) to check for macros, and call global_placement before macro_placement:

   if {[find_macros] != ""} {
     global_placement -density [place_density_with_lb_addon] -pad_left 2 -pad_right 2
     macro_placement -halo {10 10} -style corner_max_wl
   }
   

5. GDS_ALLOW_EMPTY — Set export GDS_ALLOW_EMPTY = fakeram.* so ORFS allows empty GDS cells for the macro stubs.

6. Safety flags — Large macro designs need SKIP_CTS_REPAIR_TIMING=1 and SKIP_LAST_GASP=1 to avoid OpenROAD crashes.

Behavioral SRAM Stubs (Alternative to Macro Mapping)

When the design's SRAM macros are undefined in the RTL set (e.g., Chipyard BOOM's freepdk45_sram_*, generic foundry stubs without LEF/LIB) AND the total memory bits are modest (<~256K bits), substitute behavioral flop-array implementations instead of mapping to fakeram45.

Use tools/gen_openram_behavioral_stubs.py <wrapper.v> <stubs.v> (a repo-level helper in the agent-r2g repository — not bundled with the installed skill) to auto-generate behavioral Verilog for every freepdk45_sram_<ports>_<rows>x<cols>[_<gran>] referenced. The generator handles both 1rw0r (single-port) and 1w1r (write-port + read-port, independent clocks) styles, with optional write-mask granularity.

Why behavioral instead of fakeram45:

• fakeram45 is single-port only — cannot represent BOOM's 1w1r macros.
• fakeram45 widths (32/39/64/etc.) don't match arbitrary BOOM widths (40, 44, 52, 56, 124) without padding waste.
• Behavioral stubs let Yosys infer memories cleanly; ORFS handles them as plain logic with no macro placement needed.

When NOT to use behavioral stubs:

• Total memory bits > ~50K — Yosys's memory_map pass turns each read port into a wide mux tree. At BOOM SmallSEBoom's 168K total bits, the post-mapping cell count exceeds ~1M gates and ABC's speed script grinds beyond the 4 h ORFS_TIMEOUT. The earlier guidance (512K) was too optimistic; lowered after a 2h28m ABC failure.
• Designs that need real silicon (taping out) — use real macros.
• The skill currently caps SYNTH_MEMORY_MAX_BITS per memory at 65536. Single memories larger than that should use real macros.

When behavioral stubs hit the ABC ceiling but you still want to avoid real macros, try SYNTH_HIERARCHICAL=1 before switching to fakeram45. With hierarchical synth, ABC is invoked separately per Yosys module — each freepdk45_sram_* stub becomes its own small ABC run (≤32K gates per memory) instead of one giant 1M-gate ABC pass.

Validated:

• DMA-class designs (Faraday DMA, ff_ram flop array): pass behavioral.
• AES / ibex / CRC / iscas89 (no SRAM macros, total memory ≤ ~10K bits): pass.
• Faraday RISC (3 unique tsyncram_* sizes, 87K total memory bits, 79 RTL files, dual SYSCLK/BUSCLK): synth passes in 213 s under SYNTH_HIERARCHICAL=1 + ABC_AREA=1. Yosys peak 553 MB; ABC ran per kept module (5 ABC invocations, 57 s combined). Confirms hierarchical mode lifts the behavioral ceiling well past the 50K-bit flat-mode limit when single memories stay <16K bits each.
• BOOM SmallSEBoom (17 SRAM types, 168K total memory bits, ~360K-line top): flat-mode ABC fails (2h28m wall, 5.36 GB peak). Retry with SYNTH_HIERARCHICAL=1 is the recommended next step; if that still fails, fall back to mapping the four 1rw0r macros (61K bits total) to fakeram45 and keep the thirteen 1w1r macros behavioral.
• See docs/faraday_viability.md for the per-design SRAM scale audit.

ORFS Backend Details

config.mk Format

export DESIGN_NAME = <top_module>
export PLATFORM    = nangate45

export VERILOG_FILES = <absolute_path_to_rtl>
export SDC_FILE      = <absolute_path_to_sdc>

export CORE_UTILIZATION = 30
export PLACE_DENSITY_LB_ADDON = 0.20

For macro designs, add (note the override on CDL_FILE):

export ADDITIONAL_LEFS = $(PLATFORM_DIR)/lef/fakeram45_64x32.lef
export ADDITIONAL_LIBS = $(PLATFORM_DIR)/lib/fakeram45_64x32.lib
export GDS_ALLOW_EMPTY = fakeram.*
export MACRO_PLACEMENT_TCL = /absolute/path/to/macro_placement.tcl
override export CDL_FILE = /absolute/path/to/combined_with_fakerams.cdl

The override keyword on CDL_FILE is essential: ORFS includes the platform config.mk after the design config.mk, so a bare export CDL_FILE gets silently overwritten by the platform default (which only has standard cells).

For sky130 (and any design whose RTL contains port-to-port assign out = in feedthroughs), also add the stage hook that keeps Netgen LVS top-level pins matchable:

# Split port-to-port feedthrough nets so Netgen LVS top-level pins match
export POST_GLOBAL_PLACE_TCL = <skill>/scripts/flow/orfs_hooks/buffer_port_feedthroughs.tcl

Set it in config.mk (not the shell — ORFS scrubs exported variables). It is a no-op for designs without feedthroughs, so it is safe to set unconditionally.

config.mk Validation Rules (Hard)

Before running ORFS, validate every config.mk against these rules:

1. PLACE_DENSITY_LB_ADDON ≥ 0.10 — Values below 0.10 cause placement divergence (NesterovSolve stuck). Use 0.20+ for macro designs.
2. Bus-heavy designs need CORE_UTILIZATION ≤ 15% — Crossbars, interconnect fabrics, and bus arbiters have high routing demand.
3. Large macro designs need safety flags — Designs with >50K instances or SRAM macros (swerv, black_parrot, ibex) must include:

   export SKIP_CTS_REPAIR_TIMING = 1
   export SKIP_LAST_GASP = 1
   

   Without these, OpenROAD may SIGSEGV during CTS timing repair.
4. All VERILOG_FILES must be absolute paths and point to existing files.
5. DESIGN_NAME must exactly match the RTL top module name.

constraint.sdc Format

current_design <top_module>
set clk_name  core_clock
set clk_port_name clk
set clk_period 10.0
set clk_io_pct 0.2

set clk_port [get_ports $clk_port_name]
create_clock -name $clk_name -period $clk_period $clk_port

set non_clock_inputs [all_inputs -no_clocks]
set_input_delay  [expr $clk_period * $clk_io_pct] -clock $clk_name $non_clock_inputs
set_output_delay [expr $clk_period * $clk_io_pct] -clock $clk_name [all_outputs]

Running ORFS

The scripts/flow/run_orfs.sh script:

1. Copies RTL and constraints to an ORFS-compatible design directory
2. Derives a unique FLOW_VARIANT from the project directory name (prevents collisions)
3. Runs make DESIGN_CONFIG=<config.mk> FLOW_VARIANT=<variant> with optional timeout
4. Collects results back to the project directory

Resource control via environment variables:

ORFS_TIMEOUT=7200    # Per-stage max runtime in seconds (default: 2 hours)
ORFS_MAX_CPUS=4      # Limit CPU cores via taskset (default: all)
PLACE_FAST=1         # Disable GPL_TIMING_DRIVEN/ROUTABILITY_DRIVEN — use for
                     # BOOM-class designs (>1M nets) where the timing-repair
                     # loop in gpl spins for hours after Nesterov has already
                     # converged. CTS/route still apply timing closure.
ROUTE_FAST=1         # Cap GRT iterations + skip optional GRT-stage repair
                     # passes. Sets SKIP_INCREMENTAL_REPAIR=1,
                     # SKIP_ANTENNA_REPAIR=1, DETAILED_ROUTE_END_ITERATION=10,
                     # GLOBAL_ROUTE_ARGS='-congestion_iterations 5
                     # -allow_congestion -verbose ...'. Use when GRT extra
                     # iterations cycle 1→30 forever on >1M-net designs
                     # (BOOM ChipTop). Result: GDS may have congestion/DRC
                     # but produces final ODB.
ROUTE_FAST_SKIP_DRT=1  # Combined with ROUTE_FAST: also SKIP_DETAILED_ROUTE=1.
                       # Last-resort fallback — produces DEF + GRT solution
                       # but no GDS. Use only when ROUTE_FAST still hangs in
                       # detailed routing.
FROM_STAGE=place     # Resume from a specific stage (synth|floorplan|place|cts|route|finish)

Shell Script Safety Rules (Validated Against 70 Designs)

These rules are enforced in all 6 execution scripts (run_orfs.sh, run_lvs.sh, run_rcx.sh, run_drc.sh, run_magic_drc.sh, run_netgen_lvs.sh):

1. PIPESTATUS capture — When piping timeout ... | tee, temporarily disable set -e and pipefail to capture the correct exit code:

   set +e +o pipefail
   setsid timeout ... make ... 2>&1 | tee logfile
   STATUS=${PIPESTATUS[0]}
   set -e -o pipefail
   

   Never use || true after a pipeline — it resets PIPESTATUS to 0 and masks all failures.

2. Environment isolation — Always unset SCRIPTS_DIR before calling ORFS make. ORFS Makefile uses this variable internally; an external value causes "No rule to make target synth.sh" errors.

3. Process group kills — Use setsid timeout (not bare timeout) so the entire process tree is killed on timeout. Without setsid, grandchild processes (klayout, openroad) survive as zombies, consuming memory and holding file locks.

4. LVS timeout scaling — Default LVS_TIMEOUT=3600 works for designs <100K cells. For swerv-class (~145K), use 4200s. For bp_multi_top-class (~200K), use 7200s. ORFS and RCX always complete within default timeouts.

Running Signoff After ORFS

After a successful ORFS run, the signoff scripts operate on the ORFS results in-place:

• run_drc.sh — runs make DESIGN_CONFIG=<config.mk> drc in ORFS flow directory (KLayout)
• run_magic_drc.sh — runs Magic DRC with sky130A tech file (sky130 only)
• run_lvs.sh — runs make DESIGN_CONFIG=<config.mk> lvs in ORFS flow directory (KLayout)
• run_netgen_lvs.sh — extracts SPICE via Magic, compares with Netgen (sky130 only)
• run_rcx.sh — runs OpenROAD directly with a generated Tcl script reading 6_final.odb

All scripts collect results back to the project's drc/, lvs/, rcx/ directories and also copy them into the latest backend/RUN_*/ subdirectory.