Signoff Regression Hygiene

What this topic teaches

Signoff Regression Hygiene is about converting constraints and path rules into measurable slack. Signoff quality depends on reproducible runs: same lib, SPEF, SDC, scenarios, and tool versions — tagged and archived. The hard part is never the textbook inequality; it is proving, under the right corner/mode/view, which layer — constraint, view, or implementation — broke the contract.

The senior-engineer question

When run-to-run WNS drift, database tag consistency, closure memo completeness moves, can you identify the failing path group, the STA layer, the responsible owner, and the smallest experiment that proves the root cause?

STA CLOSURE FLOW — Signoff Regression Hygiene

constraints (SDC)
        |
        v
timing graph (cell + net delays)
        |
        v
MMMC scenarios (corner x mode x view)
        |
        v
variation + SI adjustments
        |
        v
slack (WNS/TNS) -> ECO -> regression

Debug rule: always state corner, mode, view, and database tag with slack.

Picture the timing closure flow

Start every study session by drawing the behavior before reading signals. The diagrams below are the mental models to reproduce on a whiteboard.

Run manifest

STA RUN MANIFEST
design: chip_top tag v2.4.1
sdc:    constraints_v18.git
spef:   cworst_v3
libs:   ss_125 / ff_m40 / ...
tool:   PrimeTime Q-2020.03
scenarios: 36 (full matrix)

Timing path sequence

TIMING PATH — Signoff Regression Hygiene

launch FF --logic stages--> capture FF
     |                           |
  launch clock                capture clock
     |                           |
  required: arrival < capture_edge - setup
  metric: run-to-run WNS drift, database tag consistency, closure memo completeness

Who owns which layer

STA OWNERSHIP LAYERS — Signoff Regression Hygiene

layer              owns                         failure mode
----------------   --------------------------   ---------------------
constraints        clocks, IO, exceptions       false violations
implementation     cells, nets, placement       real path delay
extraction         SPEF/RC corners              view mismatch
signoff policy     derate, MMMC matrix          margin disputes
closure           ECO order, regression        fix breaks other corner

Evidence to collect

• Primary metric: run-to-run WNS drift, database tag consistency, closure memo completeness.

• Primary artifact: run manifest, git/DB tag log, signoff checklist.

• Owners to bring into review: STA lead, CAD owner, program management.

• Spec clause or requirement ID for every constraint claim.

• One failing report_timing excerpt and one reduced path that isolates the rule.

Ownership map

OWNERSHIP MAP — Signoff Regression Hygiene

artifact              owner
----------------      -----------------
timing report       STA lead
SDC/constraint      CAD owner
implementation      program management

Every WNS cluster needs a named owner before ECO.

Subpages in this topic

Each topic is taught across mechanism, inputs/outputs, reports, debug, worked example, pitfalls, interview, checklist, theory, design space, expanded case study, walkthrough, comparison matrix, software view, and silicon PPA impact.

Key takeaways

• State corner, mode, and view with every slack number.

• Separate constraint bugs, view mismatches, and real path delays before ECO.

• Run setup and hold plus full MMMC after every change.

Common pitfalls

• Comparing runs with different SPEF or SDC tags.

• Broad false_path to hide violations.

• Setup-only ECO without hold regression.

STA deep dive

Debug hygiene separates senior STA leads from report readers.

Concept diagram

DEBUG ORDER

1. same tags?  2. constraints?  3. view?  4. real path?  5. ECO

Metric graph

ROOT CAUSE MIX (typical)

constraint bugs  ████████  35%
view mismatch    ██████    25%
real path        ██████    25%
methodology      ████      15%

Reports and artifacts

• run manifest

• SDC diff

• SPEF tag

• scenario isolation

Mini case study

Mystery regression: SPEF corner tag typo in run script — not a design bug.

Debug branches

• Manifest every run

• War room by path group

Senior review question

Ask: what corner/mode/view proves this topic is closed or failing?

Key takeaways

• State corner, mode, view, and database tag with every slack claim.

• Run setup and hold plus MMMC regression after every ECO.

Common pitfalls

• Comparing STA runs with different SPEF or SDC tags.

• Broad false_path to green-wash violations.

• Setup-only ECO without hold check.