GLS Signoff Review Checklist

Chapter 15 · Section 15.4 · Interview & Signoff Review Preparation

Project thread — this is the deliverable the whole book was building toward: the concrete GLS signoff checklist. 15.5 is a readiness self-assessment against it.

1. Why Should I Learn This?

A signoff review needs a concrete checklist — not vibes — to say the netlist is clean.

• Nine items encode every bar from the book.
• It asserts what GLS proves and names the boundaries (STA/LEC/CDC).
• GLS is one gate — reviewed alongside the others.

This is the practical signoff deliverable (15.5 self-assesses against it).

2. Real Silicon Story — the signoff that skipped the checklist

A team called their GLS "clean" and signed off — informally, no checklist. Two items had been skipped: the SDF was never verified (silently zero-delay, 4.5) and a batch of Xs had been waived wholesale (8.4). A timing-dependent bug and a real X both slipped through.

A formal checklist review would have caught both: "SDF annotation verified — 0 unmatched?" (no) and "waivers justified and tracked, not wholesale?" (no). The checklist exists precisely to stop skipped items from becoming escaped bugs.

Lesson: GLS signoff needs a concrete checklist, run down in a review — informal "it's clean" skips items and ships bugs. The checklist encodes every bar and names every boundary.

3. Concept — the signoff checklist

The nine checklist items (the concrete gate):

1. SDF annotation verified — 0 unmatched (4.5) for timed runs (else silently zero-delay).
2. No unexplained X — every X explained or waived with justification (not no X ever, 12.6).
3. No real violations — firings triaged real-vs-artifact (8.4), artifacts fixed in the run, real ones fixed in the design; not disabled wholesale.
4. Corner coverage — setup at the slow corner, hold at the fast corner; models/SDF match (3.4/4.4).
5. X/reset signoff — every state flop reset, no power-up X escapes (Ch7/14.3).
6. Power/DFT/CDC clean (where applicable) — isolation/retention/sequencing (Ch10), scan chains/patterns/test-mode reset (Ch11), synchronized crossings + no reconvergence glitch (Ch9).
7. Tiered regression run — zero-delay nightly, timed at milestones, a clean full run (13.4).
8. Waivers justified and tracked — reason/owner/scope, reviewable; not wholesale (13.5).
9. Reviewed alongside LEC/STA/static CDC — GLS is one gate; boundaries named (13.4).

What the checklist does:

• Asserts what GLS proves (items 1–8) and names the boundaries (item 9): timing → STA, equivalence → LEC, CDC → static CDC/MTBF, coverage → RTL/UVM/ATPG.
• Run in a review meeting — each item checked with evidence (the coverage log, the waiver DB, the regression report).

Scope (accuracy):

• The checklist is the GLS gate — one of several (LEC/STA/static-CDC); GLS is the dynamic gate-level input (0.3/13.4).

4. Mental Model — a pre-flight checklist signed by the crew

The GLS signoff checklist is a pre-flight checklist the crew signs before the plane (chip) departs (tapes out).

• Pilots don't eyeball the plane and say "looks fine" — they run the printed list, item by item, with evidence (gauges, indicators).
• Each item is a known failure point (fuel, flaps, instruments) — like each GLS item (SDF, X, violations, corners, reset, power/DFT/CDC, waivers).
• And the pilot's checklist is one of several — maintenance (LEC), weight-and-balance (STA), weather/ATC (CDC) all sign too. No single checklist clears the flight.
• Skipping an item (unverified SDF, wholesale-waived X) is how avoidable failures happen.

Run the printed list with evidence, sign it, and know it's one of the sign-offs — that's how the chip departs safely.

5. Working Example — the signoff checklist, filled

The checklist as a review artifact (tool-neutral):

# GLS SIGNOFF REVIEW CHECKLIST - REPRESENTATIVE (run in the review, evidence per item):
# [ ] 1. SDF annotation VERIFIED -- 0 unmatched (coverage log)                    [4.5]  evidence: annotation log
# [ ] 2. No UNEXPLAINED X -- every X explained/waived-with-justification           [12.6] evidence: X-report
# [ ] 3. No REAL violations -- firings TRIAGED (not disabled wholesale)            [8.4]  evidence: violation triage
# [ ] 4. Corner coverage -- setup=slow, hold=fast; models/SDF match                [3.4]  evidence: corner matrix
# [ ] 5. X/RESET signoff -- every state flop reset, no power-up X escapes          [7/14.3] evidence: reset report
# [ ] 6. Power/DFT/CDC clean (where applicable)                                     [10/11/9] evidence: per-domain reports
# [ ] 7. TIERED regression run -- nightly zero-delay / milestone timed / clean gate [13.4] evidence: regression report
# [ ] 8. Waivers JUSTIFIED + TRACKED (reason/owner/scope, not wholesale)            [13.5] evidence: waiver DB
# [ ] 9. Reviewed ALONGSIDE LEC / STA / static-CDC -- GLS is ONE gate               [13.4] evidence: signoff matrix
# ALL checked (with evidence) -> CLEAN-GLS SIGNOFF. Any unchecked -> NOT signed off.

Practical context (representative, tool-neutral):

# Running the signoff review (tool-neutral):
#   1) for EACH item, show the EVIDENCE (coverage log, X-report, triage, waiver DB, regression report)
#   2) any item without evidence = NOT signed off (no "looks clean")
#   3) name the BOUNDARIES (item 9): timing->STA, equivalence->LEC, CDC->static-CDC/MTBF, coverage->RTL/UVM/ATPG
#   4) GLS is ONE gate -> the review is joint with LEC/STA/static-CDC
#   -> all items checked with evidence = clean-GLS signoff

A clean-GLS-signoff state (all items evidenced), as a real waveform:

6. Debugging Session — a signoff that skipped checklist items

7. Common Mistakes

• Informal "it's clean" — no checklist, skips items.
• Not verifying SDF (item 1) — silently zero-delay (4.5).
• Wholesale-waived X (items 2/8) — masks real ones (8.4).
• Checking items without evidence — each needs a log/report.
• Treating GLS as the only gate (item 9) — it's one of LEC/STA/static CDC.

8. Industry Best Practices

• Run the checklist in a review — evidence per item, no "looks clean."
• Verify SDF, triage violations, justify waivers — the high-risk items.
• Assert X/reset and power/DFT/CDC clean where applicable.
• Confirm the tiered regression ran (13.4).
• Review alongside LEC/STA/static CDC — GLS is one gate; name the boundaries.

Senior Engineer Thinking

• Beginner: "The GLS looks clean, so we're signed off."
• Senior: "Show me the checklist with evidence: SDF verified (0 unmatched)? every X explained (not wholesale-waived)? violations triaged? corners? reset? tiered run? And this is one gate — where are LEC/STA/static CDC? No evidence, no signoff."

The senior runs the full checklist with evidence and treats GLS as one gate among several.

Silicon Impact

The signoff checklist is the deliverable that prevents escaped bugs — it converts the whole book's bars into a reviewable, evidence-based gate. The two most dangerous ways a signoff fails are exactly the ones informal review misses: an unverified SDF (item 1) that makes a "timing" run silently zero-delay (4.5), and wholesale-waived Xs (items 2/8) that silence a real bug (8.4) — either letting a bug reach silicon (0.3) behind a "clean" label. The checklist forces evidence per item, so a skipped step is a visible unchecked box rather than an invisible escape, and item 9 keeps GLS in its lane — one gate reviewed alongside LEC/STA/static CDC, with the boundaries named. This is the practical culmination of the course: not "understand GLS," but "produce the signoff artifact that says the netlist is clean," defensibly, in a review.

Engineering Checklist

• SDF verified (0 unmatched, evidence: coverage log) — item 1.
• No unexplained X; no real violations (triaged, not disabled) — items 2–3.
• Corners, X/reset, power/DFT/CDC clean where applicable — items 4–6.
• Tiered regression run; waivers justified/tracked — items 7–8.
• Reviewed alongside LEC/STA/static CDC (GLS is one gate) — item 9; evidence for every item.

Try Yourself

1. Take a GLS run and fill the nine-item checklist — provide evidence for each (coverage log, X-report, triage, waiver DB, regression report).
2. Observe: any item you can't evidence is a gap — an unchecked box, not a "probably fine."
3. Change: close each gap (verify SDF, triage Xs, justify waivers) until all nine are evidenced.
4. Expect: a defensible clean-GLS signoff — run alongside LEC/STA/static CDC. That's the deliverable the whole book built toward.

The signoff checklist is a methodology artifact, tool-independent. No paid tool required.

Interview Perspective

• Weak: "We sign off when the GLS looks clean."
• Good: "We run a checklist — SDF verified, no unexplained X, no real violations, reset clean."
• Senior: "GLS signoff is a nine-item checklist run in a review with evidence per item: SDF verified (0 unmatched), no unexplained X, no real violations (triaged not disabled), corner coverage, X/reset signoff, power/DFT/CDC clean, tiered regression, waivers justified/tracked, and reviewed alongside LEC/STA/static CDC — GLS is one gate. Any item without evidence isn't signed off. It asserts what GLS proves and names the boundaries."

9. Interview / Review Questions

10. Key Takeaways

• The GLS signoff review checklist is the concrete deliverable — the artifact that says a netlist is GLS-clean, run down in a review with evidence per item.
• Nine items encode every bar: (1) SDF verified (0 unmatched, 4.5); (2) no unexplained X (12.6); (3) no real violations (triaged, not disabled, 8.4); (4) corner coverage (3.4); (5) X/reset signoff (7/14.3); (6) power/DFT/CDC clean (10/11/9); (7) tiered regression (13.4); (8) waivers justified/tracked (13.5); (9) reviewed alongside LEC/STA/static CDC (13.4).
• It asserts what GLS proves and names the boundaries — timing → STA, equivalence → LEC, CDC → static CDC/MTBF, coverage → RTL/UVM/ATPG.
• Any item without evidence = not signed off — informal "it's clean" skips items and ships bugs (an unverified SDF or wholesale-waived X are the classic escapes).
• GLS is one gate among several — the review is joint with LEC/STA/static CDC; GLS is the dynamic gate-level input (0.3/13.4). Next: 15.5 — GLS readiness self-assessment.

Quick Revision

GLS signoff = a 9-item CHECKLIST run in a review, EVIDENCE per item: (1) SDF verified 0-unmatched (4.5); (2) no unexplained X (12.6); (3) no real violations (triaged, not disabled, 8.4); (4) corners (3.4); (5) X/reset signoff (7/14.3); (6) power/DFT/CDC clean (10/11/9); (7) tiered regression (13.4); (8) waivers justified/tracked (13.5); (9) alongside LEC/STA/static-CDC (GLS = one gate, 13.4). No evidence = not signed off. "Looks clean" skips items and ships bugs. Next: 15.5 — readiness self-assessment.