DFT · Chapter 6 · Coverage & Signoff — chapter closer
Coverage Closure & Signoff Targets
This closer turns a coverage number into a signed-off quality commitment. The signoff target is derived, not arbitrary: the DPPM goal set by the market and the process defect density work backward to a required coverage target, and there is one target per fault model, from stuck-at to at-speed to memory. Closure is the chapter's loop, run until every target is met: measure coverage, read it critically, debug the loss, improve testability, recover aborts, and re-run. Signoff is then a formal gate with a checklist covering coverage at or above target per model, clean design rules, intact chains, pattern count within the test-time budget, an honest denominator, and a proven-untestable residual that is accepted and insured. Even at target a residual remains, so coverage closure is a commitment, not a guarantee.
Intermediate13 min readDFTCoverage ClosureSignoffDPPMTest Targets
Chapter 6 · Section 6.5 · Coverage & Signoff — chapter closer
Project thread — the FSM (closed via its observe point, 6.4) is signed off here against a DPPM-derived target, and its pattern set hands to compression (Chapter 7) to cut test time.
1. Why Should I Learn This?
Signoff is where DFT becomes a quality commitment — and the target and the gate must both be right.
- The target is derived from DPPM + defect density, per model (1.5/2.1) — not arbitrary.
- Closure = the Ch6 loop: measure → read critically → debug → improve testability → recover aborts → re-run.
- Signoff is a formal gate: coverage ≥ target (per model), DRC clean, chains intact, patterns in budget, honest denominator, residual insured.
- Coverage closure is a commitment, not a guarantee — a residual always remains (insure it).
2. Real Silicon Story — 'sign off at 98%, we're out of time'
Near tapeout, a block sat at 98% test coverage and the schedule was screaming. The proposal: 'sign off at 98% — we're out of time.' But this part was going into an automotive program whose DPPM contract implied a required coverage of ~99.9% (1.5). 98% wasn't a rounding error below target — it was a different quality tier.
Signing off at 98% would have shipped escapes the automotive DPPM explicitly forbade — a contractual and safety breach, discovered in the field. Instead, the team closed the gap: a few SCOAP-guided test points (6.4) and recovered aborts (5.5) took stuck-at to target, and they added at-speed (2.3) for the timing tier. The genuine residual (proven-untestable) was accepted and insured with SLT/burn-in (1.3).
Lesson: the signoff target is derived from the market's DPPM, so you don't lower it because you're out of time — you close the gap, or renegotiate the market/insurance explicitly. Silently signing off below the DPPM-derived target ships escapes.
3. Factory Perspective — closure & signoff through each lens
- What the test engineer sees: the signoff checklist — coverage per model vs target, DRC/chain-integrity, pattern count vs test-time budget — the go/no-go gate.
- What the yield engineer sees: that coverage at target bounds the escape rate statistically (1.5), and the residual is the insured tail (SLT/burn-in).
- What the RTL/DV engineer sees: that any shortfall points back to testability (6.3/6.4) — closing it is their lever, and a clean block makes signoff routine.
- What management cares about: that signoff = the DPPM commitment — a contract, so signing off below the derived target is a quality/legal liability, not a schedule shortcut.
4. Concept — deriving the target, closing, and the gate
The target is derived (not arbitrary):
- Work the DPPM chain backward (1.5): market DPPM goal + defect density → required coverage target.
- Per model (2.1): a stuck-at target, an at-speed/transition target (2.3), a memory target (Chapter 8) — a portfolio of targets, each sized to the defect class and DPPM.
- Automotive/medical → high targets (zero-defect mindset); consumer → lower.
Closure (the Chapter 6 loop):
- Measure test coverage (6.1) per model.
- Read critically (6.2) — honest denominator, right model, DT-vs-PT.
- Debug the loss (6.3) — localize, diagnose the half, find the cause.
- Improve testability (6.4) — test points, restructure, wrap, remove redundancy.
- Recover aborts (5.5) with effort.
- Re-run → repeat until test coverage ≥ target, per model.
Signoff (a formal gate — the checklist):
- Test coverage ≥ target (per model, honest denominator — 6.1).
- Scan DRC clean (4.3) and chain integrity passing (3.3).
- Pattern count within the test-time budget (1.4).
- Residual (proven-untestable) accepted and insured — SLT/burn-in (1.3) per DPPM (1.5).
- Patterns written (STIL/WGL) for the ATE (1.3).
The residual — commitment, not guarantee:
- Even at target, two things remain: proven-untestable faults (accepted) and unmodeled defect classes (2.1).
- So coverage closure is a commitment (a statistical bound on escapes) not a guarantee — 100% of the target model ≠ 0 DPPM (1.5) → insure the tail for zero-defect markets.
The signoff gate is a stack of criteria — all must hold:
5. Mental Model — a pre-flight sign-off, to a required standard
Coverage signoff is like an aircraft's airworthiness sign-off — to a required standard set by the route, not by the clock.
- The standard is derived from the mission (the DPPM goal): a short hop (consumer) and a transatlantic flight (automotive) require different safety margins — you don't pick the standard arbitrarily.
- Closure is doing the maintenance until the aircraft meets the standard — you fix the discrepancies (testability, 6.4), you don't wave them through.
- Sign-off is a checklist gate — all items (coverage, structure, budget, honest records) must pass; a single high number isn't sign-off any more than 'the engines look fine' is airworthiness.
- 'We're out of time, sign it at 98%' is 'skip the inspection, we're behind schedule' — for a transatlantic flight (automotive DPPM), that's how you crash in the field.
- And even a signed-off aircraft isn't crash-proof — you carry insurance and redundancy (SLT/burn-in) for the residual risk.
Sign off to the standard the mission demands — close the gap or renegotiate the mission, but never quietly lower the bar.
6. Working Example — deriving a target and passing the gate
Derive the target, close, and check the gate:
# Deriving the signoff target — REPRESENTATIVE, from the DPPM chain (1.5):
Market = automotive -> DPPM goal ~ single digits (zero-defect mindset)
Defect density = ~1% -> required STUCK-AT test coverage ~ 99.9% ; + AT-SPEED target (2.3) ; + MBIST (Ch8)
Consumer contrast : DPPM higher -> stuck-at target could be lower (~99%) ; fewer models
# The target is DERIVED per model, not picked. Automotive != consumer.# Signoff checklist — REPRESENTATIVE, SIMPLIFIED, tool-neutral:
[x] Test coverage (stuck-at) = 99.92% >= 99.9% target (per model; honest denominator, 6.1)
[x] Test coverage (at-speed) = target met (2.3) (timing defect class covered)
[x] Scan DRC = CLEAN (4.3)
[x] Chain integrity = PASS (3.3)
[x] Pattern count = within test-time budget (1.4) -> compression next (Ch7)
[x] Denominator = honest (only proven-untestable removed; no masking inflation, 6.1)
[x] Residual (proven-untestable)= accepted + INSURED (SLT/burn-in, 1.3) per DPPM (1.5)
[x] Patterns written (STIL) -> ATE (1.3)
=> SIGNOFF: PASS. This is a COMMITMENT tied to DPPM -- not a guarantee (residual insured).7. Industry Flow — signoff is the gate to production test
Signoff gates the pattern set into production test and sets up compression:
8. Debugging Session — pressure to sign off below the target
Near tapeout a block is below the DPPM-derived coverage target and there is pressure to sign off anyway to hit the schedule; but the target is derived from the market DPPM, so signing off below it ships escapes the contract forbids -- the fix is to close the gap (test points, more models, recover aborts) or explicitly renegotiate the market/insurance, never to silently lower the bar
THE TARGET IS DPPM-DERIVED — DON'T SIGN OFF BELOW IT SILENTLYNear tapeout, a block is below its coverage target (e.g. 98% vs a 99.9% automotive target). Schedule pressure produces the proposal: 'sign off at 98% — we're out of time.'
The coverage target is derived from the market's DPPM goal, so it is not a soft aspiration you can lower for schedule — signing off below it ships exactly the escapes the DPPM contract forbids. The target (99.9% here) was worked backward from the automotive DPPM commitment and defect density (1.5): at that defect density, only ~99.9% test coverage keeps the escape rate within the contracted DPPM. So 98% is not '1.9% short of a nice-to-have' — it corresponds to a materially higher escape rate, a different quality tier than the part was sold as. 'Signing off at 98%' doesn't make the escapes disappear; it just stops measuring them — the defects still ship, now undetected, and surface as field returns, a contractual and (for safety parts) legal breach. The schedule pressure is real, but it argues for closing the gap or changing the commitment, not for redefining 98% as passing — the coverage number is only a quality commitment if the target reflects the DPPM, and silently lowering the bar severs that link while keeping the appearance of signoff.
Close the gap to the derived target, or explicitly renegotiate the market/insurance — never silently lower the bar. First, close it: apply the Chapter 6 loop hard — recover aborts with effort (5.5), add SCOAP-guided test points (6.4) at the flat-tail nodes (a few points often close a 1–2% gap cheaply), debug any dominant loss region (6.3), and add the models the DPPM demands (at-speed 2.3, memory Chapter 8). If the gap genuinely can't close in time, make the trade explicit and owned: renegotiate (ship into a lower-DPPM market for this revision, or descope), or add insurance (SLT/burn-in, 1.3) sufficient to bring the effective DPPM within contract — a documented, signed decision, not a quiet edit. Either way, keep the denominator honest (6.1) — don't 'reach target' by masking or removing unproven faults. The principle to lock in: the signoff target is derived from the market's DPPM goal and defect density (per model), coverage closure is the disciplined loop of measuring, debugging, and improving testability until test coverage meets that target, and signoff is a formal gate — so a block below target is closed (test points, more models, recovered aborts) or its market/insurance is explicitly renegotiated, but it is never signed off below the DPPM-derived target silently, because doing so ships the very escapes the target exists to prevent. (Testability fixes are 6.4; the honest denominator is 6.1; insurance/DPPM is 1.3/1.5; compression to recover test time is Chapter 7.)
9. Common Mistakes
- Signing off below the DPPM-derived target for schedule. The target is a commitment — close it or renegotiate explicitly.
- Treating the target as arbitrary. It's derived from DPPM + defect density, per model (1.5/2.1).
- Reaching target via an inflated denominator. Masking/unproven removals (6.1) is cheating, not closure.
- Forgetting the residual. Even at target, insure the proven-untestable + unmodeled tail (1.3/1.5).
- Signing off on coverage alone. The gate also needs DRC clean, chains intact, patterns in budget (4.3/3.3/1.4).
10. Industry Best Practices
- Derive targets from DPPM + defect density, per model (1.5) — automotive ≠ consumer.
- Close with the Ch6 loop — testability (6.4) and recovered aborts (5.5), not brute-force patterns.
- Gate on the full checklist — coverage per model, DRC, chains, pattern budget, honest denominator.
- Accept + insure the residual — SLT/burn-in for zero-defect markets (1.3/1.5).
- Never lower the bar silently — renegotiate market/insurance explicitly if you can't close.
11. Senior Engineer Thinking
- Beginner: "We're at 98% and out of time — just sign off."
- Senior: "The target is 99.9% because that's what the automotive DPPM demands — 98% is a different quality tier that ships escapes. I close the gap — recover aborts, add test points, add at-speed — or, if I truly can't, I renegotiate the market/insurance explicitly. I never sign off below the derived target silently, and I never reach it by inflating the denominator."
The senior treats the target as a DPPM commitment and the gate as formal — close it honestly or change the commitment openly.
12. Silicon Impact
Coverage closure and signoff are where all the DFT work of Chapters 2–5 becomes the part's quality contract — the moment a coverage number turns into a DPPM commitment (1.5). The two disciplines that make this real are deriving the target and honoring the gate. The target is not arbitrary: it's worked backward from the market's DPPM goal and the process defect density, per model — a stuck-at target, an at-speed target (2.3), a memory target (Chapter 8) — so automotive and consumer are different tiers, and a number like 98% vs 99.9% is not a rounding difference but a quality-class difference. Closure is the disciplined Chapter 6 loop — measure, read critically, debug, improve testability, recover aborts, re-run — that reaches the target honestly (via testability, 6.4, not an inflated denominator, 6.1). And signoff is a formal gate, not a high number: it requires coverage ≥ target per model, clean DRC and chains (4.3/3.3), pattern count within the test-time budget (1.4), an honest denominator, and an accepted, insured residual (SLT/burn-in, 1.3). The most important — and most tempting-to-violate — principle is that you do not sign off below the DPPM-derived target for schedule; the escapes don't vanish when you stop measuring them, so you close the gap or renegotiate the market/insurance explicitly. Finally, the humility the whole curriculum has built toward: coverage closure is a commitment, not a guarantee — a residual of proven-untestable and unmodeled defects always remains, which is why zero-defect markets insure it. For the RTL/DV engineer, signoff is the scoreboard of everything upstream: testable, scan-ready design (Ch4) and testability fixes (6.4) are what let a block meet its DPPM-derived target cleanly — and, with coverage now closed and signed off, Chapter 7 (compression) takes over to cut the test time of the very pattern set you committed to, closing the loop on quality and cost.
13. Engineering Checklist
- Derived the coverage target from DPPM + defect density, per model (1.5/2.1).
- Closed to target via the Ch6 loop (testability 6.4, recovered aborts 5.5) — honest denominator (6.1).
- Passed the full signoff gate: coverage per model, DRC clean (4.3), chains intact (3.3), patterns in budget (1.4).
- Accepted + insured the residual (SLT/burn-in, 1.3) per DPPM (1.5).
- Did not sign off below target silently — closed the gap or renegotiated explicitly.
14. Try Yourself
- Derive a stuck-at coverage target for an automotive (single-digit DPPM) vs a consumer part at 1% defect density (1.5).
- Run the closure loop on a block at 98%: recover aborts (5.5), add test points (6.4) → reach target.
- Build a signoff checklist and mark each item (coverage/model, DRC, chains, budget, denominator, residual).
- Argue why signing off 98% vs a 99.9% target ships escapes (not a rounding difference).
- Explain why coverage closure is a commitment, not a guarantee, and what you insure (1.3/1.5).
The targets/checklist are tool-neutral. Real coverage/DRC come from the DFT flow; DPPM from the quality plan (1.5). No paid tool required.
15. Interview Perspective
- Weak: "Signoff means hitting the coverage target."
- Good: "You close coverage to a target and check DRC and chains before signing off."
- Senior: "Signoff turns coverage into a DPPM commitment. The target is derived — market DPPM goal + defect density → required coverage, per model (stuck-at + at-speed + memory), so automotive ≠ consumer. Closure is the Chapter 6 loop — measure, read critically, debug loss, improve testability, recover aborts, re-run — until test coverage ≥ target with an honest denominator. Signoff is a formal gate: coverage per model, DRC clean, chain integrity, pattern count in the test-time budget, and the residual accepted and insured (SLT/burn-in). And it's a commitment, not a guarantee — a residual of proven-untestable and unmodeled defects remains, so I insure it. Critically, I never sign off below the DPPM-derived target for schedule — I close the gap or renegotiate the market/insurance explicitly."
16. Interview / Review Questions
17. Key Takeaways
- The signoff target is derived, not arbitrary: the market's DPPM goal + defect density work backward to a required coverage target, and it's per model (stuck-at + at-speed 2.3 + memory Ch8) — automotive/medical demand higher targets than consumer (1.5/2.1).
- Coverage closure is the Chapter 6 loop: measure (6.1) → read critically (6.2) → debug loss (6.3) → improve testability (6.4) → recover aborts (5.5) → re-run, until test coverage ≥ target, per model, with an honest denominator.
- Signoff is a formal gate with a checklist: coverage ≥ target (per model), scan DRC clean (4.3), chain integrity (3.3), pattern count within the test-time budget (1.4), honest denominator (6.1), and the residual accepted + insured (SLT/burn-in, 1.3).
- Coverage closure is a commitment, not a guarantee — even at target, a residual of proven-untestable faults and unmodeled defect classes remains (2.1), so zero-defect markets insure it (1.5).
- Never sign off below the DPPM-derived target silently (or via an inflated denominator) — close the gap or renegotiate market/insurance explicitly; this is where Chapters 2–5's DFT effort becomes the part's quality commitment, and it hands the signed-off pattern set to Chapter 7 (compression) to cut test time. Next: Chapter 7 — Test Compression (cutting the test time of the signed-off patterns).
18. Quick Revision
Coverage closure & signoff (Ch6 closer). TARGET is DERIVED (not arbitrary): market DPPM goal + defect density → required coverage, per model (stuck-at + at-speed 2.3 + memory Ch8); automotive ≫ consumer (1.5). CLOSURE = the Ch6 loop: measure (6.1) → read critically (6.2) → debug loss (6.3) → improve testability (6.4) → recover aborts (5.5) → re-run, until TEST coverage ≥ target per model (honest denominator, 6.1). SIGNOFF = a formal GATE: coverage/model + DRC clean (4.3) + chain integrity (3.3) + pattern count in budget (1.4) + honest denominator + residual accepted & INSURED (SLT/burn-in 1.3). A commitment, NOT a guarantee (residual of proven-untestable + unmodeled always remains → insure, 1.5). Never sign off below the DPPM-derived target silently — close the gap or renegotiate explicitly. Hands off to Ch7 (compression) to cut test time. Next: Chapter 7 — Test Compression.