Interviewer is testing

Flow ownership.

Bad answer

"Run the tool and waive the rest."

Better answer

"Capture clocks, resets and intent; run structural analysis; classify; fix or waive with evidence; dynamic checks; sign off on final RTL."

Strong answer (a 4)

Adds hierarchy (block versus flat), re-runs after every fix, and names each stage's output.

Follow-ups

Hierarchical versus flat trade-offs; what restarts the loop.

Common trap

Signoff on stale RTL.

Evidence to cite

The closure package, not the tool's exit code.

Taught in

The Generic CDC Flow

Interviewer is testing

Concreteness.

Bad answer

"It finds all CDC bugs."

Better answer

"Every launch-capture pair across clock nets, each with its scheme legality: synchroniser presence, no logic between flops, single-bit-only, stable data, no reconvergence."

Strong answer (a 4)

Adds completeness dependence on the clock list and what structure cannot prove (protocol under all timings).

Follow-ups

Violation classes; a category you have debugged.

Common trap

"Clean report means safe".

Evidence to cite

The classified violation list.

Taught in

Violation Classes

Interviewer is testing

Method combination.

Bad answer

"Formal is just thorough simulation."

Better answer

"Proof over all alignments and input sequences: no missed or duplicated events, no torn values, for the modelled protocol."

Strong answer (a 4)

States both boundaries (no analogue settling model; abstraction must be reviewed) and places all three methods (structure, formal, simulation) as covering each other's gaps.

Follow-ups

What the formal model abstracts; liveness versus safety.

Common trap

"Formal proves the silicon".

Evidence to cite

Proven properties plus the reviewed abstraction.

Taught in

Formal CDC Checks

Interviewer is testing

Constraints literacy.

Bad answer

"False-path everything CDC."

Better answer

"Separate the clocks into groups where there is truly no timing relationship, and bound the datapath delay where the protocol needs it."

Strong answer (a 4)

Explains why blanket false paths are dangerous (they silence real paths, including convergences and resets) and names the review: every exception justified, bounded-delay limits where skew matters.

Follow-ups

Clock groups versus false paths; max-delay on data versus control.

Common trap

The one-line "set_false_path" answer.

Evidence to cite

Reviewed constraint set plus STA coverage of the bounded paths.

Taught in

Clock Groups and False Paths

Interviewer is testing

Risk discipline.

Bad answer

"When schedule demands it."

Better answer

"When the finding is proven safe or impossible by evidence: owner, condition, expiry, residual risk."

Strong answer (a 4)

Gives a worked shape (quasi-static with stability proof, scoped, expiring on RTL change) and names two anti-patterns (urgency as justification, waiving by designer assertion).

Follow-ups

Who approves; what reopens it.

Common trap

"Low severity, so waive".

Evidence to cite

The attached proof, not the severity label.

Taught in

When to Waive and Evidence for a Waiver

Interviewer is testing

Honest metrics.

Bad answer

"Yes, zero means done."

Better answer

"Only if the zeros are trustworthy: final RTL, final intent, final constraints, no fresh waivers, trends stable."

Strong answer (a 4)

What a 4 sounds like: "Assumption: I trust no aggregate I have not decomposed. Mechanism of false confidence: waivers hiding risk, stale RTL behind a fresh report, intent gaps silently excluding crossings, ECOs after the last run. Trade-off: re-running everything costs days against the risk of a respin; the compromise is a defined gate checklist, not vibes. Evidence: the review-gate package (intent version, report version, waiver ages, ECO reconciliation) per the signoff readiness checklist. Zero is a claim; the package is the proof."

Follow-ups

Waiver aging; ECO reconciliation.

Common trap

Celebrating the number.

Evidence to cite

The gate package.

Taught in

Signoff Readiness and Honest Metrics