Silicon Bring-up · All levels

Failure Triage & Debug: Tricky Q&A

Senior interview and review questions for Failure Triage & Debug.

Section Q&A bank

Use these drills after completing all topics in Failure Triage & Debug. Answer with setup context, mechanism proof, artifact, owner, and release decision.

When a bring-up failure disappears after enabling verbose tracing, what should you conclude first?

diagram
[INT][BRINGUP][FAILURE-TRIAGE]

Q: When a bring-up failure disappears after enabling verbose tracing, what should you conclude first?

A:
Treat observability as a perturbation and assume the failure window shifted, not that the bug vanished. Verbose tracing can change timing, interrupt load, memory pressure, and power states, so you should re-establish a minimally intrusive repro path before trusting any hypothesis.

FOLLOW-UP TRAP: Declaring the issue fixed because the instrumented build now passes repeatedly.

How do you differentiate livelock from deadlock during a silicon hang investigation?

diagram
[INT][BRINGUP][FAILURE-TRIAGE]

Q: How do you differentiate livelock from deadlock during a silicon hang investigation?

A:
Look for bounded but non-productive activity. Livelock shows repeated state transitions and traffic with no completion progress, while deadlock shows a stable wait cycle with no legal escape. Heartbeat counters plus dependency graph analysis are the fastest discriminator.

FOLLOW-UP TRAP: Classifying any non-booting system as deadlocked without checking ongoing state churn.

Why is a failure-rate curve more valuable than a single deterministic reproduction for marginal bugs?

diagram
[INT][BRINGUP][FAILURE-TRIAGE]

Q: Why is a failure-rate curve more valuable than a single deterministic reproduction for marginal bugs?

A:
Marginal bugs are often probabilistic interactions of timing and environment. A curve across controlled stress axes reveals sensitivity and confidence bounds, helping you distinguish true causality from incidental correlation and quantify mitigation impact.

FOLLOW-UP TRAP: Optimizing for one deterministic setup that cannot explain field variability.

What makes an FA request likely to produce actionable evidence instead of an expensive dead end?

diagram
[INT][BRINGUP][FAILURE-TRIAGE]

Q: What makes an FA request likely to produce actionable evidence instead of an expensive dead end?

A:
A hypothesis-linked handoff with exact failing context, suspect region, and requested technique rationale. Asking for FIB or emission microscopy without a constrained failure model usually returns interesting images but weak causality for fix decisions.

FOLLOW-UP TRAP: Escalating to FA with generic 'chip fails to boot' notes and no experiment-backed suspect boundary.

Q&A drill guide

diagram
SYMPTOM -> ROOT-CAUSE CLASS -> ARTIFACT -> OWNER ACTION -> VALIDATION

Sketch while answering

diagram
TRIAGE CONVERGENCE

symptom -> classify -> isolate -> prove -> bounded fix -> replay

Common traps

  • Preserve first-failure state before reruns.

  • Use disproof-oriented experiments to collapse cause tree quickly.

  • Promote fixes only after recurrence tracking windows pass.

Key takeaways

  • Tie every bring-up claim to one reproducible setup state and one proving artifact.

  • Prefer bounded fixes with clear owner and rollback trigger over broad multi-variable edits.

Common pitfalls

  • Running parallel uncontrolled experiments and losing causality.

  • Declaring closure without replaying across representative corners.

  • Escalating severity before bench/setup hypotheses are disproven.