SerDes & High-Speed I/O · All levels

Link Training & Calibration: Tricky Q&A

Senior interview and review questions for Link Training & Calibration.

Section Q&A bank

Use these drills after completing all topics in Link Training & Calibration. Answer with workload context, mechanism proof, artifact, owner, and release decision.

What is the safest order of operations during lane bring-up?

diagram
[INT][SERDES][LINK-TRAINING-CALIBRATION]

Q: What is the safest order of operations during lane bring-up?

A:
Typically: detect partner presence, establish electrical common mode, run receiver detection, apply safe default presets, run adaptation on PRBS, verify BER/FEC, then enable alignment and mission traffic. TX must not launch full swing before partner RX is ready. Each protocol specifies fencing; violating order risks electrical overstress or false lock.

FOLLOW-UP TRAP: Enabling TX output before RX adaptation and detect complete.

Why store per-SKU coefficient profiles in firmware?

diagram
[INT][SERDES][LINK-TRAINING-CALIBRATION]

Q: Why store per-SKU coefficient profiles in firmware?

A:
Board and package variation shifts optimal presets; factory calibration reduces training time and improves yield. Field retrain then starts near known-good coefficients, avoiding local minima. Profiles must version with BOM changes and include rollback if new profile fails margin gates.

FOLLOW-UP TRAP: Using one global preset for all board variants to simplify firmware.

How does deskew buffer depth affect latency and fault behavior?

diagram
[INT][SERDES][LINK-TRAINING-CALIBRATION]

Q: How does deskew buffer depth affect latency and fault behavior?

A:
Deeper buffers absorb more skew but add fixed latency and gate complexity. Insufficient depth fails on worst-case lane delay spread. Misaligned marker lock with adequate eye per lane still corrupts data—debug must check alignment state, not only per-lane BER.

FOLLOW-UP TRAP: Debugging lane BER while alignment FSM is unlocked or slipping.

What runtime signature suggests impending link failure before hard BER collapse?

diagram
[INT][SERDES][LINK-TRAINING-CALIBRATION]

Q: What runtime signature suggests impending link failure before hard BER collapse?

A:
Rising FEC correctable error rate, coefficient wander, shrinking margin on periodic health polls, and increased retrain requests often precede hard failure. Temperature-correlated drift points to SI/PI or thermal margin loss. Trending beats single-threshold BER alarms for fleet ops.

FOLLOW-UP TRAP: Alerting only on uncorrectable errors without monitoring correctable trend.

Q&A drill guide

diagram
WORKLOAD -> SerDes SYMPTOM -> TIMING/QUEUE METRIC -> ROOT CAUSE -> FIX -> REGRESSION

Sketch while answering

diagram
LINK TRAINING CALIBRATION
lane-bringup-sequence -> coefficient-training -> closure

Key takeaways

  • Always tie controller and PHY counter shifts to application latency and throughput outcomes.

  • Lock firmware timing profile, thermal condition, and DIMM state before comparing SerDes captures.

Common pitfalls

  • Chasing peak bandwidth while ignoring p99 latency and fairness tails.

  • Changing timing guardbands without separating SI noise from scheduling issues.

  • Declaring closure without reliability gates, fault injection, and regression replay.