Debug playbook

Debug Playbook for Training & Timing Modes focuses on training margin, eye width, boot failure rate. The goal is to connect the observable symptom to protocol mechanism, ownership, and regression risk.

Protocol debug is a search for the FIRST deviation, not the loudest symptom. Timeouts and error flags are usually many cycles downstream of the real cause.

Root-cause tree

ROOT-CAUSE TREE — Training & Timing Modes

training margin, eye width, boot failure rate looks wrong
        |
   reproducible?
     /        \
   no          yes
   |            |
 flaky env   same first transaction every time?
 / seed         /            \
              yes             no
               |               |
        protocol rule     timing/reset/PVT
        or config bug     or load-dependent

1. Freeze the failing seed, firmware tag, and spec revision.

2. Find the first bad transaction, not the loudest downstream timeout.

3. Map the transaction to signals and VIP monitor events.

4. Classify the failure: protocol rule, integration config, timing/reset, or performance pressure.

5. Prove the mechanism with one reduced sequence.

6. Patch the smallest owner-controlled boundary and rerun compliance plus workload traffic.

Review memo template

STAFF PROTOCOL REVIEW MEMO — Memory Interfaces (DDR / LPDDR / HBM) / Training & Timing Modes

1. Symptom
   - Watched metric: training margin, eye width, boot failure rate
   - Failing interface: <master/slave/endpoint/controller/PHY>
   - Transaction identity: <ID/tag/address/endpoint/lane>
   - Repro setup: <sim/emulation/FPGA/silicon + firmware tag>

2. Mechanism hypothesis
   - Primary mechanism: training aligns DQS/DQ timing and voltage margins so digital transfers survive PVT and board/package variation.
   - Competing hypothesis: <timing, reset, bridge, ordering, firmware, or VIP issue>
   - Missing evidence: <waveform, analyzer trace, counter, spec clause, or log>

3. Proposed action
   - Minimal reversible change: <RTL, register setting, bridge config, scheduler, VIP check>
   - Expected metric movement: <delta and workload>
   - Regression risk: ordering, compatibility, performance, power, area, or timing

4. Signoff
   - Re-run artifact: training transcript, margin report, mode register dump
   - Required owners: bring-up owner, PHY owner, board owner
   - Final decision: fix, waive, document limitation, or escalate to architecture

Protocol deep dive

DDR bandwidth is scheduler + PHY: rows, banks, refresh, and turnarounds eat headline data rate.

Concept diagram

MEMORY PATH

masters -> controller scheduler -> PHY -> DRAM banks
              |                      |
         refresh/QoS            training/margin

Scheduler sees transactions; PHY sees picoseconds.

Metric graph

BANDWIDTH LOSS WATERFALL

peak              ████████████████████████
refresh           █████████████████████
turnaround        ██████████████████
row miss          ██████████████
effective         ██████████████

Quote the bottom bar in reviews.

Metrics and artifacts to collect

• effective BW

• row hit rate

• refresh stall %

• training margin

• ECC error log

Mini case study

Video workload lost half effective bandwidth after firmware enabled aggressive low-power refresh. Scheduler and firmware QoS had to be co-designed.

Debug branches

• If ECC errors, check training margin and address interleave first.

• If BW low with high row hit, suspect port arbitration not DRAM.

• If boot fail, stop at training step in transcript.

Senior review question

Ask: what is the first transaction that deviates, and which spec rule does it test?

Key takeaways

• Connect every protocol claim to a transaction identity and measurable metric.

• Store the artifact (waveform, log, counter) next to every signoff decision.

Common pitfalls

• Debugging timeouts without finding the first bad transaction.

• Quoting peak bus width without payload efficiency and retry overhead.

• Treating VIP compliance as a substitute for system integration replay.

Principal review addendum

Re-read Training & Timing Modes against one concrete product workload, not a synthetic directed test.

training aligns DQS/DQ timing and voltage margins so digital transfers survive PVT and board/package variation.