Most timing failures that matter are not found by reading one path report. They are found by noticing what the report does not cover, by localising why two runs disagree, by running the last weeks of closure with discipline, and by keeping results reproducible. This section is a playbook for those four activities, with a runnable check behind every pattern.
Lesson | After it you can... |
|---|---|
Count unclocked registers. | |
Spot a half cycle modelled as a full one. | |
Measure an exception's coverage. | |
Pair setup N with hold N-1. | |
Find the first divergent stage. | |
Run two sanity checks. | |
Flag suspicious improvements. | |
Classify a difference by its shape. | |
Bucket endpoints by history. | |
Assign owners and find orphans. | |
Escalate from arithmetic. | |
Separate TNS from count. | |
Reach a verdict from manifests. | |
Rank constraint changes by risk. | |
Set a tolerance. | |
Gate signoff on weights. |
Suggested order: top to bottom. Previous sections: ECO and Closure, Crosstalk and Noise and Constraints and SDC.
The forgotten multicycle hold: with a multicycle setup and no hold exception, the hold check moves out by N-1 periods.