Analysis finds violations; closure removes them. This section is about the removing: why the loop converges and where it stalls, how to fix hold without breaking setup, which kinds of change are cheap and which are risky, how to move slack with useful skew, and how to verify each change without rerunning the world. Every lesson has a runnable example with small, consistent numbers.
Lesson | After it you can... |
|---|---|
Predict where a closure loop stalls. | |
Fit hold fixes into setup slack. | |
Prefer shared segments. | |
Find endpoints delay cannot fix. | |
Buy delay at the lowest leakage. | |
Choose the number of buffers. | |
Restructure by arrival time. | |
Match each change to its checks. | |
Compute the skew window. | |
Distribute slack along a pipeline. | |
Include variation in skew. | |
Check skew per corner. | |
Compute a batch's checks. | |
Choose a fast scenario set. | |
Batch conflicting ECOs. | |
Detect a stall. |
Suggested order: top to bottom. Previous sections: Crosstalk and Noise, OCV, AOCV, POCV and Variation and Multi-Corner Multi-Mode Signoff.
Setup slack plus hold slack at the endpoint is unchanged by delay insertion in the same corner: delay trades one for the other one-for-one, but the sum must be checked per scenario because an inserted cell is slower in the slow corner than in the fast one.