Why it matters

Hold fixes insert delay, and delay on a path is delay on every check through it. If the setup slack at that endpoint is smaller than the delay needed, the fix trades a hold violation for a setup one. Checking each fix against the available setup slack prevents that.

Worked example

1
2
3
4
5
6
7
8
9
10
11

margin = 0.005
endpoints = [("e1", -0.030, 0.200), ("e2", -0.010, 0.020), ("e3", -0.050, 0.040), ("e4", 0.004, 0.500)]   # (name, hold slack, setup slack) in ns
rows = {}
for name, hold, setup in endpoints:
    need = max(0.0, margin - hold)            # delay to add so hold slack reaches the margin
    left = setup - need
    rows[name] = (round(need, 3), round(left, 3), left >= -1e-12)
assert rows == {"e1": (0.035, 0.165, True), "e2": (0.015, 0.005, True), "e3": (0.055, -0.015, False), "e4": (0.001, 0.499, True)}
fits = [n for n, r in rows.items() if r[2]]
assert fits == ["e1", "e2", "e4"]
print("PASS: fixes that fit", fits, "; needs another method:", [n for n, r in rows.items() if not r[2]])

e1 and e4 have plenty of setup slack to spend. e2 fits with 5 ps left, so it is safe but it must be marked as having used its slack. e3 needs 55 ps and has only 40 ps: inserting the delay would turn a hold violation of 50 ps into a setup violation of 15 ps. e3 needs a different approach (see order of operations), such as reducing the spread between its longest and shortest paths. A hold fix at the endpoint adds the same delay to every path that reaches it, so the check uses the worst setup slack of those paths.

Placement of the delay

What the review looks for

• Each hold ECO lists the delay added and the setup slack left; slack left near zero is flagged.

• Delay is inserted as close to the endpoint as possible unless there is a reason to share it.

• Hold fixes are done in all hold scenarios together; a fix verified in one corner can still fail in another.

Related topics

• Closure Strategy

• Next: Fixing Setup