Why it matters

The worst victims in a crosstalk report decide most of the SI closure work. Reading a victim block carefully, in particular which aggressors count and which were dropped, tells you whether the number can be trusted and where to act.

An illustrative victim block

illustrative generic format (not tool output)
victim: n_data   transition: rise   check: setup   nominal delay: 42.0 ps
aggressor   coupling (fF)   window overlap   delta (ps)
a1          3.2             yes              +6.1
a2          1.8             yes              +3.0
a3          2.5             no                0.0   (filtered: window)
total delta: +9.1 ps    delay with SI: 51.1 ps

1
2
3
4
5
6
7
8
9
10
11
12
13
14

nominal = 42.0
aggressors = [("a1", 3.2, True, 6.1), ("a2", 1.8, True, 3.0), ("a3", 2.5, False, 0.0)]
total = round(sum(d for _, _, overlap, d in aggressors if overlap), 3)
assert total == 9.1
assert round(nominal + total, 3) == 51.1
counted = [n for n, _, o, _ in aggressors if o]
filtered = [n for n, _, o, _ in aggressors if not o]
assert counted == ["a1", "a2"] and filtered == ["a3"]
coupling_counted = round(sum(c for n, c, o, _ in aggressors if o), 3)
coupling_total = round(sum(c for _, c, _, _ in aggressors), 3)
assert (coupling_counted, coupling_total) == (5.0, 7.5)
per_ff = {n: round(d / c, 2) for n, c, o, d in aggressors if o}
assert per_ff == {"a1": 1.91, "a2": 1.67}
print("PASS: total delta", total, "ps; filtered", filtered, "; delta per fF", per_ff)

The total (+9.1 ps) is the sum of the two counted aggressors (6.1 + 3.0); the third, with 2.5 fF of coupling, is filtered by window and contributes nothing. Two quick checks come from the table: the delta per femtofarad of coupling should be of similar size for similar aggressors (1.91 and 1.67 ps per fF here), and a large coupling with a tiny delta needs an explanation (a filter, a weak edge, an aggressor that is nearly always quiet).

What the review looks for

• Largest total deltas first; then the aggressors that cause most of each.

• A filtered aggressor with large coupling is checked: the filter must be the reason, not an extraction gap.

• Report columns and units are defined once and used the same way across reports, so deltas can be compared between runs.

Related topics

• Next: Windows and Filters

• Report Timing Deep Read