Why can summing every aggressor's worst-case crosstalk contribution be overly pessimistic, and what actually limits how many aggressors can realistically align?

Short Answer

Naively adding every aggressor's individually worst-case contribution assumes all of them can switch at exactly the moment needed to maximally disturb the victim, in the worst-case direction, all at once. Two things actually constrain this: each aggressor's own timing window (the real range of times its edge can occur, given its own paths and clocks) may not even overlap the victim's transition window, and the probability that every small aggressor aligns simultaneously in the same direction drops fast as the aggressor count grows — which is why the tool can use a statistical composite-aggressor model instead of a flat worst-case sum for the smaller contributors.

Technical Explanation

• Each aggressor net has its own timing window: the earliest and latest time its signal can actually transition, computed from its own launch clock, its own path delay, and any derating already applied to that path. It is not free to switch whenever it likes.
• The tool computes the victim's own arrival window the same way, then checks whether the aggressor's window and the victim's window actually overlap in time. If an aggressor's transition can only occur, say, between 9.0 and 10.0 time units while the victim has already transitioned between 7.0 and 8.0, no crosstalk delay effect between them is physically possible, no matter how large the coupling capacitance is on paper.
• OCV derating can artificially widen these arrival windows on the shared clock path leading up to the aggressor and victim, manufacturing overlaps that would not exist in the real circuit — this is exactly the same shared-clock-trunk pessimism CRPR addresses elsewhere, and pba_enable_xtalk_delay_ocv_pessimism_reduction (PT) applies CRPR to these arrival-window computations specifically, tightening the windows back toward what is physically possible.
• Separately from window overlap, summing every aggressor's full worst-case bump also over-assumes correlation: it is statistically unlikely that a large number of small, physically independent aggressors all switch in the same disturbing direction at the same instant, the same way it is unlikely that many independent coin flips all come up heads.
• The tool's composite aggressor mode addresses this directly for small aggressors: any aggressor whose individual bump height is below a set threshold (si_xtalk_composite_aggr_noise_peak_ratio, PT, default 1 percent of supply voltage) gets folded into a single composite aggressor instead of being summed at full worst case. si_xtalk_composite_aggr_mode (PT) set to statistical tells the tool to combine these small aggressors probabilistically rather than by flat worst-case addition.
• The composite bump height is set so that it is exceeded only with some chosen small probability — si_xtalk_composite_aggr_quantile_high_pct (PT), default 99.73 percent, corresponding to three standard deviations from the mean of the combined small-aggressor distribution — rather than assuming the theoretical maximum every small aggressor could contribute if they all switched together.
• What breaks: reporting every path with full worst-case, fully-aligned aggressor summation on a net with many neighbors produces phantom violations on nets that are, in the real switching behavior of the design, never actually exposed to that combined disturbance, wasting fixing effort on paths that were never really at risk.

Follow-up Question & Model Response

"If a specific critical net has only two or three large aggressors rather than many small ones, does the composite aggressor statistical argument still apply, and how would you analyze it differently?"

Candidate Model Response: No — the statistical composite argument specifically targets many small aggressors below the bump-height threshold, where the law-of-large-numbers style reasoning about unlikely simultaneous alignment actually holds. A net with only two or three large aggressors does not have enough independent contributors for that reasoning to meaningfully reduce pessimism, so each of those aggressors should be analyzed individually at its own real timing window rather than folded into a composite. I would confirm the arrival-window overlap for each of those specific aggressors against the victim directly, and if the net is genuinely critical, I would also consider set_si_delay_disable_statistical (PT) on it to guarantee full, non-composite treatment even if it happened to sit near the composite-mode threshold.