What is capacitive coupling percentage, and how does it affect delay calculation?

Short Answer

Capacitive coupling percentage is the share of a net's total capacitance that comes from coupling to neighboring nets, rather than from ground or fixed-reference capacitance. A net with a high coupling percentage is dominated by its neighbors' switching behavior, so its delay is far more sensitive to what nearby aggressors do than a net whose capacitance is mostly grounded.

Technical Explanation

Every net's total capacitance splits into two kinds, and the split matters as much as the total value.

• Ground capacitance is fixed and predictable. This portion of a net's capacitance goes to a stable reference — substrate, a power or ground plane — and does not depend on what any other signal is doing.
• Coupling capacitance depends on a neighbor's state. This portion links the net to another signal net, so its effective contribution to delay changes depending on whether that neighbor is switching, and in which direction, at the time in question.
• A high coupling percentage means a volatile net. If most of a net's capacitance is coupling rather than ground, its delay can swing significantly between a quiet-neighbor case and a worst-case switching-neighbor case — the same net can look fast or slow depending entirely on context.
• Non-SI timing uses a fixed assumption about this split. Without SI enabled, the tool typically treats total capacitance as if it behaved like ground capacitance for delay purposes, which is why a non-SI run misses the volatility that a high coupling percentage actually causes.
• Preserving coupling data requires the right read-in option. read_parasitics -keep_capacitive_coupling (PT) keeps the ground/coupling split intact from the SPEF or GPD source data; without it, the tool may collapse coupling capacitance into an equivalent grounded value, discarding the information SI analysis needs.
• Why it matters for prioritization: a net's raw capacitance value alone does not tell you how SI-sensitive it is — two nets with the same total capacitance but very different coupling percentages carry very different crosstalk risk.

Follow-up Question & Model Response

If parasitics were read in without -keep_capacitive_coupling, would report_timing still show any delay at all for a net, and would the number be reliable?

Candidate Model Response: Yes, it would still show a delay number, but that number would understate reality for any net with meaningful coupling percentage. Dropping the coupling split does not remove capacitance from the calculation — it folds coupling capacitance into an equivalent ground value, so the net still gets a delay based on total capacitance. What is lost is the ability to model that capacitance as context-dependent, so the reported delay reflects an average or nominal assumption rather than the worst-case push-out a real switching neighbor could cause. I would treat any signoff run missing -keep_capacitive_coupling as unreliable for SI-sensitive nets specifically, even though the report produces numbers without complaint.