How do you pick RC extraction corners for crosstalk signoff when the aggressor and victim need opposite worst cases?
From PDVerse STA Mentor Guide · pdVerse Mentor Guide
Short Answer
A single RC extraction corner cannot represent worst-case crosstalk on both sides of the same coupling event, because the aggressor needs a fast edge while the victim needs to look slow and easily disturbed. Signoff handles this by extracting the same physical nets under different assumptions for the aggressor and the victim inside one crosstalk check, rather than picking one corner for the whole design.
Technical Explanation
- RCmin (an extraction corner with low resistance and capacitance, giving fast, sharp edges) makes a good aggressor model, because a fast edge couples more sharply into a neighbor than a slow one.
- RCmax (an extraction corner with high resistance and capacitance, giving slow edges and more coupling exposure per unit time) makes a good victim model, because a slower, weaker-driven victim net is more easily pushed around by a neighbor's switching.
- For a setup check, a delay push-out on the victim is the concern, so the tool pairs an RCmin-like aggressor against an RCmax-like victim to get the worst push-out.
- For a hold check, a pull-in is the concern, so the pairing that maximizes pull-in is what signoff sweeps for, and it is not always the same physical corner assignment as the setup case.
- Because the same net can be modeled two different ways in one run, once as an aggressor and once as a victim, signoff tools automate this pairing internally rather than asking a designer to hand-pick per-net corners.
- Running every mode against every RC corner and SI on/off combination is usually too expensive, so signoff typically narrows to the corner pairs known to bound the aggressor/victim worst cases rather than the full matrix.
- This pairing is layered on top of the ordinary PVT (process-voltage-temperature) timing corner, so the same setup check is still evaluated at the slow-process, low-voltage, high-temperature corner it would use anyway, with the aggressor/victim RC pairing applied within that corner rather than replacing it.
- Because RC extraction itself has its own tolerance band, some flows also add a small guard-band on top of the RCmin/RCmax extremes for crosstalk-critical nets, treating the documented extraction corner as a starting point rather than an absolute physical limit.
Common Mistake
The Trap: running crosstalk analysis at only the nominal RC corner because it looks like "the real silicon corner," and missing that the worst aggressor/victim pairing usually sits at the extraction extremes, not at typical.
- A nominal-only crosstalk run can pass cleanly while the RCmin-aggressor and RCmax-victim pairing at a corner extreme still fails.
- Treating RC corner selection the same for setup and hold ignores that push-out and pull-in are maximized by different pairings.
Follow-up Question & Model Response
If RC extraction corners are separate from PVT (process-voltage-temperature) timing corners, how many total combinations does signoff actually need?
Candidate Model Response: In principle the combinations multiply, since each PVT timing corner could pair with each RC extraction corner and each SI on/off state, but signoff does not run the full cross product. Instead it identifies which RC corner and PVT combinations actually bound the worst setup and worst hold cases for crosstalk, based on which pairing maximizes push-out or pull-in, and restricts the scenario list to those. This keeps the scenario count in the tens rather than the hundreds, while still covering the physical situations that could actually happen on silicon.
Practical Example
On a design signing off at a 0.72V low-voltage corner, the setup check pairs an RCmin-extracted aggressor bus against an RCmax-extracted victim clock net, since that combination gives the largest delay push-out on the capture path, 38ps in this case. The hold check instead pairs an RCmax-extracted aggressor against an RCmin-extracted victim data net, producing a 21ps pull-in on the launch path. Running only the nominal RC corner at this voltage would have shown both paths passing with over 50ps of margin, hiding both real failures. Because both worst-case pairings sit at the same low-voltage PVT corner, the signoff scenario list adds just two extra RC-pairing runs at that one corner, rather than doubling the entire PVT corner count, keeping the added runtime manageable.
Complete STA Handbook
Master Signoff-Ready Static Timing Analysis
Get the complete 10-chapter STA handbook covering setup/hold margins, clock modeling, OCV/POCV, crosstalk noise, and PrimeTime closure.
Continue practising