Why is a per-cell IR-drop annotation treated as uncorrelated with the main rail, and what's the consequence?

Short Answer

Correlation is a physical claim, not a convenience setting. A per-cell IR-drop annotation (a local voltage droop measured at one instance's supply pin) describes an independent local event, so PrimeTime compares the full worst-case voltage difference across it instead of assuming the driver and load move together.

Technical Explanation

• Correlation has to be earned physically. When two cells share one UPF supply net, their voltages genuinely rise and fall together, so PrimeTime can safely ignore any small difference between them for signal-level checks.
• A per-cell IR-drop annotation is the opposite case. It records a local droop measured at one instance's own power grid. There is no physical reason a droop at cell U3 means cell U1 is drooping the same amount - they can easily be at different voltages at the same instant.
• What "uncorrelated" makes the tool do. Because the two voltages are independent, PrimeTime checks the full worst-case voltage difference across the annotated crossing rather than assuming the driver and load track each other.
• Worked numbers. Say the annotation gives a driver at 0.8V against a load at 1.1V in one direction, and 1.2V against 0.7V in the other. Both comparisons are evaluated at their worst case, and both come back as violations.
• Why the alternative would be dangerous. Assuming driver and load move together would be optimistic in exactly the scenario the annotation exists to model - a real electrical mismatch between two cells that genuinely differ in voltage right now.
• The practical consequence. Adding an IR-drop annotation can introduce new signal-level violations on crossings that were clean before. Those are not artifacts to filter away; they reflect a real risk the annotation just made visible.

Follow-up Question & Model Response

If two cells sharing a UPF supply net are always treated as correlated, could you ever legitimately need an uncorrelated per-cell annotation on cells that share that same net?

Candidate Model Response: Yes - sharing a nominal UPF supply definition doesn't guarantee two physically distant cells see identical instantaneous voltage once local IR drop on the physical grid is considered. If floorplan analysis shows one instance sits far from the power source and suffers a measurably different local droop under switching load, applying a per-cell annotation to that specific instance captures the real local condition even though both cells are nominally on the same rail. The correlated-by-default rule is a simplification for the common case, not a law about UPF nets; a targeted annotation is how you override it where physical measurement says otherwise.