BeginnerQuestion 67 of 95Source: Synopsys PrimeTime User Guide: Parametric On-Chip Variation (POCV)

What is statistical on-chip variation, and how does POCV differ from a flat derate percentage?

From PDVerse STA Mentor Guide · pdVerse Mentor Guide

By Tabish Iqbal · 9+ Years VLSI Physical Design Experience

Short Answer

Parametric on-chip variation, POCV, models each cell's delay as a range described by a mean value and a standard deviation, or sigma, instead of one worst-case number scaled by a fixed percentage. PrimeTime reads this mean-and-sigma data from the Liberty library's variation tables, sometimes called LVF data, and combines it statistically across a path rather than assuming every stage varies by the exact same fixed amount. That gives a derate that automatically adapts to how many stages a path has and how much each individual arc actually varies, instead of relying on one guessed percentage for the whole design.

Bell-curve diagram of a Liberty LVF variation table showing mean delay and 3-sigma spread for a timing arc, contrasted with a single flat derate line — Technical Reference DiagramWhat is statistical on-chip variation, and how does POCV differ from a flat derate percentage?

Technical Explanation

POCV replaces a guessed percentage with real, per-arc variation data from the library.

Liberty variation tables, sometimes called LVF (Liberty Variation Format), give a standard deviation for each timing arc's delay, alongside the normal mean delay value.
POCV multiplies each arc's sigma by a configurable number of standard deviations for a chosen confidence level, then combines those per-arc variations across the whole path statistically instead of by simple multiplication.
Because variation partly cancels out across independent stages, a path with many stages gets a smaller relative margin than a short path with the same per-stage sigma, without needing a separate stage-count table the way AOCV does.
set_timing_derate -pocvm_coefficient_scale_factor (SDC/PT) scales how many standard deviations of margin PrimeTime applies on top of the library's own data.
Where AOCV depends on pre-characterized stage/distance tables, POCV depends on the library shipping accurate LVF data for every relevant arc.

Common Mistake

The Trap: assuming a flat percentage derate and a POCV sigma-based derate can simply be mixed on the same path without double-counting margin.

A leftover flat set_timing_derate -late 1.10 derate on top of a POCV run that already reads Liberty sigma values stacks two variation models on the same delay, over-margining the path far beyond what either model intended.
The fix is to pick one model, flat, AOCV, or POCV, per signoff corner and remove leftover flat derate commands once POCV is enabled.

Follow-up Question & Model Response

If POCV already models variation statistically, why does PrimeTime still let you add an extra guardband on top of it?

Candidate Model Response: The library's LVF characterization is only as accurate as the silicon data it was built from, and a vendor may not have fully characterized every corner or unusual condition a chip sees in the field. The -pocvm_guardband (SDC/PT) option lets a team add a small margin on top of the statistically modeled sigma to cover that residual uncertainty, without abandoning the statistical model entirely. Teams typically start with a modest guardband and reduce it once silicon correlation confirms the library's variation numbers.

Practical Example

A 20-stage path with a POCV sigma-based model reports a 3-sigma late derate near 1.06, versus the flat 1.12 a non-statistical flow used for the same stage count, recovering about 40ps of setup slack. A shorter 3-stage path in the same design still gets a derate near 1.10 under POCV, since fewer stages give less averaging to lean on.

Complete STA Handbook