How does Liberty Variation Format (LVF) let POCV derates vary by slew and load instead of one fixed sigma?

Short Answer

A basic POCV model gives each timing arc one fixed sigma value, used no matter what input transition or output load that arc actually sees in the design. Liberty Variation Format (LVF) instead stores sigma as a small table indexed by input slew and output load, the same way nominal delay is already indexed, so the derate the tool applies can change arc-by-instance based on real operating conditions.

Technical Explanation

Delay itself already depends on slew and load in a standard Liberty model, and LVF extends that same table-lookup idea to variation data.

• A basic POCV sigma is a single number per arc. Without LVF, one sigma value represents the arc's variation regardless of whether that instance of the cell happens to see a fast, light-load switching event or a slow, heavily-loaded one.
• LVF stores sigma as a table, like delay itself. An ocv_std_dev_cell_rise (LIB) table in an LVF-enabled library gives sigma as a function of input transition and output load, the same axes the nominal cell_rise delay table already uses.
• Why variation actually depends on slew and load: a gate driving a heavy load, or receiving a slow input edge, behaves differently under the same underlying process variation than one driving a light load with a fast edge, so a single fixed sigma cannot capture both accurately.
• The tool looks up sigma the same way it looks up delay. For each specific instance of an arc in the design, at its actual computed input slew and output load, the tool interpolates the LVF table for a sigma value tailored to that operating point, not a library-wide average.
• This applies to more than just delay. LVF can model variation for delay arcs, and for setup and hold constraint checks, wherever the library provides the corresponding standard-deviation tables.
• The practical benefit is accuracy where it matters most. Arcs under unusual operating conditions — a very slow input edge on a lightly loaded gate, say — get a sigma that reflects their actual condition instead of one number meant to cover every condition on that cell.

Follow-up Question & Model Response

Two instances of the same buffer cell have very different input slews, one fast and one slow, but almost identical nominal delay after interpolation. Would a basic POCV sigma treat them the same way? Would LVF?

Candidate Model Response: A basic POCV library would likely treat them the same, since it stores one sigma value for that arc regardless of the actual slew each instance sees. An LVF-enabled library would not necessarily agree, because sigma is looked up from its own table indexed by slew and load, which does not have to track the delay table's shape exactly. Two instances with similar delay could still show measurably different sigma under LVF if their slew and load fall on different parts of the variation table — exactly the extra accuracy LVF provides.