IntermediateQuestion 16 of 112

What is the difference between graph-based analysis (GBA) and path-based analysis (PBA)?

From PDVerse STA Mentor Guide, part of the pdVerse Mentor Guide

Short Answer

GBA keeps a single worst-case slew value at each node of the timing graph and reuses that same value for every path passing through it โ€” fast, but pessimistic, since it can penalize a fast path with a slow, unrelated path's slew. PBA re-times each path individually using its own actual slews all the way along, which is more accurate but too computationally expensive to run on every path in the design.

Technical Reference DiagramWhat is the difference between graph-based analysis (GBA) and path-based analysis (PBA)?

Technical Explanation

  • What the timing graph shares by default. STA models the design as a graph of nodes (pins) connected by arcs (cell and net delays). GBA computes one worst-case slew (transition time) at each node โ€” every path through that node inherits the same slew, whether or not it actually produces the slowest edge itself.
  • Why GBA is pessimistic. If two paths converge at a shared node and one drives a much heavier load, GBA applies that slower transition's consequences to both โ€” even the path that never saw it gets timed as if it did, which can make GBA report worse slack than the design truly has.
  • What PBA does differently. Path-based analysis re-times a specific path end to end using its own actual slew at every stage instead of a shared graph value, typically recovering some of that artificial penalty.
  • Why PBA isn't run on everything. Re-timing every path individually is far more computationally expensive โ€” practical only on a limited set, usually the worst violators GBA already flagged.
  • How the two are used together. A typical signoff flow runs GBA across the whole design first, then re-times only the reported violators with PBA to see how many are real and how many were GBA's own pessimism.

Common Mistake

The Trap: Treating every GBA-reported violation as a real, physically-existing timing failure that must be fixed.

  • Some fraction of GBA violations are pessimism artifacts from shared worst-case slews, not genuine problems with that specific path.
  • Spending ECO effort fixing a path that PBA would have shown as already passing wastes engineering time and can introduce unnecessary area or power cost for a violation that was never physically real.

Follow-up Question & Model Response

If PBA typically recovers slack compared to GBA, could PBA ever report a worse slack than GBA on the same path?

Candidate Model Response: In most reported cases PBA recovers slack because it removes GBA's shared worst-case slew pessimism, but the relationship is not an absolute guarantee in every tool and mode โ€” PBA's job is to be more accurate, not necessarily always more optimistic. If GBA happened to use an unrealistically favorable slew at some node for a specific path, a fully accurate path-specific re-time could in principle expose a worse number there instead. This is why PBA results are treated as the more trustworthy number, not simply as an automatic slack bonus applied to every GBA violator.

Practical Example

A 600 MHz block (1.67 ns period) shows a GBA setup violation of โˆ’45 ps on a path through a shared clock-buffer node also used by a much higher-fanout sibling path. That sibling path's heavier load produces a 0.18 ns transition at the shared node, and GBA propagates that same 0.18 ns transition onto the reported path even though the reported path's own fanout would only produce a 0.09 ns transition. Re-timing just that one path with report_timing -pba_mode path (PT) using its own actual 0.09 ns transition recovers roughly 30 ps of slack, changing the result from a โˆ’45 ps GBA violation to a โˆ’15 ps PBA result โ€” still a real violation, but a much smaller one, and one that needs a much smaller fix than the GBA number suggested.

Complete STA Handbook

Get the complete 10-chapter STA handbook covering setup/hold margins, clock modeling, OCV/POCV, crosstalk noise, and PrimeTime closure.

Offline PDF Bundle

Want all 1109 questions offline?

Get the complete 4-book PDF bundle (PnR, STA, MMMC, Low Power) with a clickable table of contents - no ads, no internet needed.

See what's inside the bundle
Static Timing Analysis (STA) Handbook โ€” ten chaptersSTA HandbookTen chapters on setup, hold, OCV, and PrimeTime signoff.