Every timing summary reduces thousands of numbers to a few. This page defines each metric, how it is computed, and what it does not prove. It gives no target values: acceptable levels belong to a project's signoff policy, not to a reference.
Metric | Definition | What it tells you | What it does not prove |
|---|---|---|---|
Setup WNS | Smallest setup slack over endpoints | How far the worst setup path is from passing | That other endpoints are fine, or that all endpoints are checked |
Setup TNS | Sum of negative setup slacks over endpoints | The total size of the fixing job | How many endpoints fail: many small and few large can give the same TNS |
Violating endpoints (NVP, FEP) | Count of endpoints (or paths) with negative slack | How many places need work | How bad they are |
Hold WNS, TNS and count | The same three, on hold slack | The state of the hold check | That setup is fine: hold fixes consume setup slack |
Union versus per-view | Union takes the worst slack per endpoint over views; per-view reports each view separately | The union is the job size; per-view shows where to look | Never add per-view totals: an endpoint failing in two views is counted twice |
Near-critical band | Count of endpoints with slack between zero and a chosen threshold | How exposed the result is to small changes | Nothing about violations: it is a measure of fragility |
Path count | Number of paths covered by a group, exception or report | The scale of an effect, such as what an exception removed | That every path was analysed: coverage needs its own report |
Unconstrained endpoints | Endpoints with no timing check | Coverage gaps | Anything about timing: they are unchecked |
Setup plus hold slack at an endpoint | Sum of the two slacks at one endpoint in the same scenario | Whether delay insertion can fix the endpoint in that scenario; across corners the setup cost of inserted delay is larger, so the same-corner sum is only a screen | Anything across endpoints or corners |
Pulse width and period slack | Phase width or period at a pin less the cell's minimum | Clock waveform margin at macros and latches | Data path timing |
Hidden total (SI filters) | Delta delay removed by coupling or delta thresholds | How much the filters hide | That the hidden total is small unless it is reported |
Read the scenarios first: which views the number covers, and whether it is per view or union.
Read the check type: setup and hold are different problems with different fixes.
Read the coverage: endpoints checked, excepted and unconstrained.
Read the margins applied (derates, uncertainty) with the numbers.
Only then read WNS and TNS.
TNS adds negative slacks: forty violations of 5 ps and two of 100 ps both total 200 ps, but need very different fixes.