What are WNS, TNS and the number of violating endpoints, and what does each tell you?

Short Answer

WNS is the worst negative slack, the single worst endpoint. TNS is the total negative slack, the sum of the negative slacks of all failing endpoints. NVE is the number of violating endpoints. Together they tell you how bad the worst path is, how much total work is left, and whether the problem is a few paths or a wide spread.

Technical Explanation

• Slack is computed per endpoint, and each endpoint keeps only its worst slack. WNS is the most negative of those values. It tells you how far the single worst path is from passing, which usually sets the size of the fix you need, such as a Vt swap versus a restructure.
• TNS adds up the negative slacks of every failing endpoint. Passing endpoints contribute nothing. It measures the total amount of work left, and it is the better number for tracking whether an ECO loop is converging.
• NVE counts the failing endpoints. It separates a narrow problem from a wide one. A TNS of -2 ns could be 4 endpoints at -500 ps or 400 endpoints at -5 ps, and those need very different fixes.
• In PrimeTime, TNS defaults to union mode: timing_report_union_tns (PT) is true, so each endpoint contributes its worst slack once across path groups and scenarios. In every-group mode an endpoint that fails in two path groups counts twice.
• report_qor (PT) and report_qor (ICC2) both summarize timing per path group, including TNS, but their numbers differ because the timing engines and parasitics differ. Compare like with like across loops.
• report_global_timing (PT) reports WNS, TNS and violating counts gathered from violating paths only, split into categories such as in-to-reg, reg-to-reg and reg-to-out. That split quickly shows whether failures sit on I/O constraints or internal logic.
• Setup and hold have separate WNS, TNS and NVE. Always read them as two sets of three numbers, per scenario.

Formula Or Decision Rule

• Endpoint slack: slack(e) = worst slack over all paths ending at e.
• WNS = min over endpoints of slack(e), reported as a negative number only when something fails.
• TNS = sum of slack(e) over all endpoints with slack(e) < 0.
• NVE = count of endpoints with slack(e) < 0.
• Average violation = TNS / NVE, a quick way to tell a few deep failures from many shallow ones.

What To Check

• WNS, TNS and NVE for setup and for hold, per active scenario.
• Whether TNS is in union mode, so the number matches what the team is tracking.
• The spread of failures across path groups from report_global_timing (PT).
• The TNS / NVE ratio, which shows whether you face deep or shallow violations.
• The trend of all three numbers across ECO loops, not just the latest value.

Healthy, Suspicious & Hard-stop Results

• Healthy (illustrative): Setup and hold WNS at or above 0, TNS 0 and NVE 0 in every active scenario.
• Suspicious (illustrative): WNS of -10 ps with NVE under 20 at signoff, all on one or two path groups you can explain.
• Hard stop: TNS that grows between two ECO loops, or NVE in the thousands at signoff. The loop is not converging and needs a different strategy.

What The Interviewer Is Testing

• Can you define TNS precisely, including that only negative endpoint slacks add up?
• Knowing that the same TNS can mean very different problems depending on NVE.
• Awareness that report_qor (ICC2) and report_qor (PT) numbers differ and should not be mixed.

Follow-up Question & Model Response

"Why can TNS get worse while WNS gets better?"

Candidate Model Response: WNS looks at one endpoint and TNS looks at all of them. An ECO that upsizes cells on the worst path adds input capacitance to shared fan-in logic. That slows other paths through the same drivers, and some move from +2 ps to -3 ps. WNS improves because the worst path got faster, but NVE goes up and TNS gets worse. Watch all three numbers after every loop.