METRIC → HYPOTHESIS → EXPERIMENT → FIX → REGRESSION
METRIC: WNS/TNS, corner, mode, path pin names
HYPOTHESIS: ordered causes (no random tweaks)
EXPERIMENT: one report/map before changing layout
FIX: minimal physical/constraint/RTL change
REGRESSION: corners, modes, DRV, LVS, IR re-run1. Verify SDC tag + corner + mode + SI on
2. Path report: net delay vs cell delay
3. RTL pipeline (if allowed)
4. Layer up / spread placement / CTS latency
5. Size / VT swap / buffer / useful skew (hold risk)
6. ECO metal/cell — document tag1. Hold corner report (often BC fast)
2. Short data paths — insert delay/buffer
3. Skew adjust if budget (check setup regression)
4. Hold-specific opt pass — cluster by clock domain1. Map hotspot ↔ macro channel / pin escape
2. Macro rotate / widen channel / soft blockage
3. Spread placement density
4. Layer promotion / routing preference
5. Drop util 3–5% or die bumpLVS clean (connectivity)
DRC clean or approved waivers only
STA matrix: setup @ WC + hold @ BC (all modes)
IR/EM @ agreed vectors
Revision tag matches GDS + SDC + netlist
Timing: report_timing -max/-min, path slack, clock latency
Route: congestion map, DRV summary
CTS: skew/latency histogram
Power: IR map, EM ratio report
PV: DRC count by rule, LVS compare
Signoff: closure dashboard one slideLVS fail → stop tapeout timeline until clean
Wrong revision tag → stop all exports
IR > spec with timing margin thin → escalate to architect + power lead
Systematic DRC >100 → cluster fix meeting, not hero edits
Name corner before fix — every time.
Congestion before timing opt on unroutable nets.
Do not recite this sheet linearly. Use it as a mental index. When the interviewer gives a failure, jump to the corresponding fix order and narrate the evidence you need before making a change.
Timing failure: corner, mode, WNS/TNS, path report.
Physical failure: DRC/LVS rule, cluster, owner, waiver policy.
Power failure: IR/EM metric, vector, hotspot, grid fix.