After a metal-only ECO, what must be re-verified?

Short Answer

Incremental checks are fine for DRC and fill while you iterate, but connectivity is global, so LVS always runs in full. After the ECO, run incremental signoff DRC on the changed areas, remove and repair fill that now conflicts, rerun full LVS, and rerun IR and EM where current or metal changed. Before tapeout, run full-chip DRC on the final GDS, since incremental runs only look where things changed.

Technical Explanation

• signoff_check_drc -auto_eco true (ICC2) checks only the areas changed since the last run. It needs a previous signoff_check_drc (ICC2) run and a change below signoff.check_drc.auto_eco_threshold_value (ICC2), 20% by default.
• By default only violations in the changed areas are reported. Set signoff.check_drc.merge_incremental_error_data (ICC2) to true to merge them with the earlier results, so the report shows the whole block.
• ECO routes land on existing fill. signoff_create_metal_fill -remove_by_rule drc_auto (ICC2) removes track-based fill shapes that now cause DRC violations, including floating connections left behind.
• signoff_create_metal_fill -auto_eco true (ICC2) removes and redoes fill only in changed regions, with the same 20% threshold. A metal change also triggers the adjacent via layers so no floating vias remain.
• LVS runs in full. A single wrong via can merge or split nets far from the ECO area, and the extracted netlist has to match the final netlist from the ECO flow, including spare cells that were connected.
• IR and EM rerun where the ECO changed current or metal: connected spare cells draw current where there was none, and new routes may cut through or near PG straps. A local static and EM run on the region is the minimum; if a PG strap was touched, run the block.
• Isolated vias, antenna and density are part of the DRC deck, but check them explicitly if the ECO added vias or long routes. Timing and LEC are covered by the timing gate.

What To Check

• Incremental DRC merged with the baseline, not changed areas alone.
• Fill DRC and density in the ECO region after repair.
• Full LVS against the post-ECO netlist.
• IR and EM in the changed region, and whether any strap was touched.

Healthy, Suspicious & Hard-stop Results

• Healthy (illustrative): Incremental DRC clean on 0.8% changed area, fill repaired, full LVS PASS, IR in the region unchanged within 2 mV.
• Suspicious (illustrative): Incremental DRC clean but merge not enabled, so older violations elsewhere are not visible.
• Hard stop: LVS skipped because the ECO was small, or change above 20% forcing a full run that nobody scheduled.

What The Interviewer Is Testing

• Knowing which checks may be incremental and which may not.
• Knowledge of the 20% auto-ECO threshold and the merge option.
• Linking ECO changes to fill repair and IR.

Follow-up Question & Model Response

"Why is LVS never run incrementally in this flow?"

Candidate Model Response: LVS compares connectivity of the whole design. A short or open in the ECO area can change net identity anywhere those nets reach, and extraction has to rebuild connectivity from the full layout to see it. There is no safe changed-area boundary for a connectivity check the way there is for geometry. A full LVS run is also cheap compared with a respin, so there is no reason to cut it.