How do you detect and undo an ECO that displaced too much?

Short Answer

After an ECO change list is applied and legalized, run report_eco_physical_changes (ICC2) to see how far each sized or added cell moved and how much its nets grew. Cells that landed far from their connections usually cost more delay than they saved. Undo them with revert_eco_changes -cells (ICC2), which removes the added buffer or inverter pair, or restores the original cell, location and orientation.

Technical Explanation

• Scope: report_eco_physical_changes (ICC2) covers changes made with size_cell (ICC2), add_buffer (ICC2) and add_buffer_on_route (ICC2). Changes from other editing commands, including insert_buffer (ICC2), are not reported. Synopsys recommends running it after ECO legalization or ECO routing.
• Without options it prints only a global summary: total ECO cells, maximum and average displacement, the number of cells outside their connection bounding box, and the ratio of ECO net route length to estimated length.
• -type cell_displacement lists each cell's original location, current location and displacement. The original location is where the cell was inserted or sized, or its pre-legalization location if place_eco_cells (ICC2) placed it. An asterisk marks a cell outside its connection box, and those cells are listed first. -min_displacement filters the list and defaults to 5.0 times the site height.
• -type net_length reports estimated length, actual route length and their ratio for each ECO net. -min_net_length_ratio defaults to 1.1, a real length 110% of the estimate.
• -type estimated_cell_displacement predicts displacement before legalization using placement resource feasibility analysis. Cells over -min_estimated_displacement go into the collection eco_estimated_large_displacement_cells. The default threshold is infinity, so the collection stays empty until you set one, and the prediction is optimistic.
• revert_eco_changes (ICC2) handles the same three commands. A new buffer is removed and the original connections restored; inverters are removed as a pair so the logic stays the same; a sized cell returns to its original reference, location and orientation.
• Reverting brings the original violation back. Fix it again another way: a smaller cell, a different pin, or physically aware fixing in PrimeTime that only uses space that exists.

What To Check

• Global summary: maximum and average displacement, and the count outside the connection box.
• The per-cell list above your threshold, starting with the asterisked cells.
• Net length ratio on ECO nets, especially on critical paths.
• Timing on the endpoints the reverted cells were fixing, after the revert.

Healthy, Suspicious & Hard-stop Results

• Healthy (illustrative): Maximum displacement 3 um, no cells outside their connection box, net length ratios under 1.1.
• Suspicious (illustrative): A few asterisked cells, or displacement above five site heights clustered in one dense region.
• Hard stop: ECO buffers 50 um or more from their driver-load box, or length ratios above 2 on critical nets. Revert and re-plan before routing.

What The Interviewer Is Testing

• Knows which ECO commands the report and the revert cover.
• Reads the per-cell list and the asterisk, not only the average.
• A strong answer plans a different fix for the reverted endpoints instead of stopping at the revert.

Follow-up Question & Model Response

"Why revert with the command instead of moving the cell back by hand?"

Candidate Model Response: Moving a cell only changes its location, while the buffer or resize stays in the netlist. revert_eco_changes (ICC2) undoes the logical change too: it removes the buffer and reconnects the net, removes inverter pairs together, and restores the original reference. After the revert, the layout netlist no longer matches the edited netlist PrimeTime still holds. Write the netlist out again and reload it in PrimeTime so the next round starts from what is really in the layout.