What is threshold-voltage-based filler insertion?

Short Answer

It is a filler flow in which ICC2 picks the filler for each gap from the threshold-voltage types of the cells on its left and right. You label VT types with set_cell_vt_type (ICC2), write rules with set_vt_filler_rule (ICC2) and insert with create_vtcell_fillers (ICC2). The guide says this flow is typically used only for established foundry nodes; the other method is the standard create_stdcell_fillers (ICC2) flow.

Technical Explanation

• Each library cell gets a VT type label. set_cell_vt_type -lib_cells /_vtA -vt_type vtA (ICC2) tags every cell whose name ends in _vtA. A cell you never label counts as the keyword default.
• A rule names a pair of VT types and a filler list, for example set_vt_filler_rule -vt_type {vtA vtA} -filler_cells {myLib/fillerA2x myLib/fillerA1x} (ICC2). The pair is the cells to the left and right of the gap. List fillers from largest to smallest, because the tool tries them in the order given.
• The reason for the flow is implant layers. A cell's VT is set by implant shapes, and implants have their own width and spacing rules, so a filler whose implant matches its neighbours keeps the implant continuous across the gap instead of leaving a thin sliver or a notch.
• Mixed boundaries need their own rules. A gap between a vtA cell and a default cell is a different pair from vtA and vtA, so write a rule for every neighbour combination the placement actually contains and verify the result.
• Run check_legality (ICC2) first; the flow expects a legal placement. By default create_vtcell_fillers (ICC2) fills all empty space in the horizontal rows of the whole block. -region (ICC2) and -voltage_area (ICC2) restrict it, and -prefix (ICC2) tags a run so its fillers can be selected later.
• Instances are named xofiller!library_cell!number, with ! as the default separator, which -separator (ICC2) changes. To remove them all, use remove_cells [get_cells xofiller*] (ICC2); the guide describes the targeted remove_stdcell_fillers_with_violation (ICC2) for the standard flow.
• create_vtcell_fillers -clear_vt_information (ICC2) deletes the rules and the VT annotations and inserts nothing. Run it before redefining rules so old labels do not linger.

What To Check

• Every library cell carries the intended VT label; an unlabelled cell silently becomes default.
• A rule exists for every neighbour pair present in the placement, mixed pairs included.
• Filler lists in each rule run from largest to smallest.
• check_legality (ICC2) is clean before insertion.
• Signoff DRC shows no implant width or spacing errors at filler boundaries.

Healthy, Suspicious & Hard-stop Results

• Healthy (illustrative): No empty sites left, every filler matching the rule for its neighbours, and signoff DRC clean on the implant layers.
• Suspicious (illustrative): Implant errors clustered next to one cell family, which usually means that family has no VT label or one neighbour pair has no rule.
• Hard stop: Implant spacing errors along whole rows after a library update renamed cells, so the /_vtA pattern no longer matched and every former vtA cell became default.

What The Interviewer Is Testing

• Knowing the rule is keyed on the left and right neighbours, not on the filler alone.
• Mentioning labelling, filler order from largest to smallest, and legality first.
• Knowing this is an established-node flow and that the standard filler flow is the alternative.

Follow-up Question & Model Response

"When would you pick the standard filler flow instead?"

Candidate Model Response: The standard create_stdcell_fillers (ICC2) flow uses the placement legalizer, so fillers honour advanced-node placement rules, blockages and keepout margins. It also supports leakage-driven selection with -leakage_vt_order (ICC2) and decap mix control with -type_utilization (ICC2). The VT-rule flow is simpler and predictable when the main concern is matching implants between neighbours. I follow whichever flow the foundry reference methodology uses for that node.