Why extract parasitics with real metal fill before final STA?

Short Answer

Metal fill sits beside and above your signal wires, so it adds capacitance that routing-stage extraction did not see. Before fill you estimate that effect with emulation TLUPlus or virtual metal fill; after fill, the real shapes are what will be on silicon. Final timing should use extraction that includes them, which in ICC2 means non-emulation TLUPlus plus set_extraction_options -real_metalfill_extraction floating (ICC2).

Technical Explanation

• Fill shapes are floating or grounded metal placed to meet density rules. Each one near a signal wire adds capacitance to it, which raises wire delay and changes coupling. The effect is largest on long nets on layers with dense fill.
• ICC2 reads two kinds of TLUPlus files. Emulation TLUPlus is meant for use before fill insertion; non-emulation TLUPlus can be used before and after, with fill accounted for through virtual metal fill before insertion or real metal fill after.
• -real_metalfill_extraction (ICC2) takes none, floating, grounded or auto. The default is none, which ignores fill capacitance; auto detects the fill type and applies it.
• The guide gives two steps. Associate non-emulation TLUPlus files with each timing corner through set_parasitic_parameters (ICC2), then set set_extraction_options -real_metalfill_extraction floating (ICC2). Each corner takes one early and one late model, and you may switch them after fill.
• Before fill, -virtual_metalfill_extraction (ICC2) with a VMF parameter file estimates fill without inserting it, and the guide's VMF examples set real fill extraction to none at the same time. Once real fill exists, switch to real fill extraction and stop relying on the estimate.
• Floating and grounded fill behave differently. Grounded fill adds capacitance to ground, while floating fill couples neighbouring conductors to each other through the fill. Use the model that matches how the foundry's fill is built, or auto.
• Signoff timing runs in PrimeTime on StarRC SPEF, and the StarRC command file has a matching metal fill polygon setting. The ICC2 setting keeps route_opt and ECO timing close to what PrimeTime will report, so the final run holds no surprise.

What To Check

• report_extraction_options (ICC2) shows the intended real fill setting.
• report_parasitic_parameters (ICC2) shows non-emulation TLUPlus on every corner after fill.
• Setup and hold change on the top critical paths between pre-fill and post-fill extraction.
• The signoff SPEF was extracted from the filled layout with fill handling enabled.
• Timing-driven fill kept fill away from the nets you marked critical, if you used it.

Healthy, Suspicious & Hard-stop Results

• Healthy (illustrative): Post-fill WNS moves by a few ps, for example +9 ps to +3 ps after timing-driven refill, and no endpoint turns negative.
• Suspicious (illustrative): A loss of 10 ps or more concentrated on long buses with dense fill, enough to turn endpoints negative, which calls for timing-driven fill or rerouting before signoff.
• Hard stop: Final STA run on a SPEF extracted without fill, or -real_metalfill_extraction (ICC2) still at its default of none after fill.

What The Interviewer Is Testing

• Understanding that fill adds capacitance and why emulation estimates must give way to real fill.
• Knowing the default for -real_metalfill_extraction (ICC2) is none.
• Linking this to signoff: the SPEF used in PrimeTime must come from the filled layout.

Follow-up Question & Model Response

"When would you use virtual metal fill instead?"

Candidate Model Response: Before fill exists, through route_opt and early ECOs, when you want timing to anticipate fill without inserting real fill on every iteration. set_extraction_options -virtual_metalfill_extraction floating -virtual_metalfill_parameter_file vmf.param (ICC2) models fill from a parameter file. It is an estimate, so once fill is inserted I switch to real fill extraction for the final numbers. The guide's VMF examples set -real_metalfill_extraction none (ICC2) at the same time.