How do you choose activity and corners for power signoff?

Short Answer

Match the activity source to the question you are answering: default toggle rates for early budgets, SAIF from realistic use-case simulation for average power, and a VCD window around the worst event for peak power. Then pick the corner that makes that number worst, which is the fast, high-voltage, hot corner for leakage and the highest mode voltage for dynamic power.

Technical Explanation

• Default or vectorless activity: available before testbenches exist and fine for floorplan budgets, but it can be far off on real workloads.
• SAIF: toggle counts and static probabilities per net over a simulation; the right input for average power in a named use case.
• VCD: cycle-by-cycle value changes; needed for peak and time-based power, and it captures glitches when the gate-level run is timed.
• Use-case coverage: simulate modes the product actually spends time in, such as idle, video and boot, and weight them by duty cycle for battery life.
• Corners: leakage rises steeply with temperature and voltage, so FF, high voltage and hot bound it; dynamic power scales with V squared.
• PrimePower flow: set_app_var power_enable_analysis true (PT), read_saif (PT) or read_vcd (PT), update_power (PT), then report_power (PT).
• What breaks: a signoff number from default toggle rates or one short directed test cannot be defended, and thermal or battery budgets fail later.

# [PrimeTime]  pt_shell
set_app_var power_enable_analysis true
read_saif video_1080p.saif
update_power
report_power

What To Check

• Each reported number names its purpose (budget, average or peak) and its activity source.
• SAIF annotation covers nearly all registers and clock nets; the rest fall back to default rates.
• Use cases are weighted by the time the product spends in them.
• Each number uses its own worst corner: FF, high voltage, hot for leakage; the mode's highest voltage for dynamic.

Healthy, Suspicious & Hard-stop Results

• Healthy (illustrative): Video use case: 410 mW average from SAIF, 97% of nets annotated, at the 0.9 V mode voltage.
• Suspicious (illustrative): Only 70% of nets annotated, so almost a third of the design runs on default toggle rates.
• Hard stop: The signoff number comes from default toggle rates, or leakage was reported only at the typical 25 °C corner.

What The Interviewer Is Testing

• Do you match activity format to the question: budget, average or peak?
• Do you know which corner is worst for leakage and which for dynamic power?

Follow-up Question & Model Response

"Why not use one long VCD for everything?"

Candidate Model Response: A VCD from a long gate-level run is huge and slow to read, and for average power it adds little over a SAIF of the same run. A SAIF cannot give peak power, because it stores totals and probabilities rather than when each toggle happened. So you use SAIF for averages over long use cases and a short VCD window for the worst burst. The skill is choosing that window, usually from a power-over-time profile of a cheaper run.