What's the basic flow to get a signoff power number in PrimeTime?

Short Answer

Enable PrimePower with set_app_var power_enable_analysis true (PT), load the routed netlist with parasitics, constraints and the UPF, read switching activity with read_saif (PT) or read_vcd (PT), then run update_power (PT) and report_power (PT). The number is only as good as the activity behind it, so use a simulation of a real use case, not assumed toggle rates. Report by group to see where the power goes.

Technical Explanation

• License and switch: power analysis needs a PrimePower license, and power_enable_analysis (PT) must be true before any power command.
• Inputs: netlist, read_parasitics (PT) for wire capacitance, SDC for clocks, and for multivoltage designs the UPF with set_voltage (PT) values.
• Activity: SAIF carries average toggle rates per net; VCD carries the full time trace, which you need for peak or cycle-level power.
• Without activity: the tool relies on assumed toggle rates, fine for early estimates but not a signoff number.
• Groups: report_power -groups {register combinational sequential} (PT) splits internal, switching and leakage power per group.
• Clock pin power: with power_clock_network_include_register_clock_pin_power (PT) false, register clock-pin power is booked to the register, not the clock network.
• What breaks: a SAIF whose hierarchy does not match the netlist annotates little, and most nets fall back to default activity.

# [PrimeTime]  pt_shell
set_app_var power_enable_analysis true
read_verilog mychip.v
link_design
read_parasitics mychip.spef
read_sdc mychip.sdc
load_upf mychip.upf
set_voltage 1.0 -object_list [get_supply_nets VDD1p0]
read_saif video_decode.saif
update_power
report_power -groups {register combinational sequential}

What To Check

• Activity annotation covers most internal nets, not just primary inputs.
• The activity file comes from a named, realistic use case, not a reset-only test.
• Supply voltages match the power state you are reporting.
• Leakage is reported at the corner your spec names, often hot for standby.

Healthy, Suspicious & Hard-stop Results

• Healthy (illustrative): The SAIF annotates 97% of nets and total power lands at 180 mW, within 10% of the architecture estimate.
• Suspicious (illustrative): Only 40% of nets annotate: the rest run on default activity, so the total is part measurement, part guess.
• Hard stop: The number comes from a reset-only SAIF: 60 mW describes an idle chip and cannot sign off the use case.

What The Interviewer Is Testing

• Do you know the answer depends on activity more than on the netlist?
• Can you explain SAIF versus VCD and when each is enough?

Follow-up Question & Model Response

"When do you need a VCD instead of a SAIF?"

Candidate Model Response: SAIF holds average toggle counts and state probabilities, which is enough for average power over a use case. When you need power per cycle, such as peak power at wake-up or a current profile for dynamic IR, you need the time information in a VCD. VCDs are large, so teams dump a short, targeted window around the event. The rest of the PT flow stays the same apart from the read command.