ExpertQuestion 155 of 161Source: Ansys RedHawk User Manual: Low Power Design Analysis (Ramp-up Analysis, Power Gating Results); Synopsys ICC2 Implementation User Guide: RedHawk and RedHawk-SC Fusion; Synopsys ICC2 Tool Commands: connect_power_switch

What IR problems are specific to power-gated domains?

From PDVerse PnR Interview Handbook · pdVerse Mentor Guide

By Tabish Iqbal · 9+ Years VLSI Physical Design Experience

Short Answer

A gated domain has two IR problems ungated logic does not. When on, current flows through the switch cells, so their on-resistance adds drop between the always-on supply and the virtual supply. At wake-up, the whole domain capacitance charges at once and the rush current can pull down the always-on rail that neighbouring logic depends on. Switch sizing trades the first against the second, and daisy chaining spreads the turn-on over time.

A waveform of switch current and virtual supply voltage during wake-up, comparing all switches turning on at once with a daisy-chained turn-on, plus a small schematic of header switches between the always-on supply and the virtual supply with their acknowledge chain. — Technical Reference DiagramWhat IR problems are specific to power-gated domains?

Technical Explanation

Switch drop in the on state is switch current times on-resistance. More or larger switches lower it but add off-state leakage and area and make wake-up current steeper.
Rush current is the charge needed to bring the domain's cell and decap capacitance from its off voltage to the supply, delivered in however long the switches take to turn on. Turning them all on together gives a large step in current on the always-on rail.
Daisy chaining turns switches on in sequence through an acknowledge chain. connect_power_switch -mode daisy (ICC2) connects the control pins that way. A longer chain means gentler rush current but slower wake-up.
Ramp-up analysis runs in standalone RedHawk with setup analysis_mode lowpower (RH) and perform analysis -lowpower (RH). It needs switch models in SWITCH_MODEL_FILE (RH), switch-related nets in VDD_NETS and GND_NETS (RH), the powering-up instances in the GSC file, a PWL capacitance file or RAMPUP_OFFSTATE_VOLTAGE (RH), and STA timing windows for the switch control pins.
RedHawk Fusion does not support inrush current analysis, so this step needs standalone RedHawk and an Ansys license.
Outputs: virtual_domain_total_i.rpt (RH) gives total ramp-up current per switched domain, virtual_domain_worst_v.rpt (RH) the worst node voltage during ramp-up, and switch_dynamic.rpt (RH) peak current per switch. Static analysis writes switch_static.rpt (RH) with the voltage across each switch.
In ICC2, the rail instance switch cell map needs dynamic analysis and switch model files; otherwise switch cells are treated as black boxes.

What To Check

Voltage across the switches in the on state, from switch_static.rpt (RH).
Peak rush current and the always-on rail droop during wake-up.
Wake-up time against the power-management specification.
Switches that saturate or stay off at the end of the ramp-up simulation.

Command Checks & Actions

ICC2 (icc2_shell)connect_power_switch -source pmu/sw_en -port_name sw_en -mode daisy -ack_out pmu/sw_ack -ack_port_name sw_ack

Connects switch control pins as a daisy chain so turn-on is staggered (names illustrative).

ICC2 (icc2_shell)analyze_rail -voltage_drop static -nets {VDD VSS}

On-state IR including the drop across the switches.

RedHawk (redhawk)setup analysis_mode lowpower

Standalone RedHawk low-power mode, the recommended first step of a ramp-up run; it enables low-power run-time optimizations.

RedHawk (redhawk)perform analysis -lowpower

Ramp-up (rush current) simulation of the switched domain.

Healthy, Suspicious & Hard-stop Results

Healthy (illustrative): Switch drop 14 mV in the on state, always-on rail droop 22 mV during wake-up, and wake-up inside the 2 us specification.
Suspicious (illustrative): Rush current meets its limit only because the chain is so long that wake-up misses its specification.
Hard stop: Always-on rail droop during wake-up large enough to upset neighbouring logic, or no ramp-up analysis at all on a gated design.

Common Mistake

The Trap: Sizing switches only for on-state IR and never simulating wake-up. The domain powers up in 20 ns with every switch on, the always-on rail dips well past its budget, and a neighbouring always-on block loses state the first time the domain wakes.

What The Interviewer Is Testing

Can the candidate name both problems, on-state switch drop and wake-up rush current?
Understanding the daisy-chain trade between rush current and wake-up time.
Knowledge that inrush analysis needs standalone RedHawk.

Follow-up Question & Model Response

"How would you reduce rush current without lengthening wake-up much?"

Candidate Model Response: Turn on a small set of weak switches first to charge the domain slowly, then enable the main switches once the virtual supply is close to its final value. The first stage limits the current step, and the second stage lowers on-state resistance quickly. Check with ramp-up analysis that the handover point avoids a second current spike.

Practical Example

Tapeout Scenario: A gated domain holds about 2 nF of cell and decap capacitance and charges to 0.8 V, so it needs 1.6 nC (illustrative). With all 400 switches on together in 20 ns, average rush current is 80 mA and the always-on rail dips 58 mV against a 40 mV limit. Daisy chaining in eight groups spreads charging over 400 ns: average current falls to 4 mA and the dip to 22 mV. Wake-up takes 0.5 us against a 2 us specification.

PnR Flow Mentor Guide

Master the Physical Design Implementation Flow

Read the complete 8-chapter PnR Flow Mentor Guide free on the web — library setup through placement, clock tree synthesis, routing, chip finishing, hierarchical implementation, and ECO, all the way to stream-out.

Explore PnR Flow Mentor Guide →