BeginnerQuestion 10 of 50

What is a power domain?

From PDVerse Low-Power Physical Design Mentor Guide, part of the pdVerse Mentor Guide

By Tabish Iqbal, who has 9+ years of VLSI Physical Design experience

Short Answer

A power domain is a group of instances that are treated as one unit for power management, and that normally share one primary supply. Everything in it powers up, powers down or changes voltage together. The domain is also where you attach the strategies for its boundary: isolation, level shifting and retention.

Chip outline with three domain boxes from the MYCHIP example: PD_CPU on VDD0p9, PD_COP on switched VDD1p0_SW and PD_DSP on its external supply, all inside PD_MYCHIP on VDD1p0, each box labelled with its primary supply. — Technical Reference DiagramWhat is a power domain?

Technical Reference DiagramWhat is a power domain?

Technical Explanation

Power domain: IEEE 1801 defines it as a collection of instances grouped for power management, typically sharing a primary supply set.
Every instance in the design belongs to exactly one domain; a top domain usually holds the always-on glue logic.
A domain can have extra supplies beyond its primary, such as an always-on supply for isolation or retention cells.
The domain is a logical object in the UPF, not a physical region; the floorplan region is a separate voltage area.
Several domains with equivalent primary supplies can share one voltage area: list them in the shared_voltage_area attribute with set_design_attributes (ICC2).
Get the domain boundaries wrong and every downstream strategy, switch and check is attached to the wrong logic.

Common Mistake

The Trap: Treating a power domain and a voltage area as the same thing, with a strict one-to-one mapping.

The domain is intent; the voltage area is floorplan. Mixing them up leads to wrong questions about where cells may go and which domains can share a region.
Check the domain list and the voltage-area list separately; one voltage area can hold several domains with equivalent primary supplies.

Follow-up Question & Model Response

"Why not just give every block its own domain?"

Candidate Model Response: Every domain boundary costs something: isolation or level-shifter cells on crossings, always-on control routing, extra checks and more power states to verify. A domain only pays off if its logic can be switched off or run at a different voltage for a real share of the time. Blocks that are always on at the same voltage belong in the same domain. Good partitioning follows use cases, not the RTL hierarchy.

Practical Example

Design Scenario: MYCHIP (illustrative voltages): PD_MYCHIP is the always-on top on VDD1p0 at 1.0 V. PD_CPU is always on at 0.9 V on VDD0p9. PD_COP can shut down, so its primary is the switched VDD1p0_SW, and it keeps state in retention flops. PD_DSP runs on its own supply VDDdsp, which is switched off chip. Four domains, four primary supplies, one power controller in PD_MYCHIP driving the control signals. Check the partition: every instance sits in exactly one domain, and each domain has exactly one primary supply.

Low-Power & UPF Handbook