Low Power / UPF · All levels

Retention Register Design

Isolation & Retention: Retention flops preserve selected state through power collapse by copying data into retention latches powered by an always-on supply.

What this topic teaches

Retention Register Design translates low-power intent into release-grade evidence. Retention flops preserve selected state through power collapse by copying data into retention latches powered by an always-on supply. The practical challenge is proving policy correctness under real transitions and ensuring each owner closes their layer without semantic drift.

The senior-engineer question

When retention coverage %, restore mismatch count, and retained state corruption escapes regresses, can you identify the failing transition, policy owner, implementation evidence, and minimum regression that proves closure?

diagram

POWER INTENT FLOW — Retention Register Design

architecture intent
      |
      v
UPF objects (domain/supply/state/strategy)
      |
      v
RTL + synthesis + PD interpretation
      |
      v
verification + signoff evidence
      |
      v
release decision

Primary metric: retention coverage %, restore mismatch count, and retained state corruption escapes

Picture the low-power flow

Start with domain/state diagrams before diving into logs. These are the models to sketch in design and interview reviews.

Retention datapath

diagram

RETENTION FLOP MODEL

functional FF ----save----> retention latch (AON rail)
functional FF <---restore-- retention latch

retention_supply must remain alive through collapse.

Domain map

diagram

POWER DOMAIN MAP — Retention Register Design

  +------------------- always_on -------------------+
  | PMU / control / retention rail                  |
  +---------+----------------------------+----------+
            |                            |
   +--------v---------+         +--------v---------+
   | domain_A         |         | domain_B         |
   | switchable       |<------->| switchable       |
   | iso/ret controls | crossings require policy   |
   +------------------+         +------------------+

Domain boundaries are policy boundaries, not drawing boundaries.

State transition path

diagram

STATE TRANSITIONS — Retention Register Design

ON  --save-->  RETENTION  --off-->  OFF
 ^               |                    |
 |               +----restore<--------+
 +--------------------power_up--------+

Guard checks:
  - isolation asserted before OFF
  - restore before functional traffic
  - reset policy consistent with retained state

Ownership layers

diagram

LP OWNERSHIP LAYERS — Retention Register Design

layer                owns                          typical failure
------------------   ---------------------------   --------------------------
architecture         domain strategy               infeasible power states
UPF intent           policy objects + bindings     wrong/missing policy
implementation       LP cell insertion/routing     illegal physical behavior
verification         transition scenarios          uncovered LP bug
signoff governance   waiver + release decisions    late escape to silicon

Evidence to collect

Primary metric: retention coverage %, restore mismatch count, and retained state corruption escapes.
Primary artifact: retention list, save/restore controller spec, and retention verification report.
Owners to bring into review: RTL owner, UPF owner, design verification owner.
One failing transition timeline with state markers and control signals.
One report snippet proving policy intent versus implementation behavior.

Ownership map

diagram

OWNERSHIP MAP — Retention Register Design

artifact              owner
----------------      -----------------
intent policy       RTL owner
implementation      UPF owner
verification        design verification owner

Escapes happen when ownership is implicit.

Subpages in this topic

Each topic is taught across mechanism, inputs/outputs, reports, debug, worked example, pitfalls, interview, checklist, theory, design space, expanded case study, walkthrough, comparison matrix, software view, and silicon impact.

Key takeaways

Always name the transition context for every low-power metric.
Bind each policy decision to a specific owner and artifact.
Re-run LP simulation, formal, and implementation checks after changes.

Common pitfalls

Treating static pass reports as transition closure.
Fixing symptoms without checking policy binding and sequence order.
Shipping with ambiguous ownership on open LP violations.

Low-power deep dive

Isolation and retention are sequence-sensitive contracts, not static checkboxes.

Concept diagram

diagram

SEQUENCE

save -> isolate -> power-off -> power-on -> restore -> release

Metric graph

diagram

TRANSITION DEFECT MIX

polarity errors      ███████
late iso assertion   █████
restore ordering     ████

Reports and artifacts

isolation crossing report
retention list
save/restore timing
X-propagation log

Mini case study

A late isolation enable caused intermittent X leaks only on one PMU path.

Debug branches

Validate control polarity
Trace AON source
Replay transition waveform

Senior review question

Ask: what transition evidence proves this topic is closed, and which owner signs it?

Key takeaways

State transition context must accompany every low-power metric claim.
Intent changes require simulation, formal, and implementation re-validation.

Common pitfalls

Comparing results from mismatched UPF revisions.
Assuming static checks replace transition validation.
Shipping with aged waivers and unclear ownership.

Execution drill pack 1

Use this pack to rehearse low-power closure on low-power/isolation-retention/retention-register-design: transition framing, policy ownership, implementation evidence, and release confidence.

Transition checklist

State transition explicitly named with legal source/target states.
Crossing and domain ownership are mapped and agreed.
Policy controls are traced to always-on source logic.
Waveform bookmarks align controls with state timestamps.

Review prompts

Which policy object is first to deviate from intent?
Which owner can apply the smallest reversible fix?
What regression matrix proves no collateral damage?
Which waiver conditions would still block release?

Evidence capsule

diagram

LP EVIDENCE CAPSULE 1

PATH: low-power/isolation-retention/retention-register-design
STATE WINDOW: <from -> to>
POLICY OBJECT: <isolation / retention / shifter / switch>
OWNER: <name>
PRIMARY ARTIFACT: <report/waveform/formal result>
RELEASE DECISION: <close / bounded waiver / escalate>

Execution drill pack 2

Use this pack to rehearse low-power closure on low-power/isolation-retention/retention-register-design: transition framing, policy ownership, implementation evidence, and release confidence.

Transition checklist

State transition explicitly named with legal source/target states.
Crossing and domain ownership are mapped and agreed.
Policy controls are traced to always-on source logic.
Waveform bookmarks align controls with state timestamps.

Review prompts

Which policy object is first to deviate from intent?
Which owner can apply the smallest reversible fix?
What regression matrix proves no collateral damage?
Which waiver conditions would still block release?

Evidence capsule

diagram

LP EVIDENCE CAPSULE 2

PATH: low-power/isolation-retention/retention-register-design
STATE WINDOW: <from -> to>
POLICY OBJECT: <isolation / retention / shifter / switch>
OWNER: <name>
PRIMARY ARTIFACT: <report/waveform/formal result>
RELEASE DECISION: <close / bounded waiver / escalate>

Execution drill pack 3

Use this pack to rehearse low-power closure on low-power/isolation-retention/retention-register-design: transition framing, policy ownership, implementation evidence, and release confidence.

Transition checklist

State transition explicitly named with legal source/target states.
Crossing and domain ownership are mapped and agreed.
Policy controls are traced to always-on source logic.
Waveform bookmarks align controls with state timestamps.

Review prompts

Which policy object is first to deviate from intent?
Which owner can apply the smallest reversible fix?
What regression matrix proves no collateral damage?
Which waiver conditions would still block release?

Evidence capsule

diagram

LP EVIDENCE CAPSULE 3

PATH: low-power/isolation-retention/retention-register-design
STATE WINDOW: <from -> to>
POLICY OBJECT: <isolation / retention / shifter / switch>
OWNER: <name>
PRIMARY ARTIFACT: <report/waveform/formal result>
RELEASE DECISION: <close / bounded waiver / escalate>

Execution drill pack 4

Use this pack to rehearse low-power closure on low-power/isolation-retention/retention-register-design: transition framing, policy ownership, implementation evidence, and release confidence.

Transition checklist

State transition explicitly named with legal source/target states.
Crossing and domain ownership are mapped and agreed.
Policy controls are traced to always-on source logic.
Waveform bookmarks align controls with state timestamps.

Review prompts

Which policy object is first to deviate from intent?
Which owner can apply the smallest reversible fix?
What regression matrix proves no collateral damage?
Which waiver conditions would still block release?

Evidence capsule

diagram

LP EVIDENCE CAPSULE 4

PATH: low-power/isolation-retention/retention-register-design
STATE WINDOW: <from -> to>
POLICY OBJECT: <isolation / retention / shifter / switch>
OWNER: <name>
PRIMARY ARTIFACT: <report/waveform/formal result>
RELEASE DECISION: <close / bounded waiver / escalate>

Execution drill pack 5

Use this pack to rehearse low-power closure on low-power/isolation-retention/retention-register-design: transition framing, policy ownership, implementation evidence, and release confidence.

Transition checklist

State transition explicitly named with legal source/target states.
Crossing and domain ownership are mapped and agreed.
Policy controls are traced to always-on source logic.
Waveform bookmarks align controls with state timestamps.

Review prompts

Which policy object is first to deviate from intent?
Which owner can apply the smallest reversible fix?
What regression matrix proves no collateral damage?
Which waiver conditions would still block release?

Evidence capsule

diagram

LP EVIDENCE CAPSULE 5

PATH: low-power/isolation-retention/retention-register-design
STATE WINDOW: <from -> to>
POLICY OBJECT: <isolation / retention / shifter / switch>
OWNER: <name>
PRIMARY ARTIFACT: <report/waveform/formal result>
RELEASE DECISION: <close / bounded waiver / escalate>

Execution drill pack 6

Use this pack to rehearse low-power closure on low-power/isolation-retention/retention-register-design: transition framing, policy ownership, implementation evidence, and release confidence.

Transition checklist

State transition explicitly named with legal source/target states.
Crossing and domain ownership are mapped and agreed.
Policy controls are traced to always-on source logic.
Waveform bookmarks align controls with state timestamps.

Review prompts

Which policy object is first to deviate from intent?
Which owner can apply the smallest reversible fix?
What regression matrix proves no collateral damage?
Which waiver conditions would still block release?

Evidence capsule

diagram

LP EVIDENCE CAPSULE 6

PATH: low-power/isolation-retention/retention-register-design
STATE WINDOW: <from -> to>
POLICY OBJECT: <isolation / retention / shifter / switch>
OWNER: <name>
PRIMARY ARTIFACT: <report/waveform/formal result>
RELEASE DECISION: <close / bounded waiver / escalate>

Execution drill pack 7

Use this pack to rehearse low-power closure on low-power/isolation-retention/retention-register-design: transition framing, policy ownership, implementation evidence, and release confidence.

Transition checklist

State transition explicitly named with legal source/target states.
Crossing and domain ownership are mapped and agreed.
Policy controls are traced to always-on source logic.
Waveform bookmarks align controls with state timestamps.

Review prompts

Which policy object is first to deviate from intent?
Which owner can apply the smallest reversible fix?
What regression matrix proves no collateral damage?
Which waiver conditions would still block release?

Evidence capsule

diagram

LP EVIDENCE CAPSULE 7

PATH: low-power/isolation-retention/retention-register-design
STATE WINDOW: <from -> to>
POLICY OBJECT: <isolation / retention / shifter / switch>
OWNER: <name>
PRIMARY ARTIFACT: <report/waveform/formal result>
RELEASE DECISION: <close / bounded waiver / escalate>

Practice this lesson