IntermediateQuestion 25 of 60

What is the difference between light-sleep, deep-sleep, and full shutdown retention states for on-chip memory, and what do they trade off?

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

Short Answer

Light sleep, deep sleep and shutdown are three increasingly deep memory power modes. Each one saves more leakage than the one before and takes longer to wake, and only shutdown loses the stored data. The right choice depends on how long the memory will stay idle compared with the energy and time it costs to wake it up.

Technical Reference DiagramWhat is the difference between light-sleep, deep-sleep, and full shutdown retention states for on-chip memory, and what do they trade off?

Technical Explanation

  • Light sleep: periphery clocks and wordlines are gated and parts of it biased down; the bit-cell array stays near full supply.
  • Deep sleep: periphery powered off; the array drops to a retention voltage just above its data-retention limit, so data survives.
  • Shutdown: array and periphery both off; leakage is lowest, but data is lost and must be reloaded or reinitialized.
  • Wake cost grows with depth: a few cycles for light sleep, microseconds for deep sleep, reload time for shutdown.
  • Each mode has a break-even idle time: below it, the wake energy costs more than the leakage saved.
  • Mode pins and exact behavior come from the memory compiler datasheet, so read it rather than assuming a standard.
  • Driving the mode pins from logic that is itself off leaves the memory in an undefined mode.
# Conceptual (not a tool command)
if idle_us < 2:
    stay_active()
elif idle_us < 50:
    enter_light_sleep()
elif idle_us < 20000:
    enter_deep_sleep()
else:
    save_needed_lines()
    enter_shutdown()

Common Mistake

The Trap: Always picking the deepest mode because it has the lowest leakage number.

  • For short idle gaps, the wake-up energy and the reload after shutdown exceed the leakage saved.
  • Average power goes up and the CPU stalls on every wake, which shows up as both a power and a performance bug.

Follow-up Question & Model Response

"Who controls these modes, hardware or software?"

Candidate Model Response: Usually both. A hardware power controller in the always-on domain drives the memory mode pins and handles the timing between steps. Software or firmware picks the target mode based on the expected idle time. The mode pins must come from always-on logic, because the controller has to wake the memory from a state where its own domain may be off. Isolation on the memory outputs also has to be active before the array or periphery drops, or downstream logic sees X values while the memory is asleep.

Practical Example

Design Scenario: (illustrative) A 512 KB L2 SRAM leaks 10 mW when active. Light sleep cuts that to 6 mW and wakes in 3 cycles; deep sleep cuts it to 2 mW and wakes in 1 µs; shutdown cuts it to 0.1 mW but reloading lines costs about 40 µJ. For a 1 ms screen-off gap, shutdown saves only about 2 µJ over deep sleep and pays 40 µJ to reload, so deep sleep wins. Shutdown pays off only for idle periods longer than about 20 ms. The firmware policy therefore uses light sleep between CPU bursts, deep sleep with the screen off, and shutdown only when the whole cluster powers down for standby. The mode pins are driven by the power controller in PD_MYCHIP, which stays on in all three cases.

Low-Power & UPF Handbook

Read the complete low-power guide library covering power domains, level shifters, isolation clamps, state retention, and UPF signoff verification.

Offline PDF Bundle

Want all 1109 questions offline?

Get the complete 4-book PDF bundle (PnR, STA, MMMC, Low Power) with a clickable table of contents - no ads, no internet needed.

See what's inside the bundle
Low-Power VLSI & UPF Handbook — nine chaptersLow-Power & UPFDomains, isolation, retention, and multivoltage UPF.