clarification about GRTC.SYSCOUNTER when in low-power sleep

[bios-bob]

i have a trivially simple "bare-metal" (register-level) test case in which i setup GRTC as follows:

MODE.SYSCOUNTEREN = 1
TASKS.START = 1
TIMEOUT = 2

i also have the LFXTAL configured as my LFCLK....

after this, i go into an *indefinite* sleep....  no GRTC wakeups have been scheduled....

all other things being equal, i find that this (benign) setup adds 0.4uA of *extra* current consumption when sleeping....  (2.1uA without, 2.5uA with)....

i've verified that no SYSCOUNTER[i] is ACTIVE.... 

i've "discussed" this with your AI (very helpful!!!), but it ultimately said we need to ask a human to clarify....  ;-)

[Håkon Alseth]

Hi,

 

If you write to any of the registers mentioned here, it will be active in sleep mode:

https://docs.nordicsemi.com/bundle/ps_nrf54L15/page/grtc.html#d763e406

 

all other things being equal, i find that this (benign) setup adds 0.4uA of *extra* current consumption when sleeping....  (2.1uA without, 2.5uA with)....

Which mode(s) are you testing?

The expected SystemOnIdle current, with full RAM retention, with- and without GRTC, is specified to be just below 3 uA (2.7 vs 2.9):

https://docs.nordicsemi.com/bundle/ps_nrf54L15/page/chapters/current_consumption/doc/current_consumption.html#ariaid-title4

 

Kind regards,

Håkon 

[bios-bob]

i'm not touching any of the GRTC registers which would make SYSCOUNTER active....  i've also verified that the ACTIVE bit is indeed 0....

what i recently discovered was that i was *not* setting up the CLKCFG register -- which by default is set to LFCLKSRC....  now i have setup the LFCLK to use the LFXTAL -- but apparently i should also explicitly set CLKCFG to LFXTAL as well....

while everything "works", i found that explicitly setting CLKCFG saved about 0.2uA when sleeping.... my test case is *only* retaining a single 16K SRAM block, which puts me closer to the 2.2uA range....

i'm trying to match the sleep current achived in nordic's submission to the BlueJoule benchmark (2uA), found at https://github.com/em-foundation/BlueJoule/blob/main/captures/nrf-54-dk/zephyr2-3V0-J/ABOUT.md

i (wrongly) presumed that the default setting for CLKCFG would be sufficient -- especially since i already bound LFXTAL as my LFCLK....  for whatever reason, it introduced about 0.2uA of additional sleep current....

i'm still not quite in parity with nordic's code; another 0.2uA to go....  

[bios-bob]

i'm at v0.9.2 -- but do have another board coming tomorrow....  we'll pick this up then....

[Håkon Alseth]

That sounds good. Let me know how it goes!

 

Kind regards,

Håkon

[bios-bob]

everything works as expected on the v1.0.0 DK....  my measurements are consistent with yours....

earlier in this issue, i identified an "extra" step needed to setup the CLKCFG register to use LFXTAL....  (i haven't verified whether this is "not necessary" on the more recent silicon -- but for the cost of one instruction, why bother....)

[Håkon Alseth]

Hi,

 

I am glad to hear that it runs as expected at this time. Let us know if you run into any issues or have questions in the future.

I wish you a great weekend!

 

Kind regards,

Håkon

[Emil Lenngren]

It is necessary to set the CLKCFG.CLKSEL register of GRTC to LFXO instead of SystemLFCLK to get the lowest power consumption, and that applies even if the LFCLK.SRC register is set to LFXO. Otherwise, the clock signal will take a longer trip through PCLK32KI which apparently costs extra energy, instead of using the direct path called "GRTC.LFXO". Not sure if this is documented anywhere but that's my observation at least.