-
In search for the lowest power MCU. Is this .2nA microcontroller real?
Posted by
jpliew
on 03 Mar, 2024 04:04
-
Hi All,
I am a bit outdated with microcontrollers these days. Hope to get a bit of opinion from experts lurking here.
Was searching for a lowest sleep micro and found this link
https://www.microchip.com/en-us/solutions/low-power/8-bit-low-power-mcusThe site mentioned "
Sleep current down to .2 nA" but did not mention which micro.
Two months with their Client System Engineers leads to nothing.
Is this a real claim? Does anyone know which part number this refers to?
Cheers
JP
-
#1 Reply
Posted by
PCB.Wiz
on 03 Mar, 2024 04:45
-
The site mentioned "Sleep current down to .2 nA" but did not mention which micro.
0.2nA would be with no clocks at all running, and at 25'C - more a banner marketing number, than a real engineering number.
As a comparison, the SiLabs EFM8SB1 series claim 50nA no clocks and 300nA with LFO running.
-
-
Yes, true. After that, it all boils down to your exact use case. It's not a useful figure in general.
If your particular application can make use of the "sleep" mode that allows such low current for a significant duty cycle, then sure. (Although look exactly what you get here: I doubt that with 0.2nA the MCU can even retain the state of its GPIOs, so it would basically be like powering it off entirely. But looking at the DS closely will help figuring it out.)
In the end it's all a matter of duty cycle.
The absolute lowest current draw in *active* mode that I know of for now with ARM-based MCUs are the Ambiq Apollo MCUs. They're not even expensive for what they offer.
But in sleep mode(s), you may find other MCUs that draw less. Again, calculate your average consumption with your duty cycle(s) and figure out what figure you need to focus on.
As an obvious example, let's say your system is active 50% of the time and draws just 5 mA during that time, and can remain in a "sleep" mode the remaining 50% of the time.
If the sleep current is 0.2 nA, you have an average of 2.500 mA (keeping 3 decimals). If the sleep current is 2 µA, so 10,000 times more, you have an average of 2.501 mA. Yeah, ok - that's a difference of 0.04%. You get the gist. It will matter only if the duty cycle (active part) is very low.
-
#3 Reply
Posted by
jpliew
on 03 Mar, 2024 05:33
-
Thanks @PCB.Wiz and @SiliconWizard, agreed with your suggestions.
"But looking at the DS closely will help figuring it out."
more a banner marketing number
The thing that puzzled me is Microchip is not even able to tell me what that part number is. Even if it is a marketing number, without a part number, users can't even refer to the datasheet.
-
#4 Reply
Posted by
Psi
on 03 Mar, 2024 05:43
-
I've seen more than one datasheet or online example gets confused with 0.4uA / 400nA and instead writes it as 0.4 nA
-
-
The Mouser page states the same, "Sleep current down to 0.2nA",
https://www.mouser.com/new/microchip/microchip-8-bit-tools/without, of course, mentioning what model it could be.
I would also doubt the 0.2nA figure that is more likely to be 0.2µA, although I'd think they have MCUs with lower than 200nA, so, I don't know.
But 200pA looks extremely low indeed. Not sure how that could be achieved, even if powering off absolutely everything on the chip. Just the leakage from the internal power switches is likely to be more than this. Who knows.
-
#6 Reply
Posted by
Smokey
on 03 Mar, 2024 06:31
-
The AVR EA series claims 80nA (0.08uA). I'm pretty sure that's better than I can actually measure at the moment....
-
#7 Reply
Posted by
hans
on 03 Mar, 2024 08:53
-
200pA at 2V is 10G ohm resistance. That's well into the insulation resistance of many parts, not limiting to ceramic caps. The landing page of "low power" on Microchip.com only cites 9nA, so I would almost say this is a typo.. Maybe someone put an additive current for just some peripheral in there.. (like power-down +RTC +brown-out detector)
It's certainly not 0.2uA though. PIC16LF1503 is an older part and already has 20nA power-down current.
Regarding OP's question.. my first question is always: do you need low power or low energy (battery operation)?
They are not always the same. In particular low energy is an architecture problem, as many modern MCUs support a limited set of peripherals/operations in low power modes. It becomes real hard to do back-on-the-envelope calculations as estimations how much energy some project is going to consume.
I've looked at Ambiq micros before. Their CPU run modes are radically low power, but their peripheral set leaves a lot to be desired.
I'm now using a STM32U5 in a new design, as it supports on-demand clock requests to high-frequency peripherals, and DMA operations while the part is in a stop mode (sleep). MSP430 is also quite famous for supporting these power modes. And SiLabs also make quite a few..
But it depends on how much work you can put in for diminishing returns. I transitioned my design from a STM32L4 to STM32U5, and that lowered the average run power consumption from ~150uW to 70uW. Its a neat decrease, but given the amount of extra work put in, perhaps not worth it.
(But I'm in academia so there are plenty of other proportions out of whack
)
-
#8 Reply
Posted by
nali
on 03 Mar, 2024 10:04
-
The AVR EA series claims 80nA (0.08uA). I'm pretty sure that's better than I can actually measure at the moment....
That's the baseline figure i.e. complete power down. As soon as you add any peripheral that soon ramps up e.g. brownout (420nA) RTC (600nA) 32k OSC (270nA) Watchdog (270nA). I'm just starting a project with AVR64EA although I won't have a board for a little while yet and I'm going to struggle measuring sub-uA. Time for some new TE
-
-
Ultra-low sleep current alone is rarely useful as you will usually need a timer to wake periodically, and maybe a brownout detector to deal with situations where the power dips too low.
1uA is likely fine for anything with any sort of battery, so below that you're looking at energy-harvesting situations which will generally need bulletproof brownout.
When you wake, you'll need more power, so a capacitor will be necessary to provide it, and the leakage current of that could be orders of magnitude higher than the MCU's sleep current.
So the figure with some kind of oscillator and brownout detect is usually the most useful figure to compare devices.
-
-
Anything below a few tens of nA is not going to make any sense, even if you needed it, unless your resort to very expensive hardware, including the energy source, around the MCU. Even a conventional battery will leak more than this.
50 nA - 100 nA is still workable, anything below? Yeah.
-
#11 Reply
Posted by
BrianHG
on 03 Mar, 2024 21:46
-
-
#12 Reply
Posted by
Psi
on 04 Mar, 2024 04:20
-
There's plenty of MCU options for no-clock sleep around 0.3uA - 0.6uA with wake from GPIO
-
-
The site mentioned "Sleep current down to .2 nA" but did not mention which micro.
Two months with their Client System Engineers leads to nothing.
Is this a real claim? Does anyone know which part number this refers to?
Usually MCU consume 100-200 uA in active calculations and 1-5 uA in stand-by mode (waiting for event). There is also deep sleep mode which is actually RAM retention mode, it don't doing job and don't feed GPIO in that mode, just keep it's memory and waiting for interrupt to awake but even deep sleep usually requires 100-200 nA.
I suspect that MCU will be unable even keep RAM contents if it consume just 2 nA
If such mode really exists, this is probably some kind of deep sleep mode with no RAM retention. I think it just allows to awake MCU with some event, but it needs to start oscillator, reinitialize entire RAM and all peripherals.
-
#14 Reply
Posted by
brucehoult
on 04 Mar, 2024 21:11
-
I suspect that MCU will be unable even keep RAM contents if it consume just 2 nA
If such mode really exists, this is probably some kind of deep sleep mode with no RAM retention. I think it just allows to awake MCU with some event, but it needs to start oscillator, reinitialize entire RAM and all peripherals.
And you need to have saved the current state into flash or EEPROM before sleeping, and re-initialise from that.
Which you need to do from time to time anyway, if losing power is possible. Or lock up and watchdog resets you.
-
#15 Reply
Posted by
nimish
on 05 Mar, 2024 00:16
-
Be verrrry careful about what is actually running when manufacturers specify their power consumption numbers.
And don't trust em at all. Measure with something. There's all kinds of shenanigans they'll pull to lower it that basically turn the processor useless. Make sure the peripheral you want to use isn't only available at a certain DVFS state.
-
#16 Reply
Posted by
Psi
on 05 Mar, 2024 00:37
-
hehe yeah,
MCU can sleep at 50nA if you turn everything off, but there's no way to wake it back up.
But since you could actually configure the MCU like that technical that's the lowest possible sleep current it can do.
<Marketing team adds that to datasheet and engineers get overruled...>
-
#17 Reply
Posted by
wek
on 05 Mar, 2024 10:19
-
> MCU can sleep at 50nA if you turn everything off, but there's no way to wake it back up.
And that's still 2 orders of magnitude more than the 0.2nA from the opening post/MCHP's claim - which IMO is simply a typo and should've been uA.
Both are nasty reality: that physics rules, and that people make typos.
JW
-
#18 Reply
Posted by
jpliew
on 06 Mar, 2024 03:08
-
Thanks all for your replies.
I first contacted Microchip around early Jan, two months past and nothing from them other than generic response. It is actually quite disappointment that Microchip choosing to not confirming if it is a typo or they really have a real part number that can do that.
-
#19 Reply
Posted by
peter-h
on 13 Mar, 2024 15:12
-
If you are talking about "deep sleep" (or whatever) modes, then the question is meaningless because you can use an external MOSFET to turn off an arbitrary part of your circuit, and you will get "practically zero" current draw then
Certainly in the 10 nA area is possible, at room temp. Look up the Vgs=0 leakage of say a BSS84, at room temp.
The Q then becomes: how do you turn it back on, when the CPU is not running?
You need to build a low power timer. This can be an RC network, like a 100M resistor charging a capacitor (low leakage type; not trivial) or an oscillator running a 32768Hz watch crystal. This gets pretty low
https://ambiq.com/artasie-am1815I've done this a few times over the years.
But if you want a CPU that actually runs, no way to get the nA Icc area.
FWIW, 0.2nA for any kind of crystal-based timer is BS. No normal crystal will oscillate at such low power. In the past I used a 4000-series CMOS counter for delays with very low power but even that draws a lot more than this.
-
#20 Reply
Posted by
bson
on 25 Mar, 2024 19:28
-
And you need to have saved the current state into flash or EEPROM before sleeping, and re-initialise from that.
The MSP430FRxxxx devices would beg to disagree.
-
#21 Reply
Posted by
brucehoult
on 25 Mar, 2024 23:16
-
And you need to have saved the current state into flash or EEPROM before sleeping, and re-initialise from that.
The MSP430FRxxxx devices would beg to disagree.
On 2 nA? Link to device data sheet please.
-
#22 Reply
Posted by
rsjsouza
on 25 Mar, 2024 23:40
-
FRAM is non-volatile, so it features memory retention with zero power.
However, a device such as the MSP430FR2355 claims to consume 42nA in shutdown (probably with a small portion of the circuit still active to detect an external event), which is quite above the 0.2nA mark.
-
#23 Reply
Posted by
thm_w
on 26 Mar, 2024 01:02
-
-
-
battery technology is so good now you can do with alot more, IMO the ultra low power thing is not useful. A lithium coin has 200+mAH.
That is 2 years for 10uA. 20 years for 1uA.
I think people can afford a coin cell every couple of years
if you make it ridiculous you know your just gonna be fighting a dodgy battery. their chemical devices.
Better yet put a mechanical switch and forget about it, 1mA is fine
The consumer trend IMO from what i see on this forum is to just put bigger batteries in it because the small size aint worth a damn in most cases, hence the mods to power stuff of 18650s.
No one else noticed that the ultra low power stuff seems to break before you change the first battery anyway?? Either it uses so much power during daily use (car key) it don't matter or its such a piece of shit that it breaks anyway, or gets lost too.
And if its something important people are gonna change the batteries out ANYWAY because their not exactly trusted. Like I would leave a battery in for more then 2 years on a critical device like a car key. Run all the diagnostics you want, the cells just cant be trusted. One day your just gonna think "oh shit when is the last time I did this" ? then go to the pharmacy and have the best peace of mind for about 3$.
-
-
But really, any kind of battery, even the best one, will have more self-discharge than a few nA. That makes no sense.
-
#26 Reply
Posted by
hans
on 26 Mar, 2024 07:32
-
I think people can afford a coin cell every couple of years
[....]
No one else noticed that the ultra low power stuff seems to break before you change the first battery anyway??
A lot of electronics is embedded => digital => gets outdated even if it still works. Even in the space of RF where everything moves a bit slower, new regulations will require parts to require to use less transmit power, bandwidth, shorter duty cycles, tighter channel integrations etc. Old protocols will be phased out, don't scale properly, etc.
That makes 'ultra low power' IMO more about not being wasteful. Another aspect of not wanting to replace batteries is the effort and irregularity to do so. Certainly its not the first thing you think about if something stops working after 8yrs..
(my 2 cents after working on battery-free stuff for a few years)
-
#27 Reply
Posted by
Berni
on 26 Mar, 2024 08:08
-
You can make any microcontroller this low power by adding a P-FET on the Vcc rail. When you want to go to sleep let go of the gate, the FET turns off and cuts away your power supply. So your power consumption is now whatever the leakage of the FET is (and that can be in sub nA) even if the lowest power mode on your MCU is a huge >10mA
To be fair this is not much different than that lowest power state on MCUs where pretty much the whole MCU goes dead and waits for a outside signal to wake up. So for this reason it is absolutely useless to look at the lowest possible numbers. You can get the same functionality using an external FET,
The interesting part for low power MCUs is how many features of the MCU continue working in the really low power sleep modes. Like timer wakeup, brownout detection etc.. Also a big factor is how fast it can wake up. It doesn't matter how low power your sleep mode is if you burn up 99% of the power during the wakeups that take too long to do something useful and return to sleep.
-
-
Off-the-shelf discrete MOSFETs that leak < 1nA are not that easy to find. Really. And you need to read the fine prints and test it - usually that degrades rapidly with temperature, for instance.
With that said, yes, unless you have an extremely low duty cycle, what will matter is how fast you can wake up and what keeps running in this low power state. For that, ESP32 in general are hopeless, as an example.
But again, even so, I still fail to see with what kind of power source a difference of say 1nA or less will matter at all. If anyone can point me to an example of such power source, I'm a client!
-
#29 Reply
Posted by
nimish
on 26 Mar, 2024 17:26
-
But really, any kind of battery, even the best one, will have more self-discharge than a few nA. That makes no sense.
Back of the envelope:
300mAh coin cell that loses 0.5%/month is 1.5mAh/(30*24h) ~ 2uA drain
But for those long-life applications you use LiSOCl2 and
https://www.fdk.com/battery/lithium_e/coin_primery/ says they do 1% a _year_, which is closer to ~350nA.
So, a few nanoamps here and there is unlikely to do much unless you're using the smallest of cells, where the self discharge current starts to be on the order of tens of nanoamps.
-
#30 Reply
Posted by
bson
on 26 Mar, 2024 18:57
-
And you need to have saved the current state into flash or EEPROM before sleeping, and re-initialise from that.
The MSP430FRxxxx devices would beg to disagree.
On 2 nA? Link to device data sheet please.
All you need is a power switch and it's 0nA. Flip the switch, initialize, and continue on - no need to "save the current state into flash or EEPROM", no need to sleep either, just power off.
-
#31 Reply
Posted by
Berni
on 29 Mar, 2024 18:11
-
Yeah going down very low means looking for some more specialized FETs, A lot of BJTs have even lower leakage. But most transistors in general are not specifically rated for it in the datasheet. And even when using a special low leakage transistor, something as simple as not properly cleaning the assembled PCB in clean methanol might be causing more leakage than the transistor.
But as said above A few nanoamps up or down is not really going to make a difference when most of the power sources like batteries or supercaps will likely have a higher self discharge rate than that. Besides if you use one of these super low power modes where the MCU is basically dead, you will need some extra circuitry that wakes it up, so you have to budget in the power draw of that.
When we had to make a annoyingly power hungry ESP32 run for days off a supercap we ended up cutting off its power supply and then having an external low power RTC wake it up every so often. Worked great.