low power microcontrollers, RTCs, ePaper and solar cells

[FrankBuss]

As a sub-project of my Nixie clock, I plan to use a battery backup real time clock (or gold cap backup). First I thought about the DS1307, but it needs 500 nA and the STM32L4 series claims 200 nA with RTC running. This makes me wonder, which CPU has the lowest power consumption?

In general the question doesn't make sense, because if no oscillator for the RTC runs, it is of no use for my application. Is there an industry benchmark to compare different microcontrollers better? Would be nice to compare the power consumption for different tasks, like implementing a real time clock with some given accuracy like 20 ppm from 0°C to 60°C (doesn't need an external watch crystal, if the microcontroller can do the accuracy internally) and wake up on some pin interrupt, and another task with wake up each second, and so on. The microcontroller manufacturers could implement and program their MCUs to solve the benchmark tasks, so that we can be sure that it is the best possible solution. Or someone like EEVblog could organize a challenge.

BTW, Microchip lists ATmega MCUs now, too for their XLP technology. Of course, they still use the old Atmel datasheets and nowhere in it is XLP mentioned. I love marketing

[TJ232]

The lowest power consumtion RTC that I ever used, PCF8523, 150 nA: http://www.nxp.com/products/signal-chain/real-time-clocks/ic-real-time-clocks-rtc/real-time-clock-rtc-and-calendar:PCF8523

UPDATE: found it, I was missing the bookmark, this one seems to be the one with the lowest power consumtion :
http://www.microcrystal.com/index.php/news/25-cat-news/cat-articles/186-x-treme-low-power-real-time-clock-module-from-micro-crystal

• 60 nA at VDD = 3.0 V in crystal mode
• 17 nA in calibrated RC mode
• Integrated 32.768 KHz quartz crystal

Sounds nice.

[brucehoult]

RV-1805-C3 and AM0805/AM0815 seem pretty similar.

RV-1805-C3  60 nA full crystal mode, 17 nA in "crystal calibrated RC" mode.

AM0805/AM0815 55 nA in full crystal mode, 22 nA  in "crystal calibrated RC" +/- 2 PPM mode, 14 nA in pure RC +/- 16 PPM mode.

[amyk]

The lowest power MCU would probably be one of those mask ROM 4-bit ones that can be found in various battery-powered electronics.

[FrankBuss]

Thanks, the ambiq micro devices looks good. The AM18X5 have even additional power management functions, like an 1 ohm switch, and a watchdog timer. And I like that it is available with a SPI interface. The microcrystal part has only I2C, which kind of defeats low power solutions, if you want low power when you read or write to it, too. Not a problem for my Nixie clock, but maybe I will use such a RTC for other projects as well.

But it was difficult to find a distributor who sells this chip. Future Electronics sells it in batches of 3,000. But the ambiq micro homepage lists Abracon as "RTC resellers". And looks like this is the same chip, it is just called AB1815 instead of AM1815: http://www.digikey.de/short/32d7pj

The 3x3 mm QFN-16 package is not easy to hand solder, but possible. Someone did already a breakout board for it:

https://lowpowerlab.com/forum/projects/ab18x5-breakout-board/

I think it should be combined with a gold cap like this:

https://www.digikey.de/short/32d7h2

The maximum allowed discharge current is 1 mA. Is this the same for the charging current? Then it could be charged in an hour and last for months (if the self discharge of the capacitor is very low).

[splin]

Well your original question was "what is the lowest power microcontroller" not an RTC. As far as I am aware, Ambiq's Apollo 1 ARM processor is the lowest at 35uA/MHz @ 3.3V when running Coremark, 419nA deep sleep mode with 32768Hz crystal - but good luck actually getting one.

But today, look what just popped into my email intray - a link to AMD's new ultra low power ARM M4F microcontroller

http://www.analog.com/en/products/processors-dsp/microcontrollers/ultra-low-power-microcontrollers/aducm4050.html#product-overview

Don't know how long it willl be before you can buy one but the specs are very promising - 40uA/MHz and 246nA in shutdown mode, RTC enabled. That compares to aound 110uA/MHz for the STM32L4.

Lets hope they can meet the target performance spec and don't decide to put a huge premium pricetag on it as it could be a real killer in the ultra low power battery powered sector.

[kwass]

Have you looked into the shelf life (battery self discharge) or leakage current (capacitor) for your backup power source.  Almost certainly these will be the limiting factors with an ultra-low power micro or clock chip.    With a CR2032 cell, for example, if your circuit draws less than around 3uA the 2032 shelf life will be the limiting factor -- 10 years or so.

[splin]

Have you looked into the shelf life (battery self discharge) or leakage current (capacitor) for your backup power source.  Almost certainly these will be the limiting factors with an ultra-low power micro or clock chip.    With a CR2032 cell, for example, if your circuit draws less than around 3uA the 2032 shelf life will be the limiting factor -- 10 years or so.

But he didn't say he was space limited. I tested an 11 year old Pansonic D cell which had been partially used in a torch sometime previously. At 82mA discharge it delivered nearly twelve amp hours and almost 15Wh which was very impressive given I don't know how much had been used previously.

This compares to the spec of 16Ah when new! However it has to be said that probably half the unused batteries from the same batch had leaked quite badly. These batteries were made in Japan and had been stored in a wardrobe so probably mostly at 10C to 15C; there's no guarantee that modern cells made in China would have such good self discharge performance.

The Energizer alkaline handbook shows a graph of self discharge of AA cells with 20% loss of capacity after 10 years @ 20C and only 5% loss after 14 years at 0C.

Leakage seems to be a much bigger problem than self discharge (in cooler climates at least) so if very long lifetime is required then perhaps a number of cells from different manufactures could be used in parallel using transistors/low drop diodes. The wires would be soldered/welded to the cells and the whole battery put in a plastic bag to contain any leakage.

[mikeselectricstuff]

If it's a nixie clock, it will be mains power, so the backup battery will only be used when power is lost, so power draw is not a big deal. A supercap may be a better solution for longest product lifetime

[hans]

Similar to Coremarks (which is set to compete with Dhrystone benchmark), EEMBC also has ULPBench:
http://www.eembc.org/ulpbench/

Ambiq micro's do very well. Unfortunately can't buy them from Farnell, and that's where the options end for me as a hobbyist.

To be honest quite amazed that Silicon Labs Energy Micro's EFM32 are not higher up the list. PIC24 XLP is mediocre at best apparently.

The list is quite shallow though. Testing low power stuff is quite hard to do right, and also quite touchy. I imagine they have certified hardware and tests before the allow a score onto the sheets, probably much harder than it is for a Coremarks benchmark.

I'm not sure what kind of test conditions are applied in this ULPBench. I have a feeling this test favours CPU heavy micro's, perhaps not important at all when your application is I/O limited and sleep heavy.

Also an often used trick to minimize power is use DC/DC's to bring down nominal 3V lithium batteries to the minimum voltage the chip can operate at - often 1.8V or perhaps lower. Just that can mean a jump from e.g. 100uA/MHz down to 60uA/MHz.

[Teuobk]

To the EFM32 comment: I've not found the EFM32 to be notably better than other Cortex-M solutions. In real-world testing, I've found the Nordic nRF52 or even some of the STM32F4s, for example, to often outperform an EFM32 Giant Gecko in terms of work done per microamp. Of course, your mileage may vary. That said, the EFM32 isn't bad, and I appreciate how SiLabs provides a decent free IDE (which not all vendors do).

[rsjsouza]

Have you looked into the shelf life (battery self discharge) or leakage current (capacitor) for your backup power source.  Almost certainly these will be the limiting factors with an ultra-low power micro or clock chip.    With a CR2032 cell, for example, if your circuit draws less than around 3uA the 2032 shelf life will be the limiting factor -- 10 years or so.

This is a critical factor. I have two MSP430 development kits that are complete watches powered by CR2032 batteries and they were turned on in 2004 and 2005 respectively and were never powered off through the t8me the batteries lasted: around 11 years on both kits. I imagine the shelf discharge was the main factor for this.

However, why use such low power in a nixie clock?

[FrankBuss]

However, why use such low power in a nixie clock?

Might be useful for other projects and with the Nixie clock I can test it.

[westfw]

It gets complicated, because really low-power microcontrollers tend to be designed to spend most of their time in very-low-power "standby modes", which vary pretty greatly in their utility (what is left running, what can cause wakeup, etc) and especially in how much time/power it takes to go from the standby mode to the operating mode...

I'd guess that most of the "very low power" microcontrollers laugh at battery-backup these days; power is low and batteries are good.  It's "energy harvesting" that is the new low-power target application - Can it meaningfully operate off of the uW of power available from stray RF, or Peltier harvesting of small thermal differences, or the occasional mechanical vibration...

[splin]

I've just seen this linked in danadak's post reporting Cypress's new PSOC6

http://www.eetimes.com/document.asp?doc_id=1331471

What is  particularly interesting is this bit:

Cypress' proprietary ultra-low-power 40-nm SONOS process technology enables the PSoC 6 MCU architecture to feature industry-leading power consumption with 22 µA/MHz and 15 µA/MHz of active power on the Cortex-M4 and Cortex-M0+ cores, respectively.

Seems a bit like waiting ages for a bus then 3 come at once! If those numbers are correct and directly comparable, it blows Ambiq and AMD's just announced ADuCM4050 out of the water in the low power stakes. And EFM32 and ST32L4 are left even further behind.

I wonder if ST have anything comparable in development as I guess they will have to respond?

[EDIT] I see Cypress already have a Cortex M0+ S6E1C1 (FM0+) uC which is specced at 40uA/MHz. I don't know about availability but the datasheet was revised in March 2016 so it looks like they have been around for some time. It seems I have managed to miss these previously.

[julianhigginson]

the silabs/energy micro parts really disappointed me a year and a half ago... at first they showed very promising low power mode specs, on the data sheet, (and I think three product page, too...) then the datasheet errata corrected those numbers and the power down mode was shown as basically double. you'd think that an error of 100% for advertised power consumption in low power modes on a product specifically marketed as low power would want to be communicated slightly more urgently than tucked away in an errata! I wonder if they've fixed that yet?

Sent from my A0001 using Tapatalk

[Dielectric]

What's interesting about the PSoC6, is that while the sleep and hibernate numbers are just OK (blame CMOS leakage at 40nm), those run currents are way down and it supports voltage scaling on the cores.  The design intent is more like an always-on but low-power mode so you don't have to sleep/wakeup constantly.

I'm actually curious about lifetime data on a coin cell, since they're notoriously bad at supplying surge current (tends to shorten their overall lifetime) but are better at supplying a low, constant current.

The S6E1xxxx parts are from the Spansion merger, they're pretty good but somewhat unknown outside of Japan and a little bit of Europe.  Tons of peripherals on them if you dig into the manuals.

[FrankBuss]

I created a breakout board with KiCad for the AB1815 RTC: https://oshpark.com/shared_projects/UJrkUq22

It is my first KiCad board, so I hope I've done everything right. Cost at OSH-Park is $2.25 for 3 boards and free shipping, if you chose USPS and I've shared the project, so everyone who wants one can buy it. But might be good to wait until I have tested it

This is how it looks like in the KiCad 3D viewer:



That's with version 4.0.6, which I compiled from source, in older versions it didn't look that nice.

Github repository with all KiCad files: https://github.com/FrankBuss/nixie-clock/tree/master/ab18x5-breakout

[theatrus]

the silabs/energy micro parts really disappointed me a year and a half ago... at first they showed very promising low power mode specs, on the data sheet, (and I think three product page, too...) then the datasheet errata corrected those numbers and the power down mode was shown as basically double. you'd think that an error of 100% for advertised power consumption in low power modes on a product specifically marketed as low power would want to be communicated slightly more urgently than tucked away in an errata! I wonder if they've fixed that yet?

Sent from my A0001 using Tapatalk

EM (and I guess SiLabs by extension) has had issues delivering advertised specs for years (remember the EFR24 radio? ). Their software libraries are excellent otherwise so its a major bummer. Their original Gecko series performed really well and was a game changer, but it looks like they're falling behind.

[ebclr]

This processor need only a few electrons to work

http://www.silabs.com/products/development-tools/mcu/32-bit/sim3l1xx-low-power-badger-board-evaluation-kit

[FrankBuss]

This processor need only a few electrons to work

http://www.silabs.com/products/development-tools/mcu/32-bit/sim3l1xx-low-power-badger-board-evaluation-kit

Well, the datasheet says 75 nA in power mode 8 (timers running, CPU in sleep mode). But this would be pretty good. But the PIC10F320 for example claims 20 nA, and looks like this is with a timer running, which can generate an interrupt to wake up the CPU.

[nctnico]

IMHO it is odd nobody mentions TI's MSP430 series which are supposed to be very low power. Usually the ultra low power modes have everything shutdown to a useless level so the numbers quoted in datasheets aren't worth the bits needed to store them. Especially when typical values are printed without a maximum. What counts is the power requirement to still do something usefull (like running a wake-up timer for periodic checks) and the time needed to wake-up. During wake-up a microcontroller draws power but doesn't do anything because the oscillators have not started properly yet.

[brucehoult]

IMHO it is odd nobody mentions TI's MSP430 series which are supposed to be very low power.

https://www.eevblog.com/forum/microcontrollers/lowest-power-microcontroller/msg1165011/#msg1165011

Well known as a pleasant low power pseudo-PDP11 [1] in these parts, I should think.

[1] 16 registers instead of 8, and corresponding fewer addressing modes to still fit in 16 bits. But otherwise...

[ebclr]

PIC10F320 is a crappy 8 bit, with only 512 Words of Flash Program Memory  64 Bytes Data Memory

Are you kidding?

SiM3L1xx

Up to 256 kB flash
Up to 32 kB RAM
Up to 62 GPIOs
50 MHz Cortex-M3 core
Integrated dc-dc converter
Variety of wake sources
128 or 160 segment LCD controller
Sensor interface
Hardware AES, CRC, encode/decode
Enhanced DMA with DTM
10-bit 1 Msps or 12-bit 250 ksps ADC
Current mode DAC
2 x SPI, UART and I²C
Analog comparators
Enhanced PCA, timers
Voltage reference and temperature sensor
Available in 40, 64, 80-pin packages

[brucehoult]

PIC10F320 is a crappy 8 bit, with only 512 Words of Flash Program Memory  64 Bytes Data Memory

Are you kidding?

So about 10x more data memory than required, and probably a lot more code space than required too.

SiM3L1xx

Up to 256 kB flash
Up to 32 kB RAM
Up to 62 GPIOs
50 MHz Cortex-M3 core

MASSIVE overkill. And 4x the power usage. For nothing useful in this application.

Great in other applications, for sure.