Ultra low power MCU design: the NanoWatt Clock

[Howardlong]

Folks,

Here's a follow up to my earlier topic on Ultra Low Power Microcontroller Design: https://www.eevblog.com/forum/projects/happy-nerd-year!-my-first-project-of-the-year-semi-fail/

That old version managed about 5.4uW (1.8uA at 3.0V).

TL;DR: This new version managed an average 693nW for a 4-digit clock, and 896nW for a 6-digit clock.

That's 385nA at 1.8V and 498nA at 1.8V respectively.

(Note might still be rendering higher resolutions for a few hours after this posting)


The MCUs used are PIC16LF1907 (4-digit) and PIC16LF1947 (6-digit). The LCD is DE126-RS-20/7.5.

The key tricks over and above the first iteration were:

o Use an external RTC with push-pull programmable clock out (MicroCrystal RV-3028-C7)
o Decrease microcontroller voltage to 1.8V
o Find an LCD that works at 1.8V (not hard, I discovered it by accident, I don't think they're uncommon)
o Use an MCU with sleep current of 10s of nA rather than 100s, while allowing key peripherals like LCD to run

The MCU is run at 16MHz when out of sleep, the lowest uA/MHz achievable with the internal clock and fastest wake-from-sleep (~10us).

The RTC's CLKOUT operates at 2048Hz fed into the PIC's T1CLKI, used for both the LCD peripheral and Timer1. Using the external RTC reduced the raw 32kHz requirement from 170nA to 45nA: using the RTC's 2048Hz output rather than the PIC's 32kHz SOSC for timing and LCD reduced the dynamic current in the PIC by a further couple of hundred nA.

Active duty cycles are 2,000,000:1 for the 4-digit and 16,667:1 for the 6-digit.

To allow use from a 3.0V button cell I recommend either the MCP1810T-18 (now EOL) or the TPS7A02-18 LDOs. These are 20nA and 25nA Iq respectively. Unfortunately these 20nA LDOs, when used with 3.0V button cells rather blows the 1uW, but I'll keep upbeat about it, the clock itself is well under 1uW. We just need a battery chemistry with a nominal ~2.0V or so. A couple of lemons maybe?

[BrianHG]

Is that low enough power to operate on a tritium nuclear battery or radioisotope photovoltaic generator?
Maybe a nuclear diamond battery where your clock could run for 25k years...

[Someone]

Impressive to get the leakages so low with that construction!

To allow use from a 3.0V button cell I recommend either the MCP1810T-18 (now EOL) or the TPS7A02-18 LDOs. These are 20nA and 25nA Iq respectively. Unfortunately these 20nA LDOs, when used with 3.0V button cells rather blows the 1uW, but I'll keep upbeat about it, the clock itself is well under 1uW.

Switching regulators are starting to encroach on these requirements:
LTC3336
TPS62840

Since you arent constrained on size or price, building a bootstrapped buck regulator on the input, driven in the wakeup of the micro it might even be possible to go lower in power.

[mariush]

Wonder if it woudn't make more sense to use a ratiometric regulator, ex max1044 / icl7660 with a rechargeable battery that gives 3.6-4v , so you get 1.8v - 2v output ... the datasheet says 98% efficiency and 30uA supply current..

MAX1044/ICL7660 : https://www.digikey.com/en/products/detail/analog-devices-inc-maxim-integrated/ICL7660CSA-T/1303636  datasheet https://datasheets.maximintegrated.com/en/ds/ICL7660-MAX1044.pdf

example battery : https://www.digikey.com/en/products/detail/adafruit-industries-llc/1570/5054546

could add a solar cell a trickle charge the battery from ambient light.

[BrianHG]

As for the regulator, you can choose a binned N-Channel Depletionm mode J-Fet which has an ~1.8v output.  This also gives you the chance for 1.7-1.5v if the LCD will still work for even less current while this JFET style regulator has a quiescent current in the fentoamp range, not this wasteful power hungry ~20 nanoamp range.

Basically the gate is tied to GND, and source in the vin and the drain is the vout.

Another plus is the 0v dropout range.  15v completely all the way down to 1.8 vin and you will get 1.8v out.

[BrokenYugo]

Is that low enough power to operate on a tritium nuclear battery or radioisotope photovoltaic generator?
Maybe a nuclear diamond battery where your clock could run for 25k years...

Perhaps a peltier plate with a wet rag on one side? Then you can troll pedantic types by telling them it runs on water.

[Marco]

Can't you just use the comparator to see if capacitor voltage goes below 1.8V and give it a quick constant time blip with a microcontroller buck made with some low gate voltage mosfet with small drain leakage? The 1.8V doesn't have to be super precise, no need for a "proper" control loop.

[SiliconWizard]

Powering it could be a job for say, a small 1F/3V supercap.

If we assume an average consumption of 500 nA (possibly you did better), it would take over 23 days to lose 1V. So, charged to 3V, it could easily last a full month without recharging (but many of those supercaps have significant self-discharge, so don't count on a figure that optimistic), just followed by a very low power LDO. Then, you could add a small solar cell, or any other kind of small energy harvesting system that would trickle charge the supercap. Just an idea.

[BrianHG]

Hmm, did I do my math right?
A 1000ma primary lithium coin cell should run this clock for ~228 years?
Will the MCU eeprom fail first?

[SiliconWizard]

Hmm, did I do my math right?
A 1000ma primary lithium coin cell should run this clock for ~228 years?

Yep.
A typical humble CR2032 would be more like ~200 mAh, which would still be ~45 years. Unfortunately, the cell is probably not going to last that long?

[BrianHG]

I wonder if we could find a way to get something like the Oxford Electric Bell, or Clarendon Dry Pile battery to power this clock.  It has been ringing a bell for 179 years so far.

https://en.wikipedia.org/wiki/Oxford_Electric_Bell

Though we would need to down-convert the electrostatic voltage to 1.8v.

[BrianHG]

Hmmm, maybe a generator could be designed around atmospheric temperature and pressure change.

Like this mechanical clock:  https://en.wikipedia.org/wiki/Beverly_Clock

The claim a consistent 3.6 μWh over its 157 year operational time.

A simple pmg generator attached to a dry ceramic 10000uf cap should last.
(I avoided the supercap because it can dry out over time.)

[Howardlong]

Can't you just use the comparator to see if capacitor voltage goes below 1.8V and give it a quick constant time blip with a microcontroller buck made with some low gate voltage mosfet with small drain leakage? The 1.8V doesn't have to be super precise, no need for a "proper" control loop.

Nice idea.

The trick would be to only turn on the comparator and any associated voltage reference parts for a very short duty cycle: the on-chip comparator chews through 7.6uA, and FVR is 8.5uA, so not really an option for always on. It'll also need a reasonably low leakage cap.

Unfortunately the PIC16LF1907 doesn't have an on chip comparator although the '1947 does.

The '1939 does have a comparator, but lacks enough segment drivers even for a 4-digit clock.

I'll see if I can come up with something.

[muvideo]

We just need a battery chemistry with a nominal ~2.0V or so. A couple of lemons maybe?

Nice indeed,
have you considered a lithium titanate cell? Sometimes called LTO, it's a lithium ion chemistry with a 2.4V peak voltage and high durability.
Coupled with 5 small solar panels in series and a little ambient light could run indefinitely. I don't know if it's easy to find LTO based coin cells,
but I think that something similar is already used in some wristwatches.

[SiliconWizard]

Devising a very long lasting (say > 50 years) energy generator is a pretty interesting and challenging project in itself.

LTO is cool but despite taking many more charge cycles than standard Li-ion, it still has limited life. So indefinitely? I don't think so.
Low-capacity LTO batteries are commonly available as cans resembling through-hole electrolytic capacitors. And they are quite "chunky" for a given capacity. For instance, a typical 10mAh LTO battery is a cylinder of diam. 8.0*12mm.