Hacker News Reader

Comment by fisherjeff | Hacker News Reader

So if a human burns ~2000 calories per day and has 1.5-2 m^2 of skin, that means that a reasonable upper bound on (mean) thermal flux density of human skin is around 65 W/m^2. So if a watch makes contact with an area of, say, 2 square inches, that comes out to... 85 mW of available thermal energy?

Am I missing something?

View on HN

Not a direct response to you, but if you calculate Carnot thermal efficiency with a skin temp of 37 C (likely too high), and ambient temperature of 20 C (probably about right), you get 5.5%.[1] Using this result (combined with yours), the watch has to run on less than 5 mW of electrical energy.

[1] http://calistry.org/calculate/thermalEfficiencyCalculator

P.S. It would also be interesting to see how much waste heat is removed in breathing.

5mW should be enough for a simple e-ink watch. CPU and display have to stay idle most of the time though and you can't communicate a lot over radio. It should be possible, even with that little power if you can accumulate the energy in a capacitor. However it's not an easy task and poses a real engineering challenge, that's for sure.

Many of the Cortex M0s are very good at conserving power, and have a number of low-power and sleep modes, which make running at these levels quite easy. On the other hand, while Bluetooth 4.0 is low-power friendly, cost and size limits inherent to the watch will probably force the team to do a custom antenna and FCC certification, which is expensive, slow, and painful.

I would say the electronics are not too complicated or difficult, but the energy harvesting and design will be tough, and the whole project will be quite arduous.

It would be very interesting to see their development plan, and whether they are going for an iterative/modular design, or all-in.

I agree, the CPU is probably the smallest problem.

I'm curious with how little power you can get away with using a BLE radio. Especially if you constrain yourself to some kind of polling mode where the watch reaches out to the smartphone every minute or 5 minuetes or so.

A basic low-power microprocessor (like the MSP430) will run at 1MIPS at about 0.5mA. (Less when sleeping.) At 3V, that's 1.5mW. A Sharp epaper display, like on the Pebble, consumes basically nil when not changing, although I can't find data on energy use when updating. A traditional STN LCD uses about 1mW.

So I reckon it's plausible, just --- thermoelectric generators have traditionally been really inefficient, and I don't know whether getting 2.5mW of electricity out of 85mW of heat is possible. Particularly if you're like me and wear the watch quite loosely.

Edit: Datasheet: http://www.ti.com/lit/ds/slase54a/slase54a.pdf

That's an FRAM model with non-volatile RAM, so it can power down completely without losing memory. It claims it'll sleep with the clock on at 450nA. (That's nano, not milli.) It also claims it'll run with the processor on at 120 µA/MHz, so the 1 MIPS figure I quoted above must be about 4MHz. I've never run one below that but I gather they'll go down to about 500kHz. Enabling peripherals costs more, of course, but you can leave them powered off until you want to use them.

So, yeah. Totally plausible.

There is no epaper in pebble. You can get ~25 microwatts per cm2 per K temp difference (according to Linear). Enough for the Sharp memory display and simple clock circuit.

Yeah, it's the screen I'm curious about. The renders on their Indiegogo page make the screen look pretty awesome.

FWIW, the original Pebble is rated at 2.8 mW and the Pebble Time is 4.6 mW.

Most smartwatches are also paired with phones. So there's extra power usage for bluetooth, too.

There are CPUs out there with sleep currents well below 0.1 mA, for example the Microchip nanoWatt XLP. The product page advertises 9 nA in a best case sceneario.

Well, the MSP430 sleeps at about 1µA --- below if you turn the clock off!

But as you're not really getting anything done when the processor's not asleep, I quoted the power consumption when running. (Of course it's possible to use a capacitor to soak up energy and then run the processor at higher currents, in burst. You'd probably want to do this if you want radio. But on average, you're still getting about the same amount of work done.)

Slight digression, but the MSP430 is obsolete; modern low power ARM microcontrollers are superior in every dimension.[1]

[1] I have worked on products with both uC, and the company where I work buys both by the reel and tray.

Would those be the new Atmel SAM L21? I see a reference to 35 µA/MHz and 200 nA sleep mode:

http://www.atmel.com/products/microcontrollers/arm/sam-l.asp...

So:

(a) yay! super low power processors! Bet you can run these off a couple of long wires and radio scavanging;

(b) ARMs are so dull. [sad face] The MSP430 is a lovely processor to write machine code for. ARMs are just... dull.

Anyway, thanks for the suggestion. I think the MSP430s have a monopoly on FRAM, which makes a vast difference in usability, but I'll check these out.

The low power Kinetis processor can run at low speed for 32 µA, and down to 1.7 V. Many of the ARM manufacturers produce similar parts.

Interesting. Is it still able to wake up regularly without an external source to trigger the interrupt? Or does it have a timer with a dedicated oscillator / clock for that purpose?

Most modern low-power microcontrollers have low power timers which can be left on with the rest of the processor asleep; this low power timer is usually capable of running very slowly, and generating occasional interrupts to wake the processor.

To a first approximation that looks fine[1]. To a second approximation one might consider that heat doesn't exit any endotherm in remotely equal surface fashion[2]. And if anything the wrists appear as local minima.

[1] https://en.wikipedia.org/wiki/Spherical_cow

[2] http://i.stack.imgur.com/CsnO5.png

Air and clothing vs a metal plate in skin contact makes that slightly less important. The real probably is you probably get 2% efficiency due to minimal temperature differences involved, and people take their watches off.

you can divide that by ~100, unless you manage to keep temperature difference at 100'

Reply on news.ycombinator.com