Launchpad Takes Ultra Low Power To The Extreme

We’ve all known the MSP430s under the Launchpad are designed to be low power, but who wants to bet how long the chip can last on only 20F worth of capacitors? A couple of hours? A day at max? [Kenneth Finnegan] setup a MSP430 with supercaps to find out. To make sure the chip is actually running, [Kenneth] programmed it to count from 0 to 9 over a period of 10 seconds, and then reset. To get it ultra low power, the chip is in sleep mode most of the time, and a raw low current LCD is used to display the output. While [Kenneth] simply checks the chip every few hours to see if it’s still counting, a setup much like the Flash Destroyer, tracking a clock and then storing the current value would get a more exact time of death. Either way, it’s been over 3 weeks…and still counting. Video after the rift.


62 thoughts on “Launchpad Takes Ultra Low Power To The Extreme

  1. Rail guns often use ni-cads and various other battery chemistry variants in place of capacitors.

    When you need a bunch of current right NOW, big caps just don’t have the right discharge curves.

    Nothing compares with flash-over in a big steel cage after something goes wrong – KaPOW!

  2. Hah, “Only 20F…”

    That’s a lot of capacitance there buddy!

    50 years ago, my high school science teacher was told that a 1F capacitor would have to be the size of a football field!

    Of course, that’s surface area, which you can clearly still cram into something small. But its a funny quote considering how common these super and ultracapacitors have become!

    Still, 20F is a lot by pretty much anyone’s standards.

  3. farads aren’t amp-seconds, that’s coulombs. Farads are coulombs per volt. Q = It and C = Q/V. super caps have a very high capacitance but a low maximum voltage, so they don’t store ex much energy as it would initially appear, especially since energy stores is proportional to the square of the voltage across the cap

  4. @cgmark: I lack any kind of measuring device to measure current accurately down to fractions of microamps, and there are more variables than that: the current consumption is very spiky (the controller runs in 32Hz bursts, so what are you actually measuring on an amp meter?), and theres the self-discharge of the caps, and how low the controller can *really* run below the rated 1.8V, etc etc.

    So I decided that just running it would give a more meaningful result than a single current measurement.

    Others have already made the point that supercaps are ridiculously huge capacitors, but still orders of magnitude smaller than chemical batteries…

  5. Also add to the comments that Microchips product line boasts the lowest sleep current industry wide. Well that was the case when I last researched it. (about a month ago)

    maybe its just me , but lately the quantity of posts is high, but its stuff like jamming a hub in a mouse.
    Personally I wouldnt call that a hack . But the idea had potential no less to become a hack .

    To be a hack I would say you need to add functionality to a device , not add more devices.

    This is one of my favorite sites to page threw, Just some constructive criticism.

  6. the problem with just running it without measuring current and voltage during the process is all that it tells you is what that exact code and circuit will do. It doesn’t tell you how that will apply to other circuits powered by the same capacitors. Unless you know what it is consuming at the various times there is nothing to use for comparison.

    For measuring something that powers on and off you take a period of time for the measurement and multiply it. Measure what it uses for 1,5,10 seconds and use that to figure out the usage for longer periods.

    If you don’t have meters that can measure down to microamps then you can use an opamp to make a current to voltage converter. They can measure into the picoamps range before it gets difficult. Just an opamp and a few resistors.

  7. I thought this was pretty cool and can be turned into something useful quite easily. There sure are a lot of grumpy people reading this website. If you don’t like it then don’t comment.

  8. Still think its comical so much testing and todoo over this when its not even the best out there. I would be curious to see one of our math wizards do a comparison between the TI chips here and a Microchip XLP. Obviously Im referring to doing the math for both not waiting 3 weeks for the answers.

    I suppose the point of this is to benchmark the incredibly well priced launchpad. But still if people are awed by the results it is worth noting that low power is one of Microchips (many) strong points.

    So any math wizards wanna break it down?

  9. ok so it wont let me post the link directly to microchips comparison chart comparing PIC24 and MSP430.

    Goto microchip dot com , under applications and markets click on XLP .

    I dont think there really is much comparison. But by all means check the graph out your self ;)

  10. Q = C * V
    Q = I * T

    C = 20F
    V = 5V

    Q = 100C

    With discharge current say 1uA:

    T = 100C / 1uA = 100,000,000 seconds

    T(seconds) / 60 / 60 / 24 / 7 = 165 weeks to discharge to 68% of max capacity.

    At current draw 100uA the value is 11 weeks.

  11. Actually, I think that specsmanship on power consumption of thing has gotten so complex that running the exact target application on real hardware, for a capacitor charged to a known point, is about the only way to really tell how things are going to work out. C values of far less than 20F are likely to be more useful, though!

    The microchip graph claiming to compare “instruction set efficiency” by counting the percentage of instructions that execute “in a single cycle” was particularly laughable. Especially since it included PIC16 and PIC18 families that don’t actually have ANY “single cycle” instructions (unless you’re willing to allow that a cycle is 4 clocks…) I like PICs, but the MSP430 instruction set blows the 8bit PICs out of the water…

  12. @smoker_dave

    Thanks for making my Pepsi squirt through my nose. lol

    Also, IIRC, there’s a way to modify the clock cycle to reduce the effect of EMI. Forgot where I saw it though.

  13. This is pretty amazing, but a rock doesn’t use much power either. I would think a real test would be to actually have it doing something a little more complex than incrementing a register, like running a multifaceted benchmark program, and monitor a heartbeat pulse. Then you start getting an idea of instructions per watt. Use something like the old “self-winding watch” mechanisms to generate current, and you’ve a great portable device.

  14. Rallen, I would not trust this kind of technology with a critical function like heartbeat monitoring. However for something like a home automation sensor / actuator which simply turns an IO pin on or off once every so often I feel this is a good simulation.

  15. @smoker_dave – I didn’t mean “use it to monitor human heart activity”. :) I meant, “have the device send out a regular pulse that resets a watchdog timer, as a function of its normal operation”. That’s a heartbeat pulse. Informal slang. It can be fun. Try it, sometime. Confuse the hell out of your co-workers. :)

  16. I do a lot of MSP430 work at uni, and the correct way to measure current in a setup like this is to take the integral of the current. In his case it should be an easy step function, so effectively:

    (i_sleep * %_sleeping) + (current_wake * %_awake) = i_total. Unfortunately the current draw is a function of Vcc in CMOS tech, so it varies a bit with the current level of the caps and stuff, but it should be close enough for a first order approximation. The MSP430s are fairly power efficient if you use the low power modes as Ken did, and with 20F backing it, I wouldn’t be surprised it runs for a very very long time.
    Hell, I bet the caps are self-discharging more current than the MSP430+LCD is drawing.

  17. PIC are not the lowest on power consumption…
    deep sleep mode is useless, it wont save ram, registers etc. pic can be woken up just by external interrupt – in fact it is reset and start from scratch. In TI’s micros you can go down in sleep mode but you can woke it up by software side (counter), the same (by functionality) mode in PIC’s is taking more power. The Microchip youtube demo video is a scam, it shows conditions newer existing in real life!!.

  18. Alright, first I’m no math wizard really…
    (couldn’t get through calculus…)
    I just looked at datasheets.
    I don’t know the MSP430, but his datasheet say
    With internal oscillator @1Mhz = from .2ma – .3ma
    (change with / voltage) so let’s guess .25ma
    Yeah, there the lcd module that would be easy
    To measure since consumption is steady.
    Let’s say it take 100 clock to get a task done
    (assuming a more complexe task is done)
    = ~100us @ 1Mhz, which give 100us/1s = 0.01% duty cycle
    = 0.01 * .25mA = 2,5uA average
    or ((2.7+1.8)/2)V * 2,5uA= 5.6uWatts

    Capacitor leakage :
    We dont what it is, it seem to change wildly from
    One 10F from another :
    Nichicon = .5 * C at 2.7volt = .5 * 10 = 5mA…
    They say it’s after half an hour.
    Maxwell = 0,03ma after 72 hours
    So duuhh it must be reduced when voltage drop…

    Here you also have to take into account ESR
    (Equivalent Serial Resistor) so it’s like a resistor .
    Nichicon = .3 ohm
    Maxwell = .075 ohm
    We take that in account in the capacitor
    Discharge formula ( as R)

    Q = C* V * (1-e^(-t/(RC) ) )
    Where Q= is the charge Amp/s?
    C = 10F
    V = 2.7 at start, 0 or with our mcu 1.7V
    e = natural number = ~ 2.71828…
    t = time in second
    R= in this case, ESR + mcu load (V=RI)
    Then you need to consider the leakage of both cap.

    So it’s not impossible, it took like 10min to get these info,
    Take 20 more and you might have some approximation

    I did my part.

  19. also another useful trick i came up with is reducing the external clock below 32 kHz, it seems that all that is required is a simple “equivalent crystal” composed of resistors inductors and capacitors with a resonant frequency in the low audio range (say 1kHz) as the current draw scales linearly with reducing frequency.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.