7

Inspired by this question I would like to know how low power you could go with a counter + 32 kHz oscillator (possibly made by yourself).

I found a nice oscillator circuit on a BJT reportedly drawing less than 1.2 µA from 3V.

Unfortunately the counter and/or prescaler parts are a little more tricky. I do not believe you can make low power flip-flops from discrete transistors but most normal logic ICs are not very efficient either (standard counter ICs all draw around 80 µA at room temperature).

I do not want a counter that is integrated into a microcontroller (PIC or AVR or ARM).

Community
  • 1
jpc
  • 5,302
  • 1
  • 24
  • 45

6 Answers6

7

There are several issues here...

  1. The oscillator circuit on a BJT doesn't quite spit out +3.3v logic levels. Fortunately, you want to use a lower voltage to get lower power consumption. +1.8v logic levels would be compatible with that oscillator-- but then you'd need +1.8 and +3.3v power rails (and probably loose all benefits in your voltage conversion inefficiency.

  2. Dynamic power consumption (the power used when things switch) is mostly from charging and discharging the parasitic caps on the various signal lines. The way to reduce that is to use shorter, thinner wires. And by shorter & thinner I mean don't use wires and instead use a chip. Building this from a collection of chips and transistors instead of one chip that does everything will drive your power consumption up.

  3. You said no microcontroller, but honestly that's the best way to do this. TI has an ultra low power MSP430 that would run at 32.768 KHz at less than 1.5 uW. As you've already seen, this type of performance is really hard to beat. After a microcontroller, my next choice would be a Xilinx Coolrunner-II CPLD-- but I doubt that meets your requirements either.

To summarize, an MCU will give you the lowest total power consumption. Otherwise your best choice is to use standard logic parts and suffer with something in the several hundred micro-watt range. Making something out of discrete transistors isn't going to be better.

  • The MSP430 you linked to is quite impressive but I cannot verify you 1.5uW power consumption (the lowest thing I could glance from the datasheet is about 15uW) and it does not have a 32kHz crystal oscillator. – jpc Apr 08 '11 at 20:56
  • You are aware the crystal oscillator is a separate part? – joeforker Apr 08 '11 at 21:13
  • @jpc I did a back of the napkin calculation to get 1.5 uW. The web page, not the PDF, says 0.9v and 45 mA/MHz. At 0.032768 MHz you get 1.47 uA. Then * 0.9v you get 1.32 uW. Then I rounded up to 1.5 uW. It's not super accurate, but nothing is when you're talking about power consumption on a CPU. It at least gets you into the proper ballpark. –  Apr 08 '11 at 21:19
  • @joeforker I'm not sure about the xtal osc. It's not obvious from the datasheet, but other TI products are just as non-obvious yet still support xtal's. If not, you could use that osc that JPC found and still be super low power. –  Apr 08 '11 at 21:21
  • @David You may be right that this is the lowest power way to go but it is strange that a whole uC uses less power than every single logic chip (even simple gates). The interesting thing is that the limiting factor in all of them seems to be the quiescent current and not the dynamic one. – jpc Apr 08 '11 at 21:26
  • 1
    @jpc It's strange what an entire uC is better and cheaper at. Recently I was looking for an 8 channel, 12 bit, 200 KHz ADC. It turns out that the cheapest way to get one is an MSP430. It's much cheaper than equivalent dedicated ADC parts. Of course, uC's are so much better than 555's (a friend calls them 666 timers, because they are evil). –  Apr 08 '11 at 21:32
  • @David True, but 555 timers in particular are still cheaper than uCs. – jpc Apr 08 '11 at 21:39
  • IANAEE, but I think the MSP430s will have a built-in 32 kHz oscillator that is low precision, but you can add a watch crystal for accuracy. – joeforker Apr 08 '11 at 21:57
  • 1
    @jpc Ya got me there. 555's are about $0.10, while the cheapest uCs are about $0.30 in 100 piece wholesale pricing. I still would rather use a uC over a 555. –  Apr 08 '11 at 22:01
  • @jpc - 555's are not exactly low power. Their CMOS counterparts, 7555, are, but I think they're more expensive as well. – stevenvh Apr 09 '11 at 15:18
  • @jpc - I once used a MSP430F1101 running off a 32kHz watch crystal, which consumed 5uA @ 3V, or 15uW. It used the on-chip oscillator. – stevenvh Apr 09 '11 at 15:21
  • @stevenh Thanks for your datapoint. @David Maybe you could mention it in your answer? – jpc Apr 09 '11 at 20:57
  • MSP430 specs for ADC are not that tight and RTC power is not worst-case spec says IOSC VCC = 0.9 V to 1.65 V 20 kHz 0.6 µA nom – Tony Stewart EE75 Jul 11 '12 at 23:19
  • 1
    The MSP430 idea is really an excellent one. That model, the MSP430L092, is *not* the one you'd want to use as it's a specialty ROM-only model. Nearly all the others have program FLASH for easy programming/debugging. You can use the G-series [MSP430G2001](http://www.ti.com/product/msp430g2001) which has a 14-pin DIP for experimentation and surface mount versions for production @$.40. – Brent Faust Oct 04 '13 at 00:35
5

Something better than HCMOS: NXP's Advanced Ultra-low Power (AUP) series. I'll presume you have a 32.768 kHz clock and want prescalers/counters to get 1 Hz from it.

The AUP series is not as extended as the HCMOS but the 74AUP2G80 may be all we need. It contains two D-flip-flops with inverted Q outputs, so that we can make a :2 divider. There's no Q output but that doesn't matter.

To divide by 2\$^{15}\$ we'll need 8 devices, that's a supply current of 4 µA maximum (0.5 µA per device). Dynamic power dissipation depends on the supply voltage, and since we want low power we choose the lowest possible: 0.8 V.

Then dynamic power dissipation is given by

enter image description here

C\$_{PD}\$ is power dissipation capacitance and is 1.8 pF at 0.8 V. Then we have

\$ P_D = 1.8 \times 0.8^2 \times 0.032768 \times 1 + (2 \times 0.6) \times 0.8^2 \times 0.016384 = 0.050 \mu W \$

The two times 0.6 pF is the load of the D input of the current FF plus that of the D input of the next stage. The power of the next stage is half this one's, since only the frequencies are halved, the rest is the same. So that's 0.025 µW for the 16.384:2 stage, and so on. The sum for 15 stages is 0.101 µF. (Binary people don't need a calculator for this: N + N/2 + N/4 + N/8 + ... = 2N). Add the static power of 4 µA \$\times\$ 0.8 V and we have a total of 3.3 µW, which is almost 2 orders of magnitude better than the 192 µA with the HCMOS devices (see my other answer).

stevenvh
  • 145,145
  • 21
  • 455
  • 667
3

For the oscillator I'd use an RTC like NXP PCF8563, which provides a buffered 32kHz output and only consumes about 300nA @ 2V.

For the counter/divider I looked at HCMOS. According to NXP's HCMOS family specification quiescent current for a flip-flop is 4uA maximum, but that's @ Vdd=6V, so it should be lower at 3V. If you use 74HC93s (4 FF per device, you'll need 4 of those to get 1Hz) total quiescent current is 64uA maximum, which agrees with the 80uA you mentioned.
The good news is that at this low frequency dynamic power is far less than that.

Minimum supply voltage for HCMOS is 2V, so working at this voltage should also reduce the current. The PCF8563 even operates at Vdd=1V (since you don't need the interface).

stevenvh
  • 145,145
  • 21
  • 455
  • 667
  • 2
    Square wave open-drain outputs aren't exactly conducive to low-power operation. If one is supplied by 3 volts and uses a 1Meg pullup current dissipation from the resistor alone will be 1.5 microamps. Too bad none of the parts I've seen have a low-duty-cycle pulse wave option. A 1024Hz pulse wave with a 16us low time (half a 32Khz cycle) would only use less than 0.05 microamps--a huge savings. – supercat Apr 09 '11 at 16:43
  • @stevenh But the guaranteed quiescent current over the commercial temperature range is 10 times higher. And an RTC manages to get 400nA over the whole industrial temperature range. Why NXP and Fairchild advertise new ultra-low-power logic gates that are only 2 times better in this regard? It is not that it is impossible to do better (like the RTC and MSP430 chips prove). MSP430 has the ability to go fast if needed, has normal output current capability and still does not sacrifice low quiescent current. – jpc Apr 09 '11 at 20:55
  • 1
    @supercat Yup, your low duty OC outputs are a really nice idea. – jpc Apr 09 '11 at 20:56
2

Many impressive solutions offered to date. I could offer a 100 nano watt RTC solution but it does not meet the requirement to be CPU free.

Even the major watch makers such as Casio use CPU cores in their ASIC watch chips.

Once dynamic power is negligible with 32KHz, the drain is controlled essentially by transistor leakage and supply voltage, so the best solutions will be Vdd=1V or less.

  1. Microchip: nanoWatt XLP in all non-DSP PIC's 800 nA Real-Time Clock and Calendar
  2. SiliconGate: SGC22300 RTC - Real Time Clock Nano Power Series

I also have research articles with schematics of 0.3uW clocks only, but not commercial ICs.

If interested; I can supply more info;

If I may be so bold as to inquire; why stipulate restrictions on implementation if the requirement is for lowest power?

Tony Stewart EE75
  • 1
  • 3
  • 54
  • 182
  • One of the best solutions was the MSP430 so I am willing to lift the CPU-free requirement a little. ;) – jpc Jul 11 '12 at 23:02
  • 1
    THat's ok. but if sleep power over wide temp range in an RTC is critical , best to use guaranteed specs.. call these guys who excel in nano-power RTC's http://www.silicongate.pt/product/sgc22300_rtc_real_time_clock_nano_power_series/13 In the end.. on time impedance is ~ >1MΩ is only <1uΩ @ 1V and sleep modes of Microchip are a big advantage too. normal and deep sleep. – Tony Stewart EE75 Jul 11 '12 at 23:43
1

Technically, the method to make the most power efficient (and fastest) circuit (anything) would be a Full Custom IC using the tiniest technology you can find. You'd have to design your own IC layout and have a foundry build a wafer for you.

There are universities that have agreements with some foundries, maybe you could research your local schools for information on that. If they are willing to help you then it would be free.

Static Free's Electric is a great free software to start building your own custom ICs.

Arturo Gurrola
  • 90
  • 3
  • I wish there were some way to design something nice and psychically convince a microcontroller manufacturer that they should build it into their product. It would seem that a few cascaded quadrature graycode counters (within a single chip) should be able to count using less current than would be required by a synchronous counter, but with a cleaner output than would be offered by a ripple counter. My ideal device would probably have a 48-bit resettable counter which was readable as either the top or bottom 32 bits, along with a 32-bit compare register. – supercat Apr 08 '11 at 22:37
  • The counter and about 16 bytes (4x32 bits) of RAM would be separately powered, and should be able to run for decades off a very small battery, or for days if not weeks off a supercap. From a programming standpoint, it would be nice if the compare register could do a magnitude compare rather than just an equality compare, but provided there aren't weird synchronization issues, software should be able to work just fine using a latching equality comparer. – supercat Apr 08 '11 at 22:40
0

If you think about simple ICs, you can get decent power consumption on CMOS chips. You may take single inverter in a chip, and a counter. Run it at some 1.8V, and hopefully it will eat very little :-)

BarsMonster
  • 3,267
  • 4
  • 45
  • 79