My record so far is dipping down to 0.498v and successfully recovering blinking i/o at 1.05v, which I find amazing.
That just means the RAM managed data retention, not that the core was still running.
Static data retention will be some level below the slow-clock-dynamic, but absolute level will depend on leakages and be strongly temperature dependent.
Normally RC osc stop at roughly 2x the logic threshold whilst RAM hold can be down to ~logic threshold, which your ~0.5V and ~1.0V numbers tend to confirm.
If you can find an external Osc of suitable low power and Vcc range, you may be able to clock the prop below that 1.05v, as that's likely the RC osc halt point.
What's interesting is that article the other day about zombie memory. A guy hooked an AVR up to an LED to see how long it would hold memory contents after removing the power. It turns out that the LED was backfeeding the power bus via the photovoltaic effect and the clamping diodes.
So, that makes me wonder how much some LEDs could contribute to the Prop in such an environment?
I think Ken or Chip has said that noise from the internal oscillator can be observed on the power or IO pins when the oscillator is running. If you've got a decent oscilloscope, it might be worth looking for.
Marty
There is certainly noise there. I'm running "commando" (no bypass caps) and power is fed in via a 100kohm, so it's easy to see the noise, but I think its mostly coming from the lab supply. I'm still getting used to what I'm looking for, but it appears there is more of a 'slow wander' in the dc level (ie dc supply current via 100k) when operating above 1.05v, compared to low voltages, that might be the clue.
I should point out most of my tests yesterday were without CTRA running (for observing osc freq), and it looks like results and thresholds are a bit different when it is running. Interestingly, CTRA recovers from supply voltages as low as 0.340v, but the program itself isn't running just the counter.
That just means the RAM managed data retention, not that the core was still running.
Static data retention will be some level below the slow-clock-dynamic, but absolute level will depend on leakages and be strongly temperature dependent.
Normally RC osc stop at roughly 2x the logic threshold whilst RAM hold can be down to ~logic threshold, which your ~0.5V and ~1.0V numbers tend to confirm.
If you can find an external Osc of suitable low power and Vcc range, you may be able to clock the prop below that 1.05v, as that's likely the RC osc halt point.
Ok that makes sense.
For a while yesterday i was using a 220uF cap connected to P14 (configured to "output high" as the "power source" supporting the memory retention. After everything else is disconnected only power coming through that gpio pin could be supporting the core. Given the i/o functionality appears to die around 1.02v, I guess its the protective diodes leaking enough voltage in to keep the memory cells alive.
For most of the later tests I got rid of the 220uF and fed power in via the usual Vdd pins
Clocking XIN form an external osc is a good idea, I'll give that a go later once I have a way of reading counts
What's interesting is that article the other day about zombie memory. A guy hooked an AVR up to an LED to see how long it would hold memory contents after removing the power. It turns out that the LED was backfeeding the power bus via the photovoltaic effect and the clamping diodes.
So, that makes me wonder how much some LEDs could contribute to the Prop in such an environment?
Fascinating. I'd sure believe it after these tests
Yes I think at these kind of levels leds could indeed make a difference. Certainly something like a pair of BPW34's (50uA ish?) could be useful too.
Is this done to eliminate possible leakage currents from the caps, or do you just like pressing your luck?
-Phil
Does it have to be "or" - can't I entertain both options, and then some?
I have to say whenever I've pressed my "luck" with the Prop, it seems to have been up to the task. The H bridge driving two small motors using 8 paralleled gpio per leg springs to mind. Whether that kind of "luck" would survive Ikea style repetitive failure testing, let alone make it into any kind of commercial product, I doubt.
I guess the initial plan was to focus in on one or two areas and keeping in absolutely minimal (means less to write up). It is interesting watching the CTRA output waveform as the voltage sinks down though. This lab supply has detents, and the waveform shape kind of "bounces" as the RCSLOW osc gasps at its new lower voltage setting. That could also be related to the supply I guess, given there is next to no load.
Fascinating. I'd sure believe it after these tests
Yes I think at these kind of levels leds could indeed make a difference. Certainly something like a pair of BPW34's (50uA ish?) could be useful too.
A bpw34 has a short circuit current of near 2mA in full sun. It'll be plenty. I'd also look at the amorphous silicon solar cells that Digikey sells. They'll work better under high-efficiency lighting.
Yesterday I did some tests using RCFAST, but starved of supply voltage, so it operates at just a few kHz. I was able to get the operating current down to about 8.3 uA at 1.01~1.02v.
The oscillator kind of splutters along and is a bit irregular, but it did keep going overnight. GPIO was also working at this voltage level.
What's the advantage compared to the more intuitive RCSLOW, you ask? It appears the static current is lower down at "stasis levels", measured at 0.31uA compared with about 0.44uA for RCSLOW. That was measured at Vdd=0.473v and it was possible to recover GPIO and CTRA toggling from that point (ie the program appears intact), once Vdd goes back above 1.01v or so.
I need to move on to other things, get some better test gear in and repeat all these experiments formally with what I've learned, but that might take a while. I thought I'd put this RCFAST idea out there in case its useful for anyone else.
Yesterday I did some tests using RCFAST, but starved of supply voltage, so it operates at just a few kHz. I was able to get the operating current down to about 8.3 uA at 1.01~1.02v.
The oscillator kind of splutters along and is a bit irregular, but it did keep going overnight. GPIO was also working at this voltage level.
There is a uA / KHz curve, so some Vcc points where the Oscillator looks to be stable, could be useful.
- ie Minimum stable Vcc = ??uA & ?? kHz and maybe +100mV and +200mV to see how strongly Vcc affects this.
Agreed. I hope to put together a curve like tracey did but extending down to the 0.4 volt level, and adding a curve for RCfast. But get one of EEVblog Dave's ucurrent shunts in first
+100 / 200mv would have a huge effect in the RCFAST mode. It may be instructive to plot the kHz vs current as the Vdd sits around 1.02v, and the oscillator frequency rapidly grows if more current is available.
Ultimately battery chemistry & voltage will dictate operating point, so it might be useful to characterize data points around say 1.2v, and perhaps 1.8v too.
jmg, you were right about the external clocking. I used an external sine via 33k resistor into xin. With that in place, in xinput mode it was possible to operate with GPIO down to (at least) 0.87v successfully, with an i/o pin toggling.
Current consumption at that point was relatively high, above 10uA apparently. I'll look into that in more detail later.
The other thing is, the xtal1, 2, 3 modes seem to work down to 1.8 volts approximately. I don't know yet whether the prop will accept an xtal under 4 MHz at low supply voltages.
jmg, you were right about the external clocking. I used an external sine via 33k resistor into xin. With that in place, in xinput mode it was possible to operate with GPIO down to 0.87v successfully, with an i/o pin toggling.
Current consumption at that point was relatively high, above 10uA apparently. I'll look into that in more detail later.
The other thing is, the xtal1, 2, 3 modes seem to work down to 1.8 volts approximately. I don't know yet whether the prop will accept an xtal under 4 MHz at low supply voltages.
When doing this sort of test, it's important to keep slew rates fastest, and probably use the extclk mode so any feed-back resistor is disabled.
Slow slew signals means longer in the linear region, and that's why sine give high Icc
I's not easy finding low Icc, low freq Oscillators, but if you just want 'any freq', some of the nano power comparators as ring oscillators could be worth a look.
Some RTCs have 32KHz op modes, but often that is not push-pull, so you add the pullup Icc cost.
Google did find a PCF85063, CMOS drive, which has Icc curves for CLKOUT on and off, and it allows CLKOUT to
be any of
32768 / 16384 / 8192 / 4096 / 2048 / 1024 / 1 Hz
Package looks challenging tho, but it could make a nice nano-power Test Osc
some of the nano power comparators as ring oscillators could be worth a look.
They certainly can work well at <100KHz speeds, but they usually have slow output edges in the 1uS or greater ball-park. I've found that they only work well clocking logic if they're buffered with a Schmidt trigger gate or a gate with lots of positive feedback. They're also op-amp based, so are less likely to work below specified voltages.
The PCF85063 is a nice part. Wonder if it would work as an oscillator using a 128KHz or 256KHz crystal? Also wonder if there is something similar in a package with exposed pins...
I wish they would not put RTCs in DFN packages. I've just had an extended go-around about that with Intersil in reference to one of their RTCs. Pins of DFNs, underneath, trap a lot of flux in narrow interstices and are difficult to clean. Micro-power oscillator circuits are notoriously sensitive to leakage between the pins. Flux has to be removed completely, especially if it is water based, hygroscopic, or else it is trouble in the making. Either the clock won't work on day one, or it will stop when exposed to high humidity.
I see that the PCF85063 data sheet recommends only a small area of solder paste on the central land. That would result in less excess flux release and also perhaps allow more of a channel underneath for flow-through cleaning under pressure. I'd consider leaving paste off the central pad completely, or no land at all under the chip. It is not connected to anything electrically and it serves no thermal purpose.
Along with other measures, we are applying a spot silicone conformal coating around the RTC on our boards.
(PCF85063 too needs a pullup resistor on the INT output, so high ohms resistor --> slower rising edges)
For 555 timer collectors, I ran across this one in the micro-power enhanced category. It claims to run on typical 2.2µA at 1.5V, and includes a 6 decade delay counter and a 1-byte eeprom to program the delay.
This morning I tried a 3.27MHz crystal successfully and a 2 MHz unsuccessfully. There are a few more values to try in between, once the courier gets here.
I also tried hooking up IR leds as visual indicators, when viewed through my phone camera. They have a forward voltage ~ 1.2 volts. That gives me a way of seeing activity, although it blows the power budget somewhat
And i've started experimenting with a multistage voltage multiplier (Dickson multiplier), taking advantage of the complementary NCO counter outputs the prop counter can generate. This may give us the ability to generate a higher voltage rail for low current peripherals that do need a higher than 1.3 volts to operate.
The other nice thing is even when the program gets destroyed by going down below 0.4v, when you raise the voltage again the counters recover what they were doing, so that higher power rail could re-establish automatically. There may therefore be some way to auto-recover from such an event (would need a low voltage "prop loader" circuit). Perhaps...
The 5 stage Dickson multiplier using Schottky diodes works well, using an Eneloop AA battery as the power source.
Overall efficiency is somewhere around 65% delivering to a 3 volt load, but keep in mind this is at an extremely low power level. The efficiency may get slightly worse as the battery depletes. On the other hand I have not tried to optimise cap size, frequency, nor Schottky diode selection.
I get about 5 volts out, unloaded, so one less stage may be better for eeprom boot. I am charging a largish cap with the output, and this cap could be sized to contain enough storage for the prop to go through a complete boot cycle from eeprom, start the prop, and quickly switch to RCslow.
I think this would work well with a portable version of the prop running from a single AA or AAA NiMh cell (Eneloop or other).
I can't help thinking that it would work better, higher efficiency, with magnetics, without all the diode drops. I've worked with the LTC3108 and I recall that you have experimented with that part too, the energy harvestor chip that kicks in as low as 20mV. I like the Coilcraft transformers that are designed for use with that part. That transformer might be a good one for the low voltage booster directly from the Prop. I've used one of those with the 1:10 ratio, a Prop running at 3.3V, and a standard voltage multiplier, to boost up to ~300V at nanoamps to bias a capacitance microphone. The secondary of the transformer in that case swings about ±20V, so the diode drops are less of a factor.
I agree with Tracy. A little Taiyo Yuden NGR inductor and a BSH103 from Philips driven by the Propeller could make a nice little 1v to 3-25v boost supply.
I think you guys are right, we could make a nice prop driven boost converter driven by the the prop, that would provide a more solid voltage rail.
I'm interested whether I can get the dickson multiplier to boot/reboot the prop from eeprom, and whether its good enough for flashing a high efficiency led (output low for led on, high impedance input for led off).
Tracy yes I really like those coilcraft coupled inductors too. Wurth magnetics have something similar. Its unfortunate the LTC3108 is so expensive but it does a great job
Yes, you can easily drive red LED's with a doubler circuit. At these voltages I've put a schottky diode and and LED in series from Vdd to Vss. I then drive the common node of the diode/LED from an I/O pin via a capacitor. Twiddle the I/O pin to light the LED. Works from about 0.75 volts with a 74HC2G14 chip, but crow-bars the power rails at about 1.75 volts.
That sounds interesting, Marty. Running from a single cell you'd steer clear of the crowbar region, just fine. Comparator looks really useful too, thanks for that.
This morning I was able to run from a 2.4576 MHz crystal down as low as 1.53v. Below that the oscillation appears to die and supply needs to be brought above ~2.2v to recover.
Edit: Was later able to get a 1.84 MHz crystal to operate down to ~1.55v.
Comments
That just means the RAM managed data retention, not that the core was still running.
Static data retention will be some level below the slow-clock-dynamic, but absolute level will depend on leakages and be strongly temperature dependent.
Normally RC osc stop at roughly 2x the logic threshold whilst RAM hold can be down to ~logic threshold, which your ~0.5V and ~1.0V numbers tend to confirm.
If you can find an external Osc of suitable low power and Vcc range, you may be able to clock the prop below that 1.05v, as that's likely the RC osc halt point.
So, that makes me wonder how much some LEDs could contribute to the Prop in such an environment?
There is certainly noise there. I'm running "commando" (no bypass caps) and power is fed in via a 100kohm, so it's easy to see the noise, but I think its mostly coming from the lab supply. I'm still getting used to what I'm looking for, but it appears there is more of a 'slow wander' in the dc level (ie dc supply current via 100k) when operating above 1.05v, compared to low voltages, that might be the clue.
I should point out most of my tests yesterday were without CTRA running (for observing osc freq), and it looks like results and thresholds are a bit different when it is running. Interestingly, CTRA recovers from supply voltages as low as 0.340v, but the program itself isn't running just the counter.
Thanks for the cro suggestion, Lawson.
-Phil
Ok that makes sense.
For a while yesterday i was using a 220uF cap connected to P14 (configured to "output high" as the "power source" supporting the memory retention. After everything else is disconnected only power coming through that gpio pin could be supporting the core. Given the i/o functionality appears to die around 1.02v, I guess its the protective diodes leaking enough voltage in to keep the memory cells alive.
For most of the later tests I got rid of the 220uF and fed power in via the usual Vdd pins
Clocking XIN form an external osc is a good idea, I'll give that a go later once I have a way of reading counts
Fascinating. I'd sure believe it after these tests
Yes I think at these kind of levels leds could indeed make a difference. Certainly something like a pair of BPW34's (50uA ish?) could be useful too.
Does it have to be "or" - can't I entertain both options, and then some?
I have to say whenever I've pressed my "luck" with the Prop, it seems to have been up to the task. The H bridge driving two small motors using 8 paralleled gpio per leg springs to mind. Whether that kind of "luck" would survive Ikea style repetitive failure testing, let alone make it into any kind of commercial product, I doubt.
I guess the initial plan was to focus in on one or two areas and keeping in absolutely minimal (means less to write up). It is interesting watching the CTRA output waveform as the voltage sinks down though. This lab supply has detents, and the waveform shape kind of "bounces" as the RCSLOW osc gasps at its new lower voltage setting. That could also be related to the supply I guess, given there is next to no load.
A bpw34 has a short circuit current of near 2mA in full sun. It'll be plenty. I'd also look at the amorphous silicon solar cells that Digikey sells. They'll work better under high-efficiency lighting.
Marty
The oscillator kind of splutters along and is a bit irregular, but it did keep going overnight. GPIO was also working at this voltage level.
What's the advantage compared to the more intuitive RCSLOW, you ask? It appears the static current is lower down at "stasis levels", measured at 0.31uA compared with about 0.44uA for RCSLOW. That was measured at Vdd=0.473v and it was possible to recover GPIO and CTRA toggling from that point (ie the program appears intact), once Vdd goes back above 1.01v or so.
I need to move on to other things, get some better test gear in and repeat all these experiments formally with what I've learned, but that might take a while. I thought I'd put this RCFAST idea out there in case its useful for anyone else.
There is a uA / KHz curve, so some Vcc points where the Oscillator looks to be stable, could be useful.
- ie Minimum stable Vcc = ??uA & ?? kHz and maybe +100mV and +200mV to see how strongly Vcc affects this.
+100 / 200mv would have a huge effect in the RCFAST mode. It may be instructive to plot the kHz vs current as the Vdd sits around 1.02v, and the oscillator frequency rapidly grows if more current is available.
Ultimately battery chemistry & voltage will dictate operating point, so it might be useful to characterize data points around say 1.2v, and perhaps 1.8v too.
Current consumption at that point was relatively high, above 10uA apparently. I'll look into that in more detail later.
The other thing is, the xtal1, 2, 3 modes seem to work down to 1.8 volts approximately. I don't know yet whether the prop will accept an xtal under 4 MHz at low supply voltages.
When doing this sort of test, it's important to keep slew rates fastest, and probably use the extclk mode so any feed-back resistor is disabled.
Slow slew signals means longer in the linear region, and that's why sine give high Icc
I's not easy finding low Icc, low freq Oscillators, but if you just want 'any freq', some of the nano power comparators as ring oscillators could be worth a look.
Some RTCs have 32KHz op modes, but often that is not push-pull, so you add the pullup Icc cost.
Google did find a PCF85063, CMOS drive, which has Icc curves for CLKOUT on and off, and it allows CLKOUT to
be any of
32768 / 16384 / 8192 / 4096 / 2048 / 1024 / 1 Hz
Package looks challenging tho, but it could make a nice nano-power Test Osc
Pin pitch is 25% larger than the P2 so gotta get used to that anyway.
They certainly can work well at <100KHz speeds, but they usually have slow output edges in the 1uS or greater ball-park. I've found that they only work well clocking logic if they're buffered with a Schmidt trigger gate or a gate with lots of positive feedback. They're also op-amp based, so are less likely to work below specified voltages.
The PCF85063 is a nice part. Wonder if it would work as an oscillator using a 128KHz or 256KHz crystal? Also wonder if there is something similar in a package with exposed pins...
Marty
I see that the PCF85063 data sheet recommends only a small area of solder paste on the central land. That would result in less excess flux release and also perhaps allow more of a channel underneath for flow-through cleaning under pressure. I'd consider leaving paste off the central pad completely, or no land at all under the chip. It is not connected to anything electrically and it serves no thermal purpose.
Along with other measures, we are applying a spot silicone conformal coating around the RTC on our boards.
(PCF85063 too needs a pullup resistor on the INT output, so high ohms resistor --> slower rising edges)
http://www.aldinc.com/circuitideas.php?id=13
This morning I tried a 3.27MHz crystal successfully and a 2 MHz unsuccessfully. There are a few more values to try in between, once the courier gets here.
I also tried hooking up IR leds as visual indicators, when viewed through my phone camera. They have a forward voltage ~ 1.2 volts. That gives me a way of seeing activity, although it blows the power budget somewhat
And i've started experimenting with a multistage voltage multiplier (Dickson multiplier), taking advantage of the complementary NCO counter outputs the prop counter can generate. This may give us the ability to generate a higher voltage rail for low current peripherals that do need a higher than 1.3 volts to operate.
The other nice thing is even when the program gets destroyed by going down below 0.4v, when you raise the voltage again the counters recover what they were doing, so that higher power rail could re-establish automatically. There may therefore be some way to auto-recover from such an event (would need a low voltage "prop loader" circuit). Perhaps...
Overall efficiency is somewhere around 65% delivering to a 3 volt load, but keep in mind this is at an extremely low power level. The efficiency may get slightly worse as the battery depletes. On the other hand I have not tried to optimise cap size, frequency, nor Schottky diode selection.
I get about 5 volts out, unloaded, so one less stage may be better for eeprom boot. I am charging a largish cap with the output, and this cap could be sized to contain enough storage for the prop to go through a complete boot cycle from eeprom, start the prop, and quickly switch to RCslow.
I think this would work well with a portable version of the prop running from a single AA or AAA NiMh cell (Eneloop or other).
Marty
I'm interested whether I can get the dickson multiplier to boot/reboot the prop from eeprom, and whether its good enough for flashing a high efficiency led (output low for led on, high impedance input for led off).
Tracy yes I really like those coilcraft coupled inductors too. Wurth magnetics have something similar. Its unfortunate the LTC3108 is so expensive but it does a great job
Oh and a MAX9064 comparator with ref would be perfect for the boost-converter feedback.
Marty
This morning I was able to run from a 2.4576 MHz crystal down as low as 1.53v. Below that the oscillation appears to die and supply needs to be brought above ~2.2v to recover.
Edit: Was later able to get a 1.84 MHz crystal to operate down to ~1.55v.