Nano current inrush switch/low voltage cutout

[Howardlong]

I have a target MCU running a simple application that runs just fine on 400nA and 1.8V, except for one thing: to start, it needs at least 17uA to start for about 3ms.

It's solar cell based, with a supercap backup, and the solar cell will produce oodles over 17uA in good light, but I also want it to be able to start in relatively low light, say a couple of uA.

So I'm looking for circuit suggestions for a super low power switch that only provides current when, say, the input voltage is over 2.0V, so there'll be a brief dip during MCU startup but there'll be plenty of charge in the supercap for this brief time. And, of course, the switch itself needs to be really low current.

[Marco]

This hackaday project mentions a potential direction for discrete solutions. You can use LEDs as voltage references (together with some small signal BJTs and a MOSFET) which are a little more stable than just using the Vbe of a transistor, as long as you black out the LED any way. It also mentions a couple IC's which should work for you, "XC6135, XC6136 or an Intersil ISL880xx series".

[Howardlong]

Coincidentally, a BJT and LED was what I was experimenting with yesterday, but my rudimentary circuit was very leaky. Thanks for the link, this looks worth a try.

[David Hess]

Circuits like power supply controllers which rely on bootstrapping their own supply use under voltage lockout circuits to do exactly what you are describing.  The power source charges a reserve like a capacitor and when a voltage trip point is reached, the capacitor is connected to the circuit.  It really is not that difficult.  A simple implementation might rely on a micropower comparator/reference like the LTC1540 to switch a MOSFET.

[Howardlong]

It really is not that difficult.

I don't agree when, as in this case, you are measuring power budget in nano amps. Circuits such as potential dividers, voltage references and bias circuits you'd typically use in such designs become very expensive in terms of power requirements.

A simple implementation might rely on a micropower comparator/reference like the LTC1540 to switch a MOSFET.

Thanks, I'll take a look, although it's not fully clear in the DS how it behaves on power up, although there is a brief section on this.

[Marco]

The Torex parts seems hard to source, so the LTC1540 or ISL88001/ISL88014 seem most appropriate. No MOSFET needed, just use the output directly as the micro's power supply.

It should be possible to make a low current bandgap circuit out of discrete transistors which trips at 2V ... but doing it with the minimum amount of transistors is a challenge.

[Howardlong]

I’m going to give the discrete circuit a go. The ISL chips are power hungry from what I could see, 5uA. The LTC is better, but the DS wasn’t clear about what happens at low voltage, so it would need characterising. By the time you’ve added a potential divider, I could see it topping 1uA total, three times as much as the payload.

The discrete circuit is very clever. It’s using the leakage in reverse biased diodes as high value resistors. I measured a 1N4148, it’s around 3nA at 1 or 2 volts.

More detail here https://www.edn.com/design/power-management/4334386/Solar-powered-motor-runs-on-10-nA and here https://m.eet.com/media/1135665/31804di.pdf

[bson]

If it were me and I have the time I'd investigate why it needs so much power during startup.  It's possible for example that during initialization  a bunch of peripherals, PLLs, clocking, and other circuitry is switched on all at once and perhaps a more staggered init would make a difference.  Maybe if you run a PLL, start the oscillator first, let it settle, then configure the PLL in a second step rather than all at once.  But of course, in the end it's going to need more during power on, without doubt, but if you manage to cut it in half while doubling  3ms to 6ms, maybe that's a win.

[Howardlong]

If it were me and I have the time I'd investigate why it needs so much power during startup.  It's possible for example that during initialization  a bunch of peripherals, PLLs, clocking, and other circuitry is switched on all at once and perhaps a more staggered init would make a difference.  Maybe if you run a PLL, start the oscillator first, let it settle, then configure the PLL in a second step rather than all at once.

Already looking into it. It seems like it’s part of the MCU’s startup procedure, with a current probe I can see it oscillating, as if it’s making sure the oscillator’s stable, alhough I’ve turned off the power up timer and clock monitor in the config bits. The config bits also tell it to go straight into the low power internal oscillator, there should be no interim higher speed oscillator. I have a number of other options to investigate, but I suspect it’s a feature. It’s not too late to change the MCU, but the device was already carefully chosen from a large number of other options, both from data sheet analysis and real world current drain... just not at power up!

[Marco]

I'd be a little scared about the temperature dependence of that EDN circuit, leakage current especially is very temperature dependent.

How about using a variation of David Johnson's ultra low power oscillator to turn on a xx431 say once a second for a couple of milliseconds, which would then pull down the gate on a low threshold voltage MOSFET once the voltage goes high enough? (I think his circuit needs some base resistors, the transistors will very quickly shut off again as it stands.)

[Howardlong]

Thanks for the link.

Typical temperature is room temperature, but I'll certainly keep that in mind. This is more of an academic exercise at the moment, although I do have some specific applications in mind for it if I can make if work as a system with the MCU.

[SiliconWizard]

The Torex parts seems hard to source, so the LTC1540 or ISL88001/ISL88014 seem most appropriate. No MOSFET needed, just use the output directly as the micro's power supply.

There's also the TPS3839x at just 150nA and fixed thresholds (no need for external resistors), if the 2.325V threshold is OK.

The problem with both LTC1540 and TPS3839x if directly using their output is that their high-level output voltage (min) is 400mV below the supply voltage, which is a significant drop-out for such an application I guess. Given the very low power consumption of the system, the drop-out voltage is probably much lower but you're pretty much on your own (outside of the datasheet figures). The other issue is that there is only one threshold, whereas, if I get it well, the OP would like the system to be supplied if the voltage gets above 2V, but would like for it to keep operating down to 1.8V.

[Marco]

The way the supply current for the LTC1540 rises during <2V operation is also a bit nasty.

One more nanowatt comparator with voltage reference, TS9001.

[Howardlong]

Yes, the low voltage behaviour is quite key, otherwise I won't have achieved anything other than added more complexity.

I breadboarded up the EDN article on p102 here https://m.eet.com/media/1135665/31804di.pdf.

It does indeed work: I set it up with a simple red LED load, and had to increase the trigger voltage by adding a further red LED in series with the green one.

I used 1N4148 which aren't quite as good as 1N4007 in terms of reverse leakage current. It wouldn't work at 10nA but would at 20nA, although it would take about 150 hours to charge up the reservoir cap sufficiently to trigger!

[David Hess]

I used 1N4148 which aren't quite as good as 1N4007 in terms of reverse leakage current. It wouldn't work at 10nA but would at 20nA, although it would take about 150 hours to charge up the reservoir cap sufficiently to trigger!

Gold doped diodes like the switching 1N4148 and fast recovery rectifiers have much higher leakage currents.  If you want a low leakage diode, then use a 2N3904 base-collector or base-emitter junction, or something like a BAS116 or BAS416.

[Howardlong]

I used 1N4148 which aren't quite as good as 1N4007 in terms of reverse leakage current. It wouldn't work at 10nA but would at 20nA, although it would take about 150 hours to charge up the reservoir cap sufficiently to trigger!

Gold doped diodes like the switching 1N4148 and fast recovery rectifiers have much higher leakage currents.  If you want a low leakage diode, then use a 2N3904 base-collector or base-emitter junction, or something like a BAS116 or BAS416.

Thanks for the tip. I did measure the reverse bias current over an indicative range of voltages for both the 1N4148 and 1N4007 as mentioned earlier. It was just a bit easier to breadboard the 1N4148. I could use resistors, but the biggest I have in stock are 100Mohm. Practically speaking, reverse biased pn junctions are going to be cheaper than resistors in this range anyway.

[Howardlong]

With a 2N3904, I measure 35pA at 3v BC and 48pA BE, so yes, not bad!

[SiliconWizard]

What lab instrument are you using to measure currents in the tens of pA range?

[David Hess]

With a 2N3904, I measure 35pA at 3v BC and 48pA BE, so yes, not bad!

That must be a bad one.  I usually get more like 2pA.   

2N3904s from some manufacturers are better than others.

If you want an inexpensive low leakage diode with guaranteed specifications, then the 2N4117 JFET is the way to go.

What lab instrument are you using to measure currents in the tens of pA range?

There are a couple of easy ways to make measurements in this range.  A standard voltmeter with 100 microvolt resolution and a 10 megohm input resistance has a resolution of 10 picoamps when used in voltage mode to make current measurements.

If I need something better, then I configure an LMC6081 as an integrator, have it measure its own input bias current which will typically be about 2 femtoamps, and then use it to measure whatever I am interested in.

[Howardlong]

What lab instrument are you using to measure currents in the tens of pA range?

Keithley 237 SMU with triaxial probing.

I do a bit of nano amp ultra low power microcontroller stuff, I used to make do with DMMs and uCurrent, but an SMU is really the way to go. Recently I bought a Current Ranger, which is OK but it loses accuracy quite badly the closer you get toward the bottom of the uA and nA ranges.

[Howardlong]

First thanks for the input on this thread.

Attached is my design which seems to tick the right boxes.

o Behaves throughout the input voltage range (0V - 3.3V), including output always off in low voltage and no negative resistance characteristic.

o Push-pull output

o Low current (80nA at 1V, 94nA at 3.3V)

o ~30mV hysteresis (D3 & D4)

o Switch voltage 1.92-1.95V

The 1N4148 diodes are used in reverse bias mode for their leakage as they're a lot cheaper than ~100Mohm resistors.

You need to be careful when measuring current because the LEDs also act as photodiodes, potentially giving inconsistent measurements.