Boost converter: high ripple at extremely low current

[nsayer]

I've got a battery powered project I've got a rather successful design for at the moment. It's the Crazy Clock - I think I've talked about it before here. The design works, but I have somewhat of an academic question that may lead to a design improvement perhaps.

In a nutshell, it's an ATTiny85 powered by a NCP1402 based 3.3 volt boost converter driven by a single AA battery. The controller sleeps most of the time, and most of that time it draws less than a microamp - closer, in fact, to 100 nA. I've used the microCurrent Gold and established that the average battery draw is around 230 micro-amps, which means the battery will last easily longer than a year, which is just fine.

Since the circuit works, I haven't really paid much attention beyond that, but I happened to take a look at the output voltage of the boost converter and saw that the ripple was actually quite high - something like 150 mV P-P. Hmm.

One of the things I have in the works is that I'm going to swap out the MBRA140 Schottky for a BAT54. My motivation for that was to reduce the BOM cost given that I obviously don't need a 2A rated part for this. I'm also going to swap out the output filter cap for a cheaper one, but I do note that the ESR has gone up to 1 ohm, which is higher than the NCP datasheet recommends (~.25 ohm).

In any event... I suppose the ripple isn't impacting anything because the ATTiny has such a wide supply voltage tolerance, and this isn't in any way an analog circuit. But in the interest of better manufacturing reliability, should I try and reduce that ripple? The only way I would think it would be possible would be to increase the inductance. I suppose the up-tick in the output filter cap ESR isn't going to be helping any.

Is it just too much to ask for the boost converter to keep the ripple under control when the current load is so light?

Or is the NCP1402 perhaps swatting a fly with a jackhammer? The single AA battery is non-negotiable. What else can I do to get 3-3.3 volts regulated out of one?

[T3sl4co1l]

I don't think you can really reduce the ripple from a five pin device; you'd need control of the oscillator frequency or feedback compensation, and that's all internal junk.  What you'd be looking for is a larger value inductor (it'll still be small physically because of the minuscule energy storage at these currents), and maybe some beefier caps anyway (a couple 10uF ceramics should be more than enough?).

I don't know how much ripple is actually a problem for a AA cell.  I doubt it cares much.  But the AC is still lost, so the efficiency isn't the greatest that way.  And at the output, 5% ripple probably isn't too bad, but yeah, it could be better (and would be with a more optimal converter).

Tim

[DanielS]

Increasing inductance is one way of reducing ripple but at very low current, a larger inductor is not going to be very effective.

What size capacitor are you using in there? At 300µA on a presumed 3.3V output, you should not need much of a capacitor to smooth things out and a 1µF MLCCs can be had for less than $0.05 in quantities of 100+. If your PWM operates at 50kHz, your output ripple should be around 10mVpp.

[nsayer]

The OFC at the moment is 68 uF. It's a Kemet T495C686M016ATE250, but in the new design I am going for an AVX F930J686MBA to reduce both cost and size. The inductor is a CBC3225T470MR from Taiyo Yuden

[DanielS]

While tantalum caps might be nice for dampening noise, their high ESR fares poorly at bluntly filtering current coming out of a switched inductor. You need the low impedance of an MLCC cap for that.

Tantalum caps are usually part of a tiered decoupling setup with electrolytic caps providing bulk decoupling, tantalum providing dampening and MLCCs handing fast transients near the devices being powered or around switched loads/outputs.

You might want to try using an MLCC in your application and see how much ringing you get - you might find out you did not need a tantalum cap at all. 10µF MLCCs are available in packages all the way down to 0603 and cost under $0.10 in lots of 100+.

[nsayer]

From the NCP1402 datasheet:

Output Capacitor
The output capacitor is used for sustaining the output voltage when the internal MOSFET is switched on and smoothing the ripple voltage. Low ESR capacitor should be used to reduce output ripple voltage. In general, a 47 mF to 68 mF low ESR (0.15 [ohm] to 0.30 [ohm]) Tantalum capacitor should be appropriate. For applications where space is a critical factor, two parallel 22 mF low profile SMD ceramic capacitors can be used.

From what they're saying, 10 uF wouldn't be enough. Are you saying it might be because of the extremely low current draw (the NCP1402 designers contemplated circuits drawing 200 mA)?

I don't mind experimenting (now is certainly the time rather than later).

Yageo's CC0805MKX5R5BB226  would be a contender (I gravitate towards 0805 as a standard footprint where possible), but would one be enough?

[T3sl4co1l]

Yeah, it's designed for optimal performance at rated load.  Hence the small inductor and big currents.  If that were more like 10mA ripple, it would still be overkill for your purposes (I'm guessing), but would be a lot closer.  And then, even 1uF would be huge, let alone 10 or 68.  But the switcher is probably not designed for those values.

'Spose I should try my hand some time at a low current Joule Thief style circuit.  Something roughly like this would be a start,



But with better bias to get the low idle current.

Tim

[nsayer]

That looks physically much larger than what I've got. Physical space in this application is at an absolute premium.

In investigating the MLCC alternatives, I was directed down a Digikey training propaganda powerpoint. I am beginning to be convinced that trading out the tants for MLCC is worth a try. The space and cost savings are major.

[DanielS]

From what they're saying, 10 uF wouldn't be enough. Are you saying it might be because of the extremely low current draw (the NCP1402 designers contemplated circuits drawing 200 mA)?

If you do the energy storage calculation, 10uF at 3.3V is 54.45µJ of stored energy.
300µA at 3.3V for 20µs (50kHz switching frequency) is 20nJ worst-case drain per cycle

Subtract 20nJ from 54.45µJ, convert that back to voltage, you get less than 1mV droop. As T3sl4 said, you would most likely be fine with 1µF for your tiny load if the converter could work with that.

I picked 10µF because MLCCs are very inexpensive at least up to that size and it is not too much of a drastic departure from your existing 68µF.

Keep in mind that tantalum capacitors cannot decouple noise and ripple any better than the I*R loss of their high(-ish) ESR will let them. If your switcher is ramping inductor current to 100mA before the error comparator turns off the output driver and your tantalum has 1ohm ESR, that's up to 100mV extra ripple right there compared to MLCCs which have almost negligible ESR.

[dannyf]

Is it just too much to ask for the boost converter to keep the ripple under control when the current load is so light?

You have discovered a brand new class of smps: gated oscillator smps. Its grand daddy is lt1073 - check out its datasheet.

The answer to your question is that there is no cure -> it is just the price you pay for low idle current.

[mzzj]

From the NCP1402 datasheet:

Output Capacitor
The output capacitor is used for sustaining the output voltage when the internal MOSFET is switched on and smoothing the ripple voltage. Low ESR capacitor should be used to reduce output ripple voltage. In general, a 47 mF to 68 mF low ESR (0.15 [ohm] to 0.30 [ohm]) Tantalum capacitor should be appropriate. For applications where space is a critical factor, two parallel 22 mF low profile SMD ceramic capacitors can be used.

From what they're saying, 10 uF wouldn't be enough. Are you saying it might be because of the extremely low current draw (the NCP1402 designers contemplated circuits drawing 200 mA)?

I don't mind experimenting (now is certainly the time rather than later).

Yageo's CC0805MKX5R5BB226  would be a contender (I gravitate towards 0805 as a standard footprint where possible), but would one be enough?

just remember that high capacitance ceramics have really nasty voltage coeffient.
http://www.edn.com/design/analog/4402049/2/Temperature-and-voltage-variation-of-ceramic-capacitors--or-why-your-4-7--F-capacitor-becomes-a-0-33--F-capacitor

[mikerj]

Another reason not to use tantalums in very low current applications is the leakage current, which in your case is almost certainly higher than the actual load current.

The datasheet for your current part suggests leakage is <= 0.01CV, so could be up to (0.01*68E-6*3.3) = 2.2uA.  A ceramic cap should be at least a couple of orders of magnitude lower than this.

[nsayer]

From the NCP1402 datasheet:

Output Capacitor
The output capacitor is used for sustaining the output voltage when the internal MOSFET is switched on and smoothing the ripple voltage. Low ESR capacitor should be used to reduce output ripple voltage. In general, a 47 mF to 68 mF low ESR (0.15 [ohm] to 0.30 [ohm]) Tantalum capacitor should be appropriate. For applications where space is a critical factor, two parallel 22 mF low profile SMD ceramic capacitors can be used.

From what they're saying, 10 uF wouldn't be enough. Are you saying it might be because of the extremely low current draw (the NCP1402 designers contemplated circuits drawing 200 mA)?

I don't mind experimenting (now is certainly the time rather than later).

Yageo's CC0805MKX5R5BB226  would be a contender (I gravitate towards 0805 as a standard footprint where possible), but would one be enough?

just remember that high capacitance ceramics have really nasty voltage coeffient.
http://www.edn.com/design/analog/4402049/2/Temperature-and-voltage-variation-of-ceramic-capacitors--or-why-your-4-7--F-capacitor-becomes-a-0-33--F-capacitor

But isn't that voltage effect related to the voltage rating of the cap? That is, if I use, in this case, a 6.3 volt rated cap and am filtering 3.3 volts DC......

That, and I'm still going with the data sheet's recommendation of two 22 uF MLCCs in parallel. If what people are saying is right, it's more than is necessary for a low-current application.

I have a board into OSHPark (it shrank quite a bit - which pleases me greatly) and we'll see how it comes out.

[T3sl4co1l]

More by size than voltage rating.  A 0.1uF 16V is pretty much identical to a 50V part, if they're both in 0603 or whatever.  But a 16V 0805 will be good to much more than 16V, while a 16V 0402 won't be good to 6V.

Why they even rate them by voltage, I don't know: most ceramics break down (regardless of rating) at several kV, provided you do the test under dielectric fluid.

A fair and consistent rule would be to rate them at the voltage where capacitance drops by a fixed proportion, such as rated tolerance, or something else convenient (20% or 50%).  This is generally how inductors are rated*.  Alas, no international body regulates either type of component, and we are left with a dearth of bad or completely absent data in both fields.

*But not ferrite beads, which, being ungapped ferrimagnetic materials, saturate easily, and are an even better analogy to ceramic caps, which are themselves, ungapped ferroelectric materials.

Tim

[nsayer]

The variant of the board with the two 22 µF output and 10 µF input 0805 MLCCs works very well. It even seems to average 30 µA less current - perhaps the leakage that mikerj referred to. The ripple even seems like it's a little better. The BOM is cheaper and the board is quite a bit smaller now (25.4 x 12.7 mm now).

[Marco]

Have you taken a look at the scope at how this ripple is created?

I just don't see how a single pulse could do this ... the current for a single pulse will be less than 160 mA by my reckoning, that should give you a couple mV at most on the capacitor from charging and 40 mV from ESR.

[nsayer]

I haven't attempted to measure the current of the micro controller in the built version of the circuit. I have measured the equivalent circuit with through-hole parts on a breadboard with a bench supply. What I have measured is the output voltage (ac and dc coupled) and the input current of the boost converter.

With the MLCCs, the ripple is around 120 mV - measured directly on the ATTiny's Vcc pin. If I zoom in with the scope I can very distinctly see the difference between idle and running. Idle has the characteristic sawtooth of discontinuous mode or hysteretic switching. During the active phase the absolute magnitude of the ripple doesn't change, but its frequency goes up quite a bit.

120 mV is ~3.6%, so I don't have much to complain about - particularly given that the acceptable voltage range for this chip is so wide and given the application.

[DanielS]

120 mV is ~3.6%, so I don't have much to complain about - particularly given that the acceptable voltage range for this chip is so wide and given the application.

Most chips have separate specs for ripple and nominal voltage so you cannot assume the chip will always work properly just because your ripple is still within the chip's input voltage range - the ripple spec will often be only a fraction of the nominal voltage tolerance.

[nsayer]

Most chips have separate specs for ripple and nominal voltage so you cannot assume the chip will always work properly just because your ripple is still within the chip's input voltage range - the ripple spec will often be only a fraction of the nominal voltage tolerance.

Sure, but the change to MLCCs has reduced the ripple, and the design even before then was field-proven successful.

[DanielS]

Sure, but the change to MLCCs has reduced the ripple, and the design even before then was field-proven successful.

Field-proven on how many samples? If you exceed the manufacturer's ripple specifications, there is no guarantee all specimens can be expected to be happy with it and you may end up with unexpected field failure/glitch rates.

[nsayer]

Field-proven on how many samples? If you exceed the manufacturer's ripple specifications, there is no guarantee all specimens can be expected to be happy with it and you may end up with unexpected field failure/glitch rates.

Turns out the ATTiny datasheet doesn't have a specification for ripple that I could find.

In any event it's all moot. I figured out where the high ripple was coming from.

I've been doing all of this testing with my bench supply. Well, just by itself, there's ~50-60 mV P-P of ripple with no load (I never said it was an *awesome* bench supply).

I repeated all of the measurements with a AA battery and all variants of the circuit under all conditions exhibit less than 20 mV P-P ripple at the controller's Vcc pin.