Surprising charge pump

[cdw]

I needed a small capacitance (50-100nF, not discharged aggressively) periodically charged to 1200V as part of a project so I put together a little flyback converter with an off-the-shelf surface mount flyback transformer which works nicely.

Chatting later, a physicist friend asked why I couldn't just use a voltage multiplier. Turns out he didn't mean a few stages on the end of a flyback/boost/resonant driver, but literally chopping the 12V rail into an epic 100x charge pump!

I tried to use spice with vaguely realistic models of diodes, capacitor losses and parasitics to show why such long ladders don't work out in practice, but it stubbornly insisted on performing well in simulation.

Real-world demonstrations are more fun anyway, so I dropped 120x cheap B1040X 40V Schottky diodes (2 pence each on LCSC) and 120x 10u 35V 0603 MLCCs (2 pence each on LCSC; <1uF with DC bias of course, but that's fine) on a 30mm x 45mm board, probably stretching the creepage rules a little.

It stubbornly insists on working well in reality too. Driven at 500kHz from 12V with a little bootstrapped 4 x NMOS full-bridge from my spares pile, it charges fully in under 20ms. Stopping it at 1200V and discharging to 1000V through a 100k test load shows an effective capacitance of around 40nF, rising as it discharges further.

That's about right for the theoretical 4C/n with C roughly 1u - as predicted by Simsurfing with the 20V DC bias on these little capacitors, which I know are pushing the limits of class II dielectrics.

I'm very surprised that such a long charge pump works well, let alone ends up small and cheap. I guess my prejudice here is outdated... or maybe it has always been wrong?

Are they ever used directly like this without an inductive boost stage, e.g. in lower power applications? Feels like they might be quite good on the EMI front?

[KE5FX]

Not uncommon for negative rail generation, but also normally done with only one or two C-W stages.  Haven't heard of a 40-stage multiplier, but the proof is in the proverbial pudding!

[ejeffrey]

I've definitely seen many stage multipliers work fine for a static bias voltage.  It's the easiest way to build something with whatever you have on hand.  A small wall transformer and a handful of diodes is enough to get started, no active semiconductors required at all.  Obviously it will work a lot better with a higher frequency oscillator but you can already do it with 60 Hz.  It's also very suitable for moderately high voltages since with leaded components you can build it in free space.

As long as you need negligible current and you don't need to operate at high temperature where diode leakage will become a problem you can do a lot.

[voltsandjolts]

a little flyback converter with an off-the-shelf surface mount flyback transformer

What's the off-the-shelf transformer?

[cdw]

a little flyback converter with an off-the-shelf surface mount flyback transformer

What's the off-the-shelf transformer?

Sumida 0399-T208: https://www.digikey.co.uk/en/products/detail/sumida-america-inc/0399-T208/9960781

1:30 (or more accurately, 1:10+10+10) with Lpri 40u and Isat somewhere over 2A, though the datasheet is too rubbish to include that explicitly. Wish it had a Coilcraft style datasheet, but they seem to stick at 1:10 for capacitor-charging type transformers, presumably because that fits the photoflash people better.

They're designed for the fancy integrated, isolated LT8304 controller with primary-side sensing, but they work fine on boring direct feedback converters. I've never actually tried using it with the LT8304 it was designed for: the lack of unloaded output stability puts me off primary-side sensing and it's quite a pricy chip.

[cdw]

As long as you need negligible current and you don't need to operate at high temperature where diode leakage will become a problem you can do a lot.

I've never wanted more than low single digit watts into tiny capacitors at these voltages (e.g. 100nF to 1200V in 50ms is 1.4W) so I've never gone beyond negligible current.

My gut instinct was that 10x 12V would work fine, 30x would start to sag badly, and 120x would just be fantasy land, but I'm eating my words. The other thing that amuses me about the charge pump is that, instead of charging an output capacitor, you can use it *as* the high-voltage storage capacitor.

Seeing "10uF 35V" in 0603 for 2p makes me feel quite old too, even if it does drop by 90% with dc bias.

When I lay out a flyback, I keep the ground plane out from under the secondary diode and capacitor to avoid parasitic coupling with the giant dV/dt secondary spikes into the diode. But every node in a charge pump just has some (possibly large) DC bias overlaid on AC <= the +/-12V input, so presumably ground plane is good rather than bad there?

[mtwieg]

Last time I build a charge multiplier, I found that it worked great on a breadboard with leaded components spread out, but once it was translated to a compact design using SMT components its performance was very inconsistent and overall poor. I etched the PCBs myself (no soldermask), and I'm guessing insufficient cleaning after etching was causing leakage currents. So probably not a realistic example, but I wonder how such a circuit on a "good" PCB would hold up after operating for thousands of hours in a normal atmosphere. I'm guessing a conformal coating would help a lot.

[741]

Fundamentally - why are charge-pumps typically 'weedier' (worse compliance) than inductor based SMPS?

Loosely - I think it matters that inductors have a rate-dependent V:I relationship, you can design using this property to advantage.

With capacitors, all you can do is wait for them to charge, then switch them in series or parallel. Well..admittedly there is also that surprising circuit that slightly boosts its input voltage using only R&C; I suspect that is necessarily very weedy.