These circuits take a lot of parts to do a job that you can do with modern high frequency stuff with a lot lower cost and parts count.
The normal point of a capacitive doubler is either to give you a voltage you need without a lot of extra parts count (often negative) or to generate a very high voltage.
I did a tapped inductor boost last year to take 3V input to 80V output (at not-much output current, I forget exactly what it was but it was mostly a bias voltage; also, the actual output voltage was DAC-set and could be quite low, so the loop dynamics were unpleasant). It was definitely annoying to wrap my head around, and very annoying to select the inductor (Würth has a nice OTS series, at the usual Würth prices; HVM would likely want a custom or semicustom design) but it just plain worked the first try and continued working through the usual stress tests and also the unusual stress tests of the Very Expensive Load™ getting itself Very Expensively Killed™ (for non-power-supply reasons). I was really happy with that converter, that kind of step-up ratio isn't easy and it just worked.
A basic search coughs up a bunch of papers from academic paper mills, and I don't see obvious links to an OTS series from Wurth Elektronik.
I misremembered what design I actually ended up building. It was nominal 5V to 75V but was tested at a wider range (including the mentioned 3V to 80V). The variable voltage was implemented later, in a linear stage, which did in fact have a number of issues.
ADI app note AN-1126 was very helpful in this design. It pushes for a different topology, yet compares with this one and others in some detail. Its arguments against tapped inductor designs are threefold: (1) sometimes EMI issues on the switching node (fair, and a real concern, but not interesting for what I needed to do), (2) the inductor is annoying to source (very true), and (3) the main technical objection is that demands on the output rectifier are high and might require you to use something crappier than a Schottky. That last one is true in general but for this design in this decade, I was comfortably within medium voltage Schottky territory, and so their main objection was a complete nonissue. That looked good to me so I went ahead and built it, and was not disappointed.
The inductor was 744889030330 (say that three times fast) from the WE-MTCI family which worked a dream. I don't remember why this exact switcher chip was chosen but I do remember having a lot of candidates and so the choice was kind of arbitrary. The control scheme type it uses is important, though. I don't think the zener did anything but being paranoid I wanted to have the footprint there for the prototype build rather than not have it.
THE PREVIOUSLY PROMISED WARNINGS WHICH ARE ACTUALLY REALLY IMPORTANT SO I AM USING CAPS:
1. 80V can kill you. Really, it can. Use caution!
2. This is the design I sent off for fab (and perhaps not even the final version, if the folder notes are any guide). It is not the redlined, working version. I believe this stage was OK. Maybe it needed a bit more output capacitance? Certainly I know the following stage, not shown here, had severe issues (most of them stupid and obvious once noticed). So treat this as good inspiration, not ground truth.
3. Did I mention that 80V can kill you? USE CAUTION.
I will note that we're all talking about 10+ year old chips and technology--mostly prior to GaN.
Apparently, I'm going to have to dig through a bunch of stuff to see what is current. While the topologies don't change, the tradeoffs between them do as technology changes.
Edit: Stare at the LT7890/1/2/3 series for GaN stuff by way of comparison
It is also uncompromising and brutally difficult to get working well. GaN-FETs love to commit suicide in new and entertaining ways. And LTC7890 is not something I would want to implement in any design (though of course I'd suck it up and do it if it was the right choice).
99% of the market is traditional boring stuff because 99% of the market is well served by traditional boring stuff.
They have insane power density and lack of wound inductors means there's nothing that causes problems as you push into the kilovolts.
But really they're just more efficient (integer ratios, where `n-1` not prime) than classical Buck/boost topologies.
GaN stuff can be 99%+ efficiency. The frequencies are multiple MHz which shrinks the inductors significantly--sometimes allowing PCB based coils (See Anker 120W teardowns).
Integer ratios can be done with GaN and class 1 ceramics and stray inductance to charge the GaN drain, so they only experience ZVS (they don't need ZCS). Use a resonant gate drive. Efficiencies around 99.8% are feasible for a doubler/halver based on the "3L FC buck" topology, in the 10~30 MHz range.
The datasheet mentions low profile a lot. That does make sense as one can make a flat, high quality, capacitor. Making a flat high quality inductor is harder and probably more expensive and likely consumes more volume overall. I can imagine some applications where being flat is important, like the back of a panel.
EMC
https://cdrdv2-public.intel.com/632833/ec2650qi-datasheet.pd...
Enpirion's first products - integrated-inductor DC/DC converters - were limited to fairly low input voltages. This allowed them to run very fast with the technology of the time; they needed to run fast, because the integrated inductor was not very large. But this was severely limiting Enpirion's market: if you wanted to make something 5V-powered, they were great. But a PC motherboard application with 12V input?
This device was the answer: convert 12V to 6V with this switched capacitor halver, and then use other Enpirion parts. I don't think it was super successful, though, because at this point you were no longer winning on board complexity by using integrated inductors.
FWIW: the didactic trick of imagining the floating capacitor "carrying" charge from one "place" to another was really good. That's not the way most treatments talk about charge pumps, and I think it's a lot cleaner.
All circuits are just very basic circuit element behavior. In fact a charge pump is a decidedly counter-intuitive thing, up there with things like "why a long-tailed pair makes a differential amplifier" or "how tf does a buck/boost regulator work?".
This is like looking at a 30-line function implementing a FFT and announcing "it's just very basic C code".
Compared to an FFT, which would require a lot of math insight it seems to me a charge pump is trivial because charging and discharging is what capacitors do, it is the most basic thing you can do with a cap. In programming it would not be the equivalent of an FFT, but more like the equivalent of putting a value in a variable and expecting that value to still be there a while later.
Now, I came into software from hardware, to me software is having an infinite parts box so what's trivial to someone that started in hardware is probably entirely different from the view a software person has of the hardware world. Which is one reason why analog is such a barrier to programmers: they tend to look for the place where the state of the system is stored, rather than that the whole circuit is a continuous function of its input and what happened before. Getting analog circuitry to behave predictably can be quite tricky because of that and just hanging a probe of it will influence the circuit.
Resistors are easy, for the most part grade school math. Capacitors harder but not a lot harder, high school math, some basic integration required. Coils are where it gets more interesting and difficult math wise. And of course parasitic components are everywhere and can make your life harder, especially at higher frequencies or, in the case of single pulses shorter rise and fall times. A piece of wire is a resistor, capacitor and a coil for free all at once. Stripline takes advantage of that fact but without tooling it is next to impossible to work out the math and get predictable results (again, for me, I know people that can do this intuitively).
Multi-pole filters, high gain amplifiers, clean oscillators, those rise to the level of that FFT (for me).
The long tailed pair I agree with, that's a puzzler.
If you don't have one yet (and I assume you do, but just in case): get a scope. They're pretty cheap nowadays and a good scope will do wonders for such insights. I only got one when I was an adult and finally could afford it and it was an absolute game changer. If I had had one earlier I would have gotten a lot more mileage out of my time, especially when troubleshooting broken hardware. You don't need a fancy one, 100 MHz dual trace is more than good enough but that's table stakes now.
One big insight for me is that hardware 'is' and software 'does'. You don't tell a piece of hardware what to do, you design it to 'be' something and a computer simply (ok, maybe not so simply) needs to be told what to 'do'. This took me a long time to really grok after making the jump from hardware to software.
A good model for charge pump and duct switching converter understanding.
If you charge the bottom cap more and the top one less, you can jack the voltage toward the power rail.
A buck as well as boost-buck converter could be produced without inductors.
Indeeed, I found an article about exactly this: https://www.allaboutcircuits.com/technical-articles/boosting...
The article references the LTC3265 IC, whose datasheet says "The LDO output voltages can be adjusted using external resistor dividers" (connected to the ADJ pins).
2x the battery charge voltage is requested from the power supply, e.g. 8.6V if the phone is trying to apply 4.3V to the battery. This way, the phone doesn't need to run any complicated and heat-generating voltage regulation, just the halver, while still being able to request more than the standard 5V over the cable (allowing it to draw more than 15W of power over a standard cable).
You can e.g. stack two half bridges, connect their switching nodes with a flying capacitor, and declare the connection of the stack the new switching node. If you use a series induction from the switching node to the load, that's a "3L FC buck". The inductor is not needed for halving.
[0] https://www.analog.com/media/en/technical-documentation/data...