Horribly high frequency switching regulators - why?

[nrxnrx]

Hi!

What applications call for a supply switching at 10MHz?

Before I realized I could waste some very knowledgeable people's non-billable hours, I tried google. Came up with fast ("reaction" time) regulators and wireless power.
You need to control a lot of power _very_ fast? When? I can't imagine 20A pulses right next to a processor would make it happy.

I guess 10MHz would be necessary to modulate/control the supply to other very fast acting "things". What could those be?

blueskull?

--
Mihai

[Ice-Tea]

- smaller inductor
- smaller decoupling caps
- better or at least raster transien response
- perhaps also for something like ADSL or vdsl: switching noise outside the baseband.

[fivefish]

Higher frequencies allows use of smaller inductors & caps = more compact design.

[wiss]

At 10 MHz switching you could probably reach MHz GBW

[T3sl4co1l]

Physical size of a coherent design goes approximately as sqrt(f).  The transistors don't get much smaller, but they do have to be faster for their size; mainly, bypass caps and filter inductors get smaller, but not proportionally smaller due to losses.

Power level is a volume scale, and the speed of light is a linear distance scale.  Dynamics in a switching circuit are fundamentally limited by the speed of light, because the speed of light is inversely proportional to the inductance and capacitance of space itself, and any wires or paths running through space.  This limits the frequency and power level for a given technology:
0. Familiar discrete designs.  Limited by layout size, component leads and such.  Works up to a few MHz at modest power levels (100W?), or ludicrous power levels (> MW?) at lower frequencies (< 10kHz?).
1. Switch size is minimized by moving the transistors/diodes onto the same chip (monolithic internal-switch regulators).  Where we are now; good up to perhaps 10s of MHz for POL converters.
2. The main bypass cap must be moved onto the die as well.  And probably the inductor, or part of it.  Transmission line effects in the pins, connecting traces, and in the inductors themselves, will limit the usefulness of proper inductors in the 100MHz range.  Plus EMI issues (those frequencies don't stay in wires very well, and it would be foolish to trust a PCB layout jokey to handle it!).
3. Finally, even a monolithic design is too large (and perhaps too lossy as well) for full power in a single converter, and multiple phases must be built, with onboard inductors too.  We're not at this level yet, but expect fully monolithic converters pushing >100MHz (at reasonable efficiency and power) in some years.

(We're already seeing the beginnings of such devices: ADI's isoPower chips for example, which run rather fast (~250MHz??), and also at low power levels and rather poor efficiency.  Which is what you should expect from an early, ongoing development.)

There's no fundamental limit to "transformium", i.e., a block of arbitrary matter, which converts V/I on one side to another V/I on the other, through any arbitrary means (e.g., distributed nano-converters?).  The problem is that, each individual converter cell must be smaller and smaller, as frequency rises, and more and more of the necessary components must be integrated with them (not just the controller, driver and switches, but input and output filtering as well).  They can be arbitrarily small (~GHz, say?), but the power level per cell must necessarily be small as well (so a great many of them are necessary, in parallel, for useful total power levels).

(In reference to "computronium", a block of arbitrary matter which possesses computational powers, supplied by, say, electrical power, or a temperature difference.  A series of Dyson spheres, made of such material, could turn a star's power into "The Matrix" for quite a population.)

Tim