High frequency buck regulators and MLCC frequency behaviour

[justanothername]

Hi.
So I recently thought about some of those high frequency buck modules for a very space constrained application and I stumbled across this nice device:
http://www.ti.com/lit/ds/symlink/tps82085.pdf
Switching at 2.4 MHz. Would you look at the recommended output capacitor from the datasheet, it is this type:
https://product.tdk.com/de/search/capacitor/ceramic/mlcc/info?part_no=C2012X7S1A226M125AC
What is going on? The capacitance seems to drop significantly above 1MHz. As stated in the datasheet, I should, after considering the bias effect, provide a Cout of at least 8µF.
What am I missing?

[T3sl4co1l]

The capacitance mainly matters to the control loop, which should roll off somewhere below 1MHz; in any case, it would only need to be compensated against ESL in the cap, which is what's causing the capacitance to flip.  And that's not really a problem, as ESL causes a phase lead, not lag, so if anything it tends to help with compensation.

At high frequencies, all that's really needed is a low impedance.  The phase angle (capacitive, resistive, inductive) doesn't really matter.  At a couple (2 or 3) milliohms, it's quite low.

Mind, at very high frequencies (100s MHz), the ESL becomes relatively significant, and switching edges are let through.  The same goes for trace lengths and ground paths, which can generate common mode or ground loop noise in the surrounding circuit.  To prevent this, place components close together, over solid ground plane, and if possible, arrange the input and output filter capacitors so they are nearby (acting to keep the input and output voltages as near to the same ground point as possible).  Add another LC filter stage (to both input and output) if necessary.

Tim

[justanothername]

Thank you T3sl4co1l.
Judging from several datasheets those types of high frequency switchers seem to work well above the resonant frequencies of their output capacitors.
However, it is difficult for me to understand how the capacitor is working as capacitor since the inductance becomes dominant.
Maybe it just does not have to be capacitive for the switching frequency but for the maximum transient frequency of the load and the control loop.
Will it work right right at the self-resonant frequency when it is basically a short?

EDIT: It seems that my question is answered here:
https://www.eevblog.com/forum/projects/ceramic-capacitor-characteristics/
short: Impedance matters at the switching frequency no matter what kind of.

[OM222O]

my main question would be: why on earth would you run it at that frequency?
There are a lot of modules with crazy high switching frequencies, but that is to allow for more phases with smart doublers which basically chop that signal to multiple phases; e.g: chop 600KHz to 300KHz signals for 2 phases.

If you have a look at any high power buck converter, the sweet spot is that 300KHz to have low ripple without throwing out the entire efficiency of the module (higher switching frequencies = higher switching loss and it's pretty linear!) the ripple is further suppressed by chokes (inductors) and capacitors which are tuned for the application to allow good transient response as well. You can have a look at any motherboard / GPU power delivery. they are purely buck convertors as the main supply rail is 12v and silicon chips require something like 1 to 1.5 volts (dangerous territory!).

[justanothername]

@OM222O
as INITIALLY said: space constrained application where every mm² counts -> smaller inductors. In this case it is a die embedded into a pcb with inductor mounted ontop. 2A @ 2x2mm
and: that is NOT the main question. The main question was how this specific product works with the capacitor stated in the datasheet, not why anyone would use it.
if you ask my grandmother (if she would be still alive) she would say that the MAIN QUESTION is why on earth anyone would sit in front of a computer and how could that be called "work".

[OM222O]

well I really hope you have the heat sinking to deal with the horrific efficiency at those frequencies ... 2.4MHz is insanely huge as I mentioned before, you should expect a lot of output heat. have you done any thermal simulation / testing?

[justanothername]

thank you for your concerns

[thinkfat]

Have you tried TIs online Workbench for their power products? I "designed" a buck converter using an LMR14030, it worked quite well with the suggested components and layout. Might clear some doubts about component choice...


Gesendet von meinem Nokia 6.1 mit Tapatalk

[NorthGuy]

Although this doesn't answer the main question, TI makes insanely small DC converters, such as TPS8267xxx series. They call it MicroSIP. They're a little bit bigger than 2x2mm.

edit: TPS82085 is a good example. It can do up to 3A. It switches at 2.4MHz, but it is still about 3x3mm and it does require extra parts.

edit2: and now I have read the original post - TPS82085 is how it all started. LoL!

[magic]

Judging from several datasheets those types of high frequency switchers seem to work well above the resonant frequencies of their output capacitors.
However, it is difficult for me to understand how the capacitor is working as capacitor since the inductance becomes dominant.

I used to worry about such things until I realized that it basically boils down to capacitance being large enough that capacitive reactance is even lower than ESL reactance. For power filtering low capacitive reactance is exactly what's desired, as long as achieving it doesn't compromise ESL (larger package / electrolytic). As T3sl4co1l said, you could start to worry at higher frequencies where inductive reactance of MLCCs becomes several ohms. I've seen appnotes out there on how to bypass many-megahertz digital circuits with multiple MLCCs resonant at appropriate frequencies because simply slapping a 100nF on it just doesn't work anymore.

[Ice-Tea]

well I really hope you have the heat sinking to deal with the horrific efficiency at those frequencies ... 2.4MHz is insanely huge as I mentioned before, you should expect a lot of output heat. have you done any thermal simulation / testing?

I'm not sure why you bring this up... It's an established TI product with efficiencies between 80-90%. It has a nice thermal pad. Connected to a decent GND plane there shouldn't be much trouble.

[ocset]

I remember once  measuring output ripple voltage of a 1MHz SMPS, (with ceramic MLCC capacitors) and reported that the ripple voltage was far less when a much lower  output capacitance bank  was used…..(because the ESL of the lower output capacitance bank was lower, since it had lower value ceramic caps in it).
This nearly resulted in me getting sacked, until they were brought to  realise  that the ESL was at play.
(They had mistakenly just assumed that “Higher output capacitance means lower output ripple voltage”, which of course is folly)

[justanothername]

I'm not sure why you bring this up... It's an established TI product with efficiencies between 80-90%. It has a nice thermal pad. Connected to a decent GND plane there shouldn't be much trouble.

irritating, isn't it?  Not even remotely connected to my question as well..

[OM222O]

the voltage controller itself won't be an issue  the poor mosfet that has to switch at that frequency will be.
you will require active cooling at 2.4MHz for sure ... unless you're drawing something like 100mA  as I said before it's switching loss, not conduction loss  he will realize it sooner or later

[magic]

Read the datasheet, dude.

That being said, this part is indeed quite impressive although not really among the craziest things out there.

Intel's® 4th generation Core™ microprocessors are powered by Fully Integrated Voltage Regulators (FIVR). These 140 MHz multi-phase buck regulators are integrated into the 22nm processor die, and feature up to 80 MHz unity gain bandwidth, non-magnetic package trace inductors and on-die MIM capacitors. FIVRs are highly configurable, allowing them to power a wide range of products from 3W fanless tablets to 300W servers.

[thinkfat]

the voltage controller itself won't be an issue  [emoji14] the poor mosfet that has to switch at that frequency will be.
you will require active cooling at 2.4MHz for sure ... unless you're drawing something like 100mA  as I said before it's switching loss, not conduction loss  he will realize it sooner or later

Hm, the LMR14030 I mentioned allows for roughly 90% efficiency, I run it at 1.6MHz or thereabouts, and at 15W out it will get to about 55 degC. On a rather small PCB. Of course the inductor and diode are external, that helps dissipating the heat.

Gesendet von meinem Nokia 6.1 mit Tapatalk

[Ice-Tea]

the voltage controller itself won't be an issue  the poor mosfet that has to switch at that frequency will be.
you will require active cooling at 2.4MHz for sure ... unless you're drawing something like 100mA  as I said before it's switching loss, not conduction loss  he will realize it sooner or later

If only devices would exist that have the FET(s) integrated or somesuch... In that case, parisitcs would be minimized, the FET would be well matched to the application etc. etc.

Oh, if only such devices would exist!

[T3sl4co1l]

You say that, but I've seen integrated switch (regulator)s, of the synchronous switch type, that had just enough dead time, and apparently just the right doping profile, to exhibit drift step recovery behavior.  Sub-1ns pulses from a regulator, who'd have thought!

Indeed, it's frustrating that most regulators don't have separate drain pins, so you can't control the parasitics yourself.  You're forced to hard-bypass them, lest their control circuitry go crazy (or, so one would presume; alas, I've not actually tried, yet..).

On that note, though, the ones with internal LDO and external bypass pin, should be able to be snubbed, at least a little.

Tim

[jbb]

On the ceramic capacitor front: as DC bias is applied, the capacitance drops. This could move the self resonant frequency up a bit.

On the integrated modules: they’re a bit hard to rework. When you apply heat and try to remove them all the subcomponents can come unstuck.