Eensy weensy switchmode power supplies

[klugesmith]

Helping with a board bring-up at work, I found SMPS converters smaller than I knew existed.

Little buck converters with 1 uH integrated inductor, switched at 2 MHz.  Input voltage up to 17 V.  Output can be set between 0.9 and 6 volts, with up to 3 A of continuous current.

[T3sl4co1l]

Cute.  I've seen smaller, but they didn't seem to work very well.  (Some brand acquired by Intel, single chip POL converters -- very high frequency, basically enough logic-level transistors packed together to make a power switch, then a little bit of inductance somewhere.  They were incredibly sensitive to nearby EFT pulses, somehow.)

LT (now ADI) has had modules for quite a while, with even more integration than this -- less critical on external capacitance for example.  Though I've also seen a few that were an EMC nightmare...

Probably we'll be seeing more and more modules, as GaN continues to mature.  It's rather difficult to use transistors so fast, at any kind of power, but all that difficulty can be solved before the customer touches it, using a module.  (You already see integrated half-bridge modules for example, gate drivers and supply bypass included.)

Tim

[Weston]

I use this modules on my tesla coil PCBs and research PCBs, mostly to supply the voltage rails for FPGAs. They are pretty great and surprisingly cheap.

They save a lot of design time, which is nice for small quantity stuff where BOM cost is not an issue. However, are not even that much more expensive than a discrete inductor and DC/DC IC. Inductors are strangely expensive in small quantity.

Here is one on the top left of the row of regulators on top of the FPGA on my cuteQCW driver board, it almost blends in with the decoupling capacitors

[klugesmith]

Very pretty, Weston.  My first view of actual part, because the board I'm dealing with is accessible only from back side.

What drew my attention was output voltage ripple in the application (power for some clock fanout buffers).
Roughly -53 dBm at 2.1 MHz and a few harmonics; that's around 1 millivolt each.

[rikkitikkitavi]

Qute. And quite useful. The ripple is quite amazing, I wonder if they use a post filter?

[klugesmith]

Sometimes the ripple is not good enough.  Here is the measured spectrum of a reference clock, shown with two different sources of Vcc for a distribution buffer IC.
Yellow:  a TPS82130.  The spurs are at Vcc ripple frequency and its harmonics, and represent deterministic jitter and/or AM.
Blue: a benchtop lab supply.

Fortunately, it appears that when the Enable pin of TPS buck converter is pulled low, the main Up and Down FETs are both turned off.  So V_out is not clamped to ground, and can be driven to normal voltage by another source.

[johnf]

Yes at 68 ish dB down it is in spec
But boy would i not want that in a radio!!!

[T3sl4co1l]

If the sidebands are in phase, it really is AM, and that's the definition of what you're looking at -- a square wave with varying height.  Input thresholds vary proportionally, so it's no problem for local timing.  It may be a problem crossing supply domains (AM-PM mixing), or for a reference standard (making a nice clean e.g. 10MHz sine), where I guess you should use very quiet supplies and a DDS?

Tim

[klugesmith]

Yeah, spectrum analyzer doesn't show the +/- sideband phasing, which could let us separate AM from PM.  Was plumbed to one side of a differential PECL signal, whose common mode ripple stands out in spectrum at baseband, and whose voltage swing sensitivity to Vcc could be measured.  Looking forward to view with differential input on oscilloscope (plumbed, not probed).  How clearly can we resolve the 2.1 MHz perturbation, whose amplitude is less than 1 ps even if entirely PM?  Mainly for academic interest, because it's no big deal to get rid of the switching Vcc source.

[T3sl4co1l]

Ahh, PECL, then my assessment isn't quite true; I had CMOS in mind.  Though I think either the difference between levels, or the mean, or both, still varies with supply, so it might not be that far off after all?  And yep, a full differential measurement might help, too.

Agreed, a few inductors and capacitors are quite a bit cheaper than a picosecond resolution scope.

Tim