2
\$\begingroup\$

Buck regulators often specify the value and placement of the output capacitor. Typically 10µF placed very close to the buck regulator.

However, what would the effect be of replacing the 10µF output capacitor with 10x 1µF capacitors. Where only one of those are placed immediately at the buck regulator, and the other 9 are decoupling capacitors at the load. Within an inch or so from the buck regulator.

The 1µF capacitor close by the buck regulator should handle the high frequency components of the buck switching just as well or better than a 10µF would, and the total capacitance is 10µF so the regulator loop should function properly.

For point-of-load buck regulators, this distributed output capacitor approach would have quite some practical advantages nowdays when 1µF is readily available in 0402 footprint.

Is there a downside which I am not seeing?

Clarification 1: The current at the output of a buck regulator is DC plus a small triangle wave at the switching frequency (typically 1-2MHz). Triangle waves do not have any significant HF content. Examples can be found here: http://electronicsbeliever.com/sizing-the-inductor-of-buck-converter-and-setting-its-operation/

Clarification 2: Physically smaller capacitors have lower ESL (which is better). See for example: SMD capacitor package size and high frequency performance

\$\endgroup\$
3
  • \$\begingroup\$ I don't know it really works. Is there any ripple requirement for case. ripple requirement decides actual capacitance value in vicinity.There is always a relation between LC filter frequency to actual switching frequency for regulator stability. \$\endgroup\$
    – user19579
    May 18, 2020 at 16:37
  • \$\begingroup\$ If several of the 1uF capacitors are sufficiently close to the regulator, their series inductances may be in parallel, therefore reduced, improving performance. This should be testable if you model the inductance of the most important connections in simulation. \$\endgroup\$
    – user16324
    May 18, 2020 at 16:44
  • \$\begingroup\$ Actually, just a single 0402 1µF capacitor has lower ESL than a 0805 10µF capacitor, so from a ESL perspective the distributed approach will perform better. \$\endgroup\$ May 19, 2020 at 7:31

2 Answers 2

1
\$\begingroup\$

There could be a downside, but it depends. The manufacturer's recommendation is usually based on characterization with a given capacitor or range of capacitors. The safe approach is to stay within those guidelines. Be sure you know if the capacitance specified is effective capacitance, or a specific recommended capacitor.

By deviating from the guidelines you may introduce extra ESR, moving the ESR zero and changing the response of the control loop. 0402 capacitors have worse derating with voltage and can have worse ESR) so the extra parasitics and loss of effective capacitance could affect performance.

Over temperature an 0402 biased near its voltage rating can lose more than 80% of its capacitance, depending on dielectric and value:

enter image description here

enter image description here

You may also not be satisfied with the output ripple given the extra ESL and ESR you're adding. You could run some simulations with various stray inductance and ESR values to get a feel for it.

It would likely cause less of an issue with a current mode topology vs. voltage mode, and it might be fine, but you should at least prototype it, and measure the loop and transient response before committing anything to production.

Further info:

An 0402 1uF cap does show higher impedance than an 0805 case size 50V MLC cap- From Murata's Simsurfing tool here's an 0805 in blue and an 0402 in green:

enter image description here

And here's the derating curve for an 0402 50V 1uF X5R:

enter image description here

And ESR vs. frequency, 0805 in blue again:

enter image description here

So derating might not be a factor for 1.8V rails, but with temperature and tolerance even at 3.3V you're losing significant capacitance.

\$\endgroup\$
9
  • \$\begingroup\$ 0402 generally has lower ESR and ESL than larger packages. 0402 1uF is available with 50V rating. So the voltage coefficient is not a problem. \$\endgroup\$ May 18, 2020 at 16:26
  • \$\begingroup\$ @TimmyBrolin I'm not so sure derating isn't a problem. I checked Murata's Simsurfing tool and you can be well under 60% of rated capacitance at <5V with a 50V 1uF 0402. Just another consideration to watch... \$\endgroup\$
    – John D
    May 18, 2020 at 16:57
  • \$\begingroup\$ There are X7R 0402 1uF capacitors. Yes, derating must always be considered, even for 0805 capacitors. But it is not difficult to find suitable 1uF 0402 capacitors for 3'3V or lower. Have a second look at those murata impedance diagrams you posted. The 0402 capacitors clearly have lower ESL, since their resonance frequency is shown as higher, and impedance is shown as lower at high frequencies. The slightly higher impedance you see at low frequencies is just due to slightly different capacitance at the selected operating conditions. It has nothing to do with ESL. \$\endgroup\$ May 18, 2020 at 18:54
  • \$\begingroup\$ @TimmyBrolin I think the higher impedance is higher ESR not ESL. \$\endgroup\$
    – John D
    May 18, 2020 at 19:14
  • \$\begingroup\$ no. Look at 100Hz. There is a 1kohm difference in impedance there. And the difference in impedance changes with frequency. So it is caused by a slight difference in capacitance value. \$\endgroup\$ May 19, 2020 at 6:53
0
\$\begingroup\$

The main issue with distributing that capacitance is that the ripple currents, which are often the largest unintentional radiators in a design, are now also distributed throughout the design. Depending on your other circuitry or your environment, this may cause crosstalk with other devices or sections of your own device. It's true that you're increasing ESR as well, but depending on your switching frequency ESL may be a more critical issue. Over an area of about an inch the ESR probably remains in spec.

\$\endgroup\$
7
  • \$\begingroup\$ True, but assuming there needs to be distributed decoupling capacitors on the power rail anyway, those currents will be there regardless. Will decreasing the capacitance of the capacitor closest to the buck regulator make that big of a difference for ripple currents? The high frequency component of the ripple currents should only be affected by the ESL anyway. Not the capacitance. And the ESL would actually be improved by using a physically smaller capacitor. \$\endgroup\$ May 19, 2020 at 6:59
  • \$\begingroup\$ Each cap will roughly draw an amount of ripple current proportional to its capacitance. And the ESL improvement you get from reducing the cap size will be the same as the ESL penalty for a couple mm of traces leading to said cap. \$\endgroup\$ May 19, 2020 at 16:40
  • \$\begingroup\$ Just to be clear, exactly what ripple currents are you talking about here? The high frequency ripple currents from the loads? Or the ripple current from the buck output? Realize that the buck output current loop does not have much high frequency content. There is a big inductor in the output loop which blocks high frequencies. \$\endgroup\$ May 19, 2020 at 22:48
  • \$\begingroup\$ The inductor blocks HF voltages, not current. The voltage is stabilized by letting the ripple current flow in and out of the capacitor(s). Your ripple current from the power supply would be distributed throughout the board. \$\endgroup\$ May 20, 2020 at 13:23
  • 1
    \$\begingroup\$ Really? You put AC + DC current into an inductor and you only get the DC out? Better run that through KCL. Inductors resist changes in current, but don't block them. In a buck regulator, the inductor and capacitor form a LPF that minimizes voltage variations on the output, but the ripple current still goes from the L through the C and back to ground. \$\endgroup\$ May 21, 2020 at 15:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.