Multiple capacitor values at the input of DC-DC converters

[derGoldstein]

In EEVblog #859 Dave explained the need for multiple bypass capacitors with different values if you want to minimize the noise on the input of ICs. I was wondering how much this would be necessary in small DC-DC converter circuits.
Here's an example chip, but there are many like it:
http://ww1.microchip.com/downloads/en/DeviceDoc/25173A.pdf

What I'm trying to do is to minimize the number of different parts (simplify the BOM). In the MCP16251 chip the typical application diagram shows a 4.7uF cap at the input and a 10uF cap at the output. However, by placing a 10uF on both sides I can shorten the BOM by one part (and they cost almost the same). In this particular case, I'm fairly certain it wouldn't change anything, because the difference isn't much. However, sometimes you see two different capacitors with two different values on the input, especially when you look at the schematic of demo boards where they often over-design things.

It seems like the input caps in these circuits are pulling double-duty: they have to act as the DC-DC converter circuit's input cap, but also act as the IC's bypass (since these are mixed logic-and-power ICs).
To simplify things a bit, the circuits I'm talking about only use ceramic caps, I wouldn't expect a large electrolytic to act as the bypass cap.

Some datasheets don't give you much to work with in terms of specificity. Some of them just give you a value for the total capacitance on each side  -- "place 22uF of low-ESR capacitance at the input". In those cases I really have no information to work with.

Sorry if this is over-worded... Fundamentally the question is: Do DC-DC chips need additional small-value (1uF or 100nF) caps at the input to act as bypass caps?

[amitchell]

I have never seen the need to add small values other than to reduce the high current/voltage loop area. So say you need to use a 1210 10uF 50v X7R then you can add a 0603 or 0805 1uF in between the smps ic and the 1210 to reduce the loop area. It's also a good idea if you can to use a leadless diode if possible to tighten it up further.

It is not like bypass at an MCU.

[CraigHB]

It depends on the converter, boost converters are noisy on the output side, buck converters are noisy on the input side.  So what you can get away with in minimizing components depends on the type of converter.  For the boost converter you linked to, input caps are not as critical as they are for a buck converter. 

The problem with filtering in a general sense is you need low ESR and ESL to get the high frequency stuff and you need high capacitance to get the low frequency stuff.  An electrolytic can act as your bulk capacitance to cover low frequency where an MLCC cap can provide the low ESR and ESL you need to cover high frequency noise.  Converters are generally noisy things with big changes in current at switching frequency along with high frequency oscillations due to the interaction of switch capacitance and inductance.

I have done a few converters using all MLCC caps and it's nice to get rid of those bulk capacitors, but there's a few caveats with that.  You typically need some amount of bulk capacitance on output purely for stability concerns.  MLCCs can often provide the bulk capacitance you need by running an array, however the DC bias comes into play.  It can be tricky since at higher voltages capacitance falls off and can possibly destabilize the regulator.  That being the case MLCC arrays can be a bit numerous to satisfy bulk capacitance requirements.  Also because of the DC Bias nature of MLCCs and the fact that available capacitance values are limited with that type, higher input and/or output voltages may be out of the question.

In terms of noise mitigation, higher capacitance value MLCCs can still cover you on the high frequency stuff since they generally have low ESR and ESL.  If you can get capacitance values high enough to meet bulk capacitance requirements, you can just use a single array of MLCCs to cover both high and low frequency noise.  Also MLCCs have lower ESL and putting them in an array further reduces ESL improving high frequency attenuation.

For the converter controller's supply voltage you do want to mitigate noise as much as practical.  For a boost converter the input side is not particularly noisy so the addition of a small value MLCC at the chip's Vin and ground pins is adequate.  You could possibly even leave that out if the current path for supply and return is short enough between input capacitor and controller.  I have on occasion used a simple low pass filter on the chip's supply for buck converters, but it shouldn't be a problem for a boost converter as they're pretty quiet on the input side.

[tszaboo]

There are some, high power like 10A DC-DC converter chips, which have separate switching voltage input, and chip voltage input. There you need separate capacitors. If you are talking about the few watt, and amp chips, not really. The chip lead will run the entire current, and there will be millivolts of swing for the chip's vcc. They are designed to work with that.
Now EMI is a different story. You can gather some insigth from these measurements:
https://www.onsemi.com/pub/Collateral/AND8477-D.PDF
I see a few dB reduction with the caps.

[T3sl4co1l]

FYI, some converters don't have capacitors internally, and require external parts to keep things going.  These will mostly be either very small ones (can't fit!) or very large ones (requiring bulky electrolytics that are best left unpotted?).

In either case, the capacitors are almost always necessary for filtering.  All converters are dirty.  I've only seen one (a 15W, medical grade AC-DC type) that passed CISPR 22 without external filtering (using a reasonable length of attached cable, and a typical load).

Common mode noise is the worst offender, and the least well specified.

Note that input and output ripple figures are only given in 20MHz bandwidth, which is nearly useless.

Beyond basic functionality (which is encompassed by the above -- not that that really helps you figure out what values are required; and manufacturers provide very little design data on the internals, that might help you determine this..), it's all about emitted noise and what a tolerable level is in your circuit.  Basic filtering, bandwidth, attenuation and impedance matching/terminating stuff.

Tim