MLCC capacitor size for small boost converters

[microvia]

Hi,

I need to know what is the minimum size for 4.7µF, 10µF and 22µF MLCC capacitors used around SOT-23 boost converter ICs which operate at around 1MHz to deliver 3.3V under 50mA or so (I'm using devices which can deliver 300mA to 1A and start working a 0.9V).

I'm asking this because I don't know how much ripple current they need to provide, hence their minimum size.
Thanks.

[mtwieg]

I'm pretty sure even a 0402 size MLCC will handle the ripple current from your boost converter. For example, attached is the spec sheet for a 4.7uF, 6.3V cap in a 0402 package. At 1MHz, its ripple current is rated to beyond 2.5A rms.

Though be aware that when biased to 3.3V, the effective capacitance will be reduced to around 1.1uF. If you need more output capacitance then you will need a larger package (or multiple small packages). But ripple current is rarely a limiting factor for MLCCs.

[microvia]

Thanks for your explanations.
As it happens I read the recent post about derating. I always preferred MLCCs with a working voltage a few times over the voltage they'll operate at.
In this case the input voltage between 1.5V and 6V, and the fixed output voltage 3.3V.
I consider using 0603 25V X7R MLCCs, but didn't know whether 0805 or larger ones would be required.

This brings a question around derating: for a 10µF 10V MLCC working under 6V, will the capacitance variation also depend on the package size ?
Is there an online database / search tool which allows selecting MLCCs using specific parameters that aren't used by providers (Mouser, RS, LCSC, you name it) ?

[tooki]

Thanks for your explanations.
As it happens I read the recent post about derating. I always preferred MLCCs with a working voltage a few times over the voltage they'll operate at.
In this case the input voltage between 1.5V and 6V, and the fixed output voltage 3.3V.
I consider using 0603 25V X7R MLCCs, but didn't know whether 0805 or larger ones would be required.

This brings a question around derating: for a 10µF 10V MLCC working under 6V, will the capacitance variation also depend on the package size ?
Is there an online database / search tool which allows selecting MLCCs using specific parameters that aren't used by providers (Mouser, RS, LCSC, you name it) ?

The DC-bias-dependent capacitance loss of Class 2 and 3 dielectrics* depends pretty much only on the package size! A cap with a higher voltage rating can handle a higher voltage, obviously, but the capacitance loss at lower voltages does NOT scale with the maximum voltage rating at all! If a capacitor series offers the same size and value in multiple voltage ratings, you’ll see that the capacitance loss does not improve whatsoever in the higher-voltage variants.

The smaller the package, the more pronounced the effect. For example, an X7R 0402 cap can easily lose 80% of its capacitance at 5V — irrespective of whether the maximum voltage rating is 6V, 10V, or 25V, for example. So going with a larger size is often necessary, or you have to use a bunch in parallel.

*The dielectrics of ceramic capacitors are classified into three classes. Class 1 are the “well behaved” dielectrics whose capacitance is totally insensitive to voltage, and almost entirely insensitive to temperature. This is usually the C0G (aka NP0) dielectric. Class 2 (and the rarer Class 3) are sensitive to both voltage and temperature, and their dielectric codes (like X5R, Y5U, etc) encode their temperature dependence characteristics. They’re kinda terrible in this regard. But their big advantage is that they allow massively higher capacitance values per physical volume compared to Class 1. So in practice, if a Class 1 dielectric is capable of providing the required value in the physical size desired, it’s used, and these caps can be used for all purposes, including non-critical applications like power supply decoupling, but also in critical ones like in a signal path. Class 2 and 3, on the other hand, are used when Class 1 cannot attain the value required, but we can’t use them in critical signal paths, where the voltage- and temperature-dependence cause distortion or other errors.

See also: https://www.slideshare.net/slideshow/class-1-and-class-2-mlccs/88771295

[Smokey]

Watch out for alternate sources with all this stuff.  Just because one manufacturer's parts "worked" in your circuit doesn't mean another manufacturer's parts with the same Digikey spec will behave identically.

Having an LCR meter with DC Bias for checking this stuff is pretty nice.

[Siwastaja]

When capacitance matters (and often it does), strongly recommending only using parts for which DC bias effective capacitance curve exists, so you can check what you actually get. Then you can choose - if both ESL (small size) and large amount of capacitance are important to you, then you can pick the smallest size which gives enough actual capacitance. Or compare µF/$ or whatever.

Inferring/guessing the true capacitance curve from other parameters is last resort and you really need to know how to do that. The most usual mistakes (due to incorrect advice all over the internets) are thinking that temperature rating codes (like Y5V / X5R / X7R), often semi-incorrectly/misleadingly called "dielectric types", have any predictable correlation with the DC bias curve; and that voltage rating has anything to do with it. Former has weak-but-existing correlation, latter even weaker.

In absence of real data, best indicator for how much capacitor is losing capacity is indeed volumetric energy density - how large is the nominal capacitance to begin with compared to package size (remember to include height) - if something is "too good to be true", it probably loses more capacitance under the bias, than one with smaller claimed (nominal) energy density.

So forget all the "X7R loses 50% of capacity under rated voltage, Y5V loses 80%" crap you can find online. It's easy to find examples of X7R parts that lose 80% and Y5V parts that only lose 70%. These tags are by definition temperature dependency curves, not DC bias curves.

ESR / current capability is problem only in most extreme cases.

At 3.3V, sensible minimum package sizes for 4.7uF, 10uF and 22uF would be maybe around 0603, 0805, 1206 respectively. Rough ballpark. E.g. 10uF in 0402 would probably lose a lot of capacitance at 3.3V bias. Also remember that paralleling smaller parts is sometimes better, depends on your layout.

4.7uF 0805 for ~10-20V, 10uF 0805 for <5V would be a good golden middle road start - for more capacitance, you can parallel those.

[Psi]

Yep, just make sure you fully understand DC-bias and how physical cap sizes effects it.
And select a cap that has a table/graph in the datasheet so you can figure out what the true capacitance will be. A 10uF cap may only give you 1uF of capacitance, so it really matters for DCDC converters.

I like Murata caps as they tend to have all the DC-bias graphs easily readable on their website if you search the part number.

[MT4S301]

Is there an online database / search tool which allows selecting MLCCs using specific parameters that aren't used by providers (Mouser, RS, LCSC, you name it) ?

These data are provided by manufacturer.
https://ds.murata.com/simsurfing/?lcid=en-us
https://weblib.samsungsem.com/cn/mlcc/mlcc-ec.do
https://product.tdk.com/en/technicalsupport/tdk_meister/index.html
https://ds.yuden.co.jp/TYCOMPAS/eu/specificationSearcher?SR2=LM%2CMP&cid=C&u=M

[microvia]

Thanks a lot for these links. MLCCs selection will be a lot easier. As Psi said, the graphs are really easy to deal with by mousing over and reading the exact values.
I don't know which application entry best fits boost/buck converters. Maybe the automotive power /safety ?

[tooki]

General purpose is fine for a buck/boost. Use whatever is convenient. Automotive parts are often a bit more reliable. (“Soft termination” can reduce the risk of a cap breaking due to board flexing, for example.’