Buck converter: capacitor voltage ratings

Question

I'm currently designing a PCB for a caving robot, where PCB space is rather limited.

One thing taking quite some space are the input and output capacitors of the 12 to 5VV DC-DC converter.

I basically just used the values from the typical application from the "core" of the regulator, the XL4015 (datasheet):

For now, I used just reused part numbers I already used elsewhere (to reduce cost),but I'm intending to use just one 220/330uF capacitor instead of the 2 or 3 I initially planed to save space.

I was thinking to go for tantalum capacitors (as far as I understand, they are the smallest for given capacity and voltage). Does that seem OK? If so, what voltage ratings should I use? (the lower the voltage, the smaller the capacitor, but to what point can I push without risking failure?)

Additional information:

• the "12V" is non regulated voltage from a 3S Lipo (9 to 12.6V), and also connects to 16 small (<1A) motor controllers
• the 5V will mainly power a Raspberry Pi and the linear voltage regulator providing the 3.3V for the STM32F7, and maybe a few sensors later on, maybe a few power LEDs with PWM fading (those will probably run from the 12V directly, but to be confirmed). I expect usually 2-3A, with peaks a bit higher

NB: in the typical application notes, they use:

• 50V for the input (but input voltage is up to 38V). I suppose it is completely overkill when I have only a 3S lipo as input
• 25V for output: I realized they used this same value for 5V and for 12V typical application. Is there any reason to keep it that high? If not, how low can I go? 6.3V? 10V?

EDIT:

As asked, some aspects about safety requirements:

• failure in the DC/DC converter (whatever the reason), leads to about 90% risk of loosing the robot (one might be happy and it is still near enough to retrieve it by hand, but even so it might get damage by the fall). During testing/prototyping, the robot is accessible, so no risk of loosing it, small damage if falling.
• the components for one robot will be around 400€ (NB: I'm not planning to sell, just to put it open hardware if it works)
• in the cave, there is nothing that might catch fire, so the only risk is when testing the robot, or when handling the robot by hand
• no risk of the robot harming humans in case of failure (excepted the flame/explosion of the tantalum itself if the user is still holding the robot at the start/end of mission
• I guess the risk of loosing the robot for any mission is something like 5 or 10%

Answer 1

The input current of a buck converter is a square wave whose peak value is the output current, and it comes mostly from the input caps. Therefore, the input caps should have a ripple current rating and ESR compatible with this. For a 5A buck, the input cap should have very low ESR, otherwise it will die. Some of the ripple current will be taken up by other caps nearby, so you probably don't need a 5A rated cap unless it's the only one on the board.

If the 12V rail also powers 16x 1A motor controllers, presumably using PWM, then you'll need some low ESR capacitors there too. These should have low inductance to sink inductive kickback from the motors, and combined with the high current, this means you'll probably use a power plane for your 12V. In this case the buck input cap and motor controller caps can be shared, all on the same power plane, so you should think about all the caps on +12V rail, not just the buck.

It would be interesting to know the impedance of the battery, to know how ripple current will be shared between it and the caps.

So, let's go with the caps for both the buck and the motors. You'll need pretty low ESR, so:

• Aluminium electrolytic gives low ESR at reasonable cost, but the value and size have to be large.

• Aluminium polymer caps can have very low ESR and high current rating in small sizes, at moderately increased cost.

• Tantalums have too high ESR, no way

• Tantalum polymer is smaller, more expensive, and adds a risk of fire.

• MLCCs offer the best ripple current rating, but lower capacitance.

I'd use a buck with a higher switching frequency like TPS56528 (there are many others), which shrinks the input cap, output cap, and inductor. It's synchronous, so it's more efficient and you get rid of the diode. It uses all-ceramic input and output caps, and these are pretty tiny, the datasheet says 2x22µF on the output (or 4x10µF) so that is much smaller than 3x 100µF electrolytics, while having much better HF performance and ridiculously low ESR.

Likewise at 650kHz, the impedance of a 10µF MLCC is only 25 mOhms, ESR is about 5 mOhms according to Murata, so if you got a roll of 10µF 25V X7R MLCC you can also stick a few on the input and job done. A few polymer caps should do a good job for decoupling on the motors while remaining small. Note thru-holes usually have lower ESR because the legs are thicker.

Note that MLCCs usually fail short when cracked (make sure your board is properly secured so it doesn't bend when the robot falls on the floor of the cave) and tantalums fail by becoming a tiny flamethrower that will incinerate your board.