Electronics > Projects, Designs, and Technical Stuff

Being lazy with decoupling capacitors?

(1/2) > >>


I'm wiring up an STM32 microcontroller. It's in an LQFP and each side has a Vss and Vdd. Each Vdd needs a 100nF decoupling capacitor and the docs say to put it as close as possible (one of the Vdds also needs a 4.7uF cap).

And yet, if I look at the layout of the STM32-Discovery board (http://www.st.com/internet/com/TECHNICAL_RESOURCES/TECHNICAL_LITERATURE/USER_MANUAL/CD00267113.pdf), ST themselves don't seem to be obsessing over that "close as possible" stuff. It looks as if they've clustered all four decoupling capacitors next to each other on one side of the uC.

My question is, once you've put four 100nF caps physically next to each other, and they all go to the same ground, and each Vdd that's wired to them is also connected to +3.3V, what's the practical difference between those four caps and a single 400nF cap? In short, I'm looking for an excuse to save real estate and be lazy with my decoupling caps. I suspect the professional EE answer is "don't do that", but I don't understand why.

And while I'm on the subject, one of the Vdd pins needs both a 4.7uF cap and a 100nF cap. How is that different from a 4.8uF cap? Sure, they don't make 4.8uF caps, but that's only 2% different from 4.7uF. Given the tolerances of most caps (+/-5%, +/-10%) what's the real-world benefit of putting a 100nF cap in parallel with a 4.7uF cap?


High frequency impedance. The 100nF or 4.7uF are DC values, at 500MHz the impedance is what counts (and the whole purpose of the decoupling).  The recommendations are like that so most people would get it about right with typical available parts. The real spec is something like <1.5 ohm complex impedance at 100-600MHz, but it is much harder to pick capacitors by that. It is hard to get wide attenuation range with one capacitor, so that is why they recommend two different types. That is also why they want to caps close to the pins; the impedance of the track from the pin to the cap is (very) significant. On a larger cap, they want attenuation in the lower frequencies, so the effect of the track smaller. Therefore, you get by with one - it is going to be ok from the farthest and nearest pin to the big cap. But not so on the high frequencies, and that is why you want to have a cap on each supply pin.

It's not strictly true, but the way I think of bypass caps is as follows:

* The little ones are to take high frequency noise to ground.
* The big ones smooth dips in voltage.
So you need the combination of the two.

If you only have one place for them there's little point putting more than one.

The 'near the pin' requirement differs significantly if you have a 4-layer PCB, where the ground/power planes have low enough impedance to relax the need for caps close to every pin.

As 1uf ceramics are so cheap these days, if space isn't tight I default to these everywhere, which avoids the need for larger bulk decouplers.
Just need to make sure your voltage regulator is happy with the low ESR of ceramic caps.

Thanks to all for your replies so far.

mikeselectricstuff, I don't know how many layers the STM32-Discovery has, but I'd be surprised if it were four. It's not that complex a board and they're selling it for only ~US$10.

Since the LQFP is 7mm and they've put all the caps on one side of the chip, one of the Vdds has its cap about 1cm away. Is that "far"?

And are you saying that in their layout, they would've been OK with one 100nF cap instead of 4?

- Bob


[0] Message Index

[#] Next page

There was an error while thanking
Go to full version