Electronics > Projects, Designs, and Technical Stuff
DC derating in bypass capacitors, is it a big deal?
(1/2) > >>
ehilarioc:
Hello,

When I designed PCB with small smd ceramic capacitors most of the times I used to just pick 1206 and 0805 because they were easy to solder. With time, my soldering skills improved and know I am able to solder as little as 0402 without much effort.

Also, using smaller components helped to lower parasitics and that was a plus, however I am wondering what are the tradeoffs, of using smaller caps and their DC capacitance derating for bypassing. A 100nf 0805 25v capacitor has an insignificant DC derating at 3.3v, however a 100nf 0402 6.3v cap has a derating that is at least 30% at 3.3v

I assume that for filter design this is a bigger issue, but what is more important for bypassing? To decrease the DC derating effect the quick solution is just to pick a higher voltage part, but with that the prices increases quite a bit.

For example, a manufacturer recommends a 2.2uF in an specific pin that will have 2.5V (analog reference), the pin pitch of the part is 0.5mm so a 0402 will fit nicely in the layout. However, when I pick the component: 2.2uF, 0402, 6.3V, the capacitance derating is almost 50% at 2.5V, this does not get a lot better with a 10V part that has a derating of 40% and with a price that is twice the 6.3V part.

Should I put two 2.2uF 0402 in parallel?, should I go for a bigger size cap, a 0805 25v part that not only has a derating of 5% at 2.5v but it is also cheaper?

Should I just put a 2.2uF part that is above the 2.5V (for the example) working voltage and stop worrying about this?


mikerj:
Use the biggest parts you can, and don't assume that the higher voltage parts (in the same size package) suffer less reduction in capacitance, very often they don't.
ejeffrey:
In bypassing the absolute value of the capacitor often doesn't matter, so you can just ignore it.  Even for bypassing a precision analog reference, a 50% reduction in capacitance may not be a big deal.  Depending on the part and its intended application the datasheet recommendations may be chosen to be expecting or tolerating some capacitance loss.

Don't assume that higher voltage or higher value capacitors will help.  In the same package size they may have nearly the same capacitance at your working voltage.

One often overlooked parameter is component height.  A lot of capacitors (especially in the smaller packages) are made for cell phones where component height is a big deal -- shaving a fraciton of a mm off your components can be a big deal when you are trying to make a phone 0.1 mm thinner than your competition.  For many other applications you don't care about height at all as long as they aren't too tall to assemble.  You can commonly find 0402 capacitors in heights from about 0.2 to 0.8 mm.  The latter have (drum roll...) 4x the volume in the same footprint.  Try sorting on digikey by thickness and see if you can come up with a capacitor that looses less capacitance.

I just did that and the Murata GRM155C81A225KE11D is 2.2 uF / 10V capacitor that is 0.7 mm thick.  It only loses 15% of its capacitance at 2.5V -- considerably better than the 40% you quoted.  They are pretty expensive.  If cost is your concern I think an 0603 or 0805 capacitor will be cheaper for the same capacitance but with worse ESL.
mikerj:

--- Quote from: ejeffrey on September 24, 2019, 05:40:04 pm ---I just did that and the Murata GRM155C81A225KE11D is 2.2 uF / 10V capacitor that is 0.7 mm thick.  It only loses 15% of its capacitance at 2.5V -- considerably better than the 40% you quoted.  They are pretty expensive.  If cost is your concern I think an 0603 or 0805 capacitor will be cheaper for the same capacitance but with worse ESL.

--- End quote ---

Murata have an excellent online database (SimSurfing] giving useful characteristics for their inductors, ferrite beads and MLCCs etc.
T3sl4co1l:
Ultimately what you're creating is a low impedance at the supply node of a given device.  This involves the whole supply network (some simplifications can be made, depending) and the characteristics of each component and each connection.

Briefly, the resonant impedance of a given inductance and capacitance is Zo = sqrt(L/C), and the frequency is Fo = 1 / (2 pi sqrt(L C)).  If the resonant tank is connected to a resistance of the same magnitude as Zo, the Q will be low, and the impedance will have peaks and valleys a similar ratio above or below Zo.

You can further estimate required impedance by assuming the supply voltage should change by some maximal amount dV after a current impulse or step dI: Zo <= dV/dI.  And the inductance of traces, chip components, etc. is ballpark 1nH/mm.  You don't need to be a filter expert, just reconstruct the layout in SPICE using these approximations, and tweak values and positions until it looks good.  Simulate the device as an AC current source, and do an AC sweep to look for impedance peaks (if you use a 1A source, the voltage reading is proportional to impedance in ohms).

The most common pattern is some smallish bypass up close, at some distance (relatively large inductance, say 30nH) from the supply's bulk caps, with a nearby bulk cap (with significant ESR) chosen so Cbulk > 3*Cbypass and ESR = Zo.  If the supply has the same ESR, the local bulk cap isn't needed.

Say if Cbypass = 0.1uF, then the 30nH gives Zo = 0.54Ω, so choose Cbulk > 0.3uF (1uF would be fine?) and ESR ~= 0.5Ω (tantalum would be a typical choice, or ceramic in series with an 0.47Ω say).  This would be fine for a 3.3V supply having less than 5% ripple (dV = 165mV) and step load currents around 300mA (165mV / 300mA = 0.55Ω).

If the supply has very low ESR (a potential problem that can arise with low-ESR types like aluminum polymer and ceramic), it can exacerbate resonances elsewhere in the circuit, and local (ESR-ey) bulk is desirable.  Or reducing ESL and increasing Cbypass until Zo matches what ESR it does have (which might be ~10mΩ for ceramic, and bigger al polys).

Or for low noise sections, you want some filtering away from the general supply rail, so you choose a fairly large series filter inductor (some uH).  This needs to be paired with an ESR-ey bulk cap in all the same way.  Or you can put an R+L in parallel with the inductor, or you can use a ready-made equivalent (aka ferrite bead).

A note about ferrite beads, they saturate in exactly the same way type 2 dielectrics do (so, X7R and such ceramics).  Always check the datasheet to make sure you're getting a useful value at the operating voltage (for capacitors) and current (for inductors).  FBs are usually particularly obnoxious in this respect (an average 0805 chip saturates (-30% impedance) with just ~50mA), so watch out, and choose an inductor (= maintains inductance up to a useful bias current) if need be.

Tim
Navigation
Message Index
Next page
There was an error while thanking
Thanking...

Go to full version
Powered by SMFPacks Advanced Attachments Uploader Mod