Filtering power supplies using ferrite bead

[joniengr081]

In ZYNQ UltraScale + PCB design guide document, there is a section for power supplies for PS DDR. They used a ferrite bead as shown in the attached picture, which look like an inductor. What are the parameters to chose such ferrite bead in the power supplies ?

[GigaJoe]

it exactly defined in the screenshot , your provided

[Nominal Animal]

Ferrite beads are commonly specified by their impedance at 100 MHz, and the amount of DC current they can pass and/or their DC resistance.

If the current is less than 3A, one suitable choice would be Murata BLM18SG121TZ1D or BLM18SG121TN1D, for example.  Their DC resistance is on the order of 35mΩ, so at a 1A current it will drop the voltage by 35mV = 0.035V, and generate 35mW of waste heat (not much at all).

The above links take you to the Murata product page, where you can see the frequency-impedance curve for the part.  Exactly how the high-frequency noise is reflected or attenuated is AC transmission line stuff that others can discuss much better than I can.  To simplify, a small part of the high-frequency noise is absorbed by the bead, but most are just reflected back.  Thus, the ferrite can be used for two purposes: to stop reduce high-frequency noise from entering the circuit, and to stop reduce high-frequency noise leaking as electromagnetic interference out using the supply voltage line as an antenna, basically.

When I use one – and as a hobbyist, I like to, because I do one-offs, and beads are cheap –, I look at the frequency-impedance curves, and typically pick the one common one that has high/highest impendance at the frequency range I expect noise to come in or leak out via the supply line.  You don't typically see them used in 10MHz microcontroller circuits, but at 48MHz and above, they're common.

[T3sl4co1l]

Also be aware of saturation: impedance declines with DC bias.  Typically in the 50-200mA range, lower for higher impedance and smaller chip types (e.g. 0402 1kohm might be -30% at ~10mA) and more for larger and lower (with 8-20ohm types sometimes offering several amperes of useful range).

This is independent of the DC amperage rating, which is only a matter of temperature rise.

A PLL probably doesn't draw very much current, so this is a good application for a ferrite bead; assuming little filtering is needed (expect 10-20dB of attenuation from such a filter; not remarkable, but may be worthwhile).

Tim

[srb1954]

When I use one – and as a hobbyist, I like to, because I do one-offs, and beads are cheap –, I look at the frequency-impedance curves, and typically pick the one common one that has high/highest impendance at the frequency range I expect noise to come in or leak out via the supply line.  You don't typically see them used in 10MHz microcontroller circuits, but at 48MHz and above, they're common.

From a point of view of EMC control it is sometimes better to not just use the highest impedance bead at your operating frequency but to consider what the impedance is over a wider frequency range.

Some types of beads have a very peaky, high Q response and can only achieve good attenuation over a narrow frequency range but, due to falling impedance at higher frequencies, are less effective at suppressing harmonics.  Beads with a lower Q, and hence greater resistive losses, often maintain a higher average impedance over a broader frequency range and thus are more effective at suppressing the difficult higher harmonics.

[joniengr081]

@T3sl4co1l,

I am not sure if understood about saturation. Can you please explain a bit more ?

In addition to saturation I am also wondering about  what is the selection criteria, high impedance at high frequencies or lower impedance at high frequency ? There are ferrite bead in 0603 packages, can they be used in series in DC power supplies to pass up to 2/3 A at 5 V for example ?

[Nominal Animal]

When I use one – and as a hobbyist, I like to, because I do one-offs, and beads are cheap –, I look at the frequency-impedance curves, and typically pick the one common one that has high/highest impendance at the frequency range I expect noise to come in or leak out via the supply line.  You don't typically see them used in 10MHz microcontroller circuits, but at 48MHz and above, they're common.

From a point of view of EMC control it is sometimes better to not just use the highest impedance bead at your operating frequency but to consider what the impedance is over a wider frequency range.

Fully agreed.  I do look at the frequency-impedance curve, and try to estimate (because I don't have test equipment to measure) the expected noise frequencies from clocks and switching frequencies and their harmonics/multiples; after all, a squarish edge at some frequency does have a lot of harmonics at higher frequencies (multiples of the original frequency).  Beads do seem to help most with ADC noise, and they are cheap enough that I like to use them in my one-off designs for the EMI rejection they do provide, "just in case".

Some types of beads have a very peaky, high Q response and can only achieve good attenuation over a narrow frequency range but, due to falling impedance at higher frequencies, are less effective at suppressing harmonics.  Beads with a lower Q, and hence greater resistive losses, often maintain a higher average impedance over a broader frequency range and thus are more effective at suppressing the difficult higher harmonics.

Well put.  This is easily seen in the frequency-impedance curves, even to a hobbyist like myself.

I am not sure if understood about saturation.

Saturation means that when the DC current is high, the impedance with respect to noise on top of that DC decreases: the ability of the bead to "reject" the higher frequency noise decreases.

The same happens with supply bypass capacitors and DC voltage: the higher the voltage, the lower the capacitance with respect to noise on top of that DC.  (Well, C0G/NP0 capacitors tend to not suffer from that much, but most capacitor types do.)

We want the ferrite bead to have high impedance at the target frequency range, so that it will "reject" those frequencies.
(Again, I'm using "reject" because some of it is absorbed, most reflected; and the effectiveness is not linearly dependent on the impedance –– it's not "twice the impedance means twice as effective" –– because of that AC reflection.)

For example, the two ferrites I showed won't do anything strange at 3 A current; they will drop about 0.1 V across them, and generate 315 mW of heat, and their impedance at high frequencies –– their ability to "reject" the noise –– is quite low that being their maximum rated current, but they will "work".  I chose them as examples, because they're cheap, about 0.06€ in lots of ten at Mouser, very common, have nice wide impedance peaks up to 1 GHz, and seemed to me to fulfill the demands.

The similar ones with the highest current rating I could find at Mouser are Pulse Electronics BBUP00321611121Y00 (link to datasheet, 6A current rating).  It costs about 0.09€ in lots of ten at Mouser.

If we compare its frequency-impedance curve to the Murata ones I linked to earlier, you'll see that the impedance peak in the Pulse one is quite sharp, just like srb1954 described.  In particular, the Murata ones have 120 Ω impedance from 80 MHz to 1000 MHz = 1 GHz, whereas the Pulse one exceeds 100Ω impedance only between about 70 MHz and 200 MHz.  Thus, higher current rating in this case comes with a much narrower effective frequency "rejection" range.

"Which one to pick, then?" is a difficult question.  What is the expected current draw on that pin?  Are there any suggested example ferrite beads mentioned, so one could check the current rating and frequency-impedance curve?  But, ferrite beads are not miraculous noise filters, so it is not like the design will fail if you choose the wrong one; you'll simply have more noise/EMI than in the best case.

Given only the information in this thread, I'd pick one of the Murata ones myself.  I hope the more experienced members in this thread will give you their opinions.

[T3sl4co1l]

For comparison, consider Laird HI0603O700R-10 -- Murata never* provides bias data but Laird has it on almost their entire product line, do recommend.

The curve is significantly depressed by just a few hundred mA, and significantly saturated at the 1400mA curve given for this part (note the impedance, at most frequencies, is about 10% of the nominal zero-bias value).

Note that the impedance remains untouched up in the GHz range, but also notice what inductance this corresponds to: a couple nH.  Literally, inductance of the body length of the part itself (or the winding within, as these may be multi-turn monolithic/multilayer construction).  Which is to say, you get as much impedance from a wire link, as from a highly saturated ferrite bead.

If you'd like to play around in a simulator, a model like this: https://seventransistorlabs.com/Modeling/SPICE/HI0603P600R_NL.ckt can be used for the purpose.  This can be used in AC analysis to find frequency response under bias (remember to set up some DC bias conditions!), or transient to see the nonlinear e.g. step response.

*They don't provide DC bias under the part page, nor the characteristics (SimSurfing), but *possibly* have those data under the "bias tee" tool, but in a form that's impossible to use/access.

Tim

[joniengr081]

I think the parameters to select a ferrite bead would be then

low dc resistance
high ac impedance at resonant frequency
low inductance

What about package ? If I am looking for a ferrite bead to be used in 10 A @ 1.8 V power supply then which package would be suitable ?

What is the power rating of 0603 packages ? 250 mW ? Are bigger packages better as their power rating is higher ?
 

[Doctorandus_P]

I guess that if you want to push 10A though a ferrite bead, that will be your first selection criteria, especially with such a low power voltage, which also necessitates minimizing voltage drop. Read some datasheets about the current rating of these things. And when SMT packages get small, then their ability to get rid of heat is greatly influenced by the copper of the PCB they are connected to.

I find these ferrite beads quite weird things. They only work between about  20MHz and Half a GHz or so and even in that range they only have a narrow bump in their impedance. All together this strongly suggests that you can't just throw in some ferrite beads and hope it works (well, you can hope...  ).

[T3sl4co1l]

"Narrow" on a log scale, perhaps?

The semilog or log scales also hide the fact that losses are never very low even at low frequencies... I think the peak Q of most is below 10.  That is, DCR dominates at low frequencies, core loss dominates at middle frequencies, core loss dominates even more at high frequencies (towards resonance).  (And it's still pretty lossy in the capacitive range, not to forget: the ferrite material has a fairly high dielectric constant, but much resistance as well.)

You want to avoid constructing a resonance in that peak band, of course; or provide enough external damping (electrolytic, high-ESR polymer, tantalum, or ceramic plus resistor) to keep it well-behaved.

Ferrite beads aren't available in stupendously high currents, and even then only reasonable in very low values (a few ohms), and saturation is still likely to occur.  You need either a much larger bead, or an inductor proper (in which case, external damping is mandatory).

Tim