EMI and EMC can be a tricky subjects, and it's often tempting to mis-apply design guidelines to try and reduce EMI. One of these areas relating to EMC in PCB design is the use of ferrite beads. These components are basically filters, and they do perform a useful function on power cords in many electronics. You're probably reading this article on a laptop that uses a ferrite to filter out conducted EMI from the supply line.
A problem begins to arise when you try to apply the same logic to other areas of a PCB. Ferrite beads are sometimes applied for EMI in two ways in attempts to minimize EMI, but the designer ends up creating a new EMI problem if these components are not used correctly. In this article, we'll go over some of the ways ferrites should not be used in a PCB, as well as how they actually operate in terms of their filtering behavior. As we'll see, the same logic that applies to ferrite cores on the input of a bridge rectifier stage in a power system does not apply to the power connection between a regulator and an integrated circuit.
Ferrite beads are magnetic components, so it is tempting to think of ferrite beads as inductors that provide low pass filtering functions. They do block high frequencies, but only in a specific band; their impedance tends to maximize and be totally resistive around 100 MHz to 1 GHz. Above that band their inherent capacitance takes over and their impedance begins to drop again. In this way, they aren't the perfect filters. However, these components can be used with other ferrites to address specific types of noise on the input power section. In fact, that’s their most common usage.
Impedance vs. frequency curve for the BLM18PG600SN1D ferrite bead from Murata.
While a bead by itself can’t make a low pass or high pass filter, they can be used for more effective low-pass filtering at lower frequencies (e.g., 60 Hz AC or 120 Hz rectified DC ripple) when combined with shunt capacitors. Then you get what is essentially an LC filter that can provide low-pass filtering functions at sufficiently low frequencies. These are sometimes used on the AC power stage of a system to provide differential-mode noise filtering to ground, i.e., as a Pi filter. In higher power systems, this same circuit design is used with inductor coils as these can generally handle several amps of current. You would then follow this with a common-mode choke and further filtering on the output from your rectifier stage to produce DC power with low ripple.
An example showing these filtering elements an AC input is shown below. Note that L2 is typically a ferritic component (either a ferrite core inductor in high current systems, or a ferrite bead in low current systems).
Why should we worry about current? The reason has to do with saturation. When high DC current is being pulled into the system, the bead can saturate and lose inductance, similar to what happens in a transformer core at high current. In the most basic application, we would have the following set of inductive components involved in filtering noise on the input power stage as shown above:
Placing these components on the input section of a power stage is much more effective than placing them on the output side. If you do use ferrites for filtering on the output side of a power converter, the acceptable use of these components depends on what you need them to do.
When placed between a power regulator circuit output and the bypass capacitor on input power pins for a digital component, you have basically formed a Pi filter. Therefore, it would be reasonable to expect low-pass behavior at switching converter frequencies. This should not be done with high speed digital components as it will create new noise problems and will not function effectively as a low pass filter.
Why is this the case? There are a few reasons for this. My thought is that the typical recommended uses of ferrite beads are being applied in areas where they are no longer effective, or where they create new problems:
An implementation for points #1 and #2 is shown below. This implementation is fine in #1 and #2 as it targets conducted EMI while allowing DC to pass. It will not function properly for point #3.
As we can see from points #1 and #2 above, there are some times where a ferrite on the output of a power regulator is acceptable. In fact, the image shown above is the typical way to use a ferrite bead for filtering power supply noise from reaching a static impedance. Using it to suppress noise from affecting a large digital IC as in point #3 is not one of them as you will create a new source of EMI due to ripple on the power rail. This is why we add capacitance to the power rail through the use of decoupling/bypass capacitors as this reduces the PDN impedance at the high frequencies associated with digital signal bandwidths. This will help suppress other noise phenomena in the design, particularly ground bounce.
Ideal power supplies have zero output impedance, meaning they do not experience any power loss and they can respond infinitely fast to transients on the power bus. In reality, power supplies do not function this way; the output response time to a disturbance on the power bus is limited, and they can only respond up to certain speeds (frequencies). This is because power supplies have non-zero output impedance, which can essentially be modeled as a series RL circuit.
Here, we're not only referring to bench supplies, we're also referring to VRM circuits on a PCB that supply voltage to fast digital chips. These components need to supply power at fast edge rates, as described above, but the control loop on the VRM will also attempt to respond to voltage transients on the power rail.
One bad design choice that is sometimes implemented with ferrite beads is to place a ferrite bead Pi filter on the output of a VRM, and then use it to supply power to a digital circuit. An example with TPS546D24S is shown below. There are two reasons this is a bad design decision:
This means that, not only are strong transients more likely to occur, but also the power supply circuit will be less able to dampen transients when they do occur.
An example of how not to use a ferrite bead on a power supply. This power supply circuit is intended to provide power for digital components at fast edge rate, and the ferrite bead will slow down the response time of the power supply.
To see some proof of the concepts in action, we can look at a measurement of the output impedance from a power supply circuit versus frequency. The graph below shows a comparison of output impedance measurements for a power supply with and without a ferrite bead used as a filtering element on the output. These data are from Omicron Lab, which they discussed in very great depth in one of their YouTube videos. It should not be a surprise that the power supply with the ferrite bead experiences a wideband impedance peak, which happens to correspond to the ferrite bead's resonant frequency.