Sunil Kumar Bahuguna writes:
My design uses many high-speed ICs that require separate analog VCC filtering. I am thinking of using ferrite beads to isolate the analog VCC supplies, but I read somewhere that ferrite beads are not recommended for high-speed, gigahertz-range applications. Can you please explain this?
Ferrite beads are often used in gigahertz filtering applications. Ferrite beads come in two flavors: high-Q, resonant beads and low-Q, nonresonant beads, also called lossy, or absorptive beads.
The high-Q type has no place in a digital circuit. These beads are used to construct RF oscillators and filters and other circuits that need highly resonant circuit elements. In a digital power-filtering application, the last thing you want is resonance. The low-Q type is commonly used for power-supply filtering, in series with the power connection. Most often, this style of filter also has a capacitor to ground on either side of the inductor (Figure 1).
(courtesy EDN Magazine)
The bead manufacturer should provide you with a curve of impedance versus frequency for your bead. From this curve, you may ascertain the efficacy of a particular ferrite bead at your frequency of interest. For your filter to work properly, the impedance of L1 should greatly exceed the impedance of C2 .
The performance of some ferrite materials (particularly the very high-permeability materials) begins to deteriorate at high frequencies. Past some critical point, the impedance no longer rises proportional to frequency as quickly as you might like. However, for a filtering application, as long as the impedance remains high enough to do the filtering job, you don't need to worry about the efficiency of the ferrite.
Beware that the parasitic capacitance from the input wire to the output wire of the bead can defeat your purpose, especially when you use a double-hole-type core wound in a U-turn or multi-turn configuration. For gigahertz-speed filtering applications, you should stay with a straight-through-type (one turn) core. A good core will look like a long, skinny cylinder with a single hole through the central axis for the signal wire. This straight-through-type topology keeps the input and output circuits as far apart as possible.
The circuit in Figure 1 models the effect of a bead's parasitic capacitance. As frequency rises, the impedance of L1 rises while the impedance of CP falls. Beyond some crucial frequency, capacitor CP begins to short out inductor L1 . You will see this effect if you probe inductor L1 , mounted as it will be used in your actual layout, with a network analyzer. The effect of parasitic capacitance may not be included in the plots of impedance versus frequency that come from the manufacturer of the bead, because they don't know how you are going to lay it out.
Capacitor C2 also suffers from parasitic layout effects. Figure 1 models the inductance LP of this capacitor's physical layout in series with the capacitance C2 . As you go up in frequency, the impedance of C2 falls while the impedance of LP rises. Beyond some crucial frequency, inductor LP prevents C2 from acting as a good short to ground.
The filter ultimately fails at a frequency high enough that the extra series impedance of LP becomes comparable to the parasitic shunt impedance of CP .
All low-pass filters fall prey to parasitic effects. At extremes of frequency, the inductors all turn into capacitors and the capacitors into inductors, reversing the action of the filter. Beyond some threshold frequency, the filter no longer prevents noise from passing through. The response of such a circuit resembles a band-stop filter more than a low-pass filter.
For example, suppose L1 is a straight-through 50-nH bead with CP =0.1 pF. Let C2 be a 0.01-uF capacitor in an 0402 package with LP=0.5 nH. This combination attenuates noise more than 30 dB from about 50 MHz to 3 GHz. Beyond 3 GHz, it works progressively less well.
To extend the response you must either use better components (with better layouts) or use a multistage filter. A multistage filter couples a number of stages in series, with each stage covering only a limited range of frequencies. Two well-designed stages can band-stop twice the range (on a logarithmic frequency scale) of a single-stage design.
As with all filter designs, you should use a network analyzer to measure the results of your final filter architecture. Low-Q beads are not supposed to resonate, but you still might get a noticeable resonance near the low-pass cutoff frequency. It's worth checking.