Hi folks,
I've enjoyed the discussion on the subject of ferrites and Vcc and
feel compelled to throw in another two cents worth.
We have tested numerous products with the ferrite bead in series with
the Vcc "island" that feeds power to a chip. In an overwhelming
majority of those cases, we were able to trace emissions problems back
to this practice.
By bonding the "island" back to the main Vcc plane (copper tape and
solder wick), we have noted significant improvement in emissions--
when this has been found to be the source of the problem.
>From our empirical observations, we have formulated two mechanisms
that may explain why this practice can be harmful:
Situation #1: RF return impedance
Since the Vcc plane forms part of the RF return path, inserting an
impedance in series with this path raises the impedance of the circuit
return path. As the currents must return to their source (and thus
travel through this impedance), a voltage potential is developed
between the island structure and the rest of the circuit. This
potential elevates the "isolated" circuitry with respect to the rest
of the circuit. An RF voltage "hill" develops (that the only way I can
describe it). This can be easily detected around such an island using
an E-field stub sniffer and a spectrum analyzer.
Situation #2: Induced voltage due to bead currents
When a gate switches, it pulls current through the bead, regardless of
how well-decoupled the Vcc island is. This current generates a voltage
across the impedance of the bead. This voltage appears essentially as
RF ripple on the Vcc structure (which powers the gates in the
"isolated" circuit); thus, this RF energy is coupled to every gate in
the isolated circuit. An emissions problem occurs when these gates
drive signals OFF of the isolated area (to other areas of the circuit
board), and, regardless of the frequency of the intended signal, the
ripple is passed along to the rest of the board.
This would not be a problem if:
- The 0V plane had a zero impedance (Situation #1), and
- Decoupling capacitors were perfect (Situation #2)
Alas, this is not the case. At RF in general (and in EMC in
particular), what may offer a benefit on paper may work in opposition
in practice.
A logic gate, when switched, requires a tremendous inrush current to
support the voltage transition. This charge must be available in the
immediate local area of the chip, and a good bit of it is supplied in
the capacitance of the Vcc/0V plane (a bit is supplied by the local
decoupling caps). As these charges are depleted, current flows from
the remainder of the Vcc/0V circuit, decoupling capacitors and power
supply to replace and equalize the voltage.
If one supplies a large enough contiguous Vcc/0V structure to support
this charge demand, the high frequency voltage developed during the
transition will be minimum. By creating an island one reduces the
available supply of charges. By inserting an impedance, one puts more
demands on the local decoupling capacitor AND creates a voltage drop
across the impedance.
The best source of this current is a fully contiguous, low Z Vcc and
0V structure.
Finally, a historical observation: This concept of ferrites and power
bus filtering arose in the early 80's when a clock speed of a few MHz
was state-of-the art, and the upper frequency of emissions problems
was some tens of MHz. I think that the design practice has become
ingrained to a certain extent and wherein it didn't matter what you
did `in the old days', with 100 MHz fundamental frequencies, these RF
problems arise.
Mike Violette
Washington Labs
mi...@wll.com
