1
\$\begingroup\$
  1. How quick does a signal have to be in order to be considered a high-speed signal?
  2. What are the limits and threshold where you can get away without special PCB layout?
\$\endgroup\$
  • \$\begingroup\$ Remember that it's not just about the switching speed or frequency but also mainly the rise/fall time. Don't see digital signals as a 1 or 0, but see them as a combination of a large number of analog waves, all with different frequencies. \$\endgroup\$ – Remco Vink Nov 15 at 7:26
  • 1
    \$\begingroup\$ Re: 1. This depends on who you ask. 2. That depends on which effects on the signal your system can and can't deal with. No general answer is possible! Also, you'd need to define "special" first. \$\endgroup\$ – Marcus Müller Nov 15 at 8:41
  • 1
    \$\begingroup\$ My rule of thumb that puts a signal in transmission line territory is when the length of track is longer than 1/6 of the rise / fall time of the signal (the strict definition is much shorter). Typical propagation velocity on 'ordinary' epoxy FR4 is about 160 ps per inch (surface) / 170 ps per inch (internal layers). \$\endgroup\$ – Peter Smith Nov 15 at 9:24
  • \$\begingroup\$ It's not an "official" term. It's a useful construct used to indicate that your design might go wrong if you don't take the frequency of the signals into account. High-speed can mean different things for different reasons. If you're trying to make advanced logic operations out of discrete transistors, it means one thing, and if you're building a wave guide for microwave transmissions, it means another. \$\endgroup\$ – Scott Seidman Nov 15 at 14:50
  • \$\begingroup\$ For digital world depends on time delay of the trace to rise time of source. conservative rule is delay of the trace max should be rise time * 0.25 . see these 1. protoexpress.com/blog/… 2. electronics.stackexchange.com/questions/258765/… 3. \$\endgroup\$ – user19579 Nov 15 at 15:09
2
\$\begingroup\$

For you to need to consider high-speed design (controlled impedance) rules you would need traces on your PCB that are physically longer than the wavelength/10 of the highest frequency component of your signal. For continuous wave RF (and long-pulse RF waveforms) this will correspond to:

$$ L \ge\frac{\lambda}{10} $$

For digital waveforms the rule of thumb is the highest frequency component is related to the signal rise time \$T_R\$:

$$ f_{knee} = \frac{0.5}{T_{R}} $$

Of course these need to be calculated for the particular effective permittivity of your substrate material because the propagation velocity in the substrate changes based on the material used.

\$\endgroup\$
2
\$\begingroup\$

If you have a vacuum-tube computer with 100 nanosecond edges, then 10nanoSec or 20nanoSec round trip delays and reflections will be hidden in the edge waveform. In that case, your computer can be 5 feet across (5 + 5 nanosecond round trip) to 10 feet across (10 + 10 nanosecond round trip).

This vacuum-tube computer will have UGLY waveforms (lots of reflections) with 50 foot distances (100 nanosecond round trip) unless you provide some method to absorb the fast energy. Your long cables may couple into other long cables, with fast edge energy being the best transferred between wires; in that case, your signals (your logic levels) will show lots of trash coupled onto the "0" and onto the "1" levels, and you'll have slow edges because of the wire-wire transfer of fast energy; the slow edges may be clean, because the fast energy was stolen by adjacent wiring.

If you build a computer with 100 picosecond edges, you'll see two effects. (1) the FR-4 epoxy-fiberglass insulation becomes an absorber of your high-speed energy, and your fast edges become slow, thus reflections are less of a bother. (2) the fast edges will not have time to fully penetrate the copper foil, and the thinner region used by the fast energy results in higher resistance and lots of heating losses; again your fast edges become slow, and reflections are less of a bother. { using loop inductors on both sides of single-layer PCB foil, I observed about 150 nanoSeconds propagation delay; Jackson (E&M book) agrees/predicts this much delay THRU the foil. }

In between the cable-bundle-losses and the FR-4/skin_effect losses, you should use Roker Pivic's rule.

\$\endgroup\$
1
\$\begingroup\$

If your rise/fall time is much longer than the signal propagation time over the length of your traces, you're good to "get away without special PCB layout". So, basically, have a short traces and slow rise times.

  1. there are no special thresholds: the physics is always the same, it's just that at certain conditions you'll see some effects more strongly. Some people recommend the rule of thumb, though, where if the trace length is longer than a 10% of the wavelength to be a "high-speed" situation.
\$\endgroup\$

Not the answer you're looking for? Browse other questions tagged or ask your own question.