Sunday, September 7, 2008

Signal Routing for High Speed Signals

An important part of ensuring signal integrity is in the physical routing of signal traces. The PCB designer is often under pressure, not only to shrink designs but also to maintain signal integrity. Finding the balance is a matter of knowing where problems can occur and how far the envelope can be pushed before the system fails.

High-speed currents cannot cope with discontinuities in the signal trace. Among the most common and problematic discontinuity is the right-angled corner as shown below picture A. Whilst right-angled corners work without problem at low-frequencies, at high-speeds they radiate. Instead, right-angles can be replaced by a mitered 90º corner (Picture B), or by two spaced 45º corners (Picture C).

Corners tighter than 90º should not even be considered for high-speed signals.

Another common problem is stub traces. Unless there is a specific reason for using them, all stubs should be eliminated from the board. The problem is that at high frequencies, stubs can radiate as well as creating a host of impedance problems for signal traces.

Yet another key area in high-speed design is the routing of differential pairs. Differential pairs operate by driving two signal traces in a complementary fashion. Differential pairs offer excellent immunity to noise and improved S/N ratio. However there are two constraints in realising these advantages:

  1. The two traces must be routed adjacent to each other; and
  2. The two traces must be matched in length.

Problems rise when a pair has to be routed around a bend as shown in below.

The problem is to route a differential pair between two components that aren’t aligned. The solution in Picture A is flawed because the track on the outside is clearly longer than the track on the inside. The correct solution is shown in Picture B. Here a left-hand turn is followed by a right-hand turn so both tracks are forced to be equal length. This illustrates a general rule in routing differential pairs: follow each bend by another in the opposite direction.

No comments:

Post a Comment