Yes, size does matter; sometimes, shorter is better.

MOST SIGNAL INTEGRITY problems stem from reflections of the signal at impedance discontinuities. Luckily, this is also the easiest problem to fix. By following a simple rule of thumb, the reflection problems created by stubs can be avoided.

We use termination strategies to manage the reflections that will arise at the ends of the traces. In point-to-point architectures, the most commonly used termination strategy is a source series resistor, with a resistance selected so that its resistance plus the output impedance of the driver equals the characteristic impedance of the transmission line.

Source series termination is a powerful termination strategy that can work with more than just point-to-point architecture. It can also be used in a bus architecture, but with one caveat.

Suppose we have a driver that has a rise time of 0.5 nsec, source series terminated to a 12 inches long, 50 [omega] interconnect, and then we realize we need to have a fanout of two, with the second device placed near the midpoint of the trace, as illustrated in FIGURE 1. If this trace is 3 inches long, we get the waveform also shown in FIGURE 2.

[FIGURES 1-2 OMITTED

This is a complicated mess. Neither received signal looks very good. By just branching off from the main trace to create a second trace, we messed up both received signals. By adding that branch connection, we created an impedance discontinuity where the branch meets the main trace.

Any signal--from any direction--landing on this branch connection point will see an instantaneous impedance of 25 [omega]. This will cause reflections. Any signal entering the branch will rattle back and forth between its open far end and the 25 [omega] branch point end. Even the signal going to the far end of the mainline trace will rattle back and forth between the open far end and the branch point, distorting the received signal.

This simple topology of just one branch point is difficult to analyze by hand. The only way to get a realistic measure of the distortion is by circuit simulation, such as the one illustrated in Figure 2.

There is more than one right way of fixing this problem of routing and terminating for a fanout of two. If we wish to use a bus topology, where the driver drives one end of the bus and the bus is terminated (either at the source or far end), we must keep the branch length short.

No matter its length, there will always be reflections at its two ends. But if we can keep it short enough, the reflections will be smeared out over the rising edge and they may not be large enough to cause problems.

When the branch is short, we call it a stub. Here's a rough rule of thumb: If we can keep the time delay of a stub shorter than 20% of the rise time of the signal, the multiple reflections might smear out and not distort the received signals beyond the noise budget. FIGURE 3 shows the received signals for a source-terminated bus with a stub whose time delay is 20% of the rise time.

[FIGURE 3 OMITTED]

If the speed of the signal is about 6 inches/nsec, then the condition of the stub time delay shorter than 20% of the rise time of the signal translates into this simple rule of thumb: The maximum stub length in inches, Len, should be kept shorter than the rise time in nsec, RT, or Len < RT. If the rise time is 0.5 nsec, the maximum stub length should be shorter than 0.5 inches.

As rise times decrease, the maximum allowable stub length will decrease. Size does matter and in this instance, shorter is better.

DR. ERIC BOGATIN is the CTO at IDI, and president of Bogatin Enterprises. Many of his papers are available on his web site, www.BeTheSignal.com; [e-mail].