Tricks of the trade: a little known trick for reducing crosstalk.

NEAR-END CROSSTALK in a transmission line bus can often exceed the allocated noise budget. This can be a disaster if it isn't caught before the board goes to fab.

For example, in a 3.3 v CMOS system, the allocated noise budget for crosstalk might be about 160 mV, or 5% of the signal swing. If the bus uses 50 [ohm] lines with .005" wide lines and spaces, the noise on a victim line from one adjacent trace will be about 6%, which is not too bad. But when it is part of a bus, lines on either side switching simultaneously will create more than 12% noise, more than double the allocated noise budget.

This is the origin of an easy-to-remember rule of thumb: if you never want to worry about near-end crosstalk in a 50 [OMEGA] bus fabricated with FR-4 in either microstrip or stripline, keep the spacing twice the line width. In this geometry, the near-end crosstalk on one line from all the other lines in the bus will never, ever be greater than 5%.

Wouldn't life be great if we could always follow this guideline?

Unfortunately, the downside to implementing this is that you have to space the signal lines farther apart, and this dramatically reduces the interconnect density. This may mean more layers in the board, or a larger board, alternatives that spell higher cost. This is an example of the fundamental principle: higher performance always costs more.

When crosstalk in a bus is a problem and space is a premium, there is another possible knob to tweak to offer a slightly better cost/performance trade-off. It involves using a laminate with a lower dielectric constant.

Crosstalk between transmission lines is really about the capacitive and inductive coupling per length between the adjacent signal lines. The primary term that influences the coupling is the spacing between the traces. To first order, you would not think dielectric constant influences the coupling.

After all, if the dielectric constant of the laminate decreases, the capacitance between the adjacent signal lines decreases, but so does the capacitance between the signal and return. The ratio, which determines the relative capacitive coupling, stays the same. This suggests that to first order, changing the dielectric constant should not affect the crosstalk, and it doesn't.

However, if we decrease the dielectric constant, and we want to maintain the same target impedance of 50 [ohm], with the same .005" wide traces and spaces, we have to also move the return plane closer to the signal line. The closer spacing of the signal line to the return path is what decreases the crosstalk.

Reducing dielectric constant alone does not change the crosstalk, but reducing the dielectric constant and reducing the trace-to-plane spacing will reduce crosstalk.

To analyze how much bang for the buck we can get by reducing the dielectric constant of the laminate, we can use the Polar Instruments SI8000 2D field solver to help us perform the calculations. In this example, we will use a microstrip with .005" wide lines and a .005" spacing. We'll look at the near-end crosstalk between adjacent traces.

The first step is to figure out the optimum stackup for 50 [ohm] as we change the dielectric constant. The second step is to calculate the near-end crosstalk to the adjacent trace for each dielectric constant and thickness.

The near-end crosstalk in a differential pair is, NEXT = 0.5 * (Zeven - Zodd) / (Zeven + Zodd). Using the field solver we can calculate how the NEXT changes with dielectric constant, as shown in FIGURE 1.

[FIGURE 1 OMITTED]

In each case, the dielectric thickness has been optimized to hit a 50 [ohm] target impedance. This says the lower the dielectric constant, the lower the crosstalk for a pair of lines.

In FR-4, with a dielectric constant on the order of 4.5, the crosstalk per pair might be 6%, with a total noise of 12% from the bus. If we change the laminate to a non-glass woven substrate, we might achieve a dielectric constant of 3, with a crosstalk of 4.5%, or total bus noise of 9%. This is a 25% reduction in the near end crosstalk than the same traces in FR-4.

If your design layout is crosstalk-limited, moving toward lower dielectric constant might provide more margin, or allow a smaller form factor board or fewer layers.

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