Or should we just isolate noisy signals?
There are several theories about whether a differential pair should be routed with tight coupling or loose coupling. There must be some science that can be drawn on to arrive at a rule set that makes best use of layout time, while optimizing the signal integrity of a differential pair. This article explores the advantages of tight and loose coupling.
A known industry speaker says, “Everybody knows tight coupling is best for differential signaling.” This is stated in a tone of voice that implies those who don’t know this might be lacking. I sometimes say I am from Missouri, which is the “Show-Me State.” If the need for tight coupling is true, perhaps there is some proof. I am still waiting to see it. The following discussion will look at a tightly coupled differential pair and the same pair loosely coupled.
The left side of FIGURE 1 is a 3.25Gb/s differential pair routed with the classic 5-mil line, 5-mil space rule set, and the right side is that same differential pair loosely coupled. Clearly, the tightly coupled differential pair eye diagram is not as good as the loosely coupled differential pair. What happened to the tightly coupled pair is that to achieve 50Ω on each of the two transmission lines (100Ω differential, for those who think this is important), they had to be made narrower than those in the loosely coupled pair and spacing set as 5 mils vs. 10 mils. The primary source of degradation on the left side is skin effect loss due to traces half as wide as the traces on the right.
A good question is why did the traces have to be narrowed in the tightly coupled case? When any metallic object, whether another trace or a plane fill, gets near a transmission line, its impedance is driven down due to the added capacitance to the fill or nearby trace. To get back to 50Ω, the trace must be narrowed to return to the original impedance.
Often, the loss shown in Figure 1 does not concern a designer because the frequency is low enough that loss does not matter. Another concern may matter, however. If tight coupling is chosen, a potential problem can arise. Should it be necessary to separate the traces to route the signals through a pin field where they must be separated, after separation the traces no longer interact with each other and their individual impedances jump up to 70Ω. This is an unacceptably large change. Therefore, make sure all the signals can be routed without any separation. This is rarely possible. FIGURE 3 illustrates the routing cases for the two examples in Figure 1.
Most modern designs have component pin densities that make it difficult, if not impossible, to maintain tight spacing over the entire length of differential pair paths. For this reason, it is advisable to create a routing rule (separation) that ensures the problem shown in Figure 3 does not occur. FIGURE 4 illustrates how the impedance of a transmission line decreases as another transmission line gets closer.
Across the bottom of Figure 4 is spacing to neighboring trace with 2 mils (51µm) on the left and 28 mils (710µm) on the right. The vertical axis is impedance,with 50Ω at the top and 40Ω at the bottom. This plot is for an off-center stripline with the height above the nearest plane 5 mils (127µm). Notice that at a separation of 10 mils (254µm) the impedance has dropped to 49Ω. If that decrease in impedance is acceptable, the spacing rule for routing has been established analytically, not by some rule of thumb such as the 2H.
Common Mode Noise
Claims have been made that common mode noise coupling is reduced by tightly spacing the members of a differential pair. It would be good to understand what common mode noise coupling is before exploring this topic. To have common mode noise on a differential pair, the noise source (usually a nearby trace) must induce the same amount of noise into each member of the pair. This requires the field strength of the noise from the source be the same for both wires of the pair.
FIGURE 5 is a plot of crosstalk in an off-center stripline layer as a function of height above the nearest plane and edge-to-edge separation. Notice as a victim trace moves away from the inducing trace, the magnitude of the crosstalk gets smaller. FIGURE 6 shows the two possible methods proposed for routing a differential pair: broadside at the top and coplanar at the bottom. In neither case is there common mode noise coupling. The reason is the magnitude of the offending signal diminishes as the location of the victim signal moves farther away, as shown in Figure 5.
The best rule for routing a differential pair is to ensure any offending signal that could cause excessive crosstalk is kept far enough away from either member of the differential pair, so noise targets are met.
Tight coupling of differential pairs has few, if any, benefits. The best routing rule for differential pairs is a “not closer than” rule. This ensures there will never be a situation where needing to separate the members of a pair results in an unacceptable impedance change. This same principal applies to single-ended signals on a design as well.