Managing PCB Crosstalk

Wednesday, January 16, 2019

Author: Donald Telian, SI Guys - Guest Blogger 

“Crosstalk” occurs when energy in one signal (often called an “aggressor”) couples onto another signal (the “victim”) adversely affecting the victim signal’s performance. The aggressor/victim language associated with crosstalk indicates danger is lurking, provoking a hardware engineer to consistent vigilance. How can we tame this foe? Or, more specifically, what causes crosstalk? …when does it become problematic? …and what can you do to ensure it does not ruin your product design? Let’s have a look.

The Mechanics of Crosstalk
Over the years technology has worked against us, causing typical (unmanaged) crosstalk voltages to increase from 2% to 30% as shown in Table 1. As data rates increase and voltage margins decrease, even the smallest unexpected signal disturbance becomes problematic. As such, it’s important for engineers working in all aspects of electronics design and production to have a basic understanding of the mechanics of crosstalk.

Table 1 lists the factors that contribute to crosstalk. Intuitively, the closer signals are to each other the greater their potential for coupling or crosstalk. As signals travel “close” together (referred to as “parallelism”), the amount of crosstalk increases to the point of “saturation”; and at saturation the maximum amount of crosstalk has been reached. As shown in the Table, modern technology saturates very quickly so we don’t think about this as much as we used to. Crosstalk also grows with both voltage swing and rise time, or with increasing dv/dt and di/dt. In terms of the familiar equations I=C*dv/dt and V=L*di/dt, capacitance increases as metal moves closer together and so does mutual inductance – and hence all factors continue to combine and increase crosstalk. As such, controlling signal spacing (and, if possible, voltage swing and edge rate) directly impacts the magnitude of crosstalk in your design.

Table 1: Factors Contributing to Crosstalk, then and now

Table 1: Factors Contributing to Crosstalk, then and now

To understand how the factors interact and which factors are dominant, try entering the Table 1 values into this stripline crosstalk calculator (H=10 mils, h1=h2). Modify the parameters and observe what changes - this will enhance your crosstalk intuition. Perhaps try out the values inherent in your design.

Surprisingly, those of us who have lived through this increase in crosstalk potential have seen an overall decrease in issues. How can that be? Like other design challenges, the technology world rallied with awareness of the problem, design rules to prevent it, and design tools to ensure those rules are followed. So before we panic, let’s put crosstalk problems into perspective.

Crosstalk in Perspective
Yes crosstalk problems are real, but you might be surprised to learn I’ve encountered only three serious issues in more than 30 years – designing all types of electronic products. All three issues were found after hardware was built, and fueled new disciplines in preventing crosstalk problems prior to implementation. As the issues are instructive, let’s take a look at what caused them.

The leading cause of system-level crosstalk failures is layer-to-layer parallelism in the Z (vertical) direction. Indeed, this caused two of the three problems. One was a long section of parallelism between a “high-speed” signal and a “low-speed” signal (watch out for this, “low-speed” signals don’t get enough attention anymore). The other problem involved two serial links signals with only 100 mils of coupling through of plane cutouts. Both problems were extremely difficult to isolate, with the “aha” moments occurring during a careful study of layer-to-layer layout artwork. While layout tools may assert they DRC (Design Rule Check) these situations, I still visually overlay and examine adjacent layers for potential issues. To me, this is a situation where brainpower and experience surpasses the capabilities of computer algorithms.  

The third crosstalk issue was in package-level bond wires (remember those?) caused by interleaved inputs and outputs buffered within the IC. Crosstalk induced the inverse of the output back onto the input, and the resulting oscillation was so powerful and predictable I applied for a patent on this novel oscillator design. Who says problems can’t become inventions?

Because crosstalk problems are difficult to isolate and correct in hardware and hence severely impact a product’s schedule, the vast majority of engineers simply design it out – often times with increasing material cost. The exception to this might be very high volume products; these design teams use detailed simulation in parallel with manual layout to minimize cost. But again, the majority of product implementation teams simplify the crosstalk problem by using design rules. 

Crosstalk Design Rules
Crosstalk design rules reduce crosstalk to acceptable levels by managing the two directions signals couple within a PCB: vertical and horizontal. Vertical crosstalk is caused by signals on other layers, or “inter-layer”. Horizontal crosstalk is caused by signals on the same layer, or “intra-layer”. Crosstalk noise from each direction is handled in different ways, as follows.

Inter-layer Crosstalk
Inter-layer crosstalk problems are prevented by placing solid ground planes (shields) between signal layers. Though adding layers adds cost, solid planes solve numerous SI problems such as controlling trace impedance, return current, power supply impedance, bypass capacitor loop current, and so on. And so extra ground layers are readily added in all but the highest volume products.  This sounds simple enough, but be advised that a “solid” plane never exists in practice. As such, it’s important to verify signals will not couple through cutouts, antipads, or other gaps in the plane. In these areas, signals on both sides of the “shield” remain susceptible to crosstalk because part of the shield has been removed.

Figure 1 quantifies the impact of inter-layer crosstalk on PCIe Gen3 eye height if a ground shield layer is not in place. The eye diagrams in the corners of Figure 1 show performance without crosstalk (upper right) and with crosstalk (lower left). Because the link is short (3”), the signals are over-equalized and hence four voltage levels are seen. With no crosstalk (upper right) the eye opening is 150 mV. With crosstalk (lower left) each of the four voltage levels are widened by ~150 mV of noise, closing the eye. Simulated eye heights with increasing crosstalk are plotted in the graph, revealing how eye height decreases as the amount of inter-layer coupling increases from 0 to 200 mils on the X axis. As the gap distance between the layers is varied (red=4 mils to gold=10 mils, in 2 mil increments), eye height decreases at the rates shown in the color-coded boxes. 

Figure 1: Inter-Layer Crosstalk on PCI Gen3 Eye Height, versus Layer Gap and Coupled Length

Figure 1: Inter-Layer Crosstalk on PCI Gen3 Eye Height, versus Layer Gap and Coupled Length

Figure 1 reveals inter-layer crosstalk can cause a 1 mV decrease in eye height per 1 mil of coupling when the layer-to-layer gap is 6 mils (blue). That means only 100 mils of parallelism can remove a generous eye margin. And so, again, extra planes are readily used. Yet even if ground shields are used, make sure diff-pairs do not overlap through gaps in the plane. For example, 100 mils of coupling can easily be accumulated through cutouts under AC capacitors.

Intra-layer Crosstalk
Intra- (same-) layer crosstalk is prevented by enforcing a spacing distance between signals greater than 5h to 7h, where “h” is the distance between the signals and their adjacent ground plane(s). The design rule is stated in terms of “h” to ensure the signal’s coupling to a nearby plane (which is good) is roughly an order of magnitude greater than its coupling to a nearby signal (which is bad). In practice, this generally requires signals to be spaced about 25 mils apart. 

To illustrate the efficacy of the “5h” design rule, Figure 2 shows a crosstalk signal-to-noise ratio on the Y axis versus the spacing distance “D” between two signals on the X axis. As the Y axis is a ratio, larger values are “good” and smaller values are “bad” as shown. The colors show “h” (the stripline trace’s distance to ground in each direction) varying from 3 mils (red) to 7 mils (black), in 1 mil increments. The horizontal line marks a constant magnitude, which is the D=5h location for all values of h. For example, the h=3 mil line (red) crosses the horizontal line at 15 mils, the h=4 mil line (blue) crosses at 20 mils, and so on. While minor non-linearity is seen with small h values, the plot demonstrates how the design rule achieves a consistent crosstalk ratio across a variety of stackups and implementations.

Figure 2: Intra-Layer Crosstalk Magnitude versus Signal Spacing and Distance to Ground

Figure 2: Intra-Layer Crosstalk Magnitude versus Signal Spacing and Distance to Ground

Figure 2 illustrates both how signal quality increases (i.e., decreasing crosstalk) as spacing between signals increases (larger D), and how an acceptable crosstalk level can be reached sooner if/when signals are closer to ground (smaller h). Again, manipulating “D” and “h” is the primary mechanism for controlling intra-layer crosstalk. Consult the Design Guidelines associated with your components or technology to determine the recommended D/h ratio. I expect you’ll find it to be in the in the 5 to 7 range, unless a constant D is used instead.

As stated previously, automated layout tools are better at enforcing intra-layer than inter-layer spacing rules. As such, ground shields are typically used vertically and spacing rules horizontally. In rare situations ground moats have been used horizontally and spacing rules vertically, and the physics involved is similar to that described above.

While it’s best to prevent problems before they happen, when confronted with crosstalk in hardware don’t forget you likely have programmatic control over SerDes/DDRx drive strength, edge rate, and equalization. You may find you can fix the problem using software. For example, simply turning off the equalization in Figure 1 can restore the eye – even without removing the crosstalk.

In Conclusion
Crosstalk problems can be real, yet are not necessarily as pervasive as one might expect as long as design best practices are followed. Here we’ve discussed the factors that exacerbate crosstalk and how to manage them in practice using design rules.  Crosstalk simulation is also used to adapt the rules when lowest cost is desired. 

SiSoft users can access a detailed Crosstalk Methodology document that explains how to perform crosstalk simulation and/or utilize “Crosstalk Scan” to quickly identify and cross-probe problem areas in a PCB. Also available is a Crosstalk Analysis video series and kit that demonstrates crosstalk solutions for QSI users. 

Donald Telian, SiGuys - Guest Blogger 1/16/2019

Did you like this post?

Sign up for our mailing list and we'll let you when we post a new one.