You are on page 1of 11

CONCEPTS OF SIGNAL INTEGRITY:

CROSSTALK

MENTOR, A SIEMENS BUSINESS

W H I T E P A P E R

P C B D E S I G N

w w w . m e n t o r . c o m
Concepts of Signal Integrity: Crosstalk

INTRODUCTION
One of the most common signal integrity issues that engineers have to deal with is crosstalk. It crops up in every
design. Defined as the electromagnetic interference between two signals, crosstalk can cause jitter, false switching,
and timing problems. In this paper, we’ll explore the essential principles of crosstalk and what you can do to reduce it.

To anticipate whether a net might be a victim of crosstalk, we’ll take a look at the root cause of crosstalk and
simplify the fundamental principle as the capacitive coupling and inductive coupling between two signals.

With fundamentals explained, crosstalk is investigated in more detail to understand what the difference is between
near-end crosstalk and far-end crosstalk. As will be seen, each type of crosstalk has its own electrical signature that
can help designers identify such phenomena simply by looking at a time-domain transient simulation.

ROOT CAUSE
Typically, a third of a design’s noise budget is allocated to noise coming from crosstalk so it’s critical to investigate
crosstalk during the signal integrity analysis of a design.

To solve crosstalk issues, it is important to understand the root cause. Coupling between two transmission lines
occurs due to fringe electric fields (E-fields) and magnetic fields (H-fields). When a signal propagates on a
transmission line, it will generate fields as shown in Figure 1. The E-fields, in blue, are lines emanating from the
signal and return paths on which the signal is propagating and couples to all surrounding metal. That E-field will
induce a voltage on any conductors lying inside the field. Similarly, the signal will also generate H-fields that will
induce currents on the surrounding metal.

Figure 1: Coupling between traces: E-fields lines in blue, H-fields loops in red.

The coupling mechanism can also be described using mutual inductances and capacitances. A signal return path loop
has a loop inductance. Any two loops in close proximity will have a loop mutual inductance between them. A signal
carrying a time-varying current, di/dt, will couple from one loop to the other through this mutual inductance. Also, the
same signal will have a time-varying voltage, dV/dt, and that will couple capacitively to neighboring traces.

Based on this, it is important to keep in mind that as a signal propagates down a trace, the coupling takes place at
the location of the transitioning edge, where the dV/dt and the dI/dt are. As a signal propagates, the edge will have
a spatial extent along the interconnect. Whether it is a falling or a rising edge, the time-varying fields exist where
that edge is. The steady state part of the signal does not contribute to coupling since it contains no time-varying
voltages or currents.

w w w. m e nto r. co m
2
Concepts of Signal Integrity: Crosstalk

It is important to mention that once a signal couples onto a trace as noise, the noise will split and propagate in
both directions as shown in Figure 2.

Figure 2: Coupled signal at signal switching edge.

NEAR-END AND FAR-END CROSSTALK


Crosstalk noise, once generated on a victim line, will propagate toward both ends of the quiet line. When it arrives
at the ends, the signature of the waveforms appearing at the two ends will be very different. To distinguish these
ends, we refer to the end near the source as the near end. Since the noise appearing at the near end has been
propagating backward compared to the direction the signal propagates, we also refer to this end as the backward
end.

The noise appearing at the end far from the source is referred to as the far-end noise, and likewise is propagating in
the forward direction compared to the direction of propagation of the signal.

In order to explore those signatures, take a look at the circuit in Figure 3. To the left are a pair of coupled
microstrips with 5-mil edge-to-edge spacing between them. The top microstrip, TL1, has a driver model attached
to it and will be the aggressor net. The bottom microstrip, TL2, is the victim net.

Figure 3: The circuits on the left show coupled microstrips; on the right are coupled striplines

w w w. m e nto r. co m
3
Concepts of Signal Integrity: Crosstalk

To the right are a pair of coupled striplines, also with 5 mil edge-to-edge spacing. The top stripline, TL3, is the
aggressor net and the bottom TL4 is the victim net. All transmission lines are terminated with 50 ohms to avoid any
effects from reflections for the time being. All the transmission lines are 10 inches long, with the microstrips having
a TD (time delay) of 1.541 ns and the striplines having a TD of 1.757 ns. The driver being used has a 10-90 rise time
of 190 ps and a 10-90 fall time of 198 ps. This means the coupled lengths are at least eight times the rise time of the
aggressor signal, thus allowing for a lot of coupling.

To simplify matters, two terms should be defined. Near-end crosstalk, or NEXT, is the noise that is measured at the
pin of the victim net that is close to the driver of the aggressor net (pins R3 and R6 in this case). Far-end crosstalk,
or FEXT, is the noise measured in the victim net away from the driver and closer to the receiver of the aggressor net
(pins R2 and R5 in this case).

As a signal edge propagates down a coupled region between two traces, it is continuously inducing a voltage and
a current in the victim net. At each time increment, the propagating edge will induce a finite amount of noise and
then proceed another time step and induce more noise. These small noise increments are circled in red in Figure 4.

Figure 4: Coupling of signal edge onto victim. (The application that generated this image is available at www.bethesignal.com)

w w w. m e nto r. co m
4
Concepts of Signal Integrity: Crosstalk

At the near end, pins R3 or R6, the small noise increments arrive one after the other and create a wide pulse. See
the dashed ellipse circled in Figure 4. On the other hand, at the far end, pins R2 or R5, the noise increments closely
shadow the propagating edge. With the edge continuously adding more noise at each increment, the noise
increments overlap and the resulting noise at the far end will look like a pulse, as circled in the dotted ellipse in
Figure 4. Keeping in mind that these two different signatures of the FEXT and NEXT can help uncover crosstalk in a
specific waveform.

A rising edge simulation is run and plotted in Figure 5. For the microstrip case, the plot in red is for the NEXT at pin
R3. As expected, it shows a wide pulse having a peak of 42 mV. The output of U1.AF24 switches up to 800 mV, so
the amount of near-end noise coupling onto TL2 is 40/800 = 5%.

Figure 5: Crosstalk due to rising edge.


For microstrip: In red, NEXT at R3; in blue, FEXT at R2.
For striplines: In orange, NEXT at R6; in green, FEXT at R5.

At the far end, the noise at R2 is plotted in blue. It shows a -360-mV negative peak followed by another ripple in the
waveform that is circled in a dashed line. The source of the ripple will be explained shortly.

The question is, since a rising edge was run, why did the far end exhibit a negative peak? Due to Lenz’s law, which
describes the direction of circulation of the current, the inductively coupled noise propagating towards the far end
will be a negative peak. Also, due to Lenz’s law, the inductively coupled noise propagating towards the near end
will be a positive peak. On the other hand, the capacitive coupled noise is positive. The sum of both results in a
negative peak since the inductively coupled noise is larger. With the aggressor signal switching up to 800 mV, the
coupled noise is 360/800 = 45%!

Because both a 360-mV negative pulse and a 40-mV wide pulse exist on the victim net, both will feed noise back
onto the aggressor net. By reciprocity, 40% of the -360 mV FEXT and 5% of the 40 mV pulse (although this value is
insignificant), will be injected as noise and propagate all the way to the driver and the receiver.

w w w. m e nto r. co m
5
Concepts of Signal Integrity: Crosstalk

The resulting noise on the aggressor is shown on the red plot in Figure 6. It is measured at ~1.54 ns which is the
time delay of the 10-inch microstrip. Once the ripple reaches the receiver, it will propagate through the package,
reflect, and then propagate back through the package and onto the microstrip. The ripple now propagating
towards the receiver R1, couples 40% FEXT noise onto the victim net. This noise reaches R2 1.6 ns later as measured
on the plot.

Figure 6: Coupled noise feeding back into aggressor net.

Look at the stripline case in Figure 5. The orange plot is for the near-end noise at R6 and has a magnitude of 66 mV.
This is 66/800 = 8.2% coupling. Notice that it is higher in magnitude and wider than the NEXT at R3. Since the
return planes are far away, the fringe fields extend more to the quiet line than in the case of the coupled
microstrips. Also, since a stripline has a longer TD than the microstrip, the medium slows the propagation velocity
and the propagating aggressor signal has more time to couple.

The far-end crosstalk at pin R5 in Figure 5 is plotted in green and is almost zero! This is indirectly answered in Figure
7. Since all the fields propagate in the same medium for a stripline, both the inductively coupled noise and the
capacitively coupled noise arrive at the receiver with equal magnitude but opposite phase. Thus, they cancel out
almost completely. This is contrary to the situation of the coupled microstrips where some of the fields propagate
in air and some inside the medium.

w w w. m e nto r. co m
6
Concepts of Signal Integrity: Crosstalk

Figure 7: Field plot: Top, coupled microstrips; bottom, coupled striplines.

The plots shown in Figure 5 are a result of a rising edge. When the polarity of the edge is reversed, the NEXT and
FEXT peaks should be reversed. As far as the magnitude is concerned nothing should change, assuming the rising
and falling edge are identical.

w w w. m e nto r. co m
7
Concepts of Signal Integrity: Crosstalk

REDUCING CROSSTALK
One of the main ways to reduce far-end crosstalk is to reduce the coupling length between the two traces. As an
experiment, the coupling length was reduced from 10 inches to 5 inches. Another rising edge is simulated and
plotted in Figure 8. As can be seen, now the far-end peak has dropped from -360 mV to -250 mV! The NEXT pulses’
magnitudes remained at 44 mV and 66mV for microstrips and striplines respectively.

The FEXT scales linearly with coupled length, so the magnitude of the peak is reduced as the coupling length is
reduced. The magnitude of NEXT does not scale with length and is unchanged when the length is changed, but it
lasts for a shorter amount of time. The NEXT signal is on for the round-trip time of the coupled length. The duration
of the NEXT decreased when the coupled length decreased.

NEXT reaches its maximum peak (saturates) when the coupling length is greater than half the rise time of the
aggressor signal. Notice that, in Figure 8, the width of the pulse for the microstrip, measured from when its edges
cross 20 mV, is now 1.56 ns, down from 3.15 ns in the 10-inch coupling case. The width of the NEXT for the stripline
case also decreased when the coupling length was reduced.

Figure 8: Crosstalk due to rising edge with coupling length reduced to 5 inches.
For microstrip: In red, NEXT at R3. In blue, FEXT at R2.
For striplines: In orange, NEXT at R6. In green, FEXT at R5.

w w w. m e nto r. co m
8
Concepts of Signal Integrity: Crosstalk

IMPACT FROM MULTIPLE AGGRESSORS


It is valuable to understand how multiple aggressors can affect NEXT and FEXT. In Figure 9, the bottom circuit
(made up of U9.AF24 and U10.48) is the victim net, with NEXT and FEXT measured at those two pins respectively.
The spacing between the transmission lines of each adjacent circuit is 5 mils. The dotted green lines show which
transmission lines are adjacent to each other.

Figure 9: Multiple aggressors on victim circuit U9.AF24 - U10.48.

The issue to investigate here is to see how far away an aggressor can be and still contribute significant NEXT or
FEXT. As a start, only U7 is turned on and is driving a rising edge. Then each of the other drivers is turned on
subsequently. The pin U9.AF24 is pegged low, so that any voltage measured at U9.AF24 or U10.48 would be due
only to crosstalk. The simulations are plotted in Figure 10 with FEXT measured at U10.48 and plotted in red, and
NEXT measured at U9.AF24 and plotted in blue.

w w w. m e nto r. co m
9
Concepts of Signal Integrity: Crosstalk

Figure 10: Measured crosstalk with aggressors turned on subsequently.


In red, FEXT at U10.48. In blue, FEXT at U9.AF24.

Looking at NEXT first, the adjacent aggressor couples 33 mV on the victim net. The other four combined add just
another 9 mV, which is only 1% of the aggressor’s voltage. This shows that the bulk of the coupling comes from the
adjacent net.

Similarly, with FEXT, the adjacent net couples 152 mV of noise, with the second-closest adding around 55 mV. The
other two, more-distant nets combined add only another 38 mV. This also shows that the highest contributor, by
far, is the adjacent net.

The waveforms of the NEXT and FEXT show multiple dips and peaks that were not visible when only a resistor was
placed on the far end. A measurement between the FEXT peak and the peak in the NEXT waveform located close
to 1.9 ns shows a time difference of 715 ps. This indicates that the FEXT signal reaches the far end, reflects back due
to the input gate capacitance of U10.48, and reaches the near end at U9.AF24, approximately one time delay of the
microstrip interconnect.

In that experiment, the aggressors were all driving with the same polarity. What would happen if some were
driving with the opposite polarity? Since the coupled noise peaks could be negative or positive, there should be
some cancellation depending on the aggressor’s polarity. To test this idea, the adjacent driver was set to drive a
rising edge, while the other drivers were driving a falling edge. As can be seen in Figure 11, both the NEXT (in
green) and FEXT (in orange) dropped in value when compared to the case when all aggressors were driving in the
same direction. This is indeed due to cancellation as per our earlier expectations.

w w w. m e nto r. co m
10
Concepts of Signal Integrity: Crosstalk

Figure 11: Measured crosstalk when all aggressors driving with same polarity.
In blue, NEXT at U9.AF24. In red, FEXT at U10.48.

CONCLUSION
This paper explained that coupling can happen through two phenomena: capacitive and inductive noise. Because
of that, only the switching edge of a signal will contribute to coupling whereas the steady state section will not.

Once a signal couples onto a victim net, the noise will propagate in both directions. The signature of a near-end
crosstalk is a wide pulse, whereas the electrical signature of a far-end crosstalk is a narrow pulse. Also shown in this
paper is that the biggest noise contributors are the closest aggressors to a victim net.

We hope the information in this paper helps you better manage signal integrity issues in your next PCB design.

For the latest product information, call us or visit: w w w . m e n t o r . c o m


©2017 Mentor Graphics Corporation, all rights reserved. This document contains information that is proprietary to Mentor Graphics Corporation and may
be duplicated in whole or in part by the original recipient for internal business purposes only, provided that this entire notice appears in all copies.
In accepting this document, the recipient agrees to make every reasonable effort to prevent unauthorized use of this information. All trademarks
mentioned in this document are the trademarks of their respective owners.

MGC 01-18 TECH16810

You might also like