TECHNICAL PUBLICATION

Adjusting Signal Timing (Part 2)

Crosstalk effects in serpentine traces
Douglas Brooks, President
UltraCAD Design, Inc.
January 2004

w w w. m e n t o r. c o m
ABSTRACT
When a signal passes through a serpentine trace with
coupling between the legs, there is an apparent
speed-up of the signal. That is, the signal appears to
pass through the serpentine section faster than the
trace length would otherwise indicate. This apparent
speed-up is caused by crosstalk coupling between the
legs of the serpentine traces. The amount of apparent
speed-up is directly related to the coupling strength Figure 1: When a trace loops back upon itself, crosstalk can occur between
between the legs and inversely related to the rise time the two legs of the trace.
of the signal passing through the section. The
apparent speed-up of the signal is not directly related "backward crosstalk"3 and it travels toward the left. As
to the coupled length. For long coupled lengths (those such, this backward crosstalk signal is clearly now in
longer than the critical length) signals may become front of the main signal. When the main signal travels
distorted as they pass through the serpentine section, to point (b), it is still coupling a backward crosstalk
but the degree of distortion is a complex function of signal into the opposite trace segment, still in front of
the frequency of the signal. Signals pass relatively the main signal.
undistorted through short coupled serpentine sections.
When the signal "rounds the corner" to position (c ),
BACKGROUND and begins moving to the left, it begins coupling a
backward crosstalk signal into the opposite trace
Part 1 of this series1 discussed signal timing issues in segment, now behind the signal. As the signal
general, and why and how we can tune trace lengths continues on, it continues to couple into the opposite
to achieve certain timing objectives. Ordinarily, it does trace segment, generating a backward crosstalk signal
not matter much what type of pattern we use to tune behind the main signal. Since this situation is
traces. But there can be unexpected signal timing and symmetrical, these two coupled backward crosstalk
integrity effects if we use timing loops that are closely signals, one in front of the main signal and one behind
spaced. It is often convenient to add trace length by it, will be almost identical in shape.
extending a trace some distance and then allow it to
fold back on itself, perhaps several times. We typically Thus, when the signal reaches the end of the entire
call this type of serpentine pattern "tromboning." When trace (arriving at the receiver), there will be three
the loops are closely spaced, crosstalk can occur components to it. There will be a crosstalk component
between them that has interesting consequences. that arrives earlier than the signal, the main signal
itself, and then a crosstalk component that arrives
It is generally known that these crosstalk effects can later than the signal.
exist. Howard Johnson has observed, "Short, coupled
switchbacks produce smaller delays than the total WHAT THE CROSSTALK COMPONENTS
trace length would indicate. Long, coupled
switchbacks distort the signals."2 It is interesting to LOOK LIKE
explore why this happens. And when we do so, some
interesting conclusions result. When one trace couples into another (for "long"
lengths), the backward crosstalk signal looks roughly
HOW CROSSTALK OCCURS trapezoidal (see Figure 2). It is truly trapezoidal if the
aggressor signal is a perfectly linear ramp (modeled in
Consider the trace pattern illustrated in Figure 1. The Hyperlynx, below.)
trace extends out and then doubles back upon itself. A
signal starts to propagate down the trace and is
shown in three possible positions, (a), (b), and (c),
respectively.

When the signal first starts down the trace (traveling to

the right), it couples into the return leg of the trace
directly opposite the signal. This coupling exists as Figure 2: A backward crosstalk signal looks roughly trapezoidal.

PAGE 1
The crosstalk signal rises from zero to some peak illustrates the results of this model for three coupled
value, then holds at the peak value throughout its lengths, 12" (d) (twice the critical length), 6" (c) (the
length, and then returns to zero. The width of the pulse critical length), and 3" (b) (one-half the critical length).
is generally understood to be twice the propagation
time through the coupled region. It is important to note The results show clearly that the backward crosstalk
that the magnitude of this peak value does not change signal for coupled regions longer than the critical
with coupled length (for "long" coupled lengths). But length resembles a trapezoid when the aggressor
the width of the pulse (twice the propagation time signal is a linear ramp. This reduces to a triangle (with
down the coupled length) does change with coupled the same magnitude) when the coupled region is
length.4 The pulse looks more rounded for "short" exactly the critical length. For coupled regions less
lengths. The boundary between "long" and "short" than the critical length, the shape resembles a
lengths is called the "critical length." trapezoid again, but it is more accurately described as
a "flattened" triangle.
The backward crosstalk pulse reaches a maximum
value (for any given set of circumstances) when the Rise time: In general, the backward crosstalk signal
coupled region is longer that the "critical length." This continues to increase as long as the aggressor signal
is the same critical length we refer to in transmission is increasing (i.e. as long as di/dt is positive). It stops
line analyses.5 It is a length of trace where the increasing as soon as the aggressor signal stops
propagation time along the length equals one-half the increasing. Thus the rise time of the backward
rise time of the aggressor signal. For example, in FR4, crosstalk signal is the same as the rise time of the
the propagation time is typically about 6"/ns. One half aggressor signal. This can be seen in traces (c) and
of that is 3". So the critical length for a one-ns rise (d) in Figure 4, where the rise time of these signals is
time signal would be 3". A 2.0 ns rise time signal the same as the rise time of the aggressor signal, 2
would have a critical length of 2 x 3" or about six ns. The slope of the backward crosstalk signal when
inches. the coupled length is less than the critical length is the
same as in the other cases; but the signal does not
Figure 3 illustrates a HyperLynx model of a coupled have time to rise to the same magnitude as when the
trace and Figure 4 illustrates the results from this coupled length is greater than the critical length.
model. The aggressor signal (a) is a linear ramp taking
2 ns to rise from zero to full value. The coupled traces
are assumed to be in a stripline environment (so there
is no forward crosstalk) with FR4 dielectric whose
relative dielectric is such that the propagation time is
6"/ns. The backward crosstalk signal reflects off the
front (left) end of the victim line and arrives at the far
end of the line one propagation delay after the
aggressor signal enters the coupled region. Figure 4

Figure 4: Model response for an aggressor signal (a) rising in a ramp-like

pattern in 2 ns inducing a backward crosstalk signal in coupled regions of 3"
(b), 6" (c), and 12" (d), respectively.

Pulse width: The width of the backward crosstalk pulse

is twice the length of the coupled region. This width is
measured from the approximate midpoint of the rise
Figure 3: Model for evaluating backward crosstalk response to a ramp and fall times of the pulse. (Alternatively, it could be
aggressor signal

PAGE 2
measured between the points where the pulse just does not have time to rise to this magnitude.
starts to rise [where the aggressor just starts to rise
entering the coupled region] to the point where the Howard Johnson shows that the maximum magnitude
pulse just starts to fall at the far end [where the of a backward crosstalk pulse is proportional to6
aggressor just stops rising and levels off at the far end
of the coupled region]). So if we had a main signal 1 1
(which plays the role of an aggressor signal in this 1+(D/H)2
type of situation) with a 2 ns rise time, and a coupled
length of 12 inches (2.0 ns), then the dimensions of We can consider this to be a crosstalk coupling
this backward crosstalk signal would be as shown in coefficient. We multiply this coupling coefficient times
Figure 5. The rise time is 2 ns and the width (as the magnitude of the aggressor signal to get an
defined) is 4.0 ns (twice the length of the coupled approximate worst-case backward crosstalk
region). The total width of the backward crosstalk magnitude. This is approximately the magnitude of the
pulse, from beginning to end, would be 6.0 ns (2 x the pulse if the coupled length exceeds the critical length.
coupled length plus one rise time). This result can be For coupled regions shorter than the critical length,
seen in the modeled results shown in traces (c) and the approximate maximum magnitude is proportionally
(d) in Figure 4. less.

Reality: These signal patterns have been represented

as trapezoids and triangles for convenience and
understanding. Real signals would look very much like
these figures if the aggressor signal had absolutely
linear rise and fall times. In truth, of course, that isn't
the case. Real signals are much more rounded in their
appearance, as shown for reference in Figure 6.

Figure 5: Timing diagram (in ns) for a backward crosstalk signal when the
aggressor signal has a 2.0 ns ramp rise time and the coupled region is 2 ns
(12 in) long.

Why do we measure the pulse width between the

midpoints on the rise and fall times? The answer is
somewhat subtle. Consider the rise time of an
"aggressor" pulse. The very first part of the rise time of
the aggressor pulse begins to couple into the victim
pulse and begins generating a backward pulse whose
length is twice the coupled length. This is the very
beginning of the backward pulse. The very last part of
the rise time also couples into the victim trace and it
also generates a backward pulse whose pulse width is
twice the coupled length. The very end of this
component of the backward crosstalk pulse is the end
of the total backward pulse. So the entire backward
crosstalk pulse length, from its beginning to its end, is Figure 6: The same model results as Figure 4 when the aggressor signal is a
twice the coupled length plus the rise time of the more typical TTL driver.
aggressor signal.
CONCEPTUAL RESULTS
Pulse Magnitude: For coupled regions longer than the
critical length, the magnitude of the backward We can now sketch a conceptual view of what the
crosstalk signal rises to a maximum constant value result might be of a signal flowing through a trace that
and then increases no further. This is shown in curves folds back on itself (as shown in Figure 1). Figure 7
(c) and (d) in Figure 4. For coupled regions shorter illustrates the signal components of Figure 1 when the
than the critical length, the backward crosstalk pulse rise time is fast relative to the coupled length (i.e. the
coupled length is greater that the critical length).

PAGE 3
 up is shown in the figure by the symbol t. The rise
Trace (a) is the signal as it enters the coupled region time is shown by Tr. The magnitude of the signal is V,
at the left side of Figure 1. Trace (b) is the same and the magnitude of the backward crosstalk pulse is
signal (component) after it rounds the corner and h. The ratio, h/V is simply the crosstalk coupling
leaves the coupled region. This is what the signal coefficient.
would look like at the end of the trace if there were no
coupling. Trace (c) is the leading crosstalk component By similar right triangles we have the relationship that
of the signal and trace (d) is the lagging crosstalk t/Tr is the same as h/V, the coupling coefficient. So,
component of the signal. The true signal is the sum of the speed-up, t, is found from:
components (b), (c), and (d). This is drawn as trace
(e). t/Tr = h/V, or
t = Tr * coupling coefficient

Therefore, stronger coupling and longer (slower) rise

time lead to greater apparent speed-up of the signal.

Figure 7: Conceptual results.

If we expand the central region of Figure 7 (see

Figure 8) the arrows point to the apparent decrease
(speed-up) in propagation time of the signal through
the serpentine trace. The signal is not actually Figure 9: The degree of speed-up is related to the rise time and the coupling
speeded up at all. It is the magnitude (level) of the coefficient.
signal that is increased by the addition of the leading
and lagging crosstalk signal components that causes Just as we have printed circuit design guidelines for
the effect of decreasing propagation time. We will see controlling crosstalk (by effectively controlling the
that the degree of apparent speed-up is related to the crosstalk coupling coefficient), we can develop similar
"tails" of the leading and lagging crosstalk signal PCB design guidelines for controlling the percentage
components (inside the circle), which are in turn speed-up (t/Tr * 100) by controlling the crosstalk
related to the coupling and to the rise time of the coupling coefficient.
signal itself. Conceptually the situation is the same for
shorter coupled regions, although that is a little more MODELING THE EFFECTS
difficult to draw.
The Hyperlynx™ LineSim software is a powerful tool
that allows us to model the effects of crosstalk when
we have closely spaced serpentine traces. But more
than modeling the result, Hyperlynx can also be used
to model the individual (three) components that can
exist when traces fold back on themselves. Looking at
the three components individually can give us much
better insight into what is happening.

Figures 10 and 11 show the basic form of the

Hyperlynx models used in this paper. Figure 10
Figure 8: Close-up of the central section of Figure 7.
illustrates the model for looking at the individual
crosstalk components. The individual components can
ESTIMATING THE DEGREE OF SPEED-UP be summed together to represent the model for the
complete result. Figure 11 models the complete result
Conceptually, the degree of speed-up can be directly.
estimated with the help of Figure 9. This is the same
as Figure 8 with different labels. The degree of speed-

PAGE 4
Individual Models: Figure 10 is a model of the the single, complete result while the other allows us to
serpentine trace shown in Figure 1 that allows us to look at the individual components making up the
look at the three components (the leading crosstalk result. Each should lead to the same overall
signal, the signal itself, and the trailing crosstalk conclusion for any given set of assumptions.
signal) individually. This particular model shows the
driver to be an ultra-fast CMOS driver with a very fast
rise time. The driver signal immediately enters a
length of (serpentine) trace that is coupled to (the
return) trace. The two lengths of trace (the outgoing
length and the serpentine return) are the same length,
3", and have, in the model, an impedance of 45.8
Ohms. The propagation time through each length is
527 ps.

Figure 11: Modeling the serpentine trace directly.

MODEL RESULTS
Influence of relative coupled length: The following
sets of figures (Figures 12-14) illustrate the results of
the model for three different assumed drivers, the
generic "USERMOD" CMOS 3.3V drivers with three
Figure 10: Modeling the individual components of the serpentine trace. different rise times (speeds.) The coupled length (3")
is the same for every case. The relative coupled
When the signal enters the coupled region, it length (compared to the critical length) changes in
immediately couples to the other side. The arrow each case because the rise time changes in each
labeled "Leading" shows where the leading crosstalk case. The left panel (a) of each figure shows the
signal can be seen. At the point in time when the driven signal as it enters the beginning of the
signal exits the coupled region, we have only seen the serpentine trace and the three individual components
first half of the leading crosstalk signal (remember it (leading crosstalk signal, the signal itself, and the
has a pulse width twice the coupled region.) And the lagging crosstalk signal) at the other end of the trace.
signal itself does not show up at the same point until it The right-hand panel shows the driven signal as it
completes the return loop 527 ps later. Therefore, to enters the beginning of the serpentine trace as well as
get the signal component and the leading crosstalk the true signal at the end of the trace compared to
component to line up correctly, we must delay the what the signal would have looked like without any
signal another 527 ps through an uncoupled line to a serpentine crosstalk. Figure 12 illustrates the case
point labeled "Signal". Finally, since the trailing where the coupled length is much greater than the
crosstalk signal looks exactly like the leading one, if critical length (relatively fast rise time), Figure 13
we delay the leading one another 527 ps (1,054 ps illustrates the case where the coupled region is
total) we can have signals in the model lined up roughly equal to the critical length, and Figure 14
exactly as we want them. The point labeled "Lagging" illustrates the case where the coupled region is much
in Figure 10 shows where we can see this in the shorter than the critical length (relatively slow rise
model. time).

Combined model: Figure 11 illustrates the combined For long coupled regions (compared to the critical
model for the same circuit. The signal simply passes length), the effect of the leading and lagging crosstalk
down through one length of coupled trace and back pulses are very pronounced. The signal at the far end
through the other length of the same trace. We look at of the region shows a pronounced step before rising,
the result at the point where the arrow labeled "Model" and a pronounced overshoot before returning to its
points. normal value. These effects become more gradual as
the length of the coupled region decreases (relative to
The difference between, and advantage of, these two the rise time).
different models of the same trace is that one gives us

PAGE 5
Figure 12a & 12b: Here the coupled region is much longer than the
critical length. The rise time is about 160 ps and the apparent speed-up
is about 40 ps.

Figure 15: Influence of rise time on the apparent speed-up of the signal.
Slower rise times, everything else equal, lead to a greater apparent
speed-up of the signal.

Figure 13a & 13b: Here the coupled region is approximately the same as
the critical length. The rise time is about 900 ps and the apparent speed-
up is about 210 ps.

Figure 14a & 14b: Here the coupled region is much shorter than the
critical length. The rise time is about 1.50 ns and the apparent speed-up
is about 310 ps.

Determinants of apparent speed-up: For any given

trace configuration, the degree of apparent speed-up
is directly related to coupling and to rise time (as was
described in Figure 9). Figures 15 and 16 illustrate Figure 16: Influence of coupling on the apparent speed-up of the signal.
Tighter coupling, all other things equal, leads to increased apparent
these effects. They both reflect a relatively long trace speed-up of the signal and increased overshoot.
length (longer than the critical region) in order to
Influence of wave shape: It is interesting to see what
eliminate the effects that shorter traces may introduce.
happens as we send a more typical wave shape (a
clock square wave, for example) through a serpentine
For coupled lengths longer than the coupled region,
trace. The following figures illustrate the results of the
the apparent speed-up of the signal does not change
model for various combinations of signal rise time and
as coupled length changes. This is shown in Figure
frequency.
17. Even though the coupled length has increased
significantly, the apparent speed-up of the signal
doesn't change at all. For traces shorter than the
critical length, the apparent speed-up of the signal
does appear to change with coupled length, but this is
totally explained by the fact that the effective coupling
is changing.

PAGE 6
Figure 17: Changing the coupled length has no effect on the apparent
speed-up of the signal.

Figure 18 illustrates the very fast rise time signal

through the 3" coupled region with a relatively slow
frequency (50 MHz). In this context, the coupled
region is long. If the frequency of the square wave is
50 MHz, the distortion is probably not too great (which
is, of course, a decision for the circuit design engineer
to make). On the other hand, at higher frequencies the
distortion can be quite complex, depending on the
interaction of the rising and falling edges of the signal
through the coupled region. For example, Figure 19
illustrates what happens to the signal for the three
relatively closely spaced frequencies of 500, 700 and
900 MHz, respectively. The serpentine trace becomes
almost tuned to specific frequencies. This is one of the
very few instances in the area of signal integrity where
the issue is truly frequency, not rise time!

Figure 19a & 19b & 19c: Waveform distortion through a coupled
serpentine section depends dramatically on frequency

For short coupled regions (shorter than the critical

length), a square wave waveform passes through
without too much distortion. Figure 20 illustrates the
result of a simulation with a 75 MHz square wave
through the 3" serpentine section driven with a
relatively slow rise time driver (about 1.5 ns). The
distortion around the rising and falling edge is not too
Figure 18: Distortion of a 50 MHz signal through a long serpentine severe.
region
PAGE 7
 FOOTNOTES
1. Brooks, Douglas; Adjusting Signal Timing (Part 1),
available at [Link]
2. Howard Johnson: "Serpentine Delays", EDN Magazine,
February 2001
3. See Brooks, Douglas; "Crosstalk, Part 1 - Understanding
Forward vs. Backward", available at
[Link] Also see Brooks,
Douglas, Signal Integrity Issues and Printed Circuit Board Design,
Prentice Hall, 2003, Chapter 12.
4. Ibid.
5. Brooks, Douglas; "Propagation Times and Critical Length;
How They Interrelate," available at
[Link] See also Brooks,
Douglas, Signal Integrity Issues and Printed Circuit Board Design,
Prentice Hall, 2003, Chapter 10.
6. Johnson, Howard; High Speed Digital Design, A
Handbook of Black Magic, Prentice Hall, 1993, p. 191.
See also Brooks, Douglas, Signal Integrity Issues and
Figure 20: Waveforms through serpentine sections whose length is Printed Circuit Board Design, Prentice Hall, 2003, p. 227.
shorter than the critical length pass relatively undistorted.

SUMMARY ABOUT THE AUTHOR

When a signal passes through a serpentine trace with
Douglas Brooks has a BS and an MS in Electrical
coupling between the legs, there is an apparent
Engineering from Stanford University and a PhD from
speed-up of the signal. That is, the signal appears to
the University of Washington. During his career has
pass through the serpentine section faster than the
held positions in engineering, marketing, and general
trace length would otherwise indicate. This apparent
management with such companies as Hughes Aircraft,
speed-up is caused by crosstalk coupling between the
Texas Instruments and ELDEC.
legs of the serpentine traces. The amount of apparent
speed-up is directly related to the coupling strength
Brooks has owned his own manufacturing company
between the legs and to the rise time of the signal
and he formed UltraCAD Design Inc. in 1992.
passing through the section. The apparent speed-up
UltraCAD is a service bureau in Bellevue, WA, that
of the signal is not directly related to the coupled
specializes in large, complex, high density, high speed
length.
designs, primarily in the video and data processing
industries. Brooks has written numerous articles
For long coupled lengths (those longer than the critical
through the years, including articles and a column for
length) signals may become distorted as they pass
Printed Circuit Design magazine, and has been a
through the serpentine section, but the degree of
frequent seminar leader at PCB Design Conferences.
distortion is a complex function of the frequency of the
His primary objective in his speaking and writing has
signal. Signals pass relatively undistorted through
been to make complex issues easily understandable
short coupled serpentine sections.
to those without a technical background. You can visit
his web page at [Link] and e-mail
him at doug@[Link].

For more information, call us or visit: [Link]/pcb

Copyright © 2004 Mentor Graphics Corporation. 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 purposed only, provided that this entire notice appears in all copies. In accepting this document, the recipient agrees to make every reasonable effort to prevent the unauthorized use of this
information. Mentor Graphics is a registered trademark of Mentor Graphics Corporation. All other trademarks are the property of their respective owners.

Corporate Headquarters Silicon Valley Headquarters Europe Headquarters Pacific Rim Headquarters Japan Headquarters
Mentor Graphics Corporation Mentor Graphics Corporation Mentor Graphics Corporation Mentor Graphics (Taiwan) Mentor Graphics Japan Co., Ltd.
8005 S.W. Boeckman Road 1001 Ridder Park Drive Deutschland GmbH Room 1603, 16F, Gotenyama Hills
Wilsonville, Oregon 97070 USA San Jose, California 95131 USA Arnulfstrasse 201 International Trade Building 7-35, Kita-Shinagawa 4-chome
Phone: 503-685-7000 Phone: 408-436-1500 80634 Munich No. 333, Section 1, Keelung Road Shinagawa-Ku, Tokyo 140
Fax: 408-436-1501 Germany Taipei, Taiwan, ROC Japan Printed on Recycled Paper
Phone: +49.89.57096.0 Phone: 886-2-27576020 Phone: 81-3-5488-3030
Fax: +49.89.57096.40 Fax: 886-2-27576027 Fax: 81-3-5488-3031 01-04-JC TECH6170-w

PAGE 8