FAIRCHILD AN-817

Fairchild Semiconductor
Application Note
April 1992
Revised May 2000
Taking Advantage of ECL Minimum-Skew Clock Drivers
CLOCK DISTRIBUTION BACKGROUND
Among the factors affecting variability are:
Digital systems have tended from their inception toward
incorporation of higher speed elements rather than architectural changes as the solution to the computational
speed problem. One result of this has been mounting pressure on the semiconductor elements of these systems for
higher speed and all that this implies. Not only the operating frequency and signal propagation delay but also the relative timing relationships of devices performing parallel
tasks were challenged. As semiconductor process
improvements pushed operating speeds higher, the evolution from individually-packaged to multiply-packaged gates
helped reduce delay and speed differences among like
devices. Even so, areas still existed where system designers demanded ever greater improvements in performance
uniformity of digital logic elements. The synchronization
signal generation system, often referred to as the clock
system, is one area in need of such improvement.
• physical size such as the length of wiring,
Most digital logic systems employ some means of synchronizing the actions carried out by the system’s elements.
The precision with which these various elements are regulated can be shown to have a direct effect on the overall
speed of the system. The less time variation that must be
allocated to the logic elements in the signal path, the faster
that operation may be executed. Therefore, the overall time
to perform all operations may be reduced or the system
speed may be increased. In addition, the coordinated timing of the synchronizing system itself must benefit from an
equivalent reduction in relative timing of its elements, or
else only limited overall benefit to system performance is
possible.
CLOCK NETWORK ELEMENTS
The typical synchronizing or clock system consists of several closely related elements:
• a primary signal source, usually a precisely regulated,
high frequency oscillator,
• frequency dividers to derive lower frequencies from the
primary source,
• distribution amplifiers to boost signal power or supply
separated loads,
• signal delay, duty-cycle alteration or re-synchronizing
elements,
• interconnect wiring, connectors and signal distribution
network,
• and a power source and distribution system for the semiconductor devices in the system.
Each of these elements contributes a degree of variability
to the overall timing precision capability of the system. By
virtue of the fact that the physical properties of each element can be controlled only within certain limits, the complete system will exhibit a variability somewhat larger than
its elements taken individually. The effect of the variations
contributed by the elements to the overall system variability
is a physical fact that may be demonstrated mathematically.
© 2000 Fairchild Semiconductor Corporation
AN010982
• electrical properties such as resistance and charge,
• internally generated and externally induced noise,
• and environmental factors like temperature and altitude.
The visible effect of variability on the system timing is popularly termed “skew”. Similarly, the individual variabilities of
each element are likewise termed “skew”. The term “skew”
is used also to describe the difference in delay between
like paths taken by signals in a logic device or system.
However, as with many useful terms when carelessly
applied, the term “skew” can lead to some confusion unless
it is more precisely defined.
When applied to the differences in propagation of electrical
signals through like paths in a logic system, the term
“skew” will be taken to mean “differential timing error”. In
terms of the overall effect on the desired timing performance of the affected system or device, “skew” will be
defined as “a deviation or asymmetry from the mean or
desired timing value”. The latter meaning of skew will
assume greater importance in the prediction and analysis
of system timing errors to be covered later in this note.
MEETING THE SKEW CHALLENGE
Control of skew in systems is at best a difficult challenge
for design engineers. At its worst, skew can be the thing
that makes or breaks a design. So, to give engineers an
advantage in controlling skew, Fairchild offers ECL devices
specifically designed to minimize the device-related component of skew. In addition, the skew properties of these
devices are specified and tested. This has been done to
reduce the design effort required to compensate for devicerelated skew effects. The results are tighter system timing
margins, more reliable operation and faster operating
speeds. This means greater system through-put and information processing efficiency.
300-SERIES ECL—The Best Weapon Against Skew
300-Series ECL has some natural advantages in the war
on skew. ECL differential amplifiers are minimum-skew by
their very circuit configuration. The fundamental property of
the differential amplifier used as a logic element is balanced switching. The amplifier, Figure 1, is symmetrical.
This results in identical switching delays regardless of output. Following stages are added symmetrically, Figure 2,
so that overall balance is maintained internally. 300-Series
devices designed for minimum-skew also employ a unique
die layout to preserve the basic amplifier balance.
Another advantage of 300-Series ECL is its precise, stable
VBB reference supply with wide operating power supply
range. This reference is compensated for stability from the
effects of changes in temperature or supply voltage. 300Series minimum-skew devices operate from −4.2V to −
5.7V supplies as do all other 300-Series logic devices.
Advanced semiconductor processing, design and manufacturing result in uniquely low power consumption for this
ECL family.
www.fairchildsemi.com
AN-817 Taking Advantage of ECL Minimum-Skew Clock Drivers
AN-817
AN-817
300-SERIES ECL—The Best Weapon Against Skew
(Continued)
outputs are fully buffered emitter-followers. They are
designed to have symmetrical transitions and skew performance under symmetrical loading conditions. Some
devices also make an external VBB reference available for
use in situations where the inputs must be separately
biased.
Augmenting the inherent balanced design of the amplifier,
300-Series minimum-skew devices feature differential
inputs and outputs. Selected devices have an additional
single-ended input and input selector for use as a test input
or auxiliary main input. However, skew tolerances are
wider when using the single-ended input. The differential
FIGURE 1. ECL Differentical Amplifier
FIGURE 2. ECL Buffered Amplifier
PACKAGE TYPE AND LAYOUT
Skew can be adversly affected by packaging. For this reason Fairchild has chosen to supply F100K minimum-skew
devices in packaging complementing their inherent performance. F100K devices are supplied in SOIC and PCC
packaging as appropriate for the intended class of service;
commercial and industrial.
www.fairchildsemi.com
Packaging and pinout also affect skew performance.
Smaller, more symmetrical packaging like PLCC and SOIC
types contribute less to skew than do DIP packages. Their
smaller size reduces parasitic inductance, a contributor to
noise and signal delay. Symmetrical pin layout equalizes
path length for all signals from the die to the circuit board
thereby reducing a skew component. And, additional output VCC pins help reduce skew by providing extra paths for
output signal return currents to ground.
2
POWER SUPPLY NETWORK DESIGN OBJECTIVES
Despite careful design, packaging and advanced processing, skew performance is worth little unless its effect on
device performance can be accurately known. Reliable
skew values, guaranteed by test, are essential in predicting
system performance. 300-Series ECL minimum-skew
devices are fully tested and specified for all standard DC
and AC parameters as well as all relevant skew parameters.
ECL minimum-skew devices can benefit especially from a
well designed power supply system. Though not unique for
minimum-skew devices, the design of ECL power supply
systems must achieve several important objectives.
1. The power system must supply constant, noise-free
voltage at adequate current levels to all active devices
in the system.
2. The power system must appear as a virtual battery to
each device. That is, it must effectively isolate devices
such that each appears to be independently powered.
Thus, noise signals resulting from operations going on
within each device will not propagate to other devices
through the power supply system.
It should be appreciated that devices designed for minimum skew values of a few tens of picoseconds are difficult
to test using ordinary IC test equipment. Therefore, Fairchild has developed advanced production test equipment
capable of measuring the AC and skew characteristics of
these high-performance devices. This results in more accurate and dependable data for design and greater uniformity
of product.
3. The power supply system also must propagate without
distortion the return or image currents for all signal-carrying transmission lines.
4. The power distribution planes within each printed circuit board (FCP) should act as electrostatic shields to
reduce RFI radiation from the transmission line system.
SYSTEM DESIGN OBJECTIVES
The best way to take advantage of minimum-skew driver
capabilities is through good system design. Minimum-skew
specified devices cannot compensate for or improve on a
poor system design. However, these devices can add extra
margin to a good design. They also permit more predictable performance in well designed system environments.
These objectives can be adequately met if reasonable care
is taken in the design of the power supply feed system, the
supply isolation and bypassing and the organization and
layout of the PCB's.
More information on power supply design considerations
can be found in the “F100K ECL Logic Databook and
Design Guide”.
The majority of problems that plague clock generation and
distribution systems lies beyond the active devices and
outside of their effect. These problems result from shortcomings in the signal transmission, power supply or
mechanical systems. A good system design should give
the minimum-skew driver a fighting chance to perform satisfactorily. This is the primary job of the system designer.
POWER SUPPLY BYPASSING
The topic of power system bypassing is generally given
inadequate treatment in logic applications literature. The
common approach taken is to simply recommend that a 0.1
µF ceramic capacitor be used for every so many devices.
In systems with frequencies predominating below 10 MHz
and edge rates on the order of 1V/ns or slower, this prescription may be adequate. However, for systems with frequencies above 10 MHz and edge rates less than 1V/ns,
the above prescription is bad medicine. The reasons are
easy to explain but require a look at the role and objective
of bypassing.
But, before the designer can effectively use these devices,
a reasonable set of electrical and mechanical specification
objectives for the overall system must first be developed.
Next, the device specifications as they apply within the
scope of the system specification must be understood thoroughly. Only then can devices with performance appropriate to the overall objectives be selected. Indeed, the
process of resolving device and system objectives may
reveal incompatibilities. Then either the system requirements or the choice of device must be changed. Lastly and
most importantly, the active devices must not be overworked or handicapped by improper load conditions or the
system environment.
Bypassing serves two general roles in the power supply
system—filtering and energy storage. Both of these are
related by the fact that, in performing filtering, the capacitor
stores and releases energy. To be more specific, the four
goals of an efficient bypassing scheme are:
A good set of system objective specifications will incorporate all of the following considerations:
1. Reducing the internal impedance of the power distribution wiring system,
• A stable, adequate, noise-free and well isolated DC
power supply and power distribution system,
2. Low-pass filtering of the DC being supplied to each
active device,
• Noise-resistant and RFI/EMI-proof design and suppression techniques,
3. Isolating active devices from each other, and
• A carefully designed signal transmission and distribution
system,
4. Controlling noise due to transient current demands on
the supply system.
• Appropriate and efficient printed circuit board layout,
• The appropriate active devices for the required performance.
These four goals are interrelated and are parts of the primary objective of having each active device appear as
though powered by an internal, ideal battery. In fact, goals
2, 3 and 4 can be considered as results of achieving goal 1.
Reducing the impedance of the power system conductors
at all frequencies in the system pass-band helps accomplish all goals of efficient bypassing.
The sections that follow present design techniques and
guidelines formulated around the above objective specifications. Colateral sources where more complete or additional design information may be found will be noted or
may be found listed in the Appendix.
The “F100K ECL Logic Databook and Design Guide” ) recommends multi-layer PCB construction with entire layers
assigned to each supply voltage and ground. While this
might appear to be an ideal power distribution method for
DC, it increasingly deviates from the ideal as frequency
• Adequate cooling for all power-dissipative devices,
• Environmental requirements like altitude, humidity, and
temperature:
3
www.fairchildsemi.com
AN-817
TESTING
AN-817
POWER SUPPLY BYPASSING
(Continued)
In the time domain, the capacitor’s behavior becomes more
increases. For frequencies below about 10 MHz, the intercomplex. Fast risetime transient pulses contain frequennal impedance of power plane layers may be in the range
cies generally far above the capacitor’s self-resonant freof 1Ω to 10Ω depending upon geometry. At higher frequenquency. During these pulses, the capacitor has an
cies, and especially above 50 MHz to 100 MHz, skin and
instantaneous impedance value ten or more times its ESR.
striction effects cause a ten-fold or greater increase in
For this reason, it is often desirable to “bandspread” the
internal impedance. This means that current changes
bypass capacitor by using several different capacitor valresulting from switching of the active devices at high freues in parallel. An example of this is placing 0.1 µF, 10 nF
quencies can induce large voltages in the power planes.
These noise voltages affect nearby circuits and signals on
and 1 nF ceramic chip capacitors in parallel. The capacitransmission lines. To eliminate this interference, a means
tors may be attached to the PCB in a sandwich, “piggyis needed for decreasing the impedance of the planes at
back” style, provided their lengths are equal. Leaded
high frequencies.
capacitors may be placed close together on the circuit
board to achieve a similar result, also.
Restricting the pass-band of a wire transmission system to
frequencies approaching DC is termed low-pass filtering.
Suitably sized capacitors placed at regular intervals on a
wire transmission system form a low-pass filter. As frequency increases, the capacitor’s reactance decreases.
This reduces the amplitude of the signal as it traverses the
transmission line. Therefore, the effect of noise generated
by an IC on the power supply voltage of its near neighbors
is reduced. In addition, lowering the internal impedance of
the power plane system reduces radiated RFI and the
effects of external interference sources on the power supply system. And, the power plane system will appear as a
more nearly uniform AC ground for signal-carrying transmission lines in adjacent layers of the PCB.
The signal frequencies being transmitted by the logic
devices and the switching frequencies of device outputs
are not the only signals that must be handled by the
bypassing. There are also frequencies associated with the
energy transfer between different functional groups of components on the PCB's and in the overall system. For
instance, a memory system may be active at one-fourth of
the rate of an associated CPU group. The peak energy
exchange rate between the memory system and the power
supply would be one-fourth that of the CPU and the supply.
Though the memory may handle signals at the same frequency as the CPU during communications, the power
supply system also sees the lower frequency as well.
Therefore, the power supply system and the memory/CPU
circuit board must have bypass capacitors for both sets of
frequencies. For this reason, larger value capacitors, usually tantalum or similar electrolytic types, are found on circuit boards. These capacitors smooth energy transfers at
lower frequencies.
Another property of the transmission line, low-pass filter is
that it delays signals traveling on the line. In addition to
attenuating AC signals and noise generated by the active
devices, the signals are also delayed. This effectively
increases the isolation between devices in the timedomain. Naturally, as the delay of the power supply transmission system increases, the closer the bypass capacitors
must be to the noise source in order for them to be optimally effective.
CLOCK SOURCE NOISE CONTROL
The clock system is normally the source of most high frequency noise in digital systems. Part of the overall design
objective for the clock system is that it not create interference with other parts of the system.
Supply bypassing is generally less critical in F100K ECL
systems than in 10K systems, especially since output load
balancing can be used. Bypassing still must perform the
functions mentioned previously, but the balanced switching
of F100K results in a quieter overall system to begin with.
The primary suppression role of bypassing is in controlling
noise from unbalanced output switching on the power system.
Isolation of the clock source and its associated distribution
amplifiers is an effective way of reducing such interference.
Insertion of a ferrite choke in the power supply feed to the
active devices forms a noise filter that reduces interference
from these devices to other logic through the power system. The filter may take the form of a Pi-network or longitudinal choke as in Figure 4. In selecting the inductor, the
core must have sufficient current rating to handle the load
without saturation. Also, the core material should be capable of operating over the frequency range of the oscillator
and associated components. Type 3B or 4B core materials
are typical for ECL applications. Suitable 6-hole ferrite
chokes are available from Fair-Rite, Siemens, Ferroxcube
and others.
The selection of bypassing components for F100K systems
should be based on the expected operating frequencies of
the various networks as well as the output transition rates.
Also, the choice of balanced versus unbalanced output
loading plays a major role.
Capacitors bypass or suppress frequencies best near their
self resonant frequency, Figure 3. At this frequency the
capacitor appears to be more nearly a short-circuit. Below
resonance the capacitive component of reactance dominates. Above resonance the inductive component plays a
bigger role. The effective impedance at resonance is the
resistive component or ESR (equivalent series resistance).
It is normally desirable to minimize the ESR so that the
capacitor is a good short circuit. However, it may be more
desirable to have some resistance in series with the capacitor in order to broaden the range of frequencies where the
capacitor is effective.
www.fairchildsemi.com
CLOCK NETWORK TOPOLOGIES
The choice of clock network topology is intimately a part of
the architecture and partitioning of the system. To recommend a particular topology would be presumptive. Therefore, the scope of this discussion will be limited to more
general considerations of the networks.
4
AN-817
CLOCK NETWORK TOPOLOGIES
(Continued)
FIGURE 3. Capacitor Reactance
π-Filter
Longitudinal Choke
FIGURE 4. Power Supply Isolation Techniques
Low impedances (less than 50Ω) require wide line widths,
a problem where small packages and dense PCB layout
are desirable. Also, lower impedances require more energy
from the driver. For a given output type, this results in lower
voltage swings into the line and higher DC and AC power
since the termination value is lower, also. An advantage of
lower impedances is that of their being less affected by
capacitive loading and crosstalk.
The general forms of networks suitable for clock distribution are: point-to-point, daisy-chain, radial (star), branching
(tree), auxiliary main input and bus. Any of these can be
implemented as single-wire-over-ground or balanced differential networks. The differential method has advantages of
less suceptability to interference from noise, temperature
effects, threshold shifts and common-mode signals. These,
plus the network’s inherent balance, mean lower skew.
F100K is particularly suited to use in differential networks.
It employs differential outputs on virtually all devices and
differential inputs particularly on clock distribution devices.
The effect of loading versus line impedance is easily illustrated. A 50Ω line in FR-4 material has an unloaded propagation delay of about 56 ps/cm. Unloaded 25Ω and 100Ω
lines in this material have the same delay. By adding only
1 pF/cm load to the 50Ω line, the delay is increased by
21 ps. The 25Ω line’s delay increases 11 ps and the 100Ω
line by 38 ps. Clearly, skew-versus-loading is easier to control if the line impedance is kept low. This is the reason
that, from all practical standpoints, 50Ω is widely considered as the best line impedance choice.
Naturally, to benefit from the advantages of the differential
network, its balance must be maintained. Loading and terminations should be applied equally to both conductors of
the differential pair. If loading is not balanced, the propagation delay of each conductor will be different. This can lead
to differences in the arrival time of the complementary signals thereby increasing skew.
Regardless of the network topology, the relationships of
network impedance, drive capability, critical loading and
termination must always be considered.
Controlling the chosen impedance and skew in practical
networks can be difficult in practice. This is particularly true
where many loads need to be driven in precise phase relationship to a source. Ideally, it would be best to drive each
load in a point-to-point connection. Except for the most
exacting applications, this can be costly in terms of board
space, power and components.
IMPEDANCE, LOADING AND INTERFERENCE
The equivalent generator resistance of the F100K output
emitter follower is approximately 7Ω. With this low value,
the output is nearly an ideal voltage source capable of driving almost any reasonable value of load impedance or
transmission line type. High impedance transmission lines
(above 50Ω) are difficult to achieve in practice on PCB’s
with thin dielectrics. However, high impedances are more
common with twisted-pair lines. The propagation velocity of
energy is more affected by capacitive loading on high
impedance lines. And high impedance lines are more susceptable to crosstalk and noise interference.
Series-terminated radial or star networks can be used in
many cases to drive multiple loads from one output. As
long as the load impedance driven does not exceed output
drive capability, 2, 3 or 4 lines may be driven. Radial nets
may also be driven from the end of another transmission
line connected to the driver output. In this case, the impedance of the line between the driver output and the starpoint must be less than the impedance of any line connected to it. Each line radiating from the star-point is driven
through a series termination. The value of the resistor is
5
www.fairchildsemi.com
AN-817
DIFFERENTIAL LINE LAYOUT
(Continued)
and tested on an individual basis to assure correct operachosen to give the minimum required voltage swing at the
tion. The laws governing such networks are the same as
input connected at the receiving end of the line. More
for the basic networks and terminations. Further informadetails for designing radial networks may be found in of the
tion on terminations, other network configurations and their
“F100K Logic Databook and Design Guide”.
design requirements can be found in the “F100K ECL Logic
Timing and skew are more difficult to control with the daisyDatabook and Design Guide”.
chain network and its bi-directional equivalent, the bus network. Differences in the individual load values attached to
the net contribute to increased skew. This is especially true
in parallel-path buses where there is another complicating
factor. The bus is normally bi-directional. Whereas the
daisy-chain net has one source and multiple loads, the bus
has multiple sources. Delay from one source to a given
destination, relative to another source to the same destination, may create difficult timing situations, especially where
all sources share a common distributed clock. In such situations, consideration should be given to re-synchronization
of the common clock relative to the other signals being
transmitted from the particular location on the bus. This retimed clock will then be properly synchronized to its related
signals and a reduction in location-dependent skew on the
bus will result.
PC BOARD LAYOUT RECOMMENDATIONS
Proper PC board layout is high on the list of important considerations when using ECL minimum-skew devices. The
performance enhancement offered by minimum-skew
devices can be wasted by inadequate layout. So, a look at
some additional layout recommendations aimed at maintaining the high performance that these ECL devices are
capable is essential.
F100K ECL minimum-skew devices are available standardly in surface mount packaging. Devices use either
SOIC and PCC packages. Each package type has its own
unique mounting requirements. However, there are some
common areas where both types can benefit from attention
to layout. These areas are bypassing location, power and
ground connections, differential line layout and power
plane isolation.
TERMINATION NETWORKS
Minimum-skew drivers cannot correct the behavior of
improperly terminated networks. In fact, operation of any
driver-network combination is impaired by faulty termination. Faulty termination corrupts signals and contributes to
skew, timing and noise problems. A brief look at the goals
of basic series and parallel termination networks will clarify
their use.
Literature available from Fairchild Semiconductor is listed
in the Appendix.
BYPASS CAPACITOR PLACEMENT
The placement of bypass capacitors is as important as
their electrical characteristics. Placement and connection
into the system can significantly alter the effectiveness and
electrical performance of the bypassing elements. Ideally,
the capacitor is placed as close as possible to the device
requiring the bypass. Only an absolute minimum of wire
length should connect the capacitor to the power planes or
the IC leads. If surface mount capacitors are used, the
mounting lands are connected directly to the power planes
through vias and omit any extra wiring. Figure 5 shows the
recommended placement and connection methods for
PCC layouts on conventional multi-layer PCB's. Figure 6
shows the recommended application where SOIC packaging is used.
The goal of parallel termination is absorbtion of all transmitted energy at the network terminus with none reflected to
the source. Any signal distortion occurring is solely due to
reflections from the loads attached to the line. The parallel
termination should have a resistive value close to that of
the network's impedance to be optimally effective. Parallel
terminations may be simply a single resistor from the terminus of the network to a voltage source (normally −2V) or
they may be Thevenin networks equivalent to the single
resistor and voltage source values.
The series termination operates on the principle of Ohm's
Law. As a voltage divider, the goal of the series termination
is to reduce the driver's output voltage transmitted into the
line by about one-half. When this voltage (referred to as the
incident wave voltage) reaches the opposite end or terminus of the network, which is unterminated and approximately an open circuit, the voltage level doubles. A
receiver at the terminus of the line sees its full input voltage
swing. However, receivers between the input and terminus
ends see only the voltage transmitted into the line. All of
the voltage arriving at the terminus is reflected back to the
source with the same polarity as that of the incident wave
voltage. As this reflected wave voltage transits the line
toward the source end, the voltage level along the line is
raised to the full swing voltage of the driver's output. Only
at the instant when this voltage wave reaches each input
will that input see its full voltage swing. Therefore, intermediate loads experience a period where input voltage is at or
near threshold level. Consequently, there is a delay of
more than one line transit time until the input voltage
reaches a proper level. For this reason, series terminations
are used almost exclusively in point-to-point networks.
POWER AND GROUND CONNECTIONS
Power and ground connection paths must be kept short
otherwise excess inductance is introduced in the IC's
power leads. Excess inductance reduces the effectiveness
of any associated bypassing and makes the power supply
seen by the IC less stable. Where multiple power or ground
pins are provided on the IC package, all should be connected to the appropriate supply potential. Figure 7 shows
examples of correct and incorrect power connections to
PCC and SOIC packaged devices.
DIFFERENTIAL LINE LAYOUT
Fairchild's minimum-skew devices generally employ differential inputs and outputs as an aid in combating skew.
When combined with properly designed differential transmission line layouts, precise system timing is possible.
However, careless layout of differential lines can cause
serious timing and skew problems. By observing a few simple rules, effective differential transmission systems can be
easily designed.
In some cases, networks may require combinations of
series and parallel terminations to achieve satisfactory performance. Networks of this type must be carefully designed
www.fairchildsemi.com
1. Differential transmission lines must be organized in
pairs. Lines originating at complementary outputs must
6
(Continued)
be paired throughout their lengths. Additional spacing
should be allowed between pairs in all layers. Approximately twice the spacing between the lines of the pair
should be used between pairs.
2. Both lines of a differential pair must be of equal length.
Changes of direction must be accompanied by either
an equal and opposite change of direction or additional
length must be added to the resulting shorter line.
change of layer, both conductors should change layers.
In general, there should be an equal number of vias in
each line as the result of layer changes. The width of
the lines may be adjusted according to preserve line
impedance.
4. Corners in lines with angles greater than 45° should be
mitered or radiused to preserve impedance uniformity.
Figure 8 illustrates these rules with several ways in which
differential lines may be implemented correctly and incorrectly.
3. Differential lines should be paired (run parallel) in the
same layer. When a direction change dictates a
BEST
ACCEPTABLE
FIGURE 5. PCC Bypass Capacitor Placement
7
www.fairchildsemi.com
AN-817
DIFFERENTIAL LINE LAYOUT
AN-817
DIFFERENTIAL LINE LAYOUT
(Continued)
BEST
ACCEPTABLE
FIGURE 6. SOIC Bypass Capacitor Placement
CORRECT, INCORRECT
FIGURE 7. Power Pin Connection
INCORRECT
CORRECT
FIGURE 8. Differential Line Layout
www.fairchildsemi.com
8
TESTING MINIMUM-SKEW CLOCK SYSTEMS
There are additional steps that can be taken to avoid skew
build-up. Perhaps the simplest is to observe the rule of
equal loading and termination. Briefly stated, all matching
delay paths must be loaded equally, both resistively and
reactively. Complementary outputs must have equal resistive termination values. Capacitive loading must be equal
for both outputs, also. This is especially true of differential
lines.
As with all types of electronic systems, the performance of
minimum-skew circuits must be verifiable and testable. The
very nature of these systems calls for more care and precision in testing than other types of logic systems. And,
greater precision is required from the test equipment and
its use.
Certainly the oscilloscope is the most frequently used piece
of test equipment for logic systems. While it does provide
much information about the quality of signals, it may be the
cause of error if improperly applied. Perhaps the single
largest error source and the least recognized is the place
where scope and circuit meet, the probe. To do justice to
the subject of probing and probes in pulse application
would require a book unto itself. One very useful book from
Tektronix is “ABC's of Probes”. In it are emphasized some
essentials for choosing a suitable probe for the job. Some
of these are:
For example and using the delay value for 50Ω lines mentioned previously, a loading difference of 5 pF between the
conductors of a differential line can result in a skew of
105 ps, or about one-fifth of a risetime. When combined
with other effects like crosstalk and impedance discontinuities, significant skew can occur.
Path delay distribution is another step that can be taken to
avoid skew build-up, Figure 9. This means that, if a clock
distribution path must pass through several amplifiers,
steps should be taken to assure that the signal does not
use the same path, pin-wise, through all amplifiers. To illustrate, assume that the path passes through three F100311,
1:9 differential drivers. The input pins in all cases are 28
(in) and 2 (in/). If the first output pair is 25 (D0) and 24 (D0/
), then choose pins 23 (D1) and 22 (D1/) for amplifier #2
and pins 20 (D2) and 19 (D2/) for the third amplifier. Additional variation can be achieved by reversing the input
phase through each amplifier. Of course, the output phase
must be similarly reversed to match the input. These points
may seem minor, but all contribute to reducing the mathematical probability of skew tolerance build-up.
• Compatibility with the scope system,
• Adequate bandwith and risetime capability exceeding
that of the system to be tested,
• Low circuit loading, which implies low capacitance at the
tip,
• High enough impedance so as to not unduly load the circuit being probed (This does not mean that FET probes
are the best choice!),
• And delay matching when multiple probes are used for
relative delay measurements.
FIGURE 9. Avoiding Skew Tolerance Buildup
signals taken at different times and places in the circuit is
important.
The choice of scope is equally important and the list of its
basic features is similar to that for the probe. Where the
precise nature of the signal to be measured is unknown,
the analog scope may provide the most useful information
initially. While it does not have all of the sophistication of
digital scopes, it offers fewer complications and a fidelity
that is hard to beat. The digital scope, especially if it has
storage capability, can be an asset where comparison of
Additional useful information on scopes and in-circuit measurements may be found in the series “Troubleshooting
Analog Circuits” by Fairchild Semiconductor's Bob Pease
in “EDN” and available as a reprint from Fairchild.
A final consideration and one that is often overlooked is the
incorporation of test points in the circuit. The increasing
9
www.fairchildsemi.com
AN-817
AVOIDING SKEW TOLERANCE BUILD-UP
AN-817 Taking Advantage of ECL Minimum-Skew Clock Drivers
980045-001 — Reliability Report Update — Small Outline
(SO) Package
use of surface mount devices and dense layouts makes
probing these circuits more difficult. Inclusion of test points
or test connectors can ease the job considerably. Just
remember to make sure that the test points do not corrupt
the signals and are themselves designed for minimumskew. Test points may even be provided by reserving one
output of an ECL minimum-skew device at each critical
point in the circuit. This provides isolation for the probe and
minimal loading on the monitored circuit. And, of course it
should go without saying that test points should be accessible without the need to mechanically or electrically alter
normal operation of the equipment being tested.
980055-004 — Surface Mount Availability Guide
F100K ECL 100 Series to 300 Series Conversion Book
Troubleshooting Analog Circuits (EDN Article
Series Reprint)
Databooks
400080 —
Reliability Handbook
400028 —
F100K ECL Logic Databook and Design
Guide
APPENDIX
BIBLIOGRAPHY
Available Literature
“F100K ECL Logic Databook and Design Guide”, Fairchild
Semiconductor Corp., 1991
Application Notes
AN-467 —
Surface Mount: From Design to Delivery
AN-468 —
Wave Soldering of Surface Mount Components
AN-469 —
Reliability of Small Outline Surface Mount
Packages
AN-470 —
Wave Soldering: Is SOIC Reliability Compromised?
“Radiotron Designer's Handbook”; Fourth Ed.; F. LangfordSmith; Amalgamated Wireless Valve Co. Pty. Ltd.; Sydney,
Australia; 1952
“The RF Capacitor Handbook”; 1st Ed.; American Technical Ceramics; 1979
“Traveling-Wave Engineering”; R.K. Moore; McGraw-Hill;
1960
“Communications Systems: An Introduction to Signals and
Noise in Electrical Communications”; A. Bruce Carlson;
McGraw–Hill; 1968
Datasheets
113295-001 — Plastic Chip Carrier Technology
“Analysis, Transmission and Filtering of Signals”; M. Javid
and E. Brenner; MrGraw-Hill; 1963
113615-001 — A Basic Guide to Surface Mounting of Electronic Components
“ABC's of Probes”; Tektronix, Inc.; 1989
Electronic Circuit Calculations Simplifed, Part 4 - RC combinations in D.C. circuits”; S.W. Amos; “Wireless World”; 9/
1975; pp. 423–426
Other Literature
570430-001 — Reliability Report Update Small Outline
(SO) Package
“Electronic Circuit Calculations Simplified, Part 5 - RC combinations in A.C. circuits”, S.W. Amos; “Wireless World”;
10/1975; pp. 475–478
570435-001 — Getting Started in Surface Mount
980040-001 — Reliability Report Update — Plastic Chip
Carrier (PCC) Package
Fairchild does not assume any responsibility for use of any circuitry described, no circuit patent licenses are implied and
Fairchild reserves the right at any time without notice to change said circuitry and specifications.
LIFE SUPPORT POLICY
FAIRCHILD’S PRODUCTS ARE NOT AUTHORIZED FOR USE AS CRITICAL COMPONENTS IN LIFE SUPPORT
DEVICES OR SYSTEMS WITHOUT THE EXPRESS WRITTEN APPROVAL OF THE PRESIDENT OF FAIRCHILD
SEMICONDUCTOR CORPORATION. As used herein:
2. A critical component in any component of a life support
device or system whose failure to perform can be reasonably expected to cause the failure of the life support
device or system, or to affect its safety or effectiveness.
1. Life support devices or systems are devices or systems
which, (a) are intended for surgical implant into the
body, or (b) support or sustain life, and (c) whose failure
to perform when properly used in accordance with
instructions for use provided in the labeling, can be reasonably expected to result in a significant injury to the
user.
www.fairchildsemi.com
www.fairchildsemi.com
10