Professional Documents
Culture Documents
SI Guide
A basic guide for SI and crosstalk for a PCB designer
Preface
What the heck is SI?
Why and what should I Simulate?
What is the strategy for SI Analysis?
Is crosstalk happens when you talk in the mobile?
What do you mean by transmission line?
Topology??? Are you kidding?
Delay?? How about a schedule change?
This is a collection of articles, which are found in web and other sources trying to answer your
questions. Hope it will be helpful
Table of Content
Basics of SI___________________________________________________________________5
1.1 When Speed is important? _____________________________________________5
1.1.1 Acceptable Voltage and timing values ________________________________5
1.2 Signal Integrity ______________________________________________________5
1.2.1 Waveform Voltage Accuracy _______________________________________5
1.2.2 Timing_________________________________________________________5
1.3 Speed of currently used logic families ____________________________________5
1.3.1 Transition Electrical Length (TEL) __________________________________6
1.3.2 Critical length ___________________________________________________6
1.3.3 What is Transmission Line? ________________________________________6
1.3.4 What is moving in a Transmission line?_______________________________6
1.3.5 Power Plane Definition____________________________________________6
1.3.6 The concept of Ground ____________________________________________7
1.4 STRIPLINE circuit with Electromagnetic field _____________________________7
1.5 RLC Transmission Line Model _________________________________________8
1.5.1 What is Impedance? ______________________________________________8
1.5.2 A Practical impedance equation for microstrip _________________________8
1.5.3 What is relative dielectric constant Er? _______________________________9
2 Interconnections for High Speed Digital Circuits _______________________________10
2.1.1 Summary______________________________________________________10
2.2 Examples of dynamic interfacing problems _______________________________10
2.3 IC Technology and Signal Integrity _____________________________________12
2.4 Speed and distance __________________________________________________14
2.5 Digital signals: Static interfacing _______________________________________15
2.6 Digital signals: Dynamic interfacing ____________________________________16
2.7 Review questions ___________________________________________________18
3 Interconnection Models____________________________________________________20
3.1 Summary__________________________________________________________20
3.2 Reference model for interconnection analysis _____________________________20
3.3 Receiver model _____________________________________________________21
3.4 RC interconnection model ____________________________________________23
3.5 Parameters of the interconnection ______________________________________25
3.6 Refined models _____________________________________________________26
3.7 Review question ____________________________________________________28
4 Transmission Line Models _________________________________________________31
4.1 Summary__________________________________________________________31
4.2 Transmission line models _____________________________________________31
4.3 Loss- less transmission lines ___________________________________________32
4.4 Critical Length _____________________________________________________34
4.5 Reference transmission line model______________________________________35
4.6 Line driving _______________________________________________________36
4.7 Propagation and reflected waves _______________________________________37
4.8 A sample system____________________________________________________39
4.9 Review questions ___________________________________________________42
Speed is important when; the edge rate (rise or fall time) of a clock is fast enough that the signal can
change from one logic state to the other in the same or less time than it takes the signal to travel the
length of the wire or net.
1.2.2 Timing
Is affected by propagation delay variations in ICs, travel time on wires, and variation in edge rates of
ICs.
Overshoot and undershoot problems will start to occur at less than 1/3 of the Transition Electrical
Length (TEL) and may require series termination of transmission lines, as well as impedance control.
EM waves do not depend on electrons for transmission, all move through vacuum at the speed of
light.
• Chassis ground
•
• Logic Ground – this is the point where all measurements are made, no need to connect to
chassis
• Analog ground – this is commonly used to represent a point in a system that analog circuits
use to measure or compare analog signals. It can and must be connected to LOGIC
GROUND at some point
Magnetic lines of force (solid lines) encircle the signal conductor; electric lines of force (dotted lines)
connect the signal conductor to the reference plane
This model assumes ground is plane of negligible inductance and resistance. The following
equations permit one to calculate the reactance of capacitors and inductors as a function of
frequency.
The equation for calculating the amount of capacitive reactance in an ac circuit is given by:
XC = 1 / (2πfC)
where:
The equation for calculating the amount of inductive reactance in an ac circuit is given
by:
XL = 2πfL
Where:
An equation for calculating Er using velocity measure with a TDR (time domain reflection)
c
er = Where c = speed of light; v = measure propagation velocity
v
1) Spikes and temporary faults appear at the output of a combinatorial circuit. The
reason is that the change of logic state is sensed with different delays by a logic
circuit connected to the same signal. This may cause transient combinations of logic
states, which were not taken into account in the design process.
2) Figure 1.1 shows two microprocessors driven by the same clock signal. That means they should
run synchronously, without problems related with marginal timing, such as metastability. In the real
circuit, for some combinations of temperature and supply voltage, the system exhibits random errors
for microprocessor 2. Swapping the devices has no effect; therefore the problem is related with the
socket, not with the device.
The random errors are caused by synchronization problems (metastability) in the information
exchange between the two microprocessors. The solution is a redesign of the clock distribution
circuit as in Figure 1.2, taking into account transmission line effects. This can guarantee the timing
margins for the synchronous circuits of the two microprocessors.
3) A full-rack length (49 cm) backplane is equipped with boards using 74F245 transceivers towards
the bus, as in Figure 1.3. Backplane signals on the scope exhibit high signal distortion (Figure 1.4).
Moreover, the bus interfaces have rather high power consumption.
Understanding the problem enables the designer to select the proper logic family for the drivers,
thus achieving better waveforms, and reducing the power consumption.
Microelectronic technology allows us to put several million devices (transistors) on a single die, but
similar advances did not occur in interconnection capability. Namely, the number of devices in a
single IC scales with the square of the (chip size)/(device size) ratio (S/l), while standard
interconnection, being placed on the die boundary, scales linearly towards the same parameter. The
two parameters are plotted in Figure 1.5. Therefore, in the design of complex high-performance
systems, we have to face an interconnection bottleneck.
This communication bottleneck is addressed by increasing the speed (bit rate) on external
interconnections, and by increasing the number of pins. But higher speed means higher dv/dt on
I/O pins. This requires high currents to charge/discharge parasitic capacitors, and the effect is
multiplied by the increase in the number of I/O pins. Higher currents mean more noise towards
other circuits and systems, and careful design of interfaces is required to sustain the nominal speed
of digital circuits in real systems.
Surface mount technology (Figure 1.6) and new packaging techniques which reduce the size of
devices and systems, and reduce parasitics by moving interconnection points from the boundary to
any die position (eg BGA and flip-chip) help to deal with these problems. Also Low Voltage (LV)
logic families, which reduce the DV between logic states help to limit EMI (and power
consumption), at the expense of reducing noise margin and therefore increasing their susceptibility
(sensitivity to external interference).
Figure 1.6 Surface mount packages (divisions on the top are 1 mm. each)
What must be actually taken into account is not the clock period, but the edge slope. The change
from one logic state to the other requires a finite time (roughly corresponding to the rise time or the
fall time of the waveform). In first approximation, the spectrum of a digital signal covers from DC
to a frequency F = 1/2tr. Current high-speed circuits, with rise/fall time below 1 ns, have significant
components at frequencies above 500 MHz. The frequency content depends on edge steepness, not
on repetition rate.
If the signal has to travel over some distance, devices tied to the same node (that is connected by a
wire, or a PCB track) may sense different logic states. This effect is completely hidden in the
conventional "logic diagrams", which assume that all logic circuits tied to a node sense the same
state at the same time.
The rise/fall time of current logic devices is about 1 ns; that means only wires shorter that 20 cm
can be considered a single "node".
This data shows how for most modern high-speed logic families the interconnection delay and the
transmission line effects interact with pure logic behavior of gates, and can cause false signaling.
These effects put upper bounds on the speed of logic systems; even if the pure logic could run at
infinite clock rate, the wiring introduces limits, as discussed in the following lessons.
The mapping between logic states and analogue voltages is defined by the static electrical
parameters.
Logic Output Electrical Input Electrical
State Parameters Parameters
LOW VOL , IO L VIL , I IL
HIGH VOH , IO H VIH , I IH
(the total current at driver output must be lower that the IOH or IOL specified by the
manufacturer).
This condition puts an upper limit on the number of inputs that can be connected to the same
output. Even if most current digital circuits use CMOS devices, with input currents practically 0 (and
therefore no limit in a purely static situation), CMOS inputs are always dynamic loads (capacitance),
and increasing their number leads to speed reduction, as discussed in the following lessons.
Propagation delays, tPLH and tPHL: respectively for LOW to HIGH and HIGH to LOW state
changes. Figure 1.10a shows the input-output relationship for an inverter. The actual signals on a
scope are in Figure 1.10b.
Correct operation of flip-flops and registers require compliance with minimum set-up time (tSU)
and hold time (tH). When these limits are fulfilled, the output state changes with a propagation delay
tP after the clock active edge.
2) Assuming a signal propagation speed of 0,6 c (c is the free-space propagation speed of the light),
the "electrical length" of one clock period at 50 MHz is · 36 mm, · 1 m, · -* 3.6 m, · 6 m. 0.6 c = 0.6
x 300 000 km/s = 180 000 km/s For a 50 MHz clock the period is 20 ns; the distance traveled in 20
ns is 180x106 m/s x 20x10-9 s = 3.6 m.
3) The "propagation time" (tPHL) is defined as: · the time required by the voltage representing a
logic state to change from VOL to VOH · *the delay between input and output state change for a
High-to-Low transition · the time required by the voltage representing a logic state to change from
VOL to 50% of VOH · the time required by the voltage representing a logic state to slew from 10%
to 90% of VOH - VOL . Propagation times represent the delay between input and output of a logic
device. The index (HL or LH) indicates the direction of the state change.
4) The "rise time" (tR ) is defined as: · the time required by the voltage representing a logic state to
change from VOL to VOH, · the time required by the voltage representing a logic state to change
from VOL to 50% of VOH, · *the time required by the voltage representing a logic state to slew
from 10% to 90% of VOH - VOL, · the input-to-output delay of the receiver. This is a definition -
just learn it !
5) To explain the "set-up time" (tSU) you could say: · *the input D must be stable for at least tSU
before the clock active edge; · the input D must be stable for at least tSU after the clock active edge;
· the output Q will change with a delay of at least tSU after the active clock edge; · the minimum
duration of the clock pulse is tSU. · the time required by the clock to slew from 10% to 90% of
VOH - VOL is at least tSU . Flip-flops require time margins around the active clock edge to avoid
6) The "disable" (tDIS) is defined as: · the time required by a three-state driver to exit from the high-
impedance state, · *the time required by a three-state driver to enter the high-impedance state, · the
input-output delay in a three-state driver to exit from the high-impedance state, · the time required
by the voltage at the output of a three-state driver to slew from 10% to 90% of VOH - VOL . Due
to internal parasitic and delays, a three-state driver requires some time to switch between active and
non-active states. These are the enable and disable times.
7) With microelectronic technology evolution, the interconnection capability (that is the product
between and ) · increases faster than local processing power, · *increases slower than local
processing power, · increases at the same rate as local processing power, · decreases. Processing
power depends on the number of active devices (transistors), which increases with the square of the
(feature size)/(chip size) ratio. Interconnection capability (for standard packages) depends on how
many active devices can be placed on the sides of the chip, which increases linearly with (feature
size)/(chip size).
We shall start the analysis from simple linear models for driver/receivers
and interconnections; this allows us to define the parameters which
describe the behavior of systems made by several interconnected logic
devices.
Before entering the driver and after the output of the receiver we have logic states (0, 1). Between
the driver output and the receiver input we describe the system in terms of electrical quantities such
as voltage and current, or devices such as resistors, capacitors, etc. For the driver-interconnection-
receiver systems we can define several models, with increasing accuracy.
The red and blue transfer characteristics in Figure 2.4 represent inverting buffers with different
thresholds (both within specifications, that is in the range VIL- VIH). When the input voltage VIN is in
the VIL- VIH range, it is not possible to state if it is above or below VTH for all logic circuits of that
family.
In this example when the input voltage V IN is within the VIL - VIH range, the "red" and "blue"
inverters A and B (with different threshold voltages) map the same input signal to different output
states. Static interfacing conditions are violated, and a Data Split may occur.
Data split happens because of faulty or poor design, or because external noise modifies the voltage
level on the interconnection.
On data split the digital circuits tied to the same node sense different logic states; the system is
forced to a state not foreseen in the design phase (logic synthesis). In turn, it causes output errors or,
for sequential circuits, unpredictable evolution or deadlocks.
The indetermination of VTH is one of the reasons for the presence of skew (time indetermination)
on interconnections. From this analysis we found that something wrong may happen when the input
voltage VIN lies in the VIL - VIH interval. A first rule for digital interface design can be derived from
this consideration:
The input voltage V IN should lie in the V IL - VIH interval for the shortest possible time. This goal can
be achieved by acting on two parameters:
A first approximation model for the interconnection is shown in Figure 2.7; it is modeled as a single
node; the capacitor CI corresponds to the parasitic input capacitance of the wire and of the receiver
(which is actually a good approximation for CMOS inputs). The circuit is a first order low-pass cell
and the step response at the receiver input is an exponential.
t=0 VC = 0, VB = 0
t --> VC = V B = V A
VB = V C = V A (1 - e ) with =RC
tT X. = ln (V A / (V A - VTH ))
The circuit is still a first order low-pass cell, but the threshold V TH is now
crossed at different times in nodes B and C (Figure 2.12). This time
difference is another type of skew: even assuming that all parameters are
fixed and known, the transmission time depends on the position of the
receiver (node B or C).
To further increase the accuracy from these basic RC models we can add
new circuit elements that represent a more detailed interconnection
structure (Figure 2.13):
The logic state changes become voltage steps at the output of logic
circuits. They are sensed by other logic inputs when the voltage crosses
the logic threshold. Due to output resistance of drivers and parasitic
capacitance of interconnections and receivers, the vo ltage step becomes
an exponential, and reaches the logic threshold after a delay called
transmission time.
• *first at the driver output (near end), the n at receiver input (far end).
• first at the receiver input, then at the driver output.
• at the same time at the driver and at the receiver.
The voltage at the near end (driver side) rises faster than the voltage at
the far end (receiver).
• the delay from the state change at driver input and voltage change
at driver output.
Receivers sense logic state changes after the transmission time, that is
with a delay related with parasitic parameters, value of the threshold,
position on the interconnection. The total variation of the transmission time
is called skew.
The transmission time and the skew depend on the electrical parameters
of the drive r and of the receiver. The logic operation is performed inside
the device, and does not affect the parameters of the interconnection.
• *tT X = RC ln (V A / (V A - VTH)),
• tT X = RC,
• tT X = RC (V A / (V A - VTH)),
• tT X = RC ln V TH.
VB = V TH = V A (1 - e ) with =RC
VTH = V A (1 - e )
VTH - VA = - VA e
- ln((VTH - VA )/ V A ) = tT X / RC
Each cell models a piece of wire (or PCB track); as the length of this piece
diminishes and the number of cells increases, the model approximation
improves. For an infinite number of very small cells, the sequence turns
into a distributed parameter model or transmission line (Figure 3.2).
They depend on inductance and capacitance per unit length (1 meter) L'
and C'.
Wider tracks have higher capacitance and lower inductance, that is lower
characteristic impedance Z0 and lower propagation speed U. The inverse
occurs with narrow tracks or when the gap between track and ground is
decreased (that is when the board thickness is reduced).
As the unit inductance L' increases (more narrow track) the characteristic
impedance Z0 increases and the propagation speed U decreases.
When many devices are connected to a track, the unit capacitance C' is
increased by the device parasitic (input and output capacitance). Also in
this case the characteristic impedance Z0 and the propagation speed U
decrease.
If the propagation time over an interconnection is far less than rise and fall
time of signals, the conductor can be considered equipotential; in this case
For each logic family a critical length (related to transition times) can be
defined: connections longer than this critical length must be modeled as
transmission lines.
The table shows the values of critical connection length for various logic
families.
TTL 60 cm
ALS 25 cm
AS 6,5 cm
HC 45 cm
ABT 15 cm
The open-circuit voltage of the line driver is a voltage step with amplitude
VA. RO is the internal output resistance of the driver.
VB is the voltage at node B (driver output, left side of the transmission line,
or near end)
VC is the voltage at node C, corresponding to the right side of the line or
far end. The termination resistance is connected to this node. The line
termination can be a specific component or the input of a receiver.
When the logic state at driver input changes (eg 0 to 1), a voltage step
(V OL to V OH, simplified as 0 to V A) appears at the driver output. The driver
is loaded by the transmission line - that is by a dynamic impedance Z0.
The actual step at node B is given by the partition of V A across RO and Z0
(Figure 3.9)
The two ends of the line usually represent the most significant
discontinuity; a typical situation for instance could be:
v = v(t,x); i = i(t,x);
The total voltage and current at time t in the position x are the sum of
incident and reflected terms:
iT = i1 + i2; vT = v1 + v2; vT /i T = RT
At t = 0 the output of the driver sees the line as impedance Z0: the voltage
at node B is the partition of V A among R0 and Z0 (Figure 3.11). This is the
first step (incident wave) with amplitude:
This signal corresponds to the first incident wave V', which moves forward
on the line at speed U.
At t = tp the step reaches the end of the line (node C). The signal at the far
end C comes from a generator V L with equivalent impedance Z0, the
transmission line. Therefore for the incident wave v/i = Z0.
At t = 2 tp the reflected wave V" comes back to the driver output (node B).
Since the driver output resistance RO is different from Z0, a new reflected
wave V"' is generated.
diagram with diagram, and travels left-to-right (from the near end to the
far end).
The voltage in any position is the sum of a ll incident and reflected waves.
The waveforms in B and C are a sequence of steps with amplitude related
with RO , Z0 , RT . The steps occur at t = 2 K tP at the near end (B), and at
(2k + 1) tP at the far end (C).
Figure 3.14Two examples of waveforms at near end (VB) and far end
(VC).
Figure 3.15 Near end (top trace) and far end waveforms.
• 2c
• *0.7 c
• c
• 1c
• *increases
• decreases
• does not cha nge
The track and the ground plane make a capacitor; when track width
decreases, the size of the capacitor decreases. The characteristic
• increases
• *decreases
• does not change
• increases
• *decreases
• does not change
The track and the ground plane make a capacitor; when the separation
between plates decreases, the value of the capacitor increases. The
• *increases
• decreases
• does not change
The track and the ground plane make a capacitor; when the separation
between plates increases, the value of the capacitor decreases. The
The rise/fall time in the HC logic family is a few ns. The distance travelled
by the electric wave in this time is 40-60 cm, higher than the
interconnection length.
• 1V
• 2 V,
• *2.5 V
• 5V
• =1
• = -1
• * = 0.6
• = -0.6
Transmission time and skew keep the definition given in lesson 2. Instead
of following an exponential, now the voltage on the line changes as a
sequence of steps, corresponding to the sequence of reflected waves
(Figure 4.1).
Fig 4.1 Waveforms at the near end (top trace) and far end of a
transmission line driven by digital signals.
In the following sections we shall analyse some simple specific cases for
the termination resistance, evaluating the transmission time and the skew.
For the first reflected wave, we can look at signals for three specific cases.
1. RT = Z0 = 0;
The line is matched: the v/i ratio is equal to Z0 both on the line and at
termination; G = 0 and there is no reflected wave. In figure 4.3 the top
trace represents the voltage at the near end (driver), and the bottom trace
is voltage at termination. Since there is no reflection, we observe a single
step, respectively at t = 0 (, node B, driver, top trace) and t = tP (node C,
termination, bottom trace).
Fig 4.3 Waveforms at the near end (top trace) and far end of a
transmission line with matched termination.
2. RT = oo = 1;
Fig 4.4 Waveforms at the near end (top trace) and far end of an open
transmission line.
The lattice diagram represents on the horizontal axis (x) the segment of
line under analysis, and on the vertical axis the time (t), increasing
towards the bottom. The signal on the line is represented by a point, which
moves in the (x,t) plane on straight lines with a slope depending on the
propagation speed U = x/t. When the signal hits a discontinuity (such as
line ends in this example), a reflected wave with an amplitude depending
on the corresponding is sent backwards.
The diagram represents the signal trajectory in the (x,t) plane, that is the (x,t) pairs that satisfy the
wave equation. The amplitude of each reflected wave is given by the incident wave multiplied by the
reflection coefficient. The total signal amplitude can be obtained by summing the amplitude of the
reflected waves sequentially generated at each end. The v(P,t) waveforms corresponds to a vertical
section of the diagram in the corresponding position (point P in this example). The various
contributions sum as they arrive at that point (that is when the lattice crosses the vertical line).
% usage of variables:
% t time % x position
% y amplitude
clear
F = 0; T = 0; t =0;
G = 1; % step amplitude
% first propagation
for t = 1:tp+1
for x = 1:L
U(t,x) = F + G*As*((x/L)<(t/tp));
end
end
G = U(t,x) - F;
F = U(t,x);
bounce = 1;
% reflections
for t = (bounce*tp):((bounce+1)*tp+1)
for x = L:-1:1
U(t,x) = F + G*Gt*(((L-x)/L)<((t-bounce*tp)/tp));
end
end
G = U(t,x) - F;
F = U(t,x);
bounce = bounce+1;
for t = (bounce*tp):((bounce+1)*tp+1)
for x = 1:L
U(t,x) = F + G*Gs*((x/L)<((t-bounce*tp)/tp));
end
end
end
T = t;
% drawing results
x=1:L;
t=1:T;
[X,Y]=meshgrid(x,t);
mesh (X,Y,U)
title (['View from termination: Gt = ',num2str(Gt), ' Gs
=',num2str(Gs)])
xlabel ('distance'), ylabel ('time'), zlabel ('amplitude')
view ([40, 30]) % view from termination
%print -dwinc
figure
mesh (X,Y,U)
title (['View from source: Gt = ',num2str(Gt), ' Gs =',num2str(Gs)])
xlabel ('distance'), ylabel ('time'), zlabel ('amplitude')
view ([-50, 30]) % view from source
%print -dwinc
Two examples of results from the simulator are give in figure 4.11 and
4.12.
Figure 4.11 shows the time evolution of the voltage on the line at different
positions for an interconnection with positive reflection coefficients at
source and at termination. All steps are positive, and the voltage at any
point of the line is a monotonic staircase.
Fig 4.12 shows the time evolution of the voltage on the line at different
positions for an interconnection with positive reflection coefficient at
termination and negative reflection coefficient at source. This situation
corresponds to a high current driver connected to an open line
(termination resistance much higher than characteristic impedance). Now
• tT Xmax = 0 ,
• tT Xmax = tP ,
• tT Xmax = 2tP ,
• *the syste m does not work.
With matched source the first step is V A / 2; due to matching at the far end
(termination) this is also the final voltage on the line. The voltage on the
line never crosses the threshold.
2) When VTH < VA / 2, the maximum transmission time for a line matched
at source and at termination is:
• tT Xmax = 0 ,
• *tT Xmax = tP ,
• tT Xmax = 2tP ,
With matching at source and at termination the first edge (incident wave)
reaches the far end in = tP, and there is no further reflection. All points on
the line are reached within this time.
The previous lesson presented how transmission line effects influence the waveform of
digital signals. Here we shall review how the designer can manage these effects, in order to
guarantee the correctness of information transfer.
Incident wave switching (IWS) occurs when the receiver can sense the logic state change on
the first step impressed to the line. To guarantee this condition, the first step must be higher
than the threshold VTH, and this means Z 0 >> RO. This can be obtained by lowering the
output resistance RO (that is using high-current drivers), and by keeping the characteristic
impedance Z 0 as high as possible.
Example 1
The designer can derive either the specification for the driver equivalent output
resistance RO, or the minimum value for driver output current IOH.
The amplitude of the first incident wave (first step) can be evaluated from the open circuit output
voltage, partitioned among the drive output resistance and the line impedance:
An alternate specification is the output current IOH; the voltage VB can be written as:
VB = Z0* IOH
To achieve IWS V B must be higher than the threshold voltage V TH, that is:
IOH = V B / Z0
IOH >= 35,7 mA
When new receivers or transceivers are connected to the line (eg in backplane buses,
as in figure 5.2), their input parasitic capacitance adds to the intrinsic line
capacitance.
to
Example 2
The minimum value for driver output current IOH to achieve IWS (other
parameters as in previous example) is now:
VB = Z0" * IOH
IOH = V B / Z0 "
IOH >= 178 mA
The current flowing in the termination resistance RT must come from (or
go into) the driver. In the previous examples the termination resistance RT
was always connected to ground. To drive a line with termination
connected to ground at values higher than V TH, the driver must have high
IOH. No current is required in the Low State.
For matched lines (RT = Z0) the current may rise to rather high values. For
instance, with a termination RT = 70 and V CC = 5 V the output current IOL
= = 70 mA. This current flows into the driver, and adds to receiver
input currents.
Due to asymmetry in the output stage, IOL is usually higher than IOH , and
therefore termination resistances are usually connected towards V OH
rather than to ground. In the case of Open Collector drivers, the
termination resistance acts also as pull-up resistance.
Several circuits can be used for terminations; some of them reduce the
output current required from the drivers:
This section describes the signals in transmission lines driven from one
end. The first case discussed is a line open at the far end ( T = 1) and
matched at the driver (RO = Z0).
The system is now in its final state; there is no further reflection because
the driver side of the line is matched. The waveforms are in figure 5.8
(where 1 tP corresponds to 2.5 horizontal divisions).
We will now analyse the same circuit as in the previous example (line
open at the far end: T = 1, with the driver no longer matched (RO < Z0,
Since G < 0, the signal reflected at the driver has opposite polarity with
respect to the original step. The steps on the line can be either positive or
negative, with sign inversion every 2 tP, (the time distance between
reflections at driver).
At the far end C, where incident and reflected wave sum with the same
polarity, the actual waveform can oscillate. These oscillations may bring
the voltage to cross several times the logic threshold V TH, thus causing
additional logic state changes.
Fig 5.9 Near end (top trace) and far end waveforms for an open line with
high-current, low-resistance driver.
Fig 5.11 Open line, low resistance driver with series termination.
The board must use a single transceiver towards the bus (figure 5.13, left).
The transceiver isolates all on-board circuits and reduces the capacitive
loading on bus lines. Direct connection of on-board devices to bus lines
(figure 5.13, right) must be avoided, since it increases the capacitive load
and lowers the characteristic impedance.
High-speed logic systems must use multiple layer Printed Circuit Boards
(PCBs), with alternated signal layer and ground/supply layers, to provide a
controlled environment for signal propagation. Example of signal track
layers and ground plane are in figure 5.15.
• RS = Z0 + RO
• RS = RO - RT
• RS = Z0 - RT
• *RS = Z0 - RO
The sum of driver output resistance and external series termination must
equal the line characteristic impedance.
Since three branches (two transmission lines and the driver output)
converge at the driving node, it is not possible to match the line at the
driver side. The only possibility is to avoid reflections at the far ends, by
matched terminations.
3) In a mid-point driven 100 ohm bused line, the amplitude of the first step,
for VA = 5 V, RO = 50 , is:
• 1 V,
• 1.25 V,
• *2.5 V,
• 5 V.
VB = 5V * 50 / (50 + 50) :
• 200 ,
• 100 ,
• 10 ,
• *35 .
1) In a bundle of wires only some of them are active (that is carry logic
transitions), but signals appear also on non-active conductors. This is a
crosstalk problem, and can be handled by proper selection of driving
devices and wiring layout.
2) A pulse is applied to one input of a buffer; the other inputs of the same
package are fixed at High or Low levels. The voltage at the outputs
corresponding to steady inputs change state when other devices in the
same package switch. This can be caused by ground bounce noise - a
common-path crosstalk problem - and can be avoided by proper layout
and decoupling of the power supply.
For this analysis the wires are modeled as two transmission lines with
characteristic impedance Z0 and both capacitive (C M ) and inductive (LM )
coupling. To avoid multiple reflections, both lines have matched
terminations RT = Z0. (figure 1.4).
The driven line (or active line, always drawn in the upper part of the
diagram) carries a signal moving at speed U. This signal is a voltage step
with rise time tr and slope dv/dt = DV/tRr. (Figure 1.5).
The effects of capacitive coupling can be evaluated with the model shown
in figure 1.6. The equivalent circuit includes a voltage generator V M , the
coupling capacitance C M , and the impedance Z0/2, which models the
resting (or passive) line.
If the voltage change VSC on the passive line is small, the whole step VS
appears through the capacitor CM. A current ICM = C M dv/dt flows from the
capacitor to the load Z0/2. This rectangular current pulse generates a
voltage pulse with the same duration tr. and amplitude V SC = ICM Z0/2. In
summary:
This pulse propagates on the passive line towards the two ends of the
passive line, thus making the Backward Capacitive (B C) and Forward
Capacitive (F C) terms of crosstalk (figure 1.7).
The total crosstalk moving towards the right end (far end, termination), in
the same direction of the disturbing signal is the forward crosstalk FT = FC
+ FL . Forward crosstalk is the sum of terms 1 and 3 above; since they
have the same polarity, if dv/dt > 0 (for the disturbing signal), the forward
crosstalk is always positive.
Backward terms sum in the time domain and make a pulse with fixed
amplitude and width related to position along the line.
When the disturbing edge reaches the far end at t = tP, it is absorbed by
the matched termination, and the corresponding noise generators turn off.
The forward term ends immediately; the various points along the line
7.5 Examples
What is Crosstalk?
Crosstalk is the interaction between signals on two different electrical nets. The one creating
Crosstalk is called an aggressor, and the one receiving it is called a victim. Often, a net is both an
aggressor and a victim.
An electrical current in a loop generates a magnetic field. If this magnetic field is changing, it can
either radiate energy by launching radio frequency waves, or it can couple to adjacent loops
("Inductive cross-talk").
When does a line or loop radiate radio frequency waves? Basically, if the line or loop
has signal energy at a frequency high enough that the line or loop represents at least a
tenth of a wavelength, there will be a measurable amount of electromagnetic
radiation ("Radio waves").
The first diagram (Fig 1.12) shows the crosstalk voltages at t = tP/3, for a
given combination of dV/dt (rise time tr of the disturbing edge), C M , and L M
(the values of these parameters shown in the box at the bottom right
corner are only for comparison of different situations).
Figure 1.13 shows the same waveforms for the same system at a later
time t = 2 tP/3 . The forward term amplitude increases (due to summation
In the third diagram (Fig. 1.14), the ratio among capacitive and inductive
coupling is inverted: the capacitive term overcomes the inductive one, and
the polarity of the resulting forward term is reversed.
The effects of edge slope are shown in figure 1.15. All parameters are the
same in both cases, the only difference being the slope of the disturbing
signal rising edge. Since crosstalk amplitude is proportional to the slew
rate, the steep edge causes a higher noise. The disturbance lasts for the
duration of the edge, and is close to 0 when the active line does not
change state.
The actual crosstalk signals in a real system are the sum of all
contributions (inductive, capacitive, forward and backward terms). Due to
different propagation directions, the waveforms depend on the observation
point: a scope connected to different points shows different wave shapes.
Summarising the previous analysis:
At the near end all terms start for t = 0; the backward term ends when the
disturbing pulse reaches the far end, and the induced noise generator is
turned of (at t = tP). This information (or the negative edge caused by
turning off the disturbing pulse at the far end) needs another tP to reach
the near end, therefore the total duration of crosstalk noise at the near end
is 2 tP (track A in figure 1.16).
The disturbing pulse (and induced noise) reaches the far end at t = 2 tP
(nothing can be seen before). The forward term (narrow pulse with width
tr) is the sum of all contributions along the line; the backward term ends
almost immediately, as the disturbing pulse reaches the far end at the
same time at t = 2 tP. Therefore the far-end crosstalk is a narrow pulse,
occurring at t = 2 tP , with sign and amplitude related with the actual L M
and C M values (track C in figure 1.15). At the intermediate position (track B
in figure 1.16) we can observe the forward term (narrow pulse, variable
height), followed by the backward term (rather flat pulse, width depending
on the position).
The backward term can be seen as the sum of all contributions along the
line, originated when the disturbing pulse travels towards the far end.
When the pulse reaches the far end (for a matched line), it is absorbed by
the termination. The backward pulse duration is therefore twice the
propagation time, which in turn depends on line length.
The forward term can be seen as the sum of all contributions along the
line, originated when the disturbing pulse travels towards the far end.
Since these elementary contributions travel towards the far end at the
same speed as the disturbing pulse, their amplitude increases as the
signal moves toward the far end. The final peak value is therefore
proportional to the length of the line.
• tr,
• 2 tr ,
• tP,
• *2 tP,
The total width of the near end crosstalk pulse depends on the time
required by the disturbing signal edge to travel along the line (tP), plus the
time required by the negative edge (the crosstalk noise turn-off) to come
back to the near end (another tP).
7) At the far end we can observe a crosstalk noise pulse with duration
• *tr,
• 2 tr ,
• tP,
• 2 tP,
The far end crosstalk pulse is the sum of all elementary contributions
along the line, which move towards the far end at the same speed as the
disturbing signal. They sum in the amplitude domain, and the pulse width
corresponds to the width of each elementary contribution (t r).
10) If the length of a couple of lines is doubled (and all other parameters
per unit length remain unaffected):
The far end crosstalk pulse is the sum of all elementary contributions
along the line, which move towards the far end at the same speed of the
disturbing signal. Therefore the amplitude of far end crosstalk is
proportional to the line length.
The physical laws that causes crosstalk cannot be modified, but proper
design can keep under control both the amount of crosstalk (passive
countermeasures), and its effects (active countermeasures). This lesson
presents the design technique for electronic systems which allows us to
reduce the errors originated by crosstalk.
The previous lesson described how mutual coupling can induce noise
pulses on conductors which should stay in a fixed logic state. These
pulses have amplitude and width related to the characteristics of the
disturbance signals and the electrical parameters of the system. For
instance, the capacitive term can be evaluated as:
If these voltages and currents cross the logic threshold of logic devices,
they may be interpreted as false logic states. The goal of the designer is to
avoid errors caused by these spurious logic states, and can be achieved
through a series of countermeasures:
Active countermeasures limit the effects of noise towards logic circuits (the
spurious pulses, even if present, are not translated into a false logic state).
These techniques exploit some characteristics of noise signals, such as
the limited duration or energy. The top level active countermeasures
include error correction techniques, which can mask errors at the electrical
level, and are not addressed here.
Both the capacitive and the inductive terms are proportional to the
disturbing signal slew rate
Increasing tr means using low-speed logic circuits (the slowest logic family
which allows to fulfilment of timing specifications). The slow transition
among logic states may cause other problems (eg. long totem-pole current
pulses - see lesson 3). A first fundamental design rule is:
The other key parameters are the capacitive and inductive coupling.
All printed circuit boards (PCBs) and cables exhibit capacitive coupling
among tracks or conductors. A proper use of screens and ground
conductors may substantially reduce these parasitic capacitances.
In digital systems signal and ground return make loops, as in figure 2.6.
Ground currents follow the lowest impedance path, that is the minimum
resistance for DC or low frequency, and minimum inductance path for high
frequency components. The inductance is proportional to the
concatenated magnetic flow of a wire loop, which depends on the loop
area. Therefore to keep inductance low we must minimise loop area by
providing a return path as close as possible to the signal conductor.
When the PCB has a continuous ground plane, the return current follows a
minimum impedance path with minimum loop area, that is as close as
possible to the signal conductor, as in figure 2.7a.
Figure 2.7 Photo of the ground plane in a multilayer PCB. The yellow line
corresponds to the signal track (on another layer).
Figure 2.8 A wide loop can be made thinner by moving the GND pin near
the signal pin.
If the loops made by signals and ground returns have significant shared
area, they make a transformer (with single-turn windings in air), with high
mutual inductive coupling. If the loops have no common area, the
inductive coupling is minimised (figure 2.9).
Flat cables, with several conductors running at close spacing for some
distance, may have strong inductive coupling. Independent ground returns
on pins close to signals, minimises the inductive coupling, while using the
same pin for return currents from several signals creates nested loops
with high mutual inductance, as shown in figure 2.10.
Figure 2.10 Signal (Si) - Ground loops in a flat cable. The number of GND
connections is the same in both cases, but b) exhibits far less inductive
coupling.
and
The input signal V1 goes to an integrator, and only when the output of the
integrator crosses the threshold V TH the output state changes the state.
Pulses shorter then a time threshold TTH cannot move the integrator output
through the threshold, and are completely ignored. The structure has an
intrinsic delay of at least TTH (to decide if a state lasts longer that TTH, it
must be observed at least for this time). If TTH = 2 TP,all spurious pulses
originated by crosstalk are filtered out.
Using separate grounds reduces the nested current loops. This reduces
mutual inductive coupling, but has no effect on capacitive coupling.
The ground tracks interleaved between signal tracks and the ground
layers above and below the track are a good approximation of a complete
screen around the signal conductor.
Using slow logic families keeps the dV/dt of signal edges; the noise
towards other parts of the systems is proportional to disturbing signal slew
rate. However, the circuit speed must comply with timing requirements
(system-operation is a requirement as mandatory as low crosstalk!)
6) Integrating receivers
• *S G S G S G S G,
• S G G S S G G S,
• G G S S S S G G,
• S S S S G G G G.
Alternating signal and ground wires minimises the common area of the
loops, thus reducing mutual inductive coupling.
The output stage of logic circuits is made with two active devices
(transistors), which act as switches with complementary commands. In a
first approximation, the equivalent circuit of a logic output is a SPDT
(Single Pole Double Throw) switch, as shown in figure 3.1. (the diagrams
show CMOS devices, but similar considerations apply also to bipolar logic
circuits).
Fig 3.2 - Current flow during logic transition (Totem Pole current spike).
The duration of the current pulse depends on the speed of the input
transition; if it is too slow (eg if the input is indefinitely kept near VTH), the
current flow may damage the device due to overheating. Since only totem-
pole outputs exhibit this behaviour (Open Collectors have only one active
device towards GND), this current is called Totem Pole Current Spike.
Inductance on each
Package
pin
DIP 14 10 nH
DIP 68 100 nH
PLCC 10 nH
Wire
1 nH
bond
Flip chip 0,1nH
Wire bond and Flip-chip have far lower parasitic inductance due to shorter
connections.
OR
where t ri is the rise time of the current. A similar drop occurs on the power
supply lead (inductance L S). The actual supply voltage at the device pin is
VALE = V AL - VS - VG.
These voltages become noise on all other nodes (outputs and inputs) in
the same package, and modify both the actual input voltage (thus causing
possible false transitions), and the outputs which should stay in a steady
state.
This effect can be seen as crosstalk (fake signals from state transitions on
nearby signals), caused by common paths, rather than by mutual inductive
and capacitive coupling as seen in lesson 1.
The shift in ground potential due to switching noise affects all nodes of the
device, and may cause false signalling. It is referred to as ground
bounce.
To change the voltage level V OUT at the output node, the output stage of
the logic gate must charge or discharge the capacitance C O associated
with that node. This again means a current flow through the output stage
from V DD (L-H transitions, figure 3.3) or towards GND (L-H transitions,
figure 3.4 ).
The following values can be used for a first rough evaluation of the actual
total capacitance:
The current flowing in the output of a driver comes through the power and
ground pins. The totem pole spike and the output capacitance charge
current occur at any output node, and the total current depends on the
number of simultaneously switching outputs. When many drivers are
included in the same package, the currents for each output must flow
through the device power and ground pin. With severe capacitive loading
Numerical example 1:
; ;
OR
This current flows in the inductance LG, and causes a voltage drop
which appears directly on the ground pin of the driver, as in figure 3.7.
The voltage drop on the ground lead appears directly at the outputs in the
LOW state, since these are connected to ground through the driver output
transistor. The HIGH outputs are connected to the power supply V CC , and
get only second order effects (due to the current spike through the totem-
pole output and parasitic coupling).
The same model can be applied for LOW-HIGH transitions, taking into
account parasitic inductance of supply interconnections. Now the voltage
drop appears directly at the outputs in the HIGH state, connected to V CC ;
LOW outputs get only second o rder effects (figure 3.9).
The induced noise is proportional to the current spike, and this in turn is
higher when several outputs change state at the same time. This
disturbance is therefore called simultaneous switching noise.
Numerical example 2
An integrated circuit contains 8 drivers, in a DIP 20 package. Each driver works in the
conditions of example 1 (a current peak of 300 mA when the output changes logic state). The
total peak current is
8 x 0.3 = 2.4 A
Assuming a triangular current waveform with total duration 4 ns (the voltage rise time), and
peak value 4,8 A (doubled, to keep the same area, which corresponds to the same total charge
moved in/out the capacitor), the slew rate of the current is:
On inverting buffers with input at the HIGH state, this effect may bring
about self-oscillations. As shown in figure 3.10, at t = T1 the actual input
voltage V'I = V I - VG crosses the threshold V TH, and that changes the
output state, starting the sustained oscillations (not shown in figure 3.10).
• 0.8 V,
• *1.6 V,
• 3.2 V,
• 4.8 V.
The slew rate dv/dt is 1V/ns; the current can be evaluated as I = C /(dv/dt).
The total capacitance is 100 pF x 16 = 1,6 nF. Since V = R x I: Ground
bounce V = 1 ohm x (1.6 nF / (1 V/ns)) = 1.6 V
Currents must flow in digital circuits to power the active devices and
charge/discharge parasitic capacitance; these same currents flow in
ground and supply conductors and may cause false signalling because of
ground bounce and simultaneous switching noise. This lesson presents
the techniques used to distribute ground and supply to ICs in such a way
as to minimise this kind of noise. Reducing the disturbance of an
electronic system towards itself reduces also the Electro Magnetic
Interference (EMI) radiated to the external world.
The peak output current depends on capacitive load, that is on the number
of inputs connected to the driver. In the case of bus lines, a transceiver
insulates the bus line from internal board loads, thus reducing ground
bounce in bus drivers.
Figure 4.3 Bypass capacitor directly on device supply pin: the voltage V2
is not affected by currents of device 1.
The bypass capacitor can be placed also in other positions, provided that
the current required by device 1 could be supplied without causing voltage
drops on the ground/supply connections of device 2. Figure 4.4 is an
example of incorrect positioning.
Even with bypass capacitors, the impedance of ground and power supply
conductors must be kept as low as possible. Reducing this impedance
reduces the voltage drop for a given current. To reduce inductance,
The impedance of all conductors in the ground -supply path must be low,
and this includes the bypass capacitor itself. The impedance of a capacitor
should decrease with frequency, but real components have series
inductive and resistive components, caused by leads and dielectric losses
(figure 4.5).
The impedance diagram for the real capacitor (figure 4.6) has a capacitive
zone (left - for low frequency the inductance can be considered a short
circuit), and an inductive zone (right - for high frequency the capacitor can
be considered a short circuit), where impedance goes up with the
frequency. Between these two, we can define a resonant frequency,
where the impedance is reduced to a minimum value (only resistive loss).
Figure
a) Different capacitance value in the same package
4.7
b) The same capacitor in different packages (with different
equivalent inductance).
A first rough rule for effective bypass is to use one multilayer ceramic 10-
100 nF, placed close to the IC (about one capacitor every 10 drivers).
Capacitor leads add inductance, and must be kept as short as possible;
surface mount devices, with direct connection to supply and ground
conductors are better than Pin Through Hole (PTH) devices.
The minimum value for a bypass capacitor can be evaluated from the
circuit in figure 4.10.
The charges from the bypass flow to the load capacitors at the output of
drivers. For a DV change on a capacitor CL the amount of charge required
is Q = DV CL . This same amount of charge causes on the supply voltage
a change DVS = Q CB, and therefore:
DV C L = DVS CB,
CB = DV C L / DVS
Higher values can be used, as long as this does not increase the parasitic
inductance. Consider that large capacitors (eg aluminum electrolytic) have
good behavior at low frequency, but the wide case causes large inductive
components. To achieve low impedance over a wide range of frequency
we must put in parallel capacitors of different values and types. Small
ones will be placed as close as possible to the switching devices, to
supply the charges through low-inductance wires (typically one capacitor
per small package). Large ones are placed on arrays of packages, or at
board level.
Bypass capacitors store charges close to the places where they are
needed (the driver IC supply and ground pins), and provide the peak
current through low-impedance paths.
The printed circuit boards must be designed with a short distance between power and ground. This
is achieved by using power and ground planes, which can be electrically approximated with
distributed capacitance. A bypass capacitor as close as possible to the power and GND pins of the
device helps to provide a low-impedance path for the transient currents.
The finger layout in figure 4.12 uses only one layer, but forces the supply
current and the signal return current to flow in wide-area loops.
Figure 4.12
Rearranging the fingers as in figure 4.13 cuts the supply current loop area,
but keeps a wide signal-return loop (and uses two different layers).
The grid scheme in figure 4.14 requires bypass capacitors on each device to reduce loop area both
for supply and return currents (at each crossing a bypass capa citor provides a path for fast-changing
currents).
Figure 4.14 Grid distribution scheme and current flow through bypass
capacitors
The star connection, with series termination on each line at the driver side,
as in figure 4.17, possibly allows the designer to insert equal delays
between the driver and each clocked device. Each interconnection must
have the same parameters and length; this equalises the transmission
delays. Several lines converge to the same point (the driver output), and it
is difficult to get line impedance matching.
• 10 pF
• 100 pF
• *100 nF
• 10 mF
The bypass capacitor must supply the charges required by the totem-pole current spike, and provide
charge or discharge to output capacitance. The actual value can change depending on the parameters
of the IC and of the interconnection, but a value of 100 nF is correct in most cases. Too small
capacitors (less than 1 nF) cannot store enough charges, and too large ones have larger size with
higher parasitic inductance.
The electric charge stored in the bypass capacitor reaches the output stage of the driver through the
ground and supply connections. The voltage drop is proportional to the resistance and inductance of
these connections. Placing the capacitor as close as possible to the IC reduces resistance and
inductance in the current paths.
The bypass capacitor must have a capacitive behaviour in the frequency range corresponding to the
harmonic content of driver current spikes. With 2 ns rise-fall times the significant power of the
harmonics is around 50 MHz and above. The best capacitors are SMD, due to very short and low-
impedance connections towards the supply and ground plane.
• F=
• *F =
• F=
• F=
For w2 = 1/LC , Z = 0.
• Pure inductance
• Series R and L
• Pure resistance
• Parallel LC
With a R L C series model, above the resonant frequency the capacitive impedance is
far less than inductive and resistive terms. The capacitor can be modeled as a R-L
serial circuit.
This lesson describes how the transmission line behaviour can be verified
with some simple laboratory experiments.
We will observe on the oscilloscope the waveforms at both ends of a transmission line driven by a
square wave generator (the square wave corresponds to a sequence of voltage steps of alternate
polarity). As transmission line we shall use a rather long coaxial cable (10 to 20 meters) to achieve
propagation times long enough to allow the use of standard low cost instruments to verify the
propagation effects (signal generator and scope).
For all measurements set the signal generator for square wave, at about 2
V peak, 200 kHz rate.
Connect the signal generator to one end of the cable, and the scope to
both ends of the cable using high impedance probes. Use adapters at
both ends to mount the required R and C components.
The experimental setup for all proposed measurements is shown in figure 5.1. A 50 W coaxial cable
(type RG 58) is used as the transmission line, and is driven by a pulse generator. The waveforms at
source and termination ends are monitored on a scope. Incident and reflected waves can thus be
verified in several different operating conditions.
Figure 5.2 The cable and the test fixtures used for the measurements
1) Verify the no load (open circuit) output amplitude VB from the signal
generator (Figure 5.3 a).
2) Connect the signal generator to a known load RL (e.g. 100 W), and
measure the new VB; compute the output impedance RO of the signal
generator (Figure 5.3 b). The expected result is about 50 W.
1. Connect the signal generator to an open line (the coaxial cable); check the
waveform at near and far ends (Figure 5.4), and compare with results from lattice
diagrams.
The top trace shows signal at source end: the two steps
Figure correspond to the incident and to the reflected wave (after
5.4 2 tP). Since the source is matched to line impedance, there
is no further reflection.
From the waveform at the near end and from the cable length compute the
wave propagation speed U.
Figure 5.6 Measurement setup for RO > Z0, line open at far end
Figure 5.7 Waveforms for RO > Z0, line open at far end.
Repeat for a generator resistance lower than characteristic impedance (put a 22 W resistance in
parallel to the generator output). Since now the reflection coefficient at source is negative, steps have
alternated direction, causing oscillations at the far end.
Figure 5.9 Waveforms for RO < Z0, line open at far end.
For a first approximation analysis, the far end capacitor can be considered a short circuit when the
step arrives at the termination (GT = -1), and an open circuit (GT = 1) after the transient. Therefore
at t = t P (for the far end) and t = 2 t P (for the near end) the waveform corresponds to a short circuit
at the far end. For t >> time constant RC the waveform corresponds to an open line.
The waveforms at the near and far end of an open transmission line with matched impedance at the
driver are in figure 5.12. The cable length can be measured from the width of the intermediate step
at the near end, which corresponds to 2t P (100 ns in this experiment). The far-end waveform shows
a single step, because the incident and the reflected wave appear at this point at the same time.
Figure 5.12 Near end (top trace) and far end waveforms for an open line
with matched driver.
If we add another segment of coaxial cable to the far end, the total
propagation time is increased (wider intermediate step in the top
waveform), and the point C becomes an intermediate point in a longer
line. The bottom waveform shows separate incident and reflected waves.
The total cable length is still measured by the width of the intermediate step at the near end (about
150 ns in this experiment). The length of the additional cable can be measured from the width of the
intermediate step at the point C (previous far end, 50 ns in this experiment). The waveform in C
(bottom trace) shows an intermediate step, because the incident and the reflected wave now appear
at this point at different times.
Clamp diodes at the termination can limit the oscillations. Since logic devices have clamp diodes at
the inputs, connecting a receiver at the end of the line changes the waveforms as in figure 5.18. This
is a correct Incident Wave Switching (IWS) working condition.
Signal Integrity (SI) analysis concentrates on the analog analysis of the switching behavior of fast
digital signals. This analysis has become common over the past few years because more system
designs are experiencing the ringing, crosstalk and ground bounce problems associated with high-
speed signal switching. By using signal integrity analysis to predict and correct SI issues up-front in
the design process, the chance of first pass success is greatly increased. The earliest PCB-level SI
tools were designed to be run after placement and routing were complete, but before the design was
fabricated. SI analysis identified signals with high-speed problems, and those nets were then rerouted
to eliminate the problems found. This technique was really only effective with boards with a small
number of highspeed nets, and enough free space on the board to accommodate rerouting signals
where problems were predicted by analysis. The original “post-route” methodology evolved to
accommodate routing only critical nets first, then performing SI analysis on the partially routed
board to determine if the performance of those nets was satisfactory. Once the critical nets
performed acceptably, the rest of the board could be routed and then post-route SI analysis run as
usual. These methodologies were adequate for PCBs with small numbers of critical nets, but are still
not adequate for today’s more demanding system designs. Modern systems are dominated by buses
with speeds of 50 MT/s and up, having hundreds of critical nets. When a system becomes this
heavily populated with high-speed nets, the traditional approach of “route-analyze-fix” becomes
impossible. For these designs, a more systematic method for defining SI requirements and ensuring
their compliance is required.
12.1.1 A modern high-speed design methodology must involve the at least the following:
1. SI analysis (simulation) software
2. Defined requirements for high-speed signal behavior (e.g. flight time, overshoot)
3. Physical design (placement and routing) tools
1. Define each signal’s SI requirements (e.g. flight time, overshoot) and then use these requirements to assess the
quality of the placement and routing processes
2. Using each signal’s SI requirements, derive a placement and routing strategy for each group of signals, and then
use that strategy to drive the placement and routing processes.
Although the two strategies appear similar at first, they are really quite different. The first strategy is
just an extension of the old route-analyze-fix approach. Defining the electrical requirements up-front
makes it possible to run simulation automatically after routing and determine if a net needs to be
rerouted — allowing the router to continue iterating until the predicted net performance is
acceptable. Once the performance for a given net is predicted as acceptable, the net is “locked
down” and further edits to that net are disabled.
The primary problem with this approach is that locking down high-speed nets will block routing
channels, making the rest of the board harder to complete, and possibly requiring extra layers in the
board to complete the design. Any edits to a high-speed net require that SI analysis be rerun to
ensure the SI performance has not been affected
Real world conditions are those elements that vary from system to system whether the designer
wants them to vary or not. These variances are the result of normal manufacturing variances and
differences in operating conditions. As an example, a PC specified for 600 MHz operation is
expected to perform with any 600 MHz processor that is plugged (thus, it must work with “fast”
and “slow” 600 MHz devices). Similarly, the computer will probably be expected to function with
silicon temperatures from about 50 F to 150 F — even though the silicon’s performance degrades as
the system warms up. If the traces on the board are impedance-controlled, the system will still be
expected to vary if trace impedance varies by +/- 10%, which is typical for impedance controlled
designs.
Solution space analysis starts by defining pin scheduling, termination (if any) and nominal circuit
parameters for a representative high-speed net. If the net is part of a high-speed bus, then all the bits
of the bus will typically have the same electrical/physical constraints, such that finding a routing
solution for one bit of the bus constitutes finding a solution for the whole bus. If there are any pre-
existing constraints for device placement, they should be taken into account at this stage. For
example, if the processor cannot be less than 3” away from its corresponding chipset, it makes no
sense to explore potential routing strategies with processor — chipset connections less than 3” in
length. By taking known placement constraints into account up front, electrically valid (but
physically invalid) solutions can be eliminated early.
12.3.2 STEP 2 — DEFINE MANUFACTURING TOLERANCES AND THEIR MIN/MAX VALUES
The next step is to determine which manufacturing tolerances will have a significant effect on the
behavior of the circuit. For instance, the typical +/- 10% variance in characteristic impedance of an
impedance-controlled board may have a substantial effect on SI, while the +/- 1% variation in the
impedance of a precision terminator may not. Because solution space analysis looks at all
combinations of variables, it makes sense to analyze only those variances that are “significant,” in
which “significant” is a design–specific factor. Sensitivity analysis can be performed at this stage to
help prioritize the effect of the different manufacturing variances on circuit behavior. This helps
identify which variables can be excluded, or left until a final, detailed validation is to be performed.
Other optimizations can also help reduce simulation run time and improve accuracy –for example, if
multiple segments of the same net are routed on the same signal layer, the impedance of the
different segments will “track” (scale together). Modeling this tracking behavior has the benefit of
reducing the number of different cases that need to be analyzed, and eliminating unnecessary
pessimism in the results. Once the variables to be analyzed and their values are determined, the
min/max data is entered as part of the circuit model.
Recall that the design variances will ultimately become the “design rules” used for PCB layout.
These rules can be specified as physical rules (net length, spacing), as well as electrical rules (segment
delay, overshoot, crosstalk) – or as a combination of both. Once determined, the different design
variables and their min/max values are entered as part of the circuit model.
Some design rules may involve a relationship between different parts of the net. For instance, if two
net segments are required to always total to a certain length, the delay of one length should be
expressed as a function of the length of the other segment. This is a similar form of optimization to
the one discussed in step 2, in that this reduces the simulation run time and improves the accura cy
of the results.
12.3.4 STEP 4 — SET UP AND RUN A NUMBER OF SIMULATION CASES
Once the circuit and its variables/relationships are set up, the process of defining all the different
cases (combinations of variables) and analyzing them should be fairly automatic. This will be
dependent on the CAD software user for analysis. Traditional solution space analysis processes have
relied on collections of commercial tools and custom “scripts” to perform this process. In the
Cadence 13.6 release, Signal Explorer Expert performs this process automatically. The number of
cases to be analyzed grows rapidly as additional variables are introduced, and can rapidly expand to
require millions of simulation runs. To keep run times down and simulations practical, some method
of randomly running “subsets” of the full simulation job is required.
This serves two purposes:
1. It keeps the number of simulation runs manageable for large, multi-variable runs.
2. It allows small subsets of a large analysis job to be run quickly to identify problems, saving the large runs for later
after the topology has converged.
12.3.5 STEP 5 — EXAMINE THE SIMULATION RESULTS, IDENTIFY WHICH CASES FAILED AND
WHY
Solution space analyses are typically run as batch jobs that take anywhere from 5 minutes to many
hours. They return simulation results for each case analyzed — usually in terms of electrical
parameters like calculated flight time, overshoot, etc. The simulation results are compared against a
design requirement to determine which cases passed and which failed. For instance, the timing
budget might show that the flight time for a high-speed net must be greater than 300pS, and less
than 3.2nS, while electrical constraints might require that overshoot be limited to less than 500mV.
The solution space process is iterative, thus, the layout-analyze-fix loop has been moved all the way
forward in the design cycle. This process continues until the topology converges into a set of design
variances that allow the design to function correctly under all combinations of real-world conditions.
Keep in mind that not all topologies will converge. Failure to converge may indicate that a
placement/routing strategy is simply not viable, and that different placement/routing/termination
rules should be investigated.
12.3.8 STEP 8 — DERIVE DESIGN RULES FOR THE TARGET CAD SYSTEM
After a solution space analysis is defined, the electrical designer determines the applicable design
rules for layout. The “real world” variances used in solution space analysis are not passed to layout,
because the analysis accounted for the actual variances in the manufactured design (e.g. trace
impedance) even when the designer tried to keep the value fixed. The design rules passed to layout
are determined by the design variances (e.g. min/max segment length) used in solution space
analysis.
In an ideal world, design rules would be defined completely before placement and routing began. In
the real world, design rules are usually not completely defined before physical design begins, and
almost always change as the design and routing process progresses. If the CAD system allows new
design rule templates to be applied against the existing design without disturbing the
placement/routing, CAD designers can quickly assess the impact of changes in design rules against
existing work, identifying which signals will need to be rerouted, and the degree to which those nets
violate the updated design rules.
An additional advantage of this technique is that SI analysis need not be rerun every time the routing
of a critical net is updated. Because the design rules were determined from up-front comprehensive
analysis of worst-case conditions, any net routing that conforms to the design rules in the database
can be expected to function correctly.
12.3.10 STEP 10 — POST LAYOUT SI ANALYSIS
Post-layout SI analysis still has its place in a solution space design flow, because unexpected
problems can still occur. The emphasis shifts considerably, because the post-layout analysis is
intended to serve as a “signoff” analysis, instead of being used as the primary vehicle for identifying
SI issues. The problems uncovered during post-layout analysis, if any, tend to be isolated and
correctable on a case-by-case basis.
12.4 CONCLUSION
This article has reviewed the traditional techniques for identifying SI problems in high-speed
designs, and outlined the “solution space” approach for defining and driving high-speed placement
and routing strategies. This technique uses comprehensive up-front SI analysis to define robust
Word Definition
Method of interfacing drivers and receivers through a series capacitor. Often
used when the differential swing between drivers and receivers is compatible,
AC Coupling but common mode voltages of driver and receiver are not. Requires that a
minimum data frequency be established based on the RC time constant,
necessitating a run length limit.
Attenuation Reduction in amplitude of a signal.
BER See Bit Error Rate
BERT See Bit Error Rate Test or Bit Error Rate Tester
A measurement of the number of errors detected at a receiver in a given
length of time, sometimes specified as a percentage of received bits;
Bit Error Rate
sometimes specified in exponential form (10E-8 to indicate 1 bit error in 10E-8
bits).
An instrument used to determine the Bit Error Rate (BER) of a device or
Bit Error Rate Test or Bit Error Rate
system under test. It is generally made up of a test pattern generator,
Tester
receiver, and analyzer.
BIST Built-In Self Test
CDR See Clock Data Recovery.
Feature of multi-channel high-speed transceivers. Allows multiple channels to
Channel Bonding
be sued together, offering a greater aggregate bandwidth.
Bit sequence, which is transmitted by a high-speed transceiver when it is not
in use. The chirp is usually a repeating pattern of IDLE characters. The
Chirp
purpose of the chirp is to keep clock recovery circuits aligned and active while
the link is not transmitting data.
Feature of most high-speed serial transceivers. At the receiver, a clock is
Clock/Data Recovery generated based on the timing of data transitions. In this way, a clock signal is
derived from the data.
CML See Current Mode Logic.
Comma K-character
The DC component of a signal. In differential channels, it is the average
Common Mode voltage of the differential pair.
Undesirable signal coupling from noisy aggressor nets to victim nets. May be
Crosstalk eliminated by increasing the spacing between the nets or reducing signal
amplitude of the aggressor net.
A differential I/O standard used in high-speed serial channels. Voltage swing
Current Mode Logic is typically from 450 mV to 1200 mV.
A channel is said to be DC Balanced if it has an equal number of 1’s and 0’s
DC Balanced transmitted across it. Encoding schemes like 8B10B are designed to ensure
this.
Method of interfacing drivers and receivers without the use of series
DC Coupling capacitors. A direct connection (through PCB trace) from driver to receiver.
The component of jitter attributable to the data pattern in the channel.
Deterministic Jitter Different digital patterns have different spectral contents. These differing
spectral contents give rise to varying amounts of signal jitter.
Differential Signaling A signaling scheme, which uses two complementary signals to transmit data.
An eye diagram of a signal overlays the signal’s waveform over many cycles.
Each cycle’s waveform is aligned to a common timing reference, typically a
clock. An eye diagram provides a visual indication of the voltage and timing
uncertainty associated with the signal. It can be generated by synchronizing
an oscilloscope to a timing reference.
System where multiple logic levels are utilized instead of just two
Multilevel Signaling (high and low). This enables the tranmission of multiple bits in a
single waveform.
Phenomenon where a signal rises to a level greater than its steady-state
Overshoot
voltage before settling to its steady-state voltage.
PCS Physical Coding Sublayer
In the case of peak-to-peak voltage, a measure of a signal's total amplitude. In
Peak-to-Peak the case of peak-to-peak jitter, a measure of the extremes of excursion of the
bit transition times.
Positive Emitter-Coupled Logic. A differential IO standard based on the ECL
standard, but which operates with a positive supply voltage (ECL uses a
PECL
negative supply voltage). PECL is used in clocking and high-speed data
applications.
PLL Phase-Locked Loop
PMA Physical Media Attachment
PRBS Pseudo-Random Bit Sequence
Pre emphasis is magnitude boosting of high frequency spectral components
before launching the signal (wave) onto the Transmission Line. Transmission
Lines embedded in most standard PCB materials (FR4, Rogers 43xx, Nelco
and Rogers) suffer varying degrees of dispersion and loss in the 1 gigahertz
spectrum. This is mostly due to conductance losses (leakage from the copper
trace to any other conducting structure) and Skin Effect. Dispersion is a
phenomenon whereby spectral components travel at different velocities. The
waveform looks smeared when it arrives at the receiver.