Professional Documents
Culture Documents
Analytical
Methodology
of Tree Microstrip
Interconnects
Modelling For
Signal Distribution
Voltage Transfer Function and S-parameter
Analyses
Analytical Methodology of Tree Microstrip
Interconnects Modelling For Signal Distribution
Blaise Ravelo
Editor
Analytical Methodology
of Tree Microstrip
Interconnects Modelling
For Signal Distribution
Voltage Transfer Function and S-parameter
Analyses
123
Editor
Blaise Ravelo
ESIGELEC Graduate School of Engineering
Sotteville-lès-Rouen, Seine-Maritime
France
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface
The present book introduces a fast analysis and modelling of the printed circuit
board (PCB) arbitrary shape interconnect. The developed methods are based on the
consideration of interconnect elementary transmission line. The methodology is
essentially built with the combination of circuit, signal and transmission theory. The
materials presented in the book constitute basic tools to solve the electronic PCB
problematics regularly exposed to graduate students, academic researchers and
industrial engineers. Efficient way to elaborate the electrical topology equivalent to
tree behaved interconnect structure is developed. Analytical approach enabling to
determine the fundamental parameters as Z and Y and T and S-matrices is detailly
explored. Sensational methodology to establish the analytical answer about the
additional problematic related to the interbranch coupling of non-symmetric
interconnect PCB is also established. The feasibility of the analysis and modelling
methodology is verified with microstrip interconnect structures.
These research works have been implemented within the frame of the “Time
Domain Electromagnetic Characterization and Simulation for EMC” (TECS) pro-
ject 2009–2013 which is part-funded by the Haute-Normandie Region (FRANCE)
and the ERDF via the Franco-British Interreg IVA programme No 4081.
Acknowledgement is made to the Upper Normandy Region for the PULSAT
2013–2015 project support of this research work through the FEDER fund.
Acknowledgement is made to the Normandy Region for the BRIDGE project
support of this research work through the FEDER fund.
v
Contents
1 General Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Blaise Ravelo
2 Basic Analysis of Single-Input Single-Output (SISO) PCB
Interconnect Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Thomas Eudes and Blaise Ravelo
3 Discrete Periodical Model of Microstrip Line with Cascaded
Elementary L-Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Blaise Ravelo and Lala Rajaoarisoa
4 Modelling of the Signal Delay Induced by PCB Interconnect
SISO Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Blaise Ravelo and Thomas Eudes
5 Analytical Modeling Methodology of Single-Input
Multiple-Output (SIMO) Symmetric Tree Interconnects
by Using Lumped Element L-Cell . . . . . . . . . . . . . . . . . . . . . . . . . . 79
Blaise Ravelo
6 Symmetric Tree Interconnects Modeling with Elementary
Distributed RC-Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Blaise Ravelo
7 Z/Y/T/S-Matrices’ Modelling of Symmetric SIMO Structure
Based on Elementary Distributed RLC-Cell . . . . . . . . . . . . . . . . . . 117
Thomas Eudes and Blaise Ravelo
8 Z/Y/T/S-Matrices’ Analysis of Non-symmetric SIMO Tree Based
on Elementary Distributed Element . . . . . . . . . . . . . . . . . . . . . . . . 137
Thomas Eudes, Blaise Ravelo, Thierry Lacrevaz and Bernard Fléchet
9 Cartographical Analyses of Reflection and Transmission
Coefficients of Shunt Coupled Lines . . . . . . . . . . . . . . . . . . . . . . . . 167
Blaise Ravelo
vii
viii Contents
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
About the Author
ix
x About the Author
2D Two dimensions
3D Three dimensions
2.5G/3G/4G Mobile phone generations
AC Alternative current (generally used for the frequency analysis)
ADS® Advanced Design System® (Simulation tool from Keysight
(ex-Agilent®) Technologies®)
BGA Ball grid array
BW Bandwidth
CMOS Complementary metal-oxide semiconductor
CPL Coupled-parallel-line
CPU Computer processor unit
CTL Coupled transmission line
CST® Computer Simulation Technology® (Simulation tool from Dassault
Aviation®)
CTF Current transfer function
Cu Copper
CUT Circuit under test
DC Direct current
DDR Double data rate
DIMM Dual inline memory module
EM Electromagnetic
EMC Electromagnetic compatibility
EMDS® Electromagnetic Design System® (Simulation tool from Keysight
Technologies®)
EMI Electromagnetic interference
FEXT Far-end cross talk
FFT Fast Fourier Transform
FR4 Epoxy dielectric substrate (In this book, used to build the microstrip
line)
xi
xii Abbreviations
Tx Transmitter
TZ Transmission zero
UWB Ultra-Wide Band
VNA Vector Network Analyzer
VLSI Very large system integration
VTF Voltage transfer function
[Y] Admittance matrix
[Z] Impedance matrix
Chapter 1
General Introduction
Blaise Ravelo
To meet the public and industrial needs, the printed circuit boards (PCBs) must oper-
ate with the higher speed and design density. With the increase of the data speed and
the design complexity, the interconnect effect becomes the dominant factor of signal
degradation. The present book addresses a basic methodology to analyse the printed
circuit board electrical interconnects. The chapters elaborate the different analyti-
cal methodologies suitable to the interconnect modelling. The methods are applied
from the simplex cases of typically single-input single-output (SISO) to complex
cases of single-input multiple-output (SIMO) structures. The SIMO interconnects
are represented as tree networks.
The analysis presented in Chaps. 2, 3 and 4 is based on the signal, circuit, system
and transmission line (TL) theories. The methodology explored is fundamentally
based on the equivalent circuit approach of the PCB electrical interconnects. As
introduced in Chap. 2, the analysis can be performed in the frequency domain or
in the time domain in function of the input signal behaviour. The equivalent circuit
must be established in function of the operating signal bandwidth and the interconnect
physical parameters. The interconnects can be assumed as ideal connection wire or
lumped element-based circuits, or distributed TL structure. The distributed TLs are
basically defined by the characteristic and propagation constant parameters. The
lumped elements are essentially a L-topology cell composed of two-port impedance
B. Ravelo (B)
Graduate Engineering School, ESIGELEC, Sotteville les Rouen, Seine-Maritime, France
e-mail: blaise.ravelo@yahoo.fr
and two-port admittance. In the other words, the interconnect structure is equivalent
to electrical multiport system. In function of the number of output nodes, two types
of system are analysed in this book.
The SISO systems are two-port block diagrams whose one port is the input and
the other one is the output. Acting as electrical system, the analytical approach is to
elaborate the relation between the input and output currents and voltages. The relation
can be expressed analytical two-dimensional (2D) matrix approach. As developed
in Chaps. 2, 3 and 4, the analysis can be proceeded in the function of the unknown
variables and the interconnect configuration, with:
– The impedance or [Z] matrix,
– The admittance or [Y ] matrix,
– The transfer or chain or [T ] matrix,
– Or the scattering or [S] matrix.
In addition, the different signal integrity parameters as attenuation loss, propaga-
tion delay, rise-time, fall-time, under-shoot and over-shoot are estimated in fast way
in Chap. 4.
– Step 1: Identification of the interconnect tree electrical topology and the consti-
tuting TL parameters,
– Step 2: Determination of the reduced circuit diagram via SIMO–SISO transform,
– Step 3: Equating the output currents and/or voltages in function of the input, and
analytical calculation and Z/Y /T /S-matrices,
– Step 4: Determination of current or voltage transfer functions (CTF and VTF).
The symmetric interconnect trees present similar electrical responses at all output
nodes. The analyses of this symmetric-type tree-type are developed in Chaps. 5, 6
and 7. The modelling of symmetric-type tree constituted by lumped RC and RLC cells
is developed in Chap. 5. Analyses of multilevel symmetric tree constituted by RC
and RLC network cells are proposed. The modelling of symmetric tree constituted
by RC- and RLC-based distributed TLs is elaborated in Chap. 6 and in Chap. 7,
respectively.
Because of the conductor and dielectric material properties, the TL parameters are
particularly sensitive to the ambient temperature variation. Meanwhile, the tem-
perature can affect the PCB interconnect behaviours. Chapter 11 introduces an
empirical study illustrating how the microstrip line behaviour varies with the ambi-
ent temperature effect. The study is based on the empirical observation of the
S-parameters.
2.1 Introduction
The last century was particularly remarkable with the spectacular progress of the
microelectronic semiconductor industries. This constant technological progress was
till now unique in the mankind history. It is nowadays a source of innovative product
developments in numerous civil sectors as the mobile phones, multimedia systems,
medical equipments and even vehicles. As an undeniable quantification of this indus-
trial development, according to Moore’s law prediction, at each generation the size
of electronic PCB is reduced of 30% and every two or three years, the number of the
transistors integrated in the microelectronic chips must be increased of four times
[1]. Despite this fascinating scenario of technological development, the structures
of the current interconnection circuitry between the logic gates, electronic chips and
digital devices, and even the buses linking the different electronic boards become
more and more complicated [2–7].
Furthermore, due to the incessant increase of the operating frequencies and the
integration density, the contributions of the interconnection electric parameters as the
resistance, capacitance and inductance effects [8–15] cannot presently be neglected
by the electronic equipment manufacturers. Indeed, the SI which represents the
numerical/digital data is not preserved especially for the high-rate communication.
This signal degradation effect needs to be taken into account by the electronic design-
ers during the manufacture processes of the high-speed electronic devices. For this
reason, different characterization techniques of the interconnection TL based on the
S-parameter extractions were introduced since two decades ago [16–21]. In this optic,
performance optimization methods dedicated to the clock tree distribution networks
were proposed [22–24] in order to minimize the undesired influences of intercon-
nection systems notably for the VLSI circuits. In addition, a compensation technique
based on the insertion of repeaters between the piece of TLs was also proposed [11,
25] in order to reduce the TL interconnection negative effects. As improvement of
this interconnection compensation technique, an alternative solution based on the
use of negative group delay (NGD) circuits is introduced in [26, 27].
Notwithstanding these various technical solutions, further simpler, faster and more
accurate methods need to be developed to model the interconnection line effects espe-
cially for the complex circuitry especially, in transient domain. One underlines that
most of existing characterization methods were developed only in frequency domain
by using the S-parameter characterization deduced from the geometrical parameters
(width, length, substrate height) of the TL and the electromagnetic properties (dielec-
tric permittivity and metal resistivity) [16–20, 28]. Whereas the existing time-domain
characterization methods are usually based on the use of the classical RC or RLC
transfer function models [8–15, 29]. So, more complete modelling method is still
necessary for the estimation of the transient response induced by the interconnection
TLs particularly for the high-speed digital and/or mixed electronic systems.
For this reason, we suggest developing further modelling process allowing to
forecast simply the influences of RF/digital interconnection circuitries on the signal
quality via the analysis of the transient response behaviours. To make this chapter
more understandable, the theoretical analysis highlighting the determination of the
microstrip TL high-frequency parameters as the characteristic impedance and con-
stant propagation from the geometrical (width, length, substrate height) and physical
(dielectric permittivity and metal resistivity) characteristics is recalled in Sect. 2.2.
Then, the calculation method enabling the extraction of the RLCG model per-unit-
length parameters is equally presented. Hence, mathematical analyses explaining the
determination of the linear transfer function and S-parameters of the interconnec-
tion transient responses are developed in Sects. 2.2, 2.3 and 2.4. To illustrate the
feasibility of the modelling concept, discussions on validation results based on the
comparison with standard electronic simulation tool SPICE results are the object of
Sect. 2.5. Finally, the last section is the conclusion of the chapter.
For the beginning, let us consider the unit L-cell composed of Z-series impedance and
Y-parallel admittance depicted in Fig. 2.1. This fundamental circuit which behaves
as a passive four-terminal network is excited by the input voltage vin at the node N in
and generates an output voltage vout at the node N out .
According to the circuit and system theory, it is well known that the transfer matrix
of the L-cell shown in Fig. 2.1 is merely written as:
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 7
vin Y v out
1+ Z ·Y Z
[T ] = . (2.1)
Y 1
The expressions of the transfer function H(s) and the input impedance Z in (s)
associated to this transfer matrix are written as:
Vout (s) 1 1
H (s) = = = , (2.2)
Vin (s) T11 1+ Z ·Y
T11 1
Z in (s) = =Z+ , (2.3)
T21 Y
For the sake of simplification, let us consider the generalized periodical network
comprised of n-element identical L-cells cascaded as illustrated in Fig. 2.2 [30] by
1 st-cell n th-cell
Z Z
o u tp u t
in p u t
Y Y
Fig. 2.2 Periodical n-element identical L-cells in cascade formed by Z-series impedance and Y-
parallel impedance [30]
8 T. Eudes and B. Ravelo
supposing that each elementary cell is formed by Z-series impedance and Y-parallel
admittance.
According to the circuit and system theory, the transfer matrix of this circuit
consisted of n-cells cascaded can be written as the matrix product:
n
1+ Z ·Y Z
[Mn ] = . (2.4)
Y 1
k=1
where the four elements of the initial matrix represent the transfer matrix of one
unique L-cell:
1+ Z ·Y Z
[M1 ] = , (2.9)
Y 1
According to the electronic circuit and system, the transfer function of this sys-
tem is the inverse of the first element M n11 . The recursive relationship between the
consecutive transfer function Tn and Tn+1 is expressed as:
Tn
Tn+1 = . (2.10)
1 + Y (Z + Tn · Mn,12 )
Figure 2.3 represents the microstrip TL structure. It presents the physical parame-
ters: geometrical length d, width w and metallization thickness t. The constituting
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 9
y
z
x
γ ( f ) = α( f ) + j · β( f ). (2.11)
where
30.66 h 0.7528
ψ(w/ h) = 6 + (2π − 6) · exp − , (2.13)
w
with η0 is the air impedance and εeff is the effective permittivity of the medium. One
points out that as reported in [31], by denoting c is the speed of light in the vacuum,
the constant propagation:
2π f
γ ( f ) = αc ( f ) + αd ( f ) + j εeff ( f ), (2.14)
c
includes the metallic conductor and the dielectric losses, respectively, given by:
2
h 1 2h Rs ( f ) 32 − wh
αc ( f ) = 1.38 · 1 + · 1 + · ln · · , (2.15)
weff π t h · Z c ( f ) 32 + w 2
h
10 T. Eudes and B. Ravelo
εr εeff − 1 c tan(δ)
αd ( f ) = 27.3 · √ · , (2.16)
εr − 1 εeff 2π f
with:
Rs ( f ) = π · μ0 · ρ · f , (2.17)
w + 1.25
π
1 + ln 2ht , if w
< 2π
1
weff = h (2.18)
w + 1.25
π
1 + ln 4πw
t
, if w
h
≥ 2π1
For the calculations including the radiating loss, more explicit analytical expressions
of the propagation constant real part or the per-unit-length loss constant α(f ) are
presented in [31–34]. Knowing Z c and γ (f ), one can extract the equivalent TL RLCG
model with the per-unit-length parameters Ru , L u , C u and Gu as described in Fig. 2.4.
As established in [28], the TL global parameters:
R = Ru · d, (2.19)
L = L u·d , (2.20)
C = Cu · d, (2.21)
and
G = G u · d, (2.22)
Ru x Lu x Ru x Lu x
Cu x Gu x Cu x Gu x
x x
γ ( f ) · Zc
L= , (2.24)
ω
γ Z( cf )
C= , (2.25)
ω
γ( f )
G= , (2.26)
Zc
where [z] and [z] are, respectively, the real and imaginary parts of the complex
number z, and
ω = 2π f, (2.27)
G = d · tan(δ) · ω · C, (2.31)
d
Zs
vs (Zu, ) vL ZL
0.35
f max ≈ . (2.33)
tr
It means that to study the analog–digital systems, one can consider its transfer
function behaviour from DC to f max . More generally, one can take into account the
Zs Ru x Lu x Ru x Lu x
vs Cu x Gu x Cu x Gu x vL ZL
Fig. 2.6 Interconnection RLCG model driven by a voltage source V s with inner impedance Z s and
loaded by Z L
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 13
transfer matrix of the TL as its baseband modelling behaviour by exploiting, for exam-
ple, the equivalent polynomial expression of the matrix [TT L ]. This approximation
can be realized, for example, by applying the second-order Maclaurin expanding of
each element of the transfer matrix. Therefore, one obtains the following expression:
1 + Z (s)·Y (s)
Z (s)
[TT L ] ≈ 2
(s) , (2.34)
Y (s) 1 + Z (s)·Y
2
with:
Z (s) = R + L · s, (2.35)
and
Y (s) = G + C · s. (2.36)
Therefore, according to the theory of electronic circuit and system, the overall
transfer matrix of the system schematized in Fig. 2.6 can be assimilated as:
⎡ ⎤
1 + 21 RG + s(LG + RC) + s 2 LC R + sL
[T ](s) = Tin · ⎣ ⎦ · [Tout ] (2.37)
G + sC 1 + 21 RG + s(LG + RC) + s 2 LC
where
1 Zs
[Tin ] = , (2.38)
0 1
and
1 0
[Tout ] = 1 , (2.39)
ZL
1
represents, respectively, the transfer matrices of the series impedance Z s and the
output shunt impedance Z L .
From the transfer matrix [T (s)] expressed in (2.37), one can demonstrate easily that
the overall VTF H (s) can be formulated as follows:
1 1
H (s) = = , (2.40)
[T11 (s)] a0 (s) + a1 (s) · s + a2 (s) · s 2
14 T. Eudes and B. Ravelo
where
(2 + R · G + 2G · Z s ) (2 + R · G)
a0 (s) = ZL + Z s + R, (2.41)
2 2
(R · C + L · G + 2C · Z s )Z L (R · C + L · G)
a1 (s) = +L+ Zs , (2.42)
2 2
L ·C
a2 (s) = (Z L + Z s ). (2.43)
2
To confirm the efficiency of this transfer function model, time-domain responses
computed with MATLAB and then compared with the results from the SPICE ADS®
simulations are discussed in Sect. 2.5.1 and 2.5.2. Thus, we will evaluate the examples
of interconnection transient parameters as rise-/fall-times and the 50% propagation
delay.
The basic configuration of the interconnect system will be examined. This latter is
composed of a source and load having impedances, respectively, denoted Z s and Z L .
Figure 2.7 depicts the schematic diagram of the system under consideration. It is
mainly comprised of an interconnect TL hypothetically having physical length d,
characteristic impedance Z c and propagation constant γ .
For the mathematical analysis of the system introduced in Fig. 2.7, the interconnect
line is assumed as its distributed RLCG model. So, the circuit under study will be
transformed as shown in Fig. 2.8.
Ru , L u , C u and Gu are, respectively, the per-unit-length resistance, inductance,
capacitance and conductance of the TL.
Fig. 2.7 Illustration circuit diagram of the whole interconnect system under consideration, driven
by a voltage source with impedance Z s loaded by Z L
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 15
Interconnect line
Source Ru x Lu x Ru x Lu x Load
vin Zs Cu x Gu x Cu x Gu x ZL vout
Fig. 2.8 Distributed RLCG model of the interconnect system shown in Fig. 2.7
So, according to the circuit and system theory [36], the voltage transfer function
associated with the ABCD-matrix can be written as:
Vout (s) 1
G(s) = = .
Vin (s) Z c (s)
+ Z s (s)
· sinh γ (s) · d + 1 + Z s (s)
· cosh γ (s) · d
Z L (s) Z c (s) Z L (s)
(2.46)
16 T. Eudes and B. Ravelo
W, h, t, r, tan , , d
SI Parameters (Delay,
Transient
over/under-shoot and
attenuation) Response Model
Output signal
Fig. 2.9 Flow diagram summarizing the methodology of the modelling method [34]
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 17
TL
load gate
FR4-substrate
To give evidence to the relevance of the previous theoretical concepts, let us con-
sider the arbitrarily folded microstrip interconnection line pictured in Fig. 2.10 [37].
This microstrip device was simulated in SPICE schematic environment of standard
microwave electronic tool ADS® from AgilentTM . This structure was printed on
the FR4 substrate having physical characteristics and arbitrary chosen geometrical
parameters summarized in Table 2.1.
By applying the RLCG model extraction formulations introduced earlier in
Sect. 2.2, one obtains the following per-unit-length values: Ru = 16.6 /m, L u =
464.7 nH/m, C u = 74.7 pF/m and Gu = 5.6 mS/m.
With this numerical value, we have implemented the above transfer function into
MATLAB program. Then, to generate the transient responses of the structure, a
periodical trapezoidal pulse presenting the temporal characteristics (t r : rise time, t f :
fall time, T p : period, t w : pulse time duration) with steady states V low = 0 V and V high
normalized at 1 V addressed in Table 2.2 was considered as the input.
18 T. Eudes and B. Ravelo
Since the rise-/fall-time of the considered input signal is about 100 ps, according to
(2.33), the maximum frequency required for the proposed polynomial model response
has been estimated of about 3.5 GHz. We verify that it enables to generate transient
response taking into account the propagating TEM modes through the structure under
test which is compliant with the second-order transfer function expressed in (2.20). In
order to demonstrate the adaptability of the proposed modelling method for various
types of the load impedance, let us analyse the computed results first, for the resistive
load:
Z L = RL , (2.47)
It is interesting to note that the static gain of the overall understudy system is equal
to:
H (0) = R L /R S , (2.49)
when the source and load impedances are purely resistive. In the present case, we
have changed the value of the resistive load RL = {10, 500 } and the length of the
microstrip line d = {8.5, 13.5 mm}. So, one gets the comparative results plotted in
Figs. 2.11. The proposed model responses are plotted in full lines, and the ADS®
simulation results are in dashed lines.
One can see that the transient response generated from the proposed model agrees
very well with the SPICE simulations. One observes that through the microstrip inter-
connection, the pulse signal is significantly degraded. Here, with the two computation
processes, one evaluates rise-/fall-times of about t r ≈ 0.14 ns for RL = 500 and t r
≈ 0.43 ns for RL = 10 . Then, one evaluates rise-/fall-time relative errors of about
1%.
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 19
voltage (V)
13.5 mm
0.5
-0.5
-1
0 0.5 1 1.5 2 2.5 3
time (ns)
(b)
2
Vs RL=10 RL=500
1.5
1
voltage (V)
0.5
-0.5
-1
0 0.5 1 1.5 2 2.5 3
time (ns)
(a)
2 Vs CL=1pF CL=10pF CL=30pF
1.5
1
voltage (V)
0.5
-0.5
-1
0 0.5 1 1.5 2 2.5 3
time (ns)
(b)
2 Vs CL=1pF CL=10pF CL=30pF
1.5
1
voltage (V)
0.5
-0.5
-1
0 0.5 1 1.5 2 2.5 3
time (ns)
Fig. 2.12 Transient responses of the microstrip structure shown in Fig. 6 for different values of the
capacitive load C L : a d = 8.5 mm and b d = 13.5 mm [37]
Figure 2.13 represents the 3D design of the considered millimetre length structure
[43]. It is composed of microstrip interconnect line under test (LUT) having phys-
ical width w = 20 μm, variable length d driven by as voltage source with internal
resistance Rs = 10 and loaded by a capacitance C L = 1 pF.
This conductor formed by Cu metallization was printed on the alumina substrate
with relative permittivity εr = 9.6 and thickness h = 250 μm. By using the extraction
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 21
Alumina Via
Substrate ground
By substituting these parameters into the reduced transfer function models expressed
in (2.5) and (2.7), one gets the results reported in the following subsections which are
performed with the electronic microwave circuit simulator ADS® from AgilentTM .
In this paragraph, frequency computation results of the circuitry shown in Fig. 2.13
are compared with those of the first- and second-order reduced model established in
previous section. For that EM and circuit SPICE co-simulations were carried out with
the LUT including the source and load parameters from DC to 40 GHz. Therefore,
one gets the magnitude and phase plotted in Fig. 2.14.
It can be seen that beyond 3 GHz base bandwidth the first-order model presents
magnitude relative errors of about 100% . However, very good agreement between the
SPICE simulations and the calculations made with the model established in previous
section is found for the second order. To get further insights about the accuracy of the
polynomial models in function of the LUT physical length, comparative computa-
tions between the frequency results from the models and the test circuit were made.
Therefore, one gets the maximum relative errors on the magnitude and the phase of
the transfer function up to 40 GHz displayed in Figs. 2.15. By varying the length of
the LUT from 1 to 10 mm, one finds that the magnitude relative errors change from
about 70% to above 200% for the first-order model, whereas the second-order model
permits to achieve relative error lower than 1%.
As shown in Fig. 2.15b, one remarks that the first-order model presents phase
relative errors going up to 80%. This latter is only lower than 4% for the second-order
model.
22 T. Eudes and B. Ravelo
Transfer function
simulation of the structure LUT
magnitude, dB
schematized in Fig. 2.13 and 1st order m odel
0 2nd order m odel
the proposed model:
a magnitude in dB and
-1 0
b phase in degree
-2 0
-3 0
0 10 20 30 40
(b) Frequency, GHz
0
Transfer function
phase, °
-1 0 0 LU T
st
1 order m odel
nd
2 order m odel
-2 0 0
0 10 20 30 40
Frequency, GHz
1 4 7 10
Length d, mm
(b)
100
relative errors, %
80
Phase
60 1st order
2nd order
40
20
0
1 4 7 10
Length d, mm
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 23
In order to evidence the accuracy of our models in function of the SI parameters (rise-
time, propagation delay, settling time, over-/under-shoot), one realizes time-domain
computations by injecting RF-/digital-or square wave pulse voltage with rise-/fall-
times equal to 5 ps. So, by using 10 Gbits/s rate analog–digital input signal, one gets
the results displayed in Fig. 2.16.
Once again, one can see that the second-order model presents relative errors lower
than 1% compared to the SPICE computation. However, the first-order one does not
enable to predict the interconnect line transient responses notably the overshoot. This
statement was also confirmed with lower rate test digital signal of 1 Gbit/s. Therefore,
one gets the transient responses displayed in Fig. 2.17.
These transient results confirm that with the reduced second-order model, one can
predict effectively the behaviour of the degradation of the digital/analog or mixed
signal for the high-speed data.
nd
ultra-high-speed 2 order m odel
analog-digital input with
10 Gbits/s rate
0
-1
0 .0 0 .2 0 .4 0 .6
Time, ns
1 1 order m odel
nd
ultra-high-speed 2 order m odel
analog-digital input with
1 Gbits/s rate
0
-1
0 1 2 3
Time, ns
24 T. Eudes and B. Ravelo
2.5.2.4 Conclusion
Polynomial models of high-speed interconnect lines for the SI predictions are inves-
tigated theoretically and numerically. The model proposed is based on the exploita-
tion of the distributed RLCG line defined from the transmission line theory. By using
Maclaurin polynomial expanding, approximation of the transfer function was real-
ized. To verify the relevance of the models, microstrip interconnect line with 20 μm
width was considered and simulated in EM and circuit environments. Then, one
extracted the per-unit-length parameters of the equivalent RLCG line by using the
technique introduced in [16]. As results, it was shown that only with the first-order
model, relative errors higher that 100% were found via both frequency and time-
domain analyses. However, with the second-order model, relative errors lower than
1% were evidenced. This illustrates that this second-order model is useful for the
fast prediction of SI parameters.
(a)
w
d h
(b)
Fig. 2.19 a 3D design of the tested serpentine interconnect line (IUT2 ) and b its photograph [37]
26 T. Eudes and B. Ravelo
600
0 1 2 3 4 5 6
56
54
52
0 1 2 3 4 5 6
40
20
0
0 1 2 3 4 5 6
Frequency (GHz)
By applying the RLCG model extraction formulations described earlier in Sect. 2.2,
the following per-unit-length parameters are obtained at 1 GHz: Ru = 32.22 /m,
L u = 621.26 nH/m, C u = 53.69 pF/m and Gu = 2.8 mS/m.
The variation of per-unit-length parameters versus frequency is sketched in
Fig. 2.20. As expected, mainly due to the skin effect, we find that the resistance
and conductance parameters are relatively sensitive to the frequency change.
With these parameters, we have applied the technique indicated by the flow work
shown in Fig. 2.15 into MATLAB program to determine the ABCD-, S- and Z-
matrices of the IUTs by considering the source and load with impedances as Z 0 =
50 . Then, in order to demonstrate the accuracy of the model, comparisons with
EM ADS-Momentum ® simulations and measurements with the fixture shown in
Fig. 2.21 were performed [37]. It is worth noting that the S-parameter measure-
ment was realized with the Agilent network analyser ENA E5071C through SOLT
calibration.
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 27
Frequency-Domain Analyses
As a result, the S-parameters plotted in Fig. 2.22 were obtained with the IUT1 . It
turns out that the model developed is in excellent agreement with the simulations
and measurements here in UWB (from DC to 6 GHz). It is important to note that
the models permit to predict the periodical aspects specific to broadband responses
of the S-parameters which cannot be done with the classical polynomial model. By
referring to the measurement results, a maximal absolute error of insertion loss S 21
of about 0.4 dB can be estimated here. Thus, compared to the simulations, due to
the numerical inaccuracy of S-parameters notably at the high frequencies, it is about
1 dB with ADS® simulation results.
By using the S-to-Z transform, the access and transfer impedances of the intercon-
nect lines including the source and load impedances are plotted. Then, comparisons
with ADS simulations and measurements were made. Once again, as illustrated in
Figs. 2.23, IUT impedance frequency responses in very good correlations were found
with the possibility of ripple effect predictions. Due to the numerical errors related
28 T. Eudes and B. Ravelo
im pedances (k )
system computed from the 0 .3
model proposed (grey full
Access
line), ADS simulation 0 .2 Z11
(dashed line) and
measurement (black full 0 .1 Z22
line) [37]
0 .0
0 1 2 3 4 5 6
F r e q u e n c y (G H z )
M odel
ADS
(b) M e a su re m e n t
150
im pedances ( )
Transfer
100
50
0
0 1 2 3 4 5 6
F re q u e n c y (G H z )
This subsection is dealing with the time-domain validations for the high-speed appli-
cations by considering the measured and modelled S-parameters. Then, transient
computations were conducted for the input square wave pulse with 0.5 Gbits/s rate
having 50 ps rise- and fall-times including the noise effects. It is worthy of note that
the source is represented by impedance Z s = 50 and the load is a parallel RC
network composed of a 500- resistor and a 4-pF capacitor. Therefore, one realizes
the transient responses of IUT1 and IUT2 plotted in Fig. 2.24 have been obtained.
As expected, the model allows an achievement of a good prediction of the transient
responses from the measured S-parameters.
In addition to this time-domain verification, more realistic computations were
made by injecting an arbitrary binary sequence of 8-bit data “10100100”. This test
signal corresponds to mixed square wave signal with 1-Gigasymbol/s rate. So, the
time-domain results displayed in Fig. 2.25 are recorded for the IUT1 and in Fig. 2.26
for the IUT2 .
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 29
IUT2
Voltage (V)
V0
V 0 /2 IUT1
0 2 4 6 8
Time (ns)
V0
Voltage (V)
V0 /2
0 5 10
Time (ns)
V0
Voltage (V)
V0 /2
0 5 10
Time (ns)
30 T. Eudes and B. Ravelo
It must be pointed out that here a relative accuracy lower than 1% between
the models proposed and SPICE computations from the measured S-parameters is
assessed.
2.5.3.4 Conclusion
A transfer function modelling method of the millimetre interconnection for the high-
speed integrity analysis in the RF-/digital PCB is investigated. The established model
is based on the exploitation of the electrical RLCG model transfer matrix which was
assumed as its second-order polynomial linear model.
To evidence the functionality of the introduced method, a microstrip intercon-
nection TL driven and loaded by the logic gates was considered and analysed. It
was described that the per-unit-length equivalent model parameters were determined
from the TL physical and the geometrical parameters as the substrate permittivity
and height, and the line width and length. Then, the equivalent transfer function lin-
ear model of the overall system composed of a TL combined with the logic gates is
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 31
mathematically established. Thus, for the testing process, it was excited with a peri-
odical trapezoidal pulse voltage with 2 Gbits/s rate. So, good agreement between the
transient responses from the established transfer function calculated with MATLAB
and those from SPICE simulations of the overall structure was found. It was shown
that the microstrip interconnection responses are completely degraded. Then, after
the sweep of the load impedance values assigned as resistance and capacitance loads,
one observes also that the calculated transient results remain well correlated to the
simulations.
One emphasizes that the design and the simulation of this type of microwave-
digital interconnections become very difficult to carry out when the circuit is com-
posed of thousands of logic gates. For this reason, it seems important to exploit the
proposed modelling method for the prediction the analog–digital signal behaviours
along the interconnections. Thanks to the accuracy and the computation time gain,
we think that the proposed method can be a good candidate for the modelling of
the complex structure of the interconnection circuitry in the high-density integrated
circuit as the clock distribution networks.
In the continuation of this work, one plans to improve this method for the esti-
mation of the RF microwave and digital electronic device interconnection effects
by taking into account the non-uniformity of the TL and the eventual crosstalk with
the neighbourhood TLs. Then, we would employ the proposed method to the devel-
opment of the equalization technique with the NGD circuit for the reduction of
the correction of the digital signal degradation with MMIC and digital integrated
systems.
References
10. A.B. Kahng, S. Muddu, An analytical delay model of RLC interconnects. IEEE Trans. Comput.-
Aided Des. 16, 1507–1514 (1997)
11. Y.I. Ismail, E.G. Friedman, Effects of inductance on the propagation, delay and repeater
insertion in VLSI circuits. IEEE Trans. VLSI Sys. 8(2), 195–206 (2000)
12. Y.I. Ismail, E.G. Friedman, J.L. Neves, Equivalent Elmore delay for RLC trees. IEEE Trans.
CAD 19(1), 83–97 (2000)
13. A. Ligocka, W. Bandurski, Effect of inductance on interconnect propagation delay in VLSI
circuits. in Proceedings of 8th IEEE Workshop SPI, (9–12 May 2004), pp. 121–124
14. P.A.W. Basl, M.H. Bakr, N.K. Nikolova, Efficient transmission line modeling sensitivity
analysis exploiting rubber cells. Prog. Electromagnet. Res. PIER B 11, 223–243 (2009)
15. H. Xie, J. Wang, R. Fan, Y. Liu, Study of loss effect of transmission lines and validity of a
Spice model in electromagnetic topology. Prog. Electromagnet. Res. PIER 90, 89–103 (2009)
16. E.O. Hammerstad, O. Jensen, Accurate models for microstrip computer aided design. IEEE
Trans. MTT. 407–409 (1980)
17. E.O. Hammerstad, Equations for microstrip circuit design. in Proceedings of 5th EuMC
(Hamburg, Germany, 1–4 September 1975), pp. 268–272
18. R.B. Marks, D.F. Williams, Interconnection transmission line parameter characterization. in
Proceedings of 40th ARTG Conference on Digest (Orlando, FL, USA, December 1992), pp. 88–
95
19. R.B. Marks, D.F. Williams, Characteristic impedance determination using propagation constant
measurement. IEEE Microwave Guided Wave Lett. 1(6), 141–143 (1991)
20. W.R. Eisenstadt, Y. Eo, S-parameter-based IC interconnect transmission line characterization.
IEEE Trans. Comp. Hybrids Manuf. Technol. 15(4), 483–490 (1992)
21. A. Deutsch, R.S. Krabbenhoft, K.L. Melde, C.W. Surovic, G.A. Katopis, G.V. Kopcsay, Z.
Zhou, Z. Chen, Y.H. Kwark, T.-M. Winkel, X. Gu, T.E. Standaert, Application of the short-
pulse propagation technique for broadband characterization of PCB and other interconnect
technologies. IEEE Trans. EMC 52(2) 266–287 (February 2010)
22. J. Cong, L. He, C.-K. Koh, P. Madden, Performance optimization of VLSI interconnect.
Integration VLSI J. 21(1–2), 1–94 (1996)
23. B. Yun, S.S. Wong, Optimization of driver preemphasis for on-chip interconnects. IEEE Trans.
CAS I 56(9), 2033–2041 (2009)
24. J. Rosenfeld, E.G. Friedman, Design methodology for global resonant H-tree clock distribution
networks. IEEE Trans. VLSI Systems 15(2), 135–148 (2007)
25. F.R. Awwad, M. Nekili, V. Ramachandran, M. Sawan, On modeling of parallel repeater-
insertion methodologies for SoC interconnects. IEEE Trans. CAS I 55(1), 322–335 (2008)
26. B. Ravelo, A. Perennec, M. Le Roy, Experimental validation of the RC-interconnect effect
equalization with negative group delay active circuit in planar hybrid technology. in Proceedings
of 13th IEEE Workshop SPI (Strasbourg, France, 12–15 May 2009), pp. 1–4
27. B. Ravelo, A. Perennec, M. Le Roy, New technique of inter-chip interconnect effects equal-
ization with negative group delay active circuits. in VLSI, Chap. 20, ed. Z. Wang (INTECH,
February 2010), pp. 409–434
28. J. Zhang, T.Y. Hsiang, Extraction of subterahertz transmission-line parameters of coplanar
waveguides. Prog. Electromagnet. Res. PIERS 3(7), 1102–1106 (2007)
29. C.V. Kashyap, C.J. Alpert, F. Liu, A. Devgan, Closed-form expressions for extending step delay
and slew metrics to ramp inputs for RC trees. IEEE Trans. CADICAS I 23(4), 509–516 (2004)
30. B. Ravelo, L. Rajaoarisoa, Numerical modeling of high-speed microelectronic interconnects
for the signal integrity analysis. Int. J. Emerg. Sci. 3(1), 1–14 (2013)
31. R.A. Pucel, D.J. Massé, C. Hartwing, Losses in microstrip. IEEE Trans. MTT 16(6), 342–350
(1968)
32. M. Kirschning, R.H. Jansen, Accurate model for effective dielectric constant with validity up
to millimeter-wave frequencies. Electron. Lett. 18, 272–273 (1982)
33. D M. Pozar, Microwave Engineering, 2nd edn. (Wiley, 1988), pp. 9–21, 154–166
34. C. Chen, J. Lillis, S. Lin, N. Chang, Interconnect analysis and synthesis (Wiley, New York,
2000)
2 Basic Analysis of Single-Input Single-Output (SISO) PCB … 33
35. T. Eudes, B. Ravelo, A. Louis, Experimental validations of a simple PCB interconnect model
for high-rate signal integrity. IEEE Tran. EMC 54(2), 397–404 (2012)
36. L.P. Carloni, A.B. Kahng, S.V. Muddu, A. Pinto, K. Samadi, P. Sharma, Accurate predictive
interconnect modeling for system-level design. IEEE Tran. VLSI 18(4), 679–684 (2010)
37. T. Eudes, B. Ravelo, A. Louis, Transient response characterization of the high-speed intercon-
nection RLCG-model for the signal integrity analysis. Prog. Electromagnet. Res. 112, 183–197
(2011)
38. G.H. Zhang, M.Y. Xia, X.M. Jiang, Transient analysis of wire structures using time domain
integral equation method with exact matrix elements. Prog. Electromagnet. Res. PIER 92,
281–298 (2009)
39. Agilent EEsof EDA, Overview: Electromagnetic Design System (EMDS) (September 2008)
[Online]. Available: http://www.agilent.com/find/eesof-emds
40. Ansoft corporation, Simulation Software: High-performance Signal and Power Integrity.
Internal Report (2006)
41. ANSYS, Unparalleled Advancements in Signal- and Power-Integrity, Electromagnetic Com-
patibility Testing (16 June 2009), [Online]. Available: http://investors.ansys.com/
42. North East Systems Associates (NESA), RJ45 Interconnect Signal Integrity (CST Com-
puter Simulation Technology AG, 2010), [Online]. Available: http://www.cst.com/Content/
Applications/Article/Article.aspx?id=243
43. B. Ravelo, T. Eudes, A.K. Jastrzebski, Investigation of Reduced Models of Capacitive Loaded
Interconnects for the High-Speed SI Applications. in Proceedings of 10th International Sym-
posium on Electromagnetic Compatibility (EMC) Europe 2011 (York, UK, 26–30 September
2011), pp. 357–361
44. 240-Pin PC-6400/PC-5300/PC-4200/PC-3200 DDR2 SDRAM Registered DIMM Design
Specification. JEDEC Standard, No. 21C, Rev. 4.04, Januay 2010. [Online]. Available: http://
www.jedec.org
45. S. Lin, E.S. Kuh, Transient simulation of lossy interconnects based on the recursive convolution
formulation. IEEE Trans. CAS 39(11), 879–892 (1992)
46. X.C. Li, J.F. Mao, M. Swaminathan, Transient analsysis of C-MOS gate driven RLCG
interconnects based on FDTD. IEEE Trans. CADICS 30(4), 574–583 (2011)
47. C.N. Chiu, I.T. Chiang, A fast approach for simulating long-time response of high-speed
dispersive and lossy interconnects terminated with nonlinear loads. Prog. Electromagnet. Res.
(PIER) 91, 153–171 (2009)
Chapter 3
Discrete Periodical Model of Microstrip
Line with Cascaded Elementary L-Cells
3.1 Introduction
The mankind society and way of life become more and more dependent on electronic
equipments. This technological dependence manifests, in particular, with the strong
needs of personal computers at work, mobile phones (2.5G, 3G and 4G) of population
anytime and anywhere around the world and boom of video games [1, 2]. This
modern daily habit change creates a motivation source of microelectronic industries
to improve with no limit the technical performances of their products. According
to the ITRS road map reports [3, 4], this development can be evaluated mainly by
decrease of device feature sizes and the increase of the operating data speed. This
renders the microelectronic system interconnections more and more complex.
With several Gigasymbols/s rates of operating data in the modern microelectronic
systems [5–9], the interconnection influences can create a severe malfunctioning of
electronic systems. In fact, interconnection lines are susceptible to generate signifi-
cant delays and losses which can be a source of signal desynchronization at different
stages of the microelectronic systems as the clock tree networks [10–12]. To over-
come this problem, deep investigations on the SI propagating in the interconnection
networks have been conducted [13–18]. Moreover, methods enabling improving the
performance optimization were also proposed [19, 20]. Till now, most of the exist-
ing methods are based on first- [10, 21, 22] or second-order [23–27] polynomial
approximations of interconnection transfer functions. In order to enhance the signal
quality, an equalization method based on the use of negative group delay circuits was
also introduced [28–30]. The interconnect models [31–34] are generally considered
and implemented for the approximation of the SI parameters such as signal delays,
rise time, overshoot/undershoot and also the attenuation. But in certain use cases,
because of the increase of the operating signal bandwidth, the second-order models
of interconnection networks are not sufficient.
For starting, let us consider the periodical discrete model of TL based on the RC-
network presented in Fig. 3.1. It is formed by n-elements or segments cascaded which
are constituted by lumped circuits with total resistance:
R = Ru x, (3.1)
and capacitance:
C = Cu x, (3.2)
x = d/n. (3.3)
Logically, if the integer number of cells n is higher, the discrete model is more
accurate.
As aforementioned, one assumes that the RC-cells constituting the discrete
network schematized in Fig. 3.1 consist of lumped resistance and capacitance,
respectively, defined as R and C.
1 st-cell n th -cell
Ru Δ x Ru Δ x
vi(t) Cu Δx Cu Δ x RL
vo (t)
Fig. 3.1 Discrete RC-line formed by n lumped elements with resistance R = Ru · Δx and capacitance
C = C u · Δx [33]
3 Discrete Periodical Model of Microstrip Line with Cascaded … 37
It means that the transfer matrix governing a unique cell is given by:
1+ R·C ·s R 1 + Ru · Cu · (Δx)2 · s Ru · Δx
[M RC ] = = . (3.4)
C ·s 1 Cu · Δx · s 1
In this case, the total transfer matrix of the interconnect network shown in Fig. 3.1
should be calculated analytically from the matrix product representing the RC-cell
transfer matrices and the load resistance RL :
n 1 0
[Mn ] = [MRC ] × 1 . (3.5)
k=1
RL
1
n
M RC11 (n) M RC12 (n)
[MRC ] = . (3.7)
M RC21 (n) M RC22 (n)
k=1
After substitution of the RC-cell matrix into Eq. (3.37), we obtain the relations
between the elements of the whole matrix [Mn ] and those of the transfer matrix:
n
[Mx ] = [M RC ], (3.8)
k=1
As the circuit network under study is constituted only by RC linear passive elec-
trical components, the global transfer function must behave as a linear polynomial
expression defined as follows:
1 1
Tn (s) = = n , (3.11)
Mn11 (s) k=0 ck (n)s
k
where ck (n) (k = {1…n}) are real coefficients corresponding to the circuit network.
By delimiting to the second-order approximation, one establishes the following recur-
sive relations between the three first coefficients, c0 , c1 and c2 which are defined
as:
c12,0 (n)
c0 (n) = c11,0 (n) + , (3.12)
RL
c12,1 (n)
c1 (n) = c11,1 (n) + , (3.13)
RL
c12,2 (n)
c2 (n) = c11,2 (n) + , (3.14)
RL
where cpq,k (n) with (pq) = {(11), (12), (21), (22)} are the element coefficients of the
transfer matrix [Mx ] defined as:
n
M RC( pq) (n) = c pq,k (n)s k . (3.15)
k=0
n+1
n
[M RC ] = [M RC ] × [M RC ]. (3.16)
k=1 k=1
where n is an integer higher than 1. The initial parameters of the iterative operation
can be determined toward the identification of [M RC (1)] with the matrix expression
shown in equation:
⎧
⎪
⎪ c11,0 (1) = 1, c11,1 (1) = R · C, c11,2 (1) = 0
⎨
c12,0 (1) = R, c12,1 (1) = 0, c12,2 (1) = 0
. (3.21)
⎪
⎪ c (1) = 0, c21,1 (1) = C, c21,2 (1) = 0
⎩ 21,0
c22,0 (1) = 1, c22,1 (1) = 0, c22,2 (1) = 0
1
Tn (0) = . (3.22)
c0 (n)
One points out that at any given time t 0 higher than the system settling time t s , this
static gain can be approximated as the output voltage attenuation corresponding to
the unit step response. Toward the previous iterative operations expressed in (3.18)–
(3.21) combined with the three expressions of c0 (n), c1 (n) and c2 (n) introduced,
respectively, in (3.14), (3.15) and (3.16), one dresses the transfer function expressions
summarized in Table 3.1. It represents the simplified model of the discrete lumped
RC-network consisted of n-cells in cascade with the arbitrarily chosen values of n =
{3, 7, 10, 15, 20}.
40 B. Ravelo and L. Rajaoarisoa
7 Tn=7 (s) =
1
1+7 RR +28RC 1+2 RR s+126 1+ RR (RCs)2 +O(s 3 )
L L L
10 Tn=10 (s) =
1
1+10 RR +55RC 1+3 RR s+99 5+8 RR (RCs)2 +O(s 3 )
L L L
15 Tn=15 (s) =
1
1+15 RR +40RC 3+14 RR s+476 5+13 RR (RCs)2 +O(s 3 )
L L L
20 Tn=20 (s) =
1
1+20 RR +70RC 3+19 RR s+1463 5+18 RR (RCs)2 +O(s 3 )
L L L
To develop the model, let us consider the interconnect system comprised of TL with
characteristic impedance Z c , propagation constant γ and physical length d driven
by a voltage source vi and loaded by an impedance Z L represented in Fig. 3.2. The
voltage across the output load of this circuit is denoted vo .
In order to investigate the integrity of mixed or analog–digital signals propagating
through the interconnection system, authentic and confident knowledge about the
analytical behavior of the equivalent transfer function is indispensable. This allows
(Zc,d )
(a)
Ii(t)
(b)
1 st-cell k th-cell n th-cell
Ruδ x Luδ x Ruδ x Luδ x Ruδ x Luδ x
Interconnect line
Fig. 3.2 Interconnect line with characteristic impedance, Z c , and physical length, d, driven by a
voltage source, vi , loaded by Z L [33]
3 Discrete Periodical Model of Microstrip Line with Cascaded … 41
Vo (s) ZL
T (s) = = , (3.24)
Vi (s) Z L cosh(γ · d) + Z c sinh(γ · d)
where γ represents the propagation constant of the TL. For this reason, a charac-
terization method based on the discrete model of interconnection line constituted
by periodical lumped RLC-cells is introduced in this subsection. For the clarity, it
is organized in three different paragraphs. Paragraph 3.3.1 develops the theoretical
analysis of the periodical structure composed of lumped RLC-network. In this case,
the formulations enabling to extract the parameters R, L and C of the interconnection
lines which can be assumed as a microstrip interconnection are used. To verify the
effectiveness of the proposed theoretic concepts, validation results are presented in
Paragraph 3.3.2. Then, the last section is the conclusion of the chapter.
L = L u x, (3.25)
42 B. Ravelo and L. Rajaoarisoa
for the piece physical length Δx defined in (3.3). Ru , L u and C u are the per-unit length
parameters of the interconnect TL. Logically, higher is the integer number of cells
n, more accurate is the discrete model.
To analyze strategically the network introduced in Fig. 3.3, let us consider the period-
ical network comprised of n identical L-cells cascaded having ABCD matrix written
as:
1 + (R + L · s) · C · s R + L · s
[M R LC ] = , (3.26)
C ·s 1
where s is the Laplace variable. With the per-unit length parameters, this transfer
matrix is written as:
1 + (Ru + L u · s) · Cu · (x)2 · s (Ru + L u · s) · x
[M R LC ] = . (3.27)
Cu · x · s 1
According to the circuit and system theory, the ABCD matrix of the circuit
consisted of n-RLC-cells cascaded can be written as the matrix product:
n n
1 + (R + L · s) · C · s R + L · s
[M(n)] = [M R LC ] = . (3.28)
C ·s 1
k=1 k=1
It is important to note that the expression of [M RLC ] does not depend on the
parameter k because the elementary RLC-cells constituting the TL are identical. For
the simplification, we denote:
M11 (n) M12 (n)
[M(n)] = . (3.29)
M21 (n) M22 (n)
Interconnect line
Fig. 3.3 Proposed discrete model formed by n segments of periodical lumped RLC-cells [33]
3 Discrete Periodical Model of Microstrip Line with Cascaded … 43
So, it can be demonstrated that the recursive expressions connecting the four
elements comprising the consecutive matrices [M(n)] and [M(n + 1)] are expressed
by:
where the four elements constituting the initial matrix [M1 ] are also the ABCD matrix
of the elementary RLC-cell:
1 + (R + L · s) · C · s R + L · s
[M(1)] = . (3.34)
C ·s 1
In this case, the total ABCD matrix of the whole RLC-network shown in Fig. 3.3
should be calculated analytically from the following matrix product:
n
1 0 MT 11 (n) MT 12 (n)
[MT (n)] = [M R LC ] × = . (3.35)
1
ZL
1 MT 21 (n) MT 22 (n)
k=1
Substituting expression (3.29) into the latter matrix relation, one gets the relations
between the elements of the whole matrix [MT (n)] and those of ABCD matrix:
n
[M(n)] = [M R LC ], (3.36)
k=1
defined as:
M12 (n)
M11 (n) + M12 (n)
[MT (n)] = ZL
M22 (n) . (3.37)
M21 (n) + ZL
M12 (n)
According to the circuit and system theory, the transfer function is the inverse of the
first element of the ABCD matrix. It means that the transfer function of the system
under study can be written as:
44 B. Ravelo and L. Rajaoarisoa
1
Tn (s) = M12 (n)
, (3.38)
1
T0n (s)
+ ZL
where T0n (n) is the transfer function of the open-ended periodical RLC-network
comprised of n-elements in cascade. The recursive relation of the transfer function
of the open-ended discrete RLC-line is given by:
1
T0(n+1) (s) = . (3.39)
1+(R+L·s)·C·s
T0n (s)
+ C · s · M12 (n)
RL
Z L (s) = , (3.40)
1 + RL · C L · s
the global transfer function of the circuit introduced in Fig. 1 will become:
1
Tn+1 (s) = 1+(R+L·s)[C·s+(1+R L ·C L ·s)/R L ]
, (3.41)
Tn (s)
− (R+L·s)(1+R L ·C LR·s)[1+R
2
L ·(C+C L )·s]
M12 (n)
L
1
Tn0 (s) = n , (3.43)
k=0 ck (n)s
k
where ck (n) (k = {1…n}) are real coefficients depending on the periodical RLC-
network parameters. We can remark that this interconnect system transfer function
can be considered for determining the frequency and time domain responses of the
system in simple way compared to the relation introduced in (3.43). One can establish
the recursive relations between the element coefficients of ABCD matrix [M(n)] by
using this polynomial relation:
n
M pq (n) = c pq,k (n) · s k , (3.44)
k=0
3 Discrete Periodical Model of Microstrip Line with Cascaded … 45
where cpq,k (n) with (p, q) = {(1, 1), (1, 2), (2, 1), (2, 2)} and k = {1…n}. For example,
the two first coefficients are given by the recursive formulae:
In order to confirm the relevance of this theoretic concept, in the next section, one
proposes to analyze an example of application based on the numerical experiment
of microstrip interconnection line.
RC-line
TRANSIENT
vi(t) (Ru ,Cu,d) vo(t) RL
Fig. 3.4 ADS schematic of the interconnection network under investigation comprised of RC-line
driven by square-wave voltage source and loaded by RL -resistance [33]
Figure 3.4 represents the schematic of the numerical test interconnection network
under investigation in this section. One can see that it consists of distributed RC-
network model (available in ADS library) with the per-unit length resistance Ru =
5 /mm, capacitance C u = 1 pF/mm, physical length d = 5 mm and loaded by a
resistance RL = 100 . The whole circuit is excited by a square-wave pulse voltage
source (with normalized amplitude V M = 1 V) and time duration t 0 = 100 ps or
10-Gbits/s rate. In order to take into account the realistic effects, this source was
assumed as a trapezoidal signal with rise/fall times equal to 10 ps.
Then, analyses of the interconnection parameter influencing the proposed numer-
ical modeling method are made in the next paragraphs and compared with SPICE
results.
In this subsection, the transient voltage responses of the circuit depicted in Fig. 3.5
are compared with the responses of the polynomial VTF models summarized in
Table 3.1 (by taking n = 20). Through time domain simulations run from t min = 0
to t max = 500 ps and step Δt = 0.2 ps, one gets the comparison results of the output
voltages displayed in Fig. 3.5 for different values of the load resistance RL = {20,
60, 100 }.
As displayed in Fig. 3.5, the transient responses from the reference SPICE compu-
tation are plotted in gray full line and those ones computed from the proposed discrete
RC-model are plotted in black dashed line. So, very good agreement between the
transient responses was observed. One evaluates here relative errors which are mainly
due to the numerical inaccuracy of about 2%.
RL = 100 [33]
40
20 SPICE
dsc model
0
0 5 10 15 20
Number of cells n
0
0 5 10 15 20
Number of cells n
48 B. Ravelo and L. Rajaoarisoa
p pSPICE pSPICE
delays a and attenuations 0.8
b versus number of cells
n [33]
|/T
0.6
0.4
|T -T
0.2
0
0 5 10 15 20
(b) Number of cells n
0.5
(t )
SPICE 0
0.4
(t )|/α
0.3
SPICE 0
0.2
|α (t )-α
0.1
0
0
0 5 10 15 20
Number of cells n
These numerical experiment results confirm that the established VTF expressions
of the interconnection discrete model converge to the ideal case of the distributed
circuit when n goes to infinity.
To complete the above study, in this subsection, one proposes to analyze the influences
of the tested interconnect per-unit length parameters Ru and C u to the model under
investigation.
By varying the per-unit length resistance Ru = {5, 10, 15, 20 /mm} of the inter-
connection circuit shown in Fig. 3.4 via sweep/transient co-simulations, one gets the
time domain results displayed in Fig. 3.9.
3 Discrete Periodical Model of Microstrip Line with Cascaded … 49
voltage,V
d = 5 mm and RL = 300 Ru=20Ω/mm
0.5
[33] SPICE
dsc model
0.0
0.0
Once again, a very good correlation between the SPICE computations and the
proposed numerical modeling method is evidenced. One underlines that when the
pulse time duration t 0 is higher than the settling time, one can estimate in more easy
way the attenuation as the system final value. Although, this latter can be deduced
in function of the parameters Ru , d and RL directly via formula (3.23).
Finally, it is noteworthy that the presented computation results were executed with
the scientist standard tool MATLAB. So, one finds that the elapsed computation times
during the calculation process were only of some tens of microseconds.
3.4.1.4.3 Conclusion
Figures 3.11a, b represents the circuit diagram of the interconnection system under
study. It is essentially composed of a microstrip line printed on the FR4 epoxy
substrate characterized by a relative permittivity εr = 4.4 and thickness 0.8 mm. The
3 Discrete Periodical Model of Microstrip Line with Cascaded … 51
(a)
Data source Interconnection line Load
w
vi (t) RL CL
d
(b)
1 st -cell k th-cell n th -cell
Ru Δ xLu Δ x RuΔ xLuΔ x RuΔ xLuΔ x
vi(t) Cu Δ x Cu Δ x Cu Δ x RL CL
Fig. 3.11 a Diagram of the interconnection circuit under study comprised of a microstrip line driven
by square-wave source and loaded by RC parallel impedance and b the considered equivalent circuit
[33]
It is worth noting that the microstrip line under study is supposed used in the context of
the high-speed mixed or digital–analog data. In this case, the digital data exciting the
interconnect line under study can be assumed as analog baseband signal delimited by
the frequency f lim inversely proportional to its rise/fall times t r defined in Eq. (2.33)
of Chap. 2.
Analytically, it means that for the microelectronic applications operating around
some Gigasymbols/s, the considerable analog bandwidth of input signals is limited
to about 6 GHz. In other words, the effectiveness of the model proposed can be
validated up to the frequency given by f lim .
So, the relevance of the proposed discrete model can be evaluated via S-parameter
simulations in baseband frequency up to 6 GHz for different values of the metalliza-
tion width w. The TL was tested with EM and circuit co-simulations. These simu-
lations consist of the global simulation of the structure shown in Fig. 3.11 by using
the full-wave EM S-parameter model of the microstrip line computed in Momentum
environment of ADS®. As consequence, by considering a discrete model comprised
of n = 10 RLC-cells in cascade, a very good agreement between the return loss S 11
and the transmission loss S 21 of the TL tested displayed in Fig. 3.12 is realized. One
can see that the TL loss is inversely proportional to the interconnect width w.
To carry out this transient analysis, the interconnection network depicted in Fig. 3.11
was excited by a square-wave pulse voltage having normalized amplitude and time
symbol duration T = 200 ps which corresponds to 5 Gigasymbols/s rate. To take into
account the practical imperfections of this high-speed signal generation, the rise/fall
time of this data source was set at 20 ps. Then, comparisons were made between the
transient responses from the EM and circuit co-simulation of the TL for w = 100 µm
presented in Fig. 3.11 and the lumped RLC-networks comprised of n = 10 RLC
segments in cascade. After transient simulations run from t min = 0 to t max = 1 ns, by
varying the load resistance RL = {100, 200, 300 } via sweep co-simulations, the
results are displayed in Fig. 3.13.
The responses from the reference SPICE computation are plotted in black dashed
line and those computed from the proposed discrete RLC-model are plotted in gray
full line. So, once again, the transient responses from the transfer function model
introduced in Sect. 3.3.4.1 are very well correlated with the EM and circuit co-
simulations of the piece of microstrip interconnect TL.
One can remark that as predicted in theory, due to the interconnection effects, the
operating signal is completely degraded. One evaluates here relative errors of about
1%, which are in fact mainly due to the numerical inaccuracy. The same numerical
investigations performed with various load capacitance values C L = {0.5, 2.5, 4.5 pF}
generate the transient results displayed in Fig. 3.14. One can see that the obtained
computation results are in very good agreement between the model proposed and
3 Discrete Periodical Model of Microstrip Line with Cascaded … 53
0 0
|S 2 1 m o d e l |, d B
|S 2 1 T L |, d B -1 -1
w=50µ m w=50µ m
-2 w=100µm -2 w=100µm
w=150µm w=150µm
w=200µm w=200µm
-3 -3
0 2 4 6 0 2 4 6
Freque ncy, G H z Freque ncy, G H z
p h a s e (S 2 1 m o d e l ), °
P h a s e (S 2 1 T L ), °
0 w=50µ m 0 w=50µ m
w=100µm w=100µm
w=150µm w=150µm
w=200µm w=200µm
-30 -30
-60 -60
0 2 4 6 0 2 4 6
Freque ncy, G H z Freque ncy, G H z
0 0
|S 1 1 m o d e l |, d B
|S 1 1 T L |, d B
-10 -10
w=50µ m w=50µ m
-20 w=100µm -20 w=100µm
w=150µm w=150µm
w=200µm w=200µm
-30 -30
0 2 4 6 0 2 4 6
Freque ncy, G H z Freque ncy, G H z
Fig. 3.12 Comparison of the proposed model S-parameters and those of the TL depicted in Fig. 3.11
for w = {50, 100, 150, 200 µm} [33]
-1 model
TL
-2
0.0 0.2 0.4 0.6 0.8 1.0
Time, ns
the EM and circuit SPICE simulations. It is noteworthy that with the circuit under
study, the quality of the analog–digital signal propagating through the interconnect
line is more and more degraded when C L increases.
54 B. Ravelo and L. Rajaoarisoa
Voltage, V
1 CL=4.5pF
[33]
0
model
TL
-1
0.0 0.2 0.4 0.6 0.8 1.0
Time, ns
According to this numerical test, we point out that the main novelty of the modeling
method presented in this chapter is based on its flexibility to operate in frequency and
time domains. It is shown with this realistic structure composed of 3D interconnect
that compared to the full-wave simulators as Momentum-ADS [34–37], the model
developed is simpler and can be executed with ten times less computation time.
To achieve more concluding numerical experiments about the accuracy of the pro-
posed model, relative error analysis about the transient voltage responses in function
of the proposed model segment number for the various values of w is proposed in this
paragraph. For that, unit step source data with rise time of about 10 ps was injected in
the circuit under test. By taking arbitrary values of the load impedance (RL = 100
and C L = 1 pF), and for the physical length d = 5 mm, the results plotted in Fig. 3.15
are obtained. It is worth noting that the transient responses from the proposed model
w=150µm
w=200µm
20
1 2 3 4 5 6 7 8 9 10
Number of cells, n
3 Discrete Periodical Model of Microstrip Line with Cascaded … 55
Relative errors, %
w = 50µm
for d varied from 0.5 mm to w = 100µm
30
20 mm [34] w = 150µm
w = 200µm
20
10
0
0 5 10 15 20
Length d, mm
converge rapidly to the reference results which is considered absolutely achieved for
infinity of cells cascaded when n = 15.
For different values of the metallization width w, the transient results present
relative errors lower than 1% when n is higher than 10. As illustrated in Fig. 3.16,
it can be underlined that the proposed model presents relative errors lower than 4%
when the TL length d is varied between 1.5 mm and 17 mm.
Moreover, one can see that the sensitivity is higher when the width w is lower. It
is interesting to note that the accuracy of the model versus d depends also on the rise
time of the operated signal data.
3.4.2.5 Conclusion
References
1. J. Ju, Wide-bandwidth video switch matrix drivers and transmission technology in consumer
electronics. IEEE Wirel. Des. Mag. 28–30 (2007)
2. J.J. Wells, Faster than fiber: the future of multi-Gb/s wireless. IEEE Microwave Mag. 104–112
(2009)
3. International Technology Roadmap for Semiconductors Update Overview. [Online] http://
www.itrs.net/Links/2016ITRS/Home2016.htm
4. A. Allan, Overall Roadmap Technology Characteristics, in ITRS Winter Conference, Makuhara,
Japan, December 2007
5. Y. Tomita, M. Kibune, J. Ogawa, W. W. Walker, H. Tamura, T. Kuroda, A 10 Gb/s receiver
with series equalizer and on-chip ISI monitor in 0.11 µm CMOS. J. Solid-State Circuits 40(4)
(2005)
6. M. Voutilainen, M. Rouvala, P. Kotiranta, T. Rauner, Multi-Gigabit serial link emissions
and mobile terminal antenna interference, in Proceedings of 13th IEEE Workshop on SPI,
Strasbourg, France, May 2009
7. W. Maichen, When digital becomes analog-interfaces in high speed test, Tutorial course, in
Proceedings of 12th IEEE Workshop on SPI, Avignon, France, May 2008
8. J. Zhang, T.Y. Hsiang, Extraction of subterahertz transmission-line parameters of coplanar
waveguides. PIERS Online 3(7), 1102–1106 (2007)
9. K.O. Kenneth, Affordable terahertz electronics. IEEE Microwave Mag. 113–116 (2009)
10. J.L. Wyatt Jr., Q. Yu, Signal delay in RC meshes, trees and lines, in Proceedings of IEEE ICAD,
pp. 15–17 (1984)
11. Y.I. Ismail, E.G. Friedman, J.L. Neves, Equivalent Elmore delay for RLC trees. IEEE Trans.
CAD 19(1), 83–97 (2000)
12. X.-C. Li, J.-F. Mao, M. Tang, High-speed clock tree simulation method based on moment
matching. PIERS 2005, Hangzhou, China, pp. 178–181, 22–26 Aug 2005
13. V. Champac, V. Avendano, J. Figueras, Built-in sensor for signal integrity faults in digital
interconnect signals. IEEE Trans. VLSI 18(2), 256–269 (2010)
14. A. Nieuwoudt, J. Kawa, Y. Massoud, Crosstalk-induced delay, noise, and interconnect pla-
narization implications of fill metal in nanoscale process technology. IEEE Trans. VLSI 18(3),
378–391 (2010)
15. A.K. Palit, V. Meyer, K.K. Duganapalli, W. Anheier, J. Schloeffel, Test pattern generation based
on predicted signal integrity loss through reduced order interconnect model, in Proceedings of
16th Workshop Test Methods and Reliability of Circuits and Systems, pp. 84–88, March 2004
3 Discrete Periodical Model of Microstrip Line with Cascaded … 57
16. A.C. Scogna, A. Orlandi, V. Ricciuti, Signal and power integrity performances of striplines in
presence of 2D EBG planes, in Proceedings of 12th IEEE Workshop on SPI, Avignon, France,
May 2008
17. A. Deutsch, High-speed signal propagation on lossy transmission lines. IBM J. Res. Develop.
34(4), 601–615 (1990)
18. F. Schnieder, W. Heinrich, Model of thin-film microstrip line for circuit design. IEEE Trans.
MTT 49(1), 104–110 (2001)
19. J. Cong, L. He, C.K. Koh, P.H. Madden, Performance optimization of VLSI interconnect layout.
Integr. VLSI J. 21(1–2), 1–94 (1996)
20. M. Qungang, Y. Yintang, L. Yuejin, J. Xinzhang, Optimal cascade lumped model of deep
submicron on-chip interconnect with distributed parameters. Microelectron. Eng. 77, 310–318
(2005)
21. W.C. Elmore, The transient response of damped linear networks. J. Appl. Phys. 19, 55–63
(1948)
22. L. Wyatt, Circuit analysis, simulation and design. North-Holland (Elsiever Science, The
Netherlands, 1978)
23. A.B. Kahng, S. Muddu, An analytical delay model of RLC interconnects. IEEE Trans. CAD
16, 1507–1514 (1997)
24. A. Ligocka, W. Bandurski, Effect of inductance on interconnect propagation delay in VLSI
circuits, in Proceedings of 8th Workshop on SPI, Heidelberg, Germany, pp. 121–124, 9–12
May 2004
25. V. Adler, E.G. Friedman, Repeater design to reduce delay and power in resistive interconnect.
IEEE Trans. CAS II, Analog Digit. Sig. Process. 54(5), 607–616 (1998)
26. Y.I. Ismail, E.G. Friedman, Effects of inductance on the propagation, delay and repeater
insertion in VLSI circuits. IEEE Trans. VLSI 8(2), 195–206 (2000)
27. B. Ravelo, A. Perennec, M. Le Roy, in New technique of inter-chip interconnect effects equaliza-
tion with negative group delay active circuits, ed by Z. Wang, INTECH Book, VLSI, Chap. 20,
February 2010, pp. 409–434
28. B. Ravelo, A. Perennec, M. Le Roy, Equalization of interconnect propagation delay with
negative group delay active circuits, in Proceedings of 11th IEEE Workshop on SPI, Genova,
Italy, pp. 15–18, May 2007
29. B. Ravelo, A. Perennec, M. Le Roy, Application of negative group delay active circuits to
reduce the 50% propagation delay of RC-line model, in Proceedings of 12th IEEE Workshop
on SPI, Avignon, France, May 2008
30. B. Ravelo, A. Perennec, M. Le Roy, Experimental validation of the RC-interconnect effect
equalization with negative group delay active circuit in planar hybrid technology, in Proceedings
of 13th IEEE Workshop on SPI, Strasbourg, France, May 2009
31. T. Eudes, B. Ravelo, A. Louis, Transient response characterization of the high-speed intercon-
nection RLCG-model for the signal integrity analysis. Prog. Electromagnet. Res. (PIER) J.
112, 183–197 (2011)
32. B. Ravelo, T. Eudes, Fast estimation of RL-loaded microelectronic interconnections delay for
the signal integrity prediction. Int. J. Numer. Model. Electron. Networks Devices Fields 25(4),
338–346 (2012)
33. B. Ravelo, Microelectronic interconnect modeling with a periodical lumped RLC-network.
Turk. J. Electr. Eng. Comput. Sci. 21:747–757 (2013)
34. Agilent EEsof EDA, Overview: Electromagnetic Design System (EMDS), (September 2008)
[Online]. Available http://www.agilent.com/find/eesof-emds
35. Ansoft corporation, Simulation software: high-performance signal and power integrity, Internal
Report (2006)
36. ANSYS, Unparalleled advancements in signal- and power-integrity, electromagnetic compat-
ibility testing (16 June 2009) [Online]. Available http://investors.ansys.com/
37. North East Systems Associates (NESA), RJ45 interconnect signal integrity (2010 CST
Computer Simulation Technology AG) [Online]. Available http://www.cst.com/Content/
Applications/Article/Article.aspx?id=243
Chapter 4
Modelling of the Signal Delay Induced
by PCB Interconnect SISO Structure
4.1 Introduction
In the late 1950s, thanks to the deep scientific and physical analyses on the semicon-
ductor materials, the mankind discovers the era of microelectronics. This industrial
period was born with the invention of the integrated circuits (ICs) by Jack S. Kilby
from Texas Instruments [1]. The impact of this invention was awarded with the
Nobel Prize in physical discipline in 2000. In the middle of the 1960s, the Intel
Co-founder Gordon Moore who is one of the pioneers in Silicon Valley [2, 3] formu-
lated an empirical law stating that the performance of ICs, including the number of
components on it, doubles every 18–24 months with the same chip price. With this
spectacular progress, the understanding of the physical effect in the components is
fundamental. Then, one of the most developed mathematical predictions enabling to
foresee the progress of the microelectronic circuits’ performance known as Moore’s
law was proposed [4].
Till now, the electronic technology develops with higher integration scale accord-
ing to this analytical prediction which manifests with the increase of the operating
frequency and the integration density [5–7]. For the better understanding in the mat-
ter of the basic functioning of this technology, a large scientific background on the
physical approach permitting an accurate analysis of the circuit and system equiva-
lent behaviours or electrical modelling, EM effects on the wave propagation, and also
the mathematical approach for the signal theory are necessary [8–13]. The synergy
of all these multiphysics fields constitutes the particularity of the SI discipline which
is one of the major steps for investigating the analog and digital high-speed systems.
Despite the spectacular technological progress of microelectronic systems, the
complexity of the IC structures including the increase of their interconnect density,
which link several million of logic gates, becomes more and more sophisticated.
T0
T (s) = , (4.1)
s2 + 2ζ · ωa · s + ωa2
with T 0 is a real constant, and ωa and ζ are, respectively, the undamped natural
angular frequency and damping ratio. But this model is valid only for the transfer
function with linear polynomial forms. So, further and deeper analysis is necessary
for the complex systems governed by typically nonlinear transfer functions as the
distributed TL. To do this, an example of technique for compensating the degradation
induced by interconnect lines is introduced in [24, 25, 28, 31]. More recently, inter-
connect effects’ equalization technique, particularly, for reducing the signal delays
[32, 33] based on the use of the negative group delay circuit whose the principle
is developed in [34, 35] has been introduced. Moreover, accurate, optimized, and
also easy to implement models enabling the prediction of these unwanted effects are
indispensable [36–39]. In this scope, new generations of design strategies and com-
mercial numerical tools for simulation and characterization of various 3D structure
geometries permitting to ensure the signal fidelity at Gbits/s speeds were recently
reported [40–44]. But, the computed results with these tools remain critical when the
designed structure like entire systems (SiP, SoC, PoP, BGA packages, etc.) presents
4 Modelling of the Signal Delay Induced by PCB Interconnect SISO … 61
many levels of integration. Nowadays, easier and more relevant modelling method
of electronic circuit interconnect network is still needed.
For this reason, a modelling method based on the distributed interconnection
network is developed in this chapter. For the clearance of readability, this chapter is
structured in three sections. Section 4.2 describes the fundamental approach enabling
to determine the transfer function of the distributed electronic network interconnec-
tion system. Then, basic matrix theory applied to the generalized periodical lumped
circuit constituting the assumed interconnect system will be proposed. In addition,
a mathematical analysis of the established model unit-step response enabling to the
calculation of the SI parameters as the propagation delay, rise/fall times, settling
time, and even the attenuation will be carried out. Hence, validations through com-
putations with the reference tool SPICE environment will be offered in Sect. 4.3.
Lastly, Sect. 4.4 is the conclusion.
(Zc,d)
Ii(t)
Rs
vi(t) interconnect line vo(t) RL CL
Fig. 4.1 Interconnect line with characteristic impedance, Z c , and physical length, d, driven by a
voltage source, vi (t), with internal impedance, Rs , loaded by RL C L parallel network
62 B. Ravelo and T. Eudes
By using the interconnect input impedance according to Ohm’s law, the input current
I i (s) injected in the circuit shown in Fig. 4.1 should be written as:
Vi (s) Vi (s)
Ii (s) = = ZL cosh(γ ·d )+Zc sinh(γ ·d )
. (4.2)
Rs + Zin (s) Rs + Zc ZL sinh(γ ·d )+Zc cosh(γ ·d )
In order to achieve more explicit analytical development, the VTF T (s) must be
expressed in function of the basic Laplace variable s. This leads us naturally to the
examination of the transfer function regarding the distributed TL model with the
per-unit-length parameters R, L, and C.
According to the microwave and TL theories, by taking into account the implicit
resistive metallic loss, and the inductive and capacitive effects of the interconnect
line, the lossy line introduced in Fig. 4.1 can be assumed as a distributed RLC network
model schematized in Fig. 4.2. The parameters Ru , L u , and C u represent, respectively,
the per-unit-length resistance, inductance, and capacitance of the interconnect line,
and δx is an infinitesimal small physical length.
In this case, the transfer function expressed is transformed as follows:
√
d (Ru + Lu · s) · Cu · s · RL
T (s) = . (4.3)
d (Ru + Lu · s) · Cu · s RL + Rs (1 + RL · C · s) cosh(d (Ru + Lu · s) · Cu · s)
+ d 2 (Ru + Lu · s) · Cu · s(1 + RL · C · s) + Rs RL sinh(d (Ru + Lu · s) · Cu · s)
By replacing s by jω, due to the presence of the terms cosh(.) and sinh(.), the
transmittance of the structure under study can be determined. From there, we can
deduce the time-domain response through FFT. But the accuracy of the FFT and
IFFT remains critical when the bandwidth of the input signal is larger than 10 GHz.
In addition, the exploitation of the frequency response to deduce the transient charac-
teristics of the transfer system (as overshoot, rise time, time response) is impossible.
For this reason, a polynomial approach permitting the extractions of these transient
parameters is developed in the next section.
d
Rs Ruδx Luδx Ruδx Luδx
Fig. 4.2 Distributed circuit equivalent of the RLC model of the system shown in Fig. 4.1
4 Modelling of the Signal Delay Induced by PCB Interconnect SISO … 63
1
T (s) = , (4.4)
c0 + c1 · s + c2 · s2 + O(s3 )
with:
Rs + Ru · d
c0 = 1 + , (4.5)
RL
Ru · Cu 6Lu + 3Rs · Cu · Ru + R2u · Cu · d
c1 = + d 2 + (Rs · Cu + Ru · CL )d , (4.6)
2 6RL
R2u · Cu · d 2 Lu Ru · L u · d R3 · Cu · d 3
c2 = + + + u · Cu · d 2
24 2 3RL 120RL
Ru · Cu2 · d Ru · Cu · CL Lu · Cu R2 · C 2 · d 2
+ Lu · CL + Rs · d + + + u u d. (4.7)
6 2 2RL 24RL
1
T0 = , (4.8)
c2
c0
ωa = , (4.9)
c2
c1
ζ = √ . (4.10)
2 c0 · c2
According to the damping ratio value ζ , one distinguishes three different cases
of unit-step responses which enable to determine the transient response parameters
prior to the evaluation of the output SI parameters.
It is interesting to note that the limit of the model presented here depends on
two parameters. First, by denoting f max , the maximal operating frequency, and v, the
speed of the signal propagating along the interconnect line, the rise time of the input
signal which must be higher than [15] is given as follows:
0.35 2.8d
tr ≥ = . (4.11)
fmax v
64 B. Ravelo and T. Eudes
In difference with the model established in [45], here, the root of the equation
vo (t) = 0 with the direct calculation enabling achieving more accurate and precise
mathematical expressions is considered. So, in the remainder of this section, the
detailed formulae on the different SI parameters are offered.
It is well known that in this case, the under-consideration system behaves as a classical
overdamped system. Clearly, it should present an overshoot here denoted by ξ at the
time, Tξ which are, respectively, expressed as:
√−πζ
ξ =e 1−ζ 2 if ζ < 1, (4.13)
π
Tξ = . (4.14)
ωa 1 − ζ 2
3
ts±5% ≈ . (4.15)
ζ · ωa
0.5Tξ
Tp ≈ √−πζ
. (4.16)
1+e 1−ζ 2
In the present case, by approximating the output voltage vo (t) as its tangential
line passing through its unique inflection point, one demonstrates the here below
rise-time formulation:
√
ωa √ ζ arctan ζ −2 −1
tr ≈ 0.8 e 1−ζ 2 . (4.17)
T0
4 Modelling of the Signal Delay Induced by PCB Interconnect SISO … 65
4.2.3.2 Case 2: ζ = 1
In this case, the normalized output unit-step response is literally written as:
For the better understanding, one denotes t = T λ the root of the equation vo (t) = λ
(λ is real constant positive lower than 1). With this expression, this parameter can be
extracted easily by considering a singular mathematical function. The corresponding
solution can be expressed as:
2
λ·ω
LW e−1 T0 a − 1 + 1
Tλ = , (4.19)
ωa
Nevertheless, simpler formulations of the 50-% propagation delay and the rise
time are estimated as:
1
Tp ≈ , (4.21)
ωa
√
ln(9) 2
tr ≈ . (4.22)
ωa
In this case, it is obvious that T (s) behaves as an underdamped system with two real
poles:
1
ω1 = = ωa ζ + ζ 2 − 1 , (4.23)
τ1
1
ω2 = = ωa ζ − ζ 2 − 1 . (4.24)
τ2
Similar to case 1, by approximating the output vo (t) as its tangential line passing
through its unique inflection point, the propagation delay demonstrated is written as
follows:
ω2
[λ2 + 1]τ2 − [λ1 + 1]τ1 0.5(ω1 − ω2 ) ln ω1
Tp ≈ + + , (4.26)
λ2 − λ1 T0 (λ2 − λ1 ) ω2 − ω1
with
ω ω−ω
1
ω2 1 2
λ1 = , (4.27)
ω1
ω ω−ω
2
ω2 1 2
λ2 = . (4.28)
ω1
Finally, knowing that in this case of underdamped system, the whole transfer
function is formed by the product of transfer function having rise times:
τa = ln(9) · τ1 , (4.29)
and:
τb = ln(9)τ2 , (4.30)
In general way, regarding the three cases cited previously, it is clear that the
normalized unit-step response voltage attenuation α(t0 ) at the given instant time t =
t 0 is explicitly defined as:
⎧ T0
⎪
⎪ 1 − (1 + ωa t0 )e−ωa t0 if ζ = 1
⎪
⎪ ⎧ ωa2
⎪
⎪ ⎡ ⎤⎫
⎪
⎪ ⎨ ⎬
⎪
⎨ T0 1 − e−ωa t0 ⎣cosh ω t ζ 2 − 1 + ζ sinh ω t ζ 2 − 1 ⎦ if ζ > 1
a 0 a 0
α(t0 ) = ωa2 ⎩ ζ 2 −1 ⎭ . (4.32)
⎪
⎪ ⎧ ⎡ ⎛ ⎞⎤⎫
⎪
⎪ ⎨ ⎬
⎪
⎪ −ζ ωa t0 2
⎪
⎪ T0
1 − e
1−ζ
sin⎣ωa t0 1 − ζ 2 + arctan⎝ ζ ⎠⎦ if ζ < 1
⎪
⎩ ωa2 ⎩ 1−ζ 2 ⎭
4 Modelling of the Signal Delay Induced by PCB Interconnect SISO … 67
The verification results presented in this subsection are obtained from a microstrip
line with width w = 0.3 mm and various lengths d = 17 mm printed on the FR4 epoxy
substrate having relative permittivity εr = 4.4. Figure 4.3a represents the schematic
diagram of the circuit tested by considering an interconnection with complex form.
(b)
mm 16
11 mm
(c)
68 B. Ravelo and T. Eudes
At noted that this interconnect corresponds to the type of PCB interconnections for
mixed-circuit as PLL or command boards using microcontrollers [19]. Figure 4.3b,
c are respectively the 3D design of the line and the photograph of the fabricated
prototype. Then, EM full-wave simulations of the 3D structure represented by the
black box of Fig. 4.3a were performed with the ADS® simulation in the EDMS®
environment.
(a)
10
Vo/Vi magnitude (dB)
R s =10 Ω Simulation
R s =30 Ω Meaurement
R s =50 Ω Model
0
-10
0.0 0.7 1.4
Frequency (GHz)
(b)
Vi
1.0 V omeasurement
Voltage (V)
V omodel
0.5
0.0
0 2 4 6
Time (ns)
Fig. 4.4 Magnitude responses of the structure tested (a) and time-domain responses (b) [48]
4 Modelling of the Signal Delay Induced by PCB Interconnect SISO … 69
shown in Fig. 4.3a. Therefore, T pmeasured ≈ 166 ps and T pmodel ≈ 159 ps meanwhile
relative error lower than 5% are achieved.
We underline that for the considered metal thickness t = 35 µm, the skin depth
is of about δ Cu (f = 1 GHz) ≈ 2.08 µm instead of δ Al (f = 1 GHz) ≈ 2.47 µm for
the aluminium. Obviously, for higher frequencies generally more than 4 GHz and
narrower lines, the interconnect resistance R is relatively important, so the frequency-
dependent R(f ) from skin depth effects should be taken into account.
In this subsection, microstrip lines with width w = 0.3 mm, and various lengths
d = {d 1 , d 2 , d 3 } are analysed. It is noteworthy that these lengths have been cho-
sen considering real cases of typical TLs for PLL circuits of DDR2 applications as
reported in [19]. According to formulations (4.10)–(4.12), we can calculate the aver-
age value of the per-unit-length parameters of these TLs up to their specific frequency
bandwidth f max (i), i = {1, 2, 3}. So, we get the results summarized in Table 4.1 for
different lengths of the line d i . During these simulations, rise times of the presented
signals are kept higher than 340 ps.
From these parameters, we compare the 50% propagation delays with the pro-
posed modelling method and the results from the standard electronic simulation tool
SPICE. For that we take the source and load parameters, respectively, as Rs = 10 ,
and L L = 1 nH. By varying the RL value, one gets the 50% propagation delays plotted
in Figs. 4.5. It is interesting to note that in each case, the symbol durations T s (i) of
the input data are assumed as a trapezoidal pulse with rise/fall times t r (i) = T s (i)/10.
These rise/fall times were set accordingly to the TL lengths d i .
We point out that the computation time duration with the proposed model was
lower than some ms by using a MATLAB program. As shown in the figures above,
we find that the computed propagation delays present very accurate values close to
SPICE simulations. We underline that the relative error over RL is higher than 5% for
the lengths d lower than 3 mm and higher than 10 mm. For this later, the error increases
considerably when RL is higher than 150 . It is interesting to note that when the load
impedance is very high, the reflections from the impedance mismatch would also
impact the 50% propagation delay. This is probably the reason for the difference
between the SPICE simulation and the polynomial approximation. Moreover, to
T (ps)
computed with SPICE model our model
50
and the proposed model for:
p
a d 1 = 4 mm, b d 2 = 6 mm,
c d 3 = 8 mm, and d d 4 = 0
0 100 200 300 400 500
10 mm for various values of
RL (Ω )
RL [48]
(a) d = 4 mm
100
SPICE
T (ps)
50 our model
p
0
0 100 200 300 400 500
RL (Ω )
(b) d = 6 mm
150
SPICE
T (ps)
100
our model
50
p
0
0 100 200 300 400 500
RL (Ω )
(c) d = 8 mm
400
SPICE
T (ps)
our model
200
p
0
50 100 150
RL (Ω )
(d) d = 10 mm
achieve more accurate delays, we must take into account frequency variations of
parameters R(f ), L(f ), and C(f ) which will be the next step of this study.
4.3.1.4 Conclusion
for interconnect lines having lengths between 4 mm and 10 mm. To achieve more
accurate, per-unit parameters versus frequency should be considered with high-order
approximation of the transfer function [20, 21]. Compared to the standard tools for
the SI analysis [24, 25], here the analysis of the structure can be performed with less
computation time and easy to implement for complex structures.
Thanks to its simplicity and its accuracy, the presented method can be useful
for the estimation of the complex structure as the tree networks constituting the
microelectronic integrated systems where the value of propagation delays needs to
be evaluated during the design process [26, 27].
Figure 4.6 depicts the design schematic of the simulated interconnection network. It
consists of the distributed RC model available in ADS library with per-unit-length
resistance Ru = 5 /mm, capacitance C u = 0.5 pF/mm, and physical length d = 5 mm,
which is loaded by RL = 100 and C L = {0.1 pF, 1 pF, 2 pF}. This interconnection
circuit is excited by a normalized square waveform pulse voltage with amplitude
V M = 1 V having internal impedance Rs = 5 and time duration t 0 = 100 ps. It
corresponds as well to a digital data source with 10 Gbit/s rate.
A-meter
Rs RC-line vo(t)
Ii(t) TRANSIENT
vi(t) (Ru,Cu,d) RL CL
Fig. 4.6 ADS schematic of the tested circuit comprised of RC line-driven by square wave source
having 10-Gbits/s rate and loaded by RL C L network [49]
72 B. Ravelo and T. Eudes
After transient simulation run from t min = 0 to t max = 300 ps and step Δt = 0.1 ps,
one gets the comparative results displayed in Fig. 4.6a for the output voltage and
shown in Fig. 4.6b for the time-dependent input current regarding the normalized
unit-step input.
The transient response from the reference SPICE computation regarding the dis-
tributed RC line model is plotted in grey full line, and those computed from the
proposed model are plotted in black dashed line. So, one observes a very good
agreement between the transient responses from the ADS model and the proposed
model responses. One evaluates here relative errors which are mainly due to the
numerical inaccuracy of lower than 5-%. In the case where Rs and C L negligible and
by varying the per-unit-length resistance Ru = {5, 10, 15 /mm} of the circuit shown
in Fig. 4.6 through ADS sweep simulation, one gets the transient results displayed in
Fig. 4.7 with the normalized unit-step input voltage. We can remark that as predicted
in theory, the 50-% propagation delay and the output attenuation increase when Ru
increases.
Furthermore, to check the effectiveness of the mathematical analysis on the tran-
sient response parameters detailed in subsection 4.4.3.1, comparisons with SPICE
computations were realized. Therefore, the results summarized in Table 4.2 are
obtained. It shows the assessments of the proposed different transient parameters
as the propagation delay, rise/fall times and also the voltage attenuation at the given
instant time t 0 (Fig. 4.8).
One points out that the results presented were calculated by implementing the
formulation established with the scientist standard tool MATLAB. The attenuation
V model(C L=2pF)
0.5 pF/mm, d = 5 mm, and 0.5 V spice (C L=2pF)
RL = 100 ) [49]
0.0
Imodel(CL=1pF)
-100
0 100 200 300
Time, ps
4 Modelling of the Signal Delay Induced by PCB Interconnect SISO … 73
can be estimated as the final value when the given pulse duration t 0 is higher than the
settling time. The steady-state final values are T (0) = {0.80, 0.67, 0.57} according
to Ru = {5, 10, 15 /mm}.
As the settling time for Ru = 5 /mm is of about t s ≈ 76 ps then lower than
the input pulse duration t 0 = 100 ps, the attenuation α(t 0 ) can be considered as
equal to T (0). As confirmed by Table 4.2, the results from the proposed model are
much closer to the SPICE computation. It means that to preserve the output voltage
amplitude as higher as possible, one should operate with digital signal having pulse
duration absolutely higher than the TL network settling time t s . In the next section,
the investigation of the RLC model by taking into account the inductive parameter
of the TL is presented.
With the same configuration as in the previous subsection, an RLC model inter-
connection network as presented earlier in Fig. 4.6 was designed and simulated.
So, the results depicted in Fig. 4.9 are realized. Similar to the former study, these
simulations were performed with the equivalent distributed network comprised of
the RLC model having per-unit-length resistance Ru = 5 /mm with various sweep
inductance values L u = {0.2, 0.4, 0.6 nH/mm} and capacitance C u = 0.5 pF/mm
and with physical length d = 5 mm. The tested RLC line is driven by a voltage
74 B. Ravelo and T. Eudes
Voltage, V
= 5 /mm, various sweep 0.5
inductances L u = {0.2, 0.4, L u=0.2nH
0.6 nH/mm}, and L u=0.4nH
capacitance C u = 0.5 pF/mm 0.0 L u=0.6nH
with physical length d =
5 mm loaded by RL = 100 0.0 0.1 0.2 0.3 0.4 0.5
[49] Time, ns
4.3.2.5 Conclusion
References
1. J.S. Kilby, Invention of the integrated circuits. IEEE Trans. Electron Devices 23, 648 (1976)
2. G.E. Moore, Cramming more components onto integrated circuits. Electron. Mag. 38(8),
114–117 (1965)
76 B. Ravelo and T. Eudes
3. G.E. Moore, Progress in digital integrated electronics, in Proceedings of the IEEE International
Electron Devices Meeting, pp. 11–13 December 1975
4. Moore’s Law. [Online]. Available http://www.intel.com/museum/archives/historydocs/
mooreslaw.htm. D.K. Sharma, B.K. Kaushik, R.K. Sharma, VLSI interconnects and their
testing: prospects and challenges ahead. J. Eng. Des. Technol. 9(1), 63–84 (2011)
5. N. Magen, A. Kolodny, U. Weiser, N. Shamir, Interconnect-power dissipation in a micro-
processor, in Proceedings of the ACM International Workshop on System Level Interconnect
Prediction, pp. 7–13, February 2004
6. X.-C. Li, J.-F. Mao, H.-F. Huang, Y. Liu, Global interconnect width and spacing optimization
for latency, bandwidth and power dissipation. IEEE Trans. Electron Devices 52(10), 2272–2279
(2005)
7. M.A. Sayed, E.Y.A. Maksoud, Interconnect synthesis in high speed digital VLSI routing. Int.
J. Open Problems Compt. Math. 2(3), 383–415 (2009)
8. M. Voutilainen, M. Rouvala, P. Kotiranta, T. Rauner, Multi-gigabit serial link emissions
and mobile terminal antenna interference, in Proceedings of 13th IEEE Workshop on SPI,
Strasbourg, France, May 2009
9. T. Sato, Y. Cao, K. Agarwal, D. Sylvester, C. Hu, Bidirectional closed-form transformation
between on-chip coupling noise waveforms and interconnect delay-change curves. IEEE Trans.
Comput. Aided Design Integrated Circ. Syst. 22(5), 560–572 (2003)
10. S. Hasan, A.-K. Palit, W. Anheier, Equivalent victim model of the coupled interconnects for
simulating crosstalk induced glitches and delays, in Proceedings of 13th IEEE Workshop on
SPI, Strasbourg, France, May 2009
11. W. Maichen, When digital becomes analog-interfaces in high speed test, Tutorial course, in
Proceedings of 12th IEEE Workshop on SPI, Avignon, France, May 2008
12. A.C. Scogna, A. Orlandi, V. Ricciuti, Signal and power integrity performances of striplines in
presence of 2D EBG planes, in Proceedings of 12th IEEE Workshop on SPI, Avignon, France,
May 2008
13. V. Champac, V. Avendano, J. Figueras, Built-in sensor for signal integrity faults in digital
interconnect signals. IEEE Tran. VLSI 18(2), 256–269 (2010)
14. T. Eudes, B. Ravelo, A. Louis, Transient response characterization of the high-speed
interconnection RLCG-model for the signal integrity analysis. PIER J. 112, 183–197 (2011)
15. B. Ravelo, T. Eudes, Fast estimation of RL-loaded microelectronic interconnections delay
for the signal integrity prediction, Int. J. Numer. Model.: Electron. Networks Devices Fields.
https://doi.org/10.1002/jnm.838
16. T. Eudes, B. Ravelo, A. Louis, Experimental validations of a simple PCB interconnect model
for high-rate signal integrity. IEEE Trans. EMC. https://doi.org/10.1109/temc.2011.2165216
17. W.C. Elmore, The transient response of damped linear networks. J. Appl. Phys. 19, 55–63
(1948)
18. L. Wyatt, Circuit analysis, simulation and design. North-Holland (Elsiever Science, The
Netherlands, 1978)
19. J.L. Wyatt Jr., Waveform bounding for fast timing analysis of MOS VLSI circuits, in
Proceedings of IEEE ISCAS, pp. 760–761 (1983)
20. J.L. Wyatt Jr., Q. Yu, Signal delay in RC meshes, trees and lines, in Proceedings of IEEE ICAD,
pp. 15–17 (1984)
21. A. Deutsch, High-speed signal propagation on lossy transmission lines. IBM J. Res. Develop.
34(4), 601–615 (1990)
22. D. Deschacht, Impact of inductance and routing orientation on timing performances of coupled
interconnect lines, in 2010 International Conference on Design & Technology of Integrated
Systems in Nanoscale Era
23. A.B. Kahng, S. Muddu, An analytical delay model of RLC interconnects. IEEE Trans. CAD
16, 1507–1514 (1997)
24. L.P.P.P. van Ginneken, Buffer placement in distributed RC-tree network for minimal Elmore
delay, in Proceedings of the IEEE International Symposium of Circuits and Systems,
pp. 865–868, May 1990
4 Modelling of the Signal Delay Induced by PCB Interconnect SISO … 77
25. V. Adler, E.G. Friedman, Repeater design to reduce delay and power in resistive interconnect.
IEEE Trans. Circuits Syst. II, Analog Digit. Sig. Process. 54(5), 607–616 (1998)
26. A. Ligocka, W. Bandurski, Effect of inductance on interconnect propagation delay in VLSI
circuits, in Proceedings of 8th Workshop on SPI, Heidelberg, Germany, pp. 121–124, 9–12
May 2004
27. A.K. Palit, V. Meyer, K.K. Duganapalli, W. Anheier, J. Schloeffel, Test pattern generation based
on predicted signal integrity loss through reduced order interconnect model, in Proceedings of
16th Workshop Test Methods and Reliability of Circuits and Systems, pp. 84–88, March 2004
28. V. Adler, E.G. Friedman, Repeater design to reduce delay and power in resistive interconnect,
IEEE Trans. Circuits Syst. II Analog Digital Sig. Process. 54(5), 607–616 (1998)
29. Y.I. Ismail, E.G. Friedman, J.L. Neves, Equivalent Elmore delay for RLC trees. IEEE Tran.
CAD 19(1), 83–97 (2000)
30. Y.I. Ismail, E.G. Friedman, Effects of inductance on the propagation, delay and repeater
insertion in VLSI circuits. IEEE Trans. VLSI 8(2), 195–206 (2000)
31. P.K. Hanumolu, G.Y. Wei, U.K. Moon, Equalizers for high speed serial links. Int. J. High Speed
Electron. Syst. 15(2), 429–458 (2005)
32. B. Ravelo, A. Perennec, M. Le Roy, New technique of inter-chip interconnect effects equaliza-
tion with negative group delay active circuits, ed by Z. Wang, INTECH Book, VLSI, Chap. 20,
pp. 409–434 (2010)
33. T. Eudes, B. Ravelo, Cancellation of delays in the high-rate interconnects with UWB NGD
active cells. Appl. Phys. Res. 3(2), 81–88 (2011)
34. B. Ravelo, Investigation on microwave negative group delay circuit. Electromagnetics 31(8),
537–549 (2011)
35. B. Ravelo, Demonstration of negative signal delay with short-duration transient pulse. Eur.
Phys. J. Appl. Phys. (EPJAP) 55(10103), 1–8 (2011)
36. A. Nieuwoudt, J. Kawa, Y. Massoud, Crosstalk-induced delay, noise, and interconnect pla-
narization implications of fill metal in nanoscale process technology. IEEE Trans. VLSI 18(3),
378–391 (2010)
37. T. Lehtonen, D. Wolpert, P. Liljeberg, J. Plosila, P. Ampadu, Self-adaptive system for addressing
permanent errors in on-chip interconnects. IEEE Trans. VLSI 18(4), 527–540 (2010)
38. D. Chen, J. Cong, Y. Fan, L. Wan, LOPASS: a low-power architectural synthesis system for
FPGAs with interconnect estimation and optimization. IEEE Trans. VLSI 18(4), 564–577
(2010)
39. L.P. Carloni, A.B. Kahng, S.V. Muddu, A. Pinto, K. Samadi, P. Sharma, Accurate predictive
interconnect modeling for system-level design. IEEE Trans. VLSI 18(4), 679–684 (2010)
40. R. Achar, Advanced modeling and simulation methodologies for signal integrity analysis, in
International Conference on Recent Advances in Microwave Theory and Applications, 2008
(MICROWAVE 2008), Jaipur, India, pp. 279–279, 21–24 Nov 2008
41. Agilent EEsof EDA, Overview: Electromagnetic Design System (EMDS), (September 2008)
[Online]. Available http://www.agilent.com/find/eesof-emds
42. Ansoft Corporation, Simulation software: high-performance signal and power integrity, Internal
Report (2006)
43. ANSYS, Unparalleled advancements in signal-and power-integrity, electromagnetic compati-
bility testing, (June 2009) [Online]. Available http://investors.ansys.com/
44. North East Systems Associates (NESA), RJ45 interconnect signal integrity, (2010 CST
Computer Simulation Technology AG.) [Online]. Available: http://www.cst.com/Content/
Applications/Article/Article.aspx?id=243
45. J.A. Davis, J.D. Meindl, Compact distributed RLC interconnect models-Part I: Single line tran-
sient, time delay and overshoot expressions. IEEE Trans. Electron Devices 47(11), 2068–2077
(2000)
46. G.H. Gonnet, D.E.G. Hare, D.J. Jeffrey, D.E. Knuth, On the Lambert W function. Adv. Comput.
Math. 5, 329–359 (1996)
47. D.J. Jeffrey, D.E.G. Hare, R.M. Corless, Unwinding the branches of the Lambert W function.
Math. Sci. 21, 1–7 (1996)
78 B. Ravelo and T. Eudes
48. B. Ravelo, T. Eudes, Fast estimation of RL-loaded microelectronic interconnections delay for
the signal integrity prediction. Int. J. Numer. Model. 25(4), 338–346 (2012)
49. B. Ravelo, Delay modelling of high-speed distributed interconnect for the signal integrity
prediction. Eur. Phys. J. Appl. Phys. (EPJAP) 57(31002), 1–8
Chapter 5
Analytical Modeling Methodology
of Single-Input Multiple-Output (SIMO)
Symmetric Tree Interconnects by Using
Lumped Element L-Cell
Blaise Ravelo
5.1 Introduction
Since the invention of the ICs by Jack S. Kilby from Texas Instruments [1], the
mankind way of life has been increasingly conditioned by the evolution of electronic
systems toward the use of personal computers and multifunction mobile gadgets.
To meet this spectacular progress, high-performance reconfigurable processors and
ultra-high-speed wired and wireless communicating systems operating up to tens of
GHz were deployed [2–9]. Due to the unceasing increase of the electronic system
integration, the modern high-speed electronic equipment meets different techno-
logical roadblocks due to the interconnect complexity [10–15]. In addition to the
investigation on the apparition of EMI and EMC, many works stating the power
loss and the interconnect delay effects, for example, in the RF/digital devices were
done [6, 16–23]. Because of the undesired interconnection perturbations, it has been
evidenced that the interconnect delays of high-speed digital IC dominate widely gate
delays [5]. During the data stream transmission, these technological issues can be
sources of signal distortions, asynchronous effects of the transmitted analog signals
and erroneous symbols. So, intensive researches were performed on the modeling
of the interconnect networks in order to predict the signal integrity (SI) [9, 11–15,
17–24]. In order to minimize the cost and energy consumption, and also the quality
of shared data and clock signals, multipath circuits play a fundamental role for the
packaged IC system design of ICs packaged in different levels this later is fundamen-
tal [25–28]. In this optic, different topologies as a typical H tree interconnect [17, 25]
were investigated. Figure 5.1 shows the implementation of the H tree structure with
four levels as a typical surface layout used in Caltech score [4] and Quicksilver’s
ACM [8].
B. Ravelo (B)
Graduate Engineering School, ESIGELEC, Sotteville les Rouen, Seine-Maritime, France
e-mail: blaise.ravelo@yahoo.fr
(a) Yn
Zn
Yk+1 Yn-1
Zk+1
Yk
Zk
Zn
Zk+1 Yn
Y2
Z2 Yk+1
N0 Z1 N1
N2
v0 Y1
Z2 Yk+1
Y2 Zk+1
Yn
Nk Zn
Zk Nk+1 Nn-1
Yk Zk+1 Nn
Yk+1 Yn-1 Zn
Yn vn
Fig. 5.1 Symmetrical tree network composed of different cells formed by Z k -series impedance
and Y k -parallel admittance (k = {1, …, n})
In order to deal with the bottleneck caused by the interconnect imperfection, inten-
sive researches for the enhancements of on-chip interconnect have been conducted
[29–36]. Different techniques enabling VLSI interconnect optimization have been
deployed. For example, with various topological approaches for wire sizing and cross
talk optimizations, signal path algorithms (Steiner tree algorithm, Greedy BST/DME
algorithm, planar clock routing) and classical models have been explored [24]. By
exploiting the moment matching of the transfer response, simulation technique of the
high-speed clock tree is presented by considering buffer insertions. But as reported
succinctly in [29], such a technique is more adapted to the lumped tree network with
few numbers of cells. Moreover, topologies of on-chip interconnects with arbitrary
numbers of levels are presented [30]. Thus, by deeming with MOS distribution net-
works, optimized computation techniques of clock tree level have been introduced
for mixed system [31]. On the other hand, based on the investigation of row and
input flit width in compiled message, a modeling method of hardware performance
analysis with Hamming product codes is presented in [32] for the improvement of
on-chip interconnect energy. Moreover, new characterization method of serial link
bus delay in mobile terminal antennas operating at multi-gigabits speed is described
5 Analytical Modeling Methodology of Single-Input … 81
Before the examination of the most important parts of this chapter which is mainly
focused on the T-tree electrical network modeling methodology, the clarification of
the preliminary fundamental theory used along the study is presented in the following
subsection.
82 B. Ravelo
Along this chapter, the topological analysis of multi- or n-level symmetrical lumped
T-tree distribution as a SIMO-type system is focused on the representation introduced
in Fig. 5.1a.
One can see that at each node Nk (k = {1, . . . , n − 1}), two identical L-cells are
connected in parallel. So, according to the voltage division rule, the input equivalent
impedance Z eq (k) seen at node N k is the half of the input impedance of the next
branch:
It means that the equivalent admittance is equal to the double of the input
admittance of the next branch:
This finding explains the electrical equivalence between the branch N 0 N n of Fig. 5.2a
as traced in gray dashed path and the SISO-type circuit depicted in Fig. 5.2b.
As highlighted in Fig. 5.3, between the consecutive planes (Pk ) and (Pk+1 ), the
piece of circuit connected in the branch (N k N k+1 ) is formed by L-cell having:
Yp = 2k Yk+1 , (5.4)
parallel admittance. For the sake of simplification, the input current and the input
impedance seen at the plane (Pk ) are, respectively, denoted by I k and Z in (k).
By considering the circuit of Fig. 5.3b, the input impedance seen at the node N n
which is located at the whole network termination is equal to:
Zin (n) = 1/ 2n−1 Yn . (5.5)
Based on the equivalent impedance calculation applied to the basic cell shown in
Fig. 5.3, it is known that the input impedance Z in (k) seen at the node N k is equal to:
5 Analytical Modeling Methodology of Single-Input … 83
(a) Yn
Zn
Yk+1 Yn-1
Zk+1
Yk
Zk
Zn
Zk+1 Yn
Y2
Z2 Yk+1
N0 Z1 N1
N2
v0 Y1
Z2 Yk+1
Y2 Zk+1
Yn
Nk Zn
Zk Nk+1 Nn-1
Yk Zk+1 Nn
Yk+1 Yn-1 Zn
Yn vn
Fig. 5.2 Reduced electrical SISO circuit equivalent to the gray dashed signal path
(Pk) (Pk+1)
Ik Nk Ik+1 Zk+1/2k
2k-1Yk Zin(k+1)
Zin(k)
Fig. 5.3 Basic cell at the kth stage of the reduced circuit shown in Fig. 5.2b
Zin (k) = 21−k /Yk // Zk+1 /2k + Zin (k + 1) . (5.6)
So that, for k = {1, …, n}, the partial input impedance can be expressed as:
2−k Zk+1 +Zin (k+1)
1+2k−1 Yk [2−k Zk+1 +Zin (k+1)]
, if k ≤ n − 1
Zin (k) = . (5.7)
1
2n−1 Yn
, if k = n
The current divider principle applied again to the piece of circuit shown in Fig. 5.4
enables to write the expression of current I k flowing through electrical branch N k −1 N k
for k = {2, …, n}. For the initial case k = 1, I 1 can be determined directly with Ohm’s
84 B. Ravelo
Cn
Ck
Rn
Rk Rn
Rk
C2 Cn
Ck
N0 R1 N1 R2
R2 N2
vin(t)
C1 Ck
C2 Cn
Rk
Rk Nk Rn
Rn Nn
Ck
vout(t)
Cn
By denoting [T k −1 ] the elementary transfer matrix of the (k − 1)th cell in the branch
N k −1 N k , similarly to the basic equation of L-cell transfer matrix expressed earlier in
(5.7), one surmises the following formulation:
1 + Z · Y Zk
Tk−1 = k−1 k k 2k−1 . (5.9)
2 Yk 1
k=n
1 + Zk · Yk 2Zk−1
k
T1,n = . (5.10)
2k−1 Yk 1
k=1
This matrix product explains that the voltage global transfer function denoted
H n (s) corresponding to this matrix must be calculated recursively from the last ele-
mentary matrix [T k,n ]. Though, the latter can be determined progressively via the
following matrix recursive relation:
[Tk ] for k = n
[Tk−1,n ] = . (5.11)
[Tk,n ] · [Tk−1 ] for k < n
For more explicit representation of literal expressions, in the next part of this
chapter, [T 1,k ] will be expressed as follows:
T11 (k) T12 (k)
T1,k = . (5.12)
T21 (k) T22 (k)
By combining former Eqs. (5.9), (5.11) and (5.12), and varying the integer k from
0 to (n − 1), the following recursive relations between the two elements T 11 and T 21
of matrices [T 1,k −1 ] and [T 1,k ] are established:
To achieve a better insight into the present theoretic approach, let us consider a
multi-level T-tree network comprised of different lumped L-cells shown in Fig. 5.4.
After the examination of this tree network configuration, we will determine the
reduced transfer function of the whole system including the load effect. Then, the
calculation formula of the high-order linear propagation delay will be elaborated.
In this case, it was introduced in Chap. 2 that for high-order structures composed
of n elementary systems in cascade as depicted in Fig. 5.4, the total propagation
delay was approximated by:
86 B. Ravelo
n
Tp = Rk · Ck . (5.15)
k=1
But this estimation presents relative errors higher than 35% compared to SPICE
simulations. This finding motivates me to develop a novel mathematical model with
higher accuracy.
For starting, Fig. 5.4 represents the tree network diagram investigated in this chapter.
It is composed of L-form lumped elements having n-levels. We can see that at each
node N k (k = {1, …, n}), two L-cells are connected in parallel. According to the
circuit theory, the input equivalent impedance at this node is the half of the input
impedance of the next branch. This finding explains the chain of electrical equivalent
circuit of the branch traced in gray path of the circuit diagram displayed in Fig. 5.4
and that one depicted in Fig. 5.5.
As highlighted in Fig. 5.6, we find that between the consecutive plans (Pk )
and (Pk+1 ), the network represented by the branch (N k N k+1 ) is formed by a series
resistance:
Rs = Rk /2k−1 , (5.16)
Cp = 2k−1 Ck . (5.17)
For the sake of simplification, the input impedance seen at the plan (Pk ) is denoted
Zk.
Fig. 5.5 Electrical equivalent network of the gray path of tree circuit diagram shown in Fig. 5.4
This subsection focus is on the modeling of the linear network introduced in Fig. 5.4.
Being given that this lumped network is a linear circuit, its transfer equation should
be governed also by a linear differential equation. This finding leads us to suppose
the elements of the ABCD matrix:
M11,n (s) M12,n (s)
[Mn (s)] =
M21,n (s) M22,n (s)
n
= MABCD,k (s) . (5.19)
k=1
n 1 + Rk · Ck · s 2Rk−1
k
= 1
k=1 2k−1 Ck ·s
1
as polynomial expressions having real coefficients. For the simplification, one adopts
that the initial matrix is defined as:
1 + R1 · C1 · s R1
[M1 (s)] = MABCD,1 (s) = . (5.20)
C1 · s 1
It means that:
So that, the first two elements of the global ABCD matrix [Mn (s)] can be written
as:
n
M11,n (s) = 1 + ak · sk , (5.23)
k=1
n
M12,n (s) = bk · sk , (5.24)
k=0
88 B. Ravelo
where ak and bk are real coefficients. For k = {1, …, n}, the above ABCD matrix of
the whole tree network can be determined with the successive matrix multiplication:
1 + Rk+1 · Ck+1 · s Rk+1
Mk+1 (s) = [Mk (s)] · 1
2k . (5.25)
2k Ck+1 ·s
1
It is interesting to recall that the physical system transfer function can be determined
from the first element of the ABCD matrix via the following expression:
1
Fn (s) = . (5.26)
M11 (s)
As illustrated in the previous subsection, the transfer function of the circuit shown
in Fig. 5.4 can be determined successively by the multiplication of elementary transfer
matrices corresponding to the constituting elementary cells. By considering the real
coefficients ak , it can be expressed as a linear function:
Vout (s) 1
Fn (s) = = n . (5.27)
Vin (s) 1 + k=1 ak (n) · sk
R1 · C1 if n = 1
a1 (n) = , (5.28)
Rn · Cn + a1 (n − 1) + 2n−1 Cn · b0 (n − 1) if n ≥ 2
0 if n = 1
a2 (n) = , (5.29)
Rn · Cn · a1 (n − 1) + a2 (n − 1) + 2n−1 Cn · b1 (n − 1) if n ≥ 2
with:
R1 if n = 1
b0 (n) = , (5.30)
Rn + b0 (n − 1) if n ≥ 2
0 if n = 1
b1 (n) = . (5.31)
21−n Rn · a1 (n − 1) + b1 (n − 1) if n ≥ 2
The propagation delay calculation process developed in this chapter is deduced from
the transfer function direct poles. An example of pole extraction is proposed in [28].
For k = {1, 2}, consider the first-order systems with voltage transfer function:
1
Fk (s) = , (5.32)
1 + τk · s
if λ ≤ 4, and if not:
ln[2λ/(λ − 1)]
Tp ≈ λ · Tp1 · . (5.36)
ln(2)
By taking:
Zk = Rk + Lk s, (5.37)
and:
Yk = Ck s, (5.38)
90 B. Ravelo
the modeling method of previous Sect. 5.2.2 can be transposed to the theoretical
analysis of the Rk L k C k -tree network distribution (k = {1, …, n}).
In this case, the SISO network exposed in Fig. 5.2b will become the lumped
Rk L k C k -networks in cascade presented in Fig. 5.7. Acting as a linear circuit, the
transfer matrix of this network should be typically governed also by linear differential
equations.
This finding leads us to suppose the elements of the transfer matrix, [T 1,k −1 ], as
polynomial expressions defined by the real coefficients denoted λ11,k and λ21,k for k
varying from 1 to (n − 1):
⎧
⎨
k
1+ λ11,k (l)sl for 1 ≤ k ≤ n − 1
T11 (k) = , (5.39)
⎩ l=1
1 + Rn · Cn · s + Ln · Cn · s for k = n
2
⎧ k
⎨
λ (l)sl for 1 ≤ k ≤ n − 1
T21 (k) = l=1 21,k . (5.40)
⎩ n−1
2 Cn · s for k = n
In this case, the whole transfer function of the circuit described in Fig. 5.7 can be
written as follows:
1
Hn (s) = n . (5.41)
1+ l=1 λ11,n (l)s
l
The real coefficients λ11,n can be calculated easily via substitutions of (5.39)
and (5.40), respectively, into the iterative relations expressed in (5.28) and (5.29)
introduced earlier. Hence, this yields the following iterative relations enabling the
determination of the first two coefficients corresponding to λ11,n−k (l) and λ21,n−k (l)
for l = {1, 2}:
Rn · Cn if k = 0 ,
λ11,n−k (1) =
λ11,n−k+1 (1) + Rn−k 21−n+k λ21,n−k+1 (1) + Cn−k , if k ≥ 1
(5.42)
Fig. 5.7 Schematic of the reduced SISO circuit equivalent to n-level RLC-tree network
5 Analytical Modeling Methodology of Single-Input … 91
⎧
⎪
⎪ L C if k = 0
⎨ n n
λ11,n−k (2) = λ11,n−k+1 (2) + Rn−k Cn−k λ11,n−k+1 (1) + 21−n+k λ21,n−k+1 (2) .
⎪
⎪
⎩ +L 2 1−n+k λ (1) + C if k ≥ 1
n−k 21,n−k+1 n−k
(5.43)
0 if k = 0
λ21,n−k (1) = , (5.44)
2 n−k−1
Cn−k + λ21,n−k+1 (1) if k ≥ 1
0 if k = 0
λ21,n−k (2) = . (5.45)
2n−k−1 Cn−k λ11,n−k+1 (1) + λ21,n−k+1 (2) if k ≥ 1
V [N (n)] 1
Hn−k (s) = = . (5.46)
V [N (k)] T11 (k)
As noted that this partial transfer function is also equal to the whole transfer
function of n-level tree distribution, H n (s) if k = 0. Hence, knowing [T 1,n ], the
global transfer function with the following relation can be deduced:
V [N (n)] 1
Hn (s) = = . (5.47)
V [N (0)] T11 (n)
Using the latter and by taking into account recursive relations (5.28) and (5.29),
literal analytical calculations were established for the four transfer functions corre-
sponding to n = {1, …, 4}. Therefore, the literal expressions of H n (s) are addressed
in Table 5.1 according to the lumped tree network formed by different Z k Y k -cells
for k = {1, …, n}.
92 B. Ravelo
As particular case of previous study, in this subsection, the T-tree interconnect net-
work with identical L-cells is investigated by taking Z k = Z and Y k = Y for all the
values of integer k. In this case, expression (5.7) of the input impedance, Z in (k) seen
at the plan (Pk ), becomes:
Zin (k+1)
21−k Z + 1+2k−1 ·Y ·Zin (k+1)
, if k ≤ n − 1
Zin (k) = . (5.48)
1
2n Y
, if k = n
Moreover, the elementary transfer matrix of the (k − 1)th cell which constitutes
the branch N k −1 N k will be simplified as:
1 + Z · Y 2k−1
Z
Tk−1 = k−1 . (5.50)
2 Y 1
It implies that the equivalent whole transfer matrix, [T 1,n ], expressed in (5.9) will
become:
5 Analytical Modeling Methodology of Single-Input … 93
k=n
1 + ZY 2k−1
Z
T1,n = . (5.51)
2k−1 Y 1
k=1
Hence, previous recursive relations (5.39) and (5.40) governing the two elements
of the [T 1,k −1 ] matrix will be expressed as follows:
Owing to these last four expressions, literal analytical calculations were realized
for the six transfer functions of n-level tree network comprised of identical L-cells
for n = {1, …, 6}. The associated analytical VTF of the L-cell-based tree structure
for these values of n is addressed in Table 5.2.
Furthermore, according to expression (5.7), the derived literal expressions of the
associated input impedance for the same number of level n are exposed in Table 5.3.
3 H3 (s) = 1
1+11ZY +7Z 2 Y 2 +Z 3 Y 3
4 H4 (s) = 1
1+26ZY +30Z 2 Y 2 +10Z 3 Y 3 +Z 4 Y 4
5 H5 (s) = 1
1+57ZY +102Z 2 Y 2 +58Z 3 Y 3 +13Z 4 Y 4 +Z 5 Y 5
6 H6 (s) = 1
1+120ZY +303Z 2 Y 2 +256Z 3 Y 3 +95Z 4 Y 4 +16Z 5 Y 5 +Z 6 Y 6
94 B. Ravelo
The modeling method of the identical RLC T-tree network can be realized from the
former subsection analysis by taking Rk = R and C k = C. So that, the two elements
of transfer matrix [T 1,n−k ], T 11 and T 21 , are supposed written as follows:
n−k
T11 (n − k) = 1 + λ11,n−k (l) · sl , (5.56)
l=1
n−k
T21 (n − k) = λ21,n−k (l)sl , (5.57)
l=1
and
By identification and turning over the k-integer values, the first two real
coefficients λ11,n−k and λ21,n−k are, respectively, expressed as:
R·C if k = 0
λ11,n−k (1) = , (5.60)
R · C + λ11,n−k+1 (1) + 21−n+k R · λ21,n−k+1 (1) if k ≥ 1
⎧
⎨L · C if k = 0
λ11,n−k (2) = R · C · λ11,n−k+1 (1) + λ11,n−k+1 (2) ,
⎩ 1−n+k
+2 R · λ21,n−k+1 (2) + L · λ21,n−k+1 (1) + L · C if k ≥ 1
(5.61)
5 Analytical Modeling Methodology of Single-Input … 95
and:
2n−1 C if k = 0
λ21,n−k (1) = , (5.62)
2n−k−1 C + λ21,n−k+1 (1) if k ≥ 1
0 if k = 0
λ21,n−k (2) = . (5.63)
2n−k−1 Cλ11,n−k+1 (1) + λ21,n−k+1 (2) if k ≥ 1
To illustrate the relevance of the literal model developed in more natural point
of view, further analysis of the RLC T-tree network time domain responses will be
made in the next subsection.
Z = R + Ls, (5.64)
and:
Y = Cs, (5.65)
For the validation of all above theoretic concepts, concrete examples of application
based on the RLC-interconnect T-tree network modeling are proposed in the next
section.
To confirm the relevance of these theoretical formulae, frequency and transient anal-
yses of a lumped RC-tree network assumed as a high-order electronic circuit are
proposed in the next section.
The results explored in this section are run in the SPICE-ADS environment of the
electrical/electronic circuit simulator in commercial provided by Agilent™. Four-
level lumped tree distribution network was first designed, analyzed and simulated.
Figure 5.8 represents the symmetric four-level RLC-tree. Similar to the practical
cases of logic gate numerical circuit input impedances, we assume that the loads Z L
are a capacitor C L . This latter is considered as a variable parameter as a capacitance
in parallel with C 4 during the calculation. The source impedance Z s is considered as
a resistance Rs = 2 . By using the transfer function expression defined in Eq. (5.27),
the responses of the following proof of concept displayed in Fig. 5.4 was calculated.
This symmetric tree proof of concept is composed of different RC-cells cascaded
with table parameters {R1 = 3 , R2 = 20 , R3 = 60 , R4 = 160 } and {C 1
= 1 pF, C 2 = 5 pF, C 3 = 5 pF, C 4 = 1.5 pF}. After the MATLAB implementation
of the mathematical model established previously in Sect. 5.2, the responses of the
circuit under study were computed both in frequency and in time domains.
Then by varying the C L value from 0 to 9 pF, we obtain the comparison results in
frequency domain with SPICE simulation plotted in Fig. 5.9. can see that the transfer
function frequency responses from the polynomial model developed in this chapter
5 Analytical Modeling Methodology of Single-Input … 97
ZL
ZL ZL ZL
C4 Output C4 C4
C4
Vout(t)
R4 R4 R4
R4 R4 ZL ZL R4
ZL R4 R4 ZL
C4 C4
R3 C3 C3 R3
C4 R3 R3 C4
C3 C3
C2 R2 C1 R2 C2
R1
Zs
Vin(t)
Input
C L =2pF
a magnitude and b phase C L =5pF
C L =7pF
C L =9pF
-40
SPICE
Model
-80
0 1 2 3 4 5
Frequency (GHz)
(b) 0
phase(Vout /Vin ) (dB)
C L (SPICE)=2pF
C L (Model) =2 pF
C L (SPICE) =9pF
-100 C L (Model) =9 pF
-200
0 1 2 3 4 5
Frequency (GHz)
are well correlated to the simulations carried out with ADS® from DC to 5 GHz.
We can see that the attenuation and the phase shift increase with the frequency. This
effect is mainly the source of the numerical signal degradation caused by the tree
network.
98 B. Ravelo
Voltage (V)
0.5 C L =0pF
C L =2pF
C L =5pF
SPICE C L =7pF
0.0 Model C L =9pF
0 2 4 6 8 10
Time (ns)
5.4.1.4 Conclusion
For the verification of the developed modeling method, the routine algorithm was
implemented into MATLAB program and the results were compared with realistic
EM and circuit co-simulations with one of the standard commercial tools. Examples
of high-speed interconnect RLC T-tree networks with different levels were analyzed
in this section. The application results proposed were designed and simulated in the
SPICE and 3D EDMS environments run with electronic microwave simulators ADS
software from Agilent™.
In this subsection, validation both in frequency and in time domains with T-tree
circuits based on the lumped RLC-networks is presented. Then, predictions of the T-
tree interconnects with levels more than tens will be made also in order to demonstrate
the relevance of the expressions developed in previous section.
As depicted in Fig. 5.12, a four-level RLC T-tree distribution network was con-
sidered. To demonstrate the feasibility of the modeling method vis-à-vis the tree
network level number, n = {3, 4, 5}-level circuits are considered. Each branch of
this RLC-tree network consists of global interconnect comprised of long wires for
deep submicron technologies proposed in [15]. As an application example, the RLC
long inter-chip interconnect with per unit length parameters Ru = 76 /cm, L u =
5.3 nH/cm and C u = 2.6 pF/cm for a physical length d = 2 mm is taken in this
section.
100 B. Ravelo
Output
Vn=4(t)
C C C C C C C C
L L
L
L
L L
L
L
R R
R
R
R R
R
R
C C C C
L
L
L
L R
R
R R
C L C L C
R R
L
R
Input
V0(t)
Fig. 5.12 Schematic of the simulated RLC-tree circuit with: R = 38 , L = 2.65 nH, C = 1.3 pF
and n = 4 [54]
As depicted in Fig. 5.13, an excellent agreement between the frequency and time
domain results was realized between the models proposed from Table 5.2 computed
in MATLAB programming environment (plotted in black dashed curve) and SPICE
simulations (plotted in full gray curve). Figures 5.13 present the frequency responses
of the RLC-tree networks shown in Fig. 5.12 from DC to 14 GHz. As can be seen
here the attenuations and phase values of the isochrone transmittances:
To carry out the time domain analysis, the understudy RLC-tree networks were
excited by a periodical trapezoidal transient source with normalized amplitude, V max
= 1 V, pulse width, T w = 0.5 ns, rise/fall time t r = 25 ps and having a time duration, T
= 4T w = 2 ns. As illustrated in Fig. 5.14, the calculated responses with the proposed
model plotted in full gray lines coincide very well with SPICE transient responses
5 Analytical Modeling Methodology of Single-Input … 101
dB(Vout/Vin)
0 n= Model
lumped RC-tree network for 3
n = {3, 4, 5}: a magnitude
n=
-10
4
and b phase responses [54] n=
5
-20
-30
0.0 3.5 7.0 10.5 14.0
Frequency (GHz)
(b)
Phase(Vout/Vin) (deg)
0
n=3
-360 n=
4
ADS
n=
Model 5
-720
0.0 3.5 7.0 10.5 14.0
Frequency (GHz)
plotted in dotted black lines. As predicted intuitively in theory, one observes that the
output tree networks, vn , are more and more degraded when n is higher.
These results confirm the exactitude of the established transfer function and the
effectiveness of the proposed method for the SI analysis. Compared to the method
introduced in [29, 52], the introduced global transfer function presents technical
benefits in terms of precision and its flexibility for the high-level tree networks. So, it
can be used by the microelectronic circuit designers notably for the fast and accurate
estimation of the distortions caused by the tree network whatever its level number.
By using expression (5.57), the tree network voltage attenuations α n = vn (T )/v0
were evaluated. To check the accuracy of this formula, comparison of SPICE com-
putation is also made in Table 5.4. So, the analytical results present relative errors
lower than 5% and fit very well to SPICE computation.
In order to evidence the operability of the proposed modeling method for very high
values of tree level, n (more than 10), quantitative time-dependent analysis of the
variations of the attenuation generated by the tree network was performed. The
MATLAB implementation of the transfer function coefficients expressed in recursive
relations (5.61)–(5.67) established earlier in Sect. 5.3.2, one can evaluate easily the
temporal response parameters for very high value of n. The value of attenuation α n
in function of the number of tree level n was calculated. This enables to investigate
the influence of n, for example, when its value is widely higher than 5. To do so, by
using expression (5.67), the plot of α n (T ) versus n displayed in Fig. 5.15 is realized.
With these examples, it can be pointed out that when the RLC-level number n is
higher than 20, the output level is strongly attenuated and less than 1% of the input
amplitude. For T = 0.5 ns and n higher than 40, the output voltages are completely
negligible compared to the input one. Otherwise, it is noteworthy that this calculation
was made over the computation in order of some ms. Compared to previous studies
[28, 52], the proposed method enables to estimate the attenuations and delays in
function of level numbers which can be in order of hundreds.
0.5 ns [54]
0.6
V (T)/V
0.4
n
0.2
0
0 10 20 30 40
Tree level, n
5 Analytical Modeling Methodology of Single-Input … 103
References
1. J.S. Kilby, Invention of the integrated circuits. IEEE Trans. Electron Devices 23, 648 (1976)
2. H. Schmit, D. Whelihan, A. Tsai, M. Moe, B. Levine, R.R. Taylor, PipeRench: a virtualized pro-
grammable datapath in 0.18 micron technology, in Proceedings of the IEEE Custom Integrated
Circuits Conference (CICC), pp. 63–66, Oct 2002
3. K. Anjo, Y. Yamada, M. Koibuchi, A. Jouraku, H. Amano, Black-bus: a new data-transfer
technique using local address on networks-on-chips, in Proceedings of IEEE International
conference on Parallel and Distributed Processing Systems, Apr 2004
4. E. Capsi, M. Chu, R. Huang, J. Yeh, J. Wawrzyne, A. DeHon, Stream computations organized
for reconfigurable execution (SCORE), in Proceedings of the Field-Programmable Logic and
Applications, pp. 605–615, Sept 2000
5. http://www.itrs.net/
6. M. Ghoneima, Y. Ismail, M.M. Khellah, J. Tschanz, V. De, Serial-link bus: a low-power on-chip
bus architecture. IEEE Trans. CAS I 56(9), 2020–2032 (2009)
7. H. Veenstra, J.R. Long, in Circuit and Interconnect Design for RF and High Bit-Rate
Applications. Analog and Signal Processing (Springer, Berlin, 2008)
8. P. Master, The age of adaptive computing is here, in Proceedings of the Field-Programmable
Logic and Applications, pp. 1–3, Sept 2002
9. T. Granberg, in Handbook of Digital Techniques for High Speed Design. Prentice Hall Modern
Semiconductor Design Series (Prentice Hall, USA, 2004)
10. B. Bottom, Assembly and Packaging White Paper on System Level Integration. ITRS white
papers (2009). Available [Online]: http://www.itrs.net/papers.html
11. J.F. Buckwalter, Predicting microwave digital signal integrity. IEEE Trans. Adv. Packag. 32(2),
280–289 (2009)
12. L.P. Carloni, A.B. Kahng, S.V. Muddu, A. Pinto, K. Samadi, P. Sharma, Accurate predictive
interconnect modeling for system-level design. IEEE Trans. VLSI 18(4), 679–684 (2010)
13. G.H. Zhang, M.Y. Xia, X.M. Jiang, Transient analysis of wire structures using time domain
integral equation method with exact matrix elements. Prog. Electromagnet. Res. PIER 92,
281–298 (2009)
14. M. Celik, L. Pileggi, A. Odabasioglu, IC Interconnect Analysis (Kluwer Academic Publisher,
Dordrecht, Germany, 2002)
15. A. Deutsch, G.V. Kopcsay, P. Restle, G. Katopis, W.D. Becker, H. Smith, P.W. Coteus,
C.W. Surovic, B.J. Rubin, R.P. Dunne, T. Gallo, K.A. Jenkins, L.M. Terman, R.H. Den-
nard, G.A. Sai-Halasz, D.R. Knebel, When are transmission-line effects important for on-chip
interconnections? IEEE Trans. MTT 45, 1836–1846 (1997)
16. S. Bendhia, M. Ramdani, E. Sicard, in Electromagnetic Compatibility of Integrated Circuits
(Springer, Berlin, 2005)
17. I. Chanodia, D. Velenis, Parameter variations and crosstalk noise effects on high performance
H-Tree clock distribution networks. Analog Integr. Circ. Sig. Process. 56, 13–21 (2008)
18. A. Deutsch, R.S. Krabbenhoft, K.L. Melde, C.W. Surovic, G.A. Katopis, G.V. Kopcsay,
Z. Zhou, Z. Chen, Y.H. Kwark, T.-M. Winkel, X. Gun, T.E. Standaert, Application of the short-
pulse propagation technique for broadband characterization of PCB and other interconnect
technologies. IEEE Trans. EMC 52, 266–287 (2010)
19. S. Hasan, A.-K. Palit, W. Anheier, Equivalent victim model of the coupled interconnects for
simulating crosstalk induced glitches and delays, in Proceedings of 13th IEEE Workshop on
SPI, Strasbourg, France, May 2009
20. M.-E. Hwang, S.-O. Jung, K. Roy, Slope interconnect effort: gate-interconnect interdependent
delay modeling for early CMOS circuit simulation. IEEE Trans. CAS I 56(7), 1428–1441
(2009)
21. B. Yun, S.S. Wong, Optimization of driver preemphasis for on-chip interconnects. IEEE Trans.
CAS I 56(9), 2033–2041 (2009)
22. B. Ravelo, Delay modelling of high-speed distributed interconnect for the signal integrity
prediction. Eur. Phys. J. Appl. Phys. (EPJAP) 57(31002), 1–8 (2012)
5 Analytical Modeling Methodology of Single-Input … 105
48. R. Gupta, B. Tutuianu, L.T. Pileggi, The Elmore delay as a bound for RC trees with generalized
input signals. IEEE Trans. CAD 16(1), 95–104 (1997)
49. T. Eudes, B. Ravelo, A. Louis, Transient response characterization of the high-speed intercon-
nection RLCG-model for the signal integrity analysis. Prog. Electromagnet. J. (PIER) 112,
183–197 (2011)
50. T. Eudes, B. Ravelo, A. Louis, Experimental validations of a simple PCB interconnect model
for high-rate signal integrity. IEEE Trans. EMC 54(2), 397–404 (2012)
51. B. Ravelo, T. Eudes, Fast estimation of RL-loaded microelectronic interconnections delay for
the signal integrity prediction. Int. J. Numer. Model 25(4), 338–346 (2012)
52. P. Restle, C. Carter, J. Eckhardt, B. Krauter, B. McCredie, K. Jenkins, A. Weger, A. Mule,
The clock distribution of the Power4 microprocessor, in Digest of Technical Papers IEEE
International Solid-State Circuits Conference (ISSCC 2002), vol. 1, pp. 144–145, 3–7 Feb
2002
53. Y.I. Ismail, E.G. Friedman, J.L. Neves, Equivalent Elmore delay for RLC trees. IEEE Trans.
CAD 19(1), 83–97 (2000)
54. B. Ravelo, Behavioral model of symmetrical multi-level T-tree interconnects. Prog. Electro-
magnet. Res. (PIER) B 41, 23–50 (2012)
Chapter 6
Symmetric Tree Interconnects Modeling
with Elementary Distributed RC-Line
Blaise Ravelo
6.1 Introduction
Over the last five decades, the semiconductor technology progress manifests with
the growth of analog and digital circuit integration density and also the increase
of operating data speed. However, research works state that this development is
accompanied by braking effects due to the signaling path complexity induced by
interconnections as buses and wire interconnection systems [1–9]. For this reason,
currently, interconnect modeling plays an important role during the design process of
high-speed integrated systems. In fact, the interconnections can degrade considerably
the SI and generate frequently, the undesirable noise, clock jitter, and clock skew
phenomena [10]. In numerical area, the latter can become sources of erroneous or
inter-symbol interferences.
In order to equalize these interconnect effects, integration of buffer circuits has
been adopted by most of industrial microelectronic companies [11, 12]. But, such
solution seems limited to certain levels of attenuated and distorted signals which
should be higher than the employed buffer threshold. For this reason, another tech-
nique based on the use of NGD circuit was introduced recently [13–16]. As reported
in [16], this equalization process depends on the use of circuits exhibiting the NGD
phenomenon in wide baseband frequency range.
As introduced in [17–20], the interconnection effects can modify and can distort
considerably the signal quality propagating through wired and wireless RF devices
with physical length more than 1/10-wavelength. To control the interconnect effects,
an accurate simulator tool enabling to analyze the complete signal path, including
drivers and receivers, in both frequency- and time-domains is necessary [21–24].
For that, the design engineers use simplified models as the popular models proposed
by Elmore [25–27] and Wyatt [28]. This prominent first-order model is typically
B. Ravelo (B)
Graduate Engineering School, ESIGELEC, Sotteville les Rouen, Seine-Maritime, France
e-mail: blaise.ravelo@yahoo.fr
less execution time consuming. In fact, these models are generally well-suited to
the lumped RC-networks and susceptible to present significant inaccuracies at GHz
frequency range [29–34]. So, lumped second-order models including the inductance
effect have been deeply deployed since the late 1990s [35–39].
Recently, there was a great interest about the design of global H-tree clock dis-
tribution networks for the optimization of clock signal quality [5, 6, 40]. Though
all these results, most of studies conducted about the clock tree distribution network
focus generally, on its impact to the technology scaling. Nevertheless, few works
were made about the analytical global modeling of the high-level tree networks.
To deal with this limitation, a modeling method of symmetrical H-tree networks
composed of distributed interconnects is developed in the proposed chapter.
Recently, there was a great interest about the design of global H-tree clock distribution
networks for the optimization of clock signal quality [5, 6, 40]. Though all these
results, most of studies conducted about the clock tree distribution network focus
generally, on its impact to the technology scaling. Nevertheless, few works were
made about the analytical global modeling of the high-level tree networks. To deal
with this limitation, a modeling method of symmetrical H-tree networks composed
of distributed interconnects is developed in the proposed chapter.
To achieve more explicit analytical approach about the H-tree modeling method
proposed, let us consider the network shown in Fig. 6.1. For the better understanding,
the pieces of TL constituting the H-tree are supposed modeled by distributed RC-line
with parameters:
Rt = R · d x, (6.1)
and
Ct = C · d x, (6.2)
and dx is the infinitesimal physical length. The parameters R and C are, respectively,
the TL per-unit length resistance and capacitance. It is well-known that:
R
Z c (s) = , (6.3)
C ·s
6 Symmetric Tree Interconnects Modeling with Elementary … 109
output
(R,C,d)
(R,C,d) (R,C,d)
N2
(R ,C,d)
(R ,C,d) (R,C,d)
N 3(R ,C,d) v(N 4)=v n
vi
N4
N 0 (R ,C,d)
N1
input
(R ,C,d)
(R ,C,d) (R ,C,d)
(R ,C,d)
(R ,C,d) (R ,C,d)
(R ,C,d)
Fig. 6.1 Example of SIMO circuit: H-tree network composed of RC distributed TL for n = 4 [44]
We demonstrate that the electrical path N 0 N 4 of the typically single input multiple
output (SIMO) circuit displayed in Fig. 6.1 is equivalent to the single input single
output (SISO) circuit depicted in Fig. 6.2. This latter enables us to establish simple
expressions of the tree understudy voltage transfer function.
110 B. Ravelo
N0 N1 N2 N3 N4
Fig. 6.2 Equivalent SISO circuit of the network shown in Fig. 6.1 [44]
1
Hn (s) = . (6.5)
1 + a1 (n)s + a2 (n)s 2
By denoting:
τ = R · C · d 2, (6.6)
the iterative relations between real coefficients ak (n) and the intermediary coefficient
bk (n) for k = {1, 2} are obtained. We can consider the initial values of the transfer
function coefficients:
4 H4 (s) = 1
27 cos h 4 (γ d)−30 cos h 2 (γ d)+4
5 H5 (s) = 1
cos h(γ d)[81 cos h 4 (γ d)−108 cos h 2 (γ d)+28]
6 H6 (s) = 1
243 cos h 6 (γ d) − 378 cos h 4 (γ d)
+144 cos h 2 (γ d) − 8
7 H7 (s) = ⎧ ⎡
1
⎤
⎨ 729 cos h 6 (γ d) − 1296 cos h 4 (γ d)
cos h(γ d)⎣ ⎦
⎩
+648 cos h 2 (γ d) − 80
6 Symmetric Tree Interconnects Modeling with Elementary … 111
√
b1 (1) = τ, (6.9)
τ2 τ τ 1.5 √
a2 (n + 1) = + a1 (n) + a2 (n) + b1 (n) + 2 τ b2 (n), (6.12)
24 2 3
√
b1 (n + 1) = τ + 2b1 (n), (6.13)
It was established previously that the second order expanding allows the determi-
nations of the two lowest degree characteristic-equation coefficients. So, the expres-
sion of the signal attenuation can be estimated. By denoting vn the n-level tree network
output voltage according to the input vin , the unit step response voltage attenuation
at the instant t = T can be expressed as follows:
vn (T ) T
αn = = 1 − exp − . (6.15)
vi (T ) a1 (n)
To verify the feasibility of the method developed by considering the models of transfer
functions based on the RLC-networks, H-tree IUT was considered with n = 3 levels.
Figure 6.3 represents the layout diagram of the IUT understudy, comprised of three
levels of microstrip line with geometrical parameters width w = 0.1 µm and length
d = 3 mm.
The IUT considered in this section is printed on the dielectric substrate having
permittivity εr = 4.4 and thickness h = 1.6 mm. The metallization is based on
the conductor Cu-material with thickness t = 35 µm. We underline that design
and simulations of structures presented in this section were made in Schematic and
Momentum environments of ADS software from Agilent™. By applying the RLCG
112 B. Ravelo
Fig. 6.3 Design of the three level H-tree interconnect under test (size: 8.0 × 7.1 mm) [44]
Figure 6.4 displays the voltage transfer function frequency responses generated from
the 3D EMDS environment of ADS and the model established from DC to 6 GHz.
ADS
Fig. 6.3 simulated with ADS -10 Model 100
and from the model proposed
[44] -30 0
-50 -100
-70 -200
0 1 2 3 4 5 6
Frequency (GHz)
6 Symmetric Tree Interconnects Modeling with Elementary … 113
Voltage (V)
Fig. 6.3 simulated with ADS Vm
and from the model proposed
[44]
0
-Vm
0 1 2 3 4
Time (ns)
We can see that good correlations were realized both for n = 3 levels for dif-
ferent values of the resistance load. Resonance effect can be forecasted at about
1.35 GHz. Furthermore, this resonance effect is more and more accentuated when
RL is increased.
References
1. M.-E. Hwang, S.-O. Jung, K. Roy, Slope interconnect effort: gate-interconnect interdependent
delay modeling for early CMOS circuit simulation. IEEE Trans. CAS I 56(7), 1428–1441
(2009)
2. M. Ghoneima, Y. Ismail, M.M. Khellah, J. Tschanz, V. De, Serial-link bus: a low-power on-chip
bus architecture. IEEE Trans. CAS I 56(9), 2020–2032 (2009)
3. B. Yun, S.S. Wong, Optimization of driver preemphasis for on-chip interconnects. IEEE Trans.
CAS I 56(9), 2033–2041 (2009)
4. J. Cong, L. He, C.K. Koh, P.H. Madden, Performance optimization of VLSI interconnect layout.
Integr. VLSI J. 21(1–2), 1–94 (1996)
5. L. Hungwen, S. Chauchin, L.J. Chien-Nan, A tree-topology multiplexer for multiphase clock
system. IEEE Trans. CAS I 56(1), 124–131 (2009)
6. N. Rakuljic, I. Galton, Tree-structured DEM DACs with arbitrary numbers of levels. IEEE
Trans. CAS I 52(2), 313–322 (2010)
7. G.F. Bo, P. Ampadu, On hamming product codes with type-II hybrid ARQ for on-chip
interconnects. IEEE Trans. CAS I 56(9), 2042–2054 (2009)
8. P.P. Sotiriadis, A.P. Chandrakasan, A bus energy model for deep submicron technology. IEEE
Trans. VLSI Syst. 10(3), 341–350 (2002)
9. J.D. Meindl, Interconnect opportunities for gigascale integration. IEEE Micro. 23(3), 28–35
(2003)
10. B. Ravelo, Delay modelling of high-speed distributed interconnect for the signal integrity
prediction. Eur. Phys. J. Appl. Phys. (EPJAP) 57(31002), 1–8 (2012)
11. V.V. Deodhar, J.A. Davis, Optimal voltage scaling, repeater insertion, and wire sizing for
wave-pipelined global interconnects. IEEE Trans. CAS I 55(4), 1023–1030 (2008)
12. D. Velenis, R. Sundaresha, E.G. Friedman, Buffer sizing for delay uncertainty induced by
process variations, in Proceedings of IEEE International Conference on Electronics, CAS,
pp. 415–418, Dec 2004
6 Symmetric Tree Interconnects Modeling with Elementary … 115
37. Y.I. Ismail, E.G. Friedman, J.L. Neves, Equivalent Elmore delay for RLC trees. IEEE Trans.
CAD 19(1), 83–97 (2000)
38. A. Ligocka, W. Bandurski, Effect of inductance on interconnect propagation delay in VLSI
circuits, in Proceedings of 8th Workshop on SPI, pp. 121–124, 9–12 May 2004
39. G. Chen, E.G. Friedman, Transient response of a distributed RLC interconnect based on direct
pole extraction. J. Circuits Syst. Comput. 18(7), 1263–1285 (2009)
40. T. Eudes, B. Ravelo, Analysis of multi-gigabits signal integrity through clock H-tree. Int. J.
Circuit Theory Appl. 41(5), 535–549 (2013)
41. T. Eudes, B. Ravelo, A. Louis, Transient response characterization of the high-speed intercon-
nection RLCG-model for the signal integrity analysis. Prog. Electromagnet. J. (PIER) 112,
183–197 (2011)
42. T. Eudes, B. Ravelo, A. Louis, Experimental validations of a simple PCB interconnect model
for high-rate signal integrity. IEEE Trans. EMC 54(2), 397–404 (2012)
43. B. Ravelo, T. Eudes, Fast estimation of RL-loaded microelectronic interconnections delay for
the signal integrity prediction. Int. J. Numer. Model 25(4), 338–346 (2012)
44. B. Ravelo, A.K. Jastrzebski, Modelling of symmetrical distributed clock RC H-tree, in Proceed-
ings of 2012 International Symposium on Electromagnetic Compatibility (EMC EUROPE),
Rome, Italy, pp. 1–6, 17–21 Sept 2012
Chapter 7
Z/Y/T/S-Matrices’ Modelling
of Symmetric SIMO Structure Based
on Elementary Distributed RLC-Cell
7.1 Introduction
Over the five last decades, the semiconductor technology advance is based on the
growth of analogical and digital circuit integration density and also the increase
of operating data speed. However, research works state that this development is
accompanied by negative effects due to the signalling path complexity such as bus and
wire interconnection systems [1–9]. For this reason, interconnect modelling plays an
important role during the design process of high-speed integrated systems. Indeed,
the interconnections can degrade considerably the SI and generate frequently, the
undesirable noise, clock jitter and clock skew phenomena [10, 11]. In numerical area,
the latter can become drastically sources of erroneous or inter-symbol interferences.
In order to equalize these interconnect effects, technical solutions based on the
use of buffers have been proposed since [12–14]. But, such solution seems limited
to certain levels of attenuated and distorted signals which should be higher than the
employed buffer threshold. For this reason, another technique based on the use of
negative group delay circuit was introduced recently [15–18]. As reported in [18],
this equalization process depends on the use of circuits exhibiting the negative group
delay phenomenon in wide baseband frequency range.
According to ITRS report [19], for most of modern high-speed RF/digital hard-
ware systems, the interconnect delays dominate widely the logic gate delays. As
introduced in [20–22], the interconnection effects can modify and distort consid-
erably the signal quality propagating through wired and wireless RF devices with
physical length more than 1/10-wavelength. In digital IC operating at multi-gigabits
per second data rate, the interconnection impact presents a significant key factor for
the signal- and power- integrities [10, 11, 23–26]. To control the interconnect effects,
an accurate simulator tool enabling to analyse the complete signal path, including
Level 5
Level 5
Level 3
Level 3
Level 2
Level 1
Level 5
Level 5
Level 4 Level 4
Clock source
For starting, let us consider the TL having a characteristic impedance Z c and physical
length d as shown in Fig. 7.2. In the remainder of this chapter, this TL will be noticed
as (Z c ,d)-TL.
According to the TL theory, it is well known that the transfer matrix of this line
is written as follows:
T11 T12 cos h(γ d) Z c sin h(γ d)
[T ] = = sin h(γ d) , (7.1)
T21 T22 Zc
cos h(γ d)
where γ is the wave propagation constant. The voltage transfer function correspond-
ing to this matrix is defined as follows:
Vo (s) 1 1
H (s) = = = , (7.2)
Vi (s) T11 cos h(γ d)
Vi Zc , d Vo
Fig. 7.2 TL with Z c -characteristic impedance and d-length and with input and output voltages,
respectively, denoted vi and vo
120 T. Eudes and B. Ravelo
T11
Z in (s) = = Z c cot h(γ d). (7.3)
T21
In the next parts of this chapter, these analytical relations will be examined for
the characterization of the distributed H-tree network.
The understudy n-level distributed H-tree network is depicted in Fig. 7.3. Since the
distribution network is symmetrical, the signals vo delivered at the 2n -outputs do not
depend on input vi . One can see that the branch Nk Nk+1 (k = {1, . . . , n − 1}) is
constituted by a piece of distributed TL having characteristic impedance Z k+1 and
Input Output
Nn
Zn, dn
Nn-1
Z k+1, dk+1 Z n-1, dn-1 Zn, dn
Vo
Nk
Z2, d2 Zk , dk
Z k+1, dk+1 Z n-1, dn-1
Zn, dn
Nn-1
Zn, dn
Nn
N1
Vi N0
Z1, d1
Nn
Zn, dn
Nn-1
Zk+1, dk+1 Zn-1, dn-1 Zn, dn
Nk
Z 2, d2 Z k , dk
Zk+1, dk+1 Zn-1, dn-1
Zn, dn
Nn-1
Zn, dn
Nn
Fig. 7.3 H-tree network with n-level symmetrical branches formed by distributed piece of lines
having Z k characteristic impedance (k = {1, …, n}) and d k length [60]
7 Z/Y /T /S-Matrices’ Modelling of Symmetric SIMO Structure … 121
physical length d k+1 . By reason of symmetry, the equivalent input impedance seen in
the plan placed at the nodes N k (k = {1, …, n − 1}) is the half of the single branch
input impedance. This implies the equivalence of two (Z n , d)-TLs in parallel as a
single (2−1 Z n , d)-TL.
By applying this operation successively, one derives the simplified circuit equivalent
to any branch of Fig. 7.3 represented by the cascaded distributed circuit shown in
Fig. 7.4.
In this case, each piece of TL between the nodes N k −1 and N k (k = {1, …, n})
should present characteristic impedance equal to:
Z eq = 21−k Z k , (7.4)
For the sake of simplification, one denotes [T (k)] the overall transfer matrix
associated to k-pieces of (21−k Z k ,d)-TLs in cascade:
k
T11 (k) T12 (k)
[T (k)] = TN p−1 N p = , (7.6)
T21 (k) T22 (k)
p=1
with:
cos h(γ d1 ) Z 1 sin h(γ d1 )
[T (1)] = TN0 N1 = sin h(γ d1 ) . (7.7)
Z1
cos h(γ d1 )
N0 N1 Nk Nn-1 Nn
Fig. 7.4 Equivalent circuit of any branch of the distributed network [60]
122 T. Eudes and B. Ravelo
sin h(γ d1 )
T21 (1) = . (7.10)
Z1
Thus, one can determine iteratively the other transfer matrix via the following
matrix product:
[T (k + 1)] = [T (k)] × TNk−1 Nk . (7.11)
T12 (k + 1) = 21−k Z k T11 (k) sin h(γ dk ) + T12 (k) cos h(γ dk ). (7.13)
T22 (k + 1) = 21−k Z k T21 (k) sin h(γ dk ) + T22 (k) cos h(γ dk ). (7.15)
According to the definition of the transfer function introduced in (7.3), one derives
from [T (k + 1)], the recursive relation permitting to determine the voltage transfer
function, H k+1 (s) in function of H k (s) expressed as follows:
⎧
⎨ cos h(γ
1
d1 )
, if k = 0
Hk+1 (s) = 1
, if k ≥ 1 . (7.16)
⎩ H (k) cos h(γ dk )+
2k−1 T12 (k) sin h(γ dk )
Zk
In addition, based on formula (7.4), input impedance Z in,k (s) of k-level symmet-
rical tree networks presented in Fig. 7.2 is written as follows:
T11 (k)
Z in,k (s) = . (7.17)
T21 (k)
So, the input impedance associated to last stage transfer matrix [TNn−1 Nn ] is given
by:
This subsection presents the analysis of particular case of H-tree network investigated
in the previous one by taking Z k = Z and d k = d for k belongs in {1, …, n}.
So, the transfer matrix of k-level tree networks schematized in Fig. 7.2 can be
expressed as follows:
Z ·t12 (k)
t11 (k)
[T (k)] = 2k−1 t21 (k)
2k−1 , (7.19)
Z
t22 (k)
with:
and:
By replacing k by (k + 1), one gets the (k + 1)-level tree transfer matrix given
by:
Z ·t12 (k+1)
t11 (k + 1)
[T (k + 1)] = 2k t21 (k+1)
2k . (7.22)
Z
t22 (k + 1)
Knowing that [T (k)] and [T (k + 1)] are linked by the recursive matrix multipli-
cation:
one establishes, thus the following expressions which enable to relate t ij (k + 1)-
coefficients in function of t ij (k)-coefficients with i, j = {1, 2}:
t11 (k + 1) = t11 (k) cos h(γ d) + 2t12 (k) sin h(γ d), (7.24)
t12 (k + 1) = t11 (k) sin h(γ d) + 2t12 (k) cos h(γ d), (7.25)
1
t21 (k + 1) = t21 (k) cos h(γ d) + t22 (k) sin h(γ d), (7.26)
2
124 T. Eudes and B. Ravelo
1
t22 (k + 1) = t21 (k) sin h(γ d) + t22 (k) cos h(γ d). (7.27)
2
It yields the corresponding transfer function, H k+1 (s) associated to [T (k + 1)] written
as follows:
1
Hk+1 (s) = . (7.28)
t11 (k) cos h(γ d) + 2t12 (k) sin h(γ d)
To get more realistic effect on the proposed structure, source Z S and terminal Z L clock
impedances are taking into account, this provides the schematic sketched in Fig. 7.5
of the whole considered system. As explained previously, a n-level symmetrical H-
tree network has balanced 21−n outputs. As a result, any path of the tree can be
represented by an equivalent interconnect line with its own transfer matrix [T (n)].
Equivalent
Source interconnect line Load
ZS
Vi Zc, , d ZL Vo
Fig. 7.5 Whole system under study includes clock source and clock terminal impedances
7 Z/Y /T /S-Matrices’ Modelling of Symmetric SIMO Structure … 125
At present time, the overall transfer matrix [T T ] is given by the following relation:
where [T S ] and [T L ] are, respectively, the source and terminal impedance matrices
having subsequent expressions:
1 ZS
[TS ] = , (7.31)
01
1 0
[TL ] = 1 . (7.32)
ZL
1
For the sake of the mathematical simplification, one can assume that all terminal
impedances are similar to the same impedance denoted Z T . Thus, this implies that
the equivalent terminal impedance Z L is equal to:
ZT
Z Teq = , (7.33)
n out
Z L = 21−n Z T . (7.34)
Consequently, the whole transfer function and input impedance are obtained by
equations:
T12 (n) ZS
TT11 (n) = T11 (n) + Z S × T21 (n) + + T22 (n), (7.35)
ZL ZL
T22 (n)
TT21 (n) = T21 (n) + . (7.36)
ZL
We can recall that in the general case, the equivalent VTF between input and output
and the input impedance of the structure introduced in Fig. 7.2 can be, respectively,
written as follows:
1
HT (n) = , (7.37)
TT11 (n)
126 T. Eudes and B. Ravelo
Table 7.1 Transfer functions and input impedances of distributed H-tree networks formed by pieces
of (Z c , d)-TLs for n = {2, 3}
n Transfer function and input impedance from T T11 and T T21
⎧ ⎫
2 ⎪ ZS
⎪
⎨ 3 cos h 2
(δd) × 1 + + 3 cos h(δd) · sin h(δd)⎪
⎪
⎬
2Z L
TT11 (2) =
⎪
⎪ ZS Zc ZS ⎪
⎪
⎩× + − −2 ⎭
Zc 2Z L 2Z L
TT21 (2) = 3
cos h 2 (δd) + Z3c cos h(δd) sin h(δd) − 2Z1 L
2Z L
3 ZS ZS Zc
TT11 (3) = 9 cos h 3 (γ d) × 1 + + cos h 2 (γ d) · sin h(γ d) × +
4Z L Zc 4Z L
5Z S 2Z S Zc
− cos h(γ d) × 8 + − sin h(γ d) × +
4Z L Zc 2Z L
9 cos h 3 (γ d)−5 cos h(γ d)
TT21 (3) = Z4c sin h(γ d) × 49 cos h 2 (γ d) − 21 + 4Z L
TT11 (n)
Z in (n) = . (7.38)
TT21 (n)
By using this expression and iterative relations (7.24)–(7.27) and overall transfer
matrix relations (7.30), (7.35) and (7.36), the analytical result examples of transfer
function and input impedance formulations are summarized in Table 7.1. These
relations correspond to the tree network as shown in Fig. 7.3 for n = {2; 3}. It is
worthy of note that the transfer functions and input impedances calculated depend on
the TL propagation constant γ and the physical length d. Moreover, one can easily
assume that polynomial approximation methods of transfer functions are difficult
to implement. Therefore, analytical models ensure sufficient accuracy and allow
including coupling effects through the RLCG model.
To get more insight the proposed equations, an example of application is studied. One
assumes that the H-tree is symmetrical and has identical pieces of (Z c , d)-TLs. Thus,
the extraction of the frequency-dependent parameters γ and Z c will be straight-
forwardly obtained from geometrical and electrical properties of the elementary
TL.
7 Z/Y /T /S-Matrices’ Modelling of Symmetric SIMO Structure … 127
The symmetric tree modelling concept is related to the RLCG model of TLs, so
it is proposed to use a wide-band model which includes frequency dispersion and
frequency loss effects according to [58]. Since this model is frequency dependent, it
yields that the application limits the considered bandwidth of characterization BW3 dB
defined (2.33). It depends on the rise time t r which is provided by the application
requirements.
The more relevant about high-speed clock distribution on PCBs is the RAM clocking
for dual inline memory module (DIMM), in particular with the DDR3-SDRAM
standard. As mentioned in [59], the DDR3 PC3-17000 version with a six-layer PCB
layout has the geometrical parameters are summarized in Table 7.2, whereas electrical
parameters are reported in Table 7.3. So, the equivalent impedance at 1066 MHz of
a piece of TL is about 46 . In addition, one assumes that value of clock source is
a pure 50 resistor, while terminal impedances are considered as a parallel 60
resistor with one 4-pF capacitor.
Let us consider the elementary interconnect line of the symmetrical H-tree network
represented by its distributed RLCG model as depicted in Fig. 7.6. It is notewor-
thy that Ru , L u , C u and Gu are, respectively, per-unit length resistor, inductance,
capacitance and conductance of the interconnect line.
128 T. Eudes and B. Ravelo
Zc, , d Gu Cu Gu Cu
du
It is well known that the characteristic impedance Z c and the propagation constant
γ of the distributed interconnect line in function of the frequency are shown. In the
remainder of this chapter, the equivalent lumped parameters of the RLCG-line are
denoted by R = Ru × d, L = L u × d, G = G u × d, C = Cu × d. It is worthy of note
that these RLCG parameters can be extracted with mathematical operations yielded
from Chap. 2. As a result, the lumped element R, L, C and G depends on the frequency.
When Z c (s) and γ (s) functions are obtained, two symmetrical H-trees, respectively,
with two-levels and three-levels, were designed within the EMDS environment of
the standard ADS microwave/electronic simulation tool from Agilent™ as shown in
Fig. 7.7. So, the RLCG model in function of the frequency was implemented within
the ADS environment in order to perform co-simulation analyses with the pure
theoretical microstrip design of ADS. The signal integrity assessment is articulated
in two steps, first simulations were run in the frequency domain from DC to 8 GHz.
Then, the time-domain response was analysed with a PC3/17000 clock signal which
meets the requirements introduced in Table 7.3.
Clock terminal
Clock source
Clock terminal
Fig. 7.7 PCB Design of the 3-level symmetrical H-tree for high-speed clock distribution [60]
7 Z/Y /T /S-Matrices’ Modelling of Symmetric SIMO Structure … 129
(a) -6
ADS
-10
-12
-14
-16
0 1 2 3 4 5 6 7 8
Frequency (GHz)
(b) -10
ADS
Transfer Function (dB)
Model
-15
-20
-25
-30
0 1 2 3 4 5 6 7 8
Frequency (GHz)
Fig. 7.8 a Frequency results for the two-level symmetrical H-tree [60]. b Frequency results for the
3-level symmetrical H-tree [60]
It is important to mention that the ADS simulation does not include dispersion mod-
els, unlike the EM ADS-Momentum tool which is a full-wave simulation. Despite
this statement, one can assume that proximity effects can distort the results, which
are not the topic of this study. For this reason, simulations with the ADS SPICE
standard were only performed in order to show the importance of the high-frequency
part.
It is worthy of note that frequency dispersion and dielectric loss effects are more
heightened for the three-level H-tree than the two-levels as shown by Fig. 7.8a, b.
A perfect correlation between ADS simulation and the proposed model for the two-
level H-tree network up to 4 GHz is made, then after dispersion and dielectric loss
affect the transfer gain of less than 0.2 dB. Due to the more important length of the
130 T. Eudes and B. Ravelo
In order to focus on the impact of frequency dispersion and dielectric effect for SI
parameters, notably attenuation and time delay, different rise times were performed
as indicated by Table 7.4. So, accordingly while the frequency bandwidth is changing
with the rise time (7.36), here it yields that SI parameters of the three-level H-tree
will be changed. In this case, more the frequency bandwidth increases, the more
differences increase.
As provided by Figs. 7.9a and 7.10a, since ADS simulation and the proposed
model have very good correlation, the time-domain responses are also well correlated.
As expected, significant differences between the model and the ADS simulation for
the three-level H-tree network were found, as sketched in Figs. 7.9b and 7.10b. The
time delay is affected by 50 ps more with frequency dispersion and dielectric loss
effects for t r1 , whereas it is only affected by 20 ps for t r4 . In addition, overshoot and
undershoot effects seem to be more important for high rise times.
Vin
ADS
Model
(a)
Vclock
Voltage (V)
Vclock /2
Vin
ADS
Model
(b)
Vclock
Voltage (V)
Vclock /2
Fig. 7.9 a Time domain results for the two-level symmetrical H-tree for four different rise times
[60]. b Time domain results for the three-level symmetrical H-tree for four different rise times [60]
and dielectric loss effect that cannot be neglected for high-speed clock signals. More-
over, the RLCG model is capable to ensure crosstalk effects to improve the accuracy
of this method when dimensions become very small. As consequence, the simulation
was performed both in frequency- and time-domains to point out that the high-speed
part is responsible of SI parameters degradation. As expected in theory, it was shown
that by taking an unbuffered tree network constituted by TLs having sub-millimetre
physical length, the output signal was strongly attenuated when the tree level number
is higher.
In summary, in the present chapter, we are interested to the modelling of the
global transfer function enabling to estimate accurately and fastly, the delays and
132 T. Eudes and B. Ravelo
ADS
(a) Model
ADS
(b) Model
V max
Normalised Voltage (V)
Fig. 7.10 a Time domain results for the two-level symmetrical H-tree with a normalized voltage
[60]. b Time domain results for the three-level symmetrical H-tree with a normalized voltage [60]
attenuations. In the continuation of the present study, the effect of the interconnect
coupling as the NEXT and FEXT will be investigated. Then, impedance variations
between the branches due to the difference in return path and in vicinity to other
traces, the different types of discontinuity will be also taken into account. This con-
cept illustrates well that classical threshold buffers will not be able to compensate
propagation effects. For this reason, NGD structures can be considered in order to
reduce the number of buffers when SI parameters became deteriorated, in particular
attenuation and time delay [15–18].
One hopes that thanks to the presented modelling method, the fidelity of analogue-
numerical signal through H-tree network in the complex integrated systems can be
7 Z/Y /T /S-Matrices’ Modelling of Symmetric SIMO Structure … 133
predicted. In the continuation of this research work, one can envisage industrial
applications of the established model to preserve the clock SI in the high-speed
RF/digital systems.
References
1. M.-E. Hwang, S.-O. Jung, K. Roy, Slope interconnect effort: gate-interconnect interdependent
delay modeling for early CMOS circuit simulation. IEEE Trans. CAS I 56(7), 1428–1441
(2009)
2. M. Ghoneima, Y. Ismail, M.M. Khellah, J. Tschanz, V. De, Serial-link bus: a low-power on-chip
bus architecture. IEEE Trans. CAS I 56(9), 2020–2032 (2009)
3. B. Yun, S.S. Wong, Optimization of driver preemphasis for on-chip interconnects. IEEE Trans.
CAS I 56(9), 2033–2041 (2009)
4. J. Cong, L. He, C.K. Koh, P.H. Madden, Performance optimization of VLSI interconnect layout.
Integr. VLSI J. 21(1–2), 1–94 (1996)
5. L. Hungwen, S. Chauchin, L.J. Chien-Nan, A tree-topology multiplexer for multiphase clock
system. IEEE Trans. CAS I 56(1), 124–131 (2009)
6. N. Rakuljic, I. Galton, Tree-structured DEM DACs with arbitrary numbers of levels. IEEE
Trans. CAS I 52(2), 313–322 (2010)
7. G.F. Bo, P. Ampadu, On hamming product codes with type-II hybrid ARQ for on-chip
interconnects. IEEE Trans. CAS I 56(9), 2042–2054 (2009)
8. P.P. Sotiriadis, A.P. Chandrakasan, A bus energy model for deep submicron technology. IEEE
Trans. VLSI Syst. 10(3), 341–350 (2002)
9. J.D. Meindl, Interconnect opportunities for gigascale integration. IEEE Micro. 23(3), 28–35
(2003)
10. W.H. Ryu, M. Wang, A co-design methodology of signal integrity and power integrity, in
DesignCon, pp. 1–23 (2006)
11. D. Gizopoulos, N. Ou, T. Farahmand, A. Kuo, S. Tabatabaei, A. Ivanov, Jitter models for the
design and test of Gbps-speed serial interconnects. IEEE Des. Test Comput. 302–313 (2002)
12. Y.I. Ismail, E.G. Friedman, Effects of inductance on the propagation, delay and repeater
insertion in VLSI circuits. IEEE Trans. VLSI Syst. 8(2), 195–206 (2000)
13. V.V. Deodhar, J.A. Davis, Optimal voltage scaling, repeater insertion, and wire sizing for
wave-pipelined global interconnects. IEEE Trans. CAS I 55(4), 1023–1030 (2008)
14. D. Velenis, R. Sundaresha, E.G. Friedman, Buffer sizing for delay uncertainty induced by
process variations, in Proceedings of IEEE International Conference on Electronics, CAS,
pp. 415–418, Dec 2004
15. B. Ravelo, A. Perennec, M. Le Roy, Equalization of interconnect propagation delay with
negative group delay active circuits, in Proceedings of 11th IEEE Workshop on SPI, Genova,
Italy, pp. 15–18, May 2007
16. B. Ravelo, A. Perennec, M. Le Roy, Application of negative group delay active circuits to
reduce the 50% propagation delay of RC-line model, in Proceedings of 12th IEEE Workshop
on SPI, Avignon, France, May 2008
17. B. Ravelo, A. Perennec, M. Le Roy, Experimental validation of the RC-interconnect effect
equalization with negative group delay active circuit in planar hybrid technology, in 13th IEEE
Workshop on SPI, Strasbourg, France, May 2009
18. B. Ravelo, A. Perennec, M. Le Roy, New technique of inter-chip interconnect effects equal-
ization with negative group delay active circuits, in VLSI Intech, Chap. 20, ed. by Z.F. Wang
(2010), pp. 409–434
19. International Technology Roadmap for Semiconductors Update Overview (2009). [Online].
Available: http://www.itrs.net/
134 T. Eudes and B. Ravelo
20. J.J. Wells, Faster than fiber: the future of multi-Gb/s wireless. IEEE Microwave Mag. 104–112
(2009)
21. R.M. Henderson, K.J. Herrick, T.M. Weller, S.V. Robertson, R.T. Kihm, L.P.B. Katehi, Three-
dimensional high-frequency distribution networks—Part II: packaging and integration. IEEE
Trans. MTT 48(10), 1643–1651 (2000)
22. M. Voutilainen, M. Rouvala, P. Kotiranta, T. Rauner, Multi-gigabit serial link emissions and
mobile terminal antenna interference, in 13th IEEE Workshop on SPI, Strasbourg, France, May
2009
23. S.Y. Kim, Modeling and screening on-chip interconnect inductance, Ph.D. thesis. Stanford
University, CA, USA (2004)
24. W. Maichen, When digital becomes analog-interfaces in high speed test, in Proceedings of 12th
IEEE Workshop on SPI, Avignon, France, May 2008
25. A. Deutsch, High-speed signal propagation on lossy transmission lines. IBM J. Res. Develop.
34(4), 601–615 (1990)
26. S.P. Sim, S. Krishnan, D.M. Petranovic, N.D. Arora, K. Lee, C.Y. Yang, A unified RLC model
for high-speed on-chip interconnects. IEEE Trans. Electron Devices 50(6), 1501–1510 (2003)
27. Agilent EEsof EDA, Overview: Electromagnetic Design System (EMDS) (2008, Sept).
[Online]. Available: http://www.agilent.com/find/eesof-emds
28. Ansoft Corporation, Simulation Software: High-performance Signal and Power Integrity.
Internal Report (2006)
29. ANSYS, Unparalleled Advancements in Signal- and Power-Integrity, Electromagnetic Com-
patibility Testing (2009, June 16). [Online]. Available: http://investors.ansys.com/
30. North East Systems Associates (NESA), RJ45 Interconnect Signal Integrity (2010 CST AG).
[Online]. Available: http://www.cst.com/Content/Applications/Article/
31. W.C. Elmore, The transient response of damped linear networks. J. Appl. Phys. 19, 55–63
(1948)
32. P.K. Chan, M.D.F. Schlag, Bounds on signal delay in RC mesh networks. IEEE Trans. CAD
8, 581–589 (1989)
33. M.A. Horowitz, Timing models for MOS pass networks, in 1983 Proceedings of IEEE ISCAS,
pp. 198–201
34. L. Wyatt, Circuit Analysis, Simulation and Design (Elsevier Science, North-Holland, The
Netherlands, 1978)
35. D. Standley, J.L. Wyatt Jr., Improved signal delay bounds for RC tree networks, in VLSI Memo,
No. 86–317 (MIT, Cambridge, MAS, USA, May 1986)
36. N.K. Jain, V.C. Prasad, A.B. Bhattacharyyaa, Delay-time sensitivity in linear RC tree. IEEE
Trans. CAS 34(4), 443–445 (1987)
37. L. Vandenberghe, S. Boyd, A. El Gamal, Optimizing dominant time constant in RC circuits.
IEEE Trans. CAD 17(2), 110–125 (1998)
38. C.A. Marinov, A. Rubio, The energy bounds in RC circuits. IEEE Trans. CAS I 46(7), 869–871
(1999)
39. A.C. Deng, Y.C. Shiau, Generic linear RC delay modeling for digital CMOS circuits. IEEE
Trans. CAD 9(4), 367–376 (1990)
40. R. Gupta, B. Tutuianu, L.T. Pileggi, The Elmore delay as a bound for RC trees with generalized
input signals. IEEE Trans. CAD 16(1), 95–104 (1997)
41. M. Celik, L. Pileggi, A. Odabasioglu, IC Interconnect Analysis (Kluwer Academic Publisher,
Dordrecht, 2002)
42. A.B. Kahng, S. Muddu, An analytical delay model of RLC interconnects. IEEE Trans. CAD
16, 1507–1514 (1997)
43. Y.I. Ismail, E.G. Friedman, J.L. Neves, Figures of merit to characterize the importance of
on-chip inductance, in 1998 Proceedings of the 35th Annual ACM IEEE Design Automation
Conference, San Francisco, CA (USA), pp. 560–565
44. Y.I. Ismail, E.G. Friedman, J.L. Neves, Equivalent Elmore delay for RLC trees. IEEE Trans.
CAD 19(1), 83–97 (2000)
7 Z/Y /T /S-Matrices’ Modelling of Symmetric SIMO Structure … 135
8.1 Introduction
To meet the industrial best performance with optimal ratio of cost quality, the semi-
conductor industry innovates constantly the PCB design and implementation tech-
niques. With the rapid progress of the electronic devices, the emerging technology
with reference to the wearable and connected objects becomes a new reference [1].
To follow this trend, the flexible and organic technology is announced to be one
of the best candidates for the future electronics [2]. However, preliminary research
work is currently in progress for the further understanding on the potential of organic
flexible electronics [3]. The existing electronic products implemented on the flexible
plastic substrate are still in the maturity level corresponding to the laboratory valida-
tion [4]. Nevertheless, the prototypes of passive devices as flexible sensors printed on
polyethylene naphthalate (PEN) were recently designed and fabricated [5]. Further-
more, flexible electronics on the chapter were also proposed by using a conductive
silver ink-filled rollerball pen [6]. This approach was promised to lower the printing
technics fabrication cost. More recently, further extension has been made with the
research work on the organic light-emitting diodes and transistors [6, 7].
However, in function of the material properties and design characteristics, the SI
through the interconnect TLs notably in the high-speed systems is more and more
critical with the design complexity [8–10]. The signal delay as skew constraints and
the distortion in particular through the clock interconnection tree imperfection can-
not be neglected. Therefore, basic standards have been established [11]. Moreover,
prediction method [12, 13] usually based on the fundamental principle from the
microwave theory should be taken into account during the design of the high-speed
PCB. Thanks to the design simplicity and the relevance of the analytical model, the
microstrip interconnect is one of the most popular technologies used during the PCB
design [14, 15–18]. The analytical circuit approaches for the SI analyses with such
a structure are substantially built with the TL RLC and RLCG models [19–21]. The
current trend is to extend the model with respect to the complexity of the structure
as the consideration of the geometrical shape of the interconnect structure [20]. Fur-
thermore, higher level of complexity as the use of signal sharing tree network can
be found for the PCB composed of high-density components. Various topologies of
mesh and tree signal distribution network were developed [22–25]. The H- and T-tree
topologies are used generally for the best synchronization of the signal distributed
through multi-level interconnection [24, 25]. In the other cases as for the microelec-
tronic packaging or PCB, to connect the thousands of logic gates, the interconnect
structures are naturally more sophisticated than the one-level comb tree investigated.
Meanwhile, more relevant models are still required for higher levels of asymmetrical
tree interconnect networks as depicted in Fig. 8.1 [26].
Therefore, the prediction of the SI parameters is necessary for the PCB designer.
For example based on the high-order polynomial models [25, 27], the voltage transfer
functions through the RLC-tree network are established. Then, the calculations of the
attenuation, rise/fall time, propagation delays from each branch of the tree networks
were derived. Despite this tremendous progress, the SI of the electrical interconnects
on the flexible substrate in function of the electric property as resistive ink remains
so far, an open question.
The topology of the asymmetrical T-tree SIMO is introduced in Fig. 8.2. The pro-
posed network is represented as a SIMO network with single input and multiple
n-outputs. It is essentially comprised of distributed input access TL denoted TLin
(with characteristic impedance Z in and physical length d in ) connected to n-output
branch TLs denoted TLk with k = {1, 2, …, n}. The tree network input access is
preceded by series impedance Z s and voltage source V in . Hence, the output elemen-
tary TLs are terminated by different lumped impedances Z k . Across this output load
impedances, we have the tree output voltages V k shared at the node M k for k = {1,
…, n}.
In the rest of the chapter, for the uniformity of the mathematical notations, one
denotes:
Ru (s) + s · L u (s)
Z ck (s) = , (8.1)
G u (s) + s · Cu (s)
140 T. Eudes et al.
Fig. 8.2 Diagram of the asymmetrical distributed interconnect tree network with source series
impedance Z s and output loads Z 1 , Z2 , and Z3 [28]
γk (s) = [Ru (s) + s · L u (s)] × [G u (s) + s · Cu (s)], (8.2)
the characteristic impedances and the propagation constants of the elementary dis-
tributed TLs TLk , respectively. The TLk physical lengths are denoted d k . It is worth
noting that the electrical lengths are denoted:
for TLin .
Similar to the theoretic approach introduced in [31], the proposed behavioral model
of the asymmetrical T-tree is fundamentally built with the ABCD matrix analyses.
Indeed, according to the circuit theory, the ABCD matrix of the whole system shown
in Fig. 8.2 could be extracted by the combination of the elementary TL characteristic
matrices and the input/output load impedances. By definition, the ABCD matrices of
TLs TLk (k = {1, 2, …, n}) are written as:
8 Z/Y /T /S-Matrices’ Analysis of Non-symmetric SIMO Tree Based … 141
cosh(θk ) Z ck · sinh(θk )
[ABC D]TLk = sinh(θk ) , (8.5)
Zc
cosh(θk )
k
with:
θk = γk · dk (8.6)
is the electrical length of the TLk . The ABCD matrix of the asymmetrical T-tree
output load impedance is defined as:
1 0
[ABC D] Z k = 1 . (8.7)
Zk
1
Therefore, the equivalent ABCD matrices of the output branches TLk -Z k are
written as:
Z
cosh(θk ) + Zckk · sinh(θk ) Z ck · sinh(θk )
[ABC D]TLk −Z k = sinh(θk ) cosh(θk ) . (8.8)
Zc
+ Zk cosh(θk )
k
In order to reduce the SIMO circuit presented in Fig. 8.2 into single output, regard-
ing the electrical branch N in to N k , the other branches of the system can be assigned
as an equivalent input admittance. This admittance is in turn connected in parallel to
the output branch of interest TLk -Z k . By means of ABCD-to-Y-matrix conversion,
the total equivalent input admittance seen at the intermediate node expected by the
TL TLk can be established with the expression:
n
1 Z ξ · sinh θξ + Z cξ · cosh θξ
Yk,node = × . (8.9)
ξ =1
Z cξ Z cξ · sinh θξ + Z ξ · cosh θξ
ξ =k
The global ABCD matrix between the general input N in and the intermediate node
can be determined from the matrix product. It implies that the input voltage transfer
function of the asymmetrical tree seen at the intermediate node is expressed as:
Vnode (s) 1
=
. (8.10)
Vin (s) Zs
sinh(θ ) + cosh(θ )
Z in in in
with s is the Laplace variable. It can be underlined that in the particular case:
Z s = Z in = Z ck = Z 0 , (8.11)
142 T. Eudes et al.
Vnode (s) 1
=
. (8.12)
Vin (s) [sinh(θin ) + cosh(θin )] · 1 + nk=1 Z k sinh(θk )+Z 0 cosh(θk )
Z 0 sinh(θk )+Z k cosh(θk )
For Z s = R, the total ABCD matrix of any M in M k electrical path of the system shown
in Fig. 8.2 can be determined with the matrix relation:
1R 1 0
[ABC D] Min Mk = × [ABC D]in × × [ABC D]k . (8.13)
01 Yk,node 1
Therefore, the four elements of the total matrix are expressed as:
Z ck
[ABC D] Min Mk (2, 2) = sinh(θk )sinh(θin ) + Z ck Yk,node sinh(θk )cosh(θin )
Z in
+ cosh(θk )cosh(θin ). (8.17)
8 Z/Y /T /S-Matrices’ Analysis of Non-symmetric SIMO Tree Based … 143
This ABCD matrix is equivalent to the SISO network system corresponding to any
electrical path M in M k . To determine the voltage transfer function of this electrical
path, we can consider the five elementary ABCD matrices in cascade. The SISO
network consists of the cascaded elements: series input impedance Z s , input access
TL TLin , the shunt equivalent input impedance (of the TL TLm=k ), the output TLs
TLk , and the terminal loads Z k . It is clear that the prediction of the output voltages
V k should be easily achieved by the voltage transfer function for any input V in . For
this reason, the characterization of the asymmetrical tree stated in this chapter will
be performed via the behavioral analytical model of the asymmetrical tree for any
arbitrary signal path M in M k . Subsequently, the overall voltage transfer function can
be extracted from the first element of the ABCD matrix with the relationship:
Vk (s) 1
Tk (s) = = . (8.18)
Vin (s) [ABC D] Min Mk (1, 1)
These tree topology models are not applicable for “unbalanced structures” as 1:n
comb tree network (with 1-input and n-outputs) as illustrated in Fig. 8.3. One reminds
that in the field of microwave engineering, this kind of multi-port distributed structure
is quasi-similar to the passive 1:n power divider.
For this reason, an innovative behavioral model of the particular case of unbal-
anced known as 1:n comb tree for the performance analysis of high-speed signal
distribution is dealing within the present chapter. First, the mathematical approach is
handled to figure out the analytical principle highlighting categorically the model
under consideration. More precisely, the VTFs corresponding to the electrical
response of each path interconnecting the input with each output will be determined.
This enables us to extract the frequency- and time-domain responses of the comb
trees. To materialize concretely the concept for SI prediction, we will examine a pro-
totype of comb tree network implemented on PCB with respect to the DDR3 RAM
clock distribution standard stated in [11]. After, further discussions on the ongoing
research work in the continuation of the present study will be proposed. Finally, the
conclusion summarizing the outlet of the chapter will be drawn.
Receiver 1
TL1
ZL1 V1
Receiver 2
Source Gate
RS N0
TL2
ZL2 V2
TL0
V0
t Vin
Receiver n
TLn
ZLn Vn
Fig. 8.4 Schematic of unbalanced and non-symmetrical two-level tree network comprised of series
resistance Rs and intermediate TL TL0 as access input branch and parallel TLs TLk loaded by Z Lk
at the output branches (k = {1, …, n}) [29]
8 Z/Y /T /S-Matrices’ Analysis of Non-symmetric SIMO Tree Based … 145
Vk (s)
Hk (s) = , (8.19)
Vin (s)
Nk
YT (k) = Y11x . (8.20)
x=1
In the other hand, one represents by Y 0 (k) represents the sum of admittances
connected at N 0 with the exception of the output branch k connected to the node N 0 .
It is worth reminding that according to the TL theory, the ABCD matrix of elementary
distributed transmission line, TLx , satisfying the quasi-TEM propagation mode is
expressed as:
A x Bx cosh(γx · dx ) Z C x · sinh(γx · dx )
[TLx ] = = sinh(γx ·dx ) , (8.21)
C x Dx ZC
cosh(γx · dx )
x
where γ x and Zcx are, respectively, the propagation constant and the characteristic
impedance of the TL supposed here having physical length d x . As explored in [20],
these parameters can be extracted straightforwardly knowing the geometrical and
physical parameters of the TLs. By definition, the ABCD matrix is analog to the
lumped series resistance Rs and receiver impedance Z L k are, respectively, given by:
1 Rs
[Rs ] = , (8.22)
01
and:
1 Z Lx
Z Lx = . (8.23)
01
Driver Gate
TL0 N1 TL1 N2 Nn-2 TLn-2 Nn-1 TLn-1 Nn
Rs
1 RS 1 0 1 0
= × [TL0 ] × × [TLk ] × 1 . (8.24)
01 Y0 (k) 1 Z Lk
1
−1
N 1 0
ABC Dk+1 = [ABC Dk ] × × [TLk ], (8.25)
YT (k) 1
k=1
Afterwards, in order to find out the VTF H k at the targeted point N k of the network
for the input at N in , the equivalent admittance Y n (k) connected at the same node n
must be extracted by using Eq. (8.21). Finally, the ABCD matrix, denoted M n,k , at
the desired point k of the node n, can be obtained by:
1 0 1 0
Mn,k = [ABC Dn ] × × [TLk ] × 1 . (8.27)
Yn (k) 1 Z Lk
1
In order to check the validity of this behavioral analytical method, in the next
section, we will consider a proof of concept and proceed with comparisons of VTFs
obtained from simulations and measurement. To do this, the established behavioral
model and the workflow displayed in Fig. 8.6 will be implemented into MATLAB
programs. Then, for the given input signals V in , we will predict the desired output
waveform V out (N k ) for any values of k by means of convolution with the adequate
VTF.
148 T. Eudes et al.
Figure 8.6 represents the workflow of the proposed computational modeling of the
asymmetrical tree. It is executed in four main steps.
• In Step 1, the methodology is fundamentally built with the frequency-dependent
RLCG(f ) model of interconnect tree branches. The frequency-dependent per-unit-
length parameters (Ru (f ), L u (f ), C u (f ), Gu (f )) of the elementary transmission lines
(TLs) constituting the branches of the tree are extracted.
• Step 2 is the transformation of the RLCG parameters into the equivalent [Z]k (s)
and [Y ]k (s) matrices. Then, the SIMO model based on the topology introduced in
Fig. 8.4 is established based on the circuit and system theory.
• Then, via the Z/Y to ABCD transform, the transfer matrices corresponding to
each electrical path M in M k (k = {1, …, n}) are calculated in Step 3. The first
element of this transfer matrix which allows to determine the equivalent SISO
ABCD matrix is determined. Then, we can establish the voltage transfer function
of the asymmetrical tree.
• Step 4 will be the calculation of the transient voltage response by the mean of the
convolution with the input signal.
During the numerical tests, the presented routine algorithm was implemented as
a MATLAB program for an arbitrary asymmetrical tree as a POC.
This section is focused on the proposed behavioral model validation inspired with
the use case of passive PCBs. As POC, design of the interconnect structure will be
introduced. The main aim of the numerical tests is to monitor the influence of the
metallic ink conductivity on the signal integrity propagating along the asymmetrical
tree. The POC is constituted of the single-input and triple-output network printed on
the Kapton plastic substrate described in the next paragraph.
three outputs of the considered asymmetrical T-tree. The circuit schematic of the
considered POC is displayed in Fig. 8.7. As we can see, it is comprised of two-level
asymmetrical SIMO tree network. The T-tree was excited with source having internal
resistance R = 5 and loaded by the lumped capacitors C k = 1 pF (k = {1, 2, 3}).
At the input, access of the tree was connected to an input resistance R in downstream
of the excitation voltage source V in .
Moreover, the layout of the asymmetrical tree structure is shown in Fig. 8.8. The
elementary TL branches present the physical lengths d in = 3 cm, d 1 = 2 cm, d 2
= 7 cm, and d 3 = 15 cm. Because of the laminated metallic thin layer roughness
and impurities, the strip line conductivity or resistivity can vary considerably. So,
we propose to investigate the effect of the resistivity varied with the following test
parameters ρ = {1 µ m, 2 µ m, 3 µ m, 10 µ m}.
After the MATLAB application of the algorithm introduced in Fig. 8.6, we realize
the frequency analyses of the thin-film single input and triple outputs’ asymmetrical
tree proposed in the previous paragraph. To do this, the frequency-dependent RLCG
parameters of each branch of the interconnect lines were extracted. A microstrip
line presenting physical width w = 308 µm and length d = 14 mm was simulated
from DC to 10 GHz in momentum environment of ADS. The EM computation
was performed by using moment method (MoM) solver. Knowing the reflection
and transmission parameters, the frequency-dependent model was extracted. The
corresponding frequency results are plotted in Fig. 8.7. As expected due to the skin
depth effect, the per-unit-length resistance plotted in Fig. 8.9a increases with the
frequency. It can be pointed out that the variation of the per-unit-length capacitance
C u and conductance Gu versus interconnect metallization resistivity is negligible.
Hence, as can be seen in Fig. 8.9c, C u is rather almost constant with the frequency
and presents the average value of about 0.1 nF/m. However, the per-unit-length
conductance Gu increases up to 45 µ−1 /m in the considered frequency band.
8
Ru ( /mm)
0
1 2 3 4 5 6 7 8 9 10
Frequency (GHz)
(b) (c)
0.7
=10µ .m 0.1
0.6
=3µ .m 0.08
Lu (µH/m)
Cu , Gu
Cu (nF/m) -1
Gu (m /m)
In addition, the behavioral model of the asymmetrical interconnect tree was com-
puted from DC to 10 GHz. The corresponding frequency responses are plotted in
Fig. 8.10. As expected, the asymmetrical tree behavior can be critical when ρ is
increased. For the laminated metal with the resistivity in order of 10 µ m, the inter-
connect line attenuation can be worse than 20 dB only from only 2 GHz, 0.5 GHz,
and 0.2 GHz, respectively, for d 1 = 2 cm, d 2 = 7 cm, and d 3 = 15 cm. This illus-
trates that the operated signal bandwidth should be limited to some GHz when the
TL length is in order of tens cm. In other words, for the input with bandwidth more
(a) 0
=10µ .m
=3µ .m
-10
|T1(j )| (dB)
=2µ .m
=1µ .m
-20
-30
-40
2 4 6 8 10
Frequency (GHz)
(b) 0
=10µ .m
-10 =3µ .m
|T2(j )| (dB)
-20 =2µ .m
=1µ .m
-30
-40
-50
-60
2 4 6 8 10
Frequency (GHz)
(c) 0
=10µ .m
=3µ .m
-20
|T3(j )| (dB)
=2µ .m
=1µ .m
-40
-60
-80
2 4 6 8 10
Frequency (GHz)
Fig. 8.10 a Magnitude of the transfer function corresponding to the asymmetrical tree branch
M in M 1 [28]. b Magnitude of the transfer function corresponding to the asymmetrical tree branch
M in M 2 [28]. c Magnitude of the transfer function corresponding to the asymmetrical tree branch
M in M 3 [28]
152 T. Eudes et al.
than 2 GHz, it can be pointed out that the interconnect tree output can be completely
distorted when ρ > 5 µ m.
Furthermore, the asymmetrical tree presents significant nonlinear transmission
phases through each branch. To highlight such an effect, the group delay analytically
defined by the opposite of the derivative of the transmission phase ∠Tk ( j f ) with
respect to the radian frequency:
∂∠Tk ( j f )
τk ( f ) = − . (8.28)
2π · ∂ f
was also computed. Figure 8.11 plots the frequency-dependent group delay through
the three branches M in M 1 , M in M 2 , and M in M 3 . It can be seen that significant dis-
crepancies are found compared to the ideal values τ 1 = τ (M in M 1 ) ≈ 0.26 ns, τ 2 =
τ (M in M 2 ) ≈ 0.53 ns, and τ 3 = τ (M in M 3 ) ≈ 0.95 ns. Moreover, the ripple is inversely
increased with the metallization resistivity.
0.4
1
M M
in
0.2
0
2 4 6 8 10
Frequency (GHz)
(b) 1
(ns)
0.8
2
M M
in
0.6
0.4
2 4 6 8 10
Frequency (GHz)
(c)
1.2
(ns)
1
3
M M
in
0.8
2 4 6 8 10
Frequency (GHz)
The time-domain analyses were performed by considering the arbitrary digital data
“010110000”. This mixed signal presents 0.5 Gbps rate. Each bit element corresponds
to a trapezoidal signal with amplitude 1 V, pulse width 2 ns, and rise/fall time t r =
0.2 ns. The extracted per-unit-length parameters of the branches versus conductivity
from DC to 10 GHz are displayed in Fig. 8.12. It can be emphasized that the output
signals through the electrical paths M in M k (k = {1, 2, 3}) present rise times higher
than 1.5 ns. The 50% propagations’ delay can exceed the data period when the
interconnect length is more than 7 cm and the strip line ink resistive is higher than ρ
= 5 m m.
(a) =10µ .m
1
=3µ .m
=2µ .m
v1, V
0.5 =1µ .m
0 5 10 15
Time (ns)
(b)
1 =10µ .m
=3µ .m
=2µ .m
v2, V
0.5 =1µ .m
0 5 10 15
Time (ns)
(c)
=10µ .m
1 =3µ .m
=2µ .m
v3, V
0.5 =1µ .m
0 5 10 15
Time (ns)
Fig. 8.12 a Transient response of the POC asymmetrical tree branch M in M 1 for the 0.5 Gbps
rate input signal [28]. b Transient response of the POC asymmetrical tree branch M in M 2 for the
0.5 Gbps rate input signal [28]. c Transient response of the POC asymmetrical tree branch M in M 3
for the 0.5 Gbps rate input signal [28]
154 T. Eudes et al.
asymmetrical T-tree
branches. a M in M 1 , 0.5
1
b M in M 2 , and c M in M 3 [28]
0
0 1 2 3 4
Time (ns)
(b)
1
vM (V)
0.5
2
0 1 2 3 4
Time (ns)
(c)
1
vM (V)
0.5
3
0 1 2 3 4
Time (ns)
8 Z/Y /T /S-Matrices’ Analysis of Non-symmetric SIMO Tree Based … 155
8.4.2.4 Conclusion
This section focus is on the experimental verification of the behavioral model and the
associated methodology introduced previously. A proof of concept will be designed,
modeled, and tested in order to confirm that the method exposed can be useful
for PCB comb tree. For the sake of the analytical computation and simplicity of
structure design, non-buffered high-level trees will be explored. As aforesaid earlier
in the introductive part, physical and electrical parameters for the high-speed clock
distribution such as RAM applications accorded to the standard reported in [11] will
be considered.
It is worth pointing out that the design and simulations performed along this
chapter were carried out with the standard tool for the microwave electronic simulator
Advanced Design System from Agilent™.
For the experimental analyses, we envisaged 1:8 comb tree structure for representing
structures equivalent to the distribution of clock signals to 8-input gates.
To apply the modeling concept developed, we considered the circuit schematic
sketched in Fig. 8.14. As we can see, this structure under test is devoted to interconnect
8 receivers represented by loads Z k (k = {1, …, 8}) with a single input attacked by
156 T. Eudes et al.
driver with voltage source V in (to be defined later in the next subsection for the
time-domain analysis) and internal series resistance Rs . The equivalent electrical
circuit is composed of cascaded 8-L-form-network consisted of elementary series
TLs TLix which connect two nodes together. For instance, TLi1 is located between
N 1 and N 2 . Then, each branch k is considered by an elementary parallel TLs TLnk
which is connected to the load Z k . Then, we materialized this schematic with a PCB
constituted by electric microstrip interconnect structures.
To design the proof of concept, the structure under test has been designed over
a PCB printed on a FR4 substrate of height h = 0.8 mm and relative permittivity
supposed ideally constant εr = 4.4. The dimensions of interconnects have been
determined with respect to the DDR3 standard indicated in [11]. Accordingly, the
adopted choices are summarized in Table 8.1. A 3D representation of the 1:8 comb
tree networks under test designed in the EMDS environment of ADS is displayed in
Fig. 8.14. The interconnect line is metallic conductors assigned as copper material
having thickness 35 µm etched on the FR4 substrate.
One assumes that the elementary TLs constituting the comb tree under test present
the same width w = 1 mm. Moreover, the TLik (for k = {1, …, 6}) and TLnk (for k
= {1, …, 7}) are supposed as identical with physical lengths respectively equal to
d i = 15.3 mm and d n = 8 mm.
As a proof of concept, the prototype of 1:8 comb tree photographed in Fig. 8.15
was manufactured and measured. The characterization of the comb tree has been
made within the 100 kHz–8.5 GHz frequency bandwidth by using a VNA Agi-
lent EC5071C. The test was conducted under SOLT calibration. By optimizing the
number of measurements, the 9-port S-matrix has been reconstructed.
As aforementioned earlier, from these measurements, comparisons were made
with the proposed model and simulations from ADS of Agilent. It is worthy of note
that connectors are not taken into account both with the model and simulations.
Figures 8.16, 8.17, 8.18, 8.19, 8.20 are the plots of some remarkable results of
respectively S 21 , S 31 , S 41 , and S 91 . We point out that very good correlations have been
found out between measurements, simulations, and the proposed model, considering
the effects of connectors.
Nevertheless, a kind of “non-physical” behavior has been identified for the S 21
phase (Fig. 8.17a and zoom in Fig. 8.17b). Indeed, the positive slope within the 0–
100 MHz frequency band indicates a negative group delay. This behavior has been
-2
-4
S21 Magnitude (dB)
-6
-8
-10
-12 Measurement
Simulation
Model
-14
-16
0 1 2 3 4 5 6 7 8 9
Frequency (GHz)
found first in simulations using ADS and the proposed model (Fig. 8.17a) and then
also was confirmed by measurements (Fig. 8.17b).
Despite this unusual behavior, the developed model gives very good results in
S-parameters according to simulations from ADS and measurements from the VNA.
It is noteworthy that for higher frequencies, differences between the model and
simulations are heightened. This behavior comes from the RLCG model in function
of the frequency used in this proposed technique. Indeed, this extraction method
uses a frequency dispersion model for losses [20]. In addition, beyond 4 GHz the
differences with measurements are higher that shows the frequency limitation of the
FR-4 substrate. The connector effects also exaggerate this situation.
For the frequency AC analysis presented in this subsection, the comb tree outputs are
loaded at each termination by a parallel RC circuit as illustrated earlier in Fig. 8.4. The
impedance source is considered as a pure resistor Rs = 10 . The physical dimensions
of the output TLs and the realistic parameters of the output RC loads employed
connected at the terminations of the comb tree under test are recalled in Table 8.2.
As expected, this justifies once again that the tree network structure under test is
obviously typically unbalanced. The simulated VTFs are extracted straightforwardly
from the AC analysis in ADS schematic environment, while VTFs from the proposed
model were calculated by implementing the algorithm in Matlab.
8 Z/Y /T /S-Matrices’ Analysis of Non-symmetric SIMO Tree Based … 159
(a) 100
0
S21 Absolute Phase (°)
-100
Simulation
Model
-200
-300
-400
-500
-600
0 1 2 3 4 5 6 7 8 9
Frequency (GHz)
(b) 50
40
30
S21 Absolute Phase (°)
Measurement
20 Simulation
Model
10
0
-10
-20
-30
-40
-50
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Frequency (GHz)
Fig. 8.17 S 21 phase between the input and receiver no. 1 (a) and negative group delay confirmed
by measurement (b) [29]
Frequency-domain comparisons are made between simulations and the mode for
each receiver. The VTFs of H 1 and H 8 are given in amplitude and phase, respectively,
in Figs. 8.21a, b and 8.22a, b.
Next, the standard deviation was calculated for each H x VTF (with x = {1, 2, …,
8}) and for different frequency bandwidth, DC—3 GHz, 3–5 GHz, and 5–10 GHz.
The results obtained are summarized in Table 8.3.
The obtained results show an excellent correlation between the simulations and
the calculations based on the developed model. As expected, the differences are more
heightened for high frequencies due to the frequency dispersion model. Nonetheless,
the accuracy obtained will be sufficient for the SI forecasting in this network.
160 T. Eudes et al.
-6
-8
S31 Magnitude (dB)
-10
-12
Measurement
Simulation
Model
-14
-16
0 1 2 3 4 5 6 7 8 9
Frequency (GHz)
-10
S51 Magnitude (dB)
-15
-20
Measurement
Simulation
Model
-25
-30
0 1 2 3 4 5 6 7 8 9
Frequency (GHz)
-10
-15 Measurement
S91 Magnitude (dB) Simulation
Model
-20
-25
-30
-35
-40
0 1 2 3 4 5 6 7 8 9
Frequency (GHz)
(a) 20
Simulation
10
Model
0
VTF Gain (dB)
-10
-20
-30
-40
0 1 2 3 4 5 6 7 8 9 10
Frequency (GHz)
(b) 200
0
Simulation
VTF Absolute Phase (°)
Model
-200
-400
-600
-800
0 1 2 3 4 5 6 7 8 9 10
Frequency (GHz)
Y-matrices, mathematical expressions of the VTF for different n-levels of comb tree
networks are established.
For validating the behavioral model, a proof of concept consisting of a PCB 1:8
comb tree microstrip circuit was designed, constructed, and measured. The imple-
mented prototype was intentionally chosen with respect to the DDR standardization
reported in [11]. Comparisons between the AC analyses of VTFs from simulations
and the calculated models were realized in the frequency band from DC to 8.5 GHz.
It was stated that due to substrate dispersion and loss which were not taken into
account in the behavioral modal, slight differences were occurred between the fre-
quency results at higher frequencies. The slight differences between the models,
simulations, and experimentations occurred at higher frequencies are mainly due to
the dispersion of the substrate employed. Furthermore, with time-domain investiga-
tions, the SI characteristics as propagation delay and rise/fall times are extracted. The
8 Z/Y /T /S-Matrices’ Analysis of Non-symmetric SIMO Tree Based … 163
(a) 20
0
-20
-40
VTF Gain (dB)
Simulation
-60 Model
-80
-100
-120
-140
-160
0 1 2 3 4 5 6 7 8 9 10
Frequency (GHz)
(b) 0
-500
VTF Absolute Phase (°)
Simulation
Model
-1000
-1500
-2000
-2500
-3000
-3500
0 1 2 3 4 5 6 7 8 9 10
Frequency (GHz)
degradations of signals obtained with the considered structure justify the necessity to
forecast SI for designing high-speed interconnects’ structures. For that, 2-Gbps rate
input was considered. For the different tests performed, good agreements have been
found between simulations from a commercial tool and the results obtained with the
proposed fast modeling.
The SI prediction is indispensable especially for complex shape interconnects
in designing high-rate integrated circuits in order to preserve the quality of data
synchronization and processing. Indeed, this technological limitation is currently
a major issue for microelectronic manufacturers. As future work, the prominent
methodology explored in this chapter will also be proposed to predict the performance
of IC advanced packages based on 3D integration technology or silicon interposer
solutions recently analyzed [26].
164 T. Eudes et al.
Table 8.3 Standard deviation in amplitude and phase between the results obtained from the
simulation and the model
VTF Frequency bandwidth VTF gain standard VTF phase standard
(GHz) deviation (dB) deviation (°)
H1 0–3 0.767 5.355
3–5 3.093 22.33
5–10 3.763 24.05
H2 0–3 0.953 7.173
3–5 2.921 19.25
5–10 3.646 23.90
H3 0–3 1.113 8.570
3–5 2.856 18.99
5–10 3.715 24.42
H4 0–3 1.333 10.356
3–5 2.941 19.59
5–10 4.363 28.85
H5 0–3 1.647 12.258
3–5 3.103 21.24
5–10 3.673 23.94
H6 0–3 1.841 13.916
3–5 3.079 20.88
5–10 3.476 22.98
H7 0–3 1.844 13.160
3–5 2.583 17.23
5–10 3.205 21.00
H8 0–3 1.576 12.170
3–5 2.935 22.93
5–10 3.539 21.91
References
1. http://www.itrs.net/Links/2013ITRS/2013Chapters/2013ExecutiveSummary.pdf. Accessed
2014
2. E. Mounier, A. Bonnabel, M. Rosina, Flexible & printed electronics: flexible applications based
on printed electronics technologies. Market & Technology Report, Yole Développement, Apr
2013
3. H.R. Khaleel, H.M. Al-Rizzo, A.I. Abbosh, Design, fabrication, and testing of flexible antennas,
in Advancement in Microstrip Antennas with Recent Applications, Chap. 5, ed. by A. Kishk
(Intech Open Book, 2013), pp. 363–383
4. http://www.corning.com/displaytechnologies/en/products/flexible.aspx. Accessed 2014
5. A. Aliane, V. Fischer, R. Coppard, Large area printed temperature sensors on flexible substrate,
in Proceedings of 5th IEEE International Workshop on Advances in Sensors and Interfaces
(IWASI), Bari, Italy, pp. 188–192, 13–14 June 2013
8 Z/Y /T /S-Matrices’ Analysis of Non-symmetric SIMO Tree Based … 165
6. A. Russo, B.Y. Ahn, J.J. Adams, E.B. Duoss, J.T. Bernhard, J.A. Lewis, Pen-on-chapter flexible
electronics. Adv. Mater. 23(30), 3426–3430 (2011)
7. M. Charbonneau, S. Jacob, M. Benwadih, J. Bablet, V. Fischer, D. Boutry, R. Coppard, I.
Chartier, S. Abdinia, E. Cantatore, G. Maiellaro, E. Ragonese, G. Palmisano, R. Gwoziecki,
Printed organic TFTs for interfacing circuits and active matrix, in Proceedings of 20th
International Display Workshops (IDW’13), Sapporo, Japan, 4–6 Dec 2013
8. F. Jun, Y. Xiaoning, J. Kim, B. Archambeault, A. Orlandi, Signal integrity design for high-speed
digital circuits: progress and directions. IEEE Trans. EMC 52(2), 392–400 (2010)
9. J.H. Kim, D. Oh, W. Kim, Accurate characterization of broadband multiconductor transmission
lines for high-speed digital systems. IEEE Trans. Adv. Packag. 33(4), 857–867 (2010)
10. J. Kim, E. Li, Special issue on PCB level signal integrity, power integrity, and EMC. IEEE
Trans. EMC 52(2), 246–247 (2010)
11. 240-Pin PC-6400/PC-5300/PC-4200/PC-3200 DDR2 SDRAM Registered DIMM Design
Specification, JEDEC Standard 21C, Jan 2010
12. J.F. Buckwalter, Predicting microwave digital signal integrity. IEEE Trans. Adv. Packag. 32(2),
280–289 (2009)
13. E. Bogatin, Essential principles of signal integrity. IEEE Microwave Mag. 12(5) (2011)
14. K.S.R. Krishna, J.L. Narayana, L.P. Reddy, ANN models for microstrip line synthesis and
analysis. Int. J. Elect. Syst. Sci. Eng. 1, 196–200 (2008)
15. M.S. Maza, M.L. Aranda, Analysis of clock distribution networks in the presence of crosstalk
and ground bounce, in Proceedings of 8th IEEE International Conference on Electronics,
Circuits and Systems, vol. 2, Malta, pp. 773–776, Aug 2001
16. I. Chanodia, D. Velenis, Effects of crosstalk noise on H-tree clock distribution networks, in
Proceedings of IEEE International Symposium on CAS (ISCAS) 2006, vol. 4, Island of Kos,
Greece, pp. 5627–5630, 21–24 May 2006
17. T. Eudes, B. Ravelo, A. Louis, Experimental validations of a simple PCB interconnect model
for high-rate signal integrity. IEEE Trans. EMC 54(2), 397–404 (2012)
18. J. Roullard, A. Farcy, S. Capraro, T. Lacrevaz, C. Bermond, G. Houzet, J. Charbonnier, C.
Fuchs, C. Ferrandon, P. Leduc, B. Flechet, Evaluation of 3D interconnect routing and stacking
strategy to optimize high speed signal transmission for memory on logic, in Proceedings of
IEEE 62nd Electronic Components and Technology Conference (ECTC), San Diego, pp. 8–13,
29 May 29–1 June 2012
19. S. Roy, A. Dounavis, RLC interconnect modeling using delay algebraic equations, in
Proceedings of IEEE CAS Workshop, Richardson, TX, pp. 1–4, 4–5 Oct 2009
20. T. Eudes, B. Ravelo, A. Louis, Transient response characterization of the high-speed
interconnection RLCG model for the signal integrity analysis. PIER 112, 183–197 (2011)
21. A. Koul, P.K.R. Anmula, M.Y. Koledintseva, J.L. Drewniak, S. Hinaga, Improved technique
for extracting parameters of low-loss dielectrics on printed circuit boards, in Proceedings of
IEEE EMC Symposium, Austin, TX, pp. 191–196, 17–21 Aug 2009
22. C.-W.A. Tsao, C.-K. Koh, A clock tree router for general skew constraints, in Proceedings of
IEEE/ACM International Conference on Computer Aided Design (ICCAD-2000), San Jose,
CA, pp. 400–405, 5–9 Nov 2000
23. A.O. Balkan, G. Qu, U. Vishkin, A mesh-of-trees interconnection network for single-chip par-
allel processing, in Proceedings of International Conference on Application-specific Systems,
Architectures and Processors, 2006 (ASAP’06), Steamboat Springs, CO, pp. 73–80, Sept 2006
24. W.-K. Loo, K.-S. Tan, Y.-K. Teh, A study and design of CMOS H-tree clock distribution
network in system-on-chip, in Proceedings of 8th IEEE International Conference on ASIC,
Changsha, Hunan, China, pp. 411–414, Oct 2009
25. B. Ravelo, Behavioral model of symmetrical multi-level T-tree interconnects. PIER B 41, 23–50
(2012)
26. R.M. Averill, K.G. Barkley, M.A. Bowen, P.J. Camporese, A.H. Dansky, R.F. Hatch, D.E.
Hoffman, M.D. Mayo, S.A. McCabe, T.G. McNamara, T.J. McPherson, G.A. Northrop, L.
Sigal, H.H. Smith, D.A. Webber, P.M. Williams, Chip integration methodology for the IBM
S/390 G5 and G6 custom microprocessors. IBM J. Res. Develop. 43(5/6), 681–706, Sept/Nov
1999
166 T. Eudes et al.
27. S.L. Wang, Y.W. Chang, Delay modeling for buffered RLY/RLC trees, in Proceedings of 2005
IEEE International Symposium on VLSI Design, Automation and Test (VLSI-TSA), Hsinchu,
Taiwan, pp. 237–240, 25–27 Apr 2005
28. B. Ravelo, Modelling of asymmetrical interconnect T-tree laminated on flexible substrate. Eur.
Phys. J. Appl. Phys. (EPJAP) 72(2)(20103), 1–9 (2015)
29. T. Eudes, B. Ravelo, T. Lacrevaz, B. Fléchet, Electrical performance modeling of unbalanced
comb tree networks on advanced PCB interconnects for high-rate clock signal distribution, in
Proceedings of 63rd IEEE Electronic Component Technology Conference (ECTC 2013), Las
Vegas, Nevada, USA, pp. 2024–2034, 28–31 May 2013
30. T. Eudes, Développement de modèles pour l’analyse de l’intégrité des signaux électriques
ultra-rapides propagés dans les réseaux d’interconnexion en arbre (in French). Ph.D. thesis,
IRSEEM/ESIGELEC, Univ. Rouen (2013)
31. B. Ravelo, F. Vurpillot, A.K. Jastrzebski, Asymmetrical PCB interconnect tree modelling
with coupling effect, in Proceedings of 2014 International Symposium on Electromagnetic
Compatibility (EMC Europe 2014), Gothenburg, Sweden, pp. 713–718, 1–4 Sept 2014
32. J. Cho, E. Song, H. Kim, S. Ahn, J.S. Pak, J. Kim, J. Kim, Mixed-mode ABCD parameters:
theory and application to signal integrity analysis of PCB-level differential interconnects. IEEE
Trans. EMC 53(3), 1–9 (2011)
33. T. Eudes, B. Ravelo, Analysis of multi-gigabits signal integrity through clock H-tree. Int. J.
Circ. Theor. Appl. 41(5), 535–549 (2013)
34. D.A. Frickey, Conversions between S, Z, Y, h, ABCD and T parameters which are valid for
complex source and load impedances. IEEE Trans. MTT 42(2), 205–216 (1994)
Chapter 9
Cartographical Analyses of Reflection
and Transmission Coefficients of Shunt
Coupled Lines
Blaise Ravelo
9.1 Introduction
The high-speed PCB design performances depend naturally on the electrical multi-
conductor interconnect coupling constraints [1]. An accurate and relevant characteri-
zation method is expected to predict the unintentional degradation due to the electrical
effects as crosstalk. Predictive modelling methods can be used during the printed cir-
cuit board design phases [2]. The PCB interconnect crosstalk can affect undesirably
the digital signal bit error rate [3, 4]. Furthermore, the interconnect input/output
line coupling can generate awful electromagnetic radiating emission [5]. Tentative
roadmaps report that more accurate predictive signal and power integrity models are
needed to realize reliable high-speed electronic circuits [6].
Different circuit analysis tools [7, 8] in particular for printed circuit board link
levels were provided to help the design engineers against the degradation phenomena
due to the undesirable effects as crosstalk. But faster and easier modelling method-
ology is required when the integration density is increased [9]. Therefore, printed
circuit board TL RLCG-based modelling methods were suggested for the responses
of pre-visualization and pre-determination of the signal reflection, attenuation, delay,
differential noise and further distortions [10–12]. In addition, optimization algorithms
built with typically RC and RLC networks were also proposed for the SI and PI anal-
yses and also to improve the printed circuit board interconnect line performances
[13, 14]. One of main applications of the RLC network modelling is the accurate
prediction of clock signal distribution electrical interconnects.
Behind the developed model, the interconnect coupling effects can be occurred
for certain parameters of the clock signalling sharing topology [15]. The advanced
interconnect network is regularly used during the digital system design phase to
ensure the signal synchronization [16]. The electrical interconnect tree topologies
B. Ravelo (B)
Graduate Engineering School, ESIGELEC, Sotteville les Rouen, Seine-Maritime, France
e-mail: blaise.ravelo@yahoo.fr
are among the most popular signal distribution for the digital circuit implementation.
Furthermore, the interconnect tree network plays an important role notably to predict
the signal synchronization and attenuation [17, 18]. However, the interconnect inter-
branch coupling remains an open issue during the implementation of the interconnect
tree in the confined space. So far, time-consuming simulation techniques of inter-
connect crosstalk were proposed [19]. The most popular of the existing simulators
are using solvers based on the time domain [20] or finite-element frequency-domain
[21] approaches. The main drawbacks of such simulation tools are the necessity
to design the entire printed circuit board structures usually with complex geometry.
Such a full-wave simulation can be accurate for certain cases of symmetrical coupled
lines. An approximation method of worst-case system-level crosstalk at higher fre-
quencies with analytical formulas is developed in [22]. Then, deeper level crosstalk
between interconnect very large-scale integration CMOS and field effect transistor
circuits was suggested by virtue of TL approach [23, 24]. A computation method
of improper radiating modes (leaky modes) based on the propagation characteris-
tics of coupled microstrip TLs is presented in [25]. However, further concepts are
recently developed to predict the interconnect couplings for various cases of com-
posite right/left-handed TL [26] and complex-layout traces [27] of printed circuit
boards. Thus, a more generalized macromodel concept of high-speed interconnect
SI analyses based on full-wave time-domain was also introduced in [28]. But the
implementation of model based on rational transfer function approximated is fairly
complex.
In complementary to the digital system signal integrity analyses, the coupled
line networks are regularly used to design microwave planar circuits thanks to its
benefits in term of the compactness [29–33]. For example, the coupled line can
be employed for the bus link used to ensure the equipartitioned energy [29]. The
design and implementation of high-performance microwave coupled line filters [30–
32] constitute one of the most attractive breakthroughs for the electronic design
and research engineers. Basically, the employed topologies are based on the stub-
embedded resonators with the synthesis of the transmission zeros (TZs) and reflection
zeros (RZs). The design concept is carried out with admittance-transformer feeds
for flexible terminations [30]. A synthesis method dedicated to the dual-wideband
bandpass filters is introduced in [31]. The particularity of this filter topology is
based on the integration of source–load coupling network. The basic filter block
is built with stepped impedance resonators. To meet the multistandard demand for
wireless communication system, more complex synthesis approach for a multiband
low temperature co-fired ceramics (LTCC) bandpass filter was also implemented
[32].
Despite this diversity of coupled line applications, further investigation is still
needed for the implementation of optimum termination networks. A tentative method
was suggested based on the coupling mode and matrix approaches [33, 34]. But
such methods are not expanded enough due to the lack of analytical understanding
on the mechanism of the coupling effects as the TZ and RZ positions. A relevant
calculation method is required during the synthesis of high-density mixed circuits.
For this reason, accurate modelling methods are still needed for the particular cases
9 Cartographical Analyses of Reflection and Transmission … 169
of coupled line structures as the parallel stub resonators in function of the coupling
level.
The present chapter addresses a complete theory on the coupled-parallel-line
(CPL) used as a parallel stub including the crosstalk phenomenon. Based on the full
coupling matrix combined with the direct input–output TL, a fast and accurate CPL
model with the theoretical exact expressions of the TZs and RZs will be established.
Then, illustrative applications will be proposed to approve the established theoreti-
cal formulations. Lastly, discussion on the potential applications of the established
modelling concept will be drawn in the conclusion.
The CPL configuration including the source and load terminals represented by the
reference impedance R0 = 50 is presented in Fig. 9.1a. In other words, it acts
as a structure of parallel stub constituted by TLs ➀–➂ and ➁–➃ with different
lengths d 1 and d 2 (with d 1 > d 2 ). The two parallel stubs are supposedly separated
by space s. It is worth emphasizing that the overall structure is implemented in
microstrip technology and each stub is terminated by the arbitrary loads R1 and
R2 . Nonetheless, the extraction of the equivalent model enabling to determine the
fundamental elements as the even- and odd-mode characteristic impedance’s Z o and
Z e is required during the design phase and implementation of electronic structures.
In this way, the CPL equivalent diagram is introduced in Fig. 9.1b. This equivalent
model takes innovatively into account the interbranch coupling.
For the sake of the analytical simplification, the TLs are supposed as a lossless
structure. It can be recalled that the physical lengths dk and the resonance radian
frequencies ωk (k = {1, 2}) of the two elementary TLs are linked by the relation:
πv
dk = , (9.1)
2ωk
by denoting v the wave speed. The main technical issue to be solved in this chapter is
the influence of the stub ➀–➁ over length on the CPL TZ and RZ. Substantially, the
electrical length of the stub over length between TL1 and TL2 (d 1 > d 2 ) is analytically
defined by:
170 B. Ravelo
Fig. 9.1 a Microstrip CPL structure under study including the source and load impedances of the
CPL structure under study [35], b equivalent diagram of the CPL structure under study [35]
πω
θa (ω) = . (9.2)
2ωa
π ω(ω2 − ω1 )
θa (ω) = . (9.3)
2ω1 · ω2
It can be recalled that the ABCD matrix of the TL having characteristic impedance
Z c and electrical length θ is defined by:
cos(θ ) Z c · sin(θ )
[ABC D]TL = . (9.4)
sin(θ)
Zc
cos(θ )
The corresponding Z- and Y-matrices can be extracted from the ABCD-to-Z and
ABCD-to-Y transforms following the circuit and system theory.
√
with j = −1 and the TL electric length is defined by:
172 B. Ravelo
Fig. 9.2 Transformed equivalent diagram of the CPL circuit introduced in Fig. 9.1b
πω
θ (ω) = . (9.7)
2ω2
This Z-matrix can serve to the derivation of the CPL input parallel impedance and
then, the corresponding global S-parameters.
The load Z L constitutes the input impedance of the over length TL of the stub ➀–➂
which presents an electrical length θa . In function of the load R1 , based on the TL
theory, the input impedance of the over length stub which can be characterized by
TL (Z c , θa ) is given by:
Z c [R1 + j Z c tan(θa )]
ZL = . (9.8)
Z c + j R1 tan(θa )
9 Cartographical Analyses of Reflection and Transmission … 173
Hence, the vector currents [I] of the CPL octopole network described in Fig. 9.2
are determined by inverting the generalized Ohm’s law:
[V ] = [Z ] · [I ], (9.9)
with the Z-matrix defined in (9.5). As Ports ➀ and ➁ are electrically connected,
therefore
Vin = V1 = V2 . (9.10)
Iin = I1 + I2 . (9.11)
Via Ohm’s law, the CPL input impedance can be deduced via the basic relation:
V1 ( jω)
Z in ( jω) = . (9.12)
Iin ( jω)
After the mathematical analyses, we can establish the branch current relationships:
I3 − I4
I2 = I1 − , (9.13)
cos(θ )
⎡ ⎤
R1 tan(θa )(Z e (2 + tan2 (θ ) + 2/ cos(θ ))
2Z e [R1 tan(θa )−Z c ]
cos(θ)
I1 + ⎣ +Z o tan2 (θ )) ⎦ I4
− j Z c (Z e (2 + tan (θ )) + Z o tan (θ ))
2 2
I3 = , (9.14)
R1 (Z o + Z e ) tan(θ ) tan(θa ) − 2R1 Z c tan(θ )
− j Z c [2Z c tan(θa ) + (Z o + Z e ) tan(θ )]
I4 2Z e [Z c (Z o tan(θ ) + Z c tan(θa )) + j R1 (Z o tan(θa ) − Z c )]
=
,
I1 R1 R2 (Z e + Z o ) tan2 (θ ) − (2 + tan2 (θ ))Z e − Z o tan2 (θ ) Z c2 tan(θa )
R Z Z (2 + tan2 (θ )) − 2R02 R1 tan(θa )
−2 R02 + R1 R2 Z c tan(θ ) + j 1 e c 2
−Z c tan (θ )(R2 (Z e + Z o ) + R1 Z o )
(9.15)
For the particular case where R1 and R2 both short-circuited, it can be demonstrated
that:
j Z e tan(θ ) 2R02 tan(θ ) + Z c tan(θa )(Z e + Z o )
Z in | R1 =0 =
. (9.19)
R2 =0 4R02 tan(θ ) + 2Z c tan(θa ) Z e − 2Z o tan2 (θ )
These equation systems permit the direct calculation of the under study CPL
S-parameters in function of the particular frequencies corresponding to the physical
lengths of the structure. Substantially, the reflection and transmission parameters are
expressed as:
−R0
S 11 ( jω) = 2Z in ( jω)+R0
. (9.24)
S 12 ( jω) = 2Z in
2Z in ( jω)+R0
and
By definition, the CPL structure under study presents TZs and RZs at the radian
frequency ω when Z in ( jω) = 0 and Z in ( jω) = ∞, respectively. Using these
conditions, the TZ and RZ existence conditions in function of the CPL parame-
ters can be established. Meanwhile, for the different configurations of R1 and R2
at the arbitrary frequency, the TZ and RZ existence conditions can be derived from
expressions (9.16)–(9.19). However, by using generalized fundamental formulations
(9.17)–(9.18), the TZ existence condition of the CPL structure introduced in Fig. 9.1a
is written as:
cot(θ)[2Z c tan(θa )+tan(θ)(Z e +Z o )]
R1 = R2 Z c2R
0 tan(θ) tan(θa )+Z c (Z e +Z o )
2
The same, the TZ existence condition can also be established when for the differ-
ence cases where R2 is short-circuited, open-ended and 50 -matched. Meanwhile,
the unified existence condition is the existence conditions which are respectively
simplified as:
tan(θ ) 2R02 tan(θa ) tan(θ ) − Z c (Z e + Z o ) = 0, (9.30)
2 tan(θa ) R1 R02 tan2 (θ ) − R0 Z c2 = Z c tan(θ )(R1 (Z e + Z o ) + R0 Z o ). (9.32)
First, the present analytical characterization is limited to the particular cases where
R2 = 0 and R2 = ∞ for the sake of simplicity. Then, by denoting k the coupling coef-
ficient between the two branches of the CPL, the even- and odd-mode characteristic
impedances by the basic relations:
⎧
⎨ Z e = R0 1+k
1−k . (9.33)
⎩ Z o = R0 1−k
1+k
tan2 (θ ) Z c2
Under the condition tan2 (θa )
≥ R02
R1 = ∞ k =1− 1
tan(θ ) tan(θa )
R2 = ∞ R1 = 0 R02 tan2 (θ )
k =1−
Z c tan(θa ) Z c tan(θa )± Z c2 tan2 (θa )−R02 tan2 (θ )
Z c2
Under the condition tan2 (θ) tan2 (θa ) ≤ R02
Based on the characteristic equations established previously in Tables 9.1 and 9.2,
the cartographies illustrating the existence areas of TZs and RZs in function of
the ratio ω2 /ω1 and the coupling factor k are elaborated. The explored numerical
computations were determined by considering that ω2 /ω1 is varied from 1 to 5. The
TZ (Fig. 3.6a) and RZ (Fig. 3.6b) radian frequencies to ω1 ratio ωzero /ω1 were also
supposedly varied from 1 to 5. As a matter of fact, the cartography of (k, ω2 /ω1 ,
ωzero /ω1 ) for the different coupled values of the loads (R1 = ∞, R2 = ∞), (R1 =
0, R2 = ∞), (R1 = ∞, R2 = 0) and (R1 = 0, R2 = 0) are, respectively, displayed
in Figs. 9.3, 9.4, 9.5 and 9.6. The RZ cartographies do not present any particular
configuration. However, TZ frequencies exist for certain range of the ratio ω2 /ω1 .
Corollary, noticeable observations on the TZ existence condition can be deduced
from:
(a) Figure 9.3a corresponds to the case (R1 = ∞, R2 = ∞); the TZs are rather
situated under the horizontal axis defined by ωTZ R1 =R2 =∞ = ω1 .
Fig. 9.3 a TZ and b RZ existence surface areas in function of the coupling factor and ω2 /ω1 for
R1 = ∞ and R2 = ∞ [35]
9 Cartographical Analyses of Reflection and Transmission … 179
Fig. 9.4 a TZ and b RZ existence surface areas in function of the coupling factor and ω2 /ω1 for
R1 = 0 and R2 = ∞ [35]
(b) Figure 9.4 corresponds to the case (R1 = ∞, R2 = 0); the TZ position should
belong in the half plane above the critical hyperbolic curve ωTZ R1 =∞,R2 =0 =
ω1 ω2 /(ω2 − ω1 ).
(c) Figure 9.5 corresponds to the case (R1 = 0, R2 = ∞), the TZ are situated above
the horizontal axis defined by ωTZ R1 =0,R2 =∞ = ω1 ;
(d) Figure 9.6 corresponds to the case (R1 = 0, R2 = 0); the TZ position should
belong in the half plane above the critical axis ωTZ R1 =0,R2 =0 = ω2 /2.
In all, it is revealed from these graphical analyses that the TZ and RZ cannot exist
for all value of parameters (k, ω2 /ω1 ). In addition to the exploitation of the analytical
formulations, these graphical plots could be used to predict the positions of the RZs
and TZs. More importantly, it enables to predict fast and accurately the influence
of the multiconductor line interbranch coupling in particular during the design of
high-density circuits.
To check the relevance of the previous theoretical concept, comparisons between
the calculated TZs and RZs from the established model and simulations will be
discussed in the next section.
180 B. Ravelo
Fig. 9.5 a TZ and b RZ existence surface areas in function of the coupling factor and ω2 /ω1 for
R1 = ∞ and R2 = 0 [35]
Fig. 9.6 a TZ and b RZ existence surface areas in function of the coupling factor and ω2 /ω1 for
R1 = 0 and R2 = 0 [35]
9 Cartographical Analyses of Reflection and Transmission … 181
More practical POCs integrating the design of parametric CPLs are analysed in
order to validate the previous theory. After the description of the designed POC, the
performed simulations were run in the microwave and electronic circuit design and
simulator ADS® from Keysight Technologies® . The comparative results illustrating
the RZ and TZ frequency shifts in function of the CPL coupling coefficient will be
discussed.
First, the considered POCs of the CPL are designed in microstrip technology. The
circuit parameter syntheses as even- and odd-impedance characteristics Z e and Z o
were made by using the Hammerstad & Jensen approach.
The CPL physical lengths were arbitrarily assigned corresponding to the stub
quarter wavelength resonance frequencies f 1 = 2 GHz and f 2 = 3 GHz. The POC
circuits were designed and printed on the FR4 substrate having dielectric permittivity
εr = 4.4, loss tangent tan(δ) = 0.02 and the thickness 1.6 mm. The considered sub-
strate was assumed to be Cu-laminated with metal thickness 35 µm. The CPL was
considered with the over length line characteristic impedance Z c = 50 . Therefore,
based on the Hammerstad & Jensen model, the synthesized physical width is equal
to w = 3.02 mm. The CPL lengths d 1 = 21.85 mm and d 2 = 14.53 mm correspond
to the resonances at f 1 and f 2 ‚ respectively. As aforementioned, three different cases
of coupling k = {−10 dB, −6 dB, −3 dB} are investigated. Table 9.3 summarizes
the synthesized geometrical characteristics for the test values of coupling coefficient
k.
For the further understanding on the feasibility of the analytical models established
previously in Sect. 9.2, parametric investigations with the variation of the coupling
level from −10 to −3 dB and the over length stub characteristic impedance will be
discussed in the next paragraph.
The numerical applications were carried out based on the MATLAB implementation
of the characteristic equations addressed in Tables 9.1 and 9.2. To do this, the calcu-
lated TZs and RZs are plotted in function of the CPL parameters R0 , Z c , f 1 and f 2 .
As practical example, by considering Z c = R0 and f 2 = 1.5f 1 , we can generate the
predictive variation of TZ and RZ versus the coupling level k displayed in Fig. 9.7.
These graphical results indicate how the TZs and RZs of the stub change. Emphat-
ically, due to the crosstalk, they are different to the initial resonance frequencies of
the stubs constituting the CPL. Moreover, the fundamental questions on the influ-
ence of the over length stub ➀–➂ on the CPL TZ and RZ can be answered with the
parametric analyses versus Z c . To do this, numerical implementation of the relations
introduced in Tables 9.1 and 9.2 for the case (R1 = ∞, R2 = ∞) was carried out by
varying the characteristic impedance Z c , from 10 to 100 . The results are depicted
in Fig. 9.8.
To illustrate more explicitly the influence of the coupling on the TZs and RZs,
comparisons between the non-coupled stub resonators and CPL were made. To do
this, different values of the stub lower frequency f 1 were considered from 1 to 5 GHz.
Then, the upper frequency f 2 was swept from f 1 to 5f 1 .
The simulated transmission and reflection coefficients for the cases (R1 = ∞, R1 =
0) are explored in Figs. 9.9 and 9.10. From these results, comparisons between the
TZs and RZs from ADS simulations and the fundamental formulations provided in
Tables 9.1 and 9.2 were carried out. As expected, a good agreement between the
values of RZs and TZs with relative errors lower than 1% was found. This result
confirms the validity of the proposed theoretical concept to predict the TZ and RZ
positions notably in microstrip technology. It can be understood from Fig. 9.9 (resp.
Fig. 9.10) that the TZ is shifted in left (resp. right) for R1 short-circuited (resp. open-
ended) when the coupling level is higher. However, for both cases, the RZ bandwidth
is shortened when R2 = 0.
Once again, a good agreement between the positions of the TZs and RZs from
ADS simulations and the analytical formulations established in Tables 9.1 and 9.2
is observed. In this case, the coupling phenomenon tends to increase the bandwidth
between the successive values of TZ (Figs. 9.11 and 9.12).
184 B. Ravelo
To complete the previous analyses on the TZ and RZ, multiple cases of the analy-
ses based on the over length stub characteristic impedances were also carried out.
Figure 9.13 introduces the plots of the ADS-simulated CPL transmission and reflec-
tion coefficients by varying Z c from 10 to 100 . It is worth noting that the funda-
mental TZ position increases with Z c . As can be seen in Table 9.4, once again, a good
agreement between the TZ and RZ positions from ADS simulations and the proposed
CPL model was realized. The numerical relative errors are lower than 0.5%.
Table 9.4 Considered microstrip CPL TZ and RZ versus Z c for f 2 = 1.5f 1 = 3 GHz
Z c () f Tzeros f Rzeros
Model (GHz) ADS (GHz) Model (GHz) ADS (GHz)
10 1.090 1.090 2.080 2.080
30 1.682 1.683 2.380 2.380
50 1.967 1.968 2.535 2.538
70 2.155 2.156 2.630 2.627
90 2.275 2.276 2.695 2.695
9 Cartographical Analyses of Reflection and Transmission … 187
References
3. C.S. Ng, T.S. Yeo, Effects of interline coupling on bit error rate for high-speed digital signal
propagation on PCB. IEEE Trans. Ind. Electron. 45(2), 362–364 (1998)
4. H. Jeong, T. Kim, Y. Yang, T. Song, G. Kim, H.-S. Won, S.-O. Jung, Offset-compensated cross-
coupled PFET bit-line conditioning and selective negative bit-line write assist for high-density
low-power SRAM. IEEE Trans. CAS I: Regul. Pap. 62(4), 1062–1070 (2015)
5. C. Su, T.H. Hubing, Calculating radiated emissions due to I/O line coupling on printed circuit
boards using the imbalance difference method. IEEE Trans. EMC 54(1), 212–217 (2012)
6. J. Fan, X. Ye, J. Kim, B. Archambeault, A. Orlandi, Signal integrity design for high-speed
digital circuits: progress and directions. IEEE Trans. EMC 52(2), 392–400 (2010)
7. B.R. Archambeault, Using signal integrity tools for EMC analysis, in PCB Design for Real-
World EMI Control, The Springer International Series in Engineering and Computer Science,
vol 696, 2002, pp. 171–186
8. S. Müller, T. Reuschel, R. Rimolo-Donadio, Y.H. Kwark, Energy-aware signal integrity analysis
for high-speed PCB links. IEEE Trans. EMC 57(5), 1226–1234 (2015)
9. R.M. Averill, K.G. Barkley, M.A. Bowen, P.J. Camporese, A.H. Dansky, R.F. Hatch, D.E.
Hoffman, M.D. Mayo, S.A. McCabe, T.G. McNamara, T.J. McPherson, G.A. Northrop, L.
Sigal, H.H. Smith, D.A. Webber, P.M. Williams, Chip integration methodology for the IBM
S/390 G5 and G6 custom microprocessors. IBM J. Res. Develop., 43(5/6), pp. 681–706 (1999)
10. A. Deutsch, G.V. Kopcsay, P. Restle, G. Katopis, W.D. Becker, H. Smith, P.W. Coteus, C.W.
Surovic, B.J. Rubin, R.P. Dunne, T. Gallo, K.A. Jenkins, L.M. Terman, R.H. Dennard, G.A. Sai-
Halasz, D.R. Knebel, When are transmission-line effects important for on-chip interconnection?
IEEE Trans. MTT 45(10), 1836–1846 (1997)
11. T. Eudes, B. Ravelo, A. Louis, Experimental validations of a simple PCB interconnect model
for high-rate signal integrity. IEEE Trans. EMC 54(2), 397–404 (2012)
12. M. Choi, J.-Y. Sim, H.-J. Park, B. Kim, An approximate closed-form transfer function model
for diverse differential interconnects. IEEE Trans. CAS I: Regul. Pap. 62(5), 1335–1344 (2015)
13. Q. Zhu, W.W.M. Dai, High-speed clock network sizing optimization based on distributed RC
and lossy RLC interconnect models. IEEE Trans. CAD Integr. Circuits Syst. 15(9), 1106–1118
(1996)
14. J.N. Tripathi, R.K. Nagpal, R. Malik, Robust optimization and reflection gain enhancement
of serial link system for signal integrity and power integrity. Int. J. Des., Anal. Tools Circuits
Syst. 2(1), pp. 70–85 (2011)
15. I. Chanodia, D. Velenis, Parameter variations and crosstalk noise effects on high performance
H-tree clock distribution networks. Analog Integr. Circ. Sig. Process 56(1–2), 13–21 (2008)
16. W.-C. Liu, T.-C. Wei, Y.-S. Huang, C.-D. Chan, S.-J. Jou, All-digital synchronization for
SC/OFDM mode of IEEE 802.15.3c and IEEE 802.11ad. IEEE Trans. CAS I: Regul. Pap.
62(4), 545–553 (2015)
17. N. Rakuljic, I. Galton, Tree-structured DEM DACs with arbitrary numbers of levels. IEEE
Trans. CAS I: Regul. Pap. 52(2), 313–322 (2010)
18. B. Ravelo, Behavioral model of symmetrical multi-level T-tree interconnects. Prog. In
Electromagn. Res. B 41, 23–50 (2012)
19. J.E. Rayas-Sanchez, A frequency-domain approach to interconnect crosstalk simulation and
minimization. Microelectron. Reliab. J. 44(4), 673–681 (2004)
20. H. Aliakbaria, A. Abdipoura, R. Mirzavand, Accurate time-domain modeling of multi-finger
pHEMT transistor based on transmission line theory. AEU—Int. J. Electron. Commun. 69(1),
215–225 (2015)
21. S.M. Musa, M.N.O. Sadiku, Finite element method analysis of symmetrical coupled microstrip
lines. Int. J. Com. Dig. Sys. 3(3), 189–195 (2014)
22. D. Bellan, S. Pignari, G. Spadacini, Characterisation of crosstalk in terms of mean value and
standard deviation. IEEE Proc.: Sci. Meas. Technol. 150(6), 289–295 (2003)
23. Y. Eo, W.R. Eisenstadt, J.Y. Jeong, O.K. Kwon, A new on-chip interconnect crosstalk model
and experimental verification for CMOS VLSI circuit design. IEEE Trans. Electron Dev. 47(1),
129–140 (2000)
9 Cartographical Analyses of Reflection and Transmission … 189
24. N. Lu, R.A. Wachnik, Modeling of resistance in FinFET local interconnect. IEEE Trans. CAS
I: Regul. Pap. 62(8), 1899–1907 (2015)
25. J. Bernal, F. Mesa, D.R. Jackson, Crosstalk in coupled microstrip lines with a top cover. IEEE
Trans. EMC 56(2), 375–384 (2014)
26. Irfanullah, S. Nariyal, S. Roy, M.M. Masud, B. Ijaz, B.D. Braaten, Analysis of the noise
voltage coupling (crosstalk) between right-handed and composite right/left-handed (CRLH)
transmission lines on printed circuit boards. IEEE Trans. EMC 55(4), 788–797 (2013)
27. S.W. Park, F. Xiao, D.C. Park Y. Kami, Crosstalk analysis method for two bent lines on a PCB
using a circuit model, IEICE Trans. Commun. E90-B(6), 1313–1321 (2007)
28. E.-P. Li, E.-X. Liu, L.-W. Li, M.-S. Leong, A coupled efficient and systematic full-wave time-
domain macromodeling and circuit simulation method for signal integrity analysis of high-
speed interconnects. IEEE Trans. Adv. Packag. 27(1), 213–223 (2004)
29. S.-C. Lin, Coupled-line filters with stub-embedded resonators using accurate admittance-
transformer feeds for flexible terminations. IEEE Trans. MTT 62(12), 2911–2922 (2014)
30. R. Zhang, L. Zhu, Synthesis of dual-wideband bandpass filters with source-load coupling
network. IEEE Trans. MTT 63(3), 441–449 (2014)
31. Y.-C. Lin, T.-S. Horng, H.-H. Huang, Synthesizing a multiband LTCC bandpass filter with
specified transmission- and reflection-zero frequencies. IEEE Trans. MTT 62(12), 3351–3361
(2014)
32. S. Hafizi-Moori, E. Cretu, Weakly-coupled resonators in capacitive readout circuits. IEEE
Trans. CAS I: Regul. Pap. 62(2), 337–346 (2015)
33. M. Khalaj Amirhosseini, Determination of capacitance and conductance matrices of lossy
shielded coupled microstrip transmission lines. Prog. Electromagn. Res. 50, 267–278 (2005)
34. J. Cho, E. Song, H. Kim, S. Ahn, J.S. Pak, J. Kim, J. Kim, Mixed-mode ABCD parameters:
theory and application to signal integrity analysis of PCB-level differential interconnects. IEEE
Trans. EMC 53(3), 814–822 (2011)
35. B. Ravelo, Theory on asymmetrical coupled-parallel-line transmission and reflection zeros.
Int. J. Circ. Theor. Appl. 45(11), 1534–1551 (2017)
Chapter 10
Analytical Modelling of Interbranch
Coupling Effect on Coupled Microstrip
Tree PCB Interconnects
Blaise Ravelo
10.1 Introduction
The EMC, EMI, SI and PI phenomena become one of the most crucial effects when
the operating frequency, processing speed and integration density of digital electronic
systems (DRAM, MPU, DIMM packages …) are increased [1–3]. Design and fabri-
cation engineers have to respect required compliances notably on the interconnection
networks as memory buses (DDR4, GDDR5, XDR, IO2 and HBM), front-side bus
(quick path interconnect and hypertransport), cable (USB, HADMI and FireWire Cat
X) and Ethernet (XAUI, XFI, CEI-6GLR and SONNET) [3]. Figure 10.1 illustrated
an example of complex signal distribution interconnect tree [4]. Such a structure in the
modern high-speed memory system is challenging on the SPI and EMC constraints
effects on interconnect lines (ILs) as channel attenuation, crosstalk, reflection, delay
and distortion [5, 6]. So, prediction methods were forwarded based on various IL
model-based approaches as RC- and RLC-lines [7, 8].
More generally, different models of interconnect tree topologies were devel-
oped [9–12]. But those models are not valid for structures presenting asymmetrical
behaviours as comb tree depicted in Fig. 10.2. So, analytical model including the
unbalanced interconnect tree was established recently in [13]. In first step, this analyt-
ical approach was initiated by assuming that the IL as equivalent to its RLCG model
by taking into account the frequency variation of per unit length parameters (Ru (f ),
L u (f ), C u (f ) and Gu (f )). In the second step, the model was fundamentally imple-
mented via the circuit equivalent approach single input multiple output (SIMO) to
single input single output (SISO) [13].
The analytical operations were handled with ABCD-to-Z matrix to generate the
IL impedance matrix, Z-to-Y matrix to extract the equivalent parallel lines and Y-to-
ABCD matrix transforms (to express the equivalent transfer parameters) proposed in
B. Ravelo (B)
Graduate Engineering School, ESIGELEC, Sotteville les Rouen, Seine-Maritime, France
e-mail: blaise.ravelo@yahoo.fr
Fig. 10.1 Electronic boards with complex interconnect tree network [2]
[14, 15]. Then, the output responses across the different path from the single input N in
to any output N m (m = {1, …, 8}) were mathematically formulated via the voltage
transfer functions:
Vm
Hm = . (10.1)
Vin
At this stage, this modelling method was applied to perfectly isolated asymmet-
rical tree structure. Nevertheless, in the context of system as high-density PCBs,
10 Analytical Modelling of Interbranch Coupling Effect on Coupled … 193
Figure 10.3a represents the circuit configuration of the 2:1 asymmetrical tree. The
single input with internal impedance Z s is excited by vin . The asymmetrical tree
outputs are loaded by impedances Z L1 and Z L2 , and the branches are essentially
comprised of pieces of input lines TL0 (connected between nodes N in and N 0 ) and
output lines TL1 (between N 0 and output node N 1 ) in parallel output lines TL21
cascaded with TL22 (between N 0 and N 2 ). We will manage a comparison between
this isolated circuit tree with the tree introduced in Fig. 10.3b in the presence of
neighbouring line TLc placed in proximity of TL21 . So, we propose to replace TL21
by piece of lines TL21 , TL21 and TL21 cascaded by choosing the length of TL21
as same as TLc here assigned as d c .
(b)
ZL3 ZL4
TLc
Zs TL0 N0
Nin TL21' TL21" TL21'"
V in TL1 TL22
dc
N1 N2
ZL1 V1 ZL2 V2
194 B. Ravelo
As described in [18], the overall coupled line impedance and admittance matrices
are denoted:
and
with j is the complex number and ω the radian frequency. The even and odd mode
impedances and electrical angles of coupled branches TL21 − TLc are denoted (Z e ,
Z o ) and
θe = γe · dc , (10.4)
θo = γo · dc . (10.5)
While the ABCD matrix of the coupled lines (TL21 − TLc ≡ TL21c ) is analytically
defined as:
T (TL21c ) = eig[Z · Y ] · [Tx ] · eig[Z · Y ]−1 k,l={1,2} , (10.8)
Figures 10.4 and 10.5 represent the SISO equivalent circuits of 1:2 tree sketched in
Fig. 10.3, respectively, for the electrical paths N in N 1 and N in N 2 .
We can see that the reduced circuits are in presence of parallel impedances Z in,m
(m = 1, 2) expressed with the classical relation between impedance and ABCD
matrices. After integration of the parallel impedances including the coupling effects,
the asymmetrical tree transfer functions are mathematically established for m = {1,
2}:
1
Hm = . (10.10)
[T (TLNm Nin )]1,1
By combining all the previous analytical elements of each piece of lines defined
before, we have the associated ABCD matrix for each output branches written as:
Nin
V in TL1
N1
ZL1 V1
(b)
Zs TL0 N0 Zin2'
Nin
V in TL1
N1
ZL1 V1
196 B. Ravelo
(b) TL0
Zs N0
Nin TL21c
V in Zin1 TL22
N2
ZL2 V2
1 Zs 1 1 1 1
T (TL N1 Nin ) = · [T (TL0 )] · · [T (TL1 )] · .
1 0 1/Z in2 0 1/Z L1 0
(10.11)
⎧
⎫
⎪
⎪ 1 Zs 1 1
⎪
⎪
⎨ 1 0 · [T (TL0 )] · 1/Z in1 0 · T (TL21 ) · ⎬
T (TL N2 Nin ) =
. (10.12)
⎪
⎪ 1 1 ⎪
⎪
⎩ T (TL21c ) · T (TL21 ) · ⎭
1/Z L2 0
Figure 10.6 displays the schematic layout illustrating the configuration of the 1:2
10 Analytical Modelling of Interbranch Coupling Effect on Coupled … 197
d3
w2
for the numerical analyses:
s
d 1 = 2 mm, d 2 = 4.5 mm, d 3 Zs
w1
= 1.5 mm, d 4 = 1 mm, d 5 =
Vin(t)
d6
6 mm, d 6 = 1.5 mm, d 7 =
2 mm, w1 = 0.3 mm, w2 =
d7
d4 d5
0.1 mm and s = 0.1 mm
asymmetrical interconnect tree considered for the numerical application. This struc-
ture is comprised of the microstrip interconnect tree driven by the numerical source
vin with internal impedance Z s = 25 and loaded by capacitors Z L1 = Z L2 = 10 pF.
The electrical paths N in N 1 and N in N 2 are set with physical lengths, respectively,
2.5 and 9 mm. The perturbation line presents 7.5 mm physical length and loaded
by Z L3 = 25 and Z L4 = 10 pF. Comparisons between the responses of the 2:1
tree without and with the perturbation coupling lines were performed by plotting the
outputs (v1 , v2 ) and (vc1 , vc2 ), respectively.
with:
− jεr tan(δ)
a= , (10.14)
ln ffHL + j f0
+ j f0
Figure 10.7 display the frequency responses of the structure presented in Fig. 10.6.
By using expressions (10.9), (10.10) and (10.12),
V1,2 ( jω)
H1,2 ( jω) = , (10.15)
Vm ( jω)
and:
Vc1,2 ( jω)
Hc1,2 ( jω) = , (10.16)
Vin ( jω)
(a)
H dB H cdB φ φc
70 100
40 -100
H 1 (f) (dB)
φ (H 1 ) (∞)
10 -300
-20 -500
-50 -700
0 2 4 6 8 10
Frequency (GHz)
(b) H dB H cdB φ φc
70 0
35 -200
φ (H 2 ) (∞)
H 2 (f) (dB)
0 -400
-35 -200
-70 -200
0 2 4 6 8 10
Frequency (GHz)
Fig. 10.7 Frequency responses of the 2:1 asymmetrical tree presented in Fig. 10.6 without (solid
lines) and with (dashed lines) coupling [21]
10 Analytical Modelling of Interbranch Coupling Effect on Coupled … 199
represent the transfer function corresponding to the electrical path N in N 1,2 without,
respectively, with the perturbation line. A significant resonance effect is occurred at
about 0.61 GHz along the path N in N 1 . This is due to the stub effect from the other
branch of the tree. As we can see, due to the crosstalk between the coupled branches
of the tree, the coupled responses are influenced notably above 5 GHz.
One emphasizes that the CPU time of the whole method implemented into MAT-
LAB runs with PC equipped by Windows 7 having Intel® Core™ i5-2467M CPU
@1.6 GHz 4 Go RAM was of about hundred milliseconds.
During the numerical test for this time-domain investigation, a high-speed mixed
signal represented by eight bits input data “01001000” was assumed as input. This
data was assigned as trapezoidal signal with 0.5 Gbps rate and 150 ps rise-/fall-times.
Figure 10.8 displays the computed results. The input is traced in solid bold blue line
and the asymmetrical tree outputs without (in solid red lines) and with (in dashed
green lines) the perturbation lines.
These results highlight and enable to predict rapidly and easily the influence of the
asymmetrical interconnect distribution tree. Emphatically, due to the asymmetrical
behaviours of the two input branches, we can see that the outputs are completely
different. In all cases, it can be underlined that the data SI was significantly degraded
with considerable distortion. Moreover, the propagation delays are of about 0.22 ns
for the electrical path N in N 1 and 0.5 ns for the path N in N 2 . A reflection effect is also
observed on the path N in N 1 which is related due to the resonance effect occurred in
Fig. 10.8a.
In inference of this study, we can emphasize that:
• The output signal shapes are rather preserved because the bandwidth of the input
data is quite lower than 6 GHz.
• The time-delay between the two responses is lower than 200 ps.
• The slight variation of the output transient voltage amplitudes of about 10% was
found when placing the perturbation at 100 µm of the tree.
200 B. Ravelo
(a)
1.0
V in
V1
V c1
Voltage (V)
0.5
0.0
0 4 8 12 16
Time (ns)
(b)
1.0 V in
V2
V c2
Voltage (V)
0.5
0.0
0
0 4 8 12 16
Time (ns)
Fig. 10.8 Eight bits data responses of the 2:1 asymmetrical tree presented in Fig. 10.6 without
(solid lines) and with (dashed lines) coupling [21]
10.2.4.5 Conclusion
data with 0.5 Gbps rate. The model is beneficial in terms of flexibility, simplicity
and computation speed. Meanwhile, the present model is helpful for the design and
manufacture engineers for assessing the degradation of the sharing high-speed signal
in asymmetrical tree.
In the continuation of this work, I am looking out on the application of the
behavioural asymmetrical model proposed hereby for the miniaturized microelec-
tronic interconnect systems packaging structures. The future study will be based on
the SI and EMC/EMI modelling principle by taking into account the undesirable EM
coupling influences. The prediction of the high-density interconnect effect allows to
optimize the packaging structures and also probably establishes a post-processing
technique for the signal degradation compensation.
Furthermore, we can also foresee to extend the analysis examined in this chapter
for generalized principle of integrated circuit. The particularity of the integrated inter-
connects as 3D TSV lies on the influence of the interbranch inductive and capacitive
couplings.
For this reason, the unbalanced tree interconnects modelling with interbranch EM
coupling effect is developed in this section [22]. Section 10.3.1 describes the compu-
tational theory of interbranch coupled unbalanced single-input double-output (1:2)
tree. Section 10.3.2 is the unbalanced 1:2 tree input impedance and the VTF model
validations with ADS® simulations. Section 10.3.3 is the conclusion.
The posed problem [22] can be traduced by the unbalanced 1:2 tree interconnect
modelling. It consists in the transformation of the initial tree network introduced in
Fig. 10.9 into the systemic model depicted in Fig. 10.10. The voltages across the
nodes M a , M b and M c are, respectively, denoted Va = VMa , Vb = VMb and Vc = VMc
(Fig. 10.11).
The system VTFs and overall input impedance are defined by:
with ω is the angular frequency. The equivalent circuit diagram of the unbalanced
1:2 tree can be elaborated by considering the impedance or Z-matrix of the coupled
TL (CTL). The modelling method can be established from the equivalent circuit dia-
gram presented in Fig. 10.12. The unbalanced tree structure can be transformed as an
electrical network mainly constituted by elementary TLs. We assume that the input
is connected with the TL TL Ma M0 characterized by TL(Z a (jω), γ a (jω)). TL M0 Mb and
TL M0 Mc , respectively, characterized by TL(Z b (jω), γ b (jω)) and TL(Z c (jω), γ c (jω))
constitute the output branches. With ξ = {a, b, c}, Z ξ (jω) is the characteristic
impedance and
θξ = βξ dξ . (10.21)
These output networks can be represented by the coupled TL matrix [Z]CTL and
the output TL TL(Z b (jω), γ (jω)) associated to the electrical length θ . This CTL
structure is assumed as an octopole with Ports ➊ and ➋ interconnected, and Ports ➌
and ➍ are connected to output loads Rb and Rc . Each access port m (m = {1, 2, 3,
4}) is traversed by branch currents I m .
Let us denote C the coupling coefficient between the coupled lines connecting
Ports ➊ and ➋ and Ports ➌ and ➍, and Z 0 (jω) is the characteristic impedance of
each elementary line. According to the TL theory, the even- and odd-characteristic
impedances Z e and Z o of the associated coupled lines constituting the unbalanced 1:2
tree is defined in (9.33). The lengths of the input and output TLs can be characterized
by the resonance angular frequencies ωξ which are linked to the physical length by
the relation:
2π v
dξ = , (10.22)
ωξ
with ξ = {a, b, c} and the wave speed v. Moreover, the physical length difference:
d = db − dc , (10.23)
θ = βb d = βb (db − dc ). (10.24)
π ω(ωc − ωb )
θ = . (10.25)
2ωb · ωc
Z 0 ( jω)
Z ( jω) = . (10.26)
tan h γ ( jω)
The adopted methodology to solve the posed problem is fundamentally based on the
calculation of the branch currents. The Z-matrix of the tree coupled branches and
10 Analytical Modelling of Interbranch Coupling Effect on Coupled … 205
the access line ABCD matrices must be expressed in function of the interconnect
structure parameters. The four-port CTL structure constituting the unbalanced 1:2
tree can be represented by the equivalent 4 × 4 Z-matrix analytically expressed as:
⎡ ⎤
Z 11 Z 12 Z 13 Z 14
⎢ Z 21 Z 22 Z 23 Z 24 ⎥
[Z ] = ⎢
⎣ Z 31
⎥ (10.27)
Z 32 Z 33 Z 34 ⎦
Z 41 Z 42 Z 43 Z 44
By denoting α the attenuation constant and d the physical length, based on the
microwave theory and due to the symmetry, the matrix elements are defined as:
⎧ Z e +Z o
⎪
⎪ Z 11 = Z 22 = Z 33 = Z 44 =
⎪
⎨Z
2 tan h(αd+ jθ)
Z e −Z o
12 = Z 21 = Z 34 = Z 43 = 2 tan h(αd+ jθ)
Z e −Z o , (10.28)
⎪
⎪ Z = Z 31 = Z 24 = Z 42 =
⎪
⎩
13 2 sin h(αd+ jθ)
Z e +Z o
Z 14 = Z 41 = Z 23 = Z 32 = 2 sin h(αd+ jθ)
with
πω
θ= . (10.29)
2ωc
The abstracted topology equivalent to the unbalanced 1:2 tree structure under
study can be represented as highlighted in Fig. 10.12. This topology enables to
realize the theorization of the problem with the analogue mathematical concept. The
branch currents [I1 , I2 , I3 , I4 , Ia , Ib ] are assumed as the unknown variables which
must be expressed in function of the input excitation source V a . Meanwhile, the
problem solution can be reformulated as the calculation of the tree branch currents
I a , I a , I 1 , I 2 , I 3 , I 4 and I b . The algebraic solution can be determined from linear
equations derived via the combination of the impedance and ABCD matrices in
(10.29), (10.31) and (10.32), and the Ohm’s laws applied to the output loads Z b and
Z c:
Vb = −Z b · Ib , (10.32)
206 B. Ravelo
V4 = −Z c · I4 . (10.33)
Ia = I1 + I2 , (10.34)
and
V1 = V2 . (10.35)
By taking this condition into account, the following synthetic equation system can
be deduced from the access line ABCD matrices associated to (10.30) and (10.31):
⎧
⎪
⎪ Va = Aa V1 + Ba (I1 + I2 )
⎨
Ia = Ca V1 + Da (I1 + I2 )
. (10.36)
⎪ V3 = Ab Vb − Bb Ib = −(Ab Z b + Bb )Ib
⎪
⎩
I3 = Cb Vb − Db Ib = −(Cb Z b + Db )Ib
[V ] = [V1 V2 V3 V4 ], (10.37)
and
[I ] = [I1 I2 I3 I4 ], (10.38)
corresponding to the configuration of the coupled lines presented in the circuit dia-
gram of Fig. 10.12 are linked to the Z-matrix defined in (10.27) by the equation
system:
⎧
⎪
⎪ V1 = Z 11 I1 + Z 12 I2 + Z 13 I3 + Z 14 I4
⎨
V1 = Z 21 I1 + Z 22 I2 + Z 23 I3 + Z 24 I4
[V ] = [Z ]CPTL [I ] ⇔ . (10.39)
⎪
⎪ V = Z 31 I1 + Z 32 I2 + Z 33 I3 + Z 34 I4
⎩ 3
V4 = Z 41 I1 + Z 42 I2 + Z 43 I3 + Z 44 I4
The combination of (10.36) and (10.39) implies the following synthetic characteristic
equation system of the posed problem mathematical solution knowing the excitation
source V a . The main access branch currents I a and I b can be yielded from the solutions
via the ABCD matrices of TLs TL Ma M0 and TLPort➌- Mb in (10.31). Consequently, the
10 Analytical Modelling of Interbranch Coupling Effect on Coupled … 207
unbalanced 1:2 tree characteristic matrix derived from (10.30) and (10.31) can be
written as:
⎧
⎪
⎪ Z 11 + ABaa I1 + Z 12 + ABaa I2 + Z 13 I3 + Z 14 I4 = VAaa
⎪
⎪
⎪
⎨ Z + Ba I + Z + Ba I + Z I + Z I = Va
21 1 22 2 23 3 24 4 Aa ,
Aa Aa (10.40)
⎪
⎪ +B
⎪
⎪ Z I
31 1 + Z I
32 2 + Z 33 − A b Z b
Cb Z b +Db
b
I 3 + Z 34 4I = 0
⎪
⎩
Z 41 I1 + Z 42 I2 + Z 43 I3 + (Z 44 + Z c )I4 = 0
Ia = Ca Va +(Da AAa a−Ba )(I1 +I2 )
, (10.41)
Ib = Cb Z−Ib +D
3
⎡ Va ⎤ ⎡ ⎤
b
Z 11 + A Ba
Z 12 + A Ba
Z 13 Z 14 0 0 ⎡ ⎤
I1
⎢ Vaa ⎥ ⎢ ⎥
A a a
⎢ Ba
⎢ A ⎥ ⎢ Z 21 + Aa Z 22 + Aaa
B
Z 23 Z 24 0 0 ⎥ ⎢I ⎥
⎢ a ⎥ ⎥ ⎢ 2⎥
⎢ 0 ⎥ ⎢ ⎢ Z Z Z −
Ab Z b +Bb
Z 34 0 0
⎥ ⎢ ⎥
⎥ ⎢ I3 ⎥
⎢ ⎥ 31 32 33 Cb Z b +Db
⎢ 0 ⎥=⎢ ⎥ · ⎢ ⎥.
⎢ ⎥ ⎢ Z Z Z Z + Zc 0 0 ⎥ ⎢ I4 ⎥
⎢ Ca ⎥ ⎢ B 41 42 43 44 ⎥ ⎢ ⎥
⎣ A Va ⎦ ⎢
⎣ a −D
a
Ba
− D a 0 0 1 0
⎥ ⎣ Ia ⎦
⎦
a Aa Aa
0 0 0 1 0 0 C b Z b + Db Ib
(10.42)
By taking:
the VTF through the 1:2 tree electrical path M a M b can be expressed as:
√
Z b ( jω)Ib ( jω) 2Z e Z b (Z c + x Z o ) 1 − x 2
Hb ( jω) = = , (10.44)
−Va ( jω) χ2b x 2 + χ1b x + χ0b
where:
⎧
⎪
⎪ Ba ((Bb + Aa Z b ) − (Db + Cb Z b )Z c )
⎪
⎪ χ2 = 2Z o
b
⎪
⎪ −Aa (Db + Cb Z b )Z e2
⎪
⎪ ⎧ ⎫
⎪
⎪ ⎪ Aa (Aa Z b + Bb − (Db + Cb Z b )Z c )Z e2 ⎪
⎪
⎨ ⎪
⎨ ⎪
⎬
+4Ba (Bb + Aa Z b )Z c
χ1 =
b
. (10.45)
⎪
⎪ ⎪
⎪ Aa (4 Aa Z b + Bb ) − (4Ba (Db + Cb Z b ) ⎪
⎪
⎪
⎪ ⎩+ Zo Ze ⎭
⎪
⎪ +Aa Db Z c + Aa Cb Z b Z c )
⎪
⎪
⎪
⎪ Ba (Bb + Aa Z b ) + Z c (Aa Bb − Ba Db
⎪
⎩ χ0b = 2Z e Ba
+(Aa2 − Ba Cb )Z b ))
Similarly, the VTF equivalent to the electrical path M a M c can be written as:
208 B. Ravelo
√
Z c ( jω)I4 ( jω) 2Z c Z e 1 − x 2 [Z o x(Cb Z b + Db ) + Bb + Aa Z b ]
Hc ( jω) = = .
−Va ( jω) χ2b x 2 + χ1b x + χ0b
(10.46)
This section is focused on the validations of the developed unbalanced 1:2 model.
Two POCs of unbalanced 1:2 trees are designed by considering the aspects with
and without interbranch coupled branches. The POC modelled computed results are
compared with simulations run in the ADS® environment of the electronic circuit
designer and simulator. AC simulations are considered by assigning the voltage
excitation source V a with 201 frequency samples from 0.1 to 2 GHz.
The POC was designed with arbitrary parameters which prove the influence of the
interbranch coupling on the two VTFs. The microstrip-line effective permittivity and
10 Analytical Modelling of Interbranch Coupling Effect on Coupled … 209
Fig. 10.13 3D design of the POC asymmetrical 1:2-tree microstrip structure a with and b without
output interbranch coupling [22]
characteristic impedances were extracted based on the microstrip TL theory. The cou-
pled branch parameters (C = −10 dB, Z e = 69.37 , Z o = 36.04 , s = 40 µm)
were extracted. The modelling and simulations were performed with these differ-
ent parameters. However, the entire proposed model-computed results were realized
with MATLAB programming. During the calculations, the ideal parameters were
supposed independent of the frequency, and the TL losses were neglected. The equiv-
alent model was developed by assuming the TL Ma M0 (wa = 227 µm, d a = 23.7 mm,
Z a = 50 ) with the frequency quarter wavelength f a = 2 GHz, and the coupled
lines TL M0 Mb (wb = 223 µm, d b = 73 mm, Z b = 56.5 ) and TL M0 Mc (wc = 223 µm,
d c = 49 mm, Z c = 56.5 ) are defined with the frequency quarter wavelengths
f b = 1 GHz and f = 2 GHz. The minimal physical width of the microstrip struc-
ture which can be fabricated with the equipment available in our laboratory is limited
to 300 µm. For this reason, the POC fabricated prototypes are not available for the
present study. The branch currents were computed.
210 B. Ravelo
In this case, the impedance loads are assigned as lumped resistors with nominal values
Z b = 50 and Z c = 100 . The VTF magnitudes |H a | and |H b | are, respectively,
displayed in Fig. 10.14a, b. These results illustrate the relevance and effectiveness of
the developed model for the interbranch coupling phenomenon prediction. Without
coupling, more accentuated resonance effects are observed at the terminal M a and
M b.
The interbranch coupling effects can be predicted by the proposed computation
method in good agreement with the simulations from very low frequencies to 2 GHz.
The same remark is found with the input impedance magnitude |Z in | of the overall
structure plotted in Fig. 10.15. The VTF model accuracy presents error absolute max-
imal value of about 1 dB. The highest absolute differences between the simulations
and modelled results are reasonably appeared around the resonance frequencies. The
main difference between the model and the reference simulations is caused by the
characteristics of the elementary TLs constituting the unbalanced tree structure. Fur-
thermore, these discrepancies increase at higher frequencies. Such effects are mainly
Fig. 10.14 Comparison of modelled and simulated resistive loaded tree interconnect VTF
magnitudes: a |H b | and b |H c | [22]
10 Analytical Modelling of Interbranch Coupling Effect on Coupled … 211
due to the influence of the frequency on the TL EM and electrical parameters as the
skin depth effect and the substrate dispersion.
These computation errors are also added to the numerical computation inaccu-
racies. To generate the modelled computed results with the assigned samples, the
computation speed was less than one millisecond by using a PC equipped with
a single-core processor Intel® CoreTM i3-3120M CPU @ 2.50 GHz and 8 GByte
physical RAM with 64-bits Windows 7.
In this case, the impedance loads are constituted by arbitrary chosen lumped resistor
Z b = 50 and capacitor Z c = 1 pF. The frequency simulations were carried out with
the unbalanced 1:2 tree structure by sweeping the AC source V a frequency. Then,
comparison between the simulations and computed models is realized. The obtained
VTF magnitudes |H a | and |H b | are, respectively, displayed in Fig. 10.16a, b. Once
again, without coupling, the resonance effects are occurred slightly at lower frequen-
cies. The simulated and modelled VTFs are in good agreement for the different types
of output loads Z b and Z c . A notable well-correlated behaviour of the VTFs versus
frequency is observed with simulations and the developed modelling methods in
the considered broadband frequency band. The comparison between the associated
input impedances magnitude |Z in | can be seen in Fig. 10.17. Similar to the previous
case, the interbranch coupling influences obviously, the unbalanced tree frequency
responses notably when the frequency is higher than 0.5 GHz. Despite the coher-
ent behaviour between the modelled and simulated results, numerical discrepancies
appear around the resonance frequency situated between 0.5 and 1 GHz.
This noteworthy deviation is mainly due to the approximation of the TL electrical
and EM characteristics which are assumed to be independent to the frequency during
the computation process.
212 B. Ravelo
Fig. 10.16 Comparison of the modelled and simulated capacitive loaded tree interconnect VTF
magnitudes: a |H b | and b |H c | [22]
10.3.3 Conclusion
matrix and the access line ABCD matrices. The equivalent topology enables to tra-
duce the system into the problem mathematical abstraction. The VTF of the tree
input-output electrical path and the overall circuit input impedance are established.
Two POCs constituted by unbalanced 1:2 tree with and without interbranch cou-
pling are designed. The modelled and simulated tree input-output VTFs and also the
input impedance are compared via AC simulations from 0.1-to-2 GHz. Good agree-
ments between simulations and the models are observed. The proposed computation
method is more efficient in terms of precision with the EM coupling influence com-
pared to the methods available in [23–25] which are dedicated to the linear tree VTF
modelling.
References
11.1 Introduction
With the tremendous trend on the electronic circuit design shirking size, the electro-
magnetic compatibility/interference (EMC/EMI), the signal integrity and the temper-
ature influence become critical effects [1–5] which must be integrated to the design
and fabrication phases. These undesirable physical phenomena are more and more
crucial for the deep submicron VLSI [1, 2]. It was found that the thermal effects
can induce particular phenomenon as the electromigration in the integrated circuits
[3, 4]. Moreover, the clock signal performances can be degraded due to the thermal
influence on the substrate [5]. More recently, electronic research and design engi-
neers experimented that the temperature effect and the moisture are susceptible to
degrade the PCBs global performance [6]. Due to the constant increase of integration
density, the substrate material characterization with the frequency and temperature
dependence becomes a challenging subject for the microwave and integrated circuit
designers and manufacturers [2, 7]. Until now, very few investigations on the tem-
perature influence to the electronic devices’ performance as the integrated [8] and
hybrid [9] circuits are available in the literature. Despite the developed EM classi-
cal modeling, simulation and test techniques [10, 11], the simultaneous influence
of electrical, EM and temperature effects on the microwave circuits and devices
remains an open question for the electronic research and design engineers. Further-
more, most of the existing classical EM characterization techniques dedicated to the
dielectric materials are based on the consideration of S-parameters by using waveg-
uide structures. Some of the available techniques are essentially carried out with the
through-reflect-line calibration [11], coplanar lines [12–14] and split-ring resonators
B. Ravelo
Graduate Engineering School, ESIGELEC, Sotteville les Rouen, Seine-Maritime, France
A. Thakur (B) · A. Saini · P. Thakur
Amity University, D-210, Gurugram, Haryana 122314, India
e-mail: athakur1@ggn.amity.edu
[15]. However, those techniques present a heavy process complexity, and they do
not include the temperature effect and require sophistical analytical approaches. The
existing measurement technique of complex permittivity which takes into account
the temperature variation as proposed in [16] was set with different shapes of res-
onant cavity. Nevertheless, such a characterization technique is particularly limited
in terms of the operating frequency. Furthermore, it cannot be used for predicting
the material characteristics in the baseband and microwave frequencies which can
be regularly required for the digital and microwave or mixed SI analysis [17]. Com-
plex EM material characterizations for EMC and EMI applications have been also
developed [18–21].
To overcome such technical limits, a relevant characterization method enabling
to predict the substrate material parameters in the UWB frequency (with baseband
frequency up to some GHz) with the use of microstrip technology can be envisaged.
In addition to the flexibility of microstrip line theoretical approach, it enables a
particularly simple experimental process as can be found in [17].
This chapter addresses the EM characterization technique of dielectric substrate
material with the temperature influence based on the microstrip line theory. For the
better understanding, the present section is principally organized in three sections.
Section 11.2 is focused on the methodological approach of the substrate material
characterization method under study. The fundamental formulas enabling to deter-
mine the permittivity will be proposed. The experimental application of the analytical
approach will constitute Sect. 11.3. A microstrip test structure printed on FR4 epoxy
will be investigated. The conclusion of the chapter is drawn in the last section.
extracted. Along with the chapter, the TL implemented in microstrip structure will
be investigated to establish the theoretical approach.
The configuration of the structure using the dielectric substrate characteristic
measurement method under study is shown in Fig. 11.1. It acts as a microstrip TL with
physical length d, metallization width w and thickness t which is printed on dielectric
substrate with height h. The substrate complex permittivity versus frequency f and
temperature T will be extracted from the TL measured two-port S-parameters.
Along the chapter, the reference impedance is denoted Z 0 = 50 . The two-port
system S-to-Z transform applied to the microstrip structure proposed in Fig. 11.1
allows to determine the access and transfer impedances. From where, the input
and transmitted impedance versus frequency f with respect to the S-parameters
measurement at the temperature T are defined as:
1 + S11 (jf , T ) − S22 (jf , T ) − S11 (jf , T ) · S22 (jf , T )
Z0 ·
+S12 (jf , T ) · S21 (jf , T )
Z11 (jf , T ) = , (11.1)
1 − S11 (jf , T ) − S22 (jf , T ) + S11 (jf , T ) · S22 (jf , T )
−S12 (jf , T ) · S21 (jf , T )
Z0 · 2S21 (jf , T )
Z21 (jf , T ) = (11.2)
1 + S11 (jf , T ) + S22 (jf , T ) − S11 (jf , T ) · S22 (jf , T )
+S12 (jf , T ) · S21 (jf , T )
On the one hand, these expressions permit to determine the TL matrix impedances
from the measured S-parameters by using a VNA. On the other hand, by assuming
that the TL is Z 0 -loaded and non-dispersive, the same input and transfer impedances
can be determined knowing the characteristic impedance Z c and the propagation
constant γ via the following expressions:
cos h(γ · d )
Z11 = Zc , (11.3)
sin h(γ · d )
Zc
Z21 = . (11.4)
sin h(γ · d )
218 B. Ravelo et al.
First and foremost, the dielectric characteristics are established from the microstrip
line analysis combined with the Bahl and Trivedi theory [17]. By definition, the
TL propagation constant is defined in function of the per-unit loss α and the phase
constant β via the basic expression:
From the member to member division of Eqs. (11.3) and (11.4), the following
expression of the propagation constant can be obtained:
1 Z11 (jf , T )
γ (jf , T ) = arg cos h . (11.6)
d Z21 (jf , T )
Then, the effective relative permittivity εreff (f , T ) of the dielectric material con-
stituting the microstrip TL can be established from the phase constant thanks to the
relation:
2
c · ∂β(f , T )
εreff (f , T ) = , (11.7)
2π ∂f
with c is the light speed in the vacuum. From where, the substrate relative permittivity
can be extracted with the expression [17]:
2εreff (f , T ) − 1 + a(w/h)
εr (f , T ) ≈ , (11.8)
1 + a(w/h)
c · α(f , T )
tan δ(f , T ) ≈ . (11.10)
π · f εreff (f , T )
The microstrip TL assumed as the circuit under test is photographed in Fig. 11.2.
It can be seen that this TL presents physical parameters w = 1.5 mm and d =
164 mm. The circuit is printed on FR4 epoxy substrate with height h = 0.8 mm and
copper etched with thickness t = 35 µm. The S-parameters of the circuit under test
were measured with the Agilent VNA 8502C from DC to 5 GHz. Moreover, after
calibration, it was placed in the furnace provided by THITEC® depicted in Fig. 11.2.
The furnace presents the physical size 50 cm × 47 cm × 50 cm or volume 117 L.
The furnace operates with a digital function allowing to program the temperature of
its internal chamber. For the present study, the temperature chamber was increased
from ambient T = 40–140 °C. In order to consider the temperature influence on the
circuit under test, the sampling of the measured S(jf, T )-parameters was recorded at
least after five minutes of the temperature change.
To check the relevance of the proposed method, preliminary numerical tests with
EM computation based on the MoM were performed.
Based on the experimental setup shown in Fig. 11.3, the DUT S-parameters were
measured from DC to 5 GHz by increasing the temperature step by step. For the
sake of simplicity and temperature influence illustration, only the measured S 11 (f,
T ) reflection and S 21 (f, T ) transmission parameters at T = {40 °C, 60 °C, 85 °C,
100 °C, 120 °C, 140 °C} are presented, respectively, in Fig. 11.4a, b.
It is noteworthy that during the test, there is no significant difference between
S 11 (f, T ) and S 21 (f, T ) for T varied from the ambient temperature of about 20–40 °C,
and the TL losses increase with the temperature. Then, the UWB characteristics of
the FR4 constituting the DUT substrate from DC to 5 GHz can be determined from
expressions (11.7) and (11.10).
Consequently, the measured dielectric constant εr (f , T ) is plotted in Fig. 11.5a. It
can be emphasized that the obtained relative permittivity εr (f , T ) is proportionally
inverse of the temperature T. Moreover, εr (f , T ) presents an absolute variation of
about 0.27 when increasing the temperature from 40–140 °C. Furthermore, Fig. 11.5b
depicts the loss tangent in the same temperature range. The loss tangent tan[δ(f, T )]
increases with T from about 0.025 with margin of about 0.02. The loss tangent is
S11, dB
-45 40°C
60°C
85°C
100°C
120°C
140°C
-60
0 1 2 3 4 5
Frequency, GHz
(b) 0
-2
S21, dB
40°C
-4 60°C
85°C
100°C
120°C
140°C
-6
0 1 2 3 4 5
Frequency, GHz
11.3.1.3 Conclusion
(a) 5.5
3.5
0 1 2 3 4 5
Frequency, GHz
(b) 0.06
40°C
60°C
85°C
100°C
0.04 120°C
140°C
tan(δ)
0.02
0.00
0 1 2 3 4 5
Frequency, GHz
Fig. 11.5 a Measured FR4 substrate dielectric constant and b loss tangent versus frequency and
temperature [22]
relevance of the method was verified with MoM numerical tests. Then, an applica-
tion with a prototype of microstrip TL printed on FR4 epoxy substrate is presented.
The dielectric constant εr (f , T ) and loss tangent tan[δ(f, T )] from DC to 5 GHz
were extracted by considering the temperature variation from 40 to 140 °C. More-
over, since the linear thermal coefficient of copper is 18.10−6 m/mK, the temperature
from 40 to 140 °C, i.e., by 100 °C will increase the length of microstrip line from
164 to 164.3 mm, i.e., by 0.18%.
The present method is limited to the prediction of the PCB substrate global char-
acteristic. The main drawback is for the characterization of typically micrometric
size materials. In addition, the method is particularly sensitive to the measurement
artefacts.
11 Temperature Effect Analysis on Microstrip Structure 223
|Vo|/|Vi| (dB)
for T = {40 °C, 100 °C, 20
140 °C} [23]
0
-20
0.5 1 1.5 2 2.5 3
Frequency (GHz)
Phase(Vo/Vi) (°)
-500
-1000
From the measured S-parameters, the transfer function was extracted based on the
MATLAB computations. Figure 11.6 displays the measured frequency responses
of the interconnect line under test from DC to 3 GHz. As expected, the microstrip
line presents strong peaks of resonance frequencies. It can be pointed out that the
temperature effects are represented by the shift of the resonance frequency. This
observation can be understood due to the influence of the temperature on the substrate
relative permittivity.
The frequency-dependent equivalent RLC model of the microstrip line versus
frequency was extracted as introduced in [8] by taking into account the substrate
parameters. To complete the performed analysis, the time-domain analyses were
carried out. The obtained results will be presented in the next paragraph.
In the first analysis, the responses of the microstrip line including the temperature
influence T = {40 °C, 100 °C, 140 °C} were generated with equivalent RLC model.
Then, nonlinear load is assumed as a varicap.
The overall system was excited by 8-bit digital serial source “01011000” with
0.1 Gbps rate. The rise/fall time was fixed equal to 0.5 ns. The transient analyses
were launched with sampling time of about 0.1 ns. When the load resistive effect
was neglected, the obtained results are displayed in Fig. 11.7. It can be seen that the
224 B. Ravelo et al.
Voltage (V)
1
-1
-2
0 20 40 60 80
Time (ns)
Further, transient analysis of coupled microstrip line combined with nonlinear load
is presented in this paragraph [23]. The layout of the coupled structure under study
is depicted in Fig. 11.9. It is comprised of a microstrip line having the same physical
characteristics (w = 1.5 mm, d = 162 mm) as the previous one shown in Fig. 11.9
neighbored with a piece of perturbation microstrip line TLp . The latter is also 35 µm
thickness Cu-metallized but the physical width w0 = 0.5 mm and length d 0 = 50 mm.
During the numerical tests, it was assumed that the two coupled lines are spaced by
0.5 mm.
11 Temperature Effect Analysis on Microstrip Structure 225
2
Voltage (V) (a): T=40°
1
-1
0 50 100 150
Time (ns)
2
Voltage (V)
(b): T=100°
1
-1
0 50 100 150
Time (ns)
2
Voltage (V)
(c): T=140°
1
-1
0 50 100 150
Time (ns)
Fig. 11.8 Time-domain response of the microstrip interconnect versus α 1 and temperature for R
= 100 and C 0 = 5 pF and α 2 = 10−6 [23]
Fig. 11.9 Layout of the coupled microstrip line with nonlinear load (R = {∞, 100 }, C 0 = 5 pF,
α 1 = 0.2 and α 2 = 10−6 ). The perturbation line is loaded by R0 = 50 [23]
226 B. Ravelo et al.
The considered structure was excited by the same source as in the previous paragraph
with 8-bit digital serial data “0101100” with 0.1 Gbps rate. The rise/fall time was
assumed to be equal to 0.5 ns. The computed results are plotted in Fig. 11.10a, b.
It can be seen that the output signal distortions are significantly visible because the
input signal rate was decreased which is equivalent to increase the signal bandwidth.
Then, the resonance effect is included in the bandwidth of the 10 ns-period input
data. The NL effect increases slightly the perturbation effect with the modification
of the over- and under-shoot. Furthermore, when the interconnect structure is open
loaded, due to the coupling influence, the resonance effect is increased when the
input is in “high level state”, vi = “1” and, contrarily, is decreased when the input is
in “low level state”, vi = “0”. However, the coupling effect increased and delayed the
instant peak time of the overshoot when R = 100 . This funding can be highlighted
in more general approach, for example, by using pseudo-random bitstream source
with 8-bit maximal sequence, one creates the eye diagram displayed in Fig. 11.11a,
b from the nonlinear loaded coupled structure shown in Fig. 11.9.
The synchronous clock signal is represented by a 0.1 Gbps rate periodic signal.
The rise/fall times and threshold voltage for detecting external trigger are, respec-
tively, 0.5 ns and 0.5 V. As can be seen in Fig. 11.9a, for the case of 0.1 Gbps rate
data, the height of the opening and the width of the cross are widely low for R =
∞. The eye pattern confirms once again that overshoot attains 100% of the input
-2
0 20 40 60 80
Time (ns)
(b) 1.5
1
Voltage (V)
0.5
-0.5
0 20 40 60 80
Time (ns)
11 Temperature Effect Analysis on Microstrip Structure 227
Voltage (V)
nonlinear loaded coupled
1
interconnect line for a R =
∞ and b R = 100 [23] 0
-1
-2
0 5 10
Time (ns)
(b) 1.5
1
Voltage (V)
0.5
-0.5
0 5 10
Time (ns)
11.3.3.3 Conclusion
An analysis methodology of the temperature effect onto the microstrip line behaviors
is developed in this chapter. The proposed thermal effect analysis is elaborated with
the combination of S-parameter theory. The reverse approach enables to realize an
extraction method of the microstrip line substrate characteristics dependently to the
frequency in function of the temperature. Then, an innovative approach of thermal
SI analysis is introduced.
References
Blaise Ravelo
B. Ravelo (B)
Graduate Engineering School, ESIGELEC, Sotteville les Rouen, Seine-Maritime, France
e-mail: blaise.ravelo@yahoo.fr