This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access

Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000.
Digital Object Identifier 10.1109/ACCESS.2017.Doi Number

Direct Time-Domain TAN Model of 3D Multilayer

Hybrid PCB: Experimental Validation
Z. Xu1, Y. Liu1, B. Ravelo1, Member, IEEE, J. Gantet2, N. Marier2 and O. Maurice3
1
Normandy University UNIROUEN, ESIGELEC, IRSEEM EA 4353, 76801 Saint Etienne du Rouvray, France
2
VALEO – GEEDS (Group Electronic Expertise and Development Services), 94000 Créteil, France
3
Ariane Group, Les Mureaux, France

Corresponding author: Z. Xu (e-mail: zhifei.xu@esigelec.fr).

This research work was supported by Euripides²-Eureka Program 2015 funded by the research project “Embedded Die Design Environment & Methodology
for Automotive Applications (EDDEMA)”.

ABSTRACT The multilayer technology constitutes the ultimate solution for the design of high density
printed circuit board (PCB). Challenging modelling method is required to predict the signal integrity (SI) of
multilayer PCB. This paper addresses an unfamiliar direct time-domain model of a 3D multilayer hybrid
PCB. The subnetwork primitive elements of the equivalent graph are constituted by lumped components,
interconnect lines, vias, pads and anti-pads. The tensorial analysis of networks (TAN) is used to solve the
problem related to the graph topology in function of the PCB design parameters. The TAN concept is based
on the interaction between the primitive elements. The mesh currents constitute the proposed computational
unknowns. The unfamiliar model is validated with a three-port network prototype constituted by six-layer
PCB including passive SMD components. In the frequency domain, S-parameter validation from 100-kHz
to 5-GHz is presented. By using 80-Mbps and 0.5-Gbps rate data patterns, the proposed TAN model is
validated by both simulations and measurements in the time domain. The transient results present vector
magnitude relative error accuracy lower than 15-%. Thanks to the computation speed and adaptability to
multilayer hybrid structures, the TAN model is a prominent approach for the SI and power integrity
analyses of 3D multilayer structures.

INDEX TERMS Circuit theory, Direct time-domain, Kron-Branin (KB) formalism, Interconnects,
Modelling method, 3D Multilayer PCB, Hybrid structure, Signal integrity (SI), Tensorial analysis of
networks (TAN).

I. INTRODUCTION unavoidably challenging [1]. Innovative design

To meet the public and industrial needs, the electronic
printed circuit boards (PCBs) must integrate various
functions in high density. Therefore, the power and signal
distribution interconnects are increasingly complex. The
high-speed PCB design and implementation become
unavoidably challenging [1]. Innovative design
methodologies [2] and modeling tools [3-4] are needed to
predict the interconnect effects on the PCB performance.
The undesirable phenomena as parasitic noises causing
signal integrity (SI), electromagnetic compatibility (EMC)
and electromagnetic interference (EMI) issues must be
investigated [5-7]. The analysis technique of SI and EMC
effects plays a major role during the design and
manufacturing of high density. Therefore, the power and
signal distribution interconnects are increasingly complex.
The high-speed PCB design and implementation become
methodologies [2] and modeling tools [3-4] are needed to
predict the interconnect effects on the PCB performance.
The undesirable phenomena as parasitic noises causing
signal integrity (SI), electromagnetic compatibility (EMC)
and electromagnetic interference (EMI) issues must be
investigated [5-7]. The analysis technique of SI and EMC
effects plays a major role during the design and
manufacturing of high-density interconnect (HDI) PCB [8].
The conventional computational methods in electronic
PCB engineering might be in good agreement with
measurements and provide better models of PCBs.
However, they are difficult to perform trouble shooting and
find the reasons of these results. The tensorial analysis of
networks (TAN) approach enables to study theoretically the
complex PCB electromagnetic behaviors. The exercises
done to compare numerical computations and equations

VOLUME XX, 2017 1

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access
formulated under the TAN formalism ensure the accuracy
of the results. In addition, the TAN formulation allows to
establish the equations of the problem quickly and
understand theoretically. This capability cannot be realized
with the nodal approaches. The TAN formalism is the
unique way to generate a compact and direct Lagrangian
expression of the circuit [9].
Recently, the designs of transmission line (TL) based-
structure and multilayer circuits using more effective
analysis technique were proposed [10-11]. With respect to
PCB technology progress, improved analysis techniques
and methodologies are necessary [12-14]. As reviewed in
[15], various computer-aided design (CAD), circuit
analyses and simulation tools have been developed.
However, global simulations of the complete multilayer
PCB using 3D full wave electromagnetic (EM) solvers are
time-consuming. Based on the TL RC- [13], RLC- [14] and
RLCG- [15] network models, analytical computational
approaches of planar PCB microstrip symmetric tree
interconnects were proposed. However, these approaches
cannot be used for higher level asymmetric trees [16]
because of the necessity to calculate successively with Z-,
Y- and T-matrices.
For this reason, modelling and simulation methods of the
multilayer interconnect structures constitute one of the
hottest topics for component packaging technology design
engineers [17-18]. Nowadays, the prediction of signal
distribution in the HDI as multilayer PCB traces is still an
open challenge [17]. Further research work is expected to
reduce the cost of SI and EMC characterizations of
multilayer PCBs with acceptable performances [19-20]. To
face up this challenge, the unfamiliar TAN was recently
introduced to analyze the SI and power integrity (PI) of
multilayer interconnect structures [21-22]. Simple tree
interconnect structures with single-input and double-outputs
were considered [21-22]. The proof-of-concept (POC)
structures were designed following the standard stack-up
described in [23]. The POCs were mainly composed of
elementary TLs, vias, pads and anti-pads. The TAN model
enables to compute rapidly the multilayer interconnect S-
parameters and voltage transfer function. The model
integrates the TL and via characteristics as proposed
respectively in [24] and [25-26]. Compared to the classical
circuit analysis methods, the TAN is advantageous in terms
of flexibility to be hybridized with different circuit
simulators [27-28]. In difference to the previous works [21-
22], this paper investigates the direct time-domain TAN
modelling of 3D multilayer hybrid PCB including SMD
components.
Before the elaboration of the paper main content, it is
worth to get clear insight about the TAN model. It acts as
an unfamiliar circuit analysis approach for most of
electrical and electronic engineers. Therefore, it would be
necessary to describe the state-of-the-art about the TAN
computational method. The TAN was initiated by the
2 VOLUME XX, 2017
2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Author Name: Preparation of Papers for IEEE Access (February 2017)
General Electric research engineer, Gabriel Kron in 1930s
[29-30]. Inspired by the concept, Kron established an
electrical machine unfamiliar modelling based on the tensor
calculus. He defined the TAN as modern physics
computational method. In the 1950s, he envisioned to
extend the TAN computational method for the universal
complex system analyses. To do this, he proposed an
innovative method to analyze the complex systems based
on the decomposition method called “diakoptics” approach.
In the 1950s, few research groups as the tensor society of
Japan explored the Kron’s formalism for the electrical and
electronics engineering [31]. In the 1960s, a complementary
approach of Kron’s formalism has been initiated by Branin
for the TL transient analysis [32]. The Branin model
enables to calculate the signal propagation through TLs by
considering the voltage source coupling between the TL
terminations [32-33]. Since the 1990s, Olivier Maurice and
his group have introduced an extension of Kron’s
formalism by creating the modified Kron’s method
(MKME) for the EMC complex system analyses [34-35].
By combining the Kron’s and Branin’s models, they
developed more general TAN computational method for
electronic systems including PCBs. More recently, the
Kron-Branin (KB) formalism was applied to the tree
interconnects of single layer planar PCB [36]. It was found
that the KB formalism enables to analyze rapidly and
accurately, the tree interconnect SI. The frequency-domain
KB model applied to simple cases of 3D multilayer PCB
was proposed in [21-22]. However, the TAN model was not
applied in fully time-domain to treat the problem of 3D
multilayer PCB structures. The present paper is devoted to
the direct time-domain TAN modelling applied to 3D
multilayer hybrid PCB with multi-branch interconnects. In
addition, the unfamiliar model is validated experimentally
with fabricated 3D six-layer hybrid PCB prototype.
The paper is organized in three main sections. Section II
focuses on the analytical modelling and the methodology of
the 3D multilayer PCB structure. After the problem
formulation, the graph topology is elaborated. Then, the
analytical abstraction of the graph with time-domain
operator is described. The validation results with a
prototype as proof-of-concept (POC) are examined in
Section III. Last, Section IV addresses the conclusion.
II. DIRECT TIME-DOMAIN TAN MODELLING
METHODOLOGY
This section focuses on the direct time-domain TAN
modelling methodology for 3D multilayer PCB. The graph
topology of this multilayer hybrid structure is elaborated.
The time-domain tensorial approach is analytically
described.
A. FORMULATION OF THE MULTILAYER PCB
STRUCTURE PROBLEM
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access
The considered hybrid structure can be assumed as a
multiport system. It is mainly constituted by the subnetworks
composed of interconnects, vias, pads and anti-pads. Each
subnetwork is transcribed in its equivalent electrical scheme.
The global system synoptic representing the posed-problem
is shown in Fig. 1. The 3D multilayer structure is a three-port
circuit. It is hybridized with surface mounted devices
(SMDs) RSMD and CSMD.
The structure is excited by the voltage source vin and
loaded by the resistor R0. The perspective 3D view of the six-
layer structure is depicted in Fig. 2(a). The electrical network
configuration can be referenced with the nodes Mn
(n={1,2,…,15}). For further illustration, Fig. 2(b) shows the
PCB configuration profile and top views. The TAN
modelling of the system under investigation is inspired from
the electrical equivalent circuit.
FIGURE 1. Global configuration of the hybrid 3D structure under study.
FIGURE 2. Illustrative configuration of (a) perspective and (b) profile
view of the structure under study.
2 VOLUME XX, 2017
2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Author Name: Preparation of Papers for IEEE Access (February 2017)
B. ELABORATION OF THE DIRECT TIME-DOMAIN TAN
MODEL
The KB graph shown in Fig. 3 is equivalent to the 3D hybrid
structure introduced in Fig. 2(a). To establish this graph, the
basic subnetworks must be defined as initial objects. The
subnetwork model combination gives the final KB graph.
With the diakoptics approach [21-22], the posed-problem can
be assumed as the combination of primitive objects. The
problem resolution is proceeded with the tensorial
mathematical conversion. The main parameters and
unknowns must be defined in function of the considered
structure.
1) TAN Model General Description
The MKME formalism [34-35] enables to group all electrical
interactions into the subsystems implemented in the same
tensor space. The abstracted TAN model is built from each
subsystem with different vector spaces of variables (as
currents and voltages) referenced in edge or mesh or branch
spaces. The extended TAN method can be summarized
mainly as follows:
• First of all, the global system can be decomposed into a
set of primitive objects;
• Each object needs to be represented as a graph element
in order to build the equivalent network comprised of
mesh, node, branch, edge and moment elements;
• The analytical tensorial translation is established with
the connection relation represented by mathematical
transformations corresponding to the scenario of
physical phenomena;
• And solve the analytical equations of the posed-problem
formulated as the tensor metrics of the system.
2) Graph Topology of the 3D Multilayer Hybrid PCB
As aforementioned, this TAN modelling must begin with the
structural segmentation analysis. It is necessary to elaborate
the graph in Fig. 3. The basic elements constituting the
primitive graph are:
• R1,2,3 = R0 are the resistive loads connected to the
interconnects on each port.
• Vias, pads and anti-pads are modelled by LC lossless
lumped networks [24-26];
• And interconnect TLs are parametrized by their
characteristic impedances Zk and their propagation
constants k with k = {1,2,…,9}.
These characteristics can be defined differently in function
of the implementation technology such as microstrip and
strip lines constituting the multilayer PCB. The internal
source covariables Em with m = {1,2,…,n} represent the
Branin’s electromagnetic force (EMF) coupling sources. The
KB graph topology presented in Fig. 3 is established from
these primitive elements. This graph includes the external
excitation voltage source Vin. It is composed of B=36
branches, M=23 meshes and nodes connecting different
components.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access
Author Name: Preparation of Papers for IEEE Access (February 2017)

FIGURE 3. KB equivalent graph topology of the 3D multilayer hybrid structure shown in Figure 2(a).

3) Mathematical Abstraction and Tensorial Object The rank 1 tensor Jm(t) with m={1,2,…,M} represents the
Definition contravariable current in the mesh space with dimension M:
The analytical abstraction consists in converting the graph
into tensorial objects. The network current and voltage  J 1 (t ) 
 2 
 J (t ) 
variables must be assigned as tensorial vectors reported in the
appropriate spaces. In this situation, the physical laws
governing the system are explored from mesh, nodal and J m (t ) =  . (4)
 M −1 
branch current relations. Based on the graph of Fig. 3, the t J ( t )
 JM t 
time-dependent excitation sources are assumed as a rank 1
 ( )
tensor covariant variable in the branch space:

Ea (t ) = Vin (t ) 0
The TAN is elaborated from the graph fictive mesh
currents. The compact multilinear relation is formulated by
the Einstein tensorial notation:
I b (t ) = Cmb J m (t ) .
In this compact tensorial equation, the rank 1 tensor Ib(t)
with b={1,2,…,B} represents the contravariable current in
branch space with dimension B:
 I 1 (t ) 
 2 
I (t ) 
I b (t ) =  . (3)
 
 B 
 I (t ) 
0 . (1) The rank 2 tensor Cmb represents the connectivity matrix. It
acts as a time-independent intrinsic parameter.
4) KB Characteristic Matrix in Mesh Space
The via tensors are merely deduced from the equivalent
lossless LC networks of the simple structure connecting the
nodes MbMb+1 as introduced in Fig. 4. The pads and via
(2)
holes are modeled as capacitors and inductors respectively.
The branch space tensor of via1 can be easily expressed
with the time-domain integro-differential operator as in (5).
This model can be extended and applied to all via structures
of the multilayer 3D structure. To extract the contravariant
mesh current, the analytical implementation of Kron’s
method allows changing the branch space into mesh space.

2 VOLUME XX, 2017

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access
Author Name: Preparation of Papers for IEEE Access (February 2017)

In branch space, the excitation sources are denoted as Ea.

Knowing the transpose of C denoted Ct, the external
covariable voltage sources are written in mesh space as
follows:

Em (t ) = C t Ea (t ) = V1 (t ) 0 0 0 0 0 . (11)
FIGURE 4. (a) Via structure and (b) its equivalent model.
The impedance comatrix in mesh space without Branin’s
1 t  internal coupling is given by:
  d () 0 0 
 C1 0  Z mn (t ) = C t Z ab (t )C . (12)
 d () 
 Z via1 (t ) =  0 L1 0  . (5)
 dt  The Branin’s internal coupling will be added and discussed
 1
t
 in the next section.
 0 0  d () 
 C2 0  C. COMPUTATIONAL APPROACH OF THE INNOVATIVE
DIRECT TIME-DOMAIN TAN
The branch space impedance matrix is denoted Zab(t). This subsection describes the most innovative part of this
Based on the Einstein notation, this relationship between paper. It is dedicated to the direct time-domain KB
the mesh and branch spaces is reduced as follows: formalism. In difference to the frequency domain, the time-
domain model is defined by the iterative time-difference
Em (t ) = Cma Ea (t ) , (6) resolution of discretized integro-differential equations.
Z mn (t ) = C Z ab (t )C ,
a b
(7) 1) Time-Domain KB Modelling Principle
m n
The time-domain analysis of the graph drawn in Fig. 3
 Z b1 (t ) 0 0 0 0  needs more exhaustive parameter analyses. The internal
 0 0 0 0  Branin’s EMF sources must be represented by the voltage
 
Z ab (t ) =  0 0 Z b10 (t ) 0 0 , (8) source vectors. All the EMFs in the considered network can
  be implemented into the internal coupling matrix in mesh
 0 0 0 0  space. Following the Branin’s model, this matrix size will
 0 0 0 0 Z bn (t )  be the same as Zmn(t) defined in (12). The mesh impedance
tensor and covariable sources are respectively expressed as:
where {a,b}=1,2,…,B and {m,n}=1,2,…,M. Zbn(t) is the
branch impedance expression. In this configuration, the   Z e (t )  
Kron’s connection matrix is defined from the relationship  1  
 
between branch and mesh currents in (4). The connectivity  
matrix compact form expressed in (10) is derived from the  
Z mn (t ) = C Z ab (t )C = 
t
 Z ek (t )  
coefficients of following mesh currents:  
 
  Z e7 (t )  
 I 1 = 1J 1 + 0 J 2 + ... + 0 J 23  
 , (13)

 I 4 = 0 J 1 + 0 J 2 + 1J 3 − 1J 4 + 0 J 5 + ... + 0 J 23 Em (t ) = C t Ea (t ) = 0 V1 (t − D0 ) 0 0 0 0 , (14)
 5
 I = 0 J + 0 J + 0 J + 1J + 0 J + ... + 0 J
1 2 3 4 5 23
with D0 is the TL induced delay. The formulation of the

 mesh tensor leads to the generalized Ohm’s law:
 I 36 = 0 J 1 + ...0 J 22 + 1J 23
(9)
Em (t ) + Em (t ) =  Zmn (t ) +  Zmn
 (t ) J n (t ) . (15)
1 0 0 0 0
 
  This relation represents the abstraction of the initial
0 0 1 −1 0 0 physical posed-problem sketched in Fig. 2(a) into a purely
C= 
0 0 0 1 0 0 mathematical problem. The mesh currents Jn(t) are the
  posed problem unknowns. In a nutshell, the problem can be
  summarized with the KB characteristic matrix:
0 0 0 0 0  (10)
 M KB (t ) =  Z mn (t ) +  Z mn (t ) . (16)

VOLUME XX, 2018 9

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access
Author Name: Preparation of Papers for IEEE Access (February 2017)

intrinsic time-domain sources of six-layer hybrid PCB

The unknown mesh currents J(t) are linked to the source shown in Figs. 2 is expressed as:
vector E(t) and the time domain impedance operator Z(t)
by:  
 
 Q10 ( t −  t ) -Q11 ( t −  t )   t / C6 +... 
 
E (t ) = Z (t ) J (t ) . (17)  
 ...+Z 4 J ( t − D4 ) -Q ( t − D4 )  t / CSMD
12 12

 
By considering the internal coupling sources from the  Q10 ( t − D4 ) -Q11 ( t − D4 )   t / C6 +... 
   
Branin’s model and the intrinsic sources, the previous  ...+Z 4 J ( t − D4 ) -Q ( t − dt )  t / CSMD
11 12 
equation becomes:  
W (t ) =  
 14 
 Q ( t −  t ) -Q ( t −  t )   t / C1 + ( Z5 − R2 ) J ( t − D5 )
15 16
T (t ) = E (t ) + W (t ) = Z (t ) J (t ) , (18) 
 
 Q ( t − D5 ) -Q ( t − D5 )   t / C1 + Z 5 J ( t − D5 ) 
14 15 15

where E(t) is the vector of the extrinsic sources and W(t) is  

 
the Wronskian vector of the intrinsic and Branin’s internal  
 Q ( t −  t ) − Q ( t −  t )  / C1 − ( R3 − Z 7 ) J ( t − D7 )
21 22 23
coupling sources. The time-domain extrinsic source vector 
 
is defined by:  Q 21 ( t − D7 ) − Q 22 ( t − D7 )  / C1 + Z 7 J 22 ( t − D7 ) 
    (23)

E (t ) = e0 (t ) e0 (t − D1 ) 0 0 . (19)
where Dk is the induced delay from the TL TLk for k =
{0,1,…,7},
2) Time-Difference Tensorial Expressions To confirm the relevance of the proposed direct time-
The time-domain KB model integrodifferential equations domain modelling, validation results with a 3D multilayer
depend on the properties of lumped via-pad inductors Lk based hybrid PCB prototype are examined in the next
and capacitors Ck with k={1,2,…} as illustrated in Figs. 5. section.
Accordingly, the time-dependent operator of:
• The inductor L is defined by: III. VALIDATION RESULTS
This section describes the validation results of the previously
t 
L  developed TAN modelling methodology. The designed and
J (t ) = V + J (t −  t )  . (20)
L t  fabricated six-layer hybrid PCB prototype is presented. The
KB model computations are validated by simulations and
In this expression, δt is the time differential, δt/L is the measurements.
inverse of impedance operator in the time domain, V is
the extrinsic source term, J(t-δt)L/δt is an intrinsic A. DESCRIPTION OF THE POC 3D HYBRID
source term, generated by the inductance load which STRUCTURE
should appear in the Wronskian vector W. As aforementioned in Section II, the device under test is a
• The capacitor time-domain operator is defined by the three-port hybrid multilayer circuit. It is a six-layer passive
instantaneous charge Q with respect to the number of structure with SMD components. The tested prototype is
meshes m and the following formula: built on the FR4 epoxy dielectric with relative permittivity
εr = 4.5, loss tangent tan(δ) = 0.012. The fabricated PCB
Q p (t ) = Q p ( t −  t ) + J p ( t ) .
metallization parameters are shown in Table I. The
(21)
interconnect TLs present the same physical width and
different lengths. Fig. 5(a) represents the SIwave® design
The mesh current J(t) is calculated at each instant time of the fabricated PCB prototype.
step with the capacitor Ck integral operator:
TABLE I
t t
 (t ) = (t −  t ) ,
t =n p m
J Q (22) PCB LAYOUT PARAMETERS
t =0
Ck Ck Board Board Board Substrate Loss Metal
width length height permittivity tangent thickness
30 mm 40 mm 1.6 mm 4.5 0.012 0.035 mm
where the left-side quantity δt/Ck is the impedance INTERCONNECT LINE PARAMETERS IN mm
operator in time domain and the right-side quantity is Width TL1 TL2 TL3 TL4 TL5 TL6 TL7
the element of Wronskian internal source vector W. 0.2 10.47 14.30 18.90 7.20 47.86 27.20 29.20
The time-domain impedance comatrix is derived from VIA AND SMD COMPONENTS PARAMETERS
these operators. Instead of the classical scalar product Via Pad Capacitor Package Resistor Package and
between the tensor elements, in this case, the computations diameter diameter and Value Value
must be performed with the finite time-difference. 0.4 mm 0.6 mm 1 nF - 0603 49.9 Ω - 0603
Accordingly, the innovative Wronskian covariable of the

VOLUME XX, 2018 9

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access
Author Name: Preparation of Papers for IEEE Access (February 2017)

The PCB stack up was designed with respect to the simulated and measured reflection and transmission
multilayer fabrication standards [23]. Layers 1, 3, 4 and 6 are parameters are plotted in Fig. 7(a) in linear scale and in Fig.
dedicated to the signal distribution. Layers 2 and 5 are the 7(b) in x-axis logarithmic scale.
power and ground planes respectively.
The signal, power, ground planes, layer interconnects,
anti-pads and vias are constituted by Cu-metallization. The
test signals are obtained via identical SMA connectors
placed at ports 1, 2 and 3. The top and bottom layers are
highlighted in Fig. 5(b). This layout figures out the
placement of the lumped SMD capacitor and resistor. The
photographs of the top and bottom views of the fabricated
PCB are displayed in Fig. 6.

(a)

FIGURE 5. (a) 3D design and (b) top/bottom layers of the fabricated PCB
prototype.

FIGURE 6. Photographs of the multilayer PCB top and bottom views.

B. S-PARAMETER FREQUENCY RESULTS

The proposed S-parameter analyses take into account the
realistic case of lossy elementary TLs. During the
computations, the SMD components are modelled by
SPICE data provided by the manufacturers. The frequency
domain computations have been performed from 100 kHz
to 5 GHz. The applicative test configurations enable to
verify the TAN model efficiency. The TL losses have been
integrated in the TAN model via the propagation constants.
For more realistic validation, comparisons of computed,
(b)
FIGURE 7. Comparisons of TAN computed, simulated and measured
transmission parameters of the fabricated PCB shown in Figs. 6: (a)
linear and (b) x-axis logarithmic scales.
As expected, the TAN computations are well-correlated to
the simulations but slightly different to the measurements.

VOLUME XX, 2018 9

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access
Author Name: Preparation of Papers for IEEE Access (February 2017)

The notable differences between the TAN computations,

simulations and measurements of S-parameters are The KB model and measurement results are in good
observed around the resonance frequencies. The deviations correlation. Significant distortions, delays and attenuations
at higher frequencies are mainly due to: are observed because of the hybrid multilayer interconnect
• the microstrip and strip line dispersion skin effects and structure effects and the real time measurement.
EM couplings, As the main part of the test signal energy is focused in
• and frequency dependent characteristic impedances and the frequency band lower than 1 GHz, the higher frequency
relative permittivity. effects are not significant in time-domain. The slight
differences on signal spikes and distortions between the
C. TIME DOMAIN RESULTS TAN and measured output are caused mainly by the
Time-domain experiments have been performed to validate numerical inaccuracies and the time steps. The difference
the developed direct time-domain unfamiliar TAN model. between the model and measurement results can be
To do this, the time domain computations described earlier predicted with the transmission coefficients plotted in Fig.
in Subsection II-D were implemented as a Python routine 7. When the input frequency bandwidth is higher, more
algorithm. The present results correspond to the circuit significant deviations appeared and represented by the
configuration with outputs at port 2 and port 3 loaded by output ripples showed in Fig. 10. Moreover, the
R0=50 Ω. The performed time-domain experimental setup is measurement results were generated with periodical
shown in Fig. 8. It consists in injecting the transient voltage sequences which charge the tested PCB capacitors in
pulses from the pattern generator to the 3D multilayer difference to the calculations performed with the initial
hybrid PCB. The input and output transient voltage plots condition vout_KB (t=0)=0.
are compared via the signal analyzer visualizations. The
pattern generator Infiniium DSA90404A provided by
Agilent® presents a maximum 600 MHz data rate.
As seen in Fig. 9, the time-domain outputs present
similar behavior. Therefore, only the output signal from
port 2 is displayed in this subsection. The time-domain
results have been visualized and recorded by a 4 GHz
Agilent® digital signal analyzer with 20 GSa/s. Two
different digital input sequence data with 80 Mbps and 0.5
Gbps rates were considered. The data pattern is typically
square wave signal corresponding to random sequences. As
illustrated in Figs. 9, it is a voltage signal having a rise time
of about 1 ns, fall-time of about 1.5 ns and amplitude of
about 2.5 V. This transient signal was injected to the port 1
input of the circuit under test. The experimental results
denoted by vout_meas and the TAN time-domain calculated
results denoted by vout_KB are performed with time-step
Δt=75 ps for 80 Mbps rate input.

FIGURE 9. Time-domain results of 80 Mbps rate input sequence with (a)

real and (b) 4-time scaled output plots.

Fig. 9(a) displays the comparison between the TAN

computation, SPICE transient simulation (“simu.”) and
measurement (“meas.”) results. The test results introduced
in Fig. 10(a) with 0.5 Gbps rate input presents the time-step
Δt=10 ps. For the better resolution, Figs. 9(b) and 10(b)
represent the plots of 4×vout_KB. The outputs of the tested
PCB are respectively displayed in solid and dashed lines.
In order to quantify these differences, the error vector
magnitude (EVM) modelled between calculation and
measurement results are evaluated via the relation:

FIGURE 8. Time-domain experimental setup.

VOLUME XX, 2018 9

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access
Author Name: Preparation of Papers for IEEE Access (February 2017)

kmax Full Numerical simulator based Cannot be used for lumped

 v
k =0
2
KB (tk ) − vmeas
2
(tk )  wave on solving Maxwell
equations which is better for
circuit
particularly
analysis
long
and
time
EVM = , (24) most of the elementary 3D consuming
kmax

v
k =0
2
KB
with tk = k t represents the sampling time varying from 0
to kmax. The EVMs related to 80 Mbps and 0.5 Gbps rate
inputs are respectively lower than 10-% and 15-%.
The performed TAN modelling computation speed is of
about tens milliseconds by using a PC equipped a single-
core processor Intel® CoreTM i7-3120M CPU @ 2.50 GHz
and 16 GBytes physical RAM with 64-bits Windows 10.
D. DISCUSSION ABOUT THE ADVANTAGES AND
DRAWBACKS OF THE TAN MODEL COMPARED WITH
THE OTHER CIRCUIT ANALYSIS SOLVERS
Table II describes the advantages and drawbacks of the
TAN modelling compared to the familiar computational
methods as SPICE, full wave and analytical approaches.
Despite the mentioned disadvantages, the authors future
work is to develop an improved method allowing to
increase the TAN model accuracy in function of the
specific applications.
( tk )
Classical
structure analysis
Based on the analytical Cannot be used to the 3D
analytical solution of TL theory which structure which cannot be
is good for the traditional represented as a TL
circuit analyses
TAN The fast solution for TL Sensitive to the inaccuracies
based on the Branin approach related to the constituting
and good for analyzing entire hybrid elements
hybrid system
IV. CONCLUSION
An innovative unfamiliar TAN modeling methodology of
3D multilayer hybrid PCB is developed in direct time-
domain. The TAN model is fundamentally based on the
graph description and tensorial analytical abstraction. In
difference to the conventional computational methods in
electronic PCB engineering, the direct TAN model enables
to treat the problem originally, here 3D hybrid PCB, with
the tensor calculus. The TAN approach basic principle is
described. The TAN model was applied to a complex 3D
multilayer PCB constituting the problem to be treated. After
the modelling of primitive sub-networks with lumped SMD
components, TLs, vias, pads, and anti-pads, the KB graph
of the structure under study is established. To solve the
problem, the graph was traduced into the tensorial objects.
The problem resolution is converted into the tensorial
calculations of the contravariant mesh currents of the graph.
To validate the developed direct time-domain TAN model,
a six-layer PCB is designed and fabricated. Then, this three-
port test prototype was modelled, simulated and measured.
As expected, TAN computed transient output voltages are
in good agreement with measurements. The time-domain
experiments have been performed with 80 Mbps and 0.5
Gbps rate square wave signal as input. A very good
prediction of the transient signal behaviors is verified by
simulations and experiments of tested PCB prototype.
In the future, the proposed TAN model could be useful
for the SI, PI and EMC analyses of multilayer HDI PCBs.

REFERENCES
[1] Roadmap, ITRS, Available Online, 2016, http://www.itrs2.net/itrs-
schedule.html
[2] R. M. Averill, K. G. Barkley, M. A. Bowen, P. J. Camporese, A. H.
FIGURE 10. Time-domain results of 0.5 Gbps rate input sequence with Dan-sky, R. F. Hatch, D. E. Hoffman, M. D. Mayo, S. A. McCabe,
(a) real and (b) 4-time scaled output plots. T. G. McNamara, T. J. McPherson, G. A. Northrop, L. Sigal, H. H.
Smith, D. A. Webber, and P. M. Williams, “Chip integration
methodology for the IBM S/390 G5 and G6 custom
TABLE II microprocessors,” IBM J. Res. Develop., vol. 43, no. 5/6, pp. 681-
COMPARISON OF THE TAN MODEL ADVANTAGES AND DRAWBACKS WITH 706, Sep./Nov. 1999.
THE FAMILIAR CIRCUIT SIMULATION SOLVER [3] B. Archambeault, C. Brench, and S. Connor, “Review of Printed-
Circuit-Board Level EMI/EMC Issues and Tools,” IEEE Trans.
Method Advantage Drawback
Electromagnetic Compatibility, vol. 52, no. 2, pp. 455-461, May
SPICE Based on the nodal analysis Difficult to analyze coupling
2010.
which is advantageous in and EM fields
[4] A. Ruan, J. Yang, L. Wan, B. Jie, and Z. Tian, “Insight into a generic
lumped linear circuit analysis
interconnect resource model for Xilinx Virtex and Spartan series
in both frequency and time
domain

VOLUME XX, 2018 9

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access
FPGAs,” IEEE Trans. CAS-II: Express Briefs, vol. 60, no. 11, Nov.
2013, pp. 801-805.
[5] J. Kim, and E. Li, “Special issue on PCB level signal integrity, power
Integrity, and EMC,” IEEE Trans. Electromagnetic Compatibility,
vol. 52, no. 2, pp. 246-247, May 2010.
[6] C. Schuster, and W. Fichtner, “Parasitic Modes on Printed Circuit
Boards and their Effects on EMC and Signal Integrity,” IEEE Trans.
Electromagnetic Compatibility, vol. 43, no. 4, pp. 416-425, Nov.
2001.
[7] A. J. McDowell, and T. H. Hubing, “A Compact Implementation of
Parasitic Inductance Cancellation for Shunt Capacitor Filters on
Multilayer PCBs,” IEEE Trans. Components, Packaging and
Manufacturing Technology, vol. 57, no. 2, pp. 257-263, Apr. 2015.
[8] J. F. Buckwalter, “Predicting microwave digital signal integrity,”
IEEE Trans. Advanced Packaging, vol. 32, no. 2, pp. 280-289, May
2009.
[9] O. Maurice, A. Reineix, P. Durand, and F. Dubois, “Kron’s method
and cell complexes for magnetomotive and electromotive forces,”
Int. J. Applied Mathematics, vol. 44, no. 4, pp. 183-191, 2014.
[10] Z. A. Khan, “A Novel Transmission Line Structure for High-Speed
High-Density Copper Interconnects,” IEEE Trans. Components,
Packaging and Manufacturing Technology, vol. 6, no. 7, pp. 1077-
1086, July 2016.
[11] Z. Liu, L. Zhu, and G. Xiao, “A Novel Microwave Attenuator on
Multilayered Substrate Integrated Waveguide,” IEEE Trans.
Components, Packaging and Manufacturing Technology, vol. 6, no.
7 pp. 1106-1112, July 2016.
[12] A. E. Ruehli, and A. C. Cangellaris, “Progress in the methodologies
for the electrical modeling of interconnects and electronic packages,”
Proceedings of the IEEE, vol. 89, no. 5, pp. 740-771, 2001.
[13] A. E. Ruehli, G. Antonini, J. Esch, J. Ekman, A. Mayo and A.
Orlandi, “Nonorthogonal PEEC formulation for time- and frequency-
domain EM and circuit modeling,” IEEE Trans. Electromagnetic
Compatibility, vol. 45, no. 2, May 2003, pp. 167–176.
[14] B. Archambeault, “Using Signal Integrity Tools for EMC Analysis,”
PCB Design for Real-World EMI Control, The Springer Int. Series in
Engineering and Computer Science, vol. 696, 2002, pp. 171-186.
[15] D. Pederson, “A historical review of circuit simulation,” IEEE
Transactions on Circuits and Systems, Vol. 31, No. 1, Jan. 1984, pp.
103-111.
[16] D. Standley, and Jr. J. L. Wyatt, “Improved Signal Delay Bounds for
RC Tree Networks,” VLSI Memo, No. 86-317, MIT, Cambridge,
MAS (USA), May 1986.
[17] A. B. Kahng, and S. Muddu, “An Analytical Delay Model of RLC
interconnects,” IEEE Trans. CAD, vol. 16, no. 12, pp. 1507-1514,
Dec. 1997.
[18] T. Eudes, and B. Ravelo, “Analysis of multi-gigabits signal integrity
through clock H-tree,” Int. J. Circ. Theor. Appl., vol. 41, no. 5, pp.
535-549, May 2013.
[19] B. Ravelo, “Modelling of asymmetrical interconnect T-tree
laminated on flexible substrate,” Eur. Phys. J. Appl. Phys. (EPJAP),
vol. 72, no. 2 (20103), pp. 1-9, Nov. 2015.
[20] S. Wen, J. Zhang, and Y. Lu, “Modeling and Quantification for
Electromagnetic Radiation of Power-Bus Structure With Multilayer
Printed Circuit Board,” IEEE Trans. Components, Packaging and
Manufacturing Technology, vol. 6, No. 1, pp. 79-86, Jan. 2016.
[21] D.-H. Kim, S.-J. Joo, D.-O. Kwak, and H.-S. Kim, “Warpage
Simulation of a Multilayer Printed Circuit Board and Microelectronic
Package Using the Anisotropic Viscoelastic Shell Modeling
Technique That Considers the Initial Warpage,” IEEE Trans.
Components, Packaging and Manufacturing Technology, vol. 6, no.
11, pp. 1667-1676, Nov. 2016.
[22] H. Tang, J.-X. Chen, H. Chu, G.-Q. Zhang, Y.-J. Yang, and Z.-H.
Bao, “Integration Design of Filtering Antenna With Load-Insensitive
Multilayer Balun Filter,” IEEE Trans. Components, Packaging and
Manufacturing Technology, vol. 6, no. 7, pp. 1408-1416, July 2016.
[23] C.-L. Cho, H.-L. Kao, L.-C. Chang, Y.-H. Wu, and H.-C. Chiu,
“Inkjet-Printed Multilayer Bandpass Filter Using Liquid Crystal
Polymer System-on-Package Technology,” IEEE Trans.
VOLUME XX, 2018
2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Author Name: Preparation of Papers for IEEE Access (February 2017)
Components, Packaging and Manufacturing Technology, vol. 6, no.
4, pp. 622-629, Apr. 2016.
[24] Z. Xu, Y. Liu, B. Ravelo, and O. Maurice, “Modified Kron’s TAN
Modeling of 3D Multilayer PCB”, in Proc. 11th International
Workshop on Electromagnetic Compatibility of Integrated Circuits,
EMC Compo 2017, St. Petersburg, Russia, 4-8 July 2017, pp. 242-
247.
[25] Z. Xu, Y. Liu, B. Ravelo, and O. Maurice, “Multilayer Power
Delivery Network Modeling with Modified Kron’s Method
(MKM)”, in Proc. of 16th Int. Symposium on Electromagnetic
Compatibility (EMC) Europe 2017, Angers, France, 4-8 Sept. 2017,
pp. 1-6.
[26] T. Nikola, “Standard 4-and 6-Layer PCB stackups,” White Paper,
Data Response, Available Online, 2013,
http://www.bitweenie.com/listings/standard-4-and-6-layer-pcb-
stackups/
[27] M. E. Goldfarb, and R. A. Pucel, “Modeling via hole grounds in
microstrip,” IEEE Microwave and Guided Wave Letters, vol. 1, no.
6, pp. 135-137, 1991.
[28] Y. Shang, C. Li, and H. Xiong, “One method for via equivalent
circuit extraction based on structural segmentation,” Int. J. Computer
Science Issues, vol. 10, no. 3, pp. 18-22, May 2013.
[29] S. Liu, W. Tang, W. Zhuang, G. Wang, and Y. L. Chow,
“Capacitance and Conductance of Through Silicon Vias With
Consideration of Multilayer Media and Different Shapes,” IEEE
Trans. Components, Packaging and Manufacturing Technology, vol.
6, no. 1, pp. 256-264, Jan. 2016.
[30] S. Huang, and L. Tsang, “Fast Broadband Modeling of Traces
Connecting Vias in Printed Circuit Boards Using Broadband Green’s
Function Method,” IEEE Trans. Components, Packaging and
Manufacturing Technology, vol. 7, no. 8, pp. 1343-1355, Aug. 2017.
[31] E. Laermans, J. De Geest, D. De Zutter, F. Olyslager, S. Sercu, and
D. Morlion, “Modeling complex via hole structures,” IEEE Trans.
Advanced Packaging, vol. 25, no. 2, pp. 206-214, May 2002.
[32] G. Kron, Tensor Analysis of Networks, Wiley Publishers, New York,
Chapman & Hall, London, 1939.
[33] G. Kron, IEEE Trans. on Circuit Theory, vol. 15, no. 3, pp. 174, Sep.
1968.
[34] K. Kondo, and Y. Ishhizuka, “Geometrical Aspects of Kron’s Non-
Riemannian Electrodynamics,” Memoirs of the unifying study of the
basic problems in engineering science by means of geometry, vol. I,
Published by Unified Study Group, G. B. Fukyu-Kai, Tokyo, Japan,
pp. 195-239, 1955.
[35] F. H. Branin, “Transient analysis of lossless transmission lines,”
Proc. of the IEEE Letters, pp. 2012-2013, 1967.
[36] J. Dobes, and L. Slama, “A modified Branin model of lossless
transmission lines,” in Proc. 52nd IEEE Int. Midwest Symp. on CAS
2009, Cancun, Mexico, 2-5 Aug. 2009, pp. 236-239.
Z. Xu holds his master’s degree in
microelectronics in 2016 in ESIEE Paris. He is pursuing a PhD degree on
multilayer PCB modelling in ESIGELEC/IRSEEM. His research area
covers signal integrity, power integrity and EMC analysis with different
models on multilayer PCBs. His current publications are focus on the
Kron-Branin model for multilayer PCB modeling applications.
9
 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI
10.1109/ACCESS.2018.2873042, IEEE Access
Author Name: Preparation of Papers for IEEE Access (February 2017)

Y. Liu received the B. Sc. degree in technics of J. Gantet holds his master's degree in electronic
control, measurement and instrumentation from Dalian University of engineering and signal processing from INPT ENSEEIHT, Toulouse in
technology, Dalian, P. R. China in July 2004, another B. Sc. degree in 2011. He is currently working as electronic engineer on innovative
electronic engineering from University of Pierre and Marie Curie (UPMC), automotive electronics packaging applications in VALEO.
Paris, France in July 2007, the M. Sc. Degree in electronic engineering
from UPMC, France in July 2009 and the Ph. D. degree at
IRSEEM/ESIGELEC from University of Rouen, France, in October 2012.
Actually, he is working as EMC engineer in the transport and energy
sector.

N. Marier holds his Master’s Degree in

B. Ravelo (MS’07 – M’09) holds his PhD
degree in 2008 from Univ. Brest and his dissertation to lead research
(“HDR=Habilitation à Diriger des Recherches”) in 2012 from Univ.
Rouen. He is currently associate professor at the graduate engineering
school ESIGELEC/IRSEEM in Rouen/France. His research interests cover
the microwave circuit design, electromagnetic compatibility (EMC) and
interference (EMI), and signal and power integrity (SI/PI) engineering. He
is a pioneer of the negative group delay (NGD) RF/analog and digital
circuits and systems. He is (co)-author of more than 210 papers and
regularly involved in national/international research projects. He co-
supervised and directed 9 PhD students whose 6 PhD candidates defended.
He is the coordinator of the European TECS project funded by the
program INTERREG VA 4081, he is in charge of the EM characterization
of the platform developed in the CareStore project funded by FP7-SME-
2012 and he participates regularly in large R&D international projects. His
current publication h-index is 18 (Reference: Google Scholar 2018).
Dr. Ravelo has been a URSI member and regularly invited to review
papers submitted to international journals (the IEEE TMTT the IEEE
ACCESS, the IEEE TCAS, the IEEE TEMC, the IEEE TIM, the IEEE
TIE, the Journal of Electromagnetic Waves and Applications, the IET
CDS, the IET MAP, and the International Journal of Electronics, etc.) and
international books (Wiley, Intech Science, etc.). He is the Scientific Chair
of the 5th International Conference on Electromagnetic Near Field
Characterization and Imaging in 2011, member of Advanced
Electromagnetic Symposium 2013-2018 technical committee.
Electronics and industrial Data Processing in Conservatoire National des
Art et Métiers in 1998 and Executive MBA in Paris-Dauphine in 2012. He
is currently Research & Innovation Project Manager on Automotive
Electronics Packaging in Valeo.
After more than 18 years of experience in Electronic Product Design for
the automotive industry, he is in charge of several European Collaborative
projects and Advanced Studies on Microelectronic Packaging topics
(Embedded technologies, FOWLP, SiP, ...), for automotive applications
within the Group of Electronic Expertise, GEEDS of the entire VALEO
group.
O. Maurice had a CNAM engineer diploma in
electronics from Paris, France in 1995, PhD degree in 2013 from Limoges
university in physics, France and his dissertation to lead research
(“HDR=Habilitation à Diriger des Recherches”) in 2016, from Ecole
Centrale de Lyon, France for his researchers over 30 years on the tensorial
analysis of networks formalism. He is currently in ARIANE Group
Company for EMC studies. He is (co)-author of more than 120 papers,
author of 3 books, co-author of 4 and author of 1 patent, co-author of 4.
Inventor of the “modified Kron’s method for EMC and of the “extended
tensorial analysis of networks”, he’s specialized in its use for various
applications. EMC expert in various French and European institutions, he
co-supervised 10 PhD students. O.Maurice has the bronze medal of the
French national defense and co-win the third prize at the European
scientific competition of Seville in 1992.

VOLUME XX, 2018 9

2169-3536 (c) 2018 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution requires IEEE permission. See
http://www.ieee.org/publications_standards/publications/rights/index.html for more information.