Finite Elements in Analysis and Design 40 (2004) 17031727

www.elsevier.com/locate/nel
A hybrid 3D thermo-elastic innite element modeling for
area-array package solder joints
D.S. Liu

, D.Y. Chiou, C.H. Lin

Department of Mechanical Engineering, National Chung Cheng University, 160 San-Hsing,
Ming-Hsiung, Chia-Yi 621, Taiwan, ROC
Received 15 December 2002; accepted 29 December 2003
Abstract
In this paper, we present a 3D thermo-elastic innite element methodology (IEM) for modeling solder
joints in area-array packaging. We also propose a new concept which involves converting the IEM into an
innite element (IE). The special element is formulated based on the conventional nite element method
(FEM) using the similarity property of stiness in isoparametric elements and matrix condensation operations.
We developed an IEFE coupling scheme and implemented it using the commercial nite element software
ABAQUS so that a complete elastostatic analysis can be conducted.
This study investigated out-of-plane thermal deformation (warpage) and its eect on thermal stress distri-
bution in the solder joints of multi-chip module (MCM) BGA assemblies. The proposed model was able to
simulate a large number of solder joints for ecient computations. The most susceptible solder joints were
identied and replaced by 3D nite element models for thermal elasticplastic re-analysis. Two types of MCM
BGA assemblies with symmetrical and non-symmetrical area-array solder joints are studied.
? 2004 Elsevier B.V. All rights reserved.
Keywords: Innite element method; Innite element; IEFE coupling method; Solder joints; MCM BGA assembly
1. Introduction
New product designs are pushing the density and speed of electronic packages to the limit. To
support this development, packaging technologies are changing at an ever more rapid pace. As the
design trend moves more towards greater miniaturization and higher performance, IC packages are
becoming smaller in size, lower in height, and lighter in weight. In addition to this, the number of
electrical interconnection (I/Os) in IC packages must be maximized.

Corresponding author. Tel.: +886-5-2720411x33305; fax: +886-5-2720589/+886-5-2721522.

E-mail address: [email protected] (D.S. Liu).
0168-874X/$ - see front matter ? 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.nel.2003.12.002
1704 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
Area-array technology oers perhaps the most attractive solution to meet the increasingly challeng-
ing demands of the next generation of electronic packaging. Area-array packaging became promising
to the electronics industry when it became clear that earlier peripherally leaded packaging had reached
its practical limits in terms of ner pitch and higher I/O pin-counts. The ball grid array (BGA) was
the rst pin-less area-array type package that was broadly accepted as the package of choice. The in-
troduction of BGA interconnection technology in the early 1990s continued the trend towards faster,
lighter, smaller, and lower-cost products that has been seen in the last few decades. Currently, BGA
packages, in which the solder joint pitch and size are continually reduced to accommodate higher I/O
density, are widely used in many applications, such as mobile phones and handheld devices, because
of their high density and high performance interconnection technology. However, during fabrication,
surface mounting, assembly, and IC operation, the packages go through temperature changes which
induce out-of-plane thermal deformation (warpage) and thermal stresses within the package. This
warpage and stress are caused by the global coecient of thermal expansion (CTE) mismatch in the
constituent materials of the packages. Since it is well known that solder joint reliability is strongly
aected by warpage, the primary issue of concern here is the thermally induced warpage and the
stress distribution on the solder joints.
Reliability analysis of package design and optimization using standard experiments, such as the
temperature cycling test (TCT), is time consuming and costly. Developing a numerical model that
can be used for virtual qualication (or called qualication-by-simulation) can reduce the cycle of
experimenting and product marketing. Extensive numerical analysis, such as the conventional nite
element method (FEM), is commonly performed in such situations, leading to a better understanding
of the stress/strain distribution, reliability implications, and enhanced package designs. However, with
decreasing package size, increasing I/O count, and complicated package congurations and materi-
als, nite element analysis requires prohibitive investment in execution time and analyst expertise,
especially when full-scale analyses have to be performed. Many researchers have expended a great
amount of eort to develop various modeling techniques to overcome these diculties. Most nu-
merical models using 2D nite element modeling to simulate the package structures have been well
developed; however, they cannot model full solder joints and packages exactly.
Full-scale 3D modeling is therefore necessary to simulate every solder joint to identify the ones
most susceptible to failure (the joints having the largest thermal deformation/stress). This 3D mod-
eling technique is often expensive and is usually replaced by alternative models for a trade-o in
computation time. These models include the following. A quarter or an octant of the 3D nite ele-
ment model, applicable only for symmetrical congurations and load conditions, was proposed [1].
This method distinguishes the quality of all the solder joints poorly. Gustafsson et al. [2] investigated
some nite element models, such as the slice model and a global model with or without a sub-model,
of BGA packages to predict their life span. Tee et al. [3] and Zhang et al. [4] discussed 3D sliced
nite element models of BGA packages to reduce computation time. This model cannot simulate all
of the solder joints and makes it dicult to impose appropriate symmetry boundary conditions that
coincide with the actual circumstances. Darbha and Dasgupta [5,6] proposed a nested FEM for stress
analysis that formulated the potential energy in localized sub-elements (nested domain) within the
main element domain. This development was restricted to 2D formulations and emphasized a local-
ized, not a full-scale, discrete characterization. Zhu et al. [7,8] introduced a global/local model for
evaluating the reliability of BGA assemblies. This method modeled the PWB and package as global
models with a relatively coarse mesh. The critical solder joints were modeled as local models with
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1705
ner mesh. Obtaining more reliable and accurate results depended on enhanced mesh renements of
both the global and local models. This method also required excessive computation cost and eort.
One model used equivalent beam elements that represented a large numbers of solder joints [9,10].
However, the overall deformation/stress/strain distribution of the replaced solder joints could not be
extracted from the response obtained from the equivalent beam element.
This research focused on developing an alternative methodology for local solution renement
at sites of interest that is simple, quick, versatile, and ecient. With the above goals in mind, the
so-called innite element method (IEM), pioneered by mathematicians in the 1970s, is employed. The
basic idea behind the IEM is to discretize the problem domain using a large number of layer-wise
elements with a designed partition technique and nodal sequence such that each element-layer is
similar to the other layers. This partition concept was proposed rst by Silvester and Cermark [11]
using the nite dierence method for solving eld problems. Thatcher [12,13] used a large number
of triangular elements and derived a recurrence relation for Laplaces equation in a systematic way
to solve problems with re-entrant corners and unbounded domains. At the same time, Ying [14,15],
Han and Ying [16], and Ying and Pan [17] independently developed the mathematical foundations
for an IEM formulation for Laplaces equation, exterior Stokes problems, corner problems and plane
elasticity problems in fracture mechanics. Guo [18] proved that the concept of similar elements could
be extended to arbitrary isoparametric elements. He also constructed a transformation procedure that
combines the local stiness matrix layer by layer. No detailed numerical procedures or examples
were presented in his paper. The authors [19] previously implemented the IEM using computer
codes to handle various classical elasticity and singularity problems. A convergence criterion was
also proposed for the IEM formulation to make the imaginary concept of innite layers practical
and to assure that the innite element stiness matrix would be convergent.
Up to the present, most related studies have focused on the development of 2D IEM formulation
without introducing thermal eects. In this paper, we describe how information appearing in a
previous study [18] is used to work out in detail and to improve 3D IEM algorithms that consider
temperature eects. The similarity partition concept is employed to automatically generate a large
number of elements, layer by layer, distributed over the IE sub-domain. All degrees of freedom
related to the IE sub-domain are condensed and transformed to form a combined element with
master node degrees of freedom only. This is regarded as a regular nite element and called an
innite element (IE).
Another contribution made by our research is the scheme that successfully converts the innite
element method to the IE. After the new IE is formulated and dened, the IE stiness matrix was
assembled into the system matrix of the FE sub-domain. In addition, commercial nite element
software (e.g. ABAQUS) is used to integrate the IE and the conventional FEM so that complete
analysis can be conducted. The advantages of modeling complex shape domains and maintaining
the highly rened meshing eect are preserved simultaneously without increasing round-o errors.
This scheme also reduces the total number of degrees of freedom. In addition, the CPU time and
PC memory storage are signicantly reduced without sacricing accuracy.
This paper presents the 3D thermo-elastic IEM and the IEFE coupling scheme that can be used to
examine MCM BGA packages with area-array type solder joints. Both full-scale 3D symmetrical and
non-symmetrical models are evaluated to predict the board-level thermal deformation of the assembly
and its eect on the stress distribution in the solder joints. The analysis predicts the solder joints most
susceptible to failure and identies which joints are subjected to stresses that exceed the elastic limits
1706 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
of the solder materials. The susceptible joints are identied and can be replaced by 3D nite element
models for thermal elasticplastic re-analysis. The proposed methodology provides online engineers
another useful and ecient numerical tool for examining a large number of warpage-aected solder
joints in electronic packages.
2. 3D IEM formulation
2.1. Similarity of 3D isoparametric elements
As mentioned earlier, the key concept in the IEM is the similar partition concept. This concept
can be illustrated using two similar eight-node hexahedral elements, shown in Fig. 1. For element
I , the local nodes i are numbered 1, 2, . . . , m, using a consistent node numbering scheme to dene
connectivity. The coordinate (x
I
i
, ,
I
i
, :
I
i
) denotes the global coordinate value of node i. m is the number
of nodes on the element (in this section, m=8). Taking the global origin O and z as the center of
the similarity and the proportionality constant, one creates element II, whose nodal coordinate value
and length dimensions are similar to those of element I. The relationship between the corresponding
nodes can be expressed as
(x
II
i
, ,
II
i
, :
II
i
) = (zx
I
i
, z,
I
i
, z:
I
i
), (1)
where z (0, 1) or z (1, ), which implies that there is a scale ratio. Based on the isoparametric
representation, the shape functions [
i
(, p, ) are used to express the coordinates of a point within
element I in terms of the nodal coordinates. Thus, we have
x
I
=
m

i=1
[
i
(, p, )x
I
i
, (2)
I
II
x
y
O
( , , )
I I I
i i i
x y z
( , , )
II II II
i i i
x y z
1
2
3
4
5
6
7
8
z
Fig. 1. Schematic diagram of similar 3D isoparametric elements.
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1707
,
I
=
m

i=1
[
i
(, p, ),
I
i
, (3)
:
I
=
m

i=1
[
i
(, p, ):
I
i
. (4)
Multiplying both sides by z, and in view of Eq. (1), we have
x
II
= zx
I
=
m

i=1
[
i
(, p, )x
II
i
, (5)
,
II
= z,
I
=
m

i=1
[
i
(, p, ),
II
i
, (6)
:
II
= z:
I
=
m

i=1
[
i
(, p, ):
II
i
. (7)
From Eqs. (5)(7), both hexahedral elements II and I can be mapped using the same cubic-shaped
master element, which is dened in natural coordinates. In other words, if one isoparametric element
has nodal coordinate values similar to the values of other elements, they are designated as similar
elements.
When the physical coordinates are mapped to the natural coordinates, the associated element
stiness matrix [ k] and the nodal temperature load [f
th
] (the superscript th represents thermal
eect) for a 3D isoparametric element are then calculated using
[k]
3m3m
=

1
1

1
1

1
1
[B]
T
3m6
[D]
66
[B]
63m
det[J] d dp d, (8)
[f
th
]
3m1
=

1
1

1
1

1
1
[B]
T
3m6
[D]
66
[]
61
1 det[J] d dp d, (9)
where
[B] =

9[
i
9x
0 0
0
9[
i
9,
0
0 0
9[
i
9:

9[
i
9,
9[
i
9x
0
0
9[
i
9:
9[
i
9,

9[
i
9:
0
9[
i
9x

, (10)
1708 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
[D] =
E
(1 + v)(1 2v)

(1 v) v v 0 0 0
v (1 v) v 0 0 0
v v (1 v) 0 0 0
0 0 0
(1 2v)
2
0 0
0 0 0 0
(1 2v)
2
0
0 0 0 0 0
(1 2v)
2

, (11)
[J] =

9x
9
9,
9
9:
9
9x
9p
9,
9p
9:
9p
9x
9
9,
9
9:
9

9[
i
9


9[
i
9p


9[
i
9

x
i
,
i
:
i

, (12)
where E is the Youngs modulus, v is the Poissons ratio, [] is the coecient matrix of thermal
expansion and 1 is the temperature dierence.
Recalling Eq. (1), and from Eqs. (10) and (12), we have
[B]
II
=
1
z
[B]
I
, (13)
det[J]
II
= z
3
det[J]
I
. (14)
When Eqs. (10)(12) are substituted into Eqs. (8) and (9), it is obvious that
[k]
II
= z[k]
I
, (15)
[f
th
]
II
= z
2
[f
th
]
I
. (16)
Eqs. (15) and (16) indicate that the stiness and thermal load terms for similar 3D isoparametric
elements are directly in dimensional dependence on the ratios z and z
2
, respectively.
2.2. 3D thermo-elastic innite element method formulation
The detailed IEM formulation with thermal eects for illustrating 3D thermo-elastic problems
is presented in this section. The pyramid-like body shown in Fig. 2 is used as an example to
illustrate the above-mentioned partition concept. The meshing steps are described as follows: rst,
the primary domain boundary surface, I
0
, is properly distributed over a total number of n master
nodes and discretized into virtual four-node quadrilateral elements. Second, when the global origin O
is chosen as the similar partition center, and a constant z is used, surfaces I
1
, I
2
, . . . , I
S
, . . . similar
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1709
0

1 s

O
layer 1
layer 2
layer s
Fig. 2. Schematic diagram of 3D innite element mesh.
to I
0
are constructed with the center O and proportionality constants z
1
, z
2
, . . . , z
S
, . . . , respectively.
The region bounded between I
i1
and I
i
is called the ith element-layer (i = 1, 2, . . . , s), where s is
number of chosen element-layers. Third, straight lines are drawn from the origin to the master nodes
and each individual I
i
is regularly discretized in a manner similar to I
0
. The nodal number and
coordinates of nodes on each individual I
i
can be determined from the master node coordinates that
have geometrically similar conditions. Fourth, every element-layer is auto-meshed into eight-node
hexahedral elements that are similar to one another in a radial direction.
When the outermost element-layer (layer 1) is considered, the element stiness matrix of each
hexahedral element can be calculated and assembled into an element-layer stiness matrix using
the conventional nite element formulation. The stiness matrix of the rst element-layer can be
expressed as

K
a
A
T
A K
b

6n6n
, (17)
where K
a
, K
b
, and A are the sub-matrices of the stiness matrix and have the identical dimension
3n 3n. A
T
is the transpose of A. Since the global layer stiness is symmetrical and banded, both
K
a
and K
b
are symmetrical and banded. The nodal displacement vector
i
of nodes on I
i
is dened
as

i
[ u
i
1
t
i
1
w
i
1
u
i
n
t
i
n
w
i
n
]
T
. (18)
The nodal force vector [
i
of nodes on I
i
is dened as
f
i
[ [
i
1x
[
i
1,
[
i
1:
[
i
nx
[
i
n,
[
i
n:
]
T
. (19)
1710 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
The nodal temperature load vector f
th
i
of nodes on I
i
is dened as
f
th
i
[ [
i(th)
1x
[
i(th)
1,
[
i(th)
1:
[
i(th)
nx
[
i(th)
n,
[
i(th)
n:
]
T
. (20)
The ith element-layer stiness matrix represents the nodal force and displacement vector relation-
ships of I
i1
and I
i
. Recalling Eq. (17) and treating the rst element-layer as an example, we
have

K
a
A
T
A K
b

f
0
+f
th
0
f
1
+f
th
1

. (21)
Two algebraic equations are extracted from Eq. (21) as follows:
K
a

0
A
T

1
=f
0
+f
th
0
, (22)
A
0
+K
b

1
=f
1
+f
th
1
, (23)
where
0
, f
0
, and f
th
0
denote the nodal displacements, traction, and thermal load on I
0
, respectively.
According to the foregoing similarity principle (Eqs. (15) and (16)), it is obvious that the stiness
and thermal load matrices of all of the element-layers are in dimensional dependence on the ratios
z and z
2
, respectively. Hence, we can express the stiness matrix of the s element-layers (from the
1st element-layer to the sth element-layer) as a set of algebraic equations, namely,

K
a
A
T
A K
b

f
0
+f
th
0
f
1
+f
th
1

for layer 1, (24-1)

z

K
a
A
T
A K
b

f
1
+ z
2
f
th
0
f
2
+ z
2
f
th
1

for layer 2, (24-2)

z
2

K
a
A
T
A K
b

f
2
+ z
4
f
th
0
f
3
+ z
4
f
th
1

for layer 3, (24-3)

.
.
.
.
.
.
z
s1

K
a
A
T
A K
b

s1

f
s1
+ z
2(s1)
f
th
0
f
s
+ z
2(s1)
f
th
1

for layer s. (24-4)

Extracting every algebraic equation, adding the second equation for the ith element-layer and the
rst equation for the (i + 1)th element-layer to generate a connection among layers, and letting
P =K
b
+ zK
a
, we have
K
a

0
A
T

1
=f
0
+f
th
0
, (25-1)
A
0
+P
1
z A
T

2
=f
th
1
+ z
2
f
th
0
, (25-2)
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1711
z A
1
+ z P
2
z
2
A
T

3
= z
2
( f
th
1
+ z
2
f
th
0
), (25-3)
.
.
.
z
i1
A
i1
+ z
i1
P
i
z
i
A
T

i+1
= z
2(i1)
( f
th
1
+ z
2
f
th
0
), (25-4)
.
.
.
z
s2
A
s2
+ z
s2
P
s1
z
s1
A
T

s
= z
2(s2)
( f
th
1
+ z
2
f
th
0
), (25-5)
z
s1
A
s1
+ z
s1
K
b

s
=f
s
+ z
2(s1)
f
th
1
F
s
. (25-6)
Let M
s
=K
b
and F
s
=f
s
+ z
2(s1)
f
th
1
. Substituting them into Eq. (25-6), we have

s
=M
1
s

A
s1
+
F
s
z
s1

. (26)
By substituting Eq. (26) into Eq. (25-5), we get
z
s2
A
s2
+ (z
s2
P z
s1
A
T
K
1
b
A)
s1
=A
T
K
1
b
( f
s
+ z
2(s1)
f
th
1
) + z
2(s2)
( f
th
1
+ z
2
f
th
0
). (27)
When Eq. (25-6) is compared with Eq. (27), two iteration formulas can be inferred
M
i
=P z A
T
M
1
i+1
A, (28)
F
i
=A
T
M
1
i+1
F
i+1
+ z
2(i1)
( f
th
1
+ z
2
f
th
0
), (29)
where i = 1, 2, 3, . . . , s 1.
By substituting the above two iteration formulas into Eq. (27), we get
z
s2
A
s2
+ z
s2
M
s1

s1
=F
s1
. (30)
Rearranging Eq. (30) and another iteration formula can be inferred as

)
= z
( )1)
M
1
)
(z
)1
A
)1
+F
)
), (31)
where ) = 1, 2, . . . , s.
Since M
s
and F
s
are known, then M
s1
, M
s2
, . . . , M
1
; F
s1
, F
s2
, . . . , F
1
can be iterated out using
Eqs. (28) and (29), respectively. From Eq. (31), we have
1
=M
1
1
(A
0
+F
1
). Substituting
1
into Eq. (25-1), we obtain the most important formula for 3D IEM, that is,
(K
a
A
T
M
1
1
A)
0
= (A
T
M
1
1
F
1
+ ( f
0
+f
th
0
)). (32)
Eq. (32) can be expressed in the concise form
K
Z

0
=F
Z
. (33)
1712 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
0

1 s

O
layer 1
layer 2
layer s
0

Infinite
Element
Fig. 3. Schematic diagram of 3D innite element formation.
where K
Z
and F
Z
denote the equivalent combined element stiness matrix and associated combined
loading vector, respectively. F
Z
contains both the eects from the outer surface traction and the
thermal loading. Once F
Z
is determined,
0
can be obtained from Eq. (32). Then
1
,
2
, . . . , and
s
can be obtained sequentially from Eq. (31).
When the numerical procedures from Eq. (25-1) to Eq. (25-6) are used, and when layers are
added layer by layer, all of the inner layer elements are condensed and transformed into only one
combined element with master nodes along the domain boundary I
0
only. This is called the IE,
as schematically shown in Fig. 3. It can be viewed as a process of element elimination. The inner
elements and nodes are imaginary and virtual, but they in fact exist and work in IEM manipulations.
Generally, the IE renement strategy can be achieved by increasing the number of master nodes
and element-layers. From a physical point of view, K
Z
can be treated as the equivalent stiness
matrix of this IE with a dimension of 3n 3n. Although the total degree of freedom of K
Z
is
largely reduced, the mesh renement eect of large amounts of innitesimal elements around the
reference point O is maintained and does not increase any corresponding round-o errors because of
its specic matrix condensation process. Moreover, to compute K
Z
, only the calculation of the rst
element-layer stiness matrix with the chosen proportionality constant z and element-layer number
s is required.
2.3. IEFE coupling method
FEM can model problems with complicated shape domains (e.g. concave shapes) and boundary
conditions but cannot manage local rened areas with large numbers of d.o.f.s well. On the other
hand, IEM is favorable for modeling local rened areas with fewer d.o.f.s, but it is applicable only
to relatively simpler shaped domains. A better modeling strategy involves combining both methods
to deal with all kinds of problems.
Therefore, an IEFE coupling scheme is proposed using the IEs to sub-divide critical parts in the
entire domain. Outside the IE sub-domains, the solutions are obtained using the conventional FEM
in the FE sub-domains. The master nodes on the outer boundary surface of each IE are converted
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1713

D
FEM
domain
IEM
domain
D
IEM
domain

FEM
domain
+
conventional
finite
elements
coupling
interface
common
nodes
Infinite
Element
0

Fig. 4. Schematic diagram of 3D IEFE coupling method.

into interface common nodes that are between the IE and FE sub-domains. Because each IE stiness
matrix is pre-determined, the elements can be regarded as regular nite elements, and their stiness
matrices are assembled into the global stiness matrix formed using the FEM.
To illustrate the coupling algorithm, shown in Fig. 4, a cracked body is partitioned into two
sub-domains, O and D, separated from the coupling interface I
0
, and modeled using the FEM
and IEM, respectively. The algebraic element equations for the FE sub-domain are given by the
well-known FE implementation
K =F, (34)
where K is the global stiness matrix of the FE sub-domain, and F and are the nodal forces and
nodal displacements, respectively. Furthermore, Eq. (34) can be expressed as

K
couple
K
T
c[
K
c[
K
fem

fem

F
0fem
F
fem

, (35)
where
0
and
fem
denote the vectors of the IEFE interface and non-interface nodal displacements,
respectively. F
0fem
and F
fem
are the associated loading vectors, respectively. From Eq. (35), two
equations can be extracted
K
couple

0
+K
T
c[

fem
=F
0fem
, (36)
K
c[

0
+K
fem

fem
=F
fem
. (37)
Conversely, the IE sub-domain produces the following algebraic equation using the IEM described
in detail in Section 2.2
K
IEM

0
=F
0iem
, (38)
where K
IEM
=K
Z
=(K
a
A
T
M
1
1
A) denotes the combined stiness matrix K
Z
of the IE sub-domain.

0
and F
0iem
denote the interface nodal displacements and the associated load vectors, respectively.
1714 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
Along the IEFE interface, the displacement compatibility and force equilibrium must be satised.
Thus, we combine Eqs. (36) and (38), resulting in the following:
(K
couple
+K
IEM
)
0
+K
T
c[

fem
=F
0iem
+F
0fem
= 0. (39)
Eqs. (37) and (39) are combined to give the nal IEFE coupled element equation in the form:

K
couple
+K
IEM
K
T
c[
K
c[
K
fem

fem

0
F
fem

. (40)
When boundary conditions are applied to the global matrix and vector, the displacements
fem
in
the FE sub-domain can be obtained using Eq. (40). Once we obtain the interface nodal displacements

0
for the IE sub-domain, the traction along IEFE interface F
0iem
can be found using Eq. (38).
Furthermore, a sequence of the nodal displacement
)
within the IE sub-domain can be obtained
using Eq. (31). The static stressstrain analysis can be performed using the stressstrain relationship
and element straindisplacement relationship.
2.4. Integration of IEFE coupling method with commercial FEM software
The implementation of the IEFE coupling method used in this study was carried out using
the commercial nite element software ABAQUS/Standard. All related IEM numerical procedures
were programmed and executed in the MATLAB v.5.3 language [20]. After the IEM mathematical
manipulations, the special elements, called innite elements (IEs), were generated (see Section 2.2).
The IEs were treated as regular nite elements, and their element properties were obtained from the
pre-determined IE stiness matrices K
Z
. The IE denition is not included in the ABAQUS element
library; therefore, the IE is designated as a user-dened element that represents a geometric part
of the model. In the ABAQUS codes, the user-dened elements are introduced using the *USER
ELEMENT option. In the simplest case, a linear user-dened element can be dened as a stiness
matrix. This matrix (K
Z
) can be distilled from the Matlab results and dened by using the *MATRIX
option. The properties of linear user-dened elements are dened using the *UEL PROPERTY option
and, optionally, the *MATRIX option. For further details on this implementation, please refer to the
ABAQUS users manual [21].
3. Verication of the IEFE coupling methodology
A MCM BGA assembly, developed by ChipMos INC. for the 128 Mb SDRAM, was investigated
in this study. The assembly employed dual 64 Mb SDRAM chips on a single-sided wiring organic
substrate, using the proven wire boding and transfer molding technologies. The external connection
terminals are an area-array structure using solder joints with a pitch of 0.8 mm and a joint diameter
of 0.35 mm. Fig. 5 shows the schematic diagram of this package. Cross-sectional views along the
length and width of the package are shown in Fig. 6.
A 3D IEFE model of the MCM BGA assembly was established using the commercial software
ABAQUS/Standard for thermo-mechanical analysis. A full model of this package was created in this
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1715
Silicon Die Molding Compound
BT Substrate
Adhesive
FR-4 PWB
Eutectic Solder Joint
Fig. 5. Schematic diagram of a novel MCM BGA assembly.
(b) Lengthwise view
(a) Widthwise view
Fig. 6. Two cross-sectional views of a MCM BGA assembly.
Table 1
Brief geometric dimensions of the MCM BGA package
Die size 9.2 mm 5.1 mm 0.25 mm
Die pitch 1.1 mm
Adhesive size 9.2 mm 1.35 mm 0.2 mm
BT substrate size 21 mm 3.13 mm 0.5 mm
FR-4 PWB size 21 mm 8.6 mm 1.6 mm
Solder joint diameter 0.35 mm
Solder joint stand-o 0.3 mm
Solder joint pitch 0.8 mm
Solder joint number 15 (row) 4 (column)
study. The IEFE model, with lengthwise, widthwise and vertical cross-sectional views, is shown in
Fig. 7. Table 1 lists the geometric dimensions of the package. The material properties used for the
package are given in Table 2. These properties were obtained from the providers (ChipMos Inc.)
database system and the literature [22]. The properties of all component materials were assumed to
be linear elastic, isotropic, and independent of temperature variation. The FE modeling components
1716 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
Fig. 7. Three cross-sectional views of a 3D IEFE model.
contained the silicon dies, BT substrate, epoxy molding compound, adhesive, and FR-4 PWB. The FE
model was composed of 118 416 elements and 142 505 nodes. The IE modeling component contained
60 eutectic solder joints. Based on the similarity property of 3D isoparametric elements, all of the
60 IEs possessed identical stiness matrices. Only one solder joint with its IE stiness matrix was
required to be calculated once, and the matrix was also valid for the other solder joints. The IE
model for each solder joint was composed of 162 master nodes and 1 innite element (IE), which
is equivalent to 9600 rened hexahedral elements and 9882 nodes. The full IEFE model, therefore,
consisted of 152 387 nodes, 118 416 hexahedral elements, and 60 innite elements, equivalent to a
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1717
Table 2
Material properties of the MCM BGA package
Component Youngs modulus Poissons ratio CTE Yielding stress
E (GPa) v :(}

C) (MPa)
BT substrate 2.3 0.3 1.7 (E-5)
Die 163.1 0.3 3 (E-6)
Molding compound 14.0 0.25 1.6 (E-5)
Adhesive 2.65 0.3 4.43 (E-5)
Solder joint 20.6 0.4 2.96 (E-5) 25 [24]
FR-4 PWB 18.2 0.19 3 (E-5)
Table 3
IEFE modeling information of the MCM BGA package
Component Node Element Equivalent node Equivalent element
BT substrate 60 612 48 720 60 612 48 720
Silicon die 16 212 10 416 16 212 10 416
Molding compound 33 081 23 796 33 081 23 796
Adhesive 9408 4380 9408 4380
FR-4 PWB 42 292 31 104 42 292 31 104
Solder joint (IEM) 9882 60 592 920 576 000
Total 152 387 118 476 735 425 694 416
total of 694 416 rened hexahedral elements and 735 425 nodes. Table 3 lists the number of nodes
and elements and the corresponding equivalent rened nodes and elements for individual package
components. When IE modeling is used instead of regular FE modeling of the solder joints, large
numbers of degrees of freedom were signicantly reduced. This also greatly reduced the required
computational cost and eort.
3.1. Modeling validation
To precisely validate the IEFE model of the package, a proven displacement measurement
method, microscopic Moir e interferometry, was employed in this research [23]. Moir e interferometry
is an optical technique that measures thermally induced in-plane displacement elds with high sensi-
tivity. Fig. 8 shows the Moir e interferometry instrument (Portable Engineering Moir e Interferometer,
PEMI), which was designed by IBM and was set up in our laboratory.
After the package was sectioned and polished to the cross-section of interest, a thin cross-lined
diraction grating was applied to the widthwise cross-section at 85

C using Tra-Bond F253 epoxy.

Based on a grating frequency of 1200 lines}mm, the spacing between interference fringes represents
a 0.417 m displacement potential. Thus, the displacement elds were presented in discrete 0.417 m
steps. The deformations were induced when the package was cooled down to 25

C (room temper-
ature) from 85

C. The detailed principles and experimental procedures using Moir e can be found
1718 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
Fig. 8. A picture of the Moir e interferometry instrument.
in the literature [24]. The PEMI, together with the phase-shifting technique, was applied to capture
the digital images of the fringe pattern and the corresponding phase diagram which represents the
continuous displacement eld. The displacement elds U (x-direction) and J (,-direction) were
then extracted based on the following relationships:
U =
1
[
N
x
, J =
1
[
N
,
, (41)
where N
x
and N
,
are the fringe numbers in the U and J eld patterns, respectively, and [ is the
frequency of the virtual reference grating.
The horizontal (U) and vertical (J) displacement elds obtained using experimental measure-
ments and computer simulations are shown in Figs. 9 and 10. Because user-dened element plotting
is not supported in ABAQUS/Post, the IEs representing solder joints cannot be displayed. Moir e
fringes were found to distort more seriously at the solder joint/BT interface than at the solder
joint/PWB interface, chiey due to the larger CTE mismatch between the solder joint and BT mate-
rials. Corresponding to Figs. 9 and 10, the numerical analysis results matched the experimental data
in magnitude as well as in overall deformation trend. The maximum dierence in the number of
fringes was not more than about ve fringe orders (2.085 m deformation). The deformation trends
for the calculated thermo-mechanical behaviors of the package based on the Moir e interferometry
experiment and IEFE analyses were reasonably identical. The proposed IEFE model was therefore
deemed to be acceptable for conducting the following analyses.
4. Case study
With a broad variety of wiring layout technology and practical engineering applications, mis-
cellaneous electronic packaging structural designs can be roughly classied into symmetrical and
non-symmetrical structures. Non-symmetrical structures are the most commonly encountered struc-
ture. In this section, a full 3D IEFE model is used to study a MCM BGA assembly with symmetrical
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1719
Fig. 9. A plot of the measured fringe pattern versus the simulated displacement contour as temperature changes from
85

C to 25

C (U eld).
Fig. 10. A plot of the measured fringe pattern versus the simulated displacement contour as temperature changes from
85

C to 25

C (J eld).
and non-symmetrical area-array type solder joints that are subjected to a temperature gradient.
This study focused on the use of IE to simulate a large number of solder joints. The board-level
thermal deformation and thermal stress distribution within the relatively susceptible solder joints are
discussed.
1720 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
Fig. 11. On-board warpage of the package (unit: mil.).
4.1. A MCM BGA assembly with symmetrical type solder joints (60-I/Os)
Consider a MCM BGA assembly with symmetrical type solder joints subjected to a temperature
gradient from 25

C to 175

C (1 = +150

C). Proper boundary conditions for the symmetrical

sections were applied in this simulation. Fig. 11 shows the thermally induced out-of-plane displace-
ment (on-board warpage) distribution of the assembly. It can be observed that the point with the
largest warpage value is the point farthest from the center, located within the four diagonal corner
areas. A graph comparing the diagonal warpage variation from the proposed simulation and from the
shadow Moir e measurement [25] is shown in Fig. 12. The gure shows that the experimental results
exhibited non-symmetrical warpage variation trends, which were perhaps due to experimental error or
constituent material dierences. In general, both the numerical and experimental results were shown
to be in close agreement. The maximum warpage at the diagonal corner was 4.63 and 4.57 (mil)
from the simulation and the experiment, respectively, and the relative dierence was estimated to be
about 1.31%.
Since the entire assembly fails once the rst joint breaks, it is necessary to identify, from the
stress/strain point of view, the joints susceptible to breaking. Fig. 13 schematically shows the ar-
rangement of the symmetrical type solder joints. After the internal stress distribution within all of
the solder joints was calculated and examined, it was observed that the corner joints, numbered 1, 2,
3, and 4, were subjected to higher von Mises stresses than were any other solder joint. This nding
is consistent with the observation that the corner joints are expected to have the largest deformation.
Due to the symmetrical conguration, it is apparent that the four corner joints have identical re-
sponses to the maximum von Mises stress of about 30 MPa. The von Mises stress distribution within
the four most susceptible solder joints is depicted in Fig. 14. The gure shows that considerable
stress concentration is peripherally distributed along the top and bottom of the solder joints, the top
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1721
Location on diagonal (mm)
W
a
r
p
a
g
e
(
m
i
l
)
0 5 10 15 20
0
1
2
3
4
5
Experiment
Proposed method
Fig. 12. Diagonal warpage variations of the package (unit: mil.).
Fig. 13. Arrangement of symmetrical area-array type solder joints.
and bottom being the interfaces between the solder joint/BT and the solder joint/PWB, respectively.
At the solder joint/PWB interface, the von Mises stress exceeds the yielding stress of the solder
material (25 MPa). Damage and failure will most likely be initiated at the solder joint/PWB inter-
face rst. Once the susceptible solder joints are identied, the joints which would undergo plastic
deformation can be replaced by regular 3D nite element models for complete thermal elasticplastic
re-analysis, as schematically shown in Fig. 15.
1722 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
-5.8
-5.7
-5.6
-5.5
-5.4
X
-5.3
-5.2
-5.1
-5
-4.9
Z
-0.1
X
Y
Z
Mises
2.50E+01
2.15E+01
1.80E+01
1.45E+01
1.11E+01
7.57E+00
4.09E+00
6.00E-01
5.4
5.5
5.6
5.7
5.8
X
-5.3
-5. 2
-5.1
-5
-4.9
Z
-0.1
0
0.1
Y
Z
X
Y
Mises
2.50E+01
2.15E+01
1.80E+01
1.45E+01
1.11E+01
7.57E+00
4.09E+00
6.00E-01
-5.8
-5.7
-5.6
-5.5
-5.4
X
0.6
0.7
0.8
0.9
1
Z
- 0.1
0
0.1
X
Y
Z
Mises
2.50E+01
2.15E+01
1.80E+01
1.45E+01
1.11E+01
7.57E+00
4.09E+00
6.00E-01
5.4
5.5
5.6
5.7
5.8
X
0.6
0.7
0.8
0.9
1
Z
-0.1
X
Y
Z
Mises
2.50E+01
2.15E+01
1.80E+01
1.45E+01
1.11E+01
7.57E+00
4.09E+00
6.00E-01
1
1 2
3
4
Fig. 14. Von-Mises stress distribution of the four susceptible solder joints.
4.2. A MCM BGA assembly with non-symmetrical type solder joints (53-I/Os)
In this case, consider a MCM BGA assembly with non-symmetrical type solder joints. As explained
in the introduction, full-scale 3D modeling is necessary for non-symmetrical structures. However, this
usually requires a large amount of computation. Therefore, this analysis serves two main purposes:
(1) to investigate the performance of the proposed IEFE coupling method in modeling a large
number of non-symmetrical type solder joints, and (2) to show this methods ability to identify the
joints most susceptible to failure.
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1723
Fig. 15. Replacement of the susceptible solder joints with 3D nite elements in the symmetrical case ( represents IEs
for solder joints).
Fig. 16. Arrangement of non-symmetrical area-array type solder joints.
Fig. 16 schematically shows the arrangement of non-symmetrical type solder joints. As shown,
some of the solder joints were randomly removed. Again, compared with the symmetrical case,
the corner joints were numbered 1, 2, 3, 4, and 5, and they were subjected to excessive von Mises
stresses that exceed the yielding stress of the solder material. The von Mises stress distribution within
these ve solder joints is depicted in Fig. 17. The gure clearly shows that there was a distinct stress
distribution in each solder joint. More serious stress concentration occurred at the solder joint/PWB
interface. A comparison of the maximum von Mises stress values within the susceptible solder joints
between the symmetrical and non-symmetrical cases was tabulated and is shown in Table 4. Number
1724 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
5.4
5.5
5.6
5.7
5.8
X
0.6
0.7
0.8
0.9
1
Z
-0.1
X
Y
Z
1
-5.8
-5.7
-5.6
-5.5
-5.4
X
-5.3
-5.2
-5.1
-5
-4.9
Z
-0.1
X
Y
Z
5.4
5.5
5.6
5.7
5.8
X
-5.3
-5.2
-5.1
-5
-4.9
Z
-0.1
0
0.1
Y
Y
Z
X
Mises
2.50E+01
2.14E+01
1.79E+01
1.43E+01
1.07E+01
7.14E+00
3.57E+00
0.00E+00
-5.8
-5. 7
-5.6
-5.5
-5.4
X
-0. 2
-0.1
0
0.1
0.2
Z
-0.1
0
0.1
Y
X
Y
Z
-5
-4. 9
-4. 8
-4.7
-4. 6
X
0.6
0.7
0.8
0.9
1
Z
-0.1
0
0.1
Y
X
Y
Z
Mises
2.50E+01
2.14E+01
1.79E+01
1.43E+01
1.07E+01
7.14E+00
3.57E+00
0.00E+00
Mises
2.50E+01
2.14E+01
1.79E+01
1.43E+01
1.07E+01
7.14E+00
3.57E+00
0.00E+00
Mises
2.50E+01
2.14E+01
1.79E+01
1.43E+01
1.07E+01
7.14E+00
3.57E+00
0.00E+00
Mises
2.50E+01
2.14E+01
1.79E+01
1.43E+01
1.07E+01
7.14E+00
3.57E+00
0.00E+00
1 2
3 4
5
Fig. 17. Von-Mises stress distribution of the ve susceptible solder joints.
1 solder joint with a maximum stress of 31.81 MPa was identied as the one most likely to fail rst.
Once the susceptible solder joints were identied, the joints which would undergo plastic deformation
can be replaced by regular 3D nite element models for complete thermal elasticplastic analysis,
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1725
Table 4
Comparison of maximum Von-Mises stress within susceptible solder joints
Maximum Von-Mises stress (MPa)
Symmetrical case Non-symmetrical case
1 29.7979 1 31.8094
2 30.1669 2 30.0029
3 29.9708
3
4
26.9240
26.9273
4 30.2087 5 29.2701
Fig. 18. Replacement of the susceptible solder joints with 3D nite elements in the non-symmetrical case ( represents
IEs for solder joints).
as schematically shown in Fig. 18. Based on these two case studies, the results showed that the
solder joints located at the peripheral corner areas may be more susceptible to failure than any other
solder joint. This nding that the reliability of solder joints is strongly inuenced by warpage is also
consistent with the experimental and numerical observations found in most of the literature.
5. Conclusion
Complete 3D thermo-elastic IEM algorithms are well developed based on the similarity partition
concept and particular matrix operations. The major contribution made in this work is the con-
cept of converting the IEM into the innite element (IE). Another contribution is the establishment
of an innovative IEFE coupling scheme that associates IE and the conventional FE through the
commercial FEM software ABAQUS for 3D thermalelastic analyses. With the proposed method,
complex conguration domains and highly rened meshing eects can be modeled and preserved
1726 D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727
simultaneously without increasing round-o errors. This scheme reduces the total number of com-
putational degrees of freedom. Moreover, the CPU time and PC memory storage requirements are
signicantly reduced.
In this research the proposed numerical method was successfully demonstrated to be workable
through the investigation of the thermal deformation/stress responses of a MCM BGA assembly. A
large number of area-array type solder joints were modeled using the IEs. Other components were
modeled using nite elements. The resulting IE stiness matrix only involved master node degrees
of freedom over the elements boundary. Since identical stiness holds for similar isoparametric
elements, only one IE stiness matrix for a solder joint needs to be calculated once, and the matrix
becomes valid for the other joints.
To validate this model, the displacement elds and fringe patterns determined from IEFE
simulations and Moir e interferometry measurements as the assembly was cooled from 85

C to ambi-
ence were compared. The comparison results showed reasonable agreement in both approaches. Two
cases involving assemblies carrying symmetrical and non-symmetrical area-array type solder joints
subjected to +150

C temperature gradients were considered. In the symmetrical case, the point with
the largest warpage value was the point farthest from the center, located in the four diagonal corner
areas. The maximum von Mises stress value within the four corner joints was found to be higher
than that within other solder joints and exceeded the yielding stress of the solder material. This nd-
ing was also consistent with the warpage observations. In the non-symmetrical case, similar results
showed that the corner joints were also subjected to higher von Mises stress than any other solder
joint, even with some of the joints randomly removed.
The proposed method was able to identify the solder joints most susceptible to failure in the
assembly. The identied susceptible solder joints can be replaced by regular 3D nite element
models for a complete thermal elasticplastic re-analysis. This innovative idea not only enhances
the modeling capabilities but also provides online engineers another useful and ecient numerical
tool for examining a large number of warpage-aected solder joints in electronic packages.
Acknowledgements
The authors would like to express their thanks to the National Science Council, Taiwan, R.O.C.,
for nancially supporting this research under Grant NSC90-2212-E-194-032. They would also like
to thank Dr. Shen at ChipMos, Ltd. Co. for his assistance in providing MCM BGA assemblies and
shadow Moir e measurements.
References
[1] J. Sommer, R. Dudek, E. Kaulfersch, A. Schubert, B. Michel, Thermalmechanical FE analysis and micro
deformation measurementsbasis for reliability assessment of microelectronic components, in: ITHERM 98, The
Sixth Intersociety Conference on Thermal Phenomena, Thermal and Thermomechanical Phenomena in Electronic
Systems, Seattle, WA, 1998, pp. 159165.
[2] G. Gustafsson, I. Guven, V. Kradinov, E. Madenci, Finite element modeling of BGA packages for life prediction,
in: Proceedings of the 50th Electronic Components and Technology Conference, 2000, Las Vegas, NV, 2000,
pp. 10591063.
D.S. Liu et al. / Finite Elements in Analysis and Design 40 (2004) 17031727 1727
[3] T.Y. Tee, K. Sivakumar, D.B.V. Antonio, Board level solder joint reliability modeling of LFBGA package,
in: Electronic Materials and Packaging (EMAP 2000), International Symposium on Electronic Materials & Packaging,
Hong Kong, China, 2000, pp. 5154.
[4] L. Zhang, S.S. Chee, A. Maheshwari, A. Funcell, Experimental and nite element analysis of cavity down BGA
package solder joint reliability, in: Proceedings of Third Electronics Packaging Technology Conference (EPTC 2000),
Singapore, 2000, pp. 391397.
[5] K. Darbha, A. Dasgupta, A nested nite element methodology (NFEM) for stress analysis of electronic products
part I: theory and formulation, ASME J. Electron. Packag. 123 (2001) 141146.
[6] K. Darbha, A. Dasgupta, A nested nite element methodology (NFEM) for stress analysis of electronic products
part II: durability analysis of ip chip and chip scale interconnects, ASME J. Electron. Packag. 123 (2001)
147155.
[7] J. Zhu, K. Ojala, A. Tuominen, The application of global/local technique in nite element simulation of BGA
assemblies, in: 1997 ASME International Mechanical Engineering Congress and Exposition, (97-WA/EEP-5), Dallas,
TX, 1997, pp. 1621.
[8] J. Zhu, Three dimensional eects of solder joints in micro-scale BGA assembly, ASME J. Electron. Packag. 121
(1999) 121302.
[9] H.C. Cheng, K.N. Chiang, M.H. Lee, An eect approach for three-dimensional nite element analysis of ball grid
array typed packages, ASME J. Electron. Packag. 120 (1998) 129134.
[10] Y.W. Chan, T.H. Ju, Saeed A. Hareb, Y.C. Lee, J.S. Wu, M.J. Lii, Reliability modeling for ball grid array assembly
with a large number of warpage aected solder joints, ASME J. Electron. Packag. 124 (2002) 246253.
[11] P. Silvester, I.A. Cermark, Analysis of coaxial line discontinuities by boundary relaxation, IEEE Trans. Microwave
Theory Technol. 17 (8) (1969) 489495.
[12] R.W. Thatcher, Singularities in the solution of Laplaces equation in two dimensions, J. Inst. Math. Appl. 16 (1975)
303319.
[13] R.W. Thatcher, On the nite element method for unbounded region, SIAM J. Numer. Anal. 15 (1978) 466477.
[14] L.A. Ying, The innite similar element method for calculating stress intensity factors, Sci. Sinica 21 (1) (1978)
1943.
[15] L.A. Ying, An introduction to the innite element method, Math. In Practice Theory 2 (1992) 6978.
[16] H.D. Han, L.A. Ying, An iterative method in the nite element, Math. Numer. Sinica 1 (1) (1979) 9199.
[17] L.A. Ying, H. Pan, Computation of K
I
and compliance of arch shaped specimen by the innite similar element
method, Acta Mech. Solids Sinica 1 (1981) 99106.
[18] Z.H. Guo, Similar isoparametric elements, Sci. Bull. 24 (13) (1979) 577582.
[19] D.S. Liu, D.Y. Chiou, A coupled IEM/FEM approach for solving elastic problem with multiple cracks, Int. J. Solids
Structure 24 (40) (2003) 19731993.
[20] Y.W. Kwon, H. Bang, The Finite Element Method Using MATLAB, CRC Press, New York, 2000.
[21] ABAQUS Users Manual Version 6.2-1, Hibbitt, Karlsson & Sorensen, Inc., USA, 2001.
[22] A.F. Skipor, S.V. Harren, J. Botsis, On the constitutive response of 63/37 Sn/Pb eutectic solder, ASME J. Eng.
Mater. Technol. 118 (1) (1996) 111.
[23] B. Han, Y. Guo, Thermal deformation analysis of various electronic packaging products by Moir e Microscopic Moir e
Interferometry, ASME J. Electron. Packag. 117 (1995) 185191.
[24] D. Post, B. Han, P. Ifju, High Sensitivity Moir e: Experimental Analysis for Mechanics and Materials, Springer,
New York, 1994.
[25] G.S. Shen, S.K. Chen, On Thermal Stresses and Reliability of a Multi-chip BGA Package, in: Third IEEE/CPMT
Area Array Packaging Technologies Workshop on Flip chip and Ball Grid Arrays, Berlin, Germany, 1999,
pp. 1517.