Available online at www.sciencedirect.

com

Computational Materials Science 43 (2008) 221228

www.elsevier.com/locate/commatsci

Theoretical and experimental investigations of microstructural

changes in lead-free solders
Thomas Bohme *, Wolfgang H. Muller

Institut fur Mechanik, Lehrstuhl fur Kontinuumsmechanik und Materialtheorie (LKM),

Technische Universitat Berlin, Einsteinufer 5, 10587 Berlin, Germany
Available online 14 September 2007

Abstract
Experiments show that the microstructure of solders changes over time. From a materials science point-of-view this phenomenon considerably aects the reliability and lifetime of microelectronic products. It is, therefore, important to quantitatively predict the amount of
microstructural change. In the present paper we concentrate on a theoretical and experimental description of spinodal decomposition as
well as subsequent phase growth in binary solder alloys. We present experimental results and a theoretical description, which is based on
an extended diusion equation of the phase eld type, and which can be interpreted as an generalization of the CAHNHILLIARD equation.
Moreover, it takes diusion of the FICKian type, surface tensions along the phase boundaries as well as local thermo-mechanical stresses
into account. In particular we turn our attention to the determination of the required material parameters, which can all be obtained
consistently from atomistic models or adopted from literature. As an example the FCC-structured lead-free solder alloy AgCu is considered and numerical results are presented.
 2007 Elsevier B.V. All rights reserved.
PACS: 81.40.Gh; 72.15.v; 66.30.h; 64.75.+g; 68.03.Cd
Keywords: Lead-free solders; Diusion; Spinodal decomposition; Coarsening

1. Introduction
There is an ongoing miniaturization in the area of
microelectronics driven by an increasing requirement for
mobility and more complex functionalities (e.g., automotive or portable industry). The minimal feature size within
semiconductors continuously decreases whereas the number of transistors rapidly grows (cf., Fig. 1). As a consequence the demands on strength and lifetime of the used
materials considerably rise while the material size is continuously reduced.
In addition to these technological trends environmental
initiatives become increasingly important. The purpose of
*

Corresponding author. Tel.: +49 30 314 26440; fax: +49 30 314 24499.
E-mail addresses: thomas.boehme@tu-berlin.de (T. Bohme),
wolfgang.h.mueller@tu-berlin.de (W.H. Muller).
1
Tel.: +49 30 314 27682; fax: +49 30 314 24499.
0927-0256/$ - see front matter  2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.commatsci.2007.07.034

these activities is the reduction of electronic waste and/or

the hazardous substances within (e.g., Pb). One example in
this context is the lead-free legislation initiative started by
July of 2006 in the EU according to RoHS, [4]. From the
materials science point-of-view this directive leads to new
tasks and problems. In particular it is necessary to, rst, evaluate adequate lead-free materials and, second, investigate
their material properties experimentally as well as
theoretically.
A central issue that considerably aects the joining
capability of microelectronics is the micromorphological
development within the solder materials during use. In
what follows we want to turn the attention to special kind
of phase separation process, namely spinodal decomposition
(SD) and coarsening (C). We briey show some experimental results and present an approach allowing for the
quantication of these processes from the theoretical
point-of-view.

222

T. Bohme, W.H. Muller / Computational Materials Science 43 (2008) 221228

Fig. 1. The development of the minimal feature size and the number of
transistors in microelectronics.

2. Microstructures in solders
2.1. General remarks
In modern electronic packaging, e.g., ip-chips, solder
balls (or bumps) play an important role, cf., Fig. 2. On
the one hand side they guarantee the electrical connection
between the chips and the electronic circuits, on the other
hand they provide the mechanical connection of the dierent electronic components on the printed circuit board.
From a microscopic point-of-view these solder balls are
basically exposed to two dierent micromorphological
changes, cf., Fig. 3: (a) the formation of scallop-shaped
intermetallic compounds (IMCs) at the interface solder/
substrate and (b) spinodal decomposition and coarsening
in the bulk.

IMCs are formed and grow due to an interfacial reaction. In the case of a Cu substrate and an Sn-containing
solder (e.g., SnAgCu) this reaction takes place between
Cu and Sn and necessitates a mass transport from the substrate to the solder, cf., Fig. 3 (left) and [6]. However, the
resulting expanding scallops can have a positive inuence on the strength and lifetime of the solder joints,
because they guarantee a dovetail connection. On the
other hand there are stress peaks around the IMCs, leading
to crack initiation in this region. Consequently it is reasonable that the positive eects are limited by a critical size.
In contrast to IMC-formation SD and C are diusion
processes exclusively driven by aspects of thermodynamical
stability and interfacial energy minimization, cf., Section
2.2 and [2]. The resulting composite of dierent phases
can be interpreted as a particle reinforced material in
which the stier phase acts as the reinforcement. Unfortunately mechanical failure, such as cracks, favorably grow
along the phase boundary which result, among other reasons, from thermal mismatching. Thus the benet of phase
dispersion is also limited by a critical phase size.
However, there is a considerable interest in predicting
the micro-morphological temporal development, which is
the basic requirement for further failure analysis of the heterogeneous material. In the following we focus on spinodal
decomposition and on the coarsening process. The binary
alloy AgCu is investigated, in which SD results in two
equilibrium phases, the a-phase (Ag-rich) and the b-phase
(Cu-rich), cf., Fig. 3 (right). It has qualitatively a similar
miscibility gap (phase stability data) as the previously used
SnPb solders, and all required material parameters can
easily be obtained.
2.2. Spinodal decomposition and coarsening
From a thermodynamical point-of-view a binary alloy
AB decomposes into two equilibrium phases a and b due
to a gain of the GIBBS free energy G(c, T), which is a function
of mass concentration c  cB (cA + cB = 1) and temperature
T. Without loss of generality we put Tmelt  Teut where Teut
denotes the eutectic temperature. In Fig. 4 two dierent

Fig. 2. Schematic illustration of a ip-chip assembly and the various kinds

of microstructures occurring in the solder ball.

Fig. 3. Left: scallop-shaped Cu6Sn5-IMCs at the interface solder (SnAgCu)/substrate (polished Cu). Right: SD and C in eutectic AgCu after 20 h
heat treatment at 1000 K.

Fig. 4. Schematic GIBBS free energy curves for two dierent temperatures.

T. Bohme, W.H. Muller / Computational Materials Science 43 (2008) 221228

G-curves are illustrated, one for T > Teut and one for
T < Teut.
Obviously the curve describing the liquid phase is completely convex. Thus a mixture with arbitrary c is stable to
all uctuations. This fact is evident considering a uctuated
two-phase-system, represented by the states B in the neighborhood of the original one-phase-state A, cf., Fig. 4. The
resulting energy G of the two-phase-mixture is given by the
energetically disadvantageous connecting line. Therefore
the system remains in the one-phase-state.
This fact changes for the solid state as illustrated in
Fig. 4. In this case G(c, T) is piecewise concave. This region
is called the spinodal area, enclosed by the spinodal concentrations csp
1=2 , and it characterizes the concentrations
for which the system is unstable. Evidently any uctuating
system represented by the connecting line of the neighboring perturbed states is energetically advantageous. Consequently the system will decompose into the equilibrium
concentrations ca/b. Note that ca/b can be constructed by
the so-called common tangent rule, which results from a
thermodynamical stability analysis, [9]:

oGc; T 
Gcb ; T  Gca ; T

:
1

oc
cb  ca
cca=b

3. Theory of spinodal decomposition and coarsening

After the whole system reaches the equilibrium concentrations coarsening begins in such a way that the number of
precipitated phase regions decreases whereas the size of
the phases increases. This process is often called OSTWALD-ripening and minimizes the interfacial energy of the
system. Fig. 5 shows various experimentally observed
stages of coarsening after 2 h, 5 h, 20 h, and 40 h of heat
treatment in eutectic AgCu.

J q0 M  rlB  lA :

223

3.1. A continuum model for diusion

In what follows we consider a binary alloy A-B in which
the (local) composition is characterized by the mass concentration c  cB. Generally SD and C in the solder bulk
can be described as a diusion process within a closed system such that:
q0

oc
rJ 0
ot

partial mass balance:

q0 denotes the mass density of the homogeneous reference

state and J  JB =  JA stands for the diusion ux in [kg/
m2s] which must be specied by a constitutive equation.
Obviously FICKian diusion, i.e., J/q0 =  D $c, cannot
describe the decomposition processes (uphill diusion),
because the elements of the diusion coecient matrix Dij
in [m2/s] are always positive and, consequently, J is directed opposite to $c. Therefore all gradients decrease which
is in contradiction to the decomposition processes.
A revision of this theory by means of the thermodynamics of irreversible processes (TIP), [9], results in the following constitutive law for binary alloys:
3

This equation can be understood as a denition for the

mobility matrix M in [m5/Js] which is usually a function
of c. Furthermore it is obvious that the diusion ux J is
driven by the gradient of the chemical potentials lB  lA.
From the thermodynamical point-of-view the chemical
potential of the mth component, lm, denes the specic
GIBBS free energy-gain/lost Dw of a unit volume resulting
from changing the mole-number by DnB. In the special case
~ T ,
of a closed, homogeneous, and binary system (w wc;
DnB = DnA) we nd that lB  lA = ow/oc, [2].
For a heterogeneous system w additionally depends on
gradients of c. Consequently, the chemical potentials have
to be reformulated by a variational derivative dw/dc, [3]. It
reads:
 
dw
J q0 M  r
:
4
dc
~ ; c; rc; rrc; e the
By choosing a dependence w wT
following equations are found according to the second
law of thermodynamics, [5]:
dw ow
ow
ow

r
rr :
;
dc
oc
orc
orrc

with:
w wconf c; e  ac; e : rrc bc; e : rcrc;
wconf w0 c wel c; e;
1
Hooke 1
r : e e : Cc : e:
wel
2
2
Fig. 5. SD and C in eutectic AgCu at 970 K. Dark: b-phase. Light: aphase.

w0 stands for the purely thermodynamical part of the GIBBS

free energy density which is usually exploited in order to

224

T. Bohme, W.H. Muller / Computational Materials Science 43 (2008) 221228

construct phase diagrams, wel identies the elastically

stored energy density resulting from mechanical loading,
r denotes the stress tensor, C is the 4th order stiness matrix, and e ~ex; t represents the (linearized) strains. Furthermore the symbols a and b are higher gradient
coecients (HGCs) which are characteristic of the surface
tensions on the (smoothly changing) phases boundaries. A
combination of Eqs. (4)(6) the following expression for
the ith component of the diusion ux results:

o ow0 wel 
o2 c
oAkl oc oc
J i q0 M ij

 2Akl
oxj
oc
oxk oxl
oc oxk oxl

2
oAkl oc oemn
o akl oeop oemn oakl o2 emn
2


;
oemn oxk oxl
oeop oemn oxk oxl
oemn oxk oxl
7
with Akl = oakl/oc + bkl.
Eqs. (2), (7) form a PDE for the concentration eld
c(x, t) the so-called extended diusion equation (EDE).
In order to determine the strain eld e(x, t) the additional
momentum balance for mechanical equilibrium, namely
T

r  rT r  Cc : e 0

must be solved together with appropriate boundary conditions, [5]. Note that thermal strains e* can be incorporated
by putting:
e!ee

with

e aT  T ref ;

where a denote the thermal expansion coecients. In this

case the elastically stored energy density and the momentum
balance
change
to
wel 12  : C : 
and
T
r  C :  0. Thus the diusion ux Ji of Eq. (7) incorporates three dierent eects:
classical FICKian diusion represented by the term ow0/
oc,
surface tension eects characterized by terms including
the HGCs and
(thermo-)mechanical loadings leading to the resulting
strains  and to wel.
It is easy to see that the FICKian part $(ow0/oc) = (o2w0/
oc )$c already allows for a description of downhill and
uphill diusion depending on the sign of the curvature
of w0 (cf., Fig. 4). In particular negative curvature within
the spinodal region results in a diusion ux amplifying
$c as required for decomposition processes.
2

e(x,t)  a(x,t)DT, (ij = 0, "i,j 5 1). Then the EDE

reduces to:


oc
o
o ow0 wel 
o2 c oA oc oc

M
 2A 2 
ot ox
ox
oc
ox
oc ox ox
2
2 
oA oc o o a o o oa o 


2
:
10
o ox ox o2 ox ox o ox2
~ 0 c;
e c; w0 w
Note the following dependencies: M M
~ el c; ; a ~ac; ; b ~bc; , and A Ac;
e .
wel w
It is worth mentioning that the restriction to line strains
requires large stresses in order to avoid deformations in the
second and third dimension. However, this 1D-case enables
us to nd a closed expression for the strains (x, t) from Eq.
(8). For that reason we assume linearity for the stiness
C  Cij11 and for the thermal expansion coecient a  a11:
Nc 1  HcNa HcNb ;

H fC; ag;

Hc

c c
cb  ca

with

c ~cx; t:

11

where Na/b are the corresponding values of the equilibrium

phases a and b. By means of HOOKEs law rij = Cij11(e  e*)
we obtain in VOIGT notation:
r11 C b11  hcC b11  C a11 e  e ;
r22 C b12  hcC b12  C a12 e  e ;
r33 C b13  hcC b13  C a13 e  e ;

12

r12 r13 r23 0;

with e* = [ab  H(c)(ab  aa)](T  Tref). Consequently the
only nontrivial solution from the momentum balance is:
dr11
0 ) r11 r0 const:
dx1

13

Thus we can nally write for the strains and for the elastically stored energy density:
r0
x; t b
;
cb cx;t
C 11  cb ca C b11  C a11
14
1
r20
:
wel c
2 C b11  cb cx;t
C b11  C a11
cb ca
With that wel does not contain thermal strains which cancel
due to the restriction to the 1D-case.
Eqs. (14)1,2 can be inserted into Eq. (10). The resulting
equation still represents a nonlinear PDE of 4th order for
the last unknown quantity, the mass concentration eld
c(x, t). A numerical solution of this equation and a subsequent quantitative investigation of SD and C requires an
exact knowledge of the occurring material data which is
the subject of the following section.

3.2. Thermo-mechanical diusion in 1D

3.3. Material parameters

In the following we want to concentrate to the onedimensional case xi = x1  x, i.e., we simply consider the
diusion process along a line. Furthermore so-called
line strains are introduced by putting 11   = (x,t) =

In order to qualitatively investigate Eqs. (10), (14) the

following material data need to be known:
fw0 c; Mc; Cc; ac; ; bc; ; Ac; g:

T. Bohme, W.H. Muller / Computational Materials Science 43 (2008) 221228

Specically we investigate the binary brazing material Ag

Cu at T = 1000 K and put
A  Ag; B  Cu; c  cCu :

225

Table 2
Characteristic concentrations, diusion coecients, and mobilities of Ag
Cu at 1000 K
h 2i
h 5i
Phase
ceut
ca/b
csp
1=2
M mJs
D ms

Note that all other (indirectly required) characteristic material data, e.g., ca=b ; csp
1=2 ; ceut , follow from w0 by applying the
common tangent rule.

a
b

3.3.1. The Gibbs free energy density

In order to determine w0 we used MTdataTM which contains information on phase diagrams from calorimetric
measurements. It provides a eld of discrete values w0(ci),
ci = {0, 0.01, 0.02, . . . ,0.99, 1} which must be tted in
order to obtain a closed form required for compute coding.
Here we perform a polynomial t according to the MARGULES-ansatz

On the other hand the mobility denotes the factor of proportionality introduced in the corrected diusion Eq.
(3). A comparison of the original FICKian law with the
FICKian part of the EDE leads to:

a=b 
o2 w0
D

Dc Mc 2 ) M a=b 2 
:
16
o w0 
oc

w0 c 1  cga cgb gc RT c ln c 1  c ln1  c

c1  cvI c vII 1  c;
15
 J 
with R 8:314 mol K , (universal gas constant). The t
parameters ga/b/c and vI/II are compiled in Table 1. The corresponding curves are illustrated in Fig. 6 (rst row) and
the resulting characteristic concentrations are presented
in Table 2.
3.3.2. Stiness and mobility
Regarding the third column of Table 2 the equilibrium
concentrations ca/b are highly Ag- or Cu-rich, respectively.
Hence we put for the stiness:
C a11  C A
11 124 GPa;

C b11  C B11 168 GPa:

Table 1
The t parameters required for w0(c)
 
 
 
ga GJ
gb GJ
gc mole
m3
m3
m3
7.27

5.20

vI
5

1.11 10

GJ
m3

2.97

vII

GJ
m3

3.01

0.29
0.29

0.063
0.945

0.19
0.79

oc2

1.01 1014
4.09 1015

7.25 1025
3.65 1025

cca=b

a/b

The diusion coecients D can easily be found in the

literature, [12], where they are determined from tracer
experiments of Cu in Ag and Ag in Cu, respectively. Table
2 shows data obtained for Da/b as well as for Ma/b. The
functional dependence C11(c) and M(c) given by the linear
interpolation of Eq. (11) is illustrated in Fig. 6 (2nd row).
3.3.3. Higher gradient coecients
The HGCs are the key quantity when determining the
coarsening rate. In particular values which are too high
lead to overvalued coarsening rates and vice versa. Unfortunately the HGCs are very poorly documented in literature and, even if found, they are frequently ad hoc
estimates the source of which is not clear. Furthermore
we could only nd constant HGCs so that Eqs. (7), (10)
would reduce to the rst two terms.
Due to these shortcomings a theoretical framework
based on atomic interactions was developed which allows
the calculation of the HGCs as functions of c and , [1].
Because of space constraints we only present the relevant
results illustrated in Fig. 7. Note that the HGCs are not
symmetric with respect to positive or negative strains.
Therefore we can distinct between the eects of compressive and tensile loadings during the diusion simulations.
The functions underlying Fig. 7 are very lengthy and
thus the calculation of the HGCs is extremely time-consuming. In order to optimize computation-time we perform
a bilinear interpolation of the form:
Nc;  k N0 k Nc c k N  k Nc   c

with

N fa; b; Ag:

17

The symbols k N0=c==c denote coecients and can be determined by a t. Here we used the atomistically calculated
HGCs of Fig. 7 at the points:
c;  fca ; 0; cb ; 0; ca ; 0:2; cb ; 0:2g:

Fig. 6. 1st row, left: GIBBS free energy density according to Eq. (15). Right:
2nd derivative of w0. 2nd row, left: Stiness as a function of c. Right:
Mobility depending on c.

It is worth mentioning that the characteristic strains

 = 20% are chosen ad hoc; here dierent values are possible. Note that the tting must be performed separately for
positive and negative strains. Table 3 shows the corresponding values.

226

T. Bohme, W.H. Muller / Computational Materials Science 43 (2008) 221228

Table 4
Numerical parameters used for AgCu at 1000 K

GJ

Grid points N

Time step D~t

2pL [lm]


w

512

1 107

0.1

0.2

m3

(ii) Application of the Discrete FOURIER Transformation

(DFT) to the resulting PDE in order to substitute
the spatial derivatives by an algebraic expression.
Here one can use the approximations:
 
o
Y
n~x sY ;
o~x
 2 
o
Y
n~x~x sY  with
o~x2
18
 
i
2ps
n~x s  sin
;
h
N

 

2
2ps
n~x~x s 2 cos
1
N
h
Fig. 7. The HGCs calculated as functions of c and  by means of the
embedded-atom method.

Table 3
Coecients for the HGCs as bilinear functions of c and 
N

k N0 in N

k Nc in N

k N in N

k Nc in N

>0
a
b
A

4.04 1011
5.84 1011
1.53 1010

8.74 1011
4.72 1011
3.74 1011

1.19 1010
1.74 1010
7.73 1010

3.31 1010
1.98 1010
2.87 1010

<0
a
b
A

4.04 1011
5.84 1011
1.53 1010

8.74 1011
4.72 1011
3.74 1011

3.07 109
4.18 v 109
2.55 109

7.82 1010
2.34 109
8.45 1010

Consequently all required quantities are determined and

we turn the attention to the numerical solution of the onedimensional EDE, i.e., Eq. (10).
4. Numerical solution

and s = 0,1, . . . ,N  1. The symbol Y[e] stands for

the DFT of the dummy variable e, h = 2p/N represents the dimensionless distance between the grid
points in real space, s is the position in FOURIER space
and N is the total number of grid points. These
approximations hold for an arbitrary function f(x).
A central dierence quotient was applied to the
derivatives of f and the results were transformed by
DFT. By means of the shift theorem Y f a b
h
i
2pbs
exp  i N  Y f a the relations in (18) follow,
[5,10]. Thus Eq. (10) is rewritten to the following
ODE in FOURIER space:
 b b

d^c
dY c
c M
a
:


c
n~x~x Y K
d~t
d~t
Ma


 
 2 

Ma
oc oK
oK
1 b
Y c 2
Y
19
o~x o~x
o~x
M
with the denition for the symbol K:

4.1. Mathematical procedure

At this point we will outline the procedure required for
solving the PDE (10) numerically. Four steps are performed:
(i) Transformation of EDE (10) to a dimensionless form.
For that reason we introduce the following quantities:
x
~x ;
L

~ conf w0 wel ;
w

w

~t

 bt
wM
;
L cb  ca
2

 have been
where the numerical parameters L and w
chosen as shown in Table 4. Now the resulting equation represents a PDE for c ~c~x; ~t depending on
the dimensionless position ~x and on the dimensionless time ~t.

~ conf
ow
2
o2 c
  2 k A0 k Ac c k A  k Ac   c 2
o~x
oc
wL
 2
 
1
oc
2
o
A
A
A
A
  2 k c k c 
  2 k  k c c
o~
x
oc
wL
wL
 2
oc
1
  2 k a k ac c

o~x
wL
"   
  2 #
2
2
o  oc
o o c

:
20
2
oc
o~x
oc o~x2

Note that spatial derivatives have been used in Eq.

(19) due to a lack of space. The full expression reads
exemplarily Y o~x c  o~x K Y fY 1 n~x^c  Y 1 n~x Y Kg.

T. Bohme, W.H. Muller / Computational Materials Science 43 (2008) 221228

227

(iii) Solving the resulting ODE in FOURIER space by using

an explicit 3rd order RUNGEKUTTA time step method
(Simpsons rule):
1
^cn1 ^cn k 1 4k 2 k 3 ;
6
1
k 2 D~t  g^cn k 1 ;
2
~
k 3 Dt  g^cn  k 1 2k 2

k 1 D~t  g^cn ;

21

where g identies the right side of Eq. (19).

(iv) Transformation of the discrete numerical solution
^cis : ^cs; ~ti ; s 1; . . . ; N into real space by means
of the inverse DFT. The resulting discrete concentration array c~xk ; ~ti at time ~ti can be interpolated in
order to achieve the concentration eld c~x; ~ti . Furthermore the solution can be transformed to c~x; t
by means of the relation ~tt 3:06  t, cf., item (i).

Fig. 9. Simulation of SD and C in AgCu at 1000 K (r0 =  5000 MPa).

Note that the application of DFT restrict the solution

c~x; t to a representative line element characterizing
the periodic continuative material, i.e., we apply periodic
boundary conditions.
4.2. Simulation of SD and C
The aforementioned procedure was implemented in Fortran 95 where we used the numerical parameters displayed
in Table 4 and the FFT-subroutines provided by numerical
recipes, [11]. Note that the number of grid points N must be
chosen suciently large in order to model the extremely
narrow phase boundary realistically. Since N and D~t are
not independent the time steps must be reduced analogously in order to achieve stable solutions.
Figs. 810 show various selected results of the simulation of SD and C. We investigated the three cases
r0 = 0/  5000/5000 MPa.

Fig. 10. Simulation of SD and C in AgCu at 1000 K (r0 = 5000 MPa).

The outermost dashed lines represent the equilibrium

concentration whereas the innermost lines identify the spinodal concentrations. The initial c-prole is given by the
(unstable) eutectic homogeneous concentration prole with
one slight uctuation in order to enforce the decomposition
process.

5. Discussion

Fig. 8. Simulation of SD and C in AgCu at 1000 K (r0 = 0 MPa).

In Figs. 810 it seems that the presence of stresses inuences the coarsening process. Especially tensile loadings
appear to have a stronger impact on coarsening than compressive ones. Indeed, the number of b-islands in the last
stage of Fig. 10 is greater than in Fig. 9. But there are indications for phase-coalescence in the immediate future
whereas the b-phases in Fig. 9 uniformly separated. However, it must be emphasized that this statement needs to

228

T. Bohme, W.H. Muller / Computational Materials Science 43 (2008) 221228

Fig. 11. A zoomed interface region of Fig. 84 in order to obtain the

interface boundary width d.

be investigated more closely. In particular longer simulation

times and 2D calculations are necessary for verication.
Furthermore it is noticeable that extremely high stresses
were applied during the simulation. This was done, rst, to
illustrate the eect of stresses within the presented simulation period, which is considerably shorter than the experimental observation period. Second, such exceeding stresses
denote a consequence of the considered line strain case as
briey indicated in Section 3.2 [8].
Moreover, note that the theoretically obtained HGCs
dier to the ones usually2 found in literature, e.g., [8],
aaij 1:45 107 dij N; ab11 ab22 9:64 108 N
and
ab33 1:35 106 N. They are considerably smaller and
result in a narrower interface boundary width d
2.3 nm,
cf., Fig. 11. This value corresponds to almost 8 atomic distances r (atomic nearest neighbor distance in Ag:
) which underscores the case of a sharp interface.
r = 2.88 A
Furthermore these small HGCs lead to two consequences regarding the simulations. First, the discretization
N must be chosen suciently large in order to model the
phase boundary realistically, which results in very small
time steps Dt for stability. This fact is also clear from the
phenomenological point-of-view considering the classical
CAHNHILLIARD equation where the higher gradient terms
have a stabilizing eect on the PDE which tends to diverge
for negative curvatures of w. This eect decreases for smaller HGCs, so that stability can only be achieved by means
of a reduction of Dt. Second, sharper interface boundaries
lead to a decelerated coarsening behavior of the system.
Therefore one has to extend the simulation time in order
to observe and quantify the OSTWALD-ripening process.
6. Conclusion and outlook
In this paper we presented a method allowing for the
theoretical and experimental quantitative analysis of spinodal decomposition and coarsening in binary alloys. An

2
There are similar HGCs, e.g., [7], in which the investigated simulation
time is also extremely short.

extended diusion equation was discussed resulting from

exploitation of the second law of thermodynamics. We
concentrated on the binary alloy AgCu and illustrated
how all required material parameters could be determined
and a numerical solution of the diusion equation in 1D
could be obtained.
In principle the 1D simulations can be used in order to
nd an average phase radius. This radius characterizes the
coarsening stage and can be used for a quantitative comparison of the simulations with experiment. However, it
is more desirable to perform 2D simulations which allow
an optical comparison as well as a direct exploitation by
means of digital image analysis. Furthermore the articial
line strain case, cf., Section 3.2 can be dropped so that
more realistic loading conditions r0 can be applied to the
simulations.
Nevertheless the discussion of the last section shows that
numerical aspects and alternative numerical methods
should be investigated more closely. Here time-adaptive
methods, such as the RUNGEKUTTAFEHLBERG method,
and/or FE techniques are conceivable for future studies.
Acknowledgements
It is a great pleasure to acknowledge the cooperation
with the Weierstrass Institute for Applied Analysis and
Stochastics (Dr.W. Dreyer). Furthermore the work is supported by the Deutsche Bundesstiftung Umwelt (DBU).
References
[1] T. Bohme, W. Dreyer, W.H. Muller, WIAS-preprint (www.wiasberlin.de), No. 1131, ISSN: 0946-8633 (2006).
[2] J.W. Cahn, Trans. Metall. Soc. AIME 242 (1968) 166180.
[3] J.W. Cahn, J. Chem. Phys. 42 (1965) 9399.
[4] Directive 2002/95/EC of the European Parliament and of the Council
of 27 January 2003 on the restriction of the use of certain hazardous
substances in electrical and electronic equipment (RoHS), Ocial
Journal of the European Union (2003) L37/19L37/23.
[5] W. Dreyer, W.H. Muller, Int. J. Solids Struct. 37 (2000) 38413871.
[6] H.K. Kim, K.N. Tu, Phys. Rev. B 53 (1996) 1602716034.
[7] T. Kupper, N. Masbaum, Acta Metall. Mater. 42 (1994) 18471858.
[8] L. Li, W.H. Muller, Comput. Mater. Sci. 21 (2001) 159184.
[9] I. Muller, Grundlagen der Thermodynamik, 3. Auage, Springer,
1999.
[10] W.H. Muller, Zur Simulation des Mikroverhaltens thermo-mechanisch fehlangepater Verbundwerkstoe, VDI Fortschritt-Berichte,
Reihe 18, Nr. 234, pp. 36 . (1997).
[11] W.H. Press, S.A. Teukolsky, W.T. Vertterling, B.P. Flannery,
Numerical Recipes in Fortran 77, second ed., The Art of Scientic
Computing, Cambridge University Press, 1992.
[12] E.A. Brandes, G.B. Brook, Smithells Metals Reference Book, seventh
ed., Reed Educational and Professional Publishing Ltd, 1992.