A Study of the Transient Liquid Phase Bonding

Process Applied to a Ag/Cu/Ag Sandwich Joint

ISAAC TUAH-POKU, M. DOLLAR, and T. B. MASSALSKI

Transient liquid phase (TLP) bonding is a process currently used for joining heat resistant alloys, for
example nickel- and cobalt-based superalloys. It involves the formation of a liquid layer between
two adjoining pieces and the formation of a solid bond as the liquid disappears during annealing at a
suitable constant temperature. In the present study, a model Ag/Cu/Ag sandwich joint associated
with a simple eutectic phase diagram was used to study the different stages of this process. The re-
suits confirm that the TLP bonding is a diffusional process occurring in clearly distinctive stages.
The two most important stages are the widening and homogenization of the previously dissolved liq-
uid interlayer, and the subsequent solidification and shrinking of the interlayer. Whereas the former
stage involves diffusional processes both in the liquid phase and in the adjoining solids, the latter is
controlled mainly by the diffusion in the solid phase. A modeling approach has been explored which
shows that in most eutectic systems there exists an optimal bonding temperature corresponding to
the shortest time needed for complete solidification. The results of a study on a Ag/Ag-20 wt pct
Cu/Ag sandwich joint provide evidence that the use of an alloy close to the eutectic composition as
an interlayer material shortens the TLP process substantially.

I. INTRODUCTION by dissolving some adjacent A. The liquid thus created

homogenizes to a composition C L~. Further diffusion of B
T R A N S I E N T liquid phase (TLP) bonding is a diffusion
into the adjoining solid causes a reversal of the widening
bonding process currently used for joining several types of
process, and solidification sets in as solute B atoms diffuse
heat resistant alloys, for example nickel- and cobalt-based
into A, the base material. This continues until the remaining
superalloys.l It has the advantage of not requiring the
liquid attains the composition C ~z on solidifying and the
rather high pressures needed in typical solid state diffusion
liquid is used up. The TLP process is illustrated in Figure 2.
bonding processes. Like brazing, it employs an "interlayer"
Lesoult has examined some simple analytical models to
material as a bonding agent. Unlike brazing, however, the
describe the TLP process. 4 Another approach, suggested by
process occurs isothermally. The interlayer melts and solidi-
Sekerka, 5 employs numerical solutions related to the two-
fies as a result of interdiffusion with the base material, all
phase moving boundary problem. 6 These models consider
at a constant temperature. As a result, the solidified bond
the influence of such parameters as the bonding tempera-
region consists of a primary solid solution similar in com-
ture, TB, the initial composition of the interlayer material,
position to the base material itself. Thus, the possibility of
and the phase relations in the phase diagram.
the formation of brittle phases in the bond region, as in braz-
Before further generalizations are attempted, it is desir-
ing or other joining processes, is largely avoided. On the
able to test the kinetics of the proposed models in systems
other hand, since the latter stages of TLP bonding involve
where some of the parameters mentioned above can be
solid diffusion the process is very slow, typically requiring
explored experimentally. In this paper, the results of ex-
several hours.
perimental work done on a A g / C u / A g sandwich joint
In principle, TLP is applicable to eutectic 1'2'3or peritectic
associated with a simple eutectic phase diagram are pre-
alloy systems, where a lowering of the liquidus tempera-
sented and correlated with some of the features of the
ture occurs during alloying. For a binary A-B eutectic (Fig-
above-mentioned models.
ure 1)* an interlayer of material B is sandwiched between
Afterward, the solid/liquid interface movement into the
*Figure 1 represents the actual binary eutectic phase diagram for the Ag sides of the sandwich reverses direction as the liquid
Ag-Cu system7 examined in the present case. The designations of the rele- zone shrinks.
vant concentrations in the figure, C ~ C L'~, C Lt3, C t3L, are used in the text
below to describe the general features of the TLP process.

pieces of A. The sandwich is then heated to a "bonding" II. T H E O R E T I C A L DESCRIPTION

temperature TB (above the eutectic temperature Te), in a
vacuum or an inert atmosphere. As a result of an intimate
contact between A and B atoms at the A/B interface, the
interlayer rapidly melts and the molten zone further widens
ISAAC TUAH-POKU is Senior Member of Technical Staff, Surface
Mount Technology, AT & T, Engineering Research Center, Princeton,
NJ 08540. M. DOLLAR (permanent address: Academy of Mining and Met-
allurgy, Krakow, Poland), Research Associate, and T. B. MASSALSKI,
Professor, are with the Department of Metallurgical Engineering and
Materials Science, Carnegie Mellon University, Pittsburgh, PA 15213.
Manuscript submitted March 9, 1987.
In the proposed models 4'5 for a binary eutectic system,
the condition of local equilibrium at the interface is as-
sumed between the solid and the liquid. In such a case, the
composition of each phase at the interface is that dictated
by the phase diagram and the equality of the respective
chemical potentials. Because of small thickness of the inter-
layer, the effect of convection in the liquid is assumed to
be negligible so that the problem can be considered as one
of pure diffusion. To simplify the analysis further, the in-
terdiffusion coefficients (Ds, DL) in the solid and in the
liquid are assumed to be independent of composition, and

METALLURGICAL TRANSACTIONS A VOLUME 19A, MARCH 1988 675


0 to z0 ao 40
1200
1ooo ~ ~ L
-
800
600
~ ~
400.
~L .L*~
~00.
o to aO aO
Ag
Fig. 1--Eutectic phase diagram for the Ag-Cu system.7 The marked temperature T = 820 ~ denotes the
actual bonding temperaturein the present study.
the partial molar volumes of A and B in all phases (a, fl,
S, L) are assumed to be equal. The latter assumption makes
it possible to express Fick's second law in terms of mole
fractions.
By considering a moving boundary problem involving
two phases (solid and liquid) the TLP process can be ana-
lyzed in four stages (see Table I):
1. Dissolution of the interlayer B
2. Homogenization of the liquid
3. Isothermal solidification
4. Homogenization of the bond region
In the simplest case, one starts with two pieces of the
base metal A to be joined by a slice of pure B (the inter-
layer material), as in Figure 2(a). The melting of B results
from mixing of A and B atoms at the A/B interface and it
starts when the interface attains the concentration equal
to C ~ Since there is an infinite concentration gradient at
the interface, the dissolution of B is very rapid. The diffu-
sion at this stage is controlled by D L. The total time t~ re-
quired for complete dissolution of the B interlayer may be
of the order of seconds. The driving force for continuous
dissolution of B is proportional to the difference C L~ - C L~,
which expresses the width of the eutectic depression at TB.
During the second stage (which takes time t2), the inter-
layer material is completely dissolved, but the resulting
liquid is inhomogeneous and supersaturated, having an
average concentration of B greater than C L~ (Figure 2(b)).
The liquid zone is not, therefore, in equilibrium with the
adjoining solid and more of it dissolves to dilute the liquid.
The B atoms diffuse into the solid A to form the (primary
solid solution) a-phase at the interface, and it melts when-
676--VOLUME 19A, MARCH 1988
A t o m i c P e r c e n t Copper
~o 60 ?o so go 10o
1~,,..e'7"r
")'80aC
)
Lp ~pL
4o ~ 6o ~o eb 9o 1oo
Weight P e r c e n t Copper Cu
ever the B concentration in the a-phase adjacent to the
liquid exceeds C aL. In this process, the diffusion in the
solid, controlled by the diffusion coefficient D s , becomes
very important. At the same time, the liquid phase under-
goes homogenization by diffusion in the liquid, which is
controlled by DL. The liquid zone continues to widen until
the concentration profile within the liquid flattens to the
equilibrium value C L" (Figure 2(c)). It has been pointed
out by Sekerka 5 that the effective diffusion coefficient dur-
ing this stage may well be a geometric mean between DL
and D s .
During stage 3, isothermal solidification occurs over
time t 3. The solid-liquid interface reverses direction and
the liquid zone begins to shrink as the liquid becomes en-
riched in A and impoverished in B (Figure 2(d)). The rate
of loss of the solute B, from the liquid to the solid a-phase,
controls the interface displacement rate. Since this is deter-
mined by the rate of diffusion of B into the a-phase, the
process is very slow, as would be expected for solid s t a t e
diffusion. The diffusion parameter which controls this
stage is clearly D s . The time t 3 needed to complete the
solidification is typically of the order of hours.
In order to homogenize the bond region further (Fig-
ure 2(f)), additional annealing may be done at an appropri-
ate temperature. Again, D s is the dominant parameter and
the annealing time, t4, during this stage may be quite long.
The bulk of the bonding time in the TLP process is there-
fore taken up by stages 3 and 4.
The first three stages of the TLP process have been
modeled several years ago by Lesoult. 4 Below, we outline
some details of this earlier approach and tabulate informa-
METALLURGICAL TRANSACTIONS A
 (a) INITIAL ASSEMBLY (b) MELTED I NTERLAYER WIDEN ING (C)LIDUID 2~ONE HOMOGENIZES TO CLr
INTO A TO MAXIMUM WIDTH

iA Cp
: ' , 9 , , ,',,..,,"
9 ; , , ," 9 9 , . ' , .
CA

LIQUID
-A
CA
[.
"~176176
cL a

A/BI-~I I A/B !

25 I II -A [ ]1
I

A la A B A B

(d)LIOUIDZONE SHRINKING DUE TO (e) COMPLETELY SOLIDIFIED (f)HOMOGENIZATION OF BOND REGION

LOSS OF SOLUTE BY DIFFUSION TO GET RID OF SOLUTE HUMP

-<

[•[La
CBM

T IXJ\~ / II
I
A B A B A B
Fig. 2--Schematic illustration of the mechanism of the TLP bonding process.

tion which will be helpful in the interpretation of the results
of the present study.
During stage one (dissolution of the interlayer), the solute
conservation requirement in the liquid makes it possible to
express the diffusion conditions through the application of
the Fick's law for the case of unidirectional diffusion:
OC L 02C L
0---~ = DL "--Oy 2
where E and F are constants determined by the specific
boundary conditions. Equation [2] implies that the interface
displacement obeys a general square root law: 4'8
Yl = K l " ~ [3]
where KI is a constant* at the selected temperature TB, dic-
*The subscript 1 in Eq. [3] and subscripts 2 and 3 in subsequent simi-
[ll lar equations refer to the respective stages of the TLP process.

where C L is the concentration of the B component in the
liquid, and y is the distance across the bond region (the ini-
tial position of the solid/liquid interface is chosen as the
origin of the y axis). The well-known solution of Eq. [1] is:
CL ~- E + F " erf(y/4X/~Lt)
tated by the respective phase diagram. 4 The values of K~,
evaluated by Lesoult, 4 for several different forms of the
eutectic phase diagram, as determined by the respective ra-
tios (C L'~ - Co)/(C ~L - C L~) and (C L~ - CL'~)/(C ~L -- CL~),
[2] are given in Table II.

Table I. Stages of the TLP Bonding Process

Controlling Feature Controlled
Stage Parameters Rapidity during the Stage
Dissolution of the interlayer Dt very fast, t . - seconds~ maximum width of the transient liquid interlayer
Homogenization of the liquid D~, Ds fast, t - minutes J
Isothermal solidification Ds slow, t - hours~
Homogenization of bond region Ds slow, t hoursJ total time needed for TLP bonding

METALLURGICALTRANSACTIONSA VOLUME 19A, MARCH1988--677

 Table II. Values of Kt (See Eq. [3]) during the Table III. Values of K3 (See Eq. [9]) during the
Dissolution of the Interlayer for Typical Values of Isothermal Solidification for Typical Values of
(C L" - Co)/(C t~ - C Lt3) and (C ta - C t ~ ) / ( C e~ - C u3) (C ~ - Co)/(C L~ - C ~)

(C L~ - Co) (C La - C L~) (C ~ - Co) (C ~z - Co)

(C ~L - C L~) (C ~L - C Lt3) Kt (C La - C ~ K3 (C Lc~ - C ~ K3
0.5 0.1 0.13 0.018 0.01 1.19 0.40
I 0.44 0.093 0.05 1.73 0.50
10 1.05 0.199 0.10 2.44 0.60
1 0.1 0.16 0.317 0.15 3.40 0.70
1 0.47 0.450 0.20 4.69 0.80
10 1.06 0.602 0.25 6.45 0.90
0.775 0.30 8.85 1.00
2 0.1 0.18 1.000 0.36 12.35 1.10
1 0.51
10 1.08 From Lesoult's report4
F r o m Lesoult's report 4

While the process described above involves the pure

During stage two (homogenization of the liquid), the so-
lution of the diffusion problem is much more complex and
no simple analytical expression has been proposed.
During stage three (isothermal solidification), the initial
boundary conditions are as follows:
metals A and B which form a suitable eutectic system, a
TLP bond sandwich can be made up also, of course, by
choosing a suitable (primary) solid solution as the base-
metal, A, and an alloy of eutectic composition as the B
metal intermediate layer.

C ~ = Co, y<0
Ca = C~ y = 0
C L = C L~, y > 0
The displacement of the interface now results from the
dilution of the B component in the solid a phase, driven
by the concentration gradient O C ~ / O y . Again, the solute
conservation requirement--now in the s o l i d - - m a k e s it pos-
sible to apply the Fick's law for the case of unidirectional
diffusion:
OC ~ 02C ~
0---~- = D s ' - - O y 2
where C ~ denotes the concentration of the B component in
the solid. The well-known solution for the composition
profile in the a-phase satisfying the above set of conditions
[4] is as given below: 4
C a = M + N. erf(y/X/4Dst)
where M and N are constants determined by the specific
boundary conditions. The corresponding interface displace-
ment during stage three will obey another form of the square
root law:
Y3 = /(3 " 4%/-~st
where D s is the diffusion coefficient in the solid phase and
K3 is a constant. It has been shown by Lesoult 4 that K3 is
the solution of the general equation:
C ~ - Co
K3" exp(-K~) 9 (1 + erf K3)%/--~ = CL~ _ C ~
Thus, for a given ratio of ( C ~ - C o ) / ( C L'~ - C~162 as
dictated by the phase diagram and the selected TB tempera-
ture, the value of K3 may be obtained using numerical meth-
ods. The values of K3 given in Table III are those derived
in Lesoult's paper. 4
[4] III. E X P E R I M E N T A L
P R O C E D U R E S A N D RESULTS
A. Choice of a Suitable Binary System
(i) In order to study the kinetics of the TLP bonding pro-
cess, several low melting binary alloy systems were con-
sidered. The selection criterion was that the A-B system
must have a suitable eutectic region, preferably with the
eutectic composition in the base-metal rich zone, and also
that there must be a substantial solubility of the solute in
the base metal.
[5]
Preliminary attempts to form a suitable sandwich joint
were made with A1-Cu, A1-Zn, A1-Ag, Pb-Sn, and Ag-Cu
systems. In the Al-based systems, the tenacious oxide at-
tached to chemically cleaned A1 surface made bonding and
precise interface displacement measurements difficult. In
addition, eutectic composition alloys in these systems are
[6] brittle and difficult to roll into foils of desired width (25 to
100/xm thickness). Pb-Sn was rather soft, making section-
ing difficult. The Ag-Cu system proved to be the easiest to
use since there is no oxide problem and the alloys are rela-
tively malleable.
(ii) Ag and Cu of 99.999 pct purity were used. Ag was
cast into cylindrical rods of 8 m m diameter and cut into
[71 wafers about 2 mm thick. The interlayer foil was cut into
small circular discs of 4 m m diameter, to fit into a Ta
washer of 4 mm ID, punched from foil of 100/xm thick-
ness. Interlayer foils, 80/.Lm thick, were made of pure Cu
and Ag-20 wt pct Cu. The Ta washer/interlayer disc com-
bination was then sandwiched between two pieces of pol-
ished Ag wafers. The assembly (Figure 3) was held in
[8]
place by a miniature clamp.
B. W i d e n i n g o f the M e l t e d I n t e r l a y e r
Preliminary attempts to observe the process of initial
dissolution of the interlayer material were unsuccessful due
to the fact that this stage of the process is very rapid and

6 7 8 - - V O L U M E 19A, MARCH 1988 METALLURGICAL TRANSACTIONS A

Fig. 3--Specimen assemblyused for the present TLP studies (the width
of the interlayer foil is exaggeratedfor clarity).

Fig. 4--Stage 2 widening. Micrographs show appearance of the liquid

may be of the order of a fraction of a second. The widening region after solidification (annealing temperature 820 ~
process which results as the liquid zone undergoes homoge-
nization is also a rapid process, but is experimentally ob-
servable if the initial interlayer material is thick enough. bond region after annealing for 10 and 15 minutes at 1093 K
By selecting a pure Cu foil of about 80/xm thickness the are shown. As illustrated in Figure 4, the liquid region is
process is prolonged to the order of minutes. still widening with time so that the effective width after 15
A Pt-10 Rh thermocouple with a digital readout was at- minutes should be larger than after 10 minutes. However, a
tached to each sample at the joint. A heating curve was narrower section of the liquid zone in the 15-minute sam-
recorded until the temperature reading was within ---5 ~ ple was selected just to show the relative Cu concentration.
of the selected Tn. Subsequently, reaction times were al-
lowed to elapse while within this temperature range. Sev- C. Shrinking of the Interlayer
eral identical samples were inserted in a pre-heated furnace As mentioned above, this stage of the process is very
(with argon atmosphere) for various durations, followed by slow because the isothermal solidification is controlled by
rapid quenching in an ice-water bath. solid state diffusion. This allows a fairly accurate determi-
In the present study, widening of the melted interlayer nation of the width and time parameters. To make an esti-
was investigated in an A g / C u / A g sandwich joint. The sam- mate of the apparent diffusion coefficient controlling this
ples were annealed at 1093 K for 5, 8, 10, 15, 20, and 30 stage of the process, an accurate assessment of the width
minutes and quenched. They were then sectioned metallo- of the liquid zone was again of prime importance. Iden-
graphically in several regions. The observed widening of tical samples of the Ag-Cu-Ag sandwich were encapsulated
the interlayer is illustrated in Figure 4. in vycor tubing under argon gas (corrected to one atmo-
The "equivalent width" of the liquid zone was measured sphere at 1093 K). At the end of the heat treatments, the
on appropriate microsections. To obtain the equivalent
width, W, it was assumed that all observed dendrites (Fig-
ure 4) were formed upon quenching (i.e., that the region Table IV. Data for Analysis of Stage 2 of the TLP Process
was completely molten before quenching). The equivalent
width was then defined as the average width of the liquid Annealing Width of Interface
zone which has the same area as a given length of the non- Time, t, at Liquid Inter- Displacement
linearly bounded liquid zone actually observed. Thus, W is 1093 K layer, Wz Log t Y2 Log Y2
given by the ratio of the area obtained from a planimetric [min] [/xm] (t[s]) [/xm] (yz[cm])
measurement divided by the length of the liquid zone under 0 79 -- -- --
consideration. The data for t and W are given in Table IV 5 266 2.48 93.5 --2.03
and the W(t) plot is shown in Figure 5. 8 300 2.68 110.5 -- 1.96
In the present study, electron microprobe analysis was 10 351 2.79 126.0 -- 1.87
carded out to investigate homogenization of the liquid ac- 15 336 2.95 158.5 --1.80
companied by the widening process. In Figure 6 the con- 20 420 3.08 170.5 -- 1.77
30 419 3.26 170.0 -- 1.77
centration profiles of Cu in the liquid zone and the adjacent

METALLURGICALTRANSACTIONSA VOLUME19A,MARCH1988--679
 -4
40C

uJ
7 300

z /
O
N

W
WIDENING OF M O L T E N ZONE
200
o
u_
o
"I-
I.-
c~ IOO

=- Ilnitiol Thickness of Foil

I I I
I0 20 30
TIME (minutes)

Fig. 5--Width of the molten zone vs annealing time at 820 ~ (stage 2).

Cu CONCENTRATION PROFILE (MICROPROBE ANALYSIS)

--- 15 Minutes
-- t 0 Minutes Fig. 7--Stage 3. Micrographs show shrinking of liquid interlayer (an-
32.0 820~ Ag-Cu A , nealing temperature 820 ~

.~,_o_--_z__~ _ _/~ . . . . . . . . . . . . .
IV. DISCUSSION
A. Dissolution of the Interlayer
We consider first Eq. [3] expressing the lateral displace-
ment of one of the interfaces:

.... \, Yl = K I " ~ [3]

Ag
Fig. 6--Concentration profiles of Cu in the liquid zone after annealing
for 10 and 15 min at 820 ~ (Concentration inferred from microstructure
at room temperature after solidification.)
The width of the liquid zone attains the initial width W0,
corresponding to the width of the inserted interlayer, after
the time tl required for complete dissolution. Equation [3]
can be rewritten as:

Wo = KI " S/4DLt~ [9]

2

vycor capsules were broken in an ice-water bath to ensure Table V. Data for Analysis of Stage 3
of the TLP Process (Wmax = 420 pm)
rapid quenching. The samples were then sectioned and the
equivalent width, W, of the liquid zone was determined as Annealing Width of Interface
in Section III-B. Time, t, at Liquid Inter- Displacement
The shrinking of the interlayer is illustrated in Figure 7. 1093 K layer, W3 Log t Y3 Log Y3
The measured values of W are given in Table V. The times [Hours] [/xm] (t[s]) [/zm] (y3[cm])
recorded in Table V have been corrected for the initial 3 354 4.03 33.0 -2.48
time required to reach 1093 --- 5 K which was between 14 5 347 4.26 36.5 - 2.44
and 16 minutes. The corresponding W(t) plot is shown in 12 290 4.64 65.0 -2.19
Figure 8. 18 253 4.81 83.5 -2.08
The logarithmic scale of time applied in Figure 9 allows 24 246 4.94 87.0 -2.06
the plot to be expressed as W vs t for both widening and 48 172 5.24 124.0 -1.91
72 123 5.41 148.5 -1.83
shrinking of the molten zone.

680--VOLUME 19A, MARCH 1988 METALLURGICALTRANSACTIONSA

 4OO rounding matrix. The lateral displacement of one of the
ISOTHERMAL SOLIDIFICATION interfaces may be expressed in terms of the width of the
::k

txl
measured molten zone, W, as:
7,
O
N 300 W - Wo
7 Y2 = 2 [lla]
t,l

--1
O As indicated earlier, no simple solution of the diffusion
20O problem during this stage of the TLP process has been pro-
posed. However, since the displacement of the interfaces
32
I--"
n
is undoubtedly diffusion controlled, it can again be consid-
ered to be governed by a general square root law:
LO0
7"
7"
Y2 = K2 9 V~4De t , [llbl
<
where now De is an effective coefficient of diffusion depen-
UJ
Cr~ I I I I I I I dent on specific conditions to be determined. Combining
I0 20 50 40 50 60 70 80
T I M E (hrs)
the effective diffusion coefficient and the constant K2 into
a general kinetic constant, a, and assuming that the time
Fig. 8 - - W i d t h of the molten zone vs annealing time at 820 ~ (stage 3).
exponent can have values other than 0.5,* we can write a

*In general, it is unlikely that the time exponent should differ signifi-
cantly from 0.5, if the process is diffusion controlled in the usual way.

HOMOGENIZATION OF MOLTEN ZONE 8~ S O L I D I F I C A T I O N more general expression:

500
3'2 = a 9 t" [12a]
According to the plot of W vs t shown in Figure 5, W at-
tains a maximum width, Wmax, after a time t 2 of between
18 and 20 minutes. At that stage one would expect a ho-
0
N 300--
mogeneous liquid of composition C L~ (about 18 pct Cu) at
z 1093 K. This expectation has been confirmed by means
bJ
F-
.J of the electron microprobe analysis. The superposition of
0 .'"
200-- the concentration profiles obtained after 10 and 15 minutes
of holding at 1093 K shows that after 10 minutes the mean
.x-
I-.
173
concentration of Cu in the liquid zone (C = 20.8 pct) is
.~ 100- close to but still higher than C L~, whereas after 15 minutes
.... llni~io, In,erloyer Thickness
the mean concentration (C = 17.8 pct) does not signifi-
~5rn I01rain 30,rain 3hr 12hr 72hr cantly differ from C L~.
10 2
b
I0 s
iI
10 4.
I i I
105 seconds
Converting Eq. [12a] into a logarithmic form:
T I M E (log)
log Y2 = log a + n log t [12b]
Fig. 9 - - W i d t h of the molten zone vs log of annealing time at 820 ~
(stages 2 and 3).
the significance of the linear relationship between log Y2
and log t for the values given in Table IV (for t -< 20 min-
utes) can be examined. Values of log a = - 3 . 1 6 and n =
0.46 are obtained from the straight line best fit to the experi-
mental data. The corresponding plot of log Y2 vs log t is
yielding: shown in Figure 10. Since the correlation coefficient*
*The data in Table IV were treated statistically in the usual way. The
tl- 16K~DL [10] value of the obtained correlation coefficient (0.99) indicates a statistical
error not exceeding 1 pct.l~
In order to obtain numerical values of t~ we make use
of Table II to derive the appropriate values of K1. For the equals 0.99, for the present results Eqs. [12a] and [12b]
Cu-Ag system at 1093 K, K1 is about 0.33. We estimate may be taken to represent an acceptable description of the
Dz = 1.2 9 10-s cm2/s from published data. 9 With the ex- time dependence of Y2. The value of the slope n - 0.46 is
perimental value of the inserted interlayer W0 = 79 9 1 0 - 4 c m , sufficiently close to 0.5, the value one would expect for a
we obtain t~ ~ 3 seconds. Thus, with respect to the subse- diffusion-controlled square root law for a moving bound-
quent widening, the dissolution process may be considered ary. However, if Eq. [12a] is expressed as Y2 = a 9 t m/2,the
to be instantaneous. initial value of the constant a = l 0 -3"16 = 6.91 9 10 4 used
in the plot of this relationship makes the straight line devi-
B. H o m o g e n i z a t i o n and W i d e n i n g o f the Liquid L a y e r
ate from the experimental data by some 30 pct.* A much

If the initially inserted and subsequently molten interlayer *The use of this value of a leads to log Y2 = - 3 . 1 6 + 0.5 log t. This
straight line deviates sharply from experimental data, as shown in Fig-
has the width W0, this width will increase over time t2 as the ure 10.
liquid/solid interfaces on both sides penetrate into the sur-

METALLURGICAL TRANSACTIONS A VOLUME 19A, MARCH 1988--681


Yz =- 3.16 + 0 . 4 6 t
-I.5 Y2 =-3.16 + 0.50t //
Y2 =-3.28+0.50t / / / / ~
-2.0
////~.7
8'
// /// //./~
._.1
-2.5
//~.//
/~.//
/~///-
-5.0
/ stage two.
/
o l'.o 2'.0 31o
Logt
Fig. 1 0 - - L o g Y: vs log t relationships (for details please see the text).
better fit is obtained (still retaining n = 0.5) if a is assumed
to be 10 -3.28 = 5 . 2 5 " 10 -4 (see Figure 10). Thus, the equa-
tion which reasonably describes the present results can be
expressed as Y2 = 5.25 9 10-4t ~
It is clear that the effective diffusion coefficient D e is likely
to be related to a combination of D L and D s at the respec-
tive compositions of liquid and solid in contact, but the exact
relationship is not known. To a first approximation, it may
be reasonable to assume that the predominant difference
between stage one and stage two melting is the difference
in the effective diffusion coefficients. If we accept that the
constants are likely to be similar in both stages, i.e., K: =
KI = 0.33, and if we use the value of a2 = 5.25 x 10-4
derived above to obtain the best fit with experimental data,
e q u a t i o n D e = (a:/2K2) 2 yields D e ~ 6.3 x 10 -7 cm2/sec.
1. Analysis of the effective diffusion coefficient
Utilizing the published literature data for diffusion in Cu-
Ag alloys, 9'11J2Dc at C L~ = 18 wt pct Cu in Ag is estimated
as 1.2 x 10-5 cm2/sec, and D~ at C ~ = 7 wt pct Cu in
Ag is estimated as 4.6 • 10 -1~ cm2/sec. If neither D~ nor
DL is dominant, D e may be expressed as a simple geometric
mean of Ds and DL, yielding
De = D~ 1/2 " ~Ln1/2 = 7.4 X 10-8 cm2/sec
which is an order of magnitude lower than the value esti-
mated in the previous section on the basis of the W vs t
analysis.
In order to increase De, the influence of D L must be
more pronounced over D~. One way to achieve this is to
consider that De could be expressed in terms of a "weighted"
a c(,b
geometric mean, DL x Ds , w h e r e a + b = 1. Now, in
order to arrive at the experimentally derived value of De,
682--VOLUME 19A, MARCH 1988
and utilizing the D L and D~ values quoted above, the re-
quired relationship becomes D e = D~ 7D ~03, which clearly
underlines the expected dominance of diffusion in the liquid
// phase. There are two processes occurring simultaneously
/v~/ during the homogenization and melting in stage two: dif-
fusion in the liquid and in the solid. The former is clearly
the faster process. In parallel processes of this type, the
rate controlling process is always the faster process, so that
the bulk of Cu atoms from the original interlayer material
remains in the liquid zone with only a limited diffusion oc-
curring into the solid. It is therefore reasonable that DL
should be the dominant factor in stage two.
The physical sense of the above analysis, as well as those
of Sekerka and Lesoult's models, 4'5 is essentially as fol-
lows: a flat interface with negligible thickness and liquid
on one side moves into the solid. The rate of interface dis-
placement is influenced by processes occurring in the liq-
uid (homogenization to C ~z) and in the solid (enrichment
to C ~ ) . Such a picture is quite limited, but it allows one to
emphasize the still dominant role of the liquid diffusion in
2. Alternative analysis of the effective diffusion coefficient
The interface description in the previous section neglects
the crucial role of processes occurring at the moving inter-
face which is likely to be neither flat nor infinitesimally
thin. An alternative is to treat the interface as a thin film
where the migration of copper into silver can be consid-
ered as a general phenomenon of surface diffusion.
Probably the most relevant representation of the physical
situation in such a case was provided by Frauenfelder's ex-
periment in Ag-Cu alloys (quoted in Reference 12), where
the diffusion of radioactive copper-64 on the surface of
silver was studied at 750 ~ Frauenfelder's measurement
yielded a diffusion coefficient Of the order of 10 -6 cm2/s.
Unfortunately, the measurements involved a good deal of
scatter. Later, Pines et al., quoted in Reference 12 as well,
studied the kinetics of copper surface diffusion on thin films
of silver --~10 nm thick and this work suggested a surface
diffusion coefficient (at TB = 1093 K) of about 1 9 10 -6
cm2/s. Since the "effective diffusion coefficient" D e =
6.3 9 10 -7 c m 2 / s derived from the present experiment is of
the same order of magnitude, the S / L interface displace-
ment in stage two could be viewed as being controlled by
the copper surface diffusion in a-solid solution.
Afterward, the solid/liquid interface movement into the
Ag sides of the sandwich reverses direction as the liquid
zone shrinks.
3. Estimation of the maximum width of the liquid interlayer
According to the data given in Table IV, the initial width
(W0) and the final maximum width (Wmax) of the molten
zone, are, respectively, 79 and 420/zm. Wmaxcan be esti-
mated in principle by applying the mass conservation equa-
tion to the molten zone. In this rough approach, the change
in volume due to liquefaction can be neglected. The mass
balance relationship dictates that:
W o C B p B = (Wma x - W0). c L a p A Jr WoCLapB [13]
where PA and PB are the densities of A and B, respectively,
and C B and C L~ are weight fraction compositions. Rear-
ranging Eq. [13] and taking C B = 1, C L~ = 0.18, p~ =
8.9 g / c m 3, and PA = 10.5 g / c m 3 we obtain:
METALLURGICAL TRANSACTIONS A

C~ _ CL~ P._~]
Wm,x = W0" 1 + CL,~ " =4.9" W0 [14]
The corresponding experimental expression is Wm,x =
5.3 " W0- The agreement is remarkably good and confirms
that mass conservation is maintained in this stage of the
TLP process. This allows reasonable predictions of ex-
pected Wmaxin terms of the phase diagram, the respective
densities, and the width W0 of the inserted bonding layer.
C. Isothermal Solidification
During stage three, the liquid zone begins to recede and
continues shrinking until all liquid is used up. Cu atoms
(component B) steadily diffuse into the Ag-base material
(component A) at the S / L interface, converting it into com-
position C ~L. Since the shrinkage of the liquid zone is con-
trolled by the rate of solute (i.e., Cu) diffusion into the
Ag-based solid solution, the interdiffusion coefficient in
the solid D ~ becomes predominant. The larger the D O], the
faster should be the rate of solidification via the moving
S / L boundary. Relative to the preceding widening process
(stage 2), the solidification process (stage 3) is slow. On
the average, the kinetics of stage 3 are at least two orders
of magnitude slower than in stage 2. However, in addition
to the value of D~, the actual atomistic process occurring
at the interface may also play a major role.
To analyze the kinetics during stage 3 we define a dis-
placement variable Y3 in terms of the experimentally mea-
sured width W:
Y3 = (Wm,x - W ) / 2 [15]
Again, we define the time dependence of Y3 (and hence W)
through a general relation:
Y3 = a 3 " tm [16]
where a 3 and m are constants. The time, t, in this case is
actually an approximation for t - t 2, but since t3 is much
greater than t2, this simplification is justified. As before, a
logarithmic form of Eq. [15] provides a convenient way
for plotting the experimental data listed in Table V:
log Y3 = log a3 + m log t [17]
Values of log a 3 = - 4 . 5 (or a 3 = 3.2 • 10-5) and m =
0.495 (r = 0.99) are obtained from a straight line best fit
to the experimental data. As in stage 2, the interface dis-
placement again follows essentially a square-root law (with
m ~ 0.5):
Y3 = a3 " t~ [18]
where a 3 can now be expressed in terms of the interdiffu-
sion coefficient in the solid at the interface D O]and a gen-
eral proportionality constant K;:
a 3 = 2K~X/-D-~
~ [19]
With the experimental value of a 3 = 3.2 9 10-5 and DO] --
4.6 9 10 -m c m 2 / s , ~2 Eq. [19] yields a value of K~ = 0.75.
As shown in Table III, the evaluation by Lesoult of a con-
stant K3 of the respective Eq. [7] for the Ag-Cu system at
1093 K suggested a value of about 0.28 which would make
the movement of the S / L interface much slower than ob-
served experimentally. One might be tempted to compen-
METALLURGICAL TRANSACTIONS A
sate for the difference between K 3 and K; by adopting a
similar approach as in stage 2, i.e., by introducing an
effective diffusion coefficient for stage 3 that would be a
suitable mean between DO] and DL. However, unlike in
stage 2, it would be difficult to postulate a physical basis
for involving the diffusion coefficient in the liquid in the
rate controlling process of solidification during stage 3. At
this stage, DL is likely only to be involved in the homoge-
nization of the shrinking solute-rich liquid zone.
A more likely explanation for the differences between
K; and K3 may be found by considering the possible mech-
anisms of the interface migration. Unlike in the process
considered by Lesoult, involving a diffusion controlled
movement of a planar interface, solidification process is
almost certain to involve a ledge-type migration mecha-
nism. 13 Here, the diffusional mobility of the interface
would be controlled by a two-dimensional nucleation and
growth.of ledges involving a cooperative movement of
groups of atoms. Such a process is likely to speed up the
advancement of the solid interface into the liquid and thus
reduce the time t 3 needed for complete solidification.
An additional factor likely to accelerate growth during
stage 3 is the polycrystalline nature of the base metal. The
liquid interlayer intrudes into the grain boundaries (as in
Figure 7) and thus p r o v i d e s - - a s in the case of l e d g e s - -
an extra surface area for diffusion.
1. Estimation of time for complete solidification
To a production engineer interested in TLP bonding, one
of the most important considerations would be the time for
complete solidification of the bond region, t 3. To estimate
this time we assume that a pool of liquid of initial width
Wmax/2 is shrinking gradually to zero over time t 3 and we
apply the mass conservation principle which was found
earlier to be valid for stage 2. According to Sekerka, 5 the
mass balance at any time during stage 3, under the assump-
tion that PA = PS, is governed by the relationship:
W ' C L~ = Wmax'C L~ + 2
f0; D o] dt [20]
where W is the width at time t, and W 9 C L~ and Wm.x '
C L~ represent the actual and the initial mass values of the
interlayer, respectively; the integral represents the leak of
atoms across one of the interfaces due to diffusion of Cu
into Ag governed bY2the diffusion constant D~ (we recall
D, = 4.6 9 10-1~ c m / s ) .
To evaluate the integral in Eq. [20] we recall the solu-
tion of the Fick's equation [5, 8]:
C(y,t) = C ~162 +erf(y/4X/~t)] [21]
Integration over time yields [5]:
fot ~ dc / 4 D st\ o 5
D~yy = - - c ~ L ~ - ~ ~) [22]
and hence:
2 9 C ~z/'4D"_t\ ~
W = Wmax C c" ~ z _ ~ ) [23]
To estimate t3, w e set W = 0 at t --- t3:
2 {cL~ 2
t3 = Wma x " [24]
\C~,] 16D~
VOLUME 19A, MARCH 1988--683
 Wmaxin terms of W0 was estimated earlier (Eq. [3] or [13]).
If we assume that PA PB' Eq. [13] is simplified to:
= a function of temperature T. Then, by choosing T = TB ,
CB
Wmax : Wo.fL a [25]
Combining Eqs. [24] and [25] gives:
t3 = 16D7" \ C ~ ] [261
For the present case of the A g / C u / A g sandwich with
the Cu interlayer (Wo = 79 9 1 0 - 4 c m ) , C B and C ~ equal
1 and 0.08, respectively. A calculation of the time t 3 using
Eq. [26] yields (with Ds~ = 4.6 • 10-1~ cm2/s) a strongly
overestimated value of t 3 ~ 1200 hours. The actual time
t 3 expected from the extrapolation of the curve in Figure 7
is some 200 hours. Thus we must emphasize here that t 3
dictated by mass conservation principle (Eq. [26]) and in-
dependently estimated from diffusion considerations is in
both cases overestimated several times. This is an obvious
consequence of the very complex conditions prevailing at
the moving S / L interface as discussed above.
Due to the nonplanar nature of the bond region and the
grain boundary trapping of liquid it is difficult to define a
definite time for complete solidification in all areas of the
sandwich. While some areas may still have liquid in the
joint region, other regions will already have solidified so
that homogenization may already be well underway in these
latter regions.
2. The combined effect o f the bonding temperature TB
and the solubility o f solute B in base metal A on the time
f o r complete solidification 03)
As shown previously, for an interlayer of initial width
W0 and initial concentration C B used to join two pieces of a
base metal, the time required for complete solidification,
t3, is given approximately by Eq. [26].
The width W0 and the concentration C s are constants so
that we can rewrite Eq. [26] as:
t3 = T / D ~ " (caL) 2 [27]
where y is constant. The temperature dependence of t 3 can
now be introduced via the Arrhenius relation: D~ = Do 9
e x p ( - Q / R T ) . The expression for t3 thus becomes:
Y [exp(Q/RT)]
t3 = ~00" t (coL)2 j [28]
If the composition dependence of Do is neglected, Eq. [28]
can be reduced further to:
=
L
I.exp(Q/Rr)[/ t29]
where Y' is a constant.
The above expression is in a convenient form which re-
lates t 3 to the solubility of the solute in the base metal (C ~
at Ts) and the bonding temperature TB. Qualitatively, as
we increase Ta, t3 will decrease via the exponential term.
However, an increase in T8 simultaneously reduces the C ~
(solidus composition in a eutectic-type phase diagram) and
hence will tend to raise t 3. This interplay between the two
opposing terms offers the possibility of minimizing t3 by
an appropriate selection of the bonding temperature, as
well as the inserted interlayer composition and thickness.
684--VOLUME 19A, MARCH 1988
In order to consider this, it is necessary to express C ~ as
we effectively select a certain value of C ~z from the phase
diagram. *
*In peritectic systems, as we increase Ts, both the exponential term and
C ~ (see Eq. [29]) will tend to reduce t3; thus t3 cannot be minimized.
In numerous eutectic systems the solidus line may be
represented to a first approximation by a straight line. We
consider a phase diagram, where ire is the eutectic temper-
ature, TM is the melting temperature of the base metal (A)
and C ~z is the solubility of B in A at TE. For a given sys-
tem irE, TM, C ~e are fixed. The corresponding straight line
solidus relationship can be written as:
C~ = C~ (TM -- T)/(TM - TE) [301
which can be written in the form:
C ~z = C - nT [31]
where:
n = C~ - T~), [32a]
and
C = C~ TMI(TM - TE). [32b]
To simplify the calculation we substitute Eq. [31] into
Eq. [29] obtaining:
t3 = y ' " [ e x p ( Q / R T ) / ( C - nT) 2] [33]
or in logarithmic form,
In t3 = In y ' + ( Q / R T ) - 2 ln(C - nT) = 7. [34]
The resulting parameter z is dimensionless, and can be
designated as a new variable. It represents now the time
for complete solidification.
By taking the first derivative of z with respect to T and
equating it to zero a quadratic equation in terms of T is
generated:
2nRT 2 -4- QnT - QC = 0. [35]
From Eq. [35], the two solutions for T are given by
1
T = [-Qn +- {(Qn) 2 + (8QCRn)} '/2] "4R---n" [36]
The positive term gives a temperature T corresponding to
a minimum of z (the minimum in t 3 ) . Tmin is therefore
determined by the activation enthalpy for diffusion and,
via C and n (Eq. [32]), the melting temperature of the
base metal.
Plots of z vs T for various binary eutectic systems (in re-
gions of selected phase diagrams) are illustrated in Fig-
ure 11. The curves give a qualitative representation of the
effect of the bond temperature on the time for complete
solidification. While the Ag-Cu system involves a local
minimum near the eutectic temperature, other systems may
have a minimum at much higher temperatures. It must be
emphasized, however, that the numerical magnitude of the
z-axis is not drawn to scale since the curves have been
shifted to make it possible to show all plots on the same
diagram. Nevertheless, the slope of these curves is qualita-
tively related to the effect on t 3 of changing the bonding
METALLURGICAL TRANSACTIONS A
 EFFECT OF SOLUBILITY (CaL)& BOND TEMP, ON TIME FOR SOLIDIFICATION
Pb-Sn (26.2)\
\ ilJ / / / ,/
c~
~-~-~r'~ d/' ""~ ""-~ . . . . - 4.0
FAI-Cu (52.6)Ja/t "r =ln B-2.~n [C eE f T . - T "~]'-4- ~
ra-O. ta..EV.CY" L ~ TM-TE./j
0
I I
50
I
JO0 150
8 = (T-TE)
I I
200
r
250
* r' x 0.5
Fig. l l - - E f f e c t of bonding temperature (and hence C ~) on total time
for solidification for different eutectic systems (for details please see
the text).
temperature. While t 3 is very temperature sensitive in the
Pb-Sn system, this effect is small in the Fe-C system.
3. The use of an Au-20 wt pct Cu alloy as the interlayer
material
The TLP experiment discussed above involves the pure
metals A (Ag) and B (Cu) which form a suitable eutectic
system. While the Ag/Cu/Ag offers a simple and suitable
joint for an investigation of the TLP bonding process, the
use of a pure material as the interlayer material makes
the process quite prolonged (e.g., the time for complete
s o l i d i f i c a t i o n , t3, the important parameter for engineer-
ing applications, is some 200 hours at 1093 K). To obtain
a significantly shorter process, the TLP bond sandwich
should involve interlayer material in the form of an alloy
close to eutectic composition. To verify the gain in overall
bonding time resulting in this case, samples were made
using a 75 /xm thick Ag-20 wt pct Cu foil. For compari-
son with the "pure Cu" situation, samples were bonded at
1093 K for 8 and 15 hours.
As expected, already after 8 hours the bond region was
solidified, except for some liquid trapped at grain bound-
aries. Such regions show a concentration peak of about
9 wt pct Cu (as in Figure 12). In the Ag-Cu phase diagram
this composition lies in a two-phase (a-Ag + Liq) region.
After 15 hours the bond region was completely solidified.
The highest concentration was about 6.4 wt pct Cu near
the initial center of the interlayer foil. At the bond tempera-
ture of 1093 K, the expected solidus composition, C ~, is
about 7 wt pct Cu. Thus, after 15 hours solidification is
not only completed, but much homogenization of the bond
region has already occurred. This confirms the expectation
that choosing an appropriate alloy for the interlayer mate-
rial will shorten the TLP process.
METALLURGICAL TRANSACTIONS A
IO.O
8.0
Q.
(D
6.0
0
D..
..r
RT 15 Hours
2.0
1 t I I I I I
-200 - I00 0 I00 200 275
Distance From The Symmetry Axis (/_tm)
Fig. 12--Concentration profiles of Cu in the resolidified interlayer after
prior annealing of the Ag/Ag-20 wt pct Cu/Ag sandwich for 8 h and
15 h at 820 ~
V. SUMMARY AND CONCLUSIONS
1. A Ag/Cu/Ag sandwich joint associated with a simple
eutectic phase diagram can be used as a suitable sample
to test the TLP bonding process.
2. The present study has confirmed earlier theoretical
modeling which treats the TLP bonding as a diffusional
process involving four distinctive stages: (i) dissolution of
the interlayer, (ii) homogenization of the liquid, (iii) iso-
thermal solidification, and (iv) homogenization of the
bond region.
3. The first stage of TLP bonding occurs almost instanta-
neously and can be accounted for by diffusion that occurs
mainly in the liquid near the interlayer interface.
4. In the second stage, the kinetic behavior of the L / S in-
terface is influenced by processes occurring in both the
liquid and the solid. An effective diffusion coefficient
can be derived from experimental data; it can either be
expressed as a "weighted" geometric mean of appropri-
ate DL and Ds coefficients, or can be viewed as being
controlled by the surface diffusion of copper in the silver-
rich a-solid solution (on the Cu side).
5. In the third stage, diffusion appears to be controlled
mainly by Ds. The experimental time for complete so-
lidification has been found to be shorter than the time
predicted and this is at least in part due to the fact that
the theoretical considerations are based on the assump-
tion of a moving planar liquid/solid interface, whereas
the actual solidification involves curved interfaces and
perhaps also a ledge-type migration mechanism.
6. A modeling approach is proposed which allows, for a
given TLP process and the associated phase diagram, to
VOLUME 19A, MARCH 1988--685
 predict an optimal bonding temperature, corresponding
to the shortest time for complete solidification. The
model suggests that some eutectic-type systems are
more suitable for TLP bonding than others.
7. A clear experimental confirmation has been obtained
showing that the use of an alloy close to eutectic compo-
sition as an interlayer material shortens the TLP process.
ACKNOWLEDGMENTS
The authors are particularly grateful to Dr. G. Lesoult
for many stimulating discussions and provision of unpub-
lished notes and calculations several years ago, which have
led to their interest in the experimental work and model-
ing described in this paper. They also wish to thank Profes-
sor R. E Sekerka for additional discussions of his interests
and contributions to this subject. The experimental work was
supported in part by a grant from the National Science
Foundation (DMR 81-06234). Support from the Center for
the Study of Materials, through the Materials Research Lab-
oratory Section, National Science Foundation, under Grant
No. DMR-8521805, is also gratefully acknowledged.
686--VOLUME 19A, MARCH 1988
REFERENCES
1. D.S. Duvall, W. A. Owczarski, and D. F. Paulonis: WeldingJournal,
1974, vol. 53, pp. 203-14.
2. J.T. Niemann and R. A. Garrett: Welding Journal, Research Suppl.,
1974, vol. 53, pp. 175-84.
3. R.R. Wells: Welding Journal, Research Suppl., 1976, vol. 55,
pp. 20-27.
4. G. Lesoult: Center for Joining of Materials Report, Carnegie Mellon
University, Pittsburgh, PA, Sept. 1976.
5. R.F. Sekerka: Carnegie Mellon University, Pittsburgh, PA, un-
published research, 1979.
6. R.F. Sekerka, C.L. Jeanfils, and R.W. Heckel, Technical Report
No. 3, Carnegie Mellon University, Pittsburgh, PA, July 1975.
7. Binary Alloy Phase Diagrams, T.B. Massalski, ed., ASM, Metals
Park, OH, 1986.
8. P.G. Shewmon: Diffusion in Solids, McGraw-Hill Book Co., New
York, NY, 1963.
9. T. Yamamura and T. Ejima: J. Japan Inst. of Metals, 1973, vol. 37,
pp. 901-07.
10. W. Volk: Applied Statisticsfor Engineers, McGraw-Hill Book Co.,
New York, NY, 1969.
11. M.G. Hall and C.W. Haworth: Trans. A1ME, 1969, vol. 245,
pp. 2476-79.
12. D.B. Butrymowicz, J.R. Manning, and M.E. Read: J. Phys.
Chem. Ref. Data, 1974, vol. 3, pp. 527-602.
13. D.P. Woodruff: The Solid-Liquid Interface, Cambridge University
Press, Cambridge, 1973.
METALLURGICAL TRANSACTIONS A