W. T.

Chen
C. W. Nelson

Thermal Stressin Bonded Joints

This paper considers the stress distributions in bonded materials induced by differentialexpansion or contraction of these
materials. The analytical approach is similar to the lap joint theories attributed to Volkersen and expanded by Goland
and Reissner. Several simple and typical analytical models are presented to bring out the relative importance of dlrerent
geometrical and material parameters and to give some insight into different modes in which the bonds might fail.

Introduction
The reliability of any electronic device depends toa great behind this paper is to keep the treatment simple. Then,
extent on its construction. The structuraldesign and fab- any or all parameters can be varied through a reasonable
rication dictate the amount of power dissipation, the num- range easily, and considerable insight can be obtained as
ber of thermal cycles, and the intensity of mechanical im- to the effects of various physical parameters on the ther-
pact and vibration which the device can sustain without mal stresses.
degrading its performance.
This paper presents several simple and typical analyti-
Since electronic devices are characterized by hetero- cal models to bring out the relativeimportance of dif-
geneous materials joined together by different methods, ferent geometrical and material parameters. It also gives
including adhesives and solder, a key consideration in the some insight into different modes in which thebonds
packaging of the devices is that the bonds between the might fail.
different materials are capable of sustaining the mechani-
cal and thermal stresses over the service life of the de- While one usually associates thermal stress with the
vice. A general discussion of the commonly used materi- stresses that arise from high temperature service condi-
als and their fabrication and assembly processes may be tions on heterogeneous materials with dissimilar thermal
found in [ 11. expansion coefficients, suchstresses canalsobe in-
troduced during the fabrication processes. One such ex-
The reliability of electronic devices and their packaging ample is the internal stress condition produced in the sili-
considerations are discussed in many technical journals con oxide film during the etching, diffusion, and metal-
and monographs. Some recentreview papers are found in lization processes. Many films show “intrinsic” stresses
[2]. This paper considers the stresses induced in bonded which greatly exceed the calculated values for thermal
materials by differential expansion orcontraction of these stresses [3].
materials. A number of simplifying assumptions made in
the present treatment are described later in this paper. Taylor and Yuan [4] studied fracture failure in some
These are believed to be good approximations leading to semiconductordevices, anddeveloped
mathematical
valid results provided that the elastic modulus of the ad- analysis for thermal stress concentrations in the silicon
herend is about ten (or more) times that of the adhesive bond area to explain the observed fracture. Theanalytical
and provided that thethickness ratio of adherend to adhe- approach adoptedby Taylor andYuan has its roots in the
sive also is about ten (or more). The guiding philosophy lap joint theory commonly attributed to Volkersen [ 5 ] .

Copyright 1979 by International Business Machines Corporation. Copying is permitted without payment of royalty provided that (1)
each reproduction is done without alteration and ( 2 ) the Journal reference and IBM copyright notice are included on the first page.
The title and abstract may be used without further permission in computer-based and other information-service systems. Permission
to republish other excerpts should be obtained from the Editor. 179

IBM J. RES. DEVELOP. VOL. 23 NO. 2 MARCH 1979 W. T. CHEN AND C . W. NELSON
 In that theory the joint (adhesive, solder) was treated as centration in shear and tensionoccurs simultaneously,
a distribution of shear springs similar to the well-known and debonding would be mixed mode in character.
Winkler foundation theory. The joined materials (adher-
ends) were treated as beams in tension. The Volkersen Analysis
lap joint theory has been further expanded and improved
in the now classic paper by Golandand Reissner [6]. Three elastic layers withtwo bondedjoints (adhesive,
With the widespread use of adhesives today, it is not sur- solder)
prising that many of the later developments aremotivated Theassumptions inherent in the Volkerson'slap joint
by calculation of stress in adhesive bonds. Fortunately, analysis are adopted here with the addition of thermal
the mathematical analyses are usually applicable to other consideration. The three layersbeing bonded together are
bond materials, given the assumption that the jointmate- assumed to be of uniform thicknesses (t,, t,, t3)and with
rial (adhesive) is less stiff than the joinedmaterials (adher- elastic moduli ( E , , E,, E3), and thermal expansion coeffi-
end). cients (a,, a,, a,). The joint between layers 1 and 2 will
have thickness 7) and shear modulus G. The joint be-
Further development of the adhesive joint analysis may tween layers l and 3 will havethickness 7' andshear
be found in the survey articles by Benson [7] and in many modulus G'.
monographs [8-121. There have been many publications
on thesubject since the work of Goland and Reissner, and As shown in Fig. 1 , the forces in the elastic layers are
an exhaustive bibliography would be outside the scopeof assumed tobe uniform through the thickness (F,, F,, F J .
this paper. Some recent contributions may be found in The bonded joints exert shear stress r between layers 1
[13, 141. The present paper expands from the work of and 2, and shear stress 7' between layers 1 and 3.
Taylor and Yuan to otherphysical conditions and geome-
tries. At some temperature To the joints are stress free. What
will be the shear stresses T and r' and forces F , , F,, F,
Process engineers and adhesive scientistsoften employ with a temperature change of T?
peel tests to assess the strengthof the bond. To relate the
measured peel force to themaximum stresses in the peel- Considering the equilibrium of forces in the x direction
ing process,the peel adhesion theories developed by (see Fig. I), the following three equations are derived:
Spies [15], Bikerman [16, 171, and Kaelble [18, 191 are
used. These theories are also founded upon the Winkler -dF1
+r"r=O.
type foundation theory for the jointmaterial. Sometimes dx
a peel test may be usedfor obtaining a bond strength to be
used as a criterion of failure in the thermal stress calcu-
3+ r = o ,
dx
lations of the present paper. However, if this is done, care
should be taken to ensure that the basic assumptions used 3 -7
,
=o.
in the theoryfor obtaining bond strength froma measured dx
peel force are compatible with the assumptions used for The stress-strain-temperature equations of the three lay-
computing the bond stress in the actual device by the ers are
methods of this paper.

The first case considered in this paper is three elastic

layers bonded together at two interfaces. The layers are
assumed to remain straight as in Volkersen's theory. The
two-layer elastic solution is also given and is found to re-
duce to Taylor and Yuan's result when one layer is as- -F 3
du -
3 + a,T
sumed to berigid. The analyses forcircular plates bonded dx Est3
together are also studied.
And forthe bonded joint materialsunder shear,the
It is shown that in the limit of radius-to-thickness ratio stress-strain relationships are
approaching infinity, the previous solution can be recov-
ered. Finally, the two elastic layers are allowed to bend
freely, such asin flexible printed circuits. It is shown that
the analysis follows naturally from Goland and Reissner's
180 work. It is particularly interesting to find that stress con-

W. T. CHEN AND C. W. NELSON IBM J . RES. DEVELOP. VOL. 23 NO. 2 MARCH 1979
The solutions of the previous set of eight equations are
straightforward. The general solution for 7 and 7' is found
to be
r = A , sinh p,x + A , sinh &x + A , cosh p,x
+ A , cosh &x,
r' = A,k, sinh p,x + A,k, sinh p,x + A3k, cosh p , x
F " Y
+ A4k4cosh &x. (4)
"

Here * p , and *& are roots of the equation

-
and the constants A i , ki (i = 1, 4) are to be determined by
boundary conditions. Let us impose the boundary condi-
tions that the bond areas are completely filled up as in
Fig. I , andthattheends x = arefree.Four of the
F, -I 1- -T
F , +;\F,

-r -1
constants are found to be zero:

Write the shear stresses as

F, - (b)
F,+AF2

sinh p,x sinh p,x

r=C1-
cosh p,e
+ c, cosh p,e '
~

Figure 1 (a) Notations on three joined layers.(b) Force balance

on three joined layers for a section dx.
sinh p,x
ZZ
+ C,k,
cosh p,f '
Thesearetheexpressionsforthe maximum shear
stresses at the edge of the two bonded joints. Discussion
of numerical exampleswill be reserved fora later section.

Two elastic layers joined by one bonded joint

The analysis follows the same steps asbefore. In the case
of a completely filled joint, the shear stress is given by
and D l , D,, D, are constants defined by ( a , - a,)TG sinh px
r=
pq cosh
- a,) + E2f2(a2- 5 ) l T
E1f1[E3t3(a3
Dl =
Eltl + E,', + E3f3 where

Physically the shear stress is zero at the center, and in-

creases gradually to a maximum at the free edge. The
value of this maximum stress is
It is clear, merely from the general form of the equations
(6) and (7), that the shear stressesr and r' vary from zero
at the middle symmetric point to a maximum at the edge
x = e. When the layer's length-to-joint thickness ratio is
very large, the stresses at the edge approach
Often it may be sufficient to take tanh p f - I and use the
estimate
r(x = e) = c, + c,,
r'(x = f ) = C,k, + C,k,. (11) 181

IBM J. RES. DEVELOP. VOL. 23 NO. 2 MARCH 1979 W . T. CHEN AND C. W. NELSON
 I I ,

The maximum stress occurs at r = R , and is equal to

/=-x + q ,
Two elastic layers with one joint allowing free flexure
When two layers expand unequally, but are bonded to-
gether, there is the natural tendency for the composite to
bend. This is the basic theory behind a bimetallic thermo-
stat analyzed by S. Timoshenko [20] many yearsago.
Timoshenko assumedthat thetwolayers behave like
beams capable of axial and bending deformations, and
that there is no slip at the interface between the two lay-
ers. In this section we consider that the two layers are
separated by an amount 77 and are filled with a material
capable of deforming under shear and tension. This is es-
Figure 2 Two circular sheets bonded together. sentially the approach taken by Goland and Reissner [6],
who dealt with externally applied forces rather than the
internally generated thermal stress.

Figure 3 shows the forces and moments acting on an

If p f couldbesmall,then the shear stress would ap- elementary section of the bonded composite. Equilibrium
proach the usual simplified estimate of of moments requires that
(a1- a p c
r= (16)
77
since (tanh p f ) / p e would approach unity. However,with
physically realistic parameter, /3f is never small. There-
fore theusual simplified estimate [Eq.(16)] is often a poor
Equilibrium of horizontal forces requires that
approximation.

Two circular sheets bonded together

The foregoing analysis has been one-dimensional. It was
considered useful to study the axisymmetric situation of
two circular sheets bonded together as shown in Fig. 2.
The analytical development is given in Appendix A. As And for equilibrium of vertical forces
may be expected the solution is in the form of modified
Bessel functions. In the case where the bond area be-
tween the two sheetsis completely filled, the shear stress
is zero at the center of the circle, and gradually rises to
the free edge with increasing r according to the formula
dV2
__ + uo = 0.
182 dX

W. T. CHEN AND C. W. NELSON IBM 1. RES.DEVELOP. VOL. 23 NO. 2 MARCH 1979

 I li
L P L P J
Figure 3 (a) Notations on joined layers. (b) Force and moment diagrams on a section dx. ( c ) Positive sense of displacement com-
ponents.

The next step is to relate moments and forces to dis- where the constants a, b , and c are defined as
placements. From elementary bending theories,

Eit:
Di= (33)
12(1 - 7 : ) '
and The solution of Eq. (30) is related to the roots of the alge-
braic equation

It can be shown that the roots to the above equation al-

Finally, the stress in the joint material is assumed to de- wayscontain one pair of complex conjugates and one
pend on the displacements (ul, u,) and (",, u,), according positive real root. In other words the solution must be in
to the equations the form

u,,= A , cosh p,x + A , sinh p1x + A , cosh P,x cos &x

+ A , sinh p,x cos &x
+ A , sinh pHxsin &x + A , cosh P,x sin &x. (34)

Here G, and E, are shear modulus and Young's modulus In general, there are six boundary conditions from which
of the joint material. to determine A , to A,. But symmetry about the plane
x = 0 would reduce the solution to
Now the stress analysis problem is fully formulated. u,,= A , cosh /3,x + A , cosh pHxcos &x
With the appropriate boundary conditions, the analysis is
complete. An inventory of Eqs. (21) to (29) discloses that + A , sinh pHx sin &x. (35)
there are twelve such equations. The above set of equa- The constants A , , A , , and A , are completely determined
tions can be reduced to a single sixth-order differential by the boundary conditions that at the edge x = e, mo-
equation for u,,.A solution of the differential equation can ments, horizontal forces, and shear forces vanish. Also
be found containing six constants of integration permit- the shear stress is given by
ting the six boundaryconditions tobe satisfied.This
sixth-order differential equation is T, = C, sinh p,x + C, sinh Px, cos pvx
d6u, G,c d4u, +E bOd2uL G,E,(bc
-
- u')u, - + C, cosh p,x sin &x. (36)
- 0,
dx6 r) dx4 r) dx' 77' Details of expressions for A , , A , , A,, and C,, C,, C, are
(30) given in Appendix B. 183

IBM J. RES. DEVELOP. VOL. 23 NO. 2 MARCH 1979 W. T. CHEN AND C. W. NELSON
 Numerical examples and discussion

Two layers without bending

Numerical examples are shown for thefollowing material
and geometrical parameters for a 100°C rise in temper-
ature:

Young'smodulus E 1 .I7 X 10"Pa 2.75 x 10" Pa

Thickness t mm 1.52
1.57 mm
Thickness
exp. coef.
1.6 x 10-5/0C 6.5 x 10-6/0C
Shear modulus

lo
5 10 15 20 25
of joint material 1.23 x IO9 Pa
Distance from center (mm)
Joint thickness 0.051 mm
Figure 4 Shear stress distribution over width of joint.
Joint width mm 51

Figure 4 shows the shear stress distribution in the joint.

As may be expected, themaximum shear stress occurs at
the edge and is found to be 5.3 X lo7 Pa. The associated
shear strain is 0.043. Often one assumes the shear modu-
lus of the joint material to be very soft compared to the
joined material; and the strain and stresscalculated under
that assumption are 0.475 and 5.85 X IO8 Pa, more than
ten times larger than the elastic solution. Also, Eq. (15)
gives a very accurate estimate (0.5% low) of the actual
shear stress. Figure 5 shows the tensile force in the layer
and how it drops to zero near the edgeof the joint. This
calculation illustrates that, in this situation, about 60% of
the joint material in the center is not really structurally

lo
5 10 15 20 25
effective. Only the section from 15mm outward is ef-
fectively stressed.
Distance from center (mm)

Figure 5 Force induced by thermal expansion. Figure 6 shows the different maximum shear stresses
where the joint thicknesses are taken to be 0.051, 0.076,
0.127, and 0.178 mm. It illustrates that while the thick-
Y" I
ness may change from 0.051mm to 0.178 mm,an in-
crease of 3.5 times, the shear stress decreases by a factor
of 1.8 only. While this may be expected from examination
of Eqs. (13), (14), and (15), it is not immediately obvious
from intuition.

Three layers without bending

It is difficult to draw anygeneralconclusionfrom the
three-layer case because so many physical parameters,
thicknesses, expansioncoefficients,andelastic moduli
could all be varying. Consider the previous case of two
layers illustrated in Figs. 4 and 5. Let us put another layer

/o
5 10 15 20 25 of material in between with modulus E = 2.62 X 10" Pa
and thickness 0.51 mm. The other joint thickness and
Distancefromcenter (mm) modulus remain the same as before. The thermal expan-
Figure 6 Shear stress distribution over width of joint for four sion coefficient is taken as 3 X 10-6/oC, lessthan the other
184 thicknesses.
joint two coefficients. The shear stressdistributions are shown

W. T. CHEN AND C. W. NELSON IBM J. RES. DEVELOP. VOL. 23 NO. 2 MARCH 1979
I 60

- 7 = 0.05 mm

-. -.- 9=0.13mm
40 MPa -

lo
5 IO 15 20 25

Distance fromcenter ( m m )

Figure 7 Shear stress distributions over width of joints.

in Fig. 7. Thesolid line may be compared toFig. 4, show-

ing that thereis a slight increase in maximum shear stress.
The dotted line shows the shear stress distribution in the
joint between the two materials with expansion coeffi-
cients 3.0 x and
6.5 X The maximum stress is
not at the edge. This example illustrates that when there listance from center (x/P)
are three layers, simple physical intuitive ideas may be Figure 8 Shear stress distribution in joint with flexure.
useful but not always applicable.

Two layers allowing free flexure

For illustrative purposes numerical results are presented
for the case of different layers allowing free flexure for the
same material properties used in the previous two-layer
case without bending.

Figures 8 and 9 show the tensile and shear stressdistri-

butions in plane strain for two different thicknesses. The

"1
I
jointmaterial is given by E = 3.45 x IO9, G = 1.23 X I
lo9 Pa. Note that the thicker joint produces the lower
stress. Particularly of note is the characteristic that the
stresses remain almost zero for most of the joint length,
except in the vicinity of the edge, where the stress rises -6

dramatically. Also, the tensile stress changes from nega-

tive to positive a small distance in from the edge; but the
shear stress doesnot change sign. This is a common char-
acteristic of this problem. Finally, it is noteworthy that
the maximum shear stresses for the two cases are 4 X lo7
and 2.56 X IO7 Pa, respectively. The corresponding shear
stresses for plane stress are 2.75 X lo7 and 1.75 X lo7 Pa,
respectively. A simple stressestimate by using differ-
ential thermal expansion dividedby joint thickness would
give a gross overestimate of 5.85 X 10' and 2.34 X IO* Pa, IDistance from center( x h )

respectively. Figure 9 Tensile stress distribution in joint with flexure. 185

IBM 1. RES.DEVELOP. 0 VOL. 23 NO. 2 MARCH 1979 W. T. CHEN AND C. W. NELSON

 well. In other words, if one is only interested in moment
and deflection but not the joint stresses, Timoshenko's
analysis is still applicable.

Summary
This paper provides some insight and tools to understand
the stress distribution in a bonded joint induced by ther-
mal expansion of dissimilar materials. The maximum
shear stress always occurs at the edgeof the joint. When
flexure is allowed, significant tensile stress may occur.
However, the flexure does decrease the maximum shear
I - - - -- Tensile stress uo stress.

It is difficult to draw conclusions from the three-layer

case, since many more physical parameter variables are
possible. The analysis still indicates that the maximum
shear stress in either of the two joints usually occurs at
the edges, and that there is an interaction from one joint
to another if the middle layer is reasonably thin.

Appendix A: Analysis of two bonded circular sheets

The physical parameters and geometrical dimensions are
shown in Fig. 3. The polar coordinates ( r , 0) are em-
ployed. Thestress-strain-temperature relationships of
layers 1 and 2 are

0.5 1.o The assumptionof uniform shear stressthrough the thick-

ness of the joint gives the relationship
Layer thickness,f 2 (mm)
7
"
u - u
-U
Figure 10 Effect of variation of layer thickness t,. G r)

The equilibrium conditions in each sheet give

Figure 10 shows the effects of varying the thickness of

layer 2 from 0.76 to 1.52 mm. The tensile stress at the free and
edge changes from positive to negative, while the shear
stress increases slightly. If the joint material is weak in - + r t, = -rr
tension, a condition often referred to aspeel stress, keep-
ing this stress component negative may be desirable. The thermal stressproblem is formulated by Eqs. ( A l )to
(A3) with the proper boundary conditions. The general
The maximum shearstresscomputed from Eq. (14) solution for the shear stress to the above equations is
turns out to be almost one half the value for this case found in terms of the modified Bessel functions,
(plane stress). This has been found characteristically so 7 = A[I,(Pr)+ BK,(Pr)l* (A41
for a number of other cases. It is concluded that free
where A and B are constants, and
bending relaxes shear stresses at the edge.

A calculation for the forces and moments in the two

layers in these numerical examples checks with the re- The stresses and displacements in layer 1 are given by
sults given by Timoshenko's well-known bimetallic ther-
mostat solution. This indicates that for the physical pa-
rameters used in this paperthe numerical resultsfor
186 forces and moments agree with this analysis extremely

W . T. CHEN AND C. W. NELSON IBM 3. RES. DEVELOP. VOL. 23 NO. 2 MARCH 1979
The expressions for corresponding quantities for layer 2
+ pH
(Pi +
sinh p,ecosp,f
1 A,

may be found by changing the corresponding indices from

1 to 2 and adding a negative sign to oneside of each equa-
tion. The constants C,, C,, D l , and D, are related by the
equations

Dl + D, = 0, (A81

2(a, - a,)TG
A ( C , + C,)= -
”P
There are fourboundary conditions, which, together with
the above two, are
sufficient to solve for
all six constants.

In the caseof a solid disk with radiusR , three of the six

constants are zero:

B = Dl = D, = 0. (A101

The condition thatat radius R the radial stresscom-

ponents are zero leads to

L j (4)
c, = (Eob a [Y,A, - Y,A,I,

The shear stress is given by

From the properties of modified Bessel function Zn,it is

clear that the shear stress reaches
a maximum at the outer
edge and is zero at the center.

Appendix B References
In the main text it has been explained that the six roots of 1 . Physical Design of Electronic Systems, Vol. 11, Materials-
Technology, Vol. 111, IntegratedDeviceandConnection
the characteristicsof Eq. (30) can be written in the forms Technology, prepared by staff of Bell Telephone Lab., Inc.,
Prentice-Hall, Inc., Englewood Cliffs, NJ, 1971.
pH + ipV, p
H - ipV3 - p H + iPv, -pH - @,. 2. Reliability of Semiconductor Devices, Proc.IEEE 62, No. 2,
1974.
The three constants A , , A , , and A , are determined from 3. E. KlokholmandB. S. Berry, “Intrinsic Stress in Evapo-
ratedMetal Films,” J . Electrochem.SOC. 115, 823-826
the following set of algebraic equations: (1968).
4. T. C. Taylor and F. L. Yuan, “Thermal Stress and Fracture
cosh p l f A l in Shear Constrained
Semiconductor Structure,” IRE
Trans. Electron. Devices ED9, 303-308, (1963).
+ [(pi - p:) cosh pH[ cos p,e 5 . 0.Volkersen, Lufifahrtforschung 15, 41 (1938).
6. M. Golandand E. Reissner,“TheStresses inCemented
- 2pHp, sinh p,e sin p,e]A, Joints,” J . Appl. Mech., Trans. ASME 11, 17-27 (1944).
7. N. K . Benson,“TheMechanics of AdhesiveBonding,”
+ [ ( p i - BG) sinh p,e sin p,f A p p l . Mech. Rev., February 1961.
8. D. D. Eley, ed.,Adhesion-Fundamentals & Practice, Mac-
+ 2pH&cosh pH[ COS P , f ] A , = 0, (B 1) laren & Sons, London, 1969. 187

IBM 1. RES. DEVELOP. VOL. 23 NO. 2 MARCH 1979 W. T. CHEN AND C. W. NELSON
 9. D. Satas, ed.,Conference on Adhesion of Polymers, Wayne 17. J. J. Bikerman, The Science of Adhesive Joints, Academic
State University, Detroit, MI, 1967. Press, Inc., New York, 1961.
10. R. Houwink and G. Salomon, eds., Adhesion & Adhesives, 18. D. H. Kaelble, “Theoryand Analysis of Peel Adhesion;
Vols. I and 11, 2nd revised edition, Elsevier, Amsterdam, Bond Stresses and Distribution,” Trans. SOC. Rheology IV,
1%7. 45-73 (1960).
1 1 . R. Patric, ed., Treatise on Adhesion & Adhesives, Vols. 1 19. D. H. Kaelble, PhysicalChemistry of Adhesions, John
and 2, Marcel Dekker, New York, 1968. Wiley & Sons, Inc., New York, 1971.
12. Lieng-Huang Lee, Recent Advances in Adhesion, Palmer- 20. S. Timoshenko, “Analysis of Bi-metal Thermostats,” J.
ton, New York, 1976. Opt. SOC. Amer.11, 233-255 (1925).
13. L. M. Keer and K. Chantaramungkorn, “StressAnalysis for
a Double Lap Joint,” Trans. ASME, Series E,J . Appl. Mech. Received November 1, 1977; revised September 7 , 1978
42, 353 (1975).
14. W. J. Renton and J. R. Vinson, “Analysis of Adhesively W . T . Chen is located at the IBM System Products Di-
Bonded Joints between Panels of CompositeMaterials,” vision laboratory, P.O. Box 6 , Endicott,New York
Trans. ASME, SeriesE, J. Appl. Mech.44, 101-106 (1977).
15. G. J. Spies, J . Aircraft Engr. 25, 64 (1953). 13760, and C.W . Nelson, who is retired, is at 610 Bassett
16. J. J. Bikerman, J. Appl. Phys. 28, 1484 (1957). Avenue, Endicott, New York 13760.

188

W. T. CHEN AND C. W. NELSON IBM J. RES. DEVELOP. VOL. 23 NO. 2 MARCH 1979