You are on page 1of 15

MEASUREMENTS OF ADHESIVE BONDLINE EFFECTIVE THERMAL

CONDUCTIVITY AND THERMAL RESISTANCE USING THE LASER FLASH METHOD


Robert C. Campbell, Stephen E. Smith
Holometris
25 Wiggins Ave.
Bedford, MA 0 1730
e-mail: info@holometris.com

Raymond L. Dietz
Diemat, Inc.
458 Boston St., Suite 1
Topsfield, MA 0 1983
e-mail: dieniat@shore.net

ABSTRACT NOMENCLATURE

Thermal modeling of device packages requires thermal resistance (mm’WW)


accurate thermophysical property data for package temperature difference (K)
materials. Accurate data for the thermal resistance of heat flow (W)
the adhesive bondline used to attach a high power area (mm2)
device to a substrate is critical because this thermal bondline thickness (pm)
resistance can be a significant part of the total thermal thermal conductivity (W/m-K)
resistance in the heat flow path from the device effective thermal conductivity (W/m-K)
junction to the package case or ambient. in-plane thermal conductivity (W/m-K)
through-plane thermal
The bondline thermal resistance can in principle be conductivity (W/m-K)
calculated by dividing the espected or measured intrinsic thermal resistance (mm2K/W)
bondline thickness by the adhesive thermal contact thermal resistance (mm2K/W)
conductivity measured on a free-standing cured total bondline thermal resistance
sample. However, at a typical bondline thickness of (mm2K/W)
15-75pm, the contact thermal resistance bctween the die (substrate) to adhesive interface
adhesive and its adherents can be significant compared temperature difference (K)
to the intrinsic thermal resistance of the adhesive and thermal diffusivity (cm2/s)
thus cannot be ignored. Also, the thermal conductivity effective thermal diffusivity (cm2/s)
measured on a free-standing cured sample may not be in-plane thermal diffusivity (cm2/s)
equivalent to the thermal conductivity of the adhesive through-plane thermal diffusivity (cm2/s)
in the bonded assembly. bulk density (g/cm3)
specific heat capacity (J/g-K)
This paper will investigate some of the variables that half rise time (ms)
determine the adhesive bondline effective thermal
conductivity and contact resistance. The results of I. INTRODUCTION
multi-layer laser flash diffusivity measurements will be
presented for a range of available adhesives in The ability to accurately predict the device junction
“sandwich” sample assemblies that simulate the temperature has become critical with the increasing
package. Thermal conductivity measurements of the power and speed and decreasing size of semiconductor
adhesives free-standing will also be obtained by the devices. For high power devices, accurate knowledge
laser flash method. of the die attach thermal resistance can be critical.
The thermal conductivity provided by adhesive vendors

IEEE
0-7803-5624-5/99/$10.0001999 83 Fifteenth IEEE SEMl-THERMTMSyrnposiurn

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
is typically the result of a nieasurement on a thick, adhesive h can also be determined from the reciprocal
free-standing molded sample. In some cases this value of the slope of the line. If the adhesive h is known
may be used to calculate the intrinsic thermal from measurements of a free-standing sample then
resistance for a given bondline. With an estimate of
the contact resistance, the total resistance can be 2 ( L ) = hot-AS I h. (2)
estimated. However, the contact resistance for
adhesive bonds of various materials is not well The y-asis intercept method requires that the thermal
characterized. Also, depending on the type of adhesive conductivity is not a function of thickness. The use of
and test sample size and preparation method the the term thermal conductivity of course implies no
“bulk” sample and its thermal conductivity may not dependence on thickness but due to the nature of the
accurately represent the adhesive bondline die attach process and the drying and curing
microstructure. conditions, voids may reduce the thermal conductivity
for thicker bondlines, effecting the accuracy of fitting
11. BONDLINE THERMAL RESISTANCE AND the slope and intercept. The subtraction method has an
EFFECTIVE THERMAL CONDCUTLVITY advantage in that realistic bondline thicknesses can be
used, but requires an assumption that the adhesive
The total thermal resistance from the surface of the die microstructure and resulting thermal conductivity are
to the surface of the substrate will be the sum of the equivalent in the bondline and in the free-standing
contact resistances and the intrinsic adhesive resistance sample.
(see Figure 1). Thermal resistance is defined as
The magnitude of the contact resistance in the
R,,,= AT I (QIA). assembly is espected to be dependent on many
variables. The chemistry and thermal conductivity of
The intrinsic thermal resistance can be calculated as the adhesive, thermal conductivity of the device and
the ratio of the bondline thickness to the adhesive substrate, processing and curing conditions, effects of
thermal conductivity thermal stress due to thermal expansion mismatch and
the surface condition (plating, surface finish) of the
device and substrate can all effect the contact
resistance. The thermal conductivity of the adhesive in
The contact,resistance is defined as the bonded assembly may also be dependent on
variables such as processing and curing conditions,
bondline thickness and area and size, shape and
orientation of filler particles.
and the total bondline thermal resistance becomes

III. EXPERIMENTAL METHOD LASER -


FLASH DIFFUSIVITY
The effective thermal conductivity is defined as
A. Single Layer Diffusivity
AS I (AX I h + 2Kon).
he,= AX I hot=
/ The laser flash diffusivity method, or more generally
is useh1 in comparing the performance of different the flash diffusivity method, originally described by
adhesive bondlines, including their contact resistance, W.J. Parker [I] is a transient heat flow technique
but is a function of bondline thickness because the primarily used to measure the thermal diffusivity of
2&,) term is fixed and begins to dominate for thin materials. Thermal diffusivity is a measure of how
bondlines. quickly a temperature disturbance can propagate
through a material and is related to the thermal
Equation (1) suggests two methods to determine %, conductivity through the following equation
from measurements of Et,, and AX, neither of which is
always valid. If Ronand are not functions of As,
then a linear fit of Equation (1) for a range of As will
produce a y-asis intercept equal to 2(Kon). The This relationship allows the thermal conductivity to be

84 Fifteenth IEEE SEMI-THERMTMSymposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
-
Figure 1 Schematic of die attach bondline

calculated from measuremcnts of the thermal For single lager samples we also employ the laser flash
diffusivity, bulk density and specific heat. method to measure the specific heat of the sample by a
calibrated method [I]. This technique involves
The laser flash method (sce Figure 2) in\-olves rapidly comparing the sample temperature rise due to the laser
heating one face of a small disk or slab of the material pulse to that of a sample of known specific heat tested
with a single pulse from a laser or other source and under the same conditions. After calibration, the
monitoring the arrival of the resulting temperature specific heat and thermal diffusivity can be measured
disturbance as a function of time on the other face of on the same sample usually at the same time.
the sample. Normally the back face surface
temperature is monitored with an infrared detector, C. Multi-layer analysis
making the measurement contactless. This is a
significant advantage over other methods especially for The method can be applied to the determination of the
thin, low thermal resistance samples. Uncertainties in thermal diffusivity, thermal conductivity and thernial
other methods related to the contact resistance between resistance of an unknown layer in a multi-layer or
heating and cooling plates and the sample, and to “sandwich” structure, provided that the thermal
accurate measurement of surface temperature properties and thickness of the other layer or layers are
differences using contact sensors can be essentially known or can be measured independently. We have
eliminated. used the approach of T.R.Lee [3] for three layer
samples. In this case the bondline represents the
Figure 3 is an esample of the normalized back surface middle layer and the contact thermal resistances are
temperature rise measured by the IR detector. In this lumped in the middle layer effective thermal diffusivity
figure, the laser pulse is initiated at time zero, there is (and conductivity). J.N. Sweet [4] has developed a
short lag as the pulse propogates through the sample three layer solution which can account for the two
thickness, and then a steep rise followed by a more contact resistances separately from the middle layer
gradual rise to a plateau where the heat from the laser difisivity. However, because changes in the contact
pulse has become equally distributed in the sample. resistances and the middle layer diffusivity tend to
effect the temperature rise curve in the same way, it is
Data analysis consists of fitting the measured back usually not possible to determine both quantities
surface temperature rise to the theoretical model to simultaneously using this solution. Although Sweet’s
determine the value of diffusivity for the sample. For model more accurately represents our samples, both
single layer samples, we have used the approach of solutions applied to the same data will result in the
J.A. Koski [2], in which corrections for radiative heat same fit value of the bondline total thermal resistance.
losses and the finite width of the laser pulse can be
applied. Data analysis consists of fitting the measured
temperature rise data to the theoretical model to
B. Specific heat determine the value of diffusivity for the bondline
lager. The Lee solution relates the back surface

a5 Fifteenth IEEE SEMI-THERMTMSyrnposiurn

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
bondline +
Si \
1-
U detector

.? 4 k. 4- IR radiation
... :. .-.,
+ sample
Si 4 4 4 4 4
. I I I I

Figure 2 - Schematic of laser flash diffusivity method

temperature rise vs. time to the thickness, density, utilized in cases where the bondline thickness
specific heat and difisivity of each layer. Normally approaches zero relative to the first and second layers.
the tso or half rise time of the three layer sample is This method has the advantage of not requiring inputs
found by measuring the apparent difisivity and is of the bondline bulk density and specific heat, but as
corrected for the finite pulse effect [2]. In Figure 3, tso the first and second layer thickness increase relative to
is the time where normalized temperature equals 0.5 or the bondline thickness, the sensitivity of the bondline
about 7 ms on this curve. Then, with inputs of tso and thermal resistance measurement is reduced. This
the thickness, density, specific heat and diffusivity of solution is exact only as the bondline thickness
each layer (the diffusivity of the bondline layer is approaches zero. However, for some of the layer ratios
initially estimated), the diffusivity of the bondline layer encountered in this work the deviation between the
is iterated until the normalized temperature rise calculated thermal resistance using the two layer
calculated with the three layer solution converges to approximation and using the three layer solution was
the value of 0.5 (for tso). Although the Lee solution only about 3%.
requires inputs of the bondline thickness, specific heat
and density, the final calculated value of bondline D. Inplane thermal diffusivity and thermal
thermal conductivity and thermal resistance is conductivity
relatively insensitive to the specific heat and density
inputs for the layer thickness ratios of the samples in In order to characterize any possible anisotropy in the
this work. free-standing adhesive samples, we have employed a
modification of the standard through-plane laser flash
Sweet and Lee have also developed solutions for two method [5, 6, 71 to measure ".y and L,,. This method
layer samples with contact resistance that can be employs a mask to collimate the laser beam and heat

86 Fifteenth IEEE SEMI-THERMTMSymposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
an approsiniately 5 nim diameter spot on the front face detector. The temperature rise measured at this
of a 25 m m diameter disk. This allows heat to flow annulus is fit to a two-dimensional model with inputs
through the thickness of the sample and spread out in of the through-plane thermal diffusivity, saniple
the plane. Another mask is employed on the opposite thickness and heating and viewing diameters to
face to define an annulus of approsiniately 8 to 10 nim determine axy.
in diameter on the back face that is esposed to the IR

temperature rise curve

0.9 -

0.8 -
0.7 -

2 0.6-
E
E 0.5 - _
d.
E 0.4 -
2
0.3-

0.2 - - - I

-.
3
0.1 - a
I

O /
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
time (seconds)

Figure 3 - Back surface temperature rise curve measured by IR detector

N. SAMPLE PREPARATION The Si-Si bonded samples were prepared by using a


manual die attach machine to align and bond two 0.3
Free standing adhesive films were prepared by doctor or 0.4 inch square silicon die with the adhesive, and
blading the adhesive paste between strips of tape (for then performing the supplier-recommended curing
thickness control) on a Teflon block and then cycle. After curing, most of the adhesive “squeeze out”
performing the supplier-recommended curing cycle. was removed from the faces and edges of the die. The
After curing, the films were released from the Teflon Si-& bonded samples were prepared by using the
block and three 0.5 inch diameter disks were die cut manual die attach machine to align and bond a 0.3
for laser flash testing. For some of the films, 1.0 inch inch square silicon die near the center of a 1.0 inch
diameter disks were also cut for in-plane diffusivity square 6061 series aluminum substrate, and then
measurements. performing the supplier-recommended curing cycle.
The “squeeze out” fillet was left intact for the Si-AI

a7 Fifteenth IEEE SEMI-THERMTMSymposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
samples.
Eight different silver-filled adhesives were investigated
Due to the partial transparency of silicon in the IR and are summarized in Table 1. These adhesives were
wavelengths, the outer Si faces of the bonded samples chosen to represent a range of binder types and silver
were sputter-coated with a 0.1 pm gold film. This loading levels. The addition of spherical silver spacers
prevents transmission of laser energy and prevents the to the pastes of adhesives 3 and 7 was an experimental
IR detector from viewing internal temperature change. attempt to achieve uniform bondline thickness control.
A graphite film approximately 2 p i thick was then These spacers, approsimately 38 pm (1.5 mils) in
applied to increase the laser energy absorption and the diameter and uniform in diameter from spacer to
emmitance of the sample surfaces. spacer serve as stand-offs during die attach. Note that
the weight and volume percentages of silver do not
V. MEASUREMENTS include the silver spacers.

The free-standing film sample's thickness, diameter A. Free-standing thermal conductivity


and mass were measured before applying the graphite
coating in order to calculate the bulk density. The The results for the free-standing adhesive films are
cured films tended to have significant thickness sumniarized in Table 2 and show some dependence of
variations so a flathall micrometer was used to make the thermal conductivity on the binder chemistry and
five thickness measurements with light contact silver loading level. The thermal conductivity for
pressure to avoid compression. Adhesives 1 and 2 was lower than expected for their
silver loading level, but for Adhesives 3 through 8 the
For the bonded samples the thickness of the Si die and thermal conductivity increases in the same order as the
A1 substrates were measured before assembly with a silver loading level. Adhesive 5 has a similar silver
flatlpoint micrometer with a precision of 0.1 mil. loading level as Adhesive 4, but has a much higher
After curing the total thickness of the Si-Si and Si-AI thermal conductivity due to its solvent and
samples was measured with the flatlpoint micrometer thermoplastic binder system which may allow for more
near each of the four comers and averaged. The intimate silver particle to particle contact [SI.
average bondline thickness can be calculated by Adhesive 6 has mixed thermoplastic and thermoset
subtracting the average total thickness from the sum of binders and higher silver loading. Adhesives 7 and 8
the Si or Si and A1 thicknesses. have extremely high thermal conductivities of 85 and
,66 W/m-K. This result on Adhesive 8 is similar to
The specific heat and thermal diffusivity of the free- previous measurements on much thicker (3 mm)
standing film samples, and the thermal diffusivity of molded samples of this adhesive. The addition of
the Si die, A1 substrate and Si-Si bonded samples were silver spacers to Adhesive 7 may have resulted in
measured on a Holonietrix Microflasli instrument. slightly higher silver loading and higher thermal
This instrument utilizes a Nd:glass laser (1.06 pm conductivity.
wavelength, 330 ps pulse width, 10 mm beam
diameter) and an InSb IR detector. The thermal The measured film thermal conductivities for
diffusivity of the Si-A1 bonded samples was measured Adhesives 2, 5 and 6 were significantly lower than
on a Holometris Thermaflash 2200 instrument which previous results on much thicker (3-10 mm) molded
has the same laser and detector but allowed heating of samples. One possible esplanation for these
most of the 1.0 inch square A1 substrate area due to the differences is the formation of a binder-rich layer or
divergence of the laser beam at large distances from gradient at one surface of the film due to the tendency
the laser head. Heating most of the A1 substrate area for the binder to migrate to the one available interface
was important to minimize in-plane heat conduction. at the surface of the Teflon block on which the films
A mask was used to eqose only the die area of the Si- were cured. A difference in color was noted between
A1 samples to the LR detector. The in-plane diffusivity the Teflon and free sides of the films that corresponded
was also measured on the Thermaflash 2200 to a binder-rich surface on the Teflon side. Due to the
instrument in a cell that allows the measurements to very low thermal conductivity of the unfilled binder,
performed in vacuum to limit heat loss. the thermal resistance of this layer may be high even
though its thickness may be fairly small compared to
VI. RESULTS AND DISCUSSION the film thickness. For the thicker samples cured in a

88 Fifteenth IEEE SEMI-THERMTMSymposiurn

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
mold, a binder-rich layer at each surface of similar longer dimension parallel to the film surface, resulting
thickness may not be thick enough compared to the in anisotropy in the thermal conductivity. The
total thickness to have a similar effect. The thermal through-plane thermal conductivity X, would be
conductivity for Adhesives 3, 4 and 8 was more similar reduced and in-plane thermal conductivity hxy
to previous measurements of much thicker samples. increased relative to a random orientation analogous to
unidirectional fibrous composites.
Another possible exqhnation for the lower film
thermal conductivity is a difference in the orientation
of the silver particles within the adhesive. Both silver B. In-plane thermal conductivity
flake and silver powders are used as fillers. The silver
flake have diameters of approximately 5-15 pm with In order to investigate possible anisotropy in the
an aspect ratio of 10 to 1. Due to processing and thermal conductivity of the free-standing samples, in-
curing conditions it may be that thicker bulk samples plane thermal diffusivity measurements were
tend to have a random orientation of the silver flake. perfomled on
In the film the silver flakes may tend to align with the

-
Table 1 Summary of tested adhesives
I Adhesive I Bindertype I Wt. % ofAg I Vol. % ofAg I Notes I
Adhesive 1 I epoxy I 80 I 31 I measured h,and L o n l y
Adhesive2 I epoxy 82 34 I measured h, and 3Lxyonly
Adhesive 3 epoxy 77 28 contains Ag spacers
Adhesive 4 epoxy 90 51
Adhesive 5 solvent, thermoplastic 90 53
Adhesive 6 solvent, mised 92 58
thermoplastic and
thermoset
Adhesive 7 solvent, epoxy I 93 62 I contains Ag spacers
Adhesive 8 solvent, epoxy 93 62

-
Table 2 Freestanding adhesive propcrties

Fifteenth IEEE SEMl-THERMTMSymposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
Adhesives 1, 2, 5, 6 and 8. The results are two theories.
summarized in table 3 and show very significant
anisotropy for these free-standing films. Note that the We are planning to perform SEM on sectioas of these
results for Adhesive 8 are on a thinner film of a films in an attempt to learn more about their
different material lot from the result in Table 2. The microstructure and its correlation to measured & and
final column in Table 3 lists the Ly/ h, ratio, which
indicates the degree of anisotropy. There appears to be
an inverse relationship between the & / h, ratio and C. Bondline effective thermal conductivity and
the silver loading level. These results could be thermal resistance
consistent will the binder-rich layer theory if one
assumes that the thin, binder-rich layer and the thicker Several groups of bonded samples were tested. The Si
silver-filled layer act as a large thermal resistance in die, substrates and adhesives tested in each group are
series with a small themial resistance in measurements summarized in Table 4. Table 5 contains the
of h,, and act as a large thermal resistance in parallel properties of the Si die and A1 substrate. Note that the
with a small thermal resistance in measurements of specific heats of the Si die and the AI substrate were
LY.The results could also be consistent with the silver- taken from the literature [9].
flake orientation theory or some combination of the

-
Table 3 Free standing adhesive inplane thermal conductivity
T a. av 1, L y LyIL
(00 (cm2/s) (cm2/s) (WIm-K) (W/m-K)
I

-
Adhesive 1 25 0.0036 0.0133 0.67 I 2.46 3.7
Adhesive 2 25 0.0088 0.0357 1.67 I 6.79 I 4.1 I
Adhesive 5 25 0.0467 0.112 6.24 I 15.0 1 2.4 I
Adhesive 6 25 0.0651 0.108 9.07 I 15.0 I 1.7 I
Adhesive 8 25 0.301 0.452 48.0 I 72.1 I 1.5 I

-
Table 4 Bonded sample summary

s tested

(Si-Si) 315
polished
mound
I ground

Group2 400x 457 ground/ I ground 5,63


(Si-Si) 400 ground
Group3 300x 610 polished ground 5
(Si-Si) 300 round
Group4 300x 610 5
(Si-Si) 300
Group5 300x 610 1 .os 965 rolled 596
(Si-AI) 300 round 1 .o

90 Fifteenth IEEE SEMI-THERMTMSymposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
Ax As p T CP . a A,
(mils) (pm) (g/cm3) ("C) (J/g-K) (cm2/s) (W/m-K)
I
Sidie(avg. of3), 15 381 2.33 25 0.715 0.860 143
Group 1
Sidie(avg. of2), 18 457 2.31 25 0.715 0.820 135
Group 2
Sidie(avg. of2), 21 610 2.31 25 0.715 0.827 137
Groups 3,4,5
A1 substrate 38 965 2.70 25 0.905 0.664 162

Group 1: bondline for thickness measurements. These optical


thickness measurements agreed fairly well with the
Group 1 included 0.315 inch square Si-Si bonds for bondline thickness calculated by subtraction as
Adhesives 3, 4 and 7. The Si-Si bond was chosen so described in Section V. The masimum difference
that the effect of adhesive binder type and silver between the two methods for the 12 bonded samples
loading level on the bondline thermal properties could was 5.6 p n ~(0.22 mils) and the average difference was
be studied. Four Si-Si replicates were tested for each 2.2 pm (0.09 mils). Espressed as a percentage of the
adhesive. The results are summarized in Table 6. The bondline thickness, the masimum difference was
contact resistance is calculated by the subtraction 18.5% and the average difference was 6.5%. Note that
method (Equation 2) using the free-standing the results in Table 6 were calculated with the optical
conductivities from Table 2. The final column is the thickness measurements.
calculated contact resistance as a percentage of the
total measured resistance. Group 2:

These results show some dependence of the effective Group 2 included 0.400 inch square Si-Si bonds for
thermal conductivity on the adhesive binder type and Adhesives 5, 6 and 8. Five Si-Si replicates were tested
silver loading level, but the results are much more for each adhesive. The results are summarized in
closely grouped than the corresponding free-standing Table 7.
results. Adhesive 3 had the lowest free-standing
thermal conductivity in this group and the highest The contact resistance for Adhesive 8 is calculated by
calculated contact resistance as espected, but Adhesive the subtraction method (Equation 2) using its free-
7 had a much higher calculated contact resistance than standing thermal conductivity from Table 2. For
Adhesive 4 despite its much higher (factor of 23) free- Adhesives 5 and 6, the free-standing thermal
standing thermal conductivity. After the thermal conductivity from previous measurements of thicker
testing x-ray images were t<aken of this group of molded samples was used because Et, was about the
samples and it was determined that the Adhesive 7 same or lower than Rhk calculated with the current
samples had significant voiding. This voiding results on the free-standing films. Use of the film
probably caused the additional bondline thermal conductivity for Adhesive 5 resulted in negative
resistance and calculated contact resistance for this calculated contact resistance for each sample while for
adhesive. Adhesive 7 contains non-reactive solvents Adhesive 6 the calculated contact resistance ranged
and experiences significant shrinkage as the solvents from negative to 0.9 m r n ' w . This is a clear
are evaporated during the cure cycle. The silver indication that the microstructure of the free-standing
spacers used in Adhesive 7 to achieve uniform films of these adhesives did not accurately represent
bondline thickness did not allow for the bondline the microstructure of the adhesives in the Si-Si
shrinkage and the voids probably developed as the bondlines. Some possible esplanations for the lower
adhesive pulled away from the silicon during the cure. through-plane thermal conductivity of these films
compared to thick molded samples were discussed in
The samples in this group were later sectioned and Section C, but clearly hrther investigation is necessary
polished in order to obtain a SEM micrograph of the

91 Fifteenth lEEE SEMI-THERMTMSymposiurn

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
to undcrstand the adhesive microstructure in the sinall range of bondline thickness. Comparing the
bondline. For this work, some unsuccessful attempts contact resistance of Adhesive 8 (2.3 mmzWW) to that
were made to produce a more realistic free-standing of Adhesive 7 (14.7 mm2WW), which is the same
film of Adhesives 5 and 6 in the bondline thickness adhesive escept for the addition of silver spacers, the
range of 2 mils. additional contact resistance or intrinsic resistance
(voiding) caused by the presence of the spacers is
Adhesive 5 had the lowest average calculated contact evident.
resistance of 2.3 mm2WW, Adhesive 8 was 2.6
mni2WW and Adhesive G was 3.5 mm’WW. At the The effective thermal conductivities are again more
higher silver loading levels of these adhesives, the closely grouped for Adhesives 5, 6 and 8 than their .
binder type appears to impact the contact resistance corresponding free-standing results. This illustrates
more than the filler level or thermal conductivity. For that as the total bondline thermal resistance decreases
Adhesive 8, with the exception of sample b, the towards the contact resistance due to the higher
calculated contact resistance for each replicate is very thermal conductivity and lower intrinsic resistance of
close to 2.3 niniZWW (note that the table results have these adhesives, the contact resistance begins to
been rounded) which indicates a linear bondline dominate.
thickness vs. total resistance relationship over this

Tablc 6 - Effective thermal conductivity and thermal resistance of Si-Si bonds (Group 1)

92 Fifteenth IEEE SEMI-THERMTMSyrnposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
Tablc 7 - Effective tliermal conductivity and thermal rcsistilncc of Si-Si bonds (Group 2)

Adhesive 5
(0.4’’ Si-0.4” Si)

E averu e

1 Adhesive 8
(0.4” Si-0.4” Si)

avera e

Groups 3 and 4: was somewhat smoother than the ground faces of the
0.400 inch square die used in Group 2. Although these
These groups included 0.300 inch square Si-Si bonds surface finish characterizations were not quantitative,
of Adhesive 5 with the same Si die but with different they should at least rank the finishes of the bonded die
sides of the die bonded. Five Si-Si replicates were surface.
tested for each group. The results are summarized in
Table 8. The contact resistance is calculated with the For Adhesive 5 , the roughest die finish resulted in an
subtraction method. average contact resistance of 2.3 mm2K/W (Group 2,
Table 7) and the polished die finish of Group 4
These die had a polished mirror finish (Group 4) on resulted in an average contact resistance of 2.9
one side and a ground finish (Group 3) on the other. mm2WW. The average contact resistance for Group 3,
This comparison was intended to show the effect of with a somewhat smoother ground die finish was 2.7
surface finish on the contact resistance of this mm2K/W. These results show a fairly small (relative
adhesive. Also, the ground finish side of this set of die to the uncertainty in these contact resistance

93 Fifteenth IEEE SEMI-THERMTMSymposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
-
Table 8 Effective thermill conductivity ant1,thcrmal resistance of Si-Si bonds (Groups 3 and 4)

Table 9 Effective thermal conductivity


- 5 of Si-AI bonds (Group )
and thermal resistance

94 Fifteenth IEEE SEMI-THERMTMSymposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
calculations) but measurable effect of surface finish on thermoplastic binder system and lower modulus of
the contact resistance for this adhesive. However, the elasticity [8], Adhesive 5 was espected to be better able
result for Group 3 would be expected to be closer to to absorb the sheer stress. The samples for this
that of the rougher ground finish (Group 2) than to the adhesive were cured at 175OC and were also not given
polished finish. Note that the die size was also additional thermal cycles
different for Group 2, but this would not be espected to
effect the contact resistance. The lower contact A portion of the increased contact resistance for both
resistance with a rougher finish is probably due to the adhesives may be due to a higher Al-adhesive contact
increased contact area between the die and adhesive at resistance without any sheer stress factor. Some
the interfaces. possible esplanations for this include any effects of the
thin aluminum ohde layer on the A1 surface,
Group 5: contamination of the A1 surface or a surface finish
effect.
This group included 0.300 inch square Si-1.0 inch
square A1 bonds of Adhesives 5 and 6. Five Si-A1 The results of future work may allow the separation of
replicates were tested for each adhesive. The results any sheer stress factors from these other potential
are summarized in Table 9. The contact resistance is factors by testing ALA1 bonds of these adhesives and
calculated with the subtraction method. comparing the contact resistance to that of the Si-Si
bonds. Also, s-ray or ultrasonic analysis of the Si-A1
This sample configuration was chosen to esamine the bonded samples may help determine whether sheer
effects of using the Si-AI combination due to the large stress has damaged these bondlines.
thermal espansion mismatch between Si and A1
(approximately 3 ppnd0C vs. 23 ppm/"C). These
results show a very significant increase in calculated VIL SUMMARY
contact resistance and decrease in effective
conductivity relative to the Si-Si bonds for the same The laser flash diffusivity method has been used to
adhesives in Table 2. The average calculated contact characterize both the free-standing thermal
resistance of Adhesive 5 is 8.9 m m 2 W and the conductivity and bondline effective thermal
average effective thermal conductivity is 3.2 W/m-K conductivity and thermal resistance of a range of silver
compared to values of 2.3 mm2K/W and 9.3 W/m-K filled adhesives. Accurate knowledge of the thermal
for the Si-Si bonds. For Adhesive 6, the Si-Ai values performance of these adhesives is critical to thermal
are 19.8 m m 2 W and 1.8 W/m-K compared to values management and modeling of high power device
of 3.5 m m 2 W and 8.5 W/m-K for the Si-Si bonds. packages.
Escluding sample d from the average for Adhesive 6
gives values of 13.0 mm2K/W and 2.2 W/m-K. The Measurements of the free-standing films have shown
very high contact resistance of 47 nim2WW calculated the effects of silver filler level, binder type and sample
for Sample d might indicate a partial adhesive failure. preparation on the thermal conductivity. Significant
Mechanical sheer strength testing done on a group of anisotropy in these films has been measured using an
similarly prepared samples also indicated some low in-plane modification of the laser flash diffusivity
strengths for a small number of bonds. method.

For Adhesive 6, which has a blended thermoplastic Measurements of bonded assemblies of Si-Si and Si-A1
and thermoset binder, the sheer stress caused by the have been used to study the effects of adhesive thermal
extreme Si-A1 thermal expansion mismatch may have conductivity, filler level and binder type, as well as
led to cracking or voiding within the bondline or at the surface finish and thermal expansion mismatch on the
interfaces, increasing the calculated contact resistance. effective thermal conductivity and contact resistance of
The samples for this adhesive were cured at 200°C but
were not given additional thermal cycles. Due to its

95 Fifteenth IEEE SEMI-THERMTMSymposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
40.0
38.0
36.0
34.0
32.0
30.0 A Adh.4(Si-Si)

28.0 ----
26.0 Adh.S(SiSi)
I..._.._._...._
24.0
+ Adh.S(SiAl)
5 22.0
20.0
-
N

E 16.0 -
- Adh.6(SiSii

16.0
o Adh.6(SiAl)
14.0
12.0
-
10.0
8.0
-.. 0 Adh.7(SiiSi)

A Adh.8(Si-Si)
6.0
I . . -
4.0
2.0
0.0
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70
Pm

Figure 3 - Total thermal resistancc vs. bondlinc thickness for Groups 1 , 2 and 5

bondlines of the adhesives. The importance of REFERENCES


accounting for the contact resistance has been
demonstrated by calculated contact resistances that [I] W.J. Parker, R.J. Jenkins, C.P. Butler, and G.L.
range from 30% to over 90% of the total bondline Abbott, "A Flash Method of Determining
thermal resistance for the adhesives and assemblies Thermal Diffusivity, Heat Capacity, and Thermal
studied. Figure 3 summarizes the total thermal Conductivity", Journal 01Applied Physics, 32
resistance vs. bondline thickness for Groups 1, 2 and 5. (9): 1961, 1679-1684.
This plot includes the data points for each replicate as [Z] J.A. Koski, "Improved Data Reduction Methods
well as a line derived from Equation I using the for Laser Pulse Diffusivity Determination with
adhesive free-standing thernial conductivity and the the Use of Minicomputers", in Proceedings of tlie
average calculated contact resistance for the sample Eighth ~vinposiiinr on Therrnophysical
configuration (note that it is not a linear fit of the data Properties, Vol 11, J.V. Sengers, Ed, The
points). This plot allows the determination of the total American Institute of Physics, 1981, 94-103.
thermal resistance at any bondline thickness for each [3] T.R. Lee, Purdue University Ph.D. Thesis
adhesive and configuration. It also shows the degree Therinal Dil/usivi@ of Dispersed and Layered
of scatter in the measured total resistance for the Cornposifes,University Microfilms International,
replicates. 1977.
[4] J.N. Sweet, Dntn Analysis Methods for Flash
Tliernrnl Dgfirsivity Experir~ients, Sandia

96 Fifteenth IEEE SEMI-THERMTMSymposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.
National Laboratories Report nunibcr SAND89-
0260 UC-25, 1989.
[5] A.B. Donaldson, "T\\lo-Din~ensionaI Thermal
Attenuation of a Laser Pulse in a Solid". Journal
of the Franklin bisritrile, 294(4), 1972. 275-281.
[6] A.B. Donaldson. "Radial Conduction Effects in
the Pulse Method of Measuring Thermal
Diffusivity", Journal of Applied Physics, 33( lo),
1972,4226-4228.
[7] A.B. Donaldson and R.E. Taylor, "Thermal
Difisivity Measurement by a Radial Heat Flow
Method", Journal of Applied PhJJsics, 46( lo),
1975,4581489.
[SI R.L. Dietz, D. Peck and P.J. Robinson, "High
Thermal Conductivity in Organic Adhesives",
Intemationnl A.licroelectronics CoiiJerence,
Japan, 1996.
[9] Y.S. Touloukian and E.H. Buyco,
"Thennophysical Properties of Matter", Vol. 4,
Specijk Heat of A4etallic Elenmits and AIloj~s,
IFWlenum, 1973.

97 Fifteenth IEEE SEMI-THERMTMSymposium

Authorized licensed use limited to: University of Catania. Downloaded on November 30,2022 at 14:40:13 UTC from IEEE Xplore. Restrictions apply.

You might also like