Full text of DOI:10.1557/JMR.2000.0033

Journal of
MATERIALS RESEARCH
Welcome
Comments
Help
Adhesion and reliability of copper interconnects with Ta and
TaN barrier layers
Michael Lane and Reinhold H. Dauskardt
Department of Materials Science and Engineering, Stanford University,
Stanford, California 94305-2205
Nety Krishna and Imran Hashim
Applied Materials Corporation, Santa Clara, California 95052
(Received 20 August 1999; accepted 25 October 1999)
With the advent of copper metallization in interconnect structures, new barrier layers
are required to prevent copper diffusion into adjacent dielectrics and the underlying
silicon. The barrier must also provide adequate adhesion to both the dielectric and
copper. While Ta and TaN barrier layers have been incorporated for these purposes in
copper metallization schemes, little quantitative data exist on their adhesive properties.
In this study, the critical interface fracture energy and the subcritical debonding
behavior of ion-metal-plasma sputtered Ta and TaN barrier layers in Cu interconnect
structures were investigated. Specifically, the effects of interfacial chemistry, Cu layer
thickness, and oxide type were examined. Behavior is rationalized in terms of relevant
reactions at the barrier/dielectric interface and plasticity in adjacent metal layers.
to their adhesive and fracture properties. Accordingly, in
I. INTRODUCTION
this study we address the adhesion and subcritical debond-
ing behavior of the barrier/SiO2 interface, together with its
microstructural and chemical determinants.
Current semiconductor technology dictates the use of
low-resistivity metal lines for multilayer interconnect
devices. Currently, Al–Cu metal layers are employed in
most devices; however, copper has emerged as an attrac-
tive alternative. Copper has a higher conductivity and
superior electromigration resistance than aluminum al-
loys leading to shorter resistance capacitance (RC) delay
times and reduced failure due to electromigration.1–3
However, integration of Cu lines into interconnect
structures poses several challenges. Cu is a more mobile
element than Al and, for example, diffuses rapidly in
SiO2 and forms silicides at temperatures as low as
200 °C.1,4,5 A reliable barrier layer is therefore needed to
inhibit the diffusion of copper into adjacent dielectric
layers and the silicon substrate. In addition, the barrier
layer must also provide adequate adhesion to both the
dielectric and metal layers. The resulting adhesive
strength must be able to withstand the high residual
stresses associated with thermal expansion mismatch
and film growth processes, as well as the high back
stresses developed during electromigration processes in
the Cu lines.2,3
By adhesion we are primarily concerned with the total
energy required to separate an interface, measured in
terms of a critical value of the debond strain energy
release rate or interface debond energy, Gc (J/m2).8 This
energy includes not only the breaking of chemical bonds
across the interface but also plasticity in any adjacent
ductile layers. Energy dissipation in debond wake proc-
esses (such as a frictional contact zone) must also be
considered. Previous work has shown that each of these
processes can significantly affect the critical interface
debond energy.9–12 In the case of Al–Cu interconnects,
plastic dissipation in the metal layer adjacent to the in-
terface of interest was found to increase the interface
fracture energy by over 300% when the Al–Cu layer
thickness was increased from 0.1 to 4.0 ␮m.9–11 Like-
wise, increasing the roughness of the interface was found
to increase interface fracture energies by 25% for root-
mean-square (rms) roughness changes of as little as
7%.12 These energy-dissipation processes typically do
not act independently but behave in a synergistic fashion.
In addition, both processes of frictional contact and plas-
ticity scale with the interfacial chemical bond strength
so that moderate changes in interfacial chemistry may
have pronounced effects on the macroscopic work of
adhesion.8,11,13–15
Tantalum and tantalum-based films have emerged as
promising diffusion barriers for copper interconnects.
While much effort has been made to characterize the
phase stability and diffusion barrier properties of these lay-
ers (e.g., Refs. 6 and 7), almost no attention has been given
J. Mater. Res., Vol. 15, No. 1, Jan 2000
© 2000 Materials Research Society
203
http://journals.cambridge.org
Downloaded: 15 Mar 2015
IP address: 131.170.6.51

M. Lane et al.: Adhesion and reliability of copper interconnects with Ta and TaN barrier layers
II. EXPERIMENTAL PROCEDURES
A. Sample preparation
Of equal importance is the phenomenon of subcritical
debonding which may lead to time-dependent failure of
device structures either during fabrication or in ser-
vice.9,11,16,17 Preexisting flaws at an interface may grow
subcritically over time at a strain energy release rate, G,
far lower than the critical interface debond energy, Gc.
Previous work on Al–Cu interconnect structures has
shown that strain energy release rates of less than 50% of
Gc may propagate debonds along an interface.9,11 Fur-
thermore, environmental factors such as moisture content
and temperature may significantly accelerate the debond
growth velocity. The mechanical driving forces for such
time-dependent delamination processes include residual
thermal expansion mismatch and growth stresses, alter-
nating stresses from thermal cycling, and electromigra-
tion back-stresses. During fabrication the mechanical
stresses and moist environments associated with chemi-
cal mechanical processing (CMP) operations may be par-
ticularly deleterious.
Thin film stacks shown schematically in Fig. 1 were
fabricated on 8-in. silicon wafers using ion-metal-plasma
sputtering (for metal and barrier layers) on thermal oxide.
One sample set was fabricated with the above stack struc-
ture on tetraethyl orthosilicale (TEOS) oxide in order to
assess the effects of oxide type. Nominal thickness for
the Ta and TaN layers were 250 Å, with 1500 Å for the
Cu layer. The TaN layer was deposited by sputtering Ta
while flowing N2 into the chamber resulting in a uniform
The objective of this paper is therefore to report on
progress toward characterizing and understanding the
adhesion in interconnect structures employing copper
metallization and a selected range of adjacent barrier
and dielectric layers. Specifically, issues of plasticity in
the copper layer, type of dielectric oxide, and barrier
layer chemistry are investigated. Adhesion and subcriti-
cal debonding data are presented for a selected range
of Ta and TaN barrier layers. The presence of N in
the barrier layer is shown to significantly improve adhe-
sion and resistance to subcritical debonding. Behavior
is rationalized in terms of relevant reactions at the bar-
rier/SiO2 interface and plasticity in the adjacent cop-
per layer.
FIG. 1. Schematic illustration of the thin-film structures with Cu met-
allization showing (a) TaN or Ta barrier layers and (b) the TaN/Ta
barrier stack.
FIG. 2. Effect of Ta and TaN barrier layers on (a) the barrier layer/
SiO2 interface adhesion and (b) the subcritical debond growth rate as
a function of the debond strain energy release rate, G.
204
J. Mater. Res., Vol. 15, No. 1, Jan 2000
Downloaded: 15 Mar 2015
http://journals.cambridge.org
IP address: 131.170.6.51

M. Lane et al.: Adhesion and reliability of copper interconnects with Ta and TaN barrier layers
FIG. 3. XPS spectra obtained from fracture surfaces of samples with the Ta barrier layer showing (a) the SiO2 side and (b) the Ta side, samples
with TaN barrier layers showing (c) the SiO2 side and (d) the TaN side, and (e) a detailed scan showing N and Ta on the SiO2 side of the fracture
surface.
B. Adhesion and subcritical debonding tests
TaN layer with 30 at.% N, which lies in the Ta2N region
of the equilibrium phase diagram [Fig. 1(a)]. Additional
TaN/Ta(N)/SiO2 samples were fabricated by first depos-
iting a 20 Å Ta layer and subsequently depositing the
TaN layer [Fig. 1(b)]. The initial N2 flow was altered so
as to produce films with 30, 40, and 50 at.% N at the
Ta(N)/SiO2 interface as measured by Rutherford backs-
cattering spectroscopy (RBS). However, the resulting N
enhancement was only in the first few angstroms adja-
cent to the Ta(N)/SiO2 interface and the TaN film
retained desirable properties for semiconductor metalli-
zation. Physical vapor deposition (PVD) processing by
first depositing 20 Å of Ta and then flowing N2 into the
chamber for subsequent deposition of the layer assisted
in keeping the Ta source free of TaN contamination.
These samples were also fabricated without the Cu layer
and heat treated to simulate subsequent sample prepara-
tion procedures in order to facilitate chemical analysis of
the interface and the TaN films. A thicker Cu layer of
1.65 ␮m was also produced using electroplating in order
to determine the effects of metal layer thickness.
The thin-film stack of interest was sandwiched be-
tween two elastic silicon substrates in order to fabricate
four-point flexure samples for adhesion testing.9,11 As
debonding occurs in these sandwiched configurations,
the films of interest are constrained by the adjacent sili-
con substrate and adhesion values are not affected by
relaxation of residual film stresses.11 Two 35 × 35 mm
squares were cleaved from each wafer and diffusion
bonded at 400 °C for 4 h at a pressure of 12 MPa in a
vacuum press. The squares were then diced into strips
3 mm wide using a high-speed diamond saw. A notch
was machined through the top Si wafer to within 20 ␮m
of the interface to initiate debonding. The side faces of
some samples were polished with an aluminum slurry to
facilitate in situ viewing of the debond process and in-
hibit fracture of the Si substrate during testing. The four-
point flexure samples were tested in a high-stiffness
micromechanical test system with fully articulating four-
point bending fixtures employing a piezoelectric actua-
tor. The samples were loaded at a displacement rate of
J. Mater. Res., Vol. 15, No. 1, Jan 2000
Downloaded: 15 Mar 2015
205
http://journals.cambridge.org
IP address: 131.170.6.51

M. Lane et al.: Adhesion and reliability of copper interconnects with Ta and TaN barrier layers
TABLE I. Properties of various glasses taken from Ref. 21.
Soda lime Borosilicate Oxynitride
nificant difference in slope, m, of the two curves. The
slope of the debond growth rate curve for the TaN/SiO2
interface is similar to that reported for moisture assisted
cracking of bulk SiO2 glasses18 and previously reported
values for TiN/SiO2 interfaces in similar interconnect
structures.9,11,16,17 In contrast, the slope of the Ta/SiO2
interface debond growth rate versus strain energy release
rate curve is very steep, much like that expected for the
onset of threshold growth rate behavior where water mol-
ecules are sterically hindered from reaching the debond
tip.19 Further evidence of this phenomenon has been ob-
served over a much wider range of temperatures and
moisture contents, and a detailed model will be presented
elsewhere.20
XPS scans of the resulting fracture surfaces
[Figs. 3(a)–3(e)] clearly reveal that debonding is nomi-
nally along the barrier/SiO2 interface rather than the bar-
rier/Cu interface. Indeed, debonding of the barrier/Cu
interface was never observed for the present sample sets.
A detailed scan centered on the binding energy for N
obtained from the SiO2 side of the TaN barrier layer
samples reveals the presence of N [Fig. 3(e)]. The pres-
ence of N on the SiO2 fracture surface indicates that N
does diffuse into the SiO2 and suggests that guidance for
understanding the increased adhesion for the TaN/SiO2
interface over Ta/SiO2 might be taken from the bulk
glass literature. In general, N-containing glasses exhibit
improved mechanical properties including increased
toughness and subcritical crack-growth resistance over
SiO2 glasses (Table I).21,22 The increase in toughness
has been attributed to the replacement of bivalent O
with trivalent N leading to a more cross-linked, mechan-
Density, ␳
2.47
68.9
5.6
0.77
0.22
2.23
64
5.8
0.76
0.34
3.8
154
10.7
0.97
0.5
Elastic Modulus, E (GPa)
Hardness, H (GPa)
Toughness, Kc (MPa m1/2
)
Stress corrosion, KTH (MPa m1/2
)
0.3 ␮m/s, and loads were measured to a resolution of
0.01 N. Critical loads were taken from the plateau of the
load–displacement curves. Plateau lengths were gener-
ally quite long and associated with several mm of debond
extension. Eight to ten tests were conducted for each
sample type. Subcritical debonding tests were facilitated
by loading the sample to a predetermined load and then
fixing the displacement. The general method has been
described previously.11 Tests were conducted in a labora-
tory air environment [relative humidity (RH) 45%] and
at 25 °C. Several millimeters of debond extension were
observed during the tests.
C. X-ray photoelectron spectroscopy analysis
X-ray photoelectron spectroscopy (XPS) scans were
made on mating fracture surfaces of failed samples. The
surfaces were characterized using a Surface Science XPS
unit with monochromatized Al K␣ x-ray radiation with
an electron escape depth of 20 Å. In order to character-
ize interface reaction products, a broad XPS (0–550 eV)
scan was made of each mating fracture surface. Detailed
scans were then made in regions of the spectra containing
peaks of interest (0, Si, N, and Ta). Finally, samples
nominally identical to the nitrogen-containing samples
but processed without the copper layer were examined.
Scans were made in the TaN film, at the interface and in
the silica.
III. RESULTS AND DISCUSSION
Interface adhesion values derived from plateaus on the
load–displacement curves for different Cu/barrier layer/
SiO2 (thermal) systems are shown in Fig. 2(a). Error bars
in the figure represent the scatter obtained from 8 to 10
measurements. Variation in the measured adhesion val-
ues for the different sample sets is attributed to interface
chemistry variation from different deposition tools used
in the study. The uniform TaN barrier layer was found to
produce approximately twice the interfacial debond en-
ergy compared to the Ta barrier layer. Subcritical
debonding data for these systems are presented in
Fig. 2(b). The TaN/SiO2 interface was found to be more
resistant to subcritical debonding than the Ta/SiO2 sys-
tem as evidenced by the lower growth rates at a given
strain energy release rate, G. Furthermore, there is a sig-
FIG. 4. Interface fracture energy as a function of nitrogen content at
the Ta/SiO2 interface in the TaN/Ta(N)/SiO2 samples.
206
J. Mater. Res., Vol. 15, No. 1, Jan 2000
Downloaded: 15 Mar 2015
http://journals.cambridge.org
IP address: 131.170.6.51

M. Lane et al.: Adhesion and reliability of copper interconnects with Ta and TaN barrier layers
ically and chemically durable glass network structure.21
XPS analysis of the resulting fracture surfaces reveals
two distinct fracture paths. Both the samples containing
30 and 40 at.% interfacial N fail nominally along the
Ta(N)/SiO2 interface similar to the Ta/SiO2 samples,
while the 50 at.% N samples failed in the SiO2. Repre-
sentative scans for the 30 and 40 at.% samples are
shown in Figs. 5(a) and 5(b). In both samples, 2–3 at.%
Si is observed on the Ta(N) fracture surface and ap-
proximately the same percentage of Ta is found on the
SiO2 side, again similar to the Ta/SiO2 interface. Evi-
dence of interfacial reactions was observed in both
samples as illustrated by an asymmetric Si peak on the
SiO2 fracture surfaces and split Ta peaks on the Ta(N)
side. The asymmetric Si peak and the split Ta peak cor-
respond to different bonding states of Si and Ta, re-
spectively. The ratio of the split peak heights changes as
the N content is increased indicating that the amount of
N present at the interface may play a role in deter-
mining the extent of interfacial reactions. A detailed
analysis of the 50 at.% N-containing sample shows one
side (SiO2 side) with no measurable Ta signal but a ro-
bust N peak ( 5 at.%), while the other side [Ta(N) side)
has both Ta and N present in small amounts [Figs. 5(c)
and 5(d)].
The increased interface fracture resistance of TaN/SiO2
over Ta/SiO2 and the presence of N at the fracture
surface in the TaN system indicates that a similar
mechanism may be occurring, as discussed in more
detail below.
The interface debond energy as a function of N content
at the Ta(N)/SiO2 interface in the TaN/Ta(N)/SiO2 sys-
tem is shown in Fig. 4. Interface fracture energy initially
decreases from 10 J/m2 for the uniform TaNx layer to
4 J/m2 for samples produced with 30 at.% N at the in-
terface. However, increasing the N content above
30 at.% at the Ta(N)/SiO2 interface clearly increases the
interface fracture energy. Specifically, increasing the N
content by 10% (30 to 40 at.%) results in a 40% increase
in interface fracture energy while increasing the N con-
tent by 20% (30 to 50 at.%) results in an increase in
interface fracture energy of 145%. Indeed, samples with
50 at.% N exhibit fracture energies similar to samples
processed with uniform TaNx compositions [Fig. 2(a)].
As previously noted, the interface chemistry is expected
to significantly effect the overall debond energy since
most energy absorbing processes scale with the work of
adhesion.14,15
FIG. 5. Representative XPS results from the fractured specimens for the 30 and 40 at.% nitrogen samples showing (a) the SiO2 side and (b) the
Ta(N) side and results for the 50 at.% nitrogen samples showing (c) the SiO2 side and (d) the Ta(N) side.
J. Mater. Res., Vol. 15, No. 1, Jan 2000
207
http://journals.cambridge.org
Downloaded: 15 Mar 2015
IP address: 131.170.6.51

M. Lane et al.: Adhesion and reliability of copper interconnects with Ta and TaN barrier layers
Before detailed consideration of the XPS results
Scans of the 30 and 40 at.% N samples were similar and
differed only in the relative areas of the peaks which
indicates different amounts of each bond type. Assign-
ments of the various peaks can be made on the basis of
literature values of binding energies as summarized in
Table II. In particular, Ta2O5 is indicated by the split
Ta(4f 7) peak and the tail on the low binding energy side
of the asymmetric O(1s) peak while Ta2Si is also indi-
cated by the split Ta(4f 7) peak and the small Si(1s) peak
at 99 eV. The percent of elements in reaction products
(TaOx, TaSiy, and SiNz) was found to increase as the
nitrogen content at the interface was increased (Fig. 7). A
schematic of the stack structure summarizing reaction
layers and fracture paths is shown in Fig. 8.
described above it should be noted that since the frac-
tured samples were exposed to air, any elemental Ta
present at the fracture surface would be expected to oxi-
dize according to
Ta + 2.502 → Ta2O5
−488.7 kcal/mol
,
(1)
where the enthalpy of formation determined from heats
of formation of reactants under standard temperature and
pressure conditions are shown to the right of the reac-
tion.23 Therefore, making direct inferences from the XPS
data based on splits in the Ta peak for fracture paths
exposed to the environment may be compromised.
As a first step toward addressing this issue and iden-
tifying interfacial reaction products, samples were exam-
ined with depth-profiling XPS. First, approximately
100 Å of the top TaN film was sputter etched away and
scans were made of Ta, Si, N, and O peaks to establish a
baseline for the TaN film. After scanning of the bulk TaN
film, the remainder of the film was slowly sputter-etched
away until a Si peak was observed indicating the pres-
ence of the Ta(N)/SiO2 interface. Scans were taken of the
same peaks as in the bulk. Representative results for the
50 at.% N-containing sample are summarized in Fig. 6.
On the basis of above results, it is possible to speculate
on reactions that might take place at the oxide/barrier
interface. Reactions to consider include oxidation of Ta,
TaN, or Ta2N by SiO2 according to
14Ta + 5SiO2 → 2Ta2O5 + 5Ta2Si
−43.9 kcal/mol
,
(2)
(3)
28TaN + 25SiO2 → 7Si3N4 + 10Ta2O5 + 4Ta2Si
855.7 kcal/mol
,
FIG. 6. Depth-profiling XPS scans taken at different locations for the 50 at.% nitrogen sample showing peaks for (a) Ta, (b) Si, (c) N, and (d) O.
208
http://journals.cambridge.org
J. Mater. Res., Vol. 15, No. 1, Jan 2000
Downloaded: 15 Mar 2015
IP address: 131.170.6.51

M. Lane et al.: Adhesion and reliability of copper interconnects with Ta and TaN barrier layers
TABLE II. Photoelectron peak binding energies taken from Ref. 28.
Ta(4f 7) (eV)
Si(1s) (eV)
O(1s) (eV)
N(1s) (eV)
Ta2O5
Ta2Si
SiO2
Ta
26.5–26.9
27.0
иии
иии
530.6
иии
иии
иии
99.4–100.5
102.0–102.7
иии
532.5–534.5
иии
иии
21.8
иии
иии
иии
Si3N4
101.4–101.9
397.4–398.0
FIG. 8. Schematic of interface chemistry interpreted from XPS studies
along with fracture path selection based on interfacial N content.
to the difficulty in separating the Ta(4p3) and N(1s)
peaks, strong evidence of an oxynitride was observed as
shown by Fig. 5(c) where a robust N peak is found on the
SiO2 fracture surface.
On the basis of reaction (5), as nitrogen diffuses to the
interface the reaction enthalpy drops significantly (while
N2 may not be present at the interface, elemental N
would make the reaction even more exothermic). From a
chemical bonding viewpoint, an interface with a more
negative reaction enthalpy is expected to have better ad-
hesion than one with a more positive reaction enthalpy.
This expectation, however, is predicated on the assump-
tion that the interfacial reaction products are fracture re-
sistant. Also, N is likely to be the limiting reagent in
reaction (5) as it must diffuse through the thin Ta layer
before the reaction may proceed. Note that the formation
of Ta2N from Ta and N2 is also a thermodynamically
favorable reaction (−130 kcal/mol) and is expected to
compete with the formation of the other reaction prod-
ucts. However, reaction (5) remains the dominant reac-
tion, and the amount of Ta2N at the interface decreases
with increasing N content as evidenced by Fig. 7, where
the amount of reacted products (TaOx, TaSiy, and SiNz)
increases with increasing N. We conclude that increasing
the atomic concentration of N at the interface drives the
reaction to the right resulting in increased reaction prod-
ucts. Further, the presence of N in the SiO2 suggests the
possibility of the formation of an oxynitride glass which
is more fracture resistant. These two effects contribute to
the improved interfacial adhesion as observed experi-
mentally.
FIG. 7. Percent of reacted elements (TaOx, TaSiy, and SiNz) as a
function of nitrogen content at the Ta(N)/SiO2 interface in the TaN/
Ta(N)/SiO2 samples.
4Ta2N + 5SiO2 → Si3N4 + 2Ta2O5 + 2Ta2Si
129.3 kcal/mol
,
(4)
(5)
26Ta + 20SiO2 + 10N2 → 5Si3N4 + 8Ta2O5 + 5Ta2Si
−609.6 kcal/mol
.
From a thermodynamic viewpoint, interfacial reac-
tions will only occur if elemental Ta is present. For
samples processed with a uniform TaN layer, initial proc-
essing involves the simultaneous deposition of elemental
Ta and N, favoring reaction (5). Samples processed with
a thin layer ( 20 Å) of Ta before N2 was introduced into
the deposition chamber would initially favor reaction (2).
However, with increasing N levels adjacent to the Ta
layer, resulting diffusion of N to the interface accommo-
dates reaction (5). It is not expected that the initial Ta
layer is fully converted to TaN or Ta2N as reactions of
those films with SiO2 are not thermodynamically favor-
able [reactions (3) and (4)] and would not account for the
observed decreases in TaNx and increase in adhesion as
discussed below. Note also that although Si3N4 was not
directly observed in the depth-profiling experiment due
While changing the type of barrier layer in a copper
interconnect structure clearly has a profound impact on
the interface fracture energy, changing the oxide type
also has a significant effect as shown in Fig. 9. Approxi-
mately 40% higher adhesion values were measured for
thermal compared to TEOS oxides. The failure in both
the TaN/thermal oxide and TaN/TEOS oxide systems
was along the barrier/oxide interface. This discrepancy in
adhesive properties may be rationalized in terms of the
J. Mater. Res., Vol. 15, No. 1, Jan 2000
Downloaded: 15 Mar 2015
209
http://journals.cambridge.org
IP address: 131.170.6.51

M. Lane et al.: Adhesion and reliability of copper interconnects with Ta and TaN barrier layers
near the crack tip experiences very large strains and in-
cludes the rupture process26 and a computational finite-
element (FEM) analysis utilizing a virtual internal bond
(VIB) rupture process.25 Both models have been found to
adequately predict the increase in interface fracture en-
ergy with increasing Al–Cu layer thickness. Similar re-
sults might be expected in copper interconnects although
the extent to which plastic dissipation increases interface
debond energy will vary due to differences in yield prop-
erties of Al and Cu thin films. Preliminary data showing
interface fracture energy as a function of copper layer
thickness is shown in Fig. 10. The interface fracture en-
ergy was increased by >400% as the copper layer thick-
ness was increased from 0.3 to 3.3 ␮m. On the basis of
modeling results of the aforementioned Al–Cu intercon-
nects,26 estimates for increases in adhesion with Cu layer
thickness may be made. Particularly, for a copper yield
stress of 180 MPa and a work of adhesion of 4 J/m2, the
relevant modeling length parameter, which scales with
the plastic zone size, is found to be 3.1 ␮m. Estimates
based on predictions of the EPZ model suggest that a Gc
of 12 times the work of adhesion (Go) or 48 J/m2 should
be anticipated for the copper thickness examined. FEM-
based computational approaches provide similar esti-
mates.27 However, a number of parameters in these
models, particularly the yield properties of the Cu layer
and the value of Go, still have a degree of uncertainty and
must be more fully addressed through direct experimen-
tal measurements. Current work is underway to clarify
these issues and measure adhesion values for a wide
range of Cu layer thicknesses with detailed modeling of
the plastic dissipation.
FIG. 9. Adhesion values for TaN on different oxides.
fracture properties of the oxides themselves and/or chem-
istry of the barrier/oxide interface. Segregation to the
interface may reduce the intrinsic work of fracture ac-
cording to24
Go ס Wad − ∑(⌬goi − ⌬gso)ci
,
(6)
where ci is the concentration of segregant per unit area of
interface and ⌬goi and ⌬gos are the Gibbs free energy of
the interface and segregant, respectively. The segregant
may be in the form of either the carbon precursor to the
silica film or moisture trapped in the glass which would
lead to hydrolytic weakening of the interface.17 In addi-
tion, the fracture properties of the oxides may be differ-
ent. Fully dense thermal oxides are expected to be more
fracture resistant than less dense TEOS oxides which
contain residual polymer precursors. Particularly,
debonds extending along an interface may seek out pores
in the SiO2 film and lower the macroscopic work of
fracture more than that predicted by a rule of mixtures.
While further studies are needed to more fully address
this issue, it clearly underlies the need to carefully ex-
amine all films of the bimaterial couple and indicates that
new silica deposition techniques must be carefully stud-
ied to ensure that adequate adhesion is maintained at
interfaces.
Plastic dissipation in surrounding ductile layers has
also been shown to have a large effect on the interface
debond energy of multilayer interconnect structures.9–11
Specifically, increasing plastic deformation was found
to increase interface debond energies by >300% in alu-
minum interconnects and drive the debond path from the
TiN/SiO2 interface into the SiO2. These results have
been modeled with both a mechanics of materials ap-
proach [embedded process zone (EPZ)], in which a zone
FIG. 10. Results showing the effect of copper layer thickness on the
adhesion of the TaN/SiO2 interface.
210
J. Mater. Res., Vol. 15, No. 1, Jan 2000
Downloaded: 15 Mar 2015
http://journals.cambridge.org
IP address: 131.170.6.51

M. Lane et al.: Adhesion and reliability of copper interconnects with Ta and TaN barrier layers
5. A. Cros, M.O. Aboelfotoh, and K.N. Tu, J. Appl. Phys. 67, 3328
IV. CONCLUSIONS
(1990).
The interface fracture energies of two important bar-
rier/dielectric layers were investigated. The TaN/SiO2 in-
terface was found to have approximately twice the
adhesion as the Ta/SiO2 interface, and failure in both
cases was along the barrier/SiO2 interface rather than the
barrier/copper interface. Furthermore, subcritical
debonding data show that the TaN/SiO2 interface is more
resistant to time-dependent debonding than the Ta/SiO2
interface. For samples produced with an initial 20 Å Ta
layer, increasing the N content adjacent to the Ta(N)/
SiO2 interface in the TaN/Ta(N)/SiO2 system was shown
to have a dramatic effect on critical adhesion values.
Evidence suggests that Ta2O5 and Ta2Si reaction prod-
ucts at the interface together with the diffusion of N into
the SiO2 layer results in a more fracture resistant struc-
ture. Effects of dielectric type and Cu layer thickness
were also investigated and are shown to have a signifi-
cant effect on adhesion. Broader implications are that
device reliability and yield may be significantly in-
creased by careful consideration of the chemistry at the
barrier/dielectric interface, the type of dielectric oxide,
and the thickness of the metallization layer.
6. E. Kolawa, J.S. Chen, J.S. Reid, P.J. Pokela, and M-A. Nicolet,
J. Appl. Phys. 70, 1369 (1991).
7. K. Holloway and P. Fryer, Appl. Phys. Lett. 57, 1736 (1990).
8. J.W. Hutchinson and Z. Suo, in Advances in Applied Mechanics,
edited by J.W. Hutchinson and T.Y. Yu (Academic Press, New
York, 1991), pp. 63–191.
9. M. Lane, R.H. Dauskardt, R. Ware, Q. Ma, and H. Fujimoto, in
Materials Reliability in Microelectronics VII, edited by
J.J. Clement, R.R. Keller, K.S. Krisch, J.E. Sanchez, Jr., and
Z. Zuo (Mater. Res. Soc. Symp. Proc. 473, Pittsburgh, PA, 1997),
pp. 21–26.
10. Q. Ma, J. Bumgarner, H. Fujimoto, M. Lane, and R.H. Dauskardt,
in Materials Reliability in Microelectronics VII, edited by
J.J. Clement, R.R. Keller, K.S. Krisch, J.E. Sanchez, Jr., and
Z. Zuo (Mater. Res. Soc. Symp. Proc. 473, Pittsburgh, PA, 1997),
pp. 3–14.
11. R.H. Dauskardt, M. Lane, Q. Ma, and N. Krishna, Eng. Fract.
Mech. 61, 141 (1998).
12. M. Lane, R.H. Dauskardt, Q. Ma, H. Fujimoto, and N. Krishna, in
Thin-Films—Stresses and Mechanical Properties VII, edited by
R.C. Cammarata, M.A. Nastasi, E.P. Busso, and W.C. Oliver
(Mater. Res. Soc. Symp. Proc. 505, Warrendale, PA, 1998), pp.
357–362.
13. A.G. Evans, and J.W. Hutchinson, Acta Metall. 37, 909 (1989).
14. A.G. Evans, M. Ruhle, B.J. Dalgleish, and P.G. Charlambides,
Metall. Trans. A 21A, 2419 (1990).
15. P. Gumbsh, Mater. Sci. Eng. A260, 72 (1999).
16. Q. Ma, J. Mater. Res. 12, 840 (1997).
17. G. Xu, M-Y. He, and D.R. Clarke, Acta Metall. (1999, in press).
18. S.M. Wiederhorn, J. Am. Ceram. Soc. 50, 407 (1967).
19. T. Michalske and B. Bunker, J. Am. Ceram. Soc. 70, 780 (1987).
20. M. Lane and R.H. Dauskardt (unpublished).
21. A. Bhatnagar, M.J. Hoffman, and R.H. Dauskardt, Am. Ceram.
Soc. (1999, in press).
22. J. Mencik, Strength and Fracture of Glass and Ceramics (Elsevier
Science Publishers, New York, 1992).
23. Handbook of Chemistry and Physics, 72nd ed., edited by D. Lide
(CRC Press, Boca Raton, FL, 1992).
ACKNOWLEDGMENTS
This work was supported by the Director, Office of
Energy Research, Office of Basic Energy Sciences, Ma-
terials Science Division of the United States Department
of Energy, under Contract No. DE-FG03-95ER45543,
and by the INTEL and Applied Materials Corps. M.L.
was supported in part by an Intel Foundation Fellowship.
24. J.R. Rice and J.S. Wang, Mater. Sci. Eng. A107, 23 (1989).
25. P. Klein and H. Gao, Eng. Fract. Mech. (1999, in press).
26. J.W. Hutchinson and A.G. Evans, Harvard University Report,
May 1999.
REFERENCES
1. S.P. Murarka and S.W. Hymes, Crit. Rev. Solid State Mater. Sci.
20(2), 87 (1995).
2. C-S. Ryu, A.L.S. Loke, T. Nogami, and S.S. Wong, Proc. IEEE-
IRPS 201 (1997).
27. M. Lane, A. Vainchtein, H. Gao and R.H. Dauskardt, Stanford
University (1999, unpublished results).
3. E.M. Zielinski, R.P. Vinci, and J.C. Bravman, J. Electron. Mater.
24, 1485 (1995).
4. L. Stolt and F.M. d’Heurle, Thin Solid Films 189, 269 (1990).
28. C.D. Wagner and D.M. Bickham, NIST X-ray Photoelectron
Spectroscopy Database (U.S. Department of Commerce, Wash-
ington, DC, 1989).
J. Mater. Res., Vol. 15, No. 1, Jan 2000
Downloaded: 15 Mar 2015
211
http://journals.cambridge.org
IP address: 131.170.6.51