Practical Fab Applications of X-ray Metrology
Dileep Agnihotri, Joseph Formica, Jesus Gallegos, Jeremy O'Dell
Jordan Valley Semiconductors, Inc.
2211 Denton Drive, Suite A, Austin, TX 78758-4532 USA
Abstract. X-ray metrology techniques have emerged from the laboratory to meet the challenges of the production fab.
X-ray Reflectometry (XRR) and X-ray Fluorescence (XRF) are non-destructive methods to probe film thickness,
density, roughness, and composition. What has kept x-ray metrology out of the fab for so long is large spot size, low
throughput, and difficult analysis. Combining small-spot XRR and XRF techniques, today's x-ray metrology covers a
wide range of applications from front-end to back-end, transparent to metal, ultrathin to micron-thick, on blanket and
patterned wafers with speed and with automated data analysis. This work focuses on copper barrier applications.
INTRODUCTION
With copper yields still lagging behind aluminum,
x-ray metrology, notably XRR and XRF, enables
many copper-related issues to be explored. Copper
x-ray metrology applications include: simultaneous
seed and barrier characterization, electrochemical
deposition (BCD) layer thickness mapping, systematic
void identification, chemical-mechanical polishing
(CMP) dishing and erosion evaluation, top barrier and
interfacial film
measurements,
and low k
characterization, including porosity and the correlation
of dielectric constant with measured layer density.
In XRR, x-rays strike a thin film at a glancing
angle. Below the critical angle, x-rays are totally
externally reflected. Beyond the critical angle, x-rays
penetrate the film and interference patterns result from
reflections at various interfaces. From such patterns
come thickness, density, surface and interface
roughness information. This technique can be used
with single films or complex stacks of all types
(transparent, opaque, or mixed; amorphous,
poly crystalline, or epitaxial) with thicknesses from
~5A to ~3000A. XRR can distinguish multiple layers
containing the same elements and is a first-principles
technique requiring no standards1.
In XRF, source x-rays eject inner-shell electrons
from the film. Outer-shell electrons take their place,
and photons are emitted whose energy corresponds to
the orbital transition. XRF is an elemental analysis
technique primarily used with metals and elements
such as Ge or P. It cannot distinguish between
different layers containing the same element. The
practical thickness range covers several hundred
angstroms to several microns. Calibration curves relate
XRF intensity to thickness and composition.
XRR can characterize copper seed and barrier
layers quickly and simultaneously, while thicker BCD
copper films can be characterized by XRF. XRR can
probe thin copper oxide and tantalum silicide layers,
related to voiding and adhesion issues. Multilayer
barriers, such as Ta/TaN, can also be probed with the
ability to resolve the individual layers. X-ray
metrology can show the relationships between layer
thickness, density, roughness, and stoichiometry (xvalue) and various process parameters (e.g. time,
temperature, gas flow rates, and RF power settings).
Since XRR can characterize both transparent and
metal layers, it is possible to simultaneously measure
top barriers (e.g. SiC, SiCN), Cu, and bottom barriers
(e.g. Ta/TaN). XRR can even characterize the ultralow
density, ultrathin layer between the Cu and SiC, which
correlates to CMP clean chemistry. When the Cu
thickness exceeds the range for XRR measurements,
XRR can still provide information about top and
bottom barriers while the Cu thickness is instead
measured by XRF.
CP683, Characterization and Metrology for VLSI Technology: 2003 International Conference,
edited by D. G. Seiler, A. C. Diebold, T. J. Shaffner, R. McDonald, S. Zollner, R. P. Khosla, and E. M. Secula
© 2003 American Institute of Physics 0-7354-0152-7/03/$20.00
660
CURRENT METROLOGY
CHALLENGES
BARRIER REQUIREMENTS
The basic role of the barrier is to prevent the
diffusion of Cu into the dielectric material (Figure 1).
Likewise, the barrier may also prevent unwanted
migration of contaminants from the dielectric (e.g.
fluorine) into the Cu. Barrier materials should have
chemical and thermal stability, low resistivity, and
good adhesion properties. Grain boundary diffusion
can be prevented if the barrier material can be
deposited as an amorphous material. Materials
explored as diffusion barriers for Cu include: Ta, TaN,
Ta/TaN, TaSiN, TiSiN, and WN.
As stated in the ITRS: 2002 Update4, one of the
"Near Term" (>65^im, through 2007) "Grand
Challenges" is the measurement of complex structures.
"Reference materials and a standard measurement
methodology are required for new, high k gate
dielectrics and capacitor dielectrics with interface
layers, thin films such as interconnect barriers and low
k dielectric layers, and other processes. Optical
measurement of gate and capacitor dielectrics averages
over too large an area and needs to characterize
interfacial layers. The same is true for measurement of
barriers. A measurement methodology is therefore
needed for complex material stacks and interfacial
properties including physical and electrical
properties." Non-x-ray film thickness measurement
techniques include four-point probe and optoacoustic.
For thin barrier films, x-ray metrology techniques
offer a number of advantages over non-x-ray
techniques.
Four-point probe techniques can characterize the
sheet resistance of a film. If film thickness is the
parameter of interest, calibration curves must be
created using secondary techniques. For multilayer
metal films, a total sheet resistance can be obtained,
but there is no possibility to extract information about
the individual layers. Though a simple and common
technique, it is somewhat limited to blanket, single
metal films, and it is contacting and destructive.
FIGURE 1. Copper interconnect structure.
With each technology node, barrier layers are
becoming increasingly thinner (Table 1). One
processing challenge is to deposit thinner, conformal
barrier and seed layers over high-aspect ratio
structures. To achieve the roadmap milestones,
conventional physical vapor deposition (PVD)
methods are expected to yield to ionized PVD methods
which will give way to chemical vapor deposition
(CVD) and ultimately atomic layer deposition (ALD)
methods2. ALD methods offer the potential to
construct barriers one atomic layer at a time. With
today's XRR, barrier films can be characterized down
to the 10A level, which makes XRR uniquely qualified
to characterize ultrathin ALD barriers.
In optoacoustic techniques, short laser pulses create
sound waves which propagate and interact at the film
interfaces. Such measurements produce plots of the
change in reflectivity versus time. To extract thickness
information, there is a rarely stated assumption of
constant material density which underlies the
modeling of optoacoustic techniques5. Yet, as films
become thinner, film density decreases from the bulk
value. This is seen in the XRR results from a series of
ultrathin TaN barriers (Figures 2-4). Depending on the
crystallographic phase, the bulk density of TaN is in
the range 14.3 - 16.3 g/cm3. XRR measurements of the
sub-50A films described in Figures 2-5 show nearly
half those values. XRR provides independent thickness
and density information, so no constant density
assumptions are required. The wafers depicted in
Figures 2-4 were processed in the same chamber, and
their thickness maps possess the same qualitative,
signature shape. Figure 5 demonstrates the sensitivity
TABLE 1. Barrier/cladding thickness (for Cu intermediate
wiring) from the ITRS: 2002 Update3.
Year of Production
2003
2004
2005
2006
2007
2010
2013
2016
Barrier Thickness [A]
120
100
90
80
70
50
35
25
* All data presented here were collected on a JVX 5200 X-ray
Metrology Tool from Jordan Valley Semiconductors, Inc. For more
information about the JVX 5200, visit www.jordanvalleysemi.com.
661
of the XRR technique to such a difference in TaN
density.
Reflectivity
2.4
2* IfcN ite, XRR
FIGURE 5. Simulated XRR reflectivity curves of a seed Cu
/ TaN barrier stack demonstrating the sensitivity of the XRR
technique to TaN density. All other film properties are kept
the same in the two curves. The thickness, density, and
roughness values shown above with the TaN density of 7.70
g/cm3 were obtained from an actual XRR measurement.
!fl
.0:
ai mm m
3,
2.8
Theta, deg.
**Nf m, J* seem m
map witfi
Tss34J7±L44 A,
FIGURE 6. Measured Ta and TaN density as function of
target barrier thickness for different process N2 flow rates.
The density-sensitivity of the XRR technique is
shown comparing 50A metal oxide CVD (MOCVD)
TiSiN films processed in single and double deposition
passes (Figures 7-8). In each case, 49-point XRR
measurements were made, and the films were modeled
as bilayers. The resulting thickness maps are
remarkably similar with sub-Angstrom agreement in
the average thickness values. The differences emerge
in the film densities. For the film deposited in a single
pass, there is a distinct density difference between the
two modeled layers. For the film deposited in two
passes, there is closer agreement in the densities of
each layer. In addition to simple linear density
grading, some TiN and NiSix films have been
measured with complicated alternating density layer
structures which XRR can successfully probe.
FIGURE 4. TaN film. XRR thickness map with notch down.
49-pt averages. T=18.35±0.84 A. D=8.34±0.33 g/cm3.
Studies of TaN barriers have shown that film
density can be correlated with process parameters,
such as nitrogen flow rates. Figure 6 shows that higher
nitrogen flow rates produced lower density films.
Also, for a given flow rate, the thinner the barrier, the
less dense the film.
662
Reflectivity
5
43
'vu
% tf*A
«,
*, g i
Film
SICN
Int-1
Cu
TaN
M-2
Si
T[A]
1 R[AJ
582.7
18J
1034.6
242.8
12.7
10.4
3.0
1,93
1.34
8.85
15.31
14.78
Subs. 2.33
11.4
3.9
1.0
1,5
.to,
* ft If * A M
Int-1 is a SiCN-Iike interfacial film.
Int~2 is a TaN-llke interfaeia! film.
0.4
FIGURE 7. One 50A-pass MOCVD TiSiN. XRR thickness
map with notch down. Modeled as two films (upper and
lower). 49-pt average thickness: 29.74±l.llA upper,
26.07±1.05 A lower. 49-pt average density: 4.46±0.12 g/cm3
upper, 3.68±0.10 g/cm3 lower.
0.8
1.2
1.6
2.0
2.4
2.8
Theta, deg.
FIGURE 9. Seed copper with lower and upper barriers. This
is an example of the ability of XRR to resolve ultrathin
interfacial films.
BENEFITS OF X-RAY METROLOGY
The XRR technique is based on first principles. No
calibration standards are required.
Small-spot XRR and XRF techniques can provide
high spatial resolution scans or maps to study acrosswafer property variations. Small-spot XRF typically
provides a sub-30um spot. The footprint of an XRR
measurement is larger owing to the glancing-angle
nature of this technique. Small-spot XRR typically
irradiates areas of 100 urn by 3-6mm, but they can be
as small as lOOum by 500um. Small-spot
measurements are not limited to blanket films, but may
also be performed on pads and scribe lines.
FIGURE 8. Two 25A-passes MOCVD TiSiN. XRR
thickness map with notch down. Modeled as two films
(upper and lower). 49-pt average thickness: 29.42±1.69A
upper, 26.17iO.84 A lower. 49-pt average density: 4.18±0.08
g/cm3 upper, 4.34±0.09 g/cm3 lower.
The ability of XRR to characterize complicated,
mixed stacks with ultrathin interfacial layers is
demonstrated in Figure 9. This capability can be
important during process development and in
production to monitor oxidation, residual clean,
adhesion, and stress-relief layers which can impact
downstream performance. XRR also offers the ability
to resolve emerging stacked barriers like Ta/TaN.
XRR also provides the ability to resolve ultrathin
interfacial layers or oxidation layers and to
characterize graded films.
Since the refractive index of the x-rays used in
XRR measurements is close to unity, XRR does not
usually suffer from thickness correlations that plague
many optical methods (e.g. oxide-nitride-oxide stacks)
or from birefringent materials (e.g. organic antireflective coatings).
Furthermore, x-ray metrology can measure Cu seed
and barrier simultaneously, since it is neither always
possible nor desirable to perform metrology steps on
the individual films. For up to a micron of Cu over
barrier, XRR measurements can provide density and
roughness information about the Cu, but not thickness
information, which can instead be obtained by XRF. In
that case, it may still be possible to characterize the
underlying barrier thickness by XRR. Combined XRR
and XRF metrology offers the possibility to measure
the range of film thicknesses from 10A to 10 urn.
The modeling of XRR is fast and fully automated.
XRF analysis of thin films is also fully automated to
obtain either relative XRF intensities for elements of
interest or thicknesses based on calibration standards.
Tools with combined XRR and XRF metrologies
enable XRF thickness calibration standards to be
characterized by XRR The combined metrologies also
provide a non-destructive way to determine x-values in
binaries such as WSix, CoSix, or NiSix.
663
Reliability and maintainability are also of prime
concern. Many conventional XRR tools employ highpower sources known as rotating anodes. The metal
anode of this type of source spins at thousands of
revolutions per minute. This rotation allows the anode
to operate at higher power settings than sealed-tube
sources without damaging the anode material. These
systems require water cooling, high vacuum, and
periodic maintenance (changing seals and filaments,
balancing anodes, etc.). Technological advances in
x-ray sources and optics now produce small x-ray
spots with low power, low maintenance, and high flux.
Furthermore, innovative x-ray metrology tool designs
minimize the number of moving mechanical parts
which are subject to wear.
CHALLENGES OF PRACTICAL X-RAY
METROLOGY FOR THE FAB
Macroscopic spot sizes, of the order of tens of
millimeters in width or diameter, are typically
employed in conventional x-ray reflectometers and
spectrometers. While suitable for some blanket wafer
applications, mapping capabilities are limited, and
patterned wafer analysis is impractical. Productionoriented micro-spot XRR and XRF systems are
available which enable in-line, patterned-wafer
measurements on pads and scribe lines.
Laboratory x-ray reflectometers and spectrometers
are often equipped with off-line data analysis software
requiring expertise and interaction to obtain results
from measured data. Production-oriented systems
allow process engineers to create recipes which
combine data collection and appropriate applicationspecific models for on-line data analysis. Once these
recipes are developed, they can be run by operators
with minimal training.
CONCLUSIONS
XRR and XRF are x-ray metrology techniques with
many practical fab applications. Modern x-ray
metrology tools have overcome most of the limitations
of spot-size, automation, and throughput which have
historically limited their use for in-line production
monitoring. X-ray metrology, especially XRR, is
capable of addressing the near- and long-term ITRS
barrier challenges.
Excessive film roughness (typically >30A) can
hamper XRR analysis by "washing out" fringe
information in the XRR data. In cases where film
roughness is great, it may still be possible to
characterize film thickness using XRF rather than
XRR.
Throughput has been a longstanding issue for x-ray
metrology
techniques.
Conventional
x-ray
reflectometers found in analytical and research
laboratories employ goniometers for precise motion of
the wafer and x-ray detector or the x-ray source and
detector. Collection of high-quality data from a
conventional reflectometer typically takes from
minutes to hours. Today's in-line XRR metrology
tools employ advances in x-ray source, optics, and
detector technologies which reduce measurement
times to seconds for most applications. Conventional
XRF systems frequently measure wafers under
vacuum, which increases tool complexity and wafer
cycling time. Wavelength-dispersive (WD) XRF
systems also employ goniometers and analyzer
crystals to diffract the wavelengths of interest onto
detectors. The benefits of WDXRF include narrow
XRF peaks (high energy resolution) and the ability to
measure very light elements (down to boron) when
vacuum is used. Energy-dispersive (ED) XRF systems
employ detectors with high energy discrimination and
fast electronics to simultaneously sample the XRF
spectrum. While EDXRF energy resolution may not be
as high as can be obtained by WDXRF, it is more than
adequate for most semiconductor metrology
applications.
REFERENCES
1. Krassimir N. Stoev and Kenji Sakurai, "Review on
grazing
incidence
X-ray
spectrometry
and
reflectometry," Spectrochimica Acta Part B 54, 41-82
(1999).
2. Peters, L., "Finding the Ultimate Copper Barrier and
Seed," Semiconductor International 24(8), 23 (2001).
3. International Technology Roadmap for Semiconductors
(ITRS):
2002 Update,
Semiconductor Industry
Association, San Jose, 2002, pp. 75-76.
4. International Technology Roadmap for Semiconductors
(ITRS):
2002 Update,
Semiconductor Industry
Association, San Jose, 2002, pp. 13-14.
5. Contestable-Gilkes, D., Merchant, S. M., and Oh, ML,
"Measurement of Metal Film Thickness for Copper
Interconnects," Semiconductor Fabtech 15, 157-162
(2001).
664