Introduction
The relentless drive in the integrated circuit industry toward
greater packing density and higher speeds has served as the impetus for
optical lithography to reduce printed image sizes from 2 µm twenty
years ago to less than 0.5 µm today. This remarkable progress has
been made possible by improved lens quality, increases in numerical
aperture, improved resist processes, and the use of increasingly
shorter exposure wavelengths. Today's photolithography uses
wavelengths of 365 or 248 nm for imaging the smallest possible feature
sizes, thus employing aggressively low lithographic
k
factors of 0.5 to 0.6. However, as image
dimension requirements drop below 0.25 µm in the next few years, it
will be necessary to consider even shorter exposure wavelengths. An
obvious candidate for extension to shorter wavelengths is the 193-nm
laser line produced by the argon fluoride (ArF) excimer
laser. Indeed, the recent Semiconductor Industry Association roadmap
lists 193 nm as one of the options for printing 0.18 µm, along with
extensions of 248 nm. While each of the alternatives has its own
advantages and risks, only those of 193 nm are discussed here.
The change to 193 nm poses challenges and opens up new possibilities,
as new photoinduced processes take place at this shorter wavelength.
Specifically, optical materials that are nominally transparent have
weak absorption and also undergo laser-induced changes, and few organic
polymers are transparent enough to serve as single-layer resists.
On the other hand, efficient photoinduced cross-linking of polymers or
oxidation of silicon-containing polymers may enable new near-surface
resist processes.
This paper reviews the progress that has been made at MIT Lincoln
Laboratory toward a production-worthy 193-nm technology at
sub-0.25-µm resolution [1]. The
Lincoln Laboratory program has
addressed in parallel both the construction of a full-field prototype
exposure system (including the evaluation of optical materials and
coatings) and the development of photoresist processes (single layer,
antireflective layer, top-surface imaging, bilayers, and all-dry
resists).
Projection system
A full-field prototype 193-nm step-and-scan exposure system was
built by SVG Lithography (SVGL), and it was installed in Lincoln
Laboratory's clean room at the end of 1993. This system is the
world's first--and at present the only--large-field 193-nm prototype
exposure tool. It has a 193-nm 0.50-numerical aperture (NA)
lithographic lens with 4× demagnification, mounted on a commercial
SVGL Micrascan® II body. The slit size is 22 × 5 mm, and the
fully scanned field is 22 × 32.5 mm. In addition to its large
field, a distinctive feature of this tool is its off-axis alignment
subsystem. It provides an overlay accuracy of better than 40 nm
(3 ),
thus enabling the fabrication of devices by all-193-nm lithography. A
mix-and-match process involving other tools for noncritical levels also
requires distortion matching. The dynamic distortion in this system has
been measured, and the largest vectors are 77 nm over the full field.
The 193-nm Micrascan uses a catadioptric lens design incorporating both
reflective and refractive optical elements. The chromatic aberration of
this design is low enough to accommodate the natural 0.5-nm excimer
laser bandwidth [2]. This is in sharp
contrast to all-refractive
deep-UV (248-nm) lithographic lenses, which typically require special
modifications to the laser to deliver an optical bandwidth of less than
0.001 nm FWHM. The ability of the catadioptric lens to use the natural
bandwidth of the ArF laser considerably eases the demands on the laser
design.
The projection optics consist of an aspheric concave mirror used as a
4× reduction lens, with a cube beam splitter between the mirror and
the water surface. Several refractive optical elements are included in
the design for aberration control. All refractive elements, including
the cube beam splitter, are made of a selected grade of fused silica.
The cube beam splitter, which contains the greatest part of the optical
path length in fused silica, was tested extensively for transparency at
193 nm before fabrication, and had a bulk absorption of
 0.4% per
centimeter. The optical transmission of the 193-nm projection optics is
approximately half that of the equivalent 248-nm optics.
The laser in the 193-nm Micrascan was built by Cymer Laser. It is able
to produce its designed 6-W average power (linearly polarized) at 350
Hz with new optical elements. As the laser ages, the power level
characteristically drops, owing to increased levels of damage in bulk
optical materials and coatings within the laser. Replacement of damaged
optical elements essentially restores the laser to its original power
level. Typical lifetimes of optical components within the laser range
from several weeks to several months, with large variations among
materials from different sources.
Optical materials
In parallel with the construction of the 193-nm Micrascan, we have
been studying the performance of optical materials at 193 nm. Few
optical materials are transparent enough at 193 nm to enable the
fabrication of a high-quality all-refractive or catadioptric system as
required for a lithographic lens. High-purity synthetic fused silica
and crystalline calcium fluoride are probably the only practical
choices [3], with fused silica
having the edge for reasons of cost
and existing processing infrastructure. For purposes of insertion into
193-nm lithography, the ideal optical material should be fully
transparent, and should remain unchanged after several billions of
pulses. In practice, fused silica has an absorption coefficient of
0.005 to 0.10 cm
with large variations from grade to
grade and even from sample to sample. Since the temperature coefficient
of the index of refraction of fused silica at 193 nm is
2
× 10 °C,
the magnitude of the allowable
absorption coefficient is determined by system considerations,
including laser power and the specific design of the projection optics.
A value of 0.10 cm
is clearly unacceptable, but the
range 0.001 to 0.01 cm
requires careful analysis with
respect to each system design. It should be noted that the physical
origin of the absorption at 193 nm is not yet clearly understood. With
metallic impurities reduced to the sub-ppm level, the main candidates
are Cl impurities (residue from the starting material
SiCl ),
hydroxyl (OH) impurities, and imperfect
stoichiometry, i.e., excess oxygen or oxygen deficiency
[4]. Similar
considerations apply to scattering losses, the most important one being
Rayleigh scattering caused by density fluctuations. The fundamental
lower limit for this process at 193 nm may be in the range
10 -10
cm
[5].
Beyond the amount of initial absorption and scattering, the optical
material should be resistant to radiation-induced changes
("damage") over the practical lifetime of a projection system. The
principal modes of laser-induced damage in fused silica are color
center formation and optical compaction [6].
The color centers are E'
centers, consisting of an unpaired electron on a silicon atom,
accompanied by oxygen vacancy. Formation of E' color centers leads to
absorption at
215 nm,
with increased optical absorption also at 193
nm. Compaction manifests itself as reduced thickness accompanied by an
increased index of refraction [6,7].
The net result is a decrease in
the optical path, which in a lithographic lens would cause wavefront
aberrations and loss of image quality. The two types of damage seem to
depend on different properties of the fused silica; a sample with good
resistance to color center formation does not necessarily have good
resistance to compaction. Typical samples seem to be more sensitive to
compaction than to color center formation. Furthermore, recent samples
of fused silica exhibit saturationlike behavior of the color centers,
with laser-induced absorption coefficients saturating at
0.001
cm, for
1-5 mJ/cm²/pulse. This added value
is less than the initial absorption coefficient, or
0.002
cm. Thus, it is
expected that compaction, and not color
centers, sets the limit on allowable power densities within the
lithographic lens.
At low values of compaction in fused silica, the compaction increases
linearly with the number of pulses and as the square of the peak
intensity I within a laser pulse. This behavior is the
signature of a two-photon process, whose coefficient
=  I
is linear in I. (The
compaction, being proportional to the product of I and
,
behaves as I².) The
coefficient is
intrinsic to the material, and has been measured to
be 2 ×
10 to 2 ×
10
cm/MW [8,9]. Because excimer laser
pulses are extremely short
(typically between 5 and 20 ns FWHM), high peak intensities are
generated by pulse trains of modest average intensities. Consequently,
may be non-negligible. For instance, at a fluence of
10 mJ/cm² and pulse duration of 20 ns, I =
0.5 MW/cm², and
10-10
cm. Estimates have
been made that a 10-year lifetime of a 193-nm fused silica lithographic
lens can be achieved with the best existing grades of fused silica if
the energy density per pulse is kept at or below 1 mJ/cm²
within any element of the lens. This restriction forces the use of
relatively sensitive photoresists; dose requirements of 25
mJ/cm² or less would be needed to achieve a
50-wafer-per-hour throughput on 8-inch wafers. Higher-dose resists
could be accommodated with future advances in damage resistance of
fused silica or possibly with substitution of
CaF in
critical elements of the lens.
Calcium fluoride (CaF)
does not suffer from compaction
because of the crystalline nature of the material. It is susceptible to
color center formation, but this susceptibility is apparently due only
to defects and impurities in the material. Current high-purity
CaF
is significantly better than that available a decade
ago. In fact, it is comparable to or better than fused silica in its
resistance to damage [10]. The main
technical barriers against
extensive use of calcium fluoride as a 193-nm lens material are
residual stress birefringence and lack of experience with its grinding
and polishing characteristics.
It should be noted that the results discussed above have been obtained
from experiments performed off-line. To date, there is no evidence that
the Micrascan optics have suffered the kind of bulk damage seen on
samples at higher fluences.
The properties of bulk optical materials are but one item in the list
of optics-related topics that have to be addressed at 193 nm. Others
include the behavior of dielectric coatings [3],
pellicles [11],
and photoinduced organic deposits on surfaces. These are also the
subject of ongoing studies at Lincoln Laboratory.
Photoresists
The most difficult challenge in bringing 193-nm lithography to
full manufacturing use may be the development of a robust photoresist
process. The resins which are typically used for I-line (365-nm) and
248-nm photoresists, e.g., novolac and poly(hydroxystyrene), have
absorption depths of 30 to 50 nm at 193 nm, and therefore are far too
opaque to be used in single-layer resists at that wavelength. Since
methacrylates are semitransparent at 193 nm, these polymers can serve
as the basis for 193-nm single-layer resists. Acid-catalyzed conversion
of t-butyl methacrylate (tBMA) into methacrylic acid (MAA) provides the
chemical underpinning for several versions of such resists, developed
under a collaboration between MIT Lincoln Laboratory and the IBM
Almaden Research Center [12,13].
These systems contain resins synthesized by free-radical solution
terpolymerization of tBMA (fraction x), methylmethacrylate
(MMA, fraction y), and MAA (fraction z), and meet
the transparency and thermal stability requirements for single-layer
resist applications. A typical composition has x/y/z
0.4/0.4/0.2, and the glass transition temperature
(T )
is in the range 140-160°C. The
absorption coefficient of these base resins at 193 nm is around 0.08
µm.
Figure 1 shows the
contrast curve of a two-component resist including such a terpolymer
and a photoacid generator.
 Figure 1
The solubility of these terpolymers can be controlled by variations in
MAA content. For example, when the MAA concentration is
20 mole
percent, the exposed films are aqueous-base-soluble only after
exposure, whereas at MAA fractions approaching 30 mole percent, the
unexposed resin itself becomes aqueous-base-soluble.
Since they are photosensitive also at 248 nm, the methacrylate-based
resists have the added advantage that they can serve as effective
dual-wavelength resists. Indeed, the dual-wavelength property has
enabled their development at 248 nm, while the 193 Micrascan has been
undergoing its own tests and modifications.
The main drawback of these photoresists is their low etch resistance in
subsequent plasma processing steps, especially in the chlorine-based
chemistries required for metal etching. Increased etch resistance can
be obtained by incorporation of high-carbon-content polymers. These
cannot be aromatics because of limitations of transparency. The
alternative is copolymerization with polymers containing pendant
alicyclics, such as adamantyl methacrylates
[14]. Resultant
tetrapolymers have been synthesized and have been shown to have
significantly lower etch rates, 1.4× that of novolac, compared to
2.2× that of novolac for the terpolymers. Another option under
study for increasing etch resistance involves the development of three
component systems (including the addition of dissolution inhibitors
which, because of their high carbon-to-hydrogen ratio, also reduce the
plasma etch rate of the resist system).
Top-surface-imaged (TSI) resists have been developed at longer
wavelengths as an alternative to bulk-imaged resists
[15]. Their main
potential advantages are a larger depth of focus, especially for
sub-0.25-µm lithography, where the thickness of single-layer resists
becomes comparable to the theoretical optical depth of focus, and the
elimination of the need for antireflective layers. TSI processes
operate via area-selective in-diffusion of a silyl amine into a
phenolic polymer to form a silyl ether. Once the silicon has been
selectively incorporated, the latent image is developed in an
anisotropic oxygen plasma etch. At 193 nm, the simplest TSI resist
scheme is a positive-tone process, based on photo-cross-linking of the
single component poly(vinylphenol) (PVP), followed by selective
silylation of the unexposed areas and plasma etching
[16].
The silylation is performed with dimethylsilyldimethylamine (DMSDMA),
at a temperature of 90°C and a pressure of 25 torr for 30 to 75
s. The un-cross-linked PVP areas incorporate a controlled amount of
silicon, whereas the cross-linked areas do not. The wafers are then
dry-developed in a high-ion-density helicon etcher in an oxygen-based
plasma. The oxygen reacts with the silylated areas to form
SiO,
which acts as an etch mask in the unexposed areas.
On the basis of initial materials characterization, we chose to use
DMSDMA to silylate PVP because of its high vapor pressure and small
molar volume, which allows it to diffuse rapidly at low temperatures.
This is useful for our high-molecular-weight PVP resist, which has a
higher T,
180°C,
than many
multiple-component chemically amplified TSI resists. One effect of this
high T
is slower diffusion rates, which can be
minimized by going to lower-molar-volume silylating agents.
The overall lithographic performance of the silylation process
depends on both the silylation step (i.e., the silylation mask shape)
and the dry-development step (i.e., selectivity, vertical/lateral etch
rates, and amount of over-etch). Figure 2
demonstrates the excellent linearity obtained with the 0.5-NA 193-nm
Micrascan and the TSI process outlined above. For a wafer temperature
in the helicon etcher of -70°C, linearity is maintained down to 0.20
µm for both 25% and 100% over-etch. For an etch temperature of
30°C, linearity is maintained down to 0.20 µm for a 25% over-etch,
but only to 0.30 µm for a 100% over-etch. This is due to an
increased isotropic etching component at 30°C. The
grating-to-isolated-line bias for a 60-s silylation time decreases from
30 nm for the 25% over-etch, to 5 nm for the 100% over-etch. The
linearity of gratings, but not of isolated lines, was extended down to
0.175 µm by reducing the silylation time from 60 to 30 s.
 Figure 2
Figure 3 is a graphic representation of the
exposure-dose matrix for 0.25-, 0.20-, and 0.175-µm gratings for a
30-s silylation time and a 25% over-etch. At best dose the depth of
focus is 1.6, 1.0, and 1.0 µm, respectively. However, this maximum
depth of focus is obtained at higher doses as the feature size
decreases. Such behavior is in qualitative agreement with aerial image
simulations and corresponding exposure-defocus plots. The simulation
predicts an 8.5% increase in best dose when the grating lines are
reduced from 0.25 to 0.20 µm. This result is in close agreement with
the 10% dose increase measured experimentally. It should be noted that
the data tabulated above are based on printed features which have near-
vertical sidewalls. Figure 4 shows
representative SEMs of 0.20-, 0.175-, and 0.15-µm resist features
obtained as described above.
 Figure 3
Conclusions
Fundamental optics principles have motivated the trend in
photolithography toward shorter wavelengths. One of the main
alternatives in the printing of 0.18-µm devices, whose mass
production is expected to start around the year 2001, is lithography at
the 193-nm wavelength of ArF excimer lasers. The transition from 248 to
193 nm is viewed as largely evolutionary, following the trend from 436
to 365 to 248 nm. Nevertheless, new infrastructures must be developed,
mainly in the areas of optical materials and photoresists. Under the
MIT Lincoln Laboratory program, a full-field 193-nm prototype
step-and-scan system has been built and is currently undergoing tests
and modifications of its subsystems. Bulk optical materials and
coatings are continuously being tested for absorption and resistance to
laser-induced damage. First-generation single-layer resist systems
(including antireflective layers [17])
and TSI processes have been
developed. Initial experimental results indicate that key requirements
of such resists, such as 0.18-µm resolution and 1-µm depth of
focus, can indeed be met. Other resist schemes have also been
demonstrated, although they are technically less mature than TSI and
therefore have not been detailed in this paper. These include bilayer
resists with polysilyne imaging layers [18]
and plasma-deposited,
plasma-developed counterparts of bilayers and of TSI ("all-dry
resists") [19,20].
However, significant additional work is
required in order to transform the current proof-of-concept
demonstrations into a robust manufacturing technology. Critical
associated technologies must be developed and tested as well. These
include overlay, metrology, and mask writing--all at the dimensions
dictated by the insertion point of the 0.18-µm critical dimension.
Some of these technologies, such as overlay, are not specific to the
193-nm actinic wavelength, while others, such as materials for
attenuating phase shifting masks, are. They must all be brought to a
significant level of maturity at an accelerated pace to permit
insertion into pilot lines within the next three years.
Acknowledgments
The work on single-layer resists was performed under a
collaboration with the IBM Almaden Research Center, mainly with R. D.
Allen, R. A. DiPietro, D. C. Hofer, and G. M. Wallraff. We thank M.
Hibbs (IBM Essex Junction) who, as a Sematech assignee at Lincoln
Laboratory, worked closely with us in the evaluation of the performance
of the 193-nm Micrascan. We also thank M. W. Horn for helpful
discussions, and D. Downs, L. Eriksen, and C. Marchi for their
technical assistance. This work was sponsored by the Defense Advanced
Research Projects Agency's Advanced Lithography Program. Opinions,
interpretations, conclusions, and recommendations are those of the
authors and are not necessarily endorsed by the United States Air
Force.
Micrascan is a registered trademark of SVG
Lithography Systems, Inc.
References
Received February 9, 1996; accepted for publication September 14,
1996
|
HTML
Figure 1
Figure 2
Figure 3