|
INTRODUCTION
TO MICROLITHOGRAPHY
By
David Collier and Wayne Pantley, Alpine Research Optics
David
Collier is President and Wayne Pantley is Sales Manager of Alpine Research
Optics, 3180 Sterling Circle, Boulder, CO 80301, 303-444-3420, FAX
303-444-1686, E-mail AROcorp@AROcorp.com, http://www.optics.org/arocorp/
Integrated
circuits, and particularly microprocessors and memory chips, are now found in
everything from desktop computers to automobiles, cellular phones and even
coffee makers. A key step in the creation of integrated circuits is a process
called optical microlithography. This article reviews the basic principles of
microlithography and discusses the present and future challenges faced by
suppliers of lasers and optics for this application.
Microlithography
Basics
|
Intel Pentium®
Processor die. Photo Courtesy of Intel Corporation.
figure 1
|
An
integrated circuit (IC) consists of various electronic components (primarily
transistors) constructed as a single, monolithic piece of semiconductor (figure
1). The detailed structure of these devices is built up layer by layer, in a
process called microlithography. The first step in microlithography is to coat
a semiconductor wafer (typically silicon) with a light sensitive layer, called
a photoresist. This surface is then exposed to light (from a laser or lamp)
that has passed through a reticle (mask). This process is analogous to
traditional masked based excimer laser micromachining. The two differences are
that the laser is used to expose the photoresist, rather than directly ablate
material, and the size of the features produced is dramatically smaller.
|
Microlithography
schematic.
figure 2
|
Since
the silicon wafers are typically between 100 to 200 mm in diameter (with 300 mm
wafers coming soon), and a typical IC is only about 6 mm square, literally
hundreds of identical IC's can be produced on a single wafer. Each reticle
contains the pattern for just a single IC. After a single exposure of the
reticle is made, a highly accurate positioning system is therefore used to move
to the next location on the wafer for exposure of another IC pattern (figure
2). This process is called step and repeat, and systems that perform this
function are called optical wafer steppers.
Once
the entire wafer is completely exposed to a particular reticle pattern, the
photoresist is chemically developed. Exposed parts are removed by this
development, resulting in a pattern of protected and unprotected surface,
corresponding to the original reticle pattern. Depending upon the specific
characteristics of the circuit under construction, the next step(s) includes
processes such as ion implantation, chemical etching, diffusion, oxidation and
deposition. In ion implantation for example, ions are accelerated into the
wafer; the ions modify the electrical characteristics of the silicon to which
they are implanted. In chemical etching on the other hand, silicon is actually
removed from the substrate for subsequent replacement by a metal. Once a given
layer of device features is produced using one or more of these techniques,
another photoresist layer is added, and the entire process is repeated as many
as 20 times to build the complete chip circuitry.
|
Moore's Law.
Microprocessor transistor density has doubled every two years since 1965.
figure 3
|
Intel's
8088 microprocessor chip, introduced in 1979 and used in the first IBM personal
computer, contained 29,000 transistors. Today, Intel's Pentium II
microprocessor contains 7.5 million transistors and is only about twice the
size. By 2010, a microprocessor with the same area is projected to contain over
one billion transistors. Achieving this mammoth increase in component density
requires that the minimum feature size produced through the microlithography
process continue to shrink. As can be seen in the graph (figure 3), the
industry has moved from 10 µm feature sizes (or linewidths) in the early 1970's
to 0.25 µm linewidths in today's most advanced processes.
Microlithography
Lasers
The
ultimate limiting factor in microlithography linewidth is the wavelength of the
light source used for exposure. Due to diffraction, the minimum image feature
size of an optical system is linearly related to wavelength; thus, cutting the
wavelength in half cuts the minimum possible linewidth in half.
At
present, many production stepper systems still utilize mercury arc lamps
sources. With a wavelength of 365 nm, they have been able to deliver IC
linewidths of 0.35 µm. However, mercury arc lamps don't represent the future of
microlithography for several reasons, the most obvious of these being their
wavelength itself. Another drawback of arc lamp sources are that they emit a
relatively broad spectrum of light, which causes problems with tight focusing.
For
these reasons, stepper system designers have turned to lasers. In particular,
the KrF excimer laser, with output at 248 nm, has now become the source of
choice for advanced microlithography systems used to produce 0.25 µm features.
Typical output specifications for microlithography KrF lasers from
manufacturers such as Cymer and Lambda Physik are 10 W of average power
delivered at 1 kHz repetition rate.
The
300 µm natural spectral linewidth of the KrF excimer laser must be narrowed to
a fraction of a picometer for use with all refractive, high numerical aperture
objectives. In response, laser manufacturers have achieved spectral linewidths
in the 0.6 to 0.8 µm using techniques such as intracavity diffraction gratings
and etalons.
Reliability,
mean time between failure (MTBF), mean time to repair (MTTR) and scheduled
maintenance downtime are also significant considerations in this application,
since production downtime in a wafer fabrication line can cost as much as
$100,000 per hour. Overall lifetime is important as well, because steppers are
typically operated virtually 24 hours a day. Given the repetition rate and
overall duty cycle, this translates into a total output of well over one
billion pulses per year.
Internal
corrosion caused by the laser gases has traditionally been the limiting factor
in excimer laser lifetime. To eliminate corrosion, manufacturers have turned to
all metal/ceramic construction for their internal tube components. As a result,
a typical KrF tube lifetime specification is now in the 3 to 5 billion pulse
range. Reduced corrosion also lengthens the life of a gas fill to 100 million
pulses, and brings the window service interval to over 5 billion pulses.
Through the use of all solid state pulse power supplies to drive these lasers,
both power supply lifetime and control of total dosage (wafer exposure) have
been greatly increased.
These
KrF lasers are expected to take the microelectronics industry to linewidths of
0.18 µm. Extending optical microlithography the next step down to 0.13 µm will
most probably require a move to ArF excimer lasers, which operate at a
wavelength of 193 nm. Unfortunately, ArF has a lower gain than KrF, and
therefore produces less output power. Thus, the main thrust of development in
ArF lasers for microlithography is focused on raising power, as well as
improving output stability, narrowing linewidth and increasing reliability. ArF
products now being offered for investigational use deliver 5 W of output power
at a 1 kHz.
Optics
Fabrication
Achieving
the incredibly small linewidths mentioned previously requires an optical system
which is essentially "perfect." It must have optical surfaces that conform
exactly to design specifications and produce minimal scatter. This translates
into typical production specifications of l/10 flatness and a surface quality
(scratch and dig) specification of 10-5. Furthermore, optics (both bulk
material and coatings) must be able to withstand exposure to multi-billion
pulse counts.
Compared
to other types of micromachining, microlithography optics are subjected to
prolonged exposure to relatively moderate fluences. With regards to
reliability, researchers at Lawrence Livermore have identified subsurface
damage (SSD) as a major factor in limiting laser damage resistance under these
circumstances. Subsurface damage consists of fractures and scratches that occur
during the grinding and polishing process which become partially or totally
hidden by the polishing redeposition layer. SSD is minimized by using a
sequence of successively finer grinding and polishing steps, and ensuring that
each step removes sufficient material to eliminate any damage caused by the
previous step.
The
only two materials useful for producing transmissive optics in the deep UV are
fused silica and calcium fluoride (CaF2).
Polishing CaF2
is a particular challenge because it is anisotropic, hygroscopic and very prone
to chipping and fracturing. Achieving a tight surface quality specification
with these characteristics requires longer polishing time than for harder
materials, such as fused silica. Unfortunately, the longer the polishing time,
the more difficult it is to hold a given flatness or surface figure. The bottom
line is that fabricating CaF2
components requires a very delicate balancing act between the various process
parameters (polishing time, spindle speed, spindle pressure, etc.) in order to
simultaneously meet both shape and surface quality specifications; at the
present time, polishing this material is as much an art as a science. At ARO,
we have also found it imperative to environmentally isolate the polishing area
for CaF2
from the rest of our production area. This prevents the possibility of
cross-contamination between our standard polishing procedures and those for CaF2.
Producing
coatings for 248 nm, and especially for 193 nm, is also hampered by the limited
number of available materials. Based on the results of extensive lifetime
testing, we have found conventional electron beam evaporation to be the most
effective production method. Ion-assisted deposition, which is often superior
at longer wavelengths, can result in the deposition of oxides that are
absorptive at short wavelengths. Stringent attention must also be paid to
substrate cleaning at every stage of production.
Conclusion
The
semiconductor industry's ongoing quest for ever smaller circuit geometries
places tremendous demands on all parts of the production process. Continued
improvements in lasers and optics for the deep UV will ensure that
microlithography can meet these needs for some time to come.
Figures
1.
Intel Pentium® Processor die. Photo Courtesy of Intel Corporation.
2. Microlithography schematic.
3. Moore's Law. Microprocessor transistor density has doubled every two years
since 1965.
|