measurement uncertainty using an ultra-low-voltage, critical-shaped
Yates and John Haywood, LSI Logic; and Galen Sapp, Dmitry Gorelikov,
and Paul Knutrud, Soluris
the IC industry approaches the 65-nm technology node, decreasing
feature size and tightening requirements for dimensional control require
a new approach to critical dimension (CD) metrology. Previously adequate
approximations are no longer sufficient, and previously insignificant
errors are now unacceptable. For sub-100-nm processes, especially advanced
logic devices, transistor performance controls the final value of the
IC. Transistor speed is a direct function of the gate length, which
is controlled by the gate polysilicon line width CD. Each additional
nanometer of CD control adds directly to the final value of the circuit.
photoresists used with 193-nm lithography exhibit significant slimming
when irradiated by the electron beam of the critical dimension scanning
electron microscope (CD-SEM). Although it is possible to estimate the
rate of slimming by examining the changes induced by repeated exposures,
the slimming caused by the initial exposure is difficult to quantify
with any certainty. The work described in this article confirms that
slimming is primarily a function of beam energy and that the use of
ultralow energy can reduce slimming to an acceptable level.
2004 International Technology Roadmap for Semiconductors maintains
that printed-gate CD control should be <4
nm at the 90-nm technology node and 2.5 nm at the 65-nm node. Achieving
that goal will not be easy because it involves both process and measurement
uncertainty. Moreover, the concept of critical dimension itself is somewhat
outmoded, a holdover from previous generations when circuit designs
could be considered to be essentially two dimensional. When the width
of a feature is much greater than its height, variations in the shape
of the edge are insignificant. But current-generation processes often
use high-aspect-ratio structures whose shape and width must be tightly
controlled to ensure optimal device performance. CD metrology has evolved
into critical-shape metrology.
article examines two sources of uncertainty in CD measurements: the
impact of the measurement process on the measurement result and the
need to control shape as well as size. While several methods exist for
determining feature shape, all of them have significant shortcomings
in process control applications. In contrast, this article describes
a CD-SEM-based approach that is precise, accurate, and fast.
issue of resist slimming is not new with 193-nm resists. Early generations
of 248-nm resists also exhibited slimming. To reduce slimming, resist
manufacturers reformulated their products to reduce their susceptibility
to the high-energy electrons present during plasma etch. As a result,
the new formulations exhibited reduced slimming during CD-SEM measurements.
resist manufacturers are trying to achieve similar slimming levels with
193-nm resists, meeting that goal is proving to be difficult for several
reasons. First, 193-nm resist is optimized to work with small features
at the expense of other parameters, such as E-beam sensitivity. Second,
advanced logic devices in 90-nm processes require physical linewidths
as small as 40 nm. At a 10% precision-to-tolerance ratio, creating such
linewidths requires CD control of 4 nm. Third, line-edge roughness has
emerged as a limiting factor in circuit performance. Unfortunately,
many of the additives that improve slimming behavior during etch and
CD-SEM also increase line-edge roughness. Given these competing effects,
it may not be possible to eliminate slimming by reformulating 193-nm
the case of CD-SEMs, it has proven possible to reduce slimming to acceptable
levels by reducing the landing energy of the beam electrons. In general,
SEM resolution is determined by the size of the spot formed on the sample
surface by the scanning electron beam. However, it is generally more
difficult to focus low-energy electrons into a small spot than higher-energy
electrons. There has been a steady improvement in the electron optical
performance of SEMs at low voltages, driven largely by the need of the
semiconductor industry to reduce the adverse effects of high beam energies
on electronic circuits. However, while it has become routine to operate
with accelerating voltages that are <1 kV (CD-SEMs typically operate
with beam energies of 600–800 eV), the elimination of resist slimming
requires energies in the 100-eV range.
solve the conflicting requirements of high beam energy for optical performance
and low energy to eliminate resist slimming, specialized electron optical
systems have been developed. For example, the dedicated ultra-low-voltage
CD-SEM used in this study, the Yosemite system from Soluris (Concord,
MA), uses high beam energies throughout most of the electron column
to maintain optical performance and a retarding field between the final
lens and the sample to reduce electron landing energy at the sample
surface. Capable of fully automated operation, the tool has landing
energies of <100 eV, making it suitable for evaluating resist slimming
as a function of landing energy.
1: Response of a typical 193-nm resist to exposure to an E-beam.
Slimming is characterized by a rapid initial phase, followed by
a slower intermediate phase and very slow final decay.
of Beam Energy on Resist Slimming. Several studies have investigated
193-nm resists in order to understand the mechanisms behind slimming
behavior.1 Figure 1 shows the response of a typical 193-nm
resist to exposure to the electron beam.2 The data were collected
at beam energies of 500 and 175 eV; all other test conditions were identical.
The test results demonstrated that slimming can be characterized by
a rapid initial phase, followed by a slower intermediate phase and very
slow final decay. A triple-exponential model was developed to explain
the experimental data:
τ1–3—the half-lives of the three phases—are
obtained through fitting. Figure 1 demonstrates the excellent fit between
the model and the experimental data.
2 shows the three phases plotted separately.2 It is apparent
that most of the observed difference between the high and low voltage
values takes place during the first phase. From the metrologist's point
of view, the first phase is the most troublesome and important one,
since it occurs on a timescale that is comparable to the time required
for a single measurement. The first phase exerts its greatest influence
in the initial exposure. Moreover, when relying only on CD-SEM measurements,
the magnitude of the initial slimming cannot be determined directly,
since there is no zero exposure value with which to compare it. Several
methods are used to estimate the initial value by applying an offset
to the first measured value. That offset is determined by extrapolating
backward from subsequent measurements.
2: The three slimming phases plotted separately. The first phase
contributes to most of the observed difference between the high
and low voltage values.
3 compares CD-SEM measurements taken at 175 and 500 eV.2
The figure also includes estimated values for the unexposed CD calculated
using various fitting models: exponential, polynomial, power, and linear.
Depending on the model chosen, the initial slimming value ranges from
0.9 to 3.3 nm. Regardless of the model chosen, the steep slope of the
curve in the region of the estimate tends to amplify any uncertainty
in the data points on which the estimate is based.
3: CD-SEM measurements taken at 175 and 500 eV. Estimated values
for the unexposed CD calculated using polynomial, exponential, linear,
and power fitting models are also included.
three phases of the slimming phenomenon are thought to correspond to
three different physical processes occurring in the resist. The first
and fastest phase is characterized by chemical reactions within the
resist that are a direct result of the interaction with beam electrons.
These reactions may include chain scission and photo acid generator
decomposition. Depending on landing energy, these processes occur within
20 nm of the resist surface, which is the penetration depth of the electrons.
These reactions may cause the resist to form an inert skin that is impervious
to further electron bombardment. In the second phase, as irradiation
by the beam continues, thermal heating occurs, causing solvent to evaporate
from the resist and leading to the loss of free volume as a result of
annealing. The third phase, which may be so slow that its effect is
negligible, probably involves additional solvent loss from tightly bound
sites and other longer-term processes.
agreement between experimental results and a slimming model based on
the Grun electron range (essentially electron penetration depth) offers
strong confirmation that the processes occurring in the first phase
are related to direct beam interaction. Figure 4 compares atomic force
microscope (AFM) measurements with the Grun slimming model.2
Each AFM data point represents the difference between measurements made
before and after CD-SEM measurement. As shown in Figure 5, penetration
depth increases from 2 nm at 100 eV to more than 30 nm at 800 eV.3
Since reformulating the resist to reduce slimming may not be as successful
for 193-nm resists as for 248-nm resists, the best way to reduce slimming
is to perform measurements at the lowest possible landing energy.
4: AFM measurements compared with a model based on the Grun electron
range. Each AFM data point is the difference between measurements
made before and after CD-SEM measurement.
5: Schematic diagram showing how the depth of penetration changes
from 2 nm at 100 eV to more than 30 nm at 800 eV.
of Slimming on Poly Lines. To gain insight into the ultimate
impact of slimming on circuit elements and to evaluate the amount of
slimming that results from the initial CD-SEM measurement, a test was
performed to determine the effect of performing a single measurement
on the CD of subsequently etched features. An experiment was designed
that compared poly linewidths from regions where the resist was measured
using CD-SEM with nearly adjacent regions that were not measured. While
etched polysilicon lines are not slimmed by exposure to an electron
beam, the electron beam slims the resist, which affects the poly lines.
investigation also sought to confirm that landing energy, and not another
electron-beam exposure parameter, is the primary contributing factor
slimming. Consequently, beam energies of 100, 200, 300, 500, and 800
eV (at beam currents of 20 and 40 pA) were examined. The number of integration
frames was varied from 48 to 64 to 96 to 128 to 256. All resist exposure
regions were separated by at least 3 Ám, and care was taken to avoid
exposing other resist regions to the beam inadvertently. All postetch
measurements on the polysilicon were performed under the same conditions:
a landing energy of 500 eV, a beam current of 20 pA, and 64 integration
expected, slimming of the resist translated directly into slimming of
the etched poly lines. Figure 6 summarizes the results for all currents
and frame exposures.4 The use of a 100-eV landing energy
resulted in slimming of slightly more than 1 nm, which was 5 nm less
than that caused by using an 800-eV landing energy. ANOVA analysis of
all experimental variables confirmed that landing energy has the greatest
influence on slimming.
6: Summary of resist slimming results for all currents and frame
exposures. Operating at 100 eV reduced slimming to slightly more
than 1 nm, which was 5 nm less than at 800 eV.
sub-100-nm processes, feature shape can be as important as width. For
example, during gate formation, the overlying polysilicon line masks
the gate region during the implant step that forms the source and drain
contacts. Although the width of the line remains the primary determinant
of gate length, the slope of the side wall has a significant effect
on the contours of the implant and can play a critical role in determining
device performance. Hence, the need to use shape metrology will become
more important as gate structures shrink and become more complex.
CD-SEMs work by extracting a measurement from the signal-intensity profile
as the beam scans across the feature. Protruding features such as edges
tend to generate higher signal levels than flat or recessed features.
Thus, when CD-SEM was developed as a metrology technique, the bright
edges of lines were convenient—and presumably accurate—indicators
of the physical position of the edge. SEMs typically have spatial resolution
of a few nanometers. When CDs were hundreds of nanometers, errors of
a few nanometers were insignificant. However, as gate CDs have shrunk
to the 40-nm range in 90-nm advanced logic applications and will shrink
further to 30 nm at the 65-nm node, every nanometer is critical.
resolution will always be limited, and conventional CD-SEM algorithms
are not sufficient for characterizing the true profile of a feature.
The shape of an edge affects the shape of its intensity profile, and
proper analysis can extract shape measurements from the signal. Conventional
CD-SEM algorithms derive an arbitrary relative value from the intensity
profile to indicate edge position. Several different values are used,
the most common of which are the point of maximum slope and the point
at which the profile reaches a fraction of its peak value (e.g., 50%).
Although these values
tend to be highly precise for defined-edge shapes in >100-nm processes,
they are not accurate if the edge shape varies or in sub-100-nm processes.
an SEM's highest-resolution signal is the secondary electron signal,
which consists of sample electrons that are scattered by interactions
with beam electrons. Secondary electrons have very low energy (<50
eV). Although they may be created anywhere within the interaction volume
(the region that can be reached by scattering beam electrons), they
can escape only if they occur within a few nanometers of the surface.
On a flat sample, most secondary electrons in the detected signal are
formed immediately below the beam spot, giving the signal its high spatial
is the primary source of contrast in secondary electron images.
example, protruding features appear bright because their shape brings
more of the surface within the secondary electron escape range of the
interaction volume. For extreme topographies, such as the sharp edge
of a photoresist line, many of the added secondary electrons originate
from points that are within the interaction volume but well outside
the beam spot. Hence, both the size of the spot and the size of the
interaction volume determine effective CD-SEM resolution (as shown in
Figure 5). Lower landing energy reduces the size of the interaction
volume and improves the correspondence between the intensity profile
and the physical profile.
of the landing energy, the intensity profile that is generated as the
beam scans across an edge represents a convolution of the physical edge
profile, the distribution of electrons within the beam, and the distribution
of primary electrons in the interaction volume. To extract the physical
profile from the intensity profile, the CD-SEM signal must be deconvoluted
using a Monte Carlo–based simulation method that is based on the
physics of the interactions. The Soluris critical-shape metrology (CSM)
algorithm was used in this work to create a library of profiles based
on a range of input parameters that were specified by the operator.
The system searched the library to find a match for the detected intensity
profile and reported the dimensions of the corresponding physical profile.
accuracy of the algorithm for identifying the shape of edge profiles
on etched polysilicon lines was evaluated by comparing CSM measurements
with focused ion beam (FIB)/SEM cross sections. A trapezoidal model
was used to represent feature shape while allowing foot length, sidewall
angle, and top-corner rounding to vary. In this evaluation, left and
right edges were matched independently to improve matching speed. However,
as feature sizes continue to shrink, it will become necessary to match
both edges together to accommodate the interaction between them.
wafers were prepared with a 130-nm etched polysilicon layer on silicon
oxide. Etch conditions were varied for each wafer to produce challenging
samples with varying profile shapes and sidewall angles for the CSM
analysis. The wafers were first measured using a Yosemite CD-SEM and
then measured again using an automated FIB/SEM.
330 die were measured on each wafer using the CD-SEM's CSM algorithm,
while 10 sites were measured on each wafer using the FIB/SEM. An additional
160 measurements were performed on one of the wafers using the FIB/SEM.
CD-SEM images were obtained at 500 V, and FIB/SEM images were obtained
at 2 kV. Integrated FIB analy-sis software was used to extract the poly
line geometry from the cross-section images.
top and bottom measurements were made using the 20-point measurement
and regression scheme illustrated in Figure 7.5 The top was
calculated at the top of the line dome, and the bottom was calculated
at the gate oxide. Both left and right sidewall angles were measured,
while the thickness of the poly was measured from the top of the gate
oxide to the top of the feature using waveform analysis. Based on those
data, the bottom CD was measured at the top of the gate oxide and at
a point 5 nm above that. The top CD was measured at a point coincident
to the peak of the feature and between points extrapolated from the
sidewall angles. Top CD was also measured 5 nm and 10 nm below that
measurement point. A final CD measurement was taken at the vertical
midpoint of the feature.
7: Extrapolated top and bottom measurements made using a 20-point
measurement and regression scheme. The top was calculated at the
top of the line dome and the bottom was calculated at the gate oxide.
shown in the Figure 8a images, wafers A and
C had sidewall angles typical of features that might be encountered
in production. Wafer B, which had a profile with low sidewall angles,
was used to test CSM performance under extreme nonnormal conditions.
The microtrench along the etched lines was an unexpected by-product
of the perturbations in the etch process. In general, the agreement
between FIB and CSM results was better for wafers A and C than for wafer
B. Reported differences between the sidewall angles on wafers A and
C ranged from 1.3° to 0.1°, while the difference between the
sidewall angles on wafer B was 4°. Similarly CSM measurements of
extrapolated top and bottom CDs showed better agreement with FIB measurements
for wafers A and C than for wafer B. Deltas for top and bottom CD ranged
from 3 nm on wafer C to 9 nm on wafer A. On wafers A and C, CSM measurements
showed consistently better agreement with FIB measurements than with
conventional maximum slope CD measurements. At up to 14 nm, the difference
between FIB and maximum slope CD measurements on wafer A was nearly
twice as large as the difference between FIB and CSM measurements.
8: FIB cross-sectional images (a) showing overetch of the feature
through the gate oxide with a microtrench depth up to 15 nm below
the oxide level; and (b) a layer of metal deposited over the line
to protect it during milling.
B had a bottom CD delta of 3 nm and a top CD delta of 12 nm. Several
factors may explain the larger top CD and sidewall-angle deltas. First,
these lines are more triangular than trapezoidal and have an extrapolated
top CD of approximately 7 nm as measured by FIB. At that size, it may
not be valid to treat the edges independently, since the signal from
one edge may be affected by the signal from the other. Second, the "necking"
visible near the top of the line may introduce error into edge detection
and the extrapolation of the FIB and CD-SEM analysis. Third, the simulation
and matching library were optimized for the more-severe angles typically
encountered in production devices.
comparing the FIB and CSM measurements, the inherent undersampling of
the FIB cross-section technique and the effects of line-edge roughness
must be taken into consideration. The once-per-site FIB measurements
are approximately 3 to 10 times less precise than CSM measurements.
CSM measurements represent an average of many profiles acquired along
a 1-Ám segment of a line. The 160 FIB measurements on wafer A exhibited
a CD variability of 9.4 nm and sidewall-angle variability of 5.4°.
The corresponding CSM values were 4.2 nm and 1.2°. Linewidth roughness,
estimated at 7.4 nm, is an additional source of variability.
control requirements for sub-100-nm processes are beginning to outstrip
the capacity of conventional CD metrology. CD requirements are at the
4-nm level, while 2.5-nm CD requirements are on the horizon. Improvements
in dimensional control translate directly into improved device performance
and increased revenues. Resist slimming and uncontrolled variations
in shape are two important sources of uncertainty. Hence, the integration
of advanced CD-SEM technology and ultra-low-voltage operation to reduce
slimming and the use of critical-shape metrology to achieve 3-D shape
control promise to reduce CD-SEM uncertainty without compromising measurement
T Kudo et al., "CD Changes of 193-nm Resists during SEM
Measurement," in Proceedings of Microlithography 2001
4345 (Bellingham, WA: SPIE, 2001), 179–189.
N Sullivan et al., "Electron Beam Metrology of 193-nm Resists
at Ultralow Voltage," in Proceedings of Microlithography
2003 5038 (Bellingham, WA: SPIE, 2003), 618–623.
G Sundaram et al., "Low Impact Resist Metrology: The Use
of Ultralow Voltage for High-Accuracy Performance," in Proceedings
of Microlithography 2004 5375 (Bellingham, WA: SPIE, 2004), 675–685.
C Yates et al., "Characterization of E-Beam Induced Resist Slimming
Using Etched Feature Measurements," in Proceedings of Microlithography
2005 5752 (Bellingham, WA: SPIE, 2005), 510–515.
D Gorelikov et al., "Application of Critical Shape Metrology
to 90 nm Process," in Proceedings of Microlithography
2005 5752 (Bellingham, WA: SPIE, 2005), 489–498.
Yates works as a photolithography development engineer in LSI
Logic's wafer fabrication facility in Gresham, OR. He has been with
the company since 1989. He holds four U.S. patents. (Yates can be reached
at 503/618-5305 or firstname.lastname@example.org.)
Haywood works on etch development at LSI Logic. He has 20 years
of experience in the semiconductor industry in the areas of photolithography
and etch. Before joining LSI Logic, he worked at Intel. He holds several
U.S. patents and has other patents pending. He received a degree in
physics from the University of Manchester, UK. (Haywood can be reached
at 503/618-3677 or email@example.com.)
Sapp is a senior applications engineer at Soluris. She began
her CS-SEM metrology career at IVS in 1985 and has continued to work
in this area ever since. She received a BS in chemistry from San Francisco
State University. (Sapp can be reached at 415/664-5393 or firstname.lastname@example.org.)
Gorelikov, PhD, is a senior scientist at Soluris (Concord,
MA). His primary interest is Monte Carlo simulations of electron beam–sample
interactions in CD-SEM. He received an MS in metallurgy from the Moscow
Steel and Alloys Institute and a PhD in materials science and engineering
from Northwestern University in Evanston, IL. (Gorelikov can be reached
at 978/318-4116 or email@example.com.)
Knutrud is technical marketing manager at Soluris. He began
his metrology career in 1983 as an applications engineer at IVS and
rose to director of marketing at Schlumberger. He holds a patent in
overlay metrology and has authored more than 10 publications in the
field. He received a BSBA in management From Babson College (Wellesley,
MA) in 1983. (Knutrud can be reached at 978/318-4041 or firstname.lastname@example.org.)