DK1758 ch15

15
Scanned Probe Microscope Dimensional

Metrology
Herschel M. Marchman*
University of South Florida, Tampa, Florida
Joseph E. Grif®th
Bell Laboratories, Lucent Technologies, Murray Hill, New Jersey
I. INTRODUCTION
Since their introduction almost two decades ago, scanning probe microscopes (SPMs)
have already deeply impacted broad areas of basic science and have become an important
new analytical tool for advanced technologies, such as those used in the semiconductor
industry. In the latter case, the metrology and characterization of integrated circuit (IC)
features have been greatly facilitated over the last several years by the family of methods
associated with proximal probes. As IC design rules continue to scale downwards, the
technologies associated with SPM will have to keep pace if their utility to this industry is to
continue. The primary goal of this chapter is to discuss the application of SPM technology
to dimensional metrology. In the past, critical-dimension (CD) metrology has mostly
involved the measurement of linewidth using either light or electron optical microscopes.
However, increased aspect ratios (height/width) of today's IC features have created the
need for measurement information in all three dimensions.
The de®nition of linewidth is not clear for structures typically encountered in IC
technology, such as the one shown in Figure 1. Features may have edges that are ragged
and side walls that can be asymmetric or even re-entrant. A more appropriate term than
linewidth is perhaps line pro®le, which describes the surface height (Z) along the X direction
at a particular Y location along the line. Alternatively, the pro®le could be given by the
width X at a particular height Z of the line (the line being a three-dimensional entity in
this case). The choice of height at which to perform a width measurement will probably
depend on the context of the fabrication process in which the feature is to be measured. For
instance, the linewidth of interest may be measured at the bottom (foot) of a resist structure
that is in a highly selective etch process. However, the width measurement point should
probably be shifted further up the pro®le for less selective etch processes. Edge roughness is
also a major factor in the uncertainty of dimensional measurements, because the measured
value will differ according to where along the line (i.e., Y coordinate) one samples.
Copyright © 2001 Marcel Dekker, Inc.

Figure 1 Three-dimensional representation of a line pro®le.
Scanning probe microscopes (SPMs) offer an attractive alternative to optical and

electron microscopes for complete three-dimensional imaging of structures. And SPMs are
capable of achieving atomic-level resolution for a wide range of materials, even in ambient
conditions. Unlike optical and electron microscopes, SPMs do not use optics or waves to
obtain images. In an SPM, a needlelike probe is brought very close (< 3 nm) to the sample
surface and typically traversed back and forth in a raster fashion. The probe rides up and
down at a constant height above the sample, so a topographic image of the surface is
obtained. High resolution in all three dimensions is achieved simultaneously. Accurate
pro®les of the surface can be extracted at any position in a completely nondestructive
manner, and a complete rendering of the surface can be created. The main characteristic
that distinguishes SPMs from optical and electron-based microscopes is that a solid body,
the probe tip, is intimately involved in the measurement process.
A. Scanning Tunneling Microscopy

Scanning tunneling microscopes (STMs) were the ®rst types of SPM to be widely used
(1±3). In this technique, a metallic probe is brought to within the electron tunneling
distance of a conductive sample surface (see Figure 2), such that a small tunneling current
¯ows when a voltage bias is applied. The tunneling current (typically < 1 nA) ¯ows to or
from a single atomic cluster at the apex of the probe. Tunneling current density decays
exponentially with distance between the two electrodes (tip and sample). Essentially, this
can be thought of as a form of ``near-®eld'' SEM. There is no need for vacuum or electron
lenses because the source and sample are brought to within the near-®eld distance from
each other and scanned while tunneling occurs. Tip-to-sample separation is adjusted by
moving either the sample or tip with a piezoelectric transducer that is controlled by a
feedback system that maintains constant current and, hence, distance. In this way, the tip
is made to track the height of the surface. An image is obtained by rastering the probe
across the ®eld of view and plotting the change in tip height position as it tracks the
surface. It is this exponential dependence of the current on tip-to-sample separation
that gives the STM its extremely high vertical resolution (subangstrom). Lateral resolution
on the atomic scale is achieved if the current emanates from a single atomic cluster (the
ideal situation) near the apex of a tip. Unfortunately, the types of samples that can usually
be imaged with the STM are limited to only those having highly conductive surfaces.

Figure 2 Energy band diagram for metal±vacuum±metal tunneling in STM. The arrows pointing
to the right between the probe and the sample represent tunneling electrons.
B. Atomic Force Microscopy

Most samples encountered in integrated circuit (IC) metrology contain poorly conduct-
ing or insulating surface regions. Photoresists and oxides are two common examples.
The surface proximity sensor for probe microscopes must therefore be able to operate
over all regions uniformly. For this reason, force sensors are the most commonly used
type of detection for proximity of the probe tip to the surface. The atomic force
microscope (1) (AFM) was developed shortly after the STM, as a result. Such
AFMs measure the topography of a surface by bringing a sharp probe very close
(within angstroms) to the sample surface to detect small forces due to the atoms on
the surface instead of passing a current. Figure 3 shows the force plotted against
distance from the surface. A strong repulsive force is encountered by the tip at dis-
tances very near to the surface atoms, due to the exclusion principle. Contact-mode
scanning is said to occur when these repulsive forces are used for regulating the tip-to-
sample distance. If the tip is moved away from the surface, a small attractive force is
encountered several nanometers back. Feedback control with these longer range forces
is employed with the noncontact, attractive, mode of imaging. It should be reiterated
that the main advantage of AFM is that insulating samples can be imaged just as well
as conductive ones.
A simpli®ed diagram of an AFM is shown in Figure 4. Forces between the tip
and sample surface are sensed as the probe tip is brought towards the sample surface,
as stated earlier. The probe-to-sample distance is regulated by maintaining a constant
force, via a feedback servo-system, once the initial approach has been completed (i.e.,

Figure 3 Diagram of the tip±sample junction and a force curve for the AFM. Details of the force
curve may depend on whether the tip is approaching or withdrawing from the surface.
the surface has been found). The vertical displacement of the cantilever acquired during
scanning is then converted into topographic information in all three dimensions. The
AFM is analogous to a stylus pro®lometer, but scanned laterally in x and y, with a
constant amount of tip-to-sample force maintained in order to obtain images.
Although resolution is not at the same level as STM, atomic-scale imaging can be
performed by AFM.
Figure 4 Simpli®ed block diagram of a generalized AFM system.

II. ATOMIC FORCE MICROSCOPY IMAGING MODES
It is clear that the characteristics of different force modes have signi®cant implications on
their application in dimensional metrology. In the repulsive-force mode, damage to the tip
is accelerated by constant contact with the surface during scanning. The tip-to-sample
distance in an attractive-mode AFM is usually an order of magnitude larger than that of
the repulsive mode. Increased tip-to-sample separation helps to minimize unwanted tip-to-
sample contact. Lateral forces from approaching sidewalls can also be sensed in the
attractive mode, which makes this type of AFM ideal for imaging high-aspect features
that are commonly found in semiconductor processing. However, contact mode may track
the actual surface more acurately under some environmental conditions, such as during
static charge or ¯uid layer buildup on the surface. There is an amazing variety of sensors
used by AFMs for detecting the presence of surface forces in either mode. The large
number of surface-proximity detectors invented in recent years attests to the importance,
and dif®culty, of surface detection in scanned probe microscopy (3±5). We will brie¯y
discuss ®ve different force sensors commonly used for measurements on wafers: (1) one-
dimensional resonant microcantilever, (2) two-dimensional resonant microcantilever,
(3) piezoresistive microcantilever, (4) electrostatic force balance beam, and (5) and
lateral/shear-force resonant ®ber.
A. One-Dimensional Resonant Microcantilever

In most atomic force microscope designs the probe is mounted on a microcantilever, which
serves as a force sensor (6,7). Forces from the surface are measured by monitoring the
¯exing of a cantilever using an optical lever. This optical lever is implemented by re¯ecting
a laser beam from the end of the cantilever onto a position-sensitive detector (8). To
improve both the sensitivity and stiffness, most designs exploit resonance enhancement.
Because of their small size, a typical resonant frequency for the cantilever is around
300 kHz. Interactions between the probe and sample introduce variation in the amplitude,
frequency, and phase of the cantilever oscillation. Any of these signals can be fed back into
the microscope and used to maintain the vertical position of the scan head. One drawback
of this design is that the cantilever is typically tilted about 10 with respect to the sample
(see Figure 5). Oscillation thus causes high-frequency lateral motion of the probe tip as
well. The actual force of interaction is not directly available from the system, but Spatz et
al. (9) have obtained an estimate through modeling of typically a few tenths of a micro-
newton. In standard-mode operation, a simple raster scan is used and servo control is
performed using the one-dimensional resonant attractive-mode technique to maintain a
constant tip-to-sample spacing of approximately 2±3 nanometers. This type of scanning
algorithm makes uniformly spaced steps across the sample surface, and the vertical tip
position is recorded at each step.
B. Two-Dimensional Resonant Microcantilever

Martin and Wickramasinghe have developed a sophisticated two-dimensional force sensor
whose design is also based on a microcantilever (10±12). This system is supplied with a
¯ared probe tip, shown in Figure 6, called a boot tip, which allows improved access to
vertical sidewalls and undercut regions (11). The force sensor operates in a noncontact
fashion that exploits resonance enhancement of the microcantilever in two directions. A
microcantilever with a spring constant of 10 N=m and Q 300 is vibrated both verti-

Figure 5 Schematic diagram of an AFM microcantilever with its optical-level force monitor.
cally and laterally at different frequencies with an amplitude of 1 nm. Cantilever vibra-
tion is detected using an optical interferometer. The force sensitivity is approximately
3 10 12 N, which allows it to detect the presence of the much weaker attractive forces
(not necessarily van der Waals only). Separation of the vertical and horizontal force
components is accomplished through independent detection of the each vibration fre-
quency. Servo control of both the horizontal and vertical tip-to-sample distances now
occurs via digitally controlled feedback loops with piezo actuators for each direction.
When the tip encounters a sidewall, their separation is controlled by the horizontal feed-
back servo system. Conversely, tip height (z direction) is adjusted by the vertical force
regulation loop.
The scan algorithm uses force component information to determine the surface
normal at each point and subsequently deduce the local scan direction (11). This allows
the scan direction to be continually modi®ed as a function of the actual topography in
order for the motion to stay parallel to the surface at each point. In this way, data is not
Figure 6 Dithering the microcantilever in two dimensions allows this microscope to determine the
local surface normal vector.

acquired at regular intervals along x anymore, but at controlled intervals along the surface
contour itself. This is essential for scanning undercut regions, because in addition to
having a ¯ared tip, it is necessary to have a two-dimensional (2-D) scanning algorithm
that can provide servoed tip motion in both the lateral and vertical scan axis directions.
One could imagine what would happen if a ¯ared tip were to try to move straight up while
under a feature overhang. The data set collected is stored as a three-dimensional mesh that
maintains a relatively constant areal density of data points, irrespective of the sample
slope. It is especially well suited for measuring the angle of a sidewall or its roughness.
Another unique capability of this system is the ability to automatically adjust scan speed
to surface slope so that it slows when scanning a nearly vertical topography and hastens
when scanning a nearly ¯at surface. Data point density (in the x direction) also increases at
sudden changes in surface height, such as feature edges (of most interest), because the
sampling rate is ®xed.
C. Piezoresistive Microcantilever
One strategy for improving AFM throughput is to design a microscope in which a large
array of probes operates simultaneously. Implementing such an array presents two tech-
nical problems. First, optical lever systems are awkward to assemble in large arrays.
Second, the height of the probe tips must be independently controllable because most
samples are not perfectly ¯at. Tortonese et al. developed a novel piezoresistive-force sensor
suitable for use in large arrays (13). The microcantilever is supported by two parallel arms,
each of which contains a piezoresistive ®lm, B-doped h100iSi. The force on the probe ¯exes
the arms, producing a small change in resistance, which serves as the feedback signal.
Minne et al. developed the independent actuator for this sensor (14). They built into each
cantilever a piezoelectric ZnO actuator, which is capable of raising or lowering the probe
by several micrometers. Thus, each probe tip in the array can independently follow the
sample surface at scan speeds up to 3 mm/s (15).
D. Balance-Beam Force Sensor

Because of their tiny size, the probe tips of microcantilevers are typically manufactured
along with the cantilever, which limits the methods and materials used. It is possible to
achieve high sensitivity in a larger sensor. An example is a centimeter-long beam held
balanced on a weak pivot by electrostatic force (Figure 7). The two sides of the balance
Figure 7 Balance-beam force sensor. The magnet holds the beam on the base plate. A force-
balance circuit balances the beam with electrostatic forces.

beam form capacitors with the base. This balance-beam force sensor (16), also known as
the interfacial force microscope (17), is an inherently unstable mechanical system stabi-
lized by a servo loop. By using force balance rather than a weak spring, this method of
force sensing uncouples the sensitivity from the stiffness. The pivot, developed by Miller
and Grif®th (18), is a pair of steel ball bearings, one on each side of the beam, held to the
substrate with a small magnet. The ball bearings are rugged, so a beam can be used almost
inde®nitely. The magnetic constraint suppresses all degrees of freedom except the rocking
motion, so the sensor is quiet. This system has been used to measure surface roughness
with RMS amplitude less than 0.1 nm.
The capacitors serve two purposes: They both detect and control the position of the
beam. The detector is a 500-kHz bridge circuit, which compares the relative values of the
two sides. The bridge circuit controls the dc voltage, V V, across the two capacitors to
keep the beam balanced. The capacitors set the size of the balance beam, since the elec-
tronics are designed for a capacitance of a few picofarads. The size of the beam is
5 mm 10 mm. The force F impressed on the tip can be calculated directly from the
dimensions of the sensor and the difference V in the voltage on the two sides. For a
typical sensor , V 1 mV implies F 2 10 8 N.
The balance-beam force microscope has several advantages. The most important of
these arises from its large size. It can accept almost any probe tip. The sensor consequently
does not constrain either the probe material or its mode of fabrication. The tip can be
several millimeters long, so the sensor does not have to be extremely close to the surface.
The sensor can be mounted on the scanner piezos, so the ®xed sample can be large. The
sensor performs well in vacuum. It can operate simultaneously as a tunneling microscope.
Magnetic constraint allows easy tip exchange, and the pivot is exceptionally rugged. The
sensor also allows the tip to be accurately aligned perpendicular to the sample, which is
important when measuring vertical sidewalls.
E. Lateral Force Resonant Fiber

In lateral force microscopy, a tapered ®ber or wire whose length axis is normal to the
surface (Figure 8) is vibrated at mechanical resonance laterally as it is brought close to the
surface (19,20). Surface forces are indicated by changes in either the phase (y) or amplitude
(A) of the ®ber resonance. In many cases a combination of the two, known as the quad-
rature component, or A cos y, is actually used for regulation. This technique offers the
Figure 8 Head assembly for the lateral/shear-force ®ber SPM.

stability of an amplitude signal with the enhanced speed and sensitivity of a phase com-
ponent. Actually, force sensing in all directions is achieved. Distance to the sample surface
is sensed vertically through shear forces and horizontally to approaching side walls
through noncontact damping (similar to resonant mode). Lateral forces between the tip
and approaching side walls are attractive, so there is less likelihood of accidental contact
(i.e., damage). Noncontact lateral-force sensing also enables quicker scanning. Shear-force
sensing is advantageous for vertical height regulation in that it is not as susceptible to
effects caused by ¯uid layers or static charges as the attractive mode.
Lateral-force probes are typically made from optical ®bers that are composed mainly
of glass. The actual imaging probe is etched from the optical ®ber's endface. In addition to
yielding high-aspect cylindrically shaped probes for metrology, optical ®bers possess sev-
eral intrinsic properties that can be exploited for simplifying SPM instrumentation. The
®ber body's natural resonance provides a means of force sensing and the ability to emit
light provides a signal for tracking the probe's motion. An optical microscope with high
NA (> 0:9) objectives are used to position the tip as the sample is translated by long-range
(> 200-mm) motion stages. The entire probe scan assembly is mounted to the optical
viewing microscope. The lateral-force SPM essentially looks and operates like an ordinary
optical microscope with a scanned probe in the same ®eld of view as the feature to be
measured. Simultaneous viewing of the tip (emitting light) and feature is achieved because
the optical ®ber axis is oriented normally to the sample surface. Throughput and probe
lifetime are greatly enhanced with this con®guration, since the need for SPM imaging to
locate and center the feature of interest is eliminated. There is relatively immunity to Abbe'
offset error because the light signal for position sensing is emitted at the proximal point.
The apparatus can also be used to collect optical intensity within the near ®eld once a
con®nement layer is added to the probe walls, thus allowing optical measurements to be
made with nanometer resolution (21).
III. SOURCES OF ERROR
As with other microscopes, undesired artifacts are also present in probe microscope
images, which adversely affect dimensional measurement performance. Two elements of
probe microscopes exibit strongly nonlinear behavior that can seriously affect measure-
ment accuracy and precision well before atomic resolution has been reached. The ®rst is
the piezoelectric actuator that is used for scanning the probe. The second, and probably
more serious, problem arises from interaction between the probe and the sample.
A. Scan Linearity
Piezoceramic actuators are used to generate the probe motion because of their stiffness
and ability to move in arbitrarily small steps. Being ferroelectrics, they suffer from hyster-
esis and creep, so their motion is not linear with applied voltage (22). Therefore, any
attempt to plot the surface height data verses piezo scan signal results in a curved or
warped image and does not re¯ect the true lateral position of the tip. A variety of tech-
niques have been employed to compensate for the nonlinear behavior of piezos. In many
instruments, the driving voltage is altered to follow a low-order polynomial in an attempt
to linearize the motion. This technique is good to only several percent and doesn't really
address the problem of creep. Attempting to address nonlinearities with a predetermined
driving algorithm will not be adaquate for dimensional metrology because of the compli-

cated and nonreproducible behavior of piezoelectric materials. Another approach is to
independently monitor the motion of piezoactuators with a reliable sensor. Several types
of systems using this approach have been reported. One monitored the motion of a ¯exure
stage with an interferometer (23). Another measured the position of a piezotube actuator
with capacitance-based sensors (24). A third group employed an optical slit to monitor the
piezotube scanner (25). Electrical strain gauges have also been used for position monitor-
ing as well.
These techniques monitor the position of the piezoactuator and not the actual point
of the probe that is in closest proximity to the surface (known as the proximal point). The
error associated with sensing position in a different plane from that of the proximal point,
referred to as Abbe' offset, is illustrated in Figure 9. Abbe' offset error is given by D sin a),
which increases with the tilt of the scan head as the probe moves. Unless special precau-
tions are taken, the position sensors will not respond to this tilting. There is a tendency in
the case of SFM systems to dismiss this error as negligible because of the submicrometer
dimensions being measured. As an example of Abbe' offset error, the angular tolerance for
a 1-cm displacement between the scan-head motion and sample planes can be computed. If
we demand less than 1-nm Abbe' offset, the allowable angular error is 10 7 radians, or
1/50 arc-second. In systems built around piezoceramic tubes, the head tilting induces an
1% nonlinearity into the monitor response (26±28). In tube scanners, the magnitude of
the tilting is not a very stable function of the lateral motion of the tube, which makes
calibration of the tube unreliable. To minimize this error the sensors are designed to be as
close as possible to the probe. One can, alternatively, add extra sensors to measure the tilt.
Because of these problems with tube scanners, a design being developed at the
National Institute for Standards and Technology employs a ¯exure stage to produce
more reliable motion (29,30). A design goal of this tool is to reduce the Abbe' offset as
much as possible. Ideally, the position of the probe apex should be directly measured; a
new tool developed by Marchman et al. achieves this ideal with an optical-position sensor
and light-emitting optical ®ber probe (31,32). Errors intrinsic to the position monitors are
not constant and vary with sample topography as well as lateral scan rangeÐfor example,
errors in capacitance measurements when the plates become far apart at the lateral scan
extremes. In the optical slit method, ``1=f '' noise of the detector at low frequencies (scan-
ning) imposes limitations on resolution as well as complications involved with the slit (e.g.,
tilting of the slit as the tube scans, or aperture edge irregularities).
Figure 9 Abbe' offset error for a tube-based scanner.

A problem common to one-dimensional scanning techniques, even optical and SEM,
arises from the imaging algorithm itself. As seen in Figure 10, data points in a single
linescan are equally spaced in increments of X in the lateral direction. The size of
each X increment is constant, regardless of the topography, so there is insuf®cient
pixel density when abrupt changes in surface height (e.g., feature sidewall) are encoun-
tered. Scanning re-entrant pro®les with even a ¯aired probe is also forbidden, because the
probe always moves forward or upwards and is not able to reverse its direction without a
2-D algorithm.
B. Probe Shape
Regardless of the speci®c force used to sense proximity to the surface, all SPMs share a
common working element: the tip. The probe tip is where the rubber meets the road in
probe microscopy. While there have been, and continue to be, important advances in other
aspects, such as detection schemes and position monitoring, improvements in the tip offer
the greatest potential for increasing SPM metrology performance. As described earlier,
current-generation probe tips are made of etched or milled silicon/silicon nitride (these
make up by far the bulk of commercial tips) or etched quartz ®bers, or else are built up
from electron-beam-deposited carbon. While all these tips have enabled signi®cant metro-
logical and analytical advances, they suffer serious de®ciencies, the foremost being wear,
fragility, and uncertain and inconsistent structure. In particular, the most serious problem
facing SPM dimensional metrology is the effect of probe shape on accuracy and precision.
Very sharp probe tips are necessary to scan areas having abrupt surface changes (33).
Most commercially available probes are conical or pyramidal in shape, with rounded
apexes. When features having high aspect ratios are scanned (see Figure 11), such as
those encountered in IC fabrication, they appear to have sloped walls or curtains. The
apparent surface is generated by the conical probe riding along the upper edge of the
feature. Even if we had known the exact shape of this probe, there is no way to recover the
true shape of the sidewalls. The fraction of the surface that is unrecoverable depends on
the topography of the surface and sharpness or aspect ratio of the tip. There exists no
universal probe shape appropriate for all surfaces. In most cases, a cylindrical or ¯aired tip
will be the preferred shape for scanning high-aspect features (34). The cross-sectional
diagram in Figure 12 demonstrates the effect of even a cylindrical probe shape (i.e., the
ideal case) on pitch and linewidth measurements. Pitch measurement is unaffected by the
probe width, but the linewidth and trench-width values are offset by the width of the
probe. Simple probe shapes provide greater ease of image correction, and the unrecover-
able regions are vastly reduced. It is essential to make the probe as long and slender as
possible in order to increase the maximum aspect ratio of feature that can be scanned.
Figure 10 One-dimensional raster scan algorithm.

Figure 11 Apparent surface generated by probe shape mixing.
Conical tips are still useful for measurements on surfaces with relatively gentle
topography. A cone having a small radius of curvature at the apex of the structure,
shown in Figure 13, can perform surface roughness measurements that cover regions of
wavelength±amplitude space unavailable to other tools (35,36). Though a conical probe
tip may not be able to access all parts of a feature, its interaction with an edge has some
advantages over cylindrical probes (33,37,38). At the upper corners of a feature with
rectangular cross section, size measurement becomes essentially equivalent to pitch mea-
surement. The uncertainty in the position of the upper edge becomes comparable to the
uncertainty in the radius of curvature of the probe apex, which can be very small. If the
sidewalls are known to be nearly vertical, then the positions of the upper edges give a good
estimate for the size of the feature. To produce sharper conical tips one can employ a
focused ion beam (FIB) technique, in which a Ga beam is rastered in an annular pattern
across the apex of an etched metal shank (39). This method routinely produces tips having
a radius of curvature at the apex of 5 nm and widening to no more than 0:5 mm at a
distance of 4 mm from the apex. Occasionally the FIB sputtering generates a tip with nearly
cylindrical shape. AFM linescans of periodic photoresist lines acquired using pyramidal
and FIB-sharpened tips are displayed in Figures 14(a) and (b), respectively. The inability
of either conical probe to measure the sidewalls of high-aspect features is clearly re¯ected
by the trapezoidal image pro®les. It is interesting to see that even though the FIB-shar-
pened tip is sharper, it still exhibits strong probe shape mixing.
Figure 12 Probe size does not affect pitch measurement, but it does affect width measurement.

Figure 13 A conical SPM tip etched on the underside of a silicon microcantilever.
Figure 14(a) Pro®le of a square grating structure obtained with a pyramidal tip.
Figure 14(b) Pro®le of a square grating structure obtained with an FIB-sharpened tip.

A cylindrical probe can be formed by chemical etching of an optical ®ber with
buffered hydro¯uoric acid (40,41). This procedure produces a nearly perfect cylinder
that can have a diameter less than 0:1 mm. Since the elastic modulus of glass is typically
70 GPa, the probes are stable against ¯exing as long as the slender section is not too long.
During scanning, these probes have proven to be very durable. An optical microscope
image of a ®ber probe before ®nal etch is shown in Figure 15. Upon ®nal etch, the same
shape is maintained and the diameter is reduced to as low as 100 nm. The ®ber's cylindrical
shape is preserved throughout the isotropic HF etch. This technique also allows ¯ared
quartz probes to be made. A scan taken with the quartz cylindrical probe is superimposed
on the corresponding SEM cross section shown in Figure 16.
A cylindrical probe that widens, or ¯ares, at the apex provides the most complete
rendition of sidewall shape. An example of such a probe tip is shown in Figure 17; it was
developed by T. Bayer, J. Greschner, and H. Weiss at IBM Sindel®ngen (7). This probe is
fabricated on a silicon microcantilever. The tolerances in the fabrication of this probe are
very tight. The two-dimensional scan algorithm described earlier must also be employed in
order to use the ¯ared tips on features with undercut regions. With all cylindrical or ¯ared
probes, the accuracy of a width measurement depends on knowledge of the probe width.
The linescan from a two-dimensional resonant AFM and ¯ared tip of the same periodic
photoresist lines shown earlier in Figure 14 is shown in Figure 18. We can now see from
the pro®le that the photoresist structures were actually re-entrant, or undercut, near the
feature tops.
C. Probe Stiffness
In semiconductor manufacturing, one routinely encounters features, such as vias, less than
0.25 micrometers wide and over 1 micrometer deep. Features with such extreme aspect
ratios pose a challeging problem for probe microscopy. The probe must be narrow enough
to ®t into the feature without being so slender that it becomes mechanically unstable. To
achieve an accurate measurement, the apex of the probe must remain ®xed relative to the
Figure 15 Optical microscope image of an etched optical ®ber probe.

Figure 16 Comparison of a cross-sectional SEM measurement and a scan line taken with the
surface/interface SNP. The line superimposed on the image is the SNP scan.
Figure 17 Flared probe tip etched from silicon cantilever.

Figure 18 Pro®le of an undercut feature obtained with ¯ared probe and 2-D scan algorithm.
probe shank. In other words, the probe must not ¯ex. Flexing introduces errors into the
measurement, in addition to causing instability of the feedback control loop.
The analysis of probe stiffness is identical to that of any cantilevered beam. Imagine
a probe tip with uniform circular cross section having radius R and length L. Let the
elastic modulus of the probe material be Y. If a lateral force F is impressed on the probe
apex, the probe will de¯ect a distance x, which obeys the following expression (42):
4FL3
x
3pYR4
Note that the geometrical factors, L and R, have the strongest in¯uence on the de¯ection.
This indicates that efforts to ®nd materials with higher modulus would only produce
marginal gains compared to the effects of the probe's size and shape.
D. Next-Generation Metrology Probes

The need to precisely control probe shape and achieve the maximum possible stiffness has
motivated recent efforts to employ carbon nanotubes as probes. Carbon nanotubes are
recently discovered materials made purely of carbon arranged in one or more concentric
graphene cylinders and having hollow interiors. This arrangement makes them essentially
giant, elongated fullerenes, a relationship made clear by the strategy they share with
spheroidal fullerenes (e.g., C60 Ðbuckminsterfullerene) for closing by incorporating a
total of 12 pentagons into their hexagonal network. Single-wall nanotubes (SWNTs)
come closest of all carbon ®bers to the fullerene ideal due to their remarkably high degree
of perfection; their poverty of defects and the special nature of carbon±carbon bonding
confers upon them material properties, such as strength, stiffness, toughness, and electrical
and thermal conductivities, that are far superior to those found in any other type of ®ber

(43). There also exists larger-diameter cousins of single-wall (fullerene) nanotubes, typi-
cally having 4±20 concentric layers, known as multi-wall nanotubes (MWNTs). These may
offer extra advantages as SPM probes due to their increased radial diameter, typically
between 20 and 100 nm. In addition to having a perfect cylindrical shape, these tubes
exhibit an unusually high elastic modulusÐin the terapascal range (44). Dai et al. were the
®rst to mount a nanotube on a microcantilever and scan with it (45). The most important
characteristic of carbon nanotube tips is their immunity to wear and erosion.
Measurement precision would be greatly enhanced if their shape remains unchanged
after accidental contact with the surface during scanning. The ability to measure their
exact shape and size will provide a new level of accuracy and precision in SPM metrology.
The transmission electron micrograph in Figure 19 shows an MWNT that has been
mounted on a pyramidal AFM tip apex.
There are two practical impediments to widespread use of these probes. First, tech-
niques for handling these tiny objects are not yet mature enough to allow mass production
of mounted tips. Manual mounting of nanotubes to silicon cantilever-based probes and
etched-quartz ®bers (46) has been performed in the developmental mode, but does not
offer the throughput for an adaquate supply in production-mode applications. Optical
microscope images taken during different steps of the tip-mounting process for etched
optical ®bers are shown in Figure 20. One can see from the ®nal step image that alignment
of the tube during mounting is critical. Second, single-walled nanotubes that tend to grow
with diameters in the 5±20-nm range are too slender to resist ¯exing. Even with an elastic
modulus of one terapascal, a 1-micrometer-long nanotube needs to have a diameter of at
least 50 nm to achieve the necessary stiffness. Fortunately the multiwalled nanotube
appears to have suf®cient diameter to offer adaquate stiffness and resistance to ¯exing
during scanning of high-aspect features. Developments in this ®eld have been rapid, and
we expect these dif®culties to be overcome in the near future. Presently, metrologic per-
formance is being gauged through studies using manually mounted nanotube tips in
different AFM force imaging modes on IC features of interest.
E. Throughput
Another very serious issue facing the implementation of scanning probe microscopy into
the semiconductor device fabrication line is low throughput. Presently, probe microscopes
image too slowly to compete with optical or electron microscopes in terms of speed. Since
Figure 19 TEM micrograph of a multiwall nanotube mounted on a pyramidal tip apex.

Figure 20 Picture sequence for nanotube tip-mounting procedure.
protection of the probe tip is of paramount importance, the system should not be driven
faster than its ability to respond to sudden changes in surface height. This is a source of
complaint from those used to faster microscopes. In many instances the probe microscope
is, however, providing information unavailable from any other tool, so the choice is
between slow and never. In addition, when comparing the speed of a probe microscope
against cross-sectional SEM, the sample preparation time for the cross sectioning should
be taken into consideration along with the fact that the sample has been irretrievably
altered.
In additon to imaging time, ®nding the desired feature is complicated because the
physical structure of most probe microscopes excludes high-magni®cation viewing of the
imaging tip on the sample surface. Microcantilever-based SPMs must also contend with
the cantilever's blocking, or shadowing, the surface region of interest. Thus the feature
must be found by imaging with the SPM at slow speed, with a limited ®eld of view (with
the exception of the lateral-force optical ®ber microscope by Marchman). In many cases,
the majority of tip wear occurs during this step, because larger scan ranges and tip speeds
are used in the search phase.
IV. ACCURACY AND CALIBRATION
The probe±sample interaction is a source of error for all measuring microscopes, but the
interaction of a solid body (stylus) with a sample offers several advantages when high
accuracy is needed. The most important advantage arises from the relative insensitivity of
a force microscope to sample characteristics such as index of refraction, conductivity, and

composition. Optical and electron beam tools are sensitive to these characteristics (47,48),
so errors can arise if a reference artifact has a composition different from the sample to be
measured. For instance, calibrating an SEM-CD measurement with a metal line may give
inaccurate results if the line to be measured consists of photoresist.
A comparison between probe microscope measurements of photoresist lines and
scanning electron microscope (SEM) measurements of the same lines was made by
Marchman et al. (46). The probe microscope measurements, made with a commercial
atomic force microscope (49), agree closely with the cross-sectional SEM measurements,
while the plan view measurements deviate from them by up to 0:1 mm, depending on the
magni®cation and how the SEM signal is interpreted. In the calibration of a probe micro-
scope there are two fundamental problems: measuring the behavior of the position sensors
and ®nding the shape of the probe tip.
A. Position Sensor Calibration

Calibrating position sensors is the easier of the two chores and we will discuss it ®rst.
Lateral position sensor calibration is equivalent to a pitch measurement, and an excellent
reference material is a grating, a two-dimensional one if possible. Periodic reference mate-
rials are available commercially, or they can be reliably fabricated through, for instance,
holographic techniques (50). Some samples are self-calibrating, in that they contain
periodic structures with a well-known period.
Vertical calibration is more troublesome because most step height standards contain
only one step. This makes it impossible to generate a calibration curve unless multiple
standards are used. In addition, the step height sometimes differs by orders of magnitude
from the height to be measured. Surface roughness measurements often involve height
differences of less than a nanometer, though they are sometimes calibrated with step
heights of 100 nm or more. The vertical gain of a piezoscanner may be substantially
different in the two height ranges.
B. Tip Shape Calibration

The image produced by a scanning probe microscope is simply a record of the motion of
the probe apex as it is scanned across the sample. The image will faithfully represent the
sample as long as the apex and the point of interaction coincide. If the interaction point
wanders away from the apex, the relationship between image and sample becomes more
complicated. An example of this is shown in Figure 21. As the probe moves across the
step, the apex follows the dashed line rather than the true step pro®le. Note that the
interaction point stays ®xed at the upper edge of the step until the apex reaches the
lower plane. The dashed line is, in fact, an inverted image of the probe. Several researchers
have analyzed this process (51±55). Villarrubia has conducted the most thorough study,
making use of results from mathematical morphology (54,55). We will use his notation.
The behavior represented by the dashed line is often called ``convolution,'' which is a
misnomer. Convolution is a linear process, while the interaction of the probe with the
sample is strongly nonlinear. The correct term for this process is dilation.
The mathematical description of dilation uses concepts from set theory. Imagine two
objects A and B, represented by sets of vectors. The individual vectors in the objects will be
denoted a and b. The dilation of A by B is given by

Figure 21 Point of interaction and apex trajectories during probe shape mixing.
[
AB A b
b2B
where A b is the set A translated by the vector b. This operation replaces each point of A
with an image of B and then combines all of the images to produce an expanded, or
dilated, set. In scanning probe microscopy, we work with the following sets: S, the sample;
I, the image; and T, the probe tip. In the analysis, one frequently encounters T, the
re¯ection of the probe tip about all three dimensional axes. Villarrubia denotes the
re¯ected tip as P. It can be shown that
I S T S P
In other words, the image is the sample dilated by the re¯ected probe tip. We can see this
in Fig. 21 where the image of the step edge includes a re¯ected copy of the probe shape.
When actually performing calculations, one must convert this abstract notation into the
following expression:
i x max
0
s x 0 t x0 x
x
where i, s, and t represent the surfaces of the sets.

If the probe shape is known, then it is often possible to arrive at an estimate of the
sample shape substantially better than that represented by the raw image. An example of
this analysis is shown in Figure 22. We know that the probe reached the position shown
because the dashed line represents the collection of points visited by the probe apex. The
sample cannot extend into region A because the sample and the probe are not allowed to
overlap. We cannot make the same claim about region B, however, because the probe was
not capable of occupying that space. The subtraction of region A from the image is an
example of erosion. The erosion of set A by set B is de®ned as
\
A B A b
b2B
In our context, it can be shown that the best estimate of the true surface we can obtain
from a probe tip T is
I P
Clearly, it is important to know the precise shape of P.

Figure 22 Sample dilation by the re¯ected probe tip.
It has been known for many years that scanning probe microscope scans can be used
to measure the probe shape. In fact, it is the best way to measure probe shape. Villarrubia
has thoroughly analyzed this process, showing the manner in which any image limits the
shape of the probe. His argument is founded on the deceptively simple identity
I P P I
If we erode the image with the probe and then dilate the result with the same probe, we get
the image back. Villarrubia shows how this expression can be used to ®nd an upper bound
for the probe. It is based on the observation that the inverted probe tip must be able to
touch every point on the surface of the image without protruding beyond that image. The
algorithm implementing this is, however, subtle and complex, so we will not give it here.
One of his publications provides full C source code for implementing the algorithm (55).
Some sample shapes reveal more about the probe than others (56,57). Those shapes
specially designed to reveal the probe shape are called probe tip characterizers. The char-
acterizer shape depends on the part of the probe that is to be measured. To measure the
radius of curvature of the apex, a sample with small spheres of known size might be used.
If the sides of the probe are to be imaged, then a tall structure with re-entrant sidewalls
should be used. A re-entrant test structure for calibrating cylindrical and ¯ared probes on
commercially available CD metrology AFMs is shown in Figure 23. The raw scan data
from such a structure is shown in Figure 24a. Of course, this pro®le is actually a combina-
tion of both the probe and characterizer shapes mixed together. If the tip characterization
structure is of known shape, it can be subtracted from the raw data in order to produce a
mirror image of the actual tip. As a result of this, we can see from Figure 24b that an
undercut ¯ared probe was used for acquiring the data. A particle sticking to the probe wall
is now visible from the extracted image. It should be noted that the characterizer walls
must be more re-entrant than the ¯ared probe's. A simpler and quicker technique used for
calibration in on-line metrology applications most often employs etched silicon ridges to
determine the bottom width of a probe. This series of ridge structures is referred to as the
Nanoedge(58). A triangular-shaped pro®le would appear in the SPM image for a tip of
zero width. In fact, a set of trapazoidal lines is obtained, as shown in Figure 25. The width
of the trapazoidal tip yields the size of the probe's bottom surface, after the width of the
feature ridge (< 2 nm) is taken into account. Therefore an offset in accuracy of only 2 nm
is achieved with this technique. Quantitative correction of subsequent CD measurements is
achieved by subtracting the bottom-width value from the raw scan data. This method is
valid as long as the proximal points are located at the bottom of the probe. Even if this is
the case initially, wear and damage can cause this to change. Fortunately, the undercut
structure can be used periodically to verify the probe shape and, hence, the validity of the

Figure 23 Flared silicon ridge (FSR) tip characterization structure.
bottom-width subtraction technique. The AFM scans in Figure 26 show the same struc-
ture before and after tip width removal from the image data. As a sanity check, one can
overlay the corrected AFM linescan data on top of an SEM cross-sectional image of the
feature to verify that the shapes do indeed match (see Figure 16).
V. SPM MEASUREMENT PRECISION
An example of the precision gauge study for a CD metrology AFM will be described in
this section (59). Currently, the primary AFM commercially available for making dimen-
sional measurements on full wafers in the fab is known as the SXM workstation. This
system is the product of more than 14 years of research and development by the IBM
Corporation in the area of scanned-probe technology (60). The SXM workstation can be
Figure 24 (a) Raw data scan of FSR and (b) ¯ared probe image after subtraction.

Figure 25 (a) SEM sectional image of the Nanoedge tip characterizer, (b) AFM scan of the
Nanoedge structure for tip width estimation.
operated in two modes: standard (1-D) and critical dimension (CD). Standard mode
operates with one-dimensional resonance force sensing and simple one-dimensional ras-
ter scanning, but CD uses two-dimensional resonant force sensing along with the 2-D
surface contour scanning algorithm described earlier (Figure 6). Tip position is obtained
accurately in all three dimensions from calibrated capacitive position sensors at each
axis. Image distortion due to piezoelectric scanner nonlinearity is minimized by using the
capacitive monitors to provide the image data. In the standard mode, precise pro®les of
sample features can be obtained as long as the half-angle of the conical tip is greater
than the structure being scanned. In the CD mode, the tip shape is cylindrical with a
¯ared end. The bottom corners of the boot-shaped CD tip sense the sidewalls of a
feature as the tip scans along them. The position of these protrusions at the bottom
corners of the tip is key for imaging the foot of a sidewall, which can be imaged if only
they remain sharp and are at the lowest part of the probe body. A signi®cant advantage
of the CD-mode AFM over other techniques, including optical and SEM, is that the

Figure 26 AFM image dilation from tip shape mixing.
number of data points can be set to increase at feature sidewalls or abrupt changes in
surface height.
Initially, screening experiments were performed in order to determine the amount of
averaging necessary during each measurement and the relative weighting of different
factors in the precision. AFM images are composed of a discrete number of line scans
(in the X±Z plane)Ðone at each value of Y. For better-precision estimates, each measure-
ment was performed as close to the same location as possible (to within the stage preci-
sion) in order to minimize the effects of sample nonuniformity. Another important
screening task was to determine how many linescans per image are necessary to provide
adaquate spatial averaging of the line-edge roughness in order to reduce the effects of
sample variation on the instrument precision estimate. However, it is desirable to require
as small of a number of linescans per image as possible due to the relatively long amount
of time necessary to acquire AFM data. To determine the minimum required number of
scans, alternate lines were successively removed from an image until a noticeable change in
the edge-roughness value was observed. The minimum spatial averaging for edge rough-
ness (sample variation) was at least 8 linescans over 2 microns of feature length. Initial
repeatability tests performed with different operators indicated that there was no obser-
vable effect with the SXM system when run in the fully automatic mode. Once an auto-

mation recipe was created, the feature was consistently located and measured with no
assistance from the operators.
Gauge precision was obtained by making measurements under different conditions.
Determining the most signi®cant factors to vary depends on one's understanding of the
physical principles of the instrument's operation. Based on the principles of scanning
probe microscopy outlined earlier, the main factors that should be included for this
study are the scanning mode, tip shape, and sample loading effects. A diagram illustrating
the nested model used in this study is seen in Figure 27. The highest level contains the
scanning mode, with the tip next. There were two scanning modes (standard and CD), two
tips for each mode, and three to ®ve loading cycles. A loading cycle value consisted of a
mean of three or ®ve static measurements. Instead of using a three-level nested model, we
found it advantageous to use a separate model for each scanning mode. Due to the three-
dimensional nature of AFM data, it was possible to obtain width (at each Z), height, and
wall angle (left and right) measurements all at the same time. The reproducibility measure-
ments were obtained on nominal 0:25-mm equal lines and spaces that were patterned in
oxide on a silicon wafer. Tip width calibration was also performed before each cycle.
Precision estimates for the standard and CD modes are shown in Tables 1 and 2, respec-
tively. The nested experimental design allows the use of analysis of variance (ANOVA) for
calculating these estimates and actually separating the various components of precision:
xijk kth measurement of the jth cycle on the ith tip

x ij average of n measurements in the kth cycle on the ith tip
x i average of b n measurements on the ith tip
x total of average of a b n measurements taken in the experiment
Figure 27 Nested loop model diagram for ANOVA sampling plan.

Table 1 Standard-Mode AFM Precision Gauge Studies

Table 2 CD-Mode AFM Precision Results

The precision of the AFM in this report is de®ned as
q
s s2e s2c s2t
where s2e is the component for error (repeatability), s2c is the cycle variance component
(reproducibility), and s2t is the tip-to-tip variance component. Estimates of these variance
components are calculated using the sum of mean squares,
P 2 Pb P n
i1 j1 k1 xijk x ij 2
MSe
ab n 1
P2 Pb
n i1 j1 x ij x i 2
MSe
a b 1
P2
bn i1 x i x 2
MSe
a 1
and
p
s^ e MSe
r
MSc MSe
s^ c
n
r
MSd MSc
s^ t
bn
The caret over each sigma indicates that each is merely an estimate of the variance
component. A negative variance component estimate usually indicates a nonsigni®cant
variance component and is set to zero. Variance component estimates can be biased, so the
results should be interpreted with caution. The standard mode height precision (see
Table 1) is on the single-nanometer level, as one would expect. Height data can be gathered
very well with the standard-mode tip. It can also be seen that the error and cycle compo-
nents of wall-angle variance are quite good. As one would expect, tip variance is the main
contributor to the overall imprecision. As described earlier, the angle of the feature side-
wall was larger than that of the probe, so imaging of the probe shape occurred instead of
imaging of the feature.
The CD-mode height measurements (Table 2) were more precise than in the standard
mode and had a variance of s 0:5 nm. The total measurement precision for wall angle
was also much better in the CD mode and were a few tenths of a degree. The top-width
precision was about 2 nm and showed no clear systematic components. A dramatic
improvement in middle- and bottom-width measurements was realized by scanning in
the 2-D mode with the ¯aired boot-shaped probes. Top- and middle-width measurement
precision values were on the single-nanometer level. The improvement in CD-mode pre-
cision values was due mainly to the ability to image the feature walls with the bottom
corners of the ¯aired probes. This ability essentially eliminated the effect of probe shape
on overall precision. The increased number of data points at the feature edges due to the 2-
D scanning algorithm also helped to improve the CD measurement precision. Automatic
algorithm threshold levels for top- and bottom-width locations can also be set more
reliably when there is a greater density of data points at the edges. The AFM top width
(AFM TOP) was determined by averaging a band of points centered at the 90% level of
each edge. The width of each band was adjustable and set to 5% for our tests (61). The

bottom-width (AFM BOT) bands were centered at the 10% level and 5% wide. The
middle width (AFM MID) is centered between the top and bottom. Tip width was
found prior to each cycle in order to calibrate the linewidth measurements. It was
found that nonuniformities of the tip characterization sample produced variations in
the reproducibility results. It is very important to note that the precision and accuracy
of the AFM critically depend on, and are limited by, the tip calibration procedure.
VI. APPLICATIONS TO CRITICAL-DIMENSION METROLOGY
Although conventional scanning electron microscopes (SEMs) have theoretical resolutions

on the nanometer scale, the determination of actual edge location in an intensity pro®le
becomes somewhat arbitrary (62). As a result, the measured linewidth, or critical dimen-
sion, may vary by as much as 50% of the nominal feature size, depending on which edge
detection algorithm, threshold, magni®cation, or beam parameters are used (63). To make
matters worse, the effects on measurement linearity and accuracy are not constant and
actually vary with sample type. It is also not practical to produce a calibration standard
for every measurement case, even if one had a priori knowledge of which one to use.
Despite these limitations, SEMs are still the dominant tool for in-line measurement of
submicron features in the cleanroom. This is due primarily to their nanometer-level
precision, high throughput, and potential for full automation.
A. Evaluation of SEM Measurement Linearity

Linearity describes how well an image corresponds to actual changes in the real feature
size. One must determine if the measurements obtained from an image correspond in a
direct fashion to reality. CD-SEM linearity is typically determined by comparing measure-
ments of tests features to those obtained by a more accurate reference, the AFM in this
case. Typically, SEM width measurements are ®rst plotted against the corresponding
reference values. A regression analysis yields the equation for a straight line that best
®ts the distribution of measured points:
Y1 aX1 b e1
where a is the slope defect, b is the offset, and e is the error component. Therefore, ®rst-
order regression analysis can be used to determine the accuracy (offset), magni®cation
calibration (slope defect), and variation (error term). This technique provides a simple
means to quantify the linearity and accuracy.
A series of etched poly-silicon lines ranging in width from 50 to 1000 nm will be used
in this discussion. The AFM surface rendering of a nominally 60-nm line in the set is
shown in Figure 28a. The effect of sample variation on measurement precision was mini-
mized with spatial averaging by performing measurements on each scan (Figure 28b) at
successive intervals along the line. CD-SEM measurements of the same structures were
then plotted against the AFM reference values, as shown in Figure 29. The degree of
linearity and accuracy is given in terms of the slope defect (a), goodness of ®t (R2 ), and
offset (b) in the ®gure. The extremely small sizes of the poly-Si lines do present a challenge
for any SEM to resolve. In addition to having a measurement offset of 56 nm, the SEM
in this example was not able to resolve the 60 nm line. It is important to note that the
assumption of process linearity is not necessary, because we are comparing the SEM
measurements to values obtained from an accurate reference tool. By plotting the SEM

Figure 28 (a) Surface rendering from AFM data of nominally 60-nm poly-silicon line. (b) A single
linescan from the image in (a).
widths against actual reference values, we should obtain a linear trend in the dataÐeven if
the actual distribution of feature sizes is not.
The relative matching, or tool-induced shift (TIS), between CD-SEMs can be studied
by repeating this process for each system. Once the most linear algorithm has been deter-
mined, the additive offset needed to make the measurement curves (hopefully of the same
shape) overlap with the reference curve must be found. The SEM-to-AFM measurement
offsets are illustrated directly in Figure 30. All three SEMs continued to track changes in
linewidth down to 50 nm in a linear fashion, but there existed offsets between the three
tools. The curves are fairly constant with respect to each other for feature sizes larger than
250 nm. To achieve matching for 300-nm lines, an offset of 6 nm would be added to the
measurements from the ®rst SEM (Vendor A, Model 1). Similarly, an offset of 6 nm is
needed for the second SEM (Vendor A, Model 2) and 8 nm for the third SEM from
Vendor B. It is interesting that the tool from Vendor B actually matched Model 1 of
Vendor A better than Model 2 from Vendor A. Even if an SEM may be from the same

Figure 29 CD SEM measurements plotted against corresponding AFM reference values.
vendor, it doesn't mean that it will be any easier to match to the previous model. It will be
shown later that matching is greatly improved and simpli®ed if one maintains a homo-
geneous tool set of the same model (hardware con®guration). The plot in Figure 30 also
demonstrates the ability of CD-SEMs to detect, or resolve, 10-nm changes in width of an
isolated poly-silicon line. The dip in all three curves at the 200-nm linewidth is highly
suggestive of an error (of about 5 nm) in the AFM tip width calibration at that site. This
reinforces the assertion made earlier that AFM performance is essentially limited by the
ability to characterize and control the tip shape.
B. CD-SEM Matching for Wafer Metrology

The edge detection algorithm that best reproduces the measurement response (i.e., the
linearity) of a reference tool can be found from a series of features having a trend in
Figure 30 SEM offsets for isolated poly-silicon lines.

sizing or pro®le. This is usually accomplished by changing the coded design size (as
was done in the previous section) or stepper focus/exposure conditions about a nom-
inal-size node of interest during lithography. In addition to varying edge pro®le, the
latter technique provides even smaller increments in width variation than can be
achieved by the former method of design size coding. In this regard, measurement
resolution (the ability to detect small changes in the feature size) is tested even more
thoroughly. By varying the lithographic conditions in this way, we are able to test the
combined effects of both the edge pro®le and size on matching. This also provides a
way to see how robust the matching between tools will be over a known range of
litho-process variation. In this algorithm determination method, the AFM is used to
obtain the surface height topographs at each sidewall pro®le (focus level) condition.
AFM topographs at the optimum focus level and at each opposite extreme are shown
in Figure 31. This technique can also be implemented on etched layers by varying the
etching conditions instead. These surface topographs illustrate a wider range in the
process variation than is ever (hopefully) encountered in the production line. Initially,
there was great dif®culty in ®nding an SEM edge algorithm that produced the same
measurement curve (CD vs. focus level) shape as that of the AFM bottom-width
curve. The middle (optimum stepper focus) pro®le in Figure 31 shows why! The
CD-SEM is capable of providing only a top-down view of the feature, so only the
maximum portion, indicated by the dotted lines, is able to be seen. This is the widest
part of the feature and will be referred to as AFM MAX. As the sidewall is varied,
the threshold point in the SEM edge detection algorithm that corresponds to desired
width will also change. Since it would not be practical to try to change this threshold
in production, a worst-case value of mismatch can be speci®ed over a given range of
process variation. The SEM and AFM width measurements are plotted in Figure 32.
Indeed, it is possible to ®nd SEM algorithms that follow AFM MAX in shape, but
not for AFM BOT. Basically, the AFM MAX-to-BOT difference (unrecoverable error)
represents the worst-case SEM contribution to the measurement of pre- to postetch
offset. The true extent of this error depends on the isotropy of the etch process.
MAX-to-BOT offset will also vary for different material layers, feature sizes, and
pattern types (isolated, periodic, holes). The bias could falsely be attributed to mate-
rial or pattern dependency of SEM charging when it is actually a purely geometric
effect.
Figure 31 Feature pro®le variation across stepper focus.

Figure 32 Comparison of different SEMs and their measurements across pro®le variation.
One of the more dif®cult features to image for all microscopes is the hole pattern,
also referred to as a contact or via. SEM (left) and AFM (right) top-down images of a
nominal 350-nm-diameter hole patterned in deep-UV photoresist on an oxide substrate
are shown in Figure 33. Cutaway views of the AFM data are shown in Figure 34. Holes
induce more charging effects with the SEM, so signal collection from the bottom is much
more dif®cult than with lines. A larger size and material dependence of matching para-
meters occurs with holes. On newer-model SEMs, an effective diameter is computed by
performing a radial average about several angles (see Figure 33). The hole structure also
presents the most dif®culty for the AFM reference tool to image. The key issues with AFM
analysis of holes are measurement scan location and angular direction, sample induced
variations, and tip shape/size effects. The usual method for imaging holes with CD-mode
AFM consists of ®rst performing a quick standard-mode overview scan of the hole. The
location of the measurement scan is then set by placing the CD-mode image indicator (the
long, white rectangular box in Figure 33) at the desired position. Essentially, a small set of
CD-mode scans (denoted by 1, 2, 3 in the ®gure) are taken within a section of the top-
down standard image. Only scanning in the horizontal image direction between left and
right are possible due to the method of lateral tip-to-sample distance control used in CD-
mode AFM. The result is that we can obtain CD imaging only within a horizontal band
through the hole center. Centering of this horizontal band through the hole turns out to be
a major component of AFM hole-diameter imprecision and error. Sensitivity to centering
depends on the hole radius, of course. Radial averaging along different scan angles is also
not possible, which can lead to sample variation (i.e., edge roughness or asymmetry)
affecting the measurement quality. Although scanning in different directions is possible
in the standard mode, diameter measurements cannot be performed, because the tip cross
section is elliptical (see the elliptical trace in the AFM image of Figure 33), as seen if one
were looking up at the bottom of a probe.
A new method for improving hole-width measurement with the CD-mode AFM has
been developed by Marchman (63). This new technique involves imaging an entire hole
with the CD-scan-mode line by line and not using the standard overview mode for mea-
surement centering. CD-mode width measurements are then made at each image linescan,

Figure 33 Top-down SEM and AFM views of DUV resist contact on oxide substrate.
except at the walls parallel to the scan direction. The AFM MAX and AFM BOT width
measurements are plotted for each scan line number in Figure 35. A polynomial is then ®t
to the data points, whose maximum indicates the hole diameter after subtraction of the
probe width. This technique eliminates issues of centering and spatial averaging, as well as
improving the static measurement averaging. Residuals from the polynomial ®t can also be
used to estimate the combined precision components of sample variation and tool random
error. The overall hole measurement reproducibility was reduced to 3 nm using this
technique. It should be noted that this technique is still susceptible to errors caused by
changes in the hole diameter along different angular directions through the hole. These
errors can be corrected somewhat by correlating the horizontal diameter in the SEM
image to the AFM value. Then relative changes in hole diameter at different angles can
be found from the SEM image, assuming beam stigmation and rotational induced shifts
(RIS) in the SEM have been corrected properly (64). As noted earlier, eccentricity in the
probe front will introduce errors in hole-diameter measurements in the vertical direction.
A more serious issue is starting to arise for sub-200-nm etched-silicon CD probesÐthey
develop a rectangular footprint. This causes the diameter measurement to be less than the
actual hole diameter when the point of interaction switches from one probe bottom corner

Figure 34 AFM surface renderings of a DUV resist hole.
to the other. A pointed probe shape characterization structure can be used to measure the
probe size in two dimensions to correct this problem.
Hole-diameter measurements from four SEMs, two ``identical'' systems of the same
model type from each vendor, are plotted against stepper focus in Figure 36. The AFM
diameter of the hole (in deep-UV photoresist on oxide substrate) was used to provide the
offset necessary to match these four systems about that point.
The last factor in CD-SEM matching variation to consider at this time is that of
material composition of the feature. The chart in Figure 37 shows the SEM-to-AFM offset
of each tool for different material layers. Measurements where performed at the optimum
feature pro®le on each material combination. This was done in order to isolate the effect
on matching due to material type. The dependence of SEM bias on material type is clearly

Figure 35 Polynomial ®tting to determine the contact diameter.
Figure 36 Contact hole matching data for CD-SEMs.

Figure 37 Process-level material induced SEM bias shifts.
different for all three SEMs. Material dependence of SEM measurement offset was also
studied for dense lines, and different behaviors were observed.
C. Photomask Metrology
Modern optical exposure tools use reduction optics to print patterns from photomasks.
Mask features are typically four times larger than the printed features, so optical micro-
scopes have suf®ced for photomask CD metrology up until the most recent lithography
generations. The alternative metrology tools, electron microscopes and scanning force
microscopes, have, however, encountered problems, arising from the tendency of the
photomask to hold electric charge. SEMs unavoidably inject charge into the sample,
often resulting in charge buildup that not only degrades the quality of the image but
may also damage the mask through electrostatic discharge.
Scanning force microscopes (SFMs) are less inclined to generate sample charging,
but they are, nevertheless, susceptible to charge because of their sensitivity to electrostatic
forces. The SFM can confuse electrostatic forces with surface forces, resulting in a scan
that does not faithfully re¯ect the shape of the sample. This effect is especially troublesome
to probe microscopes that attempt to scan at extremely tiny forces in a noncontacting
mode. The balance-beam force sensor in the surface/interface SNP measures the surface
height with the probe in contact with the surface. The repulsive probe±sample forces are
high enough to make this tool insensitive to these charging problems. The SNP may,
consequently, be operated without special charge suppression measures, such as ionization
sources.
Figure 38 shows an image of a phase-shifting mask taken with the surface/inter-
face SNP. The complicated three-dimensional structures on these masks must be held
to very tight tolerances for the mask to work properly. The ability to perform the
measurement nondestructively is especially important in this situation. Figure 39 shows
the precision achievable in photomask scans, taken on a sample different from that of
Figure 38 (65).

Figure 38 Scan of a phase-shifting mask with the surface/interface SNP. The pitch of this struc-
ture is 1301 nm, and the spaces bewteen the lines are 388 nm deep.
VII. CONCLUSION
In order to gain insight into the physical states occurring during the development of IC
fabrication processes as well as the monitoring of existing ones, it is now necessary to
measure features with nanometer precision and accuracy in all three dimensions.
Unfortunately, adequate calibration standards do not exist for submicron features on
wafers and masks. The scanning probe microscope has become a good option for provid-
ing on-line reference values to higher throughput in-line tools, such as the CD-SEM. The
accuracy of SPM metrology is not affected signi®cantly by changes in the material proper-
ties, topography, or proximity of other features.
Unfortunately, the probe shape can affect measurement uncertainty in several ways.
The radius of curvature of a conical probe must be determined to know the region of
wavelength±amplitude space that can been reached. If the width of a cylindrical probe is
uncertain, then there is a corresponding uncertainty in the width of each measured object.
Durability of the probe tip is especially important. If the probe is changing during a
measurement, it will affect the precision of the measurement as well as the acuracy.
Finally, the stability of the probe against ¯exing is important in determining the precision
of a measurement. Susceptibility to ¯exing sets a fundamental limit how deep and narrow
a feature may be probed.
The SEM will most likely continue to dominate in-line CD metrology for the next
few years due to its nanometer-scale resolution and high throughput. However, a
combination of the SEM and SPM in the future may provide both throughput and
accuracy. The primary advantage of SPM over using SEM cross sections to provide
reference pro®les is that of spatial avaraging. Essentially, each slice of the SPM image
can be thought of as independent cross sections. As feature sizes shrink, it will be
necessary to perform more measurements at each site in order to improve averaging
and minimize the effects of increasing edge roughness. A more thorough estimation of
the amount of averaging required for each technology node is given in the literature.
As fabrication processes are pushed further in order to achieve smaller critical dimen-

Figure 39 Measurement data from an optical photomask taken with the surface/interface SNP.
(a) CD measurement and (b) height measurement. Two tips, designated K40 and L30, were used to
make the measurements over seven days.
sions, the amount of sample variation and roughness as a percentage of nominal

feature size will also increase.
REFERENCES
1. G. Binnig, H. Rorher, C. Gerber, E. Weibel. Phys. Rev. Lett. 49:57±61, 1982.

2. G. Binnig, H. Rorher. Sci. Am. 253(2):50, 1985.
3. J.G. Simmons. J. Appl. Phys. 34:1793, 1963.
4. G. Binnig, C.F. Quate, Ch. Gerber. Phys. Rev. Lett. 56:930±933, 1983.

5. D. Sarid. Scanning Force Microscopy. 2nd ed. New York: Oxford University Press, 1994.
6. T.R. Albrecht, S. Akamine, T.E. Carver, C.F. Quate. J. Vac. Sci. Technol. A 8:3386±3396,
1990.
7. O. Wolter, Th. Bayer, J.J. Greschner. Vac. Sci. Technol. B 9:1353±1357, 1991.
8. G. Meyer, N.M. Amer. Appl. Phys. Lett. 53:1045±1047, 1988.
9. J.P. Spatz, S. Sheiko, M. Moller, R.G. Winkler, P. Reineker, O. Marti. Nanotechnology, 6: 40±
44, 1995.
10. Y. Martin, C.C. Williams, H.K. Wickramasinghe. J. Appl. Phys. 61:4723±4729, 1987.
11. Y. Martin, H.K. Wickramasinghe. Appl. Phys. Lett. 64:2498±2500, 1994.
12. H.K. Wickramasinge. Sci. Am. 261(4):98±105, 1989.
13. M. Tortenese, H. Yamada, R. Barrett, C.F. Quate. In: The Proceedings of Transducers '91.
IEEE, Pennington, NJ, 1991, Publication No. 91CH2817-5, p. 448.
14. S.C. Minne, S.R. Manalis, C.F. Quate. Appl. Phys. Lett. 67:3918±3920, 1995.
15. S.R. Manalis, S.C. Minne, C.F. Quate. Appl. Phys. Lett. 68:871±873, 1996.
16. G.L. Miller, J.E. Grif®th, E.R. Wagner, D.A. Grigg. Rev. Sci. Instrum. 62:705±709, 1991.
17. S.A. Joyce, J.E. Houston. Rev. Sci. Instrum. 62:710±715, 1991.
18. J.E. Grif®th, G.L. Miller U.S. Patent 5,307,693, 1994.
19. R.E. Betzig. U.S. Patent 5,254,854, Nov. 4, 1991.
20. E. Betzig, P..L Finn, J.S. Weiner. Appl. Phys. Lett. 60: 2484±2486, 1992.
21. H.M. Marchman. To be published.
22. J.E. Grif®th, H.M. Marchman, G.L. Miller. J. Vac. Sci. and Technol. B13:1100±11055, 1995.
23. H. Yamada, T. Fuji, K. Nakayama. Jpn. J. Appl. Phys. 28:2402, 1990.
24. J.E. Grif®th, G.L. Miller, C.A. Green, D.A. Grigg, P.E. Russell. J. Vac. Sci. Technol. B8:2023±
2027, 1990.
25. R.C. Barret, C.F. Quate. Rev. Sci. Instrum. 62:1391, 1991.
26. D. Nyyssonen. Proc. SPIE Microlithography 1926:324±335, 1993.
27. J.E. Grif®th, D.A. Grigg. J. Appl. Phys. 74:R83-R109, 1993.
28. EC Teague. In: C.K. Marrian, ed. The Technology of Proximal Probe Lithography. Vol. IS10.
Bellingham, WA: SPIE Institute for Advanced Technologies, 1993, pp 322±363.
29. J. Schneir, T. McWaid. T.V. Vorburger. Proc. SPIE 2196:166, 1994.
30. J. Schneir, T. McWaid, J. Alexander, B.P. Wi¯ey. J. Vac. Sci. Technol. B12:3561-3566, 1994.
31. H.M. Marchman, J.E. Grif®th, J.A. Trautman. J. Vac. Sci. Technol. B13:1106±1111, 1994.
32. H.M. Marchman, J.E. Grif®th, J.Z.Y. Guo, J. Frackoviak, G.K. Celler. J.Vac. Sci. Technol. B
12:3585±3590, 1994.
33. J.E. Grif®th, H.M. Marchman, L.C. Hopkins. J. Vac. Sci. Technol. B12:3567±3570, 1994.
34. H.M. Marchman, J.E. Grif®th. Rev. Sci. Instrum. 65:2538±2541, 1994.
35. M. Stedman. J. Microscopy 152:611±618, 1988.
36. M. Stedman, K. Lindsey. Proc. SPIE 1009:56±61, 1988.
37. K.L. Westra, A.W. Mitchell, D.J. Thomson. J. Appl. Phys. 74:3608±3610,1993.
38. J.E. Grif®th, H.M. Marchman, G.L. Miller, L.C. Hopkins, M.J. Vasile, S.A. Schwalm. J. Vac.
Sci. Technol. B11:2473±2476, 1993.
39. M.J. Vasile, D.A. Grigg, J.E. Grif®th, E.A. Fitzgerald, P.E. Russell. J. Vac. Sci. Technol.
B9:3569±3572, 1991.
40. H.M. Marchman. U.S. Patent 5,394,500, Feb. 28, 1994.
41. J.E. Grif®th, H.M. Marchman, L.C. Hopkins, C. Pierrat, S. Vaidya. Proc. SPIE 2087:107±118,
1993.
42. R.P. Feynman, R.B. Leighton, M. Sands. The Feynman Lectures on Physics. Vol. 2. Reading,
MA: Addison-Wesley, 1964.
43. B.I. Yakobson, R.E. Smalley. Am. Scientist 85:324, 1997.
44. M.M.J. Treacy, T.W. Ebbesen, J.M. Gibson. Nature, 381:678±680, 1996.
45. H. Dai, J.H. Hafner, A.G. Rinzler, D.T. Colbert, R.E. Smalley. Nature, 384:147±150, 1996.
46. H.M. Marchman, et al. To be published.
47. D. Nyysonen, R.D. Larrabee. J. Res. Nat. Bur. Stand. 92:187±204, 1987.

48. M.T. Postek, D.C. Joy. J. Res. Nat. Bur. Stand. 92:205-228, 1987.
49. Digital Instruments, Inc. Nanoscope III.
50. E.H. Anderson, V. Boegli, M.L. Schattenburg, D. Kern, H.I. Smith. J. Vac. Sci. Technol.
B9:3606±3611,1991.
51. D.J. Keller. Surf. Sci. 253:353±364, 1991.
52. G. Pingali, R. Jain. Proc. SPIE 1823:151±162, 1992.
53. D.J. Keller, F.S. Franke. Surf. Sci. 294:409±419, 1993.
54. J.S. Villarrubia. Surf. Sci. 321:287±300, 1994.
55. J.S. Villarrubia. J. Res. Natl. Inst. Stand. Technol. 102:425±454, 1997.
56. J.E. Grif®th, D.A. Grigg, M.J. Vasile, P.E. Russell, E.A. Fitzgerald. J. Vac. Sci. Technol.
B9:3586±3589, 1991.
57. D.A. Grigg, P.E. Russell, J.E. Grif®th, M.J. Vasile, E.A. Fitzgerald. Ultramicroscopy 42±
44:1616±1620, 1992.
58. Nanoedge sample is supplied with the Veeco SXM System and manufactured by IBM MTC,
FRG.
59. H.M. Marchman. Proc. SPIE 2725:527±539, 1996.
60. The SXM Workstation is currently marketed by Zygo Corporation.
61. The top location and bandwidth settings were determined by optimizing the measurement
repeatability.
62. H.M. Marchman. An overview of CD metrology for advanced CMOS process development.
In: International Conference on Characterization and Metrology for ULSI Technology.
Gaithersburg, MD: NIST, March 23±27, 1998.
63. H.M. Marchman. J. Vac. Sci. Technol. B 15:2155±2161, 1997; H.M. Marchman. Future Fab
International, Aug. 1997.
64. K.M. Monohan, R.A. Forcier, W. Ng, S. Kudallur, H. Sewell, H.M. Marchman, J.E.
Schlesinger. Proc. SPIE, 3050:54±67, 1997.
65. Data courtesy of Robert Lipari, Surface/Interface, Inc.

DK1758 ch15

Uploaded by

Copyright:

Available Formats

DK1758 ch15

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DK1758 ch15

Uploaded by

Copyright:

Available Formats

15

Scanned Probe Microscope Dimensional

Copyright © 2001 Marcel Dekker, Inc.

Scanning probe microscopes (SPMs) offer an attractive alternative to optical and

A. Scanning Tunneling Microscopy

Copyright © 2001 Marcel Dekker, Inc.

B. Atomic Force Microscopy

Copyright © 2001 Marcel Dekker, Inc.

Figure 4 Simpli®ed block diagram of a generalized AFM system.

Copyright © 2001 Marcel Dekker, Inc.

A. One-Dimensional Resonant Microcantilever

B. Two-Dimensional Resonant Microcantilever

Copyright © 2001 Marcel Dekker, Inc.

Copyright © 2001 Marcel Dekker, Inc.

D. Balance-Beam Force Sensor

Copyright © 2001 Marcel Dekker, Inc.

E. Lateral Force Resonant Fiber

Figure 8 Head assembly for the lateral/shear-force ®ber SPM.

Copyright © 2001 Marcel Dekker, Inc.

III. SOURCES OF ERROR

Copyright © 2001 Marcel Dekker, Inc.

Figure 9 Abbe' offset error for a tube-based scanner.

Copyright © 2001 Marcel Dekker, Inc.

Figure 10 One-dimensional raster scan algorithm.

Copyright © 2001 Marcel Dekker, Inc.

Copyright © 2001 Marcel Dekker, Inc.

Copyright © 2001 Marcel Dekker, Inc.

Figure 15 Optical microscope image of an etched optical ®ber probe.

Copyright © 2001 Marcel Dekker, Inc.

Figure 17 Flared probe tip etched from silicon cantilever.

Copyright © 2001 Marcel Dekker, Inc.

D. Next-Generation Metrology Probes

Copyright © 2001 Marcel Dekker, Inc.

Figure 19 TEM micrograph of a multiwall nanotube mounted on a pyramidal tip apex.

Copyright © 2001 Marcel Dekker, Inc.

IV. ACCURACY AND CALIBRATION

Copyright © 2001 Marcel Dekker, Inc.

A. Position Sensor Calibration

B. Tip Shape Calibration

Copyright © 2001 Marcel Dekker, Inc.

where i, s, and t represent the surfaces of the sets.

Copyright © 2001 Marcel Dekker, Inc.

Copyright © 2001 Marcel Dekker, Inc.

V. SPM MEASUREMENT PRECISION

Copyright © 2001 Marcel Dekker, Inc.

Copyright © 2001 Marcel Dekker, Inc.

Copyright © 2001 Marcel Dekker, Inc.

xijk  kth measurement of the jth cycle on the ith tip

Figure 27 Nested loop model diagram for ANOVA sampling plan.

Copyright © 2001 Marcel Dekker, Inc.

Copyright © 2001 Marcel Dekker, Inc.

Copyright © 2001 Marcel Dekker, Inc.

Copyright © 2001 Marcel Dekker, Inc.

VI. APPLICATIONS TO CRITICAL-DIMENSION METROLOGY

Although conventional scanning electron microscopes (SEMs) have theoretical resolutions

A. Evaluation of SEM Measurement Linearity

Copyright © 2001 Marcel Dekker, Inc.

Copyright © 2001 Marcel Dekker, Inc.

B. CD-SEM Matching for Wafer Metrology

xijk kth measurement of the jth cycle on the ith tip