You are on page 1of 113

This document serves as the principal conceptual guide for AFM training in the

CharFac. Part A contains the subject matter (46 slides excluding title/outline) of the
prelim exam received as an Excel doc (which is thus an “open book” exam). Part B
accompanies the first lab session, which is a demonstration exemplifying setup,
data acquisition and data processing. Additionally there are four “cartoon” programs
in the dropbox (three executable programs and one animated powerpoint slide) that
the trainee is instructed to run during the course of reading this guide.

1
Sections 1-3 are the core of training and the subject of the prelim exam.

Section 4 is a leader for more advanced applications, if the user wishes to pursue
these methods. (Contact Greg for such further training.)

2
3
AFM is a touching technique whereby one can image and probe the properties of surfaces. The “touching”
metaphor is a little crude, however; at subnanometer scales the nature of the tip-surface interaction can be
very subtle, and involve both attractive and repulsive forces as will be further discussed. The touching is done
by a sharp stylus or tip integrated with a microcantilever device. The bending of the lever up or down can be
used to gauge a pushing or pulling force (or selected to impart a particular force during imaging) provided that
a force balance has been achieved, which is the case of quasistatic methods (contact or peak force) but not
dynamic methods (AC/tapping) as will be discussed. The gauging is achieved by reflecting a focused laser
beam from the backside of the cantilever and into a quad array of photodiodes. Provided that the reflected
spot is straddling the junction between top and bottom halves of the diode array, one can gauge the bending
of the cantilever, known as its vertical deflection, via the vertical displacement of the laser spot at the array. By
further straddling the junction between left and right halves, the lateral spot displacement at the array can be
gauged, thereby the twisting of the cantilever and thus the lateral force on the tip.

As we shall see, the vertical displacement of laser spot will indirectly enable 3D topographic imaging of the
surface (via a feedback loop with the Z scanner, to be discussed) whereas the lateral spot displacement will
enable direct lateral force imaging and measurements. We will also see that other kinds of material
response/property imaging will be enabled by the vertical cantilever bending measurement (i.e., via vertical
force measurement per the amount and sign of bending), in addition to 3D topographic imaging. Finally, we
note that one can implement larger or smaller applied vertical forces (and, thereby, larger or smaller frictional
forces) so as to manipulate matter at surfaces (e.g., scratch), or do more exotic things like nanolithography
by locally oxidizing a surface with the tip or by depositing liquid ink via nanocapillarity at the tip-sample
interface.

Note the use of the term topographic or topography. One should not use the term topology; this latter term
refers to an advanced branch of mathematics (wiki topology), not the shape of a material surface!

4
As another preliminary note, the XYZ scanning device, such as the one shown (spm 3-4, tip
scanned AFMs) is nontrivial and nonideal in its response to the applied voltages that run it.
Most of this nonideal behavior will be exemplified in the demo session, but in Part B of this
guide there are several slides covering the topic. It is a good idea to approach AFM in a
different mindset than say SEM or light microscopy. One cannot assume that just because an
image size and location is specified by the user, there is nothing further to do to make it so.
We will see that this is not the case in AFM.

4
As hinted in the previous slide, atomic force microscopy is not just microscopy. In
fact is it really a triad of capabilities. Firstly, with AFM one is generating a digital
map of surface topography in a blind, touching fashion, not a line-of-site view (as
in scanning electron microscopy or light microscopy). Thus one can digitally
measure heights, slopes curvatures, true surface areas, etc., even Fourier analyze
a surface. Also, one can generate this information on surfaces immersed in liquids,
at high humidity, even as a function of sample temperature. Moreover, because one
uses “feeling” to explore a surface, one can image and generate quantitative
information on material properties: not only things like mechanical stiffness
(thereby elastic modulus), but also things like molecular conformations (crystalline
versus amorphous), charge state and electrical polarization, magnetization, even
thermal properties like the glass transition in polymers (the temperature at which a
polymer, upon heating, transitions from a rigid “glassy” material to a soft, pliable and
stretchable material). Furthermore, one can examine interfacial forces in liquids,
forces that govern processes important to chemical engineering or biology. Lastly,
as already hinted, one can manipulate surfaces: use AFM to pattern, stimulate self-
assembly, produce wear (de-cohesion and/or de-adhesion), and more.

As such, it should be recognized from the outset that AFM “images” are not
microscopic images in the usual sense, where the intensity of reflected, transmitted,
or re-emitted light or electrons is measured by detectors in the far field as a function
of lateral position. Instead, all of the various types of AFM “images” are actually
datasets: matrices of physical measurements. These digital measurements are

5
usually converted to colors for the purpose of viewing, whereas LMs and EMs acquire
analog intensity measurements and convert to digital picture file formats (tif, jpeg,
etc.). AFM “images” (again, datasets) must be analyzed with AFM software such as
Gwyddion. Images in picture-file format are not AFM data!

5
In this example, we compare roughness metrics obtained from AFM height (“topography”)
images of a coating that consists of a glassy polymer and a drug, as prepared (upper right) and
after different degrees of drug elution or release into an aqueous AFM liquid cell (i.e., different
time points as shown). In addition to the standard deviation of height (RMS or root-mean-
square), other metrics like skewness (derived from the cubic deviation), kurtosis (quartic
deviation) and excess surface area are shown. The dominance of protrusions or holes results
in positive or negative skewness, respectively. Kurtosis weights the extremities or tailings of
the height distribution (see insets); in our case a positive number means greater extremes of
high and/or low and fewer values near the mean (i.e., a pointy distribution with a wider base,
more of a “witch hat” shape). For a Gaussian or normal distribution, normalized along the
abscissa to give a standard deviation of 1, the raw kurtosis is 3. Here our value of kurtosis is
evaluated relative to 3 (i.e., 3 is subtracted to give the displayed numbers). One must be
mindful that some programs will instead provide the raw kurtosis (i.e., not subtract 3). The
excess surface area is the percentage by which the actual, surface-integrated area exceeds
the plan-view area. For example, in a 10x10 micron image, a value of 4.6% (A) means 104.6
square microns of actual surface area.

This latter measure of roughness is easily the “best” of the metrics shown, because this
number reflects the lateral scale. Thus in our example we see that dataset A contains the
roughest surface. RMS, by contrast, throws away all X and Y information, and treats the
dataset as a simple stream of numbers, not a matrix of numbers with meaningful distances
along X and Y (see next slide). Thus one could say (and should say) that RMS is a “dumb”
metric of surface roughness. Indeed RMS evaluates dataset C to contain the roughest surface,

6
which is incorrect.

For “good” studies of roughness, search Physical Review Letters and “roughness”.
Here you will find even more sophisticated treatments than the excess surface area
metric. One of these, the power spectral density function, is covered in two upcoming
slides.

6
As already stated, besides looking at surface topography in cross section, AFM
height images are frequently used to quantify surface roughness. Often people will
use the word “roughness” to be synonymous with the standard deviation of Z: add
up all the squared deviations of Z from its mean value across the image, divide by
the number of pixels (minus one) and take the square root. This is the well-known
“RMS roughness” (also known as Rq). But as already stated this metric has its
faults, some of which are illustrated here. It should be obvious that the RMS cannot
distinguish between protrusions and dimples (whereas the skew or skewness,
can make this distinction as a positive versus negative number), nor does RMS
consider the lateral scale over which such features are distributed (X and Y). The
following three slides introduce other metrics that address these aspects of
topography.

7
For many topographies (again not topologies!), the preceding metrics of roughness
do not adequately reveal surface differences, as these differences reside more in
the lateral (XY) information. In particular, very different “looking” surfaces may have
little difference in RMS roughness. This results, of course, from the simple fact that
this metric throws away two dimensions of information! So two surfaces with the
same heights of hills and depths of valleys, but spread over different lateral scales
(e.g., spacing between hills) will have the same RMS roughness.

To retain the XY information one needs a more sophisticated metric. (We previously
considered excess surface area, which is one improvement to RMS.) Here we
demonstrate another type of useful analysis, the power spectral density (PSD)
function. This is a Fourier decomposition of height, that is, a plot of Fourier
amplitude (squared) versus Fourier wavelength. In other words we are
mathematically reproducing the topography with sine waves of different amplitude
and wavelength, and plotting the former (squared) versus the latter. In our example,
we have exposed a glass substrate with abrasive AFM scanning (by strongly
increasing the set point and thereby the loading force), then zoomed out and
imaged at a low set point. Here we show the cantilever deflection image (instead of
the height image; these will be better distinguished in the ensuing slides) because it
highlights the texture of the bare glass as compared to the polymer film surface.
One immediately “sees” that the lateral scale of roughness on the glass – the
spacing of ripples – is smaller than on the polymer film. The Fourier decomposition
highlights these differences: the power spectra density function is higher for the
glass (red curve) over a range of spatial wavelengths (lateral scales) spanning from
about 25-80 nm, then higher for the polymer (blue curve) at larger wavelengths (i.e.,

8
lateral scales).

Integrating the PSD over all wavelengths and taking the square root reproduces the
(lateral-scale independent) RMS roughness, which by the way is essentially the same
for the two surfaces in our example above.

8
The power spectral density function also can be used to measure a dominant
lateral length scale that may be somewhat “buried” in an image. In the above
example the image contains uninteresting/irrelevant “rolling” topography at short
wave number (long wavelength) but also smaller features that generate a dominant
Fourier component, seen as a peak corresponding to a 30-nm phase segregation
length scale. (Note that Gwyddion will instead show the horizontal scale as k,
meaning 2π/λ, whereas the above leaves out the two pi.)

One should recognize that this kind of data analysis is essentially the inverse of X-
ray scattering. One is taking real-space information and generating Fourier space
spectra, intensity vs k; whereas in small-angle X-ray scattering one is collecting
scattering intensity vs k (or q, as typically denoted) and analyzing to extract real-
space information (size, periodicity, symmetry), though not full real-space images.

9
Perhaps a more obvious analysis of lateral size is a grain size distribution
function. Algorithms in Gwyddion can allow one to identify the grains and
demarcate them the surroundings using some color, then export a distribution
function (population) of grain size as enumerated by equivalent disk radius (shown),
or perimeter, surface area, volume, etc. The above example quantifies the lateral
size of about a thousand “grains”, which in this case happen to be isolated pockets
of the drug dexamethasone in a poly(butyl methacrylate) matrix, a 50:50 mixture
coating.

10
Of course with Z(X,Y) maps in hand (i.e., “height” datasets) one can choose to
render 3-dimensional “perspective” type images, perhaps combined with special
colorization. These examples span from hard to soft materials to water droplets (the
latter imaged in noncontact mode in air, a subset of attractive-regime AC/”tapping”
mode).

11
12
So, how fine is this height measurement? One can easily resolve atomic planes
separated by only one unit cell in elevation. Upon imaging this 3D surface topography,
one can also render the same data as a histogram (again, a population or distribution
function), where peak positions quantify surface elevation and peak heights tell
you how much of the surface corresponds to each elevation. We see that these
peaks can be resolved down to ~1 Angstrom of separation. And the variable
measurement of elevation can, in principle, be to a precision of 0.1 Angstrom on a
homogeneous surface. (A heterogenous surface – with different chemistries,
mechanical properties, etc. – renders suspect any statements of precision at the 0.1
Angstrom level, however. And hence, for example, statements about molecular tilt
angles in monolayers. Thus when imaging over the edge of material A onto substrate B,
the measured change of height may be affected by the change of tip-sample interaction,
as we will see later in this guide.)

Note that one may decide to simply view height image data in cross section, of course,
then choose a point on each of two adjacent atomic terraces to approximately measure
the height difference, sort of a lazy-man’s measurement. But this would involve just two
data points out of ~250,000 in a 512x512 pixel image. A histogram analyzes at least
hundreds and typically thousands of data points, which is of course a statistically
superior approach and further helps to provide error bars to a height measurement.
Note the breadth of the above histogram peaks compared to their separation. The
narrow peaks suggest that one could still resolve a 1 Angstrom difference (for some
fictional unit cell) as histogram peaks. Even if such peaks overlapped, however, with

13
AFM one has the option to create a single peak histogram from a selected
terrace, and another from a second terrace, then overlay the histograms to measure
the shift between peak centroids as a measure of mean height differences. Thus one
can measure down to 0.1 Angstrom precision, in principle. In general such spatially
resolved measurements are one of the great strengths of AFM.

13
The simple Hertzian model (which neglects attractive forces, thus adhesion) can
provide an estimate of lateral resolution. Given a loading force F, a rigid spherical
tip of radius R will indent into an elastic half-space of elastic modulus ≈(2/3)K a
distance h according to the expression #1. (Expressions for R and K for the more
general case of a deformable tip of Young’s modulus Etip, and a non-flat surface
location of radius Rsample, are shown in the gray box.) Corresponding to the
indentation depth is a certain contact diameter 2a that increases with F and R but
decreases with K. The lateral resolution is essentially given by this contact diameter.
The 1/3 power in the Hertzian expression makes the contact mechanics somewhat
“forgiving”: a decrease of modulus by three orders of magnitude (say 100 GPa to
100 MPa) results in only one order of magnitude degradation of resolution, from ~1
nm to ~10 nm. But that’s still quite a lot of degradation! So one does not talk about
“the resolution of my AFM” if doing work on soft materials, because the contact
mechanics (i.e., the physics of tip-sample interaction) and the sharpness of the tip
are usually the primary considerations (i.e., not the noise floor of the instrument).

Note that any of these resolution estimates is well above atomic resolution (~0.1
nm), and that for the rubbery case the requirement a<<R is not satisfied, such that
the Hertzian model does not strictly apply anyway (even if there is minimal
adhesion, as say when immersed in water). Moreover a rubber polymer may be
viscoelastic in its response (rate dependent), further rendering the resolution
estimate suspect. Indeed very brief contacts on soft materials, such as when using
peak force or tapping modes, often improve resolution.

14
More complex (and accurate) contact mechanics models include the effects of attractive
forces, which among other things further increase the contact area and thus degrade
resolution. The most general and rigorous models either produce multiple analytical
expressions that must be solved parametrically, or must be solved numerically from the get-
go. These models account for attraction both at the tip-sample interface (solid-solid
adhesion) and in the periphery beyond the contact zone (longer-range forces or possibly
forces due to a capillary meniscus). Two simpler models, the JKR and DMT models,
account for only one or the other. The JKR treats only the attractive forces within the
contact zone whereas the DMT treats only the attractive forces beyond the contact zone.
As a result the JKR model is more appropriate for relatively soft materials, where attraction
presses the materials together to deform and grow the contact area (such that this
attraction dwarfs the peripheral attraction beyond contact). There is, however, more to this
issue than just softness; in particular the values of radius of curvature and strength of
attraction. The DMT is actually very simple in that it adds a constant to the external load but
otherwise retains the functional form of the Hertzian model relating contact radius a to
force, modulus K and radius of curvature R. (I.e., it assumes the same shape of the contact
as the Hertzian shape.) The JKR by comparison contains four terms, two under a square
root. Three of these terms contain the work of adhesion γ. If this parameter is zero (no
attraction) these terms drop out such that one retrieves the Hertzian expression. The
expression for indentation distance h also has a term containing γ that reduces to the
Hertzian expression if γ=0.

The JKR and DMT models actually are extremes within a continuum of parameter space

15
that spans from rigid materials, small R and small γ at the one end (DMT) to soft
materials, large R and large γ at the other (JKR). As one can imagine there are many
other cases within this three-dimensional parameter space such that, strictly
speaking, the DMT and JKR models often do not properly apply to the actual system
being studied. Nevertheless they are frequently used for ballpark estimates or to
analyze trends, being considerably better than the simple Hertzian treatment.

15
Besides contact mechanics, the finite shape of the tip results in broadening or
“dilation” of images of nano-protrusions, or lateral shrinking of nano-chasms,
for purely geometric reasons. Likewise, the finite sharpness of the tip limits how far
it can extend down into steep chasms and cracks. The result is a cusp-like apparent
topography when viewed in cross section. A rightward scanning tip starts down into
a valley until its leading side hits the other side of the valley, upon which the tip
begins its outward trajectory. Indeed, in cross-sectional profile, many surfaces
produce height images laden with these cusps, as the restriction of a tip fitting into
the cracks tends to be the rule rather than the exception.

Play with the cartoon program ProbeSimulator.exe to better perceive the above
issues.

Importantly, because the tip does not always fit down to the bottom of valleys, the
range of heights measured is a function of tip sharpness, which may be
variable (as a tip wears down) even during a single experimental session! Thus
“dumb” metrics like RMS may be even “dumber”, if one does not bother to assess
the variable sharpness of the tip. This can be done qualitatively by viewing the
“sharpness” of images (say if one returns to the first sample after say the fourth
sample of one’s session), but there is a more quantitative gauge: the strength of tip-
sample adhesion, as can be approximately measured from the force required to pull
the tip from the surface (described later). This “pull-off” force scales linearly with
the radius of curvature of the tip in the JKR and DMT adhesive contact

16
mechanics models previously described. So relative changes in tip sharpness can in
fact be quantified.

Note that of course geometry is geometry: The above-illustrated limitations are


independent of mode of operation (contact, peak force, AC/tapping). Geometry does
not magically go away just because one is using, say, tapping mode! (In fact surface
curvature can be a worse problem in AC/”tapping” mode, as shown later…) A
deformable contact may look different in say contact versus tapping modes, but
that’s a different issue. The rigid-body geometric issues, considered in this slide, are
the same for both modes of operation.

16
17
Now for the conceptual core of AFM operational principles.

There are two fundamentally different strategies of tracking surface topography via
feedback, so as to generate mappings of this topography. Here we consider cross-
sectional depictions of an idealized point-like force sensor tracking a material’s
surface topography as a contour of constant (a) force or (b) integrated normal force
gradient (dF/dZ) sensed during oscillation with respect to the vertical midpoint of the
sensor position, termed its “fly height”.

Within Strategy #1 reside contact mode and pulsed force / peak force mode (and
now there are additional vendor names besides these two). Strategy #2 applies to
dynamic (AC/”tapping”/vibrating) mode.

Run the cartoon AFMmodes.ppt to see how different these modes of tracking are.
Please note that the cartoon doesn’t quite do justice to dynamic mode regarding
vibrating cantilever shape, nor regarding the number of oscillations per pixel
(hundreds).

18
Real AFMs use X, Y and Z to move the probe or sample. (The simplest instruments, such as spm1-2, scan the sample;
but some scan the tip, such as spm3-4, particularly when extreme sample environments are sought, e.g., high humidity
or controlled sample temperature. Normally an AFM does not have a scanner at both locations, unlike the above
depiction. Some vendor models do have two, usually to separate the XY scanner from the Z scanner. There are good
reasons for doing this, but it’s more expensive.) Whereas X and Y are programmatically scanned independent of what
is happening at the tip-sample interface, Z is reactively displaced by a feedback circuit so as to keep a control signal
constant. In “contact mode” AFM, where the tip continuously slides over the surface, the vertical cantilever bending or
“deflection” measurement is the control signal that drives the feedback. Thus the instrument attempts to keep the
cantilever at some constant degree of bending by displacing Z (of probe or sample) up or down as the tip rides over the
hills and valleys of the surface. This Z displacement becomes the measure of surface “height”… the cantilever
bending does NOT provide a height measurement during imaging. The intrinsic speed of the feedback circuit, as
well as operator-selected gain settings and scan speed, largely determine how precisely this Z displacement tracks the
topography of the surface. The operator-selected cantilever deflection value determines the external loading force (the
pushing force of tip to surface by an upward bent cantilever, which is akin to a compressed spring). The intrinsic
interfacial forces between tip and sample add to the external load to determine the total load or contact force. Moreover
this interaction results in frictional forces that can be used to contrast dissimilar materials and, of course, enable studies
in tribology (friction/wear/lubrication).

Play with all of the sliders in the cartoon program AFMModel.exe to illustrate the feedback concepts in contact mode.
Turn up the Gain to improve tracking (provided you have adequately approached sample to tip) but note the feedback
oscillations that result at very large gain setting. (In the demo session we will do the same with a real AFM.)

For peak/pulse force and AC/”tapping” modes, things are a bit different: In peak force mode the quasistatic cantilever
deflection drives the feedback, just as in contact mode (i.e., still Strategy 1 in previous slide), but this deflection is
evaluated at its peak value during repeated, high-frequency (but well below resonance) approach-retract cycles. In
AC/”tapping” mode the dynamic amplitude of a resonantly vibrating cantilever (~2 orders of magnitude higher

19
frequency than peak force mode, i.e., hundreds of oscillations per pixel measurement) drives the
feedback (Strategy 2 in previous slide).

Very importantly, it should be noted that the Z displacement measurement is in truth simply
the voltage applied to the Z scanner to make it displace in Z. It is not some independent
measure of Z displacement by some sensor. (One can choose to do such a thing only on spm4,
at the expense of increased measurement noise.) Importantly, this voltage is converted to
distance units by a simple (linear) coefficient (i.e., nm’s per V). But we will see in the demo
session that the scanner is in fact NOT a linear device. Thus the nonlinearity of the Z scanner,
as well as some parasitic Z displacement from the XY scanner, will complicate the “height” image
data. These higher-order nonidealities are normally treated after the fact, during post processing
of height images (i.e., by fitting a curved mathematical surface through the image and subtracting
out), as we will do during the demo session.

Clearly, this is not SEM or light microscopy!!

19
This slide shows the feedback scheme for a tip-scanned AFM (e.g., spm3-4). As
already stated, the vertical cantilever deflection signal is a differential output, from
the top and bottom parts of the split photodiode. Each quadrant puts out a voltage
signal (strictly a photocurrent, but passing through a standard resistor to give a
voltage) proportional to the intensity of laser light impinging on the quadrant. In
detail, the top two outputs are added as are the bottom two, then the difference
taken and normalized to the sum of all four outputs (to remove total intensity
fluctuations). Then a difference is computed between this final photodiode signal
and the operator-specified setpoint; this difference is in fact an error signal – it
quantifies the departure of the cantilever deflection from the operator-specified
value. This error signal drives the feedback circuit to displace the Z scanner in an
attempt to null the error. This reactive Z displacement is measured for each pixel
location and rendered as colors to form a height image (left, of an ultrathin polyvinyl
alcohol film containing a dewetted layer on a wetting layer that covers a mica
substrate).

At the same time, the error signal itself is also measured for each pixel and
rendered as colors to form a deflection image (right). The operator tweaks the
feedback circuit’s gain, scan rate and set point so as to minimize the contrast in
the deflection image, because it is error! (Nevertheless one notes that the deflection
image at right is visually appealing, because it accents transitions of height at edges
of film domains, as if the reflections and shadows of light due to illumination from
the left. Deflection is proportional to the local slope δZ/δX, i.e., can be called
“differential topography”. Indeed a numerical differential of the height image data set
Z(X,Y), along X, is essentially identical to the deflection image.) Obviously AFMs

20
use multichannel data acquisition systems for synchronous capturing of multiple
images, here the height and deflection images, and in strict analogy the height and
amplitude images in AC/”tapping” (dynamic) AFM.

Also note the approximate ranges of XY and Z scanner elements shown in the slide
at lower left.

20
Of course, in this universe, one’s sample surface is not going to be exactly and
fortuitously parallel to the scanner’s XY plane! Thus the raw “height” data will
contain an arbitrary tilt (as the feedback necessarily reacts to the overall surface
tilt to displace the Z scanner during back and forth scanning, just as it reacts to the
local surface roughness). This tilt usually will need to be subtracted out of the data
via a “planefit” to enable a better view of the topography of interest. Another reality
is the limited Z range of the scanner, usually a few microns. Of course this means
that coarse approach of tip to sample will be actuated some other way, namely
by the Z motor (stepper). (On spm1-2 the motor lowers the head, to which the
cantilever chip is attached, towards the sample; on spm3-4 the motor raises the
sample plate towards the tip.) Finally it is worth noting that the ability to measure
cantilever deflection (and thus, ultimately, the range of measurable force) is limited
by the split photodiode scheme to at most 100’s of nanometers. But of course we
are not dependent on the cantilever deflection range for topographic imaging; rather,
for force measurements or force settings during scanning procedures (e.g., the
maximum pushing force attainable if seeking to produce wear or plastic indentation,
or maximum pulling force achievable to measure adhesion on a very sticky
material).

(a) Raw 35x35-µm height image of gold and aluminum strips lithographically grown
on silicon. (b) Same image after a best-fit mathematical plane has been subtracted
from (a).

21
Right: Schematic depiction of the series relationship between Z-displacing elements:
Z motor, Z scanner and cantilever spring / tip.

Higher “order” mathematical surface subtractions will be treated in Part B of this


document.

21
At the extreme of small vertical displacements, it is worth noting that the finest
increment of Z, as can be actuated by the Z scanner, is not achieved by default.
The data acquisition system is 16 bit, such that the Z range (again, several microns
if fully utilized) is subdivided into 216=65536 increments. This means the finest
increment is of order 1 Angstrom… and thus on very smooth surfaces the
digitization becomes apparent in the data (usually after a plane subtraction), as
shown in the top plot from an image of a silicon wafer with native oxide. But one has
the option to give up Z range so as to obtain finer digital increments of Z (e.g.,
bottom plot). If one does so, then one must be all the more mindful of sample tilt, or
for that matter time-dependent vertical drift (due say to the relaxing sample mount
such as tape), because now one has correspondingly less Z range to accommodate
this tilt or drift. (You don’t get something for nothing!)

22
23
Lateral force measurement during contact mode can be very powerful because of
extreme sensitivity to material composition. In the above example (lower right
image), a highly crystalline, semi-continuous surface layer of polyvinyl alcohol
exhibits much lower friction (here rendered darker) than the more amorphous
underlayer. The reason is molecular- to atomic-scale degrees of freedom: the
passing tip’s kinetic energy is partially transferred to local motions (e.g., bond
vibrations, rotational isomeric conformational transitions), what could be called
conversion of tip motion to heat. The resistive shear force that is sensed is the
manifestation of this energy transfer as a physical measurement (of force). A lesser
availability of molecular motions in crystalline domains results in lesser friction force.

Even on homogeneous, atomically flat, hard materials, the above concept is


physically at play. Here atomic-scale “stick-slip” motions are occurring at the tip-
sample interface, and these processes are irreversible in the thermodynamic sense
and thus energy dissipative (i.e., energy is taken away by crystal vibrations excited
in the solid by slip events). The friction force relates to a shear stress which
modifies the energy landscape: lowers thermal activation barriers for slip. Thus the
friction force goes up at faster scan speeds, because this allows fewer “attempts”
(at a frequency determined by cantilever’s torsional resonance frequency in contact)
for the tip-sample contact to yield to slip per measurement of lateral force. For an
analogous reason the friction force goes down with increasing temperature,
because there is more kT energy to get over the barrier to slip. So clearly the
measured friction force, as well as its rate and temperature dependence, will
depend on the surface polarity (which affects the “grab” of the tip) as well as the
lattice characteristics (e.g., phononic motions). In other words, friction force HAS

24
TO BE sensitive to material structure!

Lateral force imaging thus provides the very important ability to detect surface
contamination as patchy domains. These domains could be relatively high friction
force (perhaps because the contaminant is sticky, or because it is soft or loosely
bound and thus mobilized by AFM shearing, enhancing dissipation); or they could be
relatively low friction force (due to low surface energy, i.e. apolar character, and/or
lubricating behavior). As stated earlier, the quad photodiode in conjunction with the
cantilever’s twisting degree of freedom enables quantification of lateral forces,
whether during sliding or shear modulation (static friction regime with small shear
wiggle, ~1 nm, sensitive to shear stiffness). In basic training we will only discuss
sliding friction (and briefly).

24
It should be obvious that either the trace (usually meaning left-to-right tip scan) or
retrace (right-to-left) scan will provide materials contrast. Some applications have
stopped at this point and simply used this contrast (whether during trace or retrace)
to qualitatively identify the ingredients in coatings, for example, given some
knowledge of material differences that should yield lesser or greater friction. For
now it’s useful to see a couple of examples. One is the case of a multilayer film-on-
substrate system that has been cleaved and stood on end for cross sectional
imaging. Here we see that even fairly similar semiconductors, like GaAs and
InGaP, exhibit frictional contrast (in air). It was reported that these cross-sectional
friction maps could resolve layers down to 5 nm resolution, which is comparable to
TEM images (but without the lengthy sample prep required to make a TEM section
for imaging).

25
If you think about it, lateral forces must consist of both “true” friction as discussed
so far – shear forces that resist motion (i.e., are energy dissipative) – and
nonfrictional lateral forces that derive from the lateral component of the tip-sample
normal force and scale with the slope of topography dZ/dX along the fast scan axis.
See illustration at top right. But these latter topographic effects can be removed, to
first approximation, by subtracting trace and retrace lateral force signals (provided
adequately slow scanning and proper tracking via feedback gain setting). The
remaining data predominantly contain differences in true friction: shear forces that
resist motion, that is, are energy dissipative.

As exemplified in the preceding two slides, energy dissipation can be very sensitive
to the molecular scale structure of materials (e.g., amorphous vs. crystalline) and to
surface chemistry. Indeed, a histogram of the lateral force difference image (lower
right) looks very much like spectroscopy, one could say “friction spectroscopy”; peak
positions, breadths, and lineshapes can be analyzed as is done in traditional
spectroscopy. Conversely, adding the trace and retrace lateral force signals
removes the material-specific frictional (dissipative) forces and retains slope (dZ/dX)
derived forces, which naturally “decorate” the edges of nanoscale objects, film
domains, etc. You might call this latter force “sandpaper friction”, as it trivially
derives from topographic slope (thus sometimes called “asperity collisions”, i.e., the
tip as an “asperity” running into an edge of a sample “asperity”), irrespective of
material differences. Usually one is more interested in the right image, the material-
sensitive matrix of measurements. Indeed the information in the left image may be

26
(redundantly) obtained by digitally taking the first derivative of the height image, or by
collecting the deflection image as shown a few slides back.

26
In large images of reflective samples, the raw lateral force images also contain a
quasiperiodic optical background that derives from interferometry between laser
light reflected off the sample and off the cantilever. This background may also
contain an overall tilt. This optical artifact of course does not know which way the tip
is moving, thus is independent of scan direction so that the background can be
removed by subtracting trace and retrace lateral force signals. So, both (i) slope-
derived lateral force components (previous slide) and (ii) optical artifacts are largely
removed by this mathematical operation. Often the result reveals domains of
variable chemistry or other properties that are difficult to discern in the raw trace or
retrace images, as exemplified above.

It should be noted that the above example was obtained using a closed loop
scanner (further discussed in Part B and during the demo session), which is better
for very precise subtraction of trace and retrace images. This type of scanner can
be trivially used on spm4, with brief training (few minutes, provided one has been
trained to use spm4). For spm1-2 we have a 3rd party closed loop scanner with its
own electronics and software, thus requiring substantial training. (And it must be
scanned a factor of ~2 slower.)

27
In contact mode, shear forces can cause cohesive failure (rupture, tearing) in the
near-surface portion of bulk materials or adhesive failure for thin films (peeling from
substrate). On the good side, this can be used for analytical measurements of
cohesive or adhesive strength. (Greg can provide advice on such applications.) In
many cases the onset of such “wear” phenomena can be quantitatively seen as an
abrupt increase of slope in a plot of (kinetic) friction force versus vertical applied
force (load). Here each datum in the plot at upper right was computed by integrating
lateral force around complete friction loops across an entire image or a subdomain
of an image (ignoring the static friction turnaround data, i.e., only assessing kinetic
friction), an example of one loop being shown in the upper left.

In most soft-materials work with AFM, the above sorts of wear are, of course, not
being sought. They may, however, be unavoidable. (Try repeatedly raster scanning
polystyrene in contact mode.) The intrinsic friction (shear) forces may unavoidably
tear the material or cause adsorbed contaminants to move around or even desorb.
Peak force mode and dynamic (AC /“tapping”) mode AFM are principally used to
avoid such problems. Here the contact time is very small such that there is little
lateral motion, thus little shear force, during each contact. In dynamic (AC/”tapping”)
AFM the contact time, if there is contact at all, can be submicrosecond such that the
lateral scan distance during a given contact is subangstrom. Thus the extremes of
nonperturbative AFM are achieved using dynamic AFM.

28
Of course to measure or control the vertical applied force, one needs to know the zero of this quantity (see horizontal scale of
previous plot of friction vs vertical applied force), and one needs to further account for any intrinsic vertical force, usually due
to attraction of tip to sample (e.g., van der Waals and capillary forces). In contact and peak force modes these issues are
graphically analyzed in a force curve (also known as force versus distance curve, force-displacement curve, force
spectroscopy, and other names), as treated above. As depicted in the top cartoon, the tip begins far from the surface (no XY
scanning involved here), is approached to one location on the sample (or sample is approached to tip depending on instrument
design) until a touch results, then retracts until separation occurs, a full cycle. While off the surface, a flat horizontal trend (red)
denotes the value of zero force (i.e., at some finite deflection read signal value, in volts). Instead of a trivial onset of repulsive
force upon touching, and loss of contact after the force drops back down to zero during retraction, one usually observes the
manifestations of attractive forces, as follows. Firstly, during approach and just prior to contact, attraction causes a “snap to
contact” (1-2): attractive forces accelerate the tip to the surface because the resistive spring force of the cantilever cannot
maintain a force balance. In principle this could be due to a long-range attractive force gradient dF/dZ exceeding the spring
constant k of the cantilever (as suggested in some books and review articles, but strictly applying to vacuum with a very clean
tip and sample). In practice in air, adventitious and mobile contaminants on tip and sample form a capillary neck that pulls the
tip to the surface. In any event, once contact is made, further approach causes the cantilever to bend up, thus force increase (2-
3). Upon retracting, the force curve retraces itself until point #2 is again reached, but now the contact is not broken; instead the
cantilever must bend down until the generated pulling force overcomes the “stick” and snaps the tip off the surface (4-5). A
hefty pull indeed is needed because the capillary neck due to mobile surface and tip contaminants (water, hydrocarbons),
together with solid-solid adhesive forces, must be overcome. A useful result is that the snap-off force is a measure of tip
bluntness (proportional to radius of curvature); thus blunting due to usage can be assessed.
What is the magnitude of force in force units? Two calibration steps are needed: conversion of the raw deflection signal in
volts (properly zeroed) to deflection in nanometers; then multiplication by the cantilever spring constant k (N/m, i.e.,
nN/nm) to obtain force. The second step is trivial if one uses the manufacturer-specified spring constant. (We can discuss the
alternative: experimental determination of k.) The first step is performed by calibrating the vertical axis to the horizontal axis on
a rigid sample, assuming a precalibrated Z scanner and no indentation. Thus for each incremental step of Z in nm the cantilever
must deflect by the same amount in nm, such that the force curve slope in contact has unity magnitude once calibrated (nm
change in deflection per nm change in Z). The conversion factor to force this unity slope is the inverse of the “sensitivity” S
(V/nm; i.e., one must divide by S to convert deflection from units of V to units of nm). S is thus the slope of the raw force curve
in V/nm, the sensitivity of the output of the photodetector in V to changes of Z in nm.
The applied force is of course determined by the cantilever deflection relative to its zero-deflection state (red). The total contact
force, however, is usually the sum of this applied force (positive or negative, in general) and a physical attractive force pulling
the tip to the surface due to the interfacial tip-sample interaction: van der Waals attraction between tip and sample solids, plus

29
capillary attraction. (Under liquid immersion, sometimes this interfacial force is repulsive, say due to
polymeric steric forces or unscreened charged tip/sample surfaces of the same sign; also in some
cases in air above charged samples given a like-charged tip.) Often the magnitude of the pull-off force
is definitively referred to as the adhesive force (as labeled in the graph). A fine point, however, is the
fact that the pull-off force quantifies tip-sample adhesion only at pull-off, by definition; whereas the
operative adhesive force at other points within the force curve (and thus during imaging) may be
somewhat larger because the tip may be pushed into the material, growing the contact area over which
adhesion is felt.
It is worth noting that if using dynamic (AC/”tapping”) mode, the physics present in force curves (i.e.,
approach-retract cycles) does not somehow magically disappear! It is present in the measured
dynamic amplitude and phase, and their dependences on Z (which one can elect to measure, and
indirectly sample by choice of amplitude setpoint); but not in a way that one can readily extract from
such data. Thus one has not somehow bypassed this physics by operating in dynamic mode; rather,
one has lost the ability to probe this physics directly as force versus distance. So you could say
there is a further element of “blindness” that one accepts when choosing to use dynamic AFM.

29
30
In “pulsed force” (spm4) or equivalently “peak force” (spm1-2) modes, only select features of force-Z cycles are compiled into
images. As already stated, one of these is tip-sample adhesion or snap-off force. Another can be the time-location of the
maximum force at turnaround (spm4), the so-called loss angle (i.e., relating to viscoelastic loss). A third is a stiffness
measurement based on a deflection differential between two points per cycle preselected in time (spm4) or based on F-Z curve
shape analysis (spm1-2, a better implementation), usually during retraction (i.e., while still in contact). Note that this mode uses
(by necessity) a much higher data acquisition rate than contact (or dynamic) modes, 5 megasamples per second. Thus one can
further view the “ringdown” oscillation of the (diving board) cantilever after the tip breaks away from the surface. In some cases
this oscillation is still present at the subsequent touch, but this is generally not an issue. The amplitude of the ringdown of
course varies as tip-sample adhesion varies across the imaged surface, but again this usually plays no analytical or operational
role.

One can also acquire images at different setpoint values corresponding to different maximum force during contact. In particular
one may find much larger tip-sample adhesion when higher preceding forces were achieved (i.e., because of higher Fmax
setpoint), or for cases of having longer contact times, due to memory effects (which are not predicted by equilibrium
thermodynamics models of adhesive contact mechanics such as DMT or JKR).

Importantly, nowadays one can additionally acquire the entire force curve at each pixel location!! These composite “big
data” files can be as large as gigabyte-regime (best stored on a portable USB hard drive that you can tuck in a pocket) but can
also be smaller (i.e., more quickly loaded and manipulated during post processing), if one uses smaller acquisition site
densities. In post-processing, one can pull up individual force curves (again, known as “force spectroscopy”) from selected pixel
locations, and one can render new images from different parts of each force curve cycle. Obviously the time spent on data
analysis can reach new high levels, but hopefully new analytical insights can be gained about one’s materials. This is, of
course, the whole point of “big data” scientific work, and era recently entered.

Topography is imaged akin to contact mode. In peak force mode it is the maximum quasistatic cantilever deflection, and thus
maximum applied force, of each approach/retract cycle that is subsequently driving the reactive displacement of the Z scanner,
independent of the fixed-amplitude Z cycle by which force curves are examined (and potentially saved). So as this Fmax

31
deviates above or below setpoint from touch to touch, time-integrated error signal can grow (positive or
negative, on steep upward or downward regions respectively) and thus more strongly drive the
feedback displacement of the “DC” part of Z. But unlike contact mode, this setpoint maximum force in
each F-Z cycle is measured relative to the undeflected state of the cantilever when the tip was far
from the surface. Thus any drift of the latter “zero” of force bears no impact on the image, whereas in
contact mode this drift when present typically results in contact forces that drift upwards with passing
time, in some cases dramatically (i.e., by a factor of 100!). We will see this in the demo session.

So pulsed/peak force mode requires much less “babysitting” than contact mode. We will also see later
that it requires less babysitting than AC/tapping mode (for different reasons). Pulsed/peak force mode
is operationally the simplest of the three modes, and thus it is usually the “best” mode for
casual users.

31
In this pulsed force mode example, the slope of force versus time within the
selected (shaded) time interval during retraction was used to quantify stiffness
(elastic recovery after touch), as it must relate to the slope of each force curve
(force vs Z) that is being acquired in time. The maximum negative force during
retraction is a measure of tip-sample adhesion. The shown topography image is in
“differential” form, meaning either the raw error signal image, which is effectively
dZ/dX, or height image data Z(X,Y) that has been digitally differentiated to give
dZ/dX (X,Y).

Note that PE is softer than PU, but on average PE is less sticky than PU. This may
seem counterintuitive: all other things being equal, the softer material will yield a
larger contact area with the tip, such that a larger pulling force is needed to
separate tip from sample. In this example, however, all other things are not equal.
PU is a more polar polymer than PE, and the AFM tip is nominally polar (SiOx) such
that it is more strongly attracted to the PU than the PE. So this example shows how
stiffness and adhesion imaging sometimes separates mechanical and chemical
differences; whereas friction (contact mode, already discussed) and phase
(AC/tapping mode, to be discussed) will often convolve these different material
properties. Once again we see that pulsed/peak force mode has the fewest pitfalls
of the three modes, whether operational or interpretive.

32
33
The first step to understand AC/”tapping” mode is to examine the physics of a resonant
oscillator. (We will not go to the point of solving 2nd order differential equations for Newton’s 2nd
law of motion of the cantilever/tip, as a full discussion would entail.) If a small (sub-nm)
vibration is applied to the base of the cantilever precisely at its resonant frequency, a much
larger amplitude (usually tens of nm) results at the tip end. This strong amplification is
quantified in the frequency dependence of this amplitude of response (of cantilever deflection
near the tip end). For an ideal driven harmonic oscillator, with damping force proportional to
velocity, the frequency dependence of the steady-state (dynamic equilibrium) amplitude
is a Lorentzian peak (see inset equation, sorry about the arcane symbols), the sharpness of
which increases with decreasing damping. Most of the damping, for ordinary operation in air, is
in fact due to the air medium, as opposed to the intrinsic anelasticity of the cantilever itself.
(Thus the resonance is much broader if operating in liquids, and much sharper in vacuum.)
The quantification of sharpness is according to the quality factor Q; which, it turns out, is
approximately equal to the ratio of resonance frequency to the full peak width at 0.7 of
maximum. (We will return to the importance of this parameter.) The other quantity of
importance is the phase lag of the tip oscillation relative to the excitation at the base of the
cantilever. At resonance this phase lag is 90 degrees. Importantly, the phase changes steeply
to either side of resonance (important in the next few slides). The analytical form of the
frequency dependence of phase is an inverse tangent function, which means that away
from resonance there is an asymptotic approach towards 0 or 180 degrees. Note the inverted
phase scale.

In our instruments (spm1-4) it is even more confusing: the phase numbers are 90 minus the
numbers shown above. So at resonance the phase is “zeroed” in software, rather than
being set to 90 degrees as would be physically correct (and as some other vendors do, e.g.,

34
Asylum). We will return to finer points relevant to this issue.

Play with the cartoon program DrivenOscillator.exe to see how the drive frequency
and Q factor affect the vibrational motion of the modeled tip/cantilever system.

To understand the importance of these details of amplitude and phase near


resonance, we next consider how the free resonant behavior is modified upon tip-
sample interaction.

34
Given a particular free oscillation amplitude, tip and sample are initially brought
together (by the Z motor) until the amplitude is reduced to the operator-
selected set point amplitude (in raw voltage units of amplitude signal). Following
this “engagement” of oscillating tip to surface, if we imagine moving laterally (not in
Z) to an adjacent pixel location where the surface happens to be higher, then this
amplitude (intuitively) would be forced to decrease by the tip-sample interaction. But
the feedback system would react and displace in Z such that the sample is moved
away from the tip (or cantilever chip is moved away from sample, i.e., dependent on
instrument design) so as to regain the setpoint amplitude. In reality these will not be
separate, sequential actions as suggested here, rather a continual on-the-fly
process driven by the feedback circuit, including displacements in both Z directions
in order to, hopefully, reproduce the actual surface topography (with caveats yet to
come).

35
The above schematic for tapping/AC mode for a tip-scanned AFM should be
compared to the similar one for contact mode considered earlier in this guide.

Extra components here include (i) a modulation unit (gray box) that, firstly,
generates a high-frequency sinusoidal Drive signal; (ii) a mechanical oscillator that
is excited by this signal to generate a vibration that propagates to the cantilever chip
and thereby the base of the cantilever; and (iii) a lock-in amplifier (or functionally
equivalent electronic circuit, and contained in same gray box) that generates a DC
voltage signal proportional to the amplitude (whether rms, peak-peak or whatever)
of a sinusoidal input signal, and which further measures the phase shift φ between
the AC excitation Drive signal that is output from the same box and the AC
Deflection response signal that is input to the box. This latter sine wave is
generated by the split photodiode because the tip-end of the cantilever is oscillating
up and down sinusoidally in time, changing the angle of inclination of the cantilever
sinusoidally in time, causing the reflected laser spot to move up and down at the
photodiode’s top/bottom junction sinusoidally in time.

Note, importantly, that the raw phase shift φ is not equal to the phase shift between
tip and base ends of the cantilever. In Part B of this document and in the demo
session we will discuss how one ensures that the final measured phase shift is the
latter one (by subtracting out the “apparatus phase shift”), as used to generate the
phase image shown at right.

36
Regarding the feedback circuit: The DC output of the lock-in amplifier is proportional
to the amplitude of vertical tip oscillation. This output is negated and compared to a
user-specified setpoint, as in contact mode. The difference or error signal drives the
feedback circuit to displace the Z scanner in an attempt to null the error signal
(amplitude). The sense of this scheme is the opposite of contact mode (thus the
aforementioned negation): a decrease of amplitude due to an encountered rise of the
surface elevation during XY scanning must result in the displacement of oscillating tip
away from sample; whereas in contact mode an increase of raw deflection signal due
to a rise of the sample surface must result in the displacement of tip away from
sample.

36
37
Now we get to the unavoidably complicated aspects of dynamic AFM (if the preceding did not
seem complicated enough!).

To understand how the resonant cantilever behavior is modified by tip-sample interactions, one
must account for both attractive and repulsive forces. For a given oscillation amplitude, the tip
may fully penetrate into a nonmonotonic interaction potential – first attractive, then repulsive
(depicted with red curve) – as it nears and touches the surface. If the tip indeed approaches
close enough for repulsive forces to be felt (i.e., “touch”: electronic overlap between tip and
sample atoms, i.e., Pauli exclusion), it is sometimes called “true” tapping. But for the same
amplitude (say 20 nm as shown), the tip may instead only sense attractive forces because it
does not get close enough to touch (i.e., sense Pauli repulsion). This may correspond to a truly
non-contact regime, or perhaps penetration into a liquidy contaminant layer (water,
hydrocarbons) that primarily generates attractive (capillary) forces; or solid contact may have
been made, but in a glancing fashion on a very soft material, such that the attractive forces
remain dominant to the overall oscillator behavior.

How can this be? If attractive forces are encountered first, then why wouldn’t the tip simply
accelerate until it hits repulsive forces? This is exactly what we found in the force curves
considered earlier, the so-called snap-to-contact. But here’s where the physics of a resonant
oscillator is key to understanding. We must remember that in AC mode we’re dealing with a
steady-state dynamic equilibrium. In equilibrium at a particular vertical scanner position Z,
the amplitude could stabilize in one of two possible states, large amplitude or small amplitude,
as seen in the next slide. But the feedback will displace the scanner to maintain a single

38
amplitude from point to point across a surface. So obviously there must be an
interplay between the steady-state dynamics achieved and the Z position selected by
the feedback. We will delve more deeply into this reality in the next few slides.

38
Let’s consider the impact of attractive and repulsive forces on the frequency-dependent resonant response. If
only an attractive force gradient is encountered by the tip as it oscillates towards and away from the surface
(graphs at right), then the resonant response curves will shift slightly to the left, to lower frequency (dashed
curves): the oscillator has “softened”. We will not get into the mathematics; consult for example the Garcia
review article in the dropbox. A simpler, hand-waving argument is as follows. As the cantilever bends
downward during its oscillation cycle, the attractive force gradient dF/dZ between tip and sample opposes the
cantilever restoring force gradient k, effectively reducing the spring constant k of the oscillator to some k’=k-
δk. Now, a simple harmonic oscillator has a resonant frequency proportional to the square root of k, so the
resonant frequency must decrease, meaning the entire response curve for both amplitude and phase must
shift to lower frequency. This necessarily results in a reduced amplitude at the original resonance frequency f0,
as depicted with the red arrow pointing down on that plot. Recall that a reduction of amplitude is all that is
needed to operate the instrument; it is this sensation of tip to sample “engagement” that causes the computer
to stop the initial motored approach. Conversely, if repulsive forces are ultimately encountered as the tip
cycles towards the surface, the much greater steepness of these forces (see red curve) usually means the
repulsive force gradient “wins out” over the attractive force gradient earlier in the cycle, thus translating into a
net shift of the resonance curves to higher frequency. Again, one can think of the net push of the surface
adding to the restoring force (gradient) of the cantilever, thereby increasing the effective spring constant k and
thus the resonance frequency. Here again the only end result that matters, for the instrument to operate, is
that the amplitude decreases (again note red arrow).
So in actual operation, how do we know whether we are in the net attractive or net repulsive regimes? There
is no button to push or switch to flip to choose the regime! Nor is there one simple parameter setting (such as
drive frequency, discussed later) that guarantees attractive or repulsive regime. But at least diagnostically, the
phase comes to our rescue, because the phase shift differs in sign for net repulsive and net attractive (again
note red arrows). Moreover, we also can imagine how the magnitude of this phase shift would vary for
different strengths of “interaction stiffness”, whether net positive (mainly reflecting mechanical properties) or
negative (mainly reflecting dielectric/capillary/electrostatic properties). Thus we can envision phase images
that differentiate materials based on their physicochemical properties such as modulus, charge state or
polarity, akin to friction, adhesion, and/or stiffness imaging in the other modes of AFM.

39
There’s still more needed to fully understand and interpret the phase shift, but the
preceding picture is a significant start. The preceding did leave out a change of damping
(dissipation) that must occur due to tip-sample interaction: we know now that with
contact, adhesion hysteresis is taking place, and this must perform net work such that
some kinetic energy of the oscillator is lost. We know that this hysteresis can be due to
irreversible aspects of solid-solid adhesion (i.e., intrinsic tip/sample interaction) and/or
formation and break of capillary meniscus (i.e., extrinsic). The sample’s mechanical
response may also be dissipative, especially in the case of a viscoelastic material. All of
these extra sources of dissipation (in addition to air damping, that is) will act to effective
decrease the Q factor, i.e., broaden the resonance, and this will affect the phase value.
For now it is enough to simply recognize that a rigid shift of the resonant response
above, right or left as depicted, is not the only result of tip-sample interaction.

39
A “textbook” explanation for why there is a frequency shift, valid for the case of
attractive forces and small oscillation amplitude compared to range of tip-
sample force, is that the tip-sample force gradient (N/m, red) subtracts from the
effective force gradient of the cantilever (blue) as quantified by its spring constant k0
(also in N/m) to give rise to a new oscillator effective spring constant, that is, slope
of force vs distance, which is reduced and thus “softened” (green). Given the simple
relationship between oscillator stiffness k and resonance frequency (which, again,
goes as the square root of k over effective mass), a reduction in resonance
frequency must result.

This treatment is very close to correct for long-range forces, meaning electrostatic
or perhaps magnetic. In standard AFM dynamic (AC/”tapping”) mode on an
uncharged sample, however, the attractive forces are so short range (van der
Waals, capillary) that an exceedingly small oscillation amplitude would need to
be explored for the above picture to hold true, and this is not possible: such a
small, Angstrom-regime oscillation cannot be maintained in a stable fashion…

40
In truth dynamic AFM almost always uses amplitudes that are large compared to
the force range (except possibly in the presence of long-range electrostatic or
magnetic force). Nevertheless the concept of an oscillator that is effectively
softened or stiffened by the overall force gradient experienced, during the
oscillation of tip towards and away from sample, remains qualitatively valid,
as depicted in the graph at left in analogous fashion as in the previous slide. The
(net) attractive regime involves an effectively reduced oscillator stiffness and the
(net) repulsive an effectively increased oscillator stiffness. (More rigorous
mathematical relationships have been derived using fractional calculus (see Sader
article in the dropbox) and/or variational methods, for the mathematically astute
reader. For example it has been found that the frequency shift goes as the half
integral of force: an operator that is mathematically intermediate to the tip-sample
force function F(z) and its integral over a cycle of distance.)

One should again note that the above concept does not treat dissipative (damping)
tip-sample interactions, and is thus incomplete. We will return to this issue.

41
It is exceedingly important to understand that dynamic AFM, though often dubbed “tapping
mode”, can in fact be bistable within a single image. This means that at any given location
the oscillating tip can settle into one of two dynamic steady states, dominated by attractive
(perhaps purely non-contact) or repulsive (actual “tapping”, you could say) interactions. As
stated earlier, to be in the net attractive regime the tip does not reach as far into the overall tip-
sample interaction (here depicted by the blue curve for a pretty soft material) as in the net
repulsive regime. Thus for a given set point amplitude, the Z scanner must keep the probe and
sample relatively far apart compared to the net repulsive regime. If the oscillator switches from
net repulsive to net attractive steady state upon moving from material A to material B
(dissimilar layers of polyvinyl alcohol in this case) -- recognizing that this need only occur over
the course of ~Q (hundreds of) oscillations per pixel measurement -- the Z scanner necessarily
has displaced the tip and sample farther apart. This is identical, of course, to what would
happen if there were a real topographic step up (i.e., an increase in surface elevation). In the
example, Layer 1 becomes falsely “raised” by 1.3 nm, resulting in a net upward step of 0.3
nm from Layer 2 to Layer 1 instead of a downward step of 1.0 nm, the latter measured if the
net repulsive dynamic state is preserved upon moving from Layer 2 to Layer 1 (or if using
contact mode). This difference is underscored in the next slide.

The identification of these co-existing dynamic regimes within an image, and thus height
artifacts as exemplified above, is straightforward via the phase image. Because the phase data
within the net attractive or repulsive regimes lie on opposite sides of 90º phase lag, one obtains
many tens of degrees of phase contrast (dark and bright on the full 180º raw scale). Clearly,
then, one should not adjust imaging parameters so as to maximize phase contrast! In the

42
example, increasing the cantilever drive amplitude such that the free oscillation
amplitude increased from 10 to 14 nm (while correspondingly increasing the set point
amplitude such that the ratio A/Afree remained constant) transitioned the oscillator
from bistable to monostable across the imaged region.

42
This slide accompanies the previous one, to more explicitly show how a stable
repulsive regime results in a correct tracking of the 1.0-nm downward step (from left
to right) in the image; whereas a switching regime, from repulsive to attractive when
scanning left to right across the step, results in a false Z displacement such that
oscillating tip and sample move apart, equivalent to what would happen if there
were a step 0.3 nm up instead of 1.0 nm down. Thus an inverted contrast “height”
image is obtained under regime switching.

43
Region-specific bistabilities -- spatial domains of either net attractive or repulsive
dynamics -- do not necessarily result from heterogeneities in material properties
(e.g., soft versus rigid) or molecular structure (e.g., amorphous vs. crystalline). How
can this be? Topography. If there are valleys with concavity similar to the tip
(i.e., similar radius of curvature), then the relative strength of attractive forces can
be much greater within the valleys than atop the hills. Thus parameters chosen to
stabilize the repulsive regime may work just fine except in narrow valleys, where the
oscillator switches to the net attractive regime and concomitant higher flying height.
It is as if the tip were repelled by valleys!

We exemplify on a polycrystalline gold surface, which is considered by some to be a


simple, you could say canonical AFM sample. A drive frequency below resonance
was chosen to stabilize the repulsive regime, but in certain locations designated
with blue circles, the oscillator tripped into the attractive regime. At the edges of
these locations the oscillator suddenly moves away from the surface (again within
~Q oscillations); thus the amplitude image, which is in essence differential height,
highlights these locations. Such regions of net attraction can be greatly expanded
by driving the cantilever above resonance, the bottom set of images. Boundaries
demarcating regions of net attractive or repulsive interaction are also highly visible
in the amplitude image. One notes that the dark regions in phase (net attractive)
correspond to valleys as expected, but these valleys appear less deep than what is
seen in the top height image, the result of higher “fly height” due to the
displacement of the Z scanner so as to separate tip and sample. It is as if the

44
valleys are filled in by some liquid; but this is not the case.

44
This slide is a proof of concept that one can image so stably in the attractive regime
that liquid droplets can be topographically tracked in air. The example is water
nanodroplets that condense at high humidity.

45
Here we present a phase image of a triblock copolymer SIS (styrene-isoprene-styrene),
operating in the repulsive regime. The color convention, brighter being lesser phase lag
(below 90º), corresponds a greater shift of the resonant response to higher frequency. The
preceding interpretative picture then would suggest that the brighter domains are stiffer,
meaning higher in modulus (in the example, polystyrene as compared to polyisobutylene).
But as earlier hinted, a more complete and rigorous understanding of phase also accounts
for damping or dissipative effects (i.e., not just changes in interaction stiffness, be it positive
under net repulsion or negative under net attraction). Indeed an analytical expression has
been derived (see the “Famous” Cleveland article in the dropbox) based on power
considerations (energy imparted by driving force minus lost to damping in the absence of
tip-sample interaction, i.e., quantified by Qcant). As shown, this expression relates phase
shift to the energy dissipation per tap, but no further details regarding the interaction (such
as whether the dissipation is due primarily to the formation and break of a meniscus, or
rearrangements in interfacial structure, or viscous drag as the tip is pushing into the
material, etc.).

So, then, does stiffness not play a role? Sure it does, because the stiffer a material, the
smaller the deformation volume and tip-sample contact area, and thus the less the energy
dissipation (whether due to the anelasticity of deformation or interfacial irreversibility). But
saying a phase image is an image of stiffness or modulus is incorrect in general. (You will
find such glib interpretations in many older papers, especially in polymer journals.) One
could have two materials that are identical in elastic modulus but different in lossiness, or
charge, or Hamaker constant (polarizability), etc., and therefore exhibit significant phase

46
contrast. In general, of course, two materials will differ in several physicochemical
characteristics, all convolved into the phase value.

46
Imaging a surface with steep topography demonstrates the effect that slope can
have on phase data. In general this arises from two effects as follows. (1) The
operational amplitude varies above or below setpoint per the local slope (i.e., how
challenging is the topography to the feedback tracking system), and phase
necessarily changes when operational amplitude changes; and (2) energy
dissipation can change as the (small component of) shear interaction changes, as
arises from cantilever tilt relative to surface tilt (i.e., tip oscillation is not strictly
perpendicular to surface). Typically #1 is the dominant slope effect, but not always.

47
48
Phase images acquired at very different amplitude set points can highlight
surface/bulk differences. This can be seen by comparing subregions of the above
phase images. The top two boxes show a different number of bright polystyrene
cylinders because every other one is covered by a thin skin of polyisobutylene (that
has lower surface energy and thus favors the interface with air). The lower two
boxes also show different levels of contrast: three distinct levels for the left image,
two for the right. Again this is because some polystyrene cylinders are lying down
just below a thin skin of polyisobutylene, whereas those standing on end seem to be
protruding from the surface. Under “pounding” operation, the polystyrene is
perceived equally by the tip whether right at the surface or just below a thin layer of
polyisobutylene.

We also note that the boundaries between bright and dark phase are more distinct
in the left image, under very light tip-sample interaction. We expect this because the
tip-sample contact diameter should be smaller for lighter tapping, and thus lateral
resolution improved.

49
In passing it is worth noting that local anisotropy/orientation/texture (the term used
will depend on the research community) may be detected as spots in a 2D Fourier
transform. Uniform rings mean instead that there is no long-range periodicity in any
given direction in the image, only short-range variations in contrast. Again, this
analysis is akin to converting real space data into scattering (i.e., inverse space)
data, the inverse of what is actually done in the case of X-ray scattering analysis.

The radial power spectral density, considered earlier, is a 360-degree integration of


the 2D FFT around the azimuth.

50
Generally one can determine the absolute indentation distance into a compliant
material by comparing to measurements on a rigid reference material like silicon or
mica. Here we compare amplitude-Z and phase-Z curves (during approach) on the
polyisobutylene domains of SIBS and on cleaved mica. (This measurement is
analogous to force-Z measurements performed at a point while using contact mode,
or at each pixel while using peak/pulsed force mode.) We find that at A/Afree=0.5 the
tip indents 2 nm into isobutylene, assuming indentation into mica is negligible on the
Z scale shown. Given a calibrated cantilever and an estimate of tip radius of
curvature (say by imaging tip-derived broadening across a vertical edge in a
grating), one could in principle numerically model the dynamic system and compute
the modulus of the polyisobutylene (for a very brief contact, polymer mechanical
behavior being rate dependent). The phase data further indicate that the amount of
energy dissipation for A/A0=0.5 is hugely different in the two cases, reflecting
differences in indentation volume, contact area and dissipative character.

51
In addition to height artifacts due to switching between bistable regimes (net
attractive or net repulsive, considered in earlier slides), one can generally observe
very significant height artifacts in repulsive dynamic mode on mechanically
heterogeneous samples. Upon moving to a soft region, the tip penetrates farther
into the sample and thus swings farther out as well during the oscillation, i.e., the
amplitude increases. But exceeding the setpoint amplitude drives the feedback to
displace Z so as to bring oscillating tip and sample together, just as would occur if
the surface location were lower. Thus the softer regions are imaged as a “lower”
regions. This artifact is minimal at very high setpoints, those close to the free
oscillation amplitude at resonance A0, and becomes worse as setpoint is further
lowered below A0 (to around 0.5-0.6 of A0; below this setpoint the severity of the
artifact may reverse).

It should be noted in passing that of course these same mechanically derived height
artifacts can be manifest in contact or peak force modes. Imaging at a constant
applied force (or uniform applied peak force) means displacing Z to make it so, and
on a relatively softer material domain this Z displacement with be greater, such that
the tip indents farther into the sample, in order to achieve the same cantilever
deflection (i.e., applied force); this extra Z displacement is unavoidably rendered as
a “lower surface” in the “height” image. The effect is greater the higher the setpoint,
because a larger applied force produces greater indentation.

So the so-called “height” or “topography” may not be these things at all. In the

52
shown image at right, most of the “topography” is not topography; it is mechanical
compliance. So… just because your AFM software designates an image as
“height” or “topography” does not mean it is so!

52
This illustration is to clarify the differences seen in the preceding height image on
the two different polymers. The less steep schematic force-distance trend at lower
left corresponds to the softer polymer. The expected greater indentation into the
softer material would result in a larger amplitude in the absence of a Z
displacement, but the feedback forces this Z displacement so as to achieve the
same steady state amplitude as achieved on the rigid polymer. The additional 12-
nm Z displacement is thus rendered as a “height” difference per the software
vocabulary.

53
What are the critical parameters for controlling the interaction regime? For a given
cantilever (i.e., spring constant k, resonant frequency f0, quality factor Q and tip radius of
curvature R) and ambient-air medium, there are three critical operational parameters:
drive frequency (as earlier exemplified on polycrystalline gold), drive amplitude and set
point amplitude. Here we consider the latter two for the case of a ~40 N/m silicon
cantilever/tip on a silicon wafer sample.

Five approach curves are superimposed for amplitude and phase, corresponding to five
different drive amplitudes, and in all cases driving precisely at the resonant frequency.
These different drive amplitudes produce different free oscillation amplitudes quantified by
the values at the right end of the graph: a range of approximately 4-23 nm. In each case,
during approach the amplitude decreases upon tip-sample interaction (earlier for larger
amplitudes of course); later the amplitude suddenly jumps up by ~1 nm, then decreases
again with slope somewhat steeper than before. The phase data shows a dramatic jump
from above to below 90º at the value of Z at which the amplitude jumps. This tells us that
the transition to a higher amplitude yields a transition from the net attractive to net
repulsive regime.

Among the 5 cases, the jump occurs at very different values of A/A0: large in the case of
large A0 and small in the case of small A0. Thus the net repulsive regime is dominant for a
large range of amplitudes when the cantilever is driven hard, and small fraction when it
driven lightly. For each of the five cases, one could choose a set point amplitude A that
results in the net attractive or repulsive regime. Note that for fixed A/A0 but different
values of A0 (the two horizontal dashed lines), the attractive or repulsive regime can
preferentially result (see where vertical dashed lines cross the phase data). Also, had the

54
surface contained two materials, one would expect the transition amplitude for each
of the five cases to differ for the two materials. Thus a choice of A or A/A0 could yield
the repulsive regime on one material and the attractive regime on another (as shown
in the previous example).

54
The third critical parameter, drive frequency (as previously exemplified on
polycrystalline gold), has a dramatic effect on interaction regime. Consider the
phase-Z curves at right, on polyvinyl alcohol (with a different cantilever/tip than
previously). In the top graph are superimposed phase-Z curves for 13 different drive
frequencies ranging from below to above resonance, the extremes being denoted F-
and F+. At F0, we see that the net attractive regime is stabilized at first during
approach, then the oscillator transitions to net repulsive at a Z value of 28 nm. By
comparison, driving either well below or well above resonance shrinks one regime
at the expense of the other: dominant repulsion at F- and dominant attraction at F+.

The bottom graph at right compares data during approach and withdrawal
(decreasing- versus increasing-Z legs of the approach-retract cycle). Strong
hysteresis is seen, and in fact this is a general result at all drive amplitudes and
frequencies (and indeed on almost all materials). This means that an “accidental”
transition between regimes can yield a different stable state (i.e., interaction
regime). Thus one should choose parameters so as to “stay clear” of the hysteresis
loop.

55
This set of five amplitude and phase versus Z curves during approach were
acquired using a drive frequency below resonance as indicated in the inset. Note
here that negative values of (raw) phase are almost absent, except for the lowest
drive amplitude case (smallest free oscillation amplitude). Note that there are still
ranges of setpoint amplitude that result in a visual downward net shift in phase
relative to its free value, meaning a net attractive interaction as integrated over the
vertical positions explored by the tip. Here is where semantics are not universally
defined. Strictly speaking this latter situation could be called the net attractive
regime, because clearly the oscillator’s resonant response has been net shifted to
lower frequency in order for the phase value to shift downward on the shown scale.
But given no jump discontinuity to higher phase with continuing approach, we are
clearly on the “true tapping” branch of the bistability, which some will call the
repulsive regime/mode. Usually it is this avoidance of jump discontinuity in
operational performance that is most key to good (“artifact free”) operation.

56
57
Interleave is a “second pass” imaging scheme, whereby a second image (or
images) is (are) acquired in parallel with the first pass images. Each scan line is
cycled twice instead of just once, with the option to collect data on (trace or retrace
of) the second pass, as done in the first pass. In the second pass, however, the Z
feedback is turned OFF. Usually in the second pass the tip/cantilever (via the chip)
is raised, say by 10-30 nm, and the Z scanner follows the Z(X) trajectory that
has been saved into memory during the first pass as depicted above. During this
interleave pass the tip may or may not be biased relative to sample. Usually a
different feedback circuit is invoked. In one method, phase becomes the control
signal which forces adjustments in drive frequency (i.e., to keep phase constant), to
thereby image cantilever resonant frequency shift, usually due to longer range
electrostatic or magnetic forces. In the former case, these forces will of course scale
with the applied bias (often using metallized cantilever and tip). In the latter case
one works with a magnetized tip (coated with a magnetizable metal). In either case
the variable frequency shift image is a gauge of the strength of attractive force
gradient dF/dZ extending from the surface due to electrostatic or magnetic forces.

Another interleave-based method is Kelvin probe or surface potential imaging. Here


during interleave the AC mechanical excitation of the cantilever is turned off,
whereas an ac electrical bias of tip is invoked, on top of a variable dc bias. (Lower
case is intentionally used here to distinguish ac/dc biasing of the tip/cantilever from
AC mechanical excitation of the cantilever.) A feedback circuit varies the dc bias in
the attempt to null (or at least minimize) the cantilever excitation amplitude by

58
offsetting the intrinsic dc voltage difference between tip and sample due to
differences of surface potential (work function in the case of metals). Thereby one
obtains an image of surface potential.

58
In magnetic force microscopy (MFM) one typically obtains the usually
(dynamic/AC/”tapping” mode) height image, each line of the image being acquired
during each scan line’s first pass; and a frequency shift (dF/dZ) image acquired
from each scan-line’s second pass invoking the feedback circuit that adjusts drive
frequency so as to keep phase constant. In other words phase becomes a proxy for
frequency shift, and is usually kept at zero (which again is actually 90 degrees
relative to the sinusoidal drive signal) by using feedback that varies the drive
frequency to keep phase as this value. (Recall the phase-frequency relationship is
steep and essential linear around 90).

One may choose to invoke a different color-contrasting scheme for such


fundamentally different kinds of images, as shown above. Note in this example that
the granularity seen in topography image at left has little correlation with the
magnetic domain contrast seen in the MFM image at right.

59
Like MFM, EFM (electrostatic or “electric” force microscopy) uses interleave to
generate frequency shift images, but typically while biasing the tip during the
interleave (second) pass. One can choose to step this bias during an image, as
shown in the image above, and measure the corresponding change in frequency
shift, which again relates to long-range attractive force gradient dF/dZ. Then one
can plot the frequency shifts versus bias voltage, perhaps in different image regions
as done here (Au vs Si vs Al) and assess the parabolic data trend (as derives from
the energy of a capacitor, here tip/sample, ½ CV2). The curvature then relates to the
capacitive strength, whereas the offset along the tip voltage scale relates to the
intrinsic difference of tip and sample surface potential, the vertex being a measure
of this potential.

Of course one could also probe this physics above one point of the surface in
phase-Z curves acquired under different tip bias, shown in upper right. For the case
of aluminum (oxide) we see that a +1V bias results in much weaker long-range
effects than a bias of -1 V, because the intrinsic tip-sample surface potential
difference is around +0.7 V.

60
The top of this slide reiterates schematically the methodology of Kelvin probe force
microscopy (KFM), and at bottom middle demonstrates one surface potential image
across an operating organic transistor. Note how the variation of surface potential
from source S to drain D is not smooth. This relates to the polycrystalline, granular
morphology of the operative organic film. One should recognize how uniquely
sensitive and nanoscale-insightful this method is compared to simply measuring the
I-V performance of a transistor, when it comes to the research and development of
new device architectures.

61
62
More recently multifrequency dynamic imaging has become commercially available,
specifically on spm4 in the CharFac. Here the cantilever is excited at more than one
frequency. Typically one excites the fundamental resonance and uses its amplitude
to control the Z feedback circuit as usual. Then one can, for example, add a second
sinusoidal drive signal at the frequency of the next higher vibrational mode
(“eigenmode”). It turns out that the response at this frequency can be more sensitive
to details of tip-sample interaction. The example here is of two very similar
methacrylate polymers. At 92 degrees C the PEMA is likely beginning to soften and
become stickier relative to room temperature, while the PMMA has not changed as
much from its room-temp behavior. The first higher mode’s amplitude signal is more
sensitive to these subtle differences than is the usual phase signal at the
fundamental resonance frequency, thus improved materials contrast is attained.

63
Another mode utilizes both quasistatic and dynamics conditions. The tip is held in
contact at a setpoint deflection as in contact mode, but a vibration is imparted at a
resonance frequency. Here, however, this is a new resonance: a contact resonance,
where both tip and base of cantilever are pinned. The wave mechanics of the
cantilever is thus modified dramatically from that of a freely oscillating cantilever.
Just how dramatically it is modified is further a strong function of sample properties,
primarily modulus. Thus a given contact resonance can shift substantially in
frequency from one material to another. This can be imaged by monitoring the
amplitude or phase signal at some frequency near the contact resonance.

64
65
As already state, these slides are intended to complement the first lab session,
which is a demonstration of hardware and software elements. These slides can be
read in their own right, outside of any demonstration, but the topics are more easily
grasped by seeing the various effects on the screen in real time in the lab.

66
67
Upon acquiring a large image of a fairly smooth surface, one immediately finds a
strange curvature in the image, especially if examining the data in cross section
(e.g., upper right plots). To understand, we have to consider the nature and
performance of the scanning device. The 3D crystalline character of this
piezoelectric tube device actually couples X, Y and Z movement. The unit cell
cannot expand or contract along one axis without changing along the others. So our
idealized picture of a device that can independently scan the sample in three
dimensions is naïve. In many cases the cross coupling between X and Y is minimal
and/or is included in the scanning calibration scheme, meaning compensated up
front, programmatically. But there are other nonidealities that cannot be
programmatically corrected beforehand. With traditional XYZ piezotube scanners
there is no attempt to remove cross-coupling between the lateral (XY) and vertical
(Z) axes. Secondly, there is no attempt to account for the Z nonlinearity: Z is
reactively displaced by the feedback circuit, and this Z displacement is measured as
the change in the applied Z voltage (i.e., assuming linearity). A real-world quadratic
(or higher order) response in Z is unaccounted.

The false curvature can be mathematically removed from the image. In software,
one can choose portions of the image that lie on a terrace or plane (as per the
operator’s knowledge of the sample) and ask the computer to calculate a quadratic
surface that best fits through the data; then subtract this mathematical surface from
the entire image. Most programs allow the user to go to higher-order surfaces as
well, such as cubic and quartic (even as high as twelfth order, in Gwyddion), to

68
produce fits and subtract from the entire image to remove curvature from the image
(including real topographic curvature in case one wants to, say, remove rolling hills
over a microscale to retain only the nanostructure for analysis). In the above
example, if the color scale of the image in (c) is centered on the lower surface
(silicon) and compressed to reveal variations in “height”, one finds that the quadratic
subtraction did not fully remove the Z nonlinearity artifact. In fact this nonlinearity is
never a simple polynomial, especially for very large images. One can go to
increasingly higher order polynomial surface subtractions and find that this only
leaves an even higher order nonflat surface! Of course there is the ultimate real
“granularity” in almost any surface topography, but that is not what we are considering
here; we are dealing with false curvature in large images (due to scanner
nonidealities) that is simply not mathematically reproduced by a low-order polynomial
surface.

68
Let’s try to better understand how the principal nonlinearity in Z versus X and Y
arises. We must consider the nonlinear relationship in the actual displacement of Z
as a function of Z bias. The above illustrates how this relationship necessarily
produces a nonlinearity in Z versus X. (a) depicts a flat surface with a finite tilt
(exaggerated in Z) typical in AFM as noted earlier, because some finite tilt is always
present between sample plane and X-Y plane of scanner. Evenly spaced X
displacements (pixel spacing) produce evenly displaced Z steps under the usual
feedback control. But an intrinsically nonlinear relationship between Z displacement
and Z bias (b), such as experimentally observed (dependent in sign on the ramping
direction), means that unevenly spaced Z-bias steps must be produced by the
feedback control, in order to generate evenly spaced Z displacements. But in
conventional scanners it is these increments of Z bias that are taken as proxies for
the surface height; this means the topographic image must contain a nonlinear
trend, depicted in (c). Following the fitting and subtraction of a mathematically
planar surface (d), the tilt of this image is removed but not the curvature.

The upward or downward sense of this curvature is a strict function of the X-scan
direction (trace or retrace), because the Z bias ramp is in opposite directions for
scanning up or down the sloped surface. Thus the nonlinear relationship between
scanner Z bias and scanner Z displacement is further hysteretic (not shown
here; result considered in next slide).

69
Cross coupling between XY and Z movement contributes a somewhat similar
nonlinearity as that of the Z scanner; but it is fixed in sign, such that it either adds to
or subtracts from the Z nonlinearity depending on scan direction. This is best seen
following the usual 1st-order planefit subtraction, as show at right in images and
cross section for trace and retrace scans. The remaining curvature may be concave
up or down dependent on scan direction, but its severity may be different because
the curvature due to Z nonlinearity adds to or subtracts from (i.e., is
hysteretic) the curvature that is due to cross coupling.

70
Given such obvious nonlinearities in Z, one may wonder about X and Y. Indeed the
same nonlinearities are present, but normally these are handled “up front” (rather
than removing mathematically during data processing). With an open loop scanner
(spm1-3, and the default on spm4), the scanning algorithm itself is nonlinear;
thus non-constant increments of voltage are applied to the scanner so as to achieve
constant incremental displacements in X and Y. Thus the nonlinear response is
removed, to first approximation, during the data acquisition stage, not during post
processing. Another approach is to use a closed loop scanner (optional on spm4),
where otherwise constant voltage increments are modified on the fly by
measuring the actual scanner displacement (using built in sensors, capacitive or
inductive in design, and dedicated feedback circuits for X and Y).

71
All of the preceding nonlinearities can be compensated fairly well, either during
imaging via a nonlinear scanning algorithm, or during image processing via
mathematical manipulation. But the most problematic nonideality is piezoelectric
creep, meaning a time dependent change of the scanner response to applied
voltages. We demonstrate this effect following zooming and offsetting in X, all the
while monitoring the actual X position with inductive sensors (as used earlier for the
nonlinearity quantification). Firstly a centered 10-micron X scan was cycled; this
was followed by a 5-micron X offset, then followed by a -10 micron X offset, then
further 5- and 10-micron offsets. Contrary to the idealized depiction of the 1D scan
at bottom, each X offset produces a substantial creep response (graph at top). A 10-
micron offset produced a creep distance about twice as large as an ensuing 5-
micron offset. The direction of creep was always in the direction of the initial offset
that produced the creep.

With ordinary open-loop (programmatically ramped) scanners, the only recourse is


to wait for creep to subside. If frequently zooming in and out over a wide range of
image sizes, and/or if frequently offsetting X and Y, the wait time can be substantial;
or, worse, image distortion may be substantial. The only bona fide solution to this
problem is a closed-loop scanner. As already stated, such a device utilizes
inductive, capacitive or interferometric sensors to measure scanner movement; then
additional X and Y feedback circuits continually adjust the applied scanner biases to
achieve the desired scanner position for a given axis. Creep in Z also can be
compensated (on spm4), by using the Z sensor read as the measure of height,

72
rather than the applied Z bias as the measure of height.

72
We demonstrate a two-dimensional manifestation of piezocreep in AFM deflection
images of (lithographically grown) alternating gold and aluminum strips on silicon.
We compare closed-loop (left) and conventional open-loop (right) scanning upon
changing back and forth between larger (100 µm) and smaller (30 µm) scans. Both
images were acquired with fast-scan horizontal and raster-scanning upwards. Upon
zooming in or expanding back out, rampant piezoscanner creep is observed when
using open-loop scanning. Creep results in stretching or compression of
features, and can manifest itself as false curving of features or incorrect
orientations of features, because of time-dependent scanner movement (i.e., not
controlled by a scanning algorithm). None of this is manifest under closed-loop
scanning because the X and Y positioning sensors are continually measuring the
location coordinates of the scanner and modifying the applied X and Y biases to
ensure the correct positioning.

It should be clear that the benefit of closed-loop scanning is greatest in applications


where one is repeatedly zooming in and out and/or offsetting the imaging center in
the XY plane. But even if one is engaging at a point and simply zooming out,
closed-loop scanning removes the wait time (minutes) for creep effects to become
minimal. Secondly, if the measurement of angles is critical (say between atomic or
molecular steps, i.e., crystallographic information), then a closed-loop scanner can
be an important improvement.

73
The principal disadvantage of using closed loop is greater noise, originating in the
additional positional sensors (inductive on spm4). One can minimize this effect by
turning down the X and Y feedback circuit gains, but one cannot eliminate the
additional noise. Instead one must switch to open loop to achieve the lowest noise
condition. This is most readily seen in small-scale images containing distinct
boundaries such as the shown example of a block copolymer with crystalline
segments.

74
Creep also can be a problem in Z at very steep surface locations. At left we see
overshoot and subsequent creep effects in Z while scanning along the fast scan
axis, and further dependent on scan direction. One way to avoid the inclusion of
creep in the height data is to actually measure the output of a special Z sensor
(provided in XYZ closed loop scanners such as on spm4) as shown at right. Here
again, however, noise is increased if imaging the Z sensor output instead of the
applied Z bias to the scanner.

75
76
As suggested in Part A, many end uses of AFM involve measuring force versus
distance for its own sake (i.e., apart from calibration issues). Nanomechanical
property measurements may be of interest (e.g., modulus). Force versus distance is
often measured in liquids to get at intrinsic interfacial forces of interest in biophysics,
physical chemistry and other fields (or simply to reduce overall forces and thereby
improve resolution).

It is important to recognize that the Z axis presented along the horizontal is not
equal to tip-sample distance, because the cantilever deflection itself affects this
distance. To obtain tip-sample distance, one must add (or subtract, depending on
the software’s sign convention) the deflection (once calibrated to length units) and Z
displacement (which is normally precalibrated in length units, e.g., by facility
managers).

77
This example involves the measurement of attractive van der Waals forces in liquid,
and the computation of a true distance scale (as per the preceding slide). Note how
the curvature of the attractive force trend with distance (bottom graph) is not the
same as this trend with the raw Z measurement (top graph). Thus analytical
modeling of such physical effects can be quite incorrect if one were to treat a raw
force curve as a true “force-distance” curve, meaning force versus tip-sample
distance.

Do not assume the term “force-distance curve” is being used correctly in any paper.
Sometimes it is, often it is not: people will call a raw force (-Z) curve a “force-
distance” curve. If one means “distance of Z scanner displacement” then of course
the terminology is fine. But if one means tip-sample distance, then the terminology
is wrong. Please revisit previous slide if this is not abundantly clear!

78
The conceptually simplest method for determining the cantilever spring constant k
(not to be confused with wave number, as used in some slides of this document!)
experimentally is to perform a force curve on a precalibrated reference cantilever
(~$50 for a chip of 5 different cantilevers). The latter is tipless, of known modulus,
density and plan-view dimensions, such that its measured resonance frequency
determines its spring constant by a simple analytical formula. (Google Sader’s web
page for a useful calculator that does not require knowledge of modulus or density
because it further treats damping.) Obviously both cantilevers will deflect, such that
for a given change of Z, the change of deflection of the cantilever of interest is
reduced by the deflection of the reference cantilever. This correspondingly reduces
the slope of the force curve, and this slope value Sref on the reference cantilever is
entered into a simple formula to determine k given the spring constant kref of the
precalibrated reference cantilever. One may also collect a force curve atop the chip
to which the calibration cantilever is attached, to serve as the rigid reference surface
for calibrating the deflection measurement in nm/V (revisit preceding slide).

79
On reflective samples one often notes a strange wavy background in the force
curve data, if the Z stroke used is large (several hundreds of nm to microns). This
results from interferometry between laser light (wavelength 670 nm) reflected from
sample with that reflected from cantilever. One can play with sample tilt angle to
reduce or even remove the effect.

Cantilevers often are coated with metal on the top surface to increase the light
reflected from cantilever and correspondingly decrease the light incident on (and
thereby reflected from) the sample. A disadvantage of coated cantilevers is in
environmental AFM (i.e., spm 3-4): heating the sample (and thereby tip/cantilever)
or operating at variable humidity can result in bimaterial-derived bending and thus
instability.

80
Another issue that is potentially important if studying mechanics and/or adhesion
with AFM is a nonlinearity in the split-photodiode-based measurement of force. A
cubic effect can become appreciable roughly half way to the maximum positive or
negative force measurable (beyond +– 5V). If not identified as such, the data
interpretation could be that of material yield at the high end or nonphysical adhesive
contact mechanics at the low end.

81
82
Another “slope” that can be important is the relationship between amplitude and Z. If
too steep, feedback oscillations can result, as seen in height and phase images in
(a-b) above. In some cases one may have optimized the gain settings appropriately
for most of the surface being imaged (i.e., where feedback oscillations are not
seen), but at some locations (here a sticky blob of PLMA) there is a steeper
relationship between amplitude and Z and thus localized feedback
oscillations. To explicitly examine this problem in this example, a 2D array of
Amplitude versus Z curves were further acquired over the image region of (a-b). A
simultaneously generated (coarse) height mapping is shown in (c), and (d) renders
the slope of the individual A-Z trends at high amplitude value (~65 nm); thus bright
in (d) are locations where the A-Z relationship is very steep.

For imaging, rather than decrease the electronic gain setting in software to an
exceedingly low value, the operator can instead choose a lower amplitude setpoint
(<60nm), where the A-Z trend is NOT so steep, to avoid feedback oscillations.

83
During setup for dynamic (AC/tapping) mode one usually “zeroes the phase”,
meaning subtract out the “apparatus phase shift” to leave only the phase shift of
interest: that between base and tip ends of the cantilever, which strictly speaking is
90 degrees at resonance, but the Bruker (spm1-2) and Keysight (spm3-4) software
calls this zero. The apparatus phase shift may consist of several contributions
ranging from capacitance in the electronics to materials-related phase shifts at
interfaces (e.g., between the mechanical oscillator and the glue that attaches it to
other hardware) to a phase shift that is internal to the lock-in amplifier or similar
instrumentation. One makes no attempt to identify each of these possible effects;
rather one determines a single aggregate phase shift for the apparatus at large.
This number may be called the “phase offset” or the “lock-in phase” (where it is
subtracted) by the software.

84
Both the apparatus phase shift (“phase offset”) and the cantilever resonant
response itself can be unstable, particularly during the first couple of hours of
instrument warm-up, as exemplified in this slide.

85
Another source of variable resonance response is a change of air damping as a
function of coarse vertical (approach) distance. Over a scale of microns, as typically
approached following the initial cantilever tuning operation, there often will be very
substantial changes in resonance frequency, resonance amplitude, Q factor, and
apparatus phase shift (offset). Thus the initial tuning step (when tip is far from
surface) is only approximate, and can be far from adequate for quantitative work.
One should further tune the oscillator within say ~100 nm of the surface. This is not
difficult, but is best demonstrated in the lab with the system used (spm1-2 OR
spm3-4, the former being more automated).

The changes that occur from far to near the surface are due to a physical
phenomenon generally known as squeeze film damping: a growth in the viscous
effects of air on the cantilever when the latter is within distances of the surface on
the order of the cantilever’s lateral dimensions (e.g., it’s width is ~30um). You could
also call this a “confinement” effect.

86
This monograph is a reference for all training in charfac (basic to advanced
methods). A paper version in black and white is in each of the SPM labs, and a
digital version in color can be accessed through the UMN library.

87

You might also like