A glossy 3D pseudo-color map of an atomic force microscope (AFM) scan can stop a reader in their tracks. It looks like proof. But a pretty picture is not the same thing as a reliable measurement. Before you quote a single roughness number, answer a few uncomfortable questions first. Which surface quantity did you actually measure? Was the roughness calculated over the whole area or along one line? And did your scan size, pixel count, probe tip, feedback settings, or post-processing quietly change the answer?
This guide covers the parts of AFM roughness work people skip, and shows how to report them so your numbers still mean something to the next person who reads your paper.
What AFM Actually Measures
Atomic force microscopy showed up in 1986, from Binnig, Quate, and Gerber. Unlike scanning tunneling microscopy, it does not need a conducting sample, which is why it caught on for polymers, oxides, ceramics, and biological materials. The popular image of a perfectly sharp needle copying the real surface is wrong, though. The final image is a joint product of the probe, the sample, and the feedback system. Finite tip size, the probe-sample force, feedback bandwidth, scanner calibration, and data processing all leave fingerprints on what you see.
An AFM does not take a photograph. It builds a height map by dragging a sharp probe across the surface, letting the cantilever react to the interactions, and using a feedback loop to steer the scanner. The channels you usually get are height, amplitude or error, and phase. When you want a real statement about topography, height data comes first.
Start With the Raw Height Map, Not the 3D Render
AFM papers love the colorful 3D plot, but the color scheme, lighting angle, aspect-ratio stretch, and Z-axis scaling all change how dramatic a surface looks. None of those are quantitative. The number you report should come from a raw or properly normalized height map that carries clear lateral and vertical scales.
Height is the channel to trust for quantitative topography: step heights, surface relief, area roughness. Amplitude or error often looks sharper on edges, yet it reflects the feedback error signal and should not stand in for absolute height. Phase can reveal differences in dissipation and material behavior, but on its own it is not a hardness map, an adhesion map, or a composition map.
Here is the trap I see most often. Someone says, “This 3D image looks spikier and rougher, so the roughness must be higher.” It does not follow. Visual rendering is not a metric. You can only compare roughness when the scan scale, the data-processing path, and the roughness parameter are the same or clearly written down.
Sa Is Not the Same as Ra
This is the most common terminology slip in AFM writing, and it is worth fixing. Modern surface-texture standards draw a hard line between two families of numbers.
Parameters calculated from a whole two-dimensional area are areal parameters, written as Sa, Sq, and Sz. Parameters calculated from a single one-dimensional profile, a line cut, are profile parameters, written as Ra, Rq, and Rz.
| Parameter | Data object | Plain meaning | What to watch |
| Sa | 2D area | Arithmetic mean of absolute height deviation from a reference plane | Good for describing average relief across a full AFM image |
| Sq | 2D area | Root mean square of height deviation | Sensitive to large deviations; often what software calls “area RMS roughness” |
| Sz | 2D area | Total height range from highest peak to deepest valley | Very sensitive to isolated particles, dust, defects, outliers |
| Ra | 1D profile | Arithmetic mean of profile height deviation | State how the profile was chosen and what filtering or evaluation length was used |
| Rq | 1D profile | Root mean square of profile deviation | If software labels a whole-image RMS as “Rq,” say so in methods |
| Rz | 1D profile | A maximum-height class profile parameter | Definitions have varied across standards and software; name yours |
In plain math, Sa is the area average of the absolute height, and Sq is the square root of the area average of the height squared. So if your software hands you an “RMS roughness” for a full 5 micrometer by 5 micrometer image, the standardized label is closer to Sq. If you pulled one line profile and computed it, that is Rq. Software naming does not always follow the ISO symbols, which is exactly why your methods section should state the parameter definition, the software version, the processing flow, and the evaluation region.
The cleanest writing rule: do not use Ra, Rq, and Rz as a catch-all for all AFM roughness. Use Sa, Sq, and Sz when you describe a full height image statistically, and keep Ra, Rq, and Rz for line profiles.
Why the Same Sample Gives Different Roughness at Different Scales
Roughness is not a scale-free material constant. An AFM only samples a limited field of view and a limited number of pixels, so which spatial wavelengths of the surface get included changes the final statistic. A 2 micrometer by 2 micrometer scan might mostly show nanoparticles, while a 50 micrometer by 50 micrometer scan may also pull in longer-wavelength waviness, wrinkles, or a macroscopic tilt.
Four things decide what you measure. Scan size sets which spatial scales you see, and Sa or Sq from different fields cannot be compared without qualification. Pixels set the horizontal sampling interval, so a 10 micrometer field at 256 by 256 is not the same sampling as the same field at 1024 by 1024. Filtering decides which long- and short-wave components count as roughness, so plane correction, form removal, and filter rules must be reported. And locations matter because one image does not represent the whole sample; repeat scans at independent positions, especially for non-uniform films.
This is why a serious methods section cannot stop at “roughness was measured by AFM.” At minimum, state the scan area, pixel count, number of independent regions, parameter definitions, and whether plane correction or filtering was applied. If two groups used different scan sizes, their nanometer-scale roughness values may not be directly comparable even though both are in nm.
The Tip Convolution Trap
AFM probes are not infinitely sharp mathematical points. A real tip has a finite radius and cone angle, so when it scans a narrow bump, the sidewall of the tip can also make contact and inflate the apparent lateral size. When it scans a narrow groove, the tip may not reach the bottom at all. Metrology labs describe this more precisely as a morphological dilation between the surface and the tip, not a simple copy of the real geometry.
Lateral size is the most easily distorted. Particle diameter, line width, pore size, and edge sharpness all deserve caution. Height is usually more robust for wide, isolated steps, but narrow trenches, rough surfaces, soft samples, and poor feedback can still introduce error. Roughness can shift with tip radius too. An overly blunt tip bridges small valleys and underestimates short-wave roughness.
There are standards that address exactly this. One covers restoration procedures for AFM images dilated by finite probe size, and another gives a procedure for judging probe suitability when measuring the roughness of fine ceramic films. The takeaway: probe model and tip condition are not background trivia from the instrument manual. They are part of the quantitative measurement.
Contact, Tapping, or Non-Contact: The Mode Is Not Just a Name
AFM has several imaging modes, and for thin films the most common are contact mode and dynamic modes, often grouped in the lab under “tapping” or “intermittent contact.” Different modes mean different probe-sample interactions, and the result depends on sample stiffness, adhesion, surface water layers, scan speed, and feedback settings.
Contact mode keeps the probe in continuous contact and tracks static or quasi-static forces. It is fast and intuitive but can disturb soft polymers or weakly attached structures through lateral friction. Dynamic or tapping mode oscillates the cantilever and uses amplitude or frequency feedback to hold a stable interaction. It usually reduces lateral shear on soft samples, but a wrong setpoint can still cause indentation or distortion. Non-contact mode avoids sustained mechanical contact during imaging, typically through dynamic detection. It can reduce some contact perturbations, though the exact signal and feedback scheme vary by instrument.
For a polymer film, writing “tapping mode was used to avoid damage” is better than saying nothing, but still incomplete. Report the probe model, cantilever parameters, scan speed or line rate, free amplitude and setpoint where applicable, and confirm the surface was not obviously rewritten by checking trace and retrace consistency and repeated scans.
Phase Contrast Is Not a Hardness Map
In amplitude-modulation AFM, phase contrast relates to probe-sample interaction and energy dissipation, so different material regions can show clear phase differences. But phase is not a single material property. Adhesion, viscoelasticity, capillary forces, topography coupling, drive conditions, and feedback settings can all influence it.
So when two regions show different phase, a careful statement is “there are different local probe-sample interactions or dissipation behaviors.” If you want to claim modulus, adhesion, or hardness, you need calibrated force spectroscopy or nanomechanical modes with a clearly applicable model. For mechanical quantification, even the cantilever spring constant needs calibration.
Do not write “the bright region is the hard phase, the dark region is the soft phase” unless you have independent calibration and a valid model behind it. A safer path is to call it phase contrast or dissipation-related contrast first, then explain it alongside known material structure or quantitative nanomechanical data.
Flattening Is Not Photo Retouching
Raw AFM images often carry overall tilt, scanner bow, row-by-row offsets, or thermal drift, so some plane leveling or flattening is usually needed. But post-processing has a boundary. Correcting an instrumental or mounting background is not the same as deleting real long-wave structure you do not like.
Plane level to remove overall tilt is generally reasonable, especially when the sample was not perfectly level. Line-by-line correction can fix per-row bias but may also create stripes or erase real structure, and higher correction orders need justification. Mask should be considered when particles or steps are present, so large structures do not pull the reference plane out of shape. Filter changes the spatial-frequency content, so any filter used for roughness comparison must be identical across groups and reported.
If your conclusion depends heavily on how many flatten orders were used, whether stripes were removed, or whether outliers were deleted, that belongs in the methods section, not just in the software log. Save the raw data, state the processing steps, and process all groups with the same pipeline where possible.
Calibration and Drift: Precise Is Not the Same as Accurate
AFM x, y, and z displacements depend on scanner calibration. There are standards for calibrating the scanning axes of scanning probe microscopes used for geometric measurement, and for measuring the drift rate of these instruments. Metrology institutes treat AFM as an important method for traceable dimensional measurement at the nanoscale, but they stress that accurate measurement depends on scale calibration and control of probe effects.
For routine thin-film studies that only compare groups relatively, you may not need a full uncertainty budget in every paper. But you should know the difference between “the resolution is high” and “the value is traceable and accurate.” If you report absolute step heights, particle diameters, or nanoscale roughness, calibration status matters a great deal.
Building the Evidence Chain: AFM With XPS and Contact Angle
Suppose a polymer film was plasma treated. The water contact angle drops from 90 degrees to 45 degrees, XPS shows a rising O/C ratio and more oxygen-containing carbon, and the AFM Sa or Sq also changes. The safe conclusion is not “roughness increased, so wettability improved.” The safe conclusion is that both surface chemistry and surface topography changed, and either could influence the final wetting behavior.
Why not pin it on roughness alone? The same Sq value can come from many small particles or from fewer but taller island structures. Two surfaces with similar Sq can have completely different spatial distributions, peak-valley densities, correlation lengths, and local steepness. Roughness is a compressed statistic of the height distribution, not a full surface ID card.
The most over-used sentence in this field is “AFM proves the surface got rougher, therefore performance improved.” At most, AFM first proves that certain topography parameters changed under specified scales and treatment conditions. Whether that change caused the performance shift still needs mechanism, controls, and other characterization.
What a Proper AFM Methods Section Should Include
A defensible AFM methods section should at least cover:
- The AFM instrument and imaging mode.
- Probe model and material, with nominal tip radius, spring constant, or resonance frequency where relevant.
- Scan area, pixel count, and scan rate or line rate.
- Key feedback conditions for dynamic modes, such as free amplitude and setpoint, depending on the instrument.
- How many independent regions were measured per sample, instead of only showing the “best-looking” image.
- The exact flatten, plane-correction, filtering, and line-correction methods.
- Whether roughness is areal (Sa, Sq, Sz) or profile-based (Ra, Rq, Rz), with the calculating software or standard named.
- For lateral quantification like particle size or trench width, whether tip morphology effects were considered.
The Bottom Line
A striking 3D surface map is a starting point, not a conclusion. To make AFM roughness numbers worth citing, report the raw height data, use the right areal versus profile notation, state your scan scale and processing, account for tip effects, choose the imaging mode deliberately, and resist turning phase contrast into a hardness claim. Treat roughness as one compressed view of a surface, then build the rest of the story with chemistry and physics.