AFM versus SEM: what each instrument actually measures, and why serious labs end up with both

One grant line, two instruments, and they do not do the same job. A practical comparison of atomic force microscopy and scanning electron microscopy for Bangladeshi research groups deciding what to buy first.

A materials group with one equipment line in an approved grant asks the same question every year: scanning electron microscope or atomic force microscope. Catalogues file them together under nanoscale imaging, which is where the confusion begins. An SEM makes pictures. An AFM makes measurements. The overlap is far smaller than a specification sheet suggests, and the decision should follow from what your papers actually need to prove.

One builds an image, the other builds a height map

In an SEM, a focused electron beam is scanned across the specimen and detectors count the secondary and backscattered electrons that come back. The brightness of each pixel is a signal intensity, not a height. What your eye reads as topography is contrast produced by surface tilt, edge effects and average atomic number. Lateral dimensions come out well: line width, particle diameter, pore size, fibre thickness. The vertical axis carries no calibrated number at all.

In an AFM, a sharp tip on a flexible cantilever tracks the surface while a feedback loop holds the tip and sample interaction constant. The recorded output is the position of the z piezo at every point in the raster, which is a height in nanometres at every pixel. Step heights, deposited film thickness across a masked edge, RMS roughness, grain height distributions and layer counts on 2D materials all emerge as numbers with units and an uncertainty, not as an impression.

Resolution is not a single number

Axis or capability	SEM	AFM
Lateral detail	Very fine, set by the beam spot and by the interaction volume inside the sample	Set by the tip radius, so narrow features are imaged wider than they really are
Vertical information	Qualitative, unless you cut a cross-section or take stereo pairs	Direct and quantitative, with sensitivity well below one nanometre in z
Field of view	Millimetres down to nanometres in a single session	Small, a few tens of micrometres at most on a typical scanner
Time per frame	Seconds	Minutes for a high resolution scan
Composition	Yes, with an EDS detector on the same column	No, outside specialised chemical or electrical modes
Sample environment	Vacuum or low vacuum chamber	Ambient air, liquid or a controlled environment cell
Sample must conduct	Yes, or be coated, or be run in variable pressure mode	No

What each one demands from the sample and from the room

An SEM demands a dry, degassed, vacuum-compatible specimen. Wet samples, solvents and untreated biological material either will not pump down or will contaminate the chamber. Non-conductive samples charge, which appears as drifting bright bands and image distortion. There are two ways out. Coat the sample with a few nanometres of gold or carbon, which means a sputter coater and a supply of targets, a line item missing from roughly half the quotations we are asked to review. Or work in variable pressure mode, where residual gas in the chamber neutralises the charge. That second route is one reason the Hitachi SU3500 suits a shared facility taking mixed and unpredictable samples from several departments.

An AFM asks almost nothing of the sample chemically and a great deal of the building. It is a mechanical instrument measuring nanometres, so it hears everything. A truck on the road outside, a lift motor two rooms away, a diesel generator on the ground floor, a rooftop air handling unit, even the air conditioner blowing across the scanner head, all appear in the data as periodic noise. The sample must also be reasonably flat, because the z range of the scanner is short and a tilted or rough specimen will run the feedback loop out of travel.

• Vibration. Ground floor, on a slab, away from the road and the generator. An active isolation table costs less than moving the laboratory a year later.
• Stray magnetic fields. Distribution transformers, lift machine rooms and heavy cable runs interfere with the electron beam and show up as a 50 Hz wobble in the image. Ask for a field survey of the candidate room before any civil work starts.
• Temperature stability. Both instruments drift with the room. A dedicated split unit running continuously beats a large central system that is switched off at night and at weekends.
• Power quality. An automatic voltage regulator plus an online double conversion UPS, not an offline UPS. The changeover transient when the campus generator picks up load is what kills column electronics, not the outage itself.
• Humidity. Through the monsoon, keep spare cantilevers, filaments and prepared samples in a dry cabinet. Probes absorb moisture and stick to the surface, and a stuck tip is an expensive lesson.
• Consumables from day one. Filaments, apertures, O-rings, vacuum pump oil, gold targets and AFM probes. Probes are a consumable, not an accessory. Budget them every year.

The failure modes nobody mentions during the demonstration

Where the SEM misleads you

• Edge brightness makes small features look larger than they are, so a line width read off an uncalibrated image is not a measurement. Image a certified pitch standard at the same settings first.
• Beam damage is real on polymers, hydrated tissue and some pharmaceutical solids. The image taken in the first thirty seconds is not the image taken in the fifth minute.
• The conductive coating covers what it is protecting. A ten nanometre gold layer hides five nanometre surface texture completely.
• EDS on a thin film mostly reports the substrate, because the electron interaction volume at normal accelerating voltages is deeper than the film. Reduce the accelerating voltage until that volume sits inside the layer, or accept that you are analysing the silicon underneath.
• There is no calibrated height, so roughness derived from an SEM image is not roughness. Reviewers increasingly say so in writing.

Where the AFM misleads you

• Tip convolution. The recorded image is the surface broadened by the shape of the tip, so steep sidewalls, deep trenches and sharp posts come out systematically wrong in the lateral direction while remaining correct in height.
• Tip wear during a session. Results drift as the apex blunts. Re-image a reference feature at the end of a long run and compare it with the first frame.
• Feedback set badly. Wrong gains and setpoint give you a picture of the control loop rather than the specimen, and it can look entirely convincing.
• Thermal drift over a slow scan skews the image geometry. A twenty minute frame taken in a room whose air conditioner cycles is not a rectangle.
• Very small field of view. It is possible to measure the wrong part of the sample with excellent precision.

Which questions only one of them can answer

What you need to establish	Instrument that answers it
Thickness of a deposited film measured across a masked step	AFM
RMS roughness of a polished substrate, as a number with an uncertainty	AFM
Whether there is a crack, void or delamination somewhere on a ten millimetre die	SEM
Which elements are present in an unknown particle or inclusion	SEM with EDS
Local stiffness or adhesion of a soft coating	AFM in force spectroscopy mode
Line width and edge quality of a pattern printed on a mask aligner	Either, but the SEM surveys a whole wafer far faster
Whether the surface conducts where the design says it should	AFM in conductive mode
What a fracture surface looks like at high magnification for a failure report	SEM
Layer count on exfoliated 2D flakes	AFM step height, cross-checked with Raman

A buying order that holds up for a shared facility

1. SEM first if you own neither. It serves more disciplines than any other single instrument on a Bangladeshi campus: materials, mechanical, textile, pharmacy, geology, biology, food science and forensic work all use it. It produces the figures reviewers expect, and with EDS it answers the composition question that otherwise means shipping samples abroad.
2. Sample preparation in the same purchase order. Sputter coater, targets, a full set of stubs and holders, a cutting and mounting bench, an ultrasonic cleaner and a small vacuum desiccator. An SEM with no preparation capability runs at a fraction of its capacity.
3. AFM second, once you can name a specific z metrology requirement: film thickness in the tens of nanometres, roughness for optical or tribological work, layer counting, or surface mechanics on polymers and soft matter.
4. Raman third, because EDS reports which elements are present and Raman reports what compound or phase they have formed. A Renishaw inVia system also covers carbon materials, stress in silicon and polymorph identification, none of which an electron microscope will give you.
5. Specialised modes last. Magnetic, Kelvin probe, electrochemical and heating stages are worth buying when a funded project needs them, not in anticipation of one.

Invert that order if the group's real programme is surface forces, self-assembled monolayers, 2D materials or soft matter, where the AFM is the primary instrument and everything else supports it. The sequence above is written for a department facility serving many users. The mistake worth naming plainly: buying an AFM because it is the cheaper microscope, then discovering that it cannot survey a specimen, cannot find a defect and cannot say what an unknown particle is made of.

Why the two end up in the same room

The workflow in any established group runs one way: use the SEM to find the interesting place quickly across a large sample, then bring the AFM onto that place to get numbers. Reverse the order and you spend days hunting through a very small field of view. The pairing also settles arguments with reviewers, because a claim such as the film is smoother after annealing stops being two pictures placed side by side and becomes a roughness value with an error bar.

Find it with electrons, measure it with a tip. Groups that own both stop arguing about which picture looks smoother.: Common working rule in shared characterisation facilities

Keeping both alive in Dhaka is a separate discipline from buying them. Train two people on each instrument rather than one. Defend an annual consumables budget when departmental money gets reallocated. Sign the service agreement before the warranty lapses rather than after. Hold a local float of the parts that actually fail: filaments, apertures, probes, O-rings, pump oil and UPS batteries. An instrument waiting eight weeks for a part under a fresh letter of credit is an instrument that has cost a research group an entire semester.

Back to all Insights