An atomic force microscope (AFM) is a scanning probe microscope that investigates a surface by bringing a fine tip close to it and measuring the resulting mechanical interaction. The tip is usually mounted on a flexible cantilever, whose bending or oscillation changes as the probe moves over the sample. By controlling this interaction while scanning, the instrument reconstructs surface topography and maps properties such as adhesion, stiffness, electrical potential, and magnetic response. Unlike an optical microscope, it does not form an image by collecting light from the specimen.
AFM is important because it combines nanometre-scale surface imaging with localized force measurement and can operate on both conducting and insulating materials. Depending on the instrument and specimen, measurements can be made in air, liquid, controlled gas environments, or vacuum. Specialized instruments resolve individual surface atoms or intramolecular structure, whereas other configurations examine polymers, electronic devices, living cells, and molecular processes in water. Its images are measurements of a probe–sample interaction rather than direct photographs: their interpretation depends on the tip geometry, operating mode, environmental conditions, and feedback response.
Origins and development
The atomic force microscope emerged from efforts to extend the capabilities of the scanning tunneling microscope (STM). Developed by Gerd Binnig and Heinrich Rohrer in 1981, STM measures an electrical current produced by quantum tunneling across a very small gap between a conducting tip and a specimen. It demonstrated that scanning a sharply localized probe could reveal atomic-scale surface structure. Its dependence on a tunneling current, however, restricted conventional measurements to electrically conducting or semiconducting surfaces.
In 1986, Binnig, Calvin Quate, and Christoph Gerber published the first description of the atomic force microscope. Their approach replaced tunneling current between the specimen and probe with the mechanical force acting on a small spring. In the original implementation, an STM-style tunneling detector sensed the displacement of a cantilever carrying a small diamond tip. This arrangement established the central idea of AFM: extremely small interactions can produce measurable displacements if the mechanical sensing element is sufficiently compliant.
Subsequent instruments separated force sensing from tunneling detection. Optical beam-deflection systems became widely adopted because they were sensitive, comparatively simple, and compatible with interchangeable microfabricated cantilevers. The manufacture of reproducible silicon and silicon nitride probes helped turn AFM from a laboratory prototype into a broadly usable research instrument. Commercial systems became available near the end of the 1980s, followed by specialized scanners, liquid cells, environmental enclosures, and software for image acquisition and force analysis.
Operating methods also diversified. Early contact-mode measurements were followed by dynamic techniques in which the cantilever oscillates near a mechanical resonance. Frequency-modulation AFM, introduced by Thomas Albrecht and colleagues in 1991, became particularly important for high-resolution work in vacuum. Intermittent-contact methods developed into widely used approaches for imaging relatively delicate specimens in air. Quartz tuning-fork sensors, including the qPlus configuration developed by Franz Giessibl, later provided stiff, stable force sensors well suited to atomic-resolution measurements.
The historical relationship between AFM and STM sometimes causes confusion about recognition of the two inventions. The 1986 Nobel Prize in Physics recognized Binnig and Rohrer for the design of STM, sharing the prize with Ernst Ruska for electron-optical work; it was not an award for AFM. AFM subsequently became a distinct family of instruments whose applications extend well beyond atomic imaging. Its development linked precision mechanics, surface science, microfabrication, and biological measurement, with improvements often driven as much by force sensitivity and experimental stability as by nominal image resolution.
Tip–sample interactions and mechanical sensing
AFM detects the net force acting between the probe and its surroundings. That force is not a single universal interaction. At separations of several nanometres or more, van der Waals forces, electrostatic interactions, and, where relevant, magnetic interactions can contribute. At shorter separations, chemical bonding, hydration structure, and repulsion associated with overlapping electronic states become important. The dominant contribution depends on the materials, tip termination, separation, and medium.
For a sufficiently small static deflection, the cantilever behaves approximately as a linear spring. The normal force is then inferred from
where is the cantilever spring constant and is its deflection relative to an appropriate reference. A cantilever with , for example, bends by under a force of . This illustrates why microscopic springs can detect forces far smaller than those accessible to ordinary mechanical balances. The relationship alone does not establish an absolute force: both spring stiffness and displacement sensitivity require independent determination.
The interaction can also be described through a potential energy , where represents tip–sample separation:
An attractive interaction can destabilize the cantilever when its force gradient exceeds the effective restoring stiffness, producing a sudden “jump to contact.” During withdrawal, adhesion can hold the tip to the surface until a separate instability causes detachment. Consequently, approach and retraction curves often follow different paths even when the underlying materials remain unchanged.
Ambient air introduces effects absent from an ideal dry interface. Adsorbed water can form a capillary bridge between tip and sample, adding a substantial attractive force. Relative humidity, surface contamination, and wetting properties can therefore alter both apparent adhesion and the stability of imaging. In liquids, capillary forces associated with an air–water meniscus are usually removed, but hydrodynamic drag, electrical double layers, and hydration forces become important. The concentration and identity of dissolved ions can change electrostatic screening and the measured force–distance relationship.
Oscillating sensors respond to both conservative and dissipative interactions. Conservative forces change the effective restoring stiffness and therefore the resonance frequency; dissipative processes remove mechanical energy during each cycle. For small oscillation amplitudes and a force gradient that varies little over the motion,
where is the unperturbed resonance frequency and the sign follows the chosen separation coordinate. At larger amplitudes, the frequency shift reflects a weighted average of the interaction over the oscillation trajectory. Thus neither a frequency-shift image nor an amplitude image is automatically a map of force at one exact separation.
Instrument architecture and feedback control
A conventional AFM contains a probe assembly, a displacement detector, a three-dimensional positioning system, control electronics, and an isolation structure. The probe normally consists of a microfabricated cantilever carrying a sharp tip near its free end. Common probe materials include silicon and silicon nitride; coatings can make the tip conductive, magnetic, chemically selective, or more resistant to wear. Commercial tips often have nominal apex radii of a few to several tens of nanometres, although the effective shape can change during use through contamination, fracture, or material transfer.
The most common displacement detector uses optical beam deflection. A focused laser beam reflects from the back of the cantilever onto a segmented photodiode. Cantilever bending changes the reflected beam direction, shifting the light distribution on the detector. Differences between detector segments measure bending, while a perpendicular difference can measure torsion. The system is primarily sensitive to angular displacement rather than directly to the vertical position of the tip, so converting detector output into displacement requires attention to cantilever shape and loading.
Alternative detectors include interferometry, piezoresistive elements, and piezoelectric sensors. Interferometric systems compare optical path lengths and can provide highly sensitive displacement measurements. Piezoresistive cantilevers convert mechanical strain into a change in electrical resistance, avoiding an external optical lever. Quartz tuning forks generate an electrical signal through their piezoelectric response. Each approach presents trade-offs involving sensitivity, heating, alignment, bandwidth, liquid compatibility, and the mechanical stiffness of the sensor.
Relative motion between probe and specimen is commonly generated by piezoelectric actuators. Their expansion and contraction allow fine positioning along the lateral and axes and the vertical axis. Some instruments move the sample, others move the probe, and some divide these motions between separate stages. Coarse positioning brings the tip within the fine scanner’s working range. Closed-loop scanners use position sensors to reduce errors from hysteresis, creep, and nonlinear actuator response; open-loop scanners depend more strongly on calibration and correction.
During conventional imaging, the probe follows a raster pattern. A feedback controller compares a measured signal—such as cantilever deflection, oscillation amplitude, or frequency shift—with a selected setpoint. It changes the vertical position to keep the difference small. The recorded motion then supplies the principal topographic image. An error signal records departures from the setpoint and often emphasizes sharp edges, but it is not itself a reliable height map.
The feedback system has finite response speed. If lateral scanning is too fast relative to surface variation and controller bandwidth, the tip may lose contact, strike protrusions, or produce trailing features. Excessive controller gain can instead generate oscillations. Mechanical vibration, acoustic noise, temperature changes, and drift also couple into measurements because the instrument is sensitive to extremely small displacements. Vibration isolation, acoustic enclosures, stable mounting, and thermal equilibration are therefore functional parts of high-performance AFM rather than merely conveniences.
Contact, intermittent-contact, and noncontact imaging
In contact mode, the tip remains in continuous mechanical contact with the surface. A feedback loop usually maintains approximately constant cantilever deflection while the scanner follows changes in height. Contact imaging can be fast and effective on robust specimens, and the cantilever’s torsional response provides information related to lateral friction. However, the sliding tip exerts shear forces as well as normal forces. These can move weakly attached particles, deform polymers, displace biological structures, or wear either the tip or the specimen.
“Constant force” is a useful shorthand rather than an unconditional description of contact mode. Constant deflection closely corresponds to constant cantilever load, but changes in adhesion or long-range interactions can alter the actual contact force. Surface slope, detector alignment, and coupling between normal bending and torsion can complicate interpretation further. On very flat specimens, constant-height scanning can record deflection variations without rapid vertical correction, but it provides less protection against unexpected protrusions.
In amplitude-modulation AFM, the cantilever is driven near resonance and changes in its oscillation amplitude provide the feedback signal. A common implementation is intermittent-contact or tapping-mode imaging, in which the tip approaches or briefly touches the surface during each oscillation. Reducing continuous sliding often makes this mode useful for polymers, nanoparticles, and many biological specimens. Nevertheless, brief contacts can generate appreciable peak forces; a low average load does not guarantee negligible deformation or damage.
The phase difference between the drive and cantilever response can be recorded alongside height. Phase contrast may reveal regions with different adhesion, viscoelasticity, or energy dissipation, but it is not a direct, unique map of chemical identity. Operating amplitude, interaction regime, topography, and feedback settings all affect the signal. Transitions between predominantly attractive and predominantly repulsive oscillation states can also cause abrupt changes in contrast or apparent height.
In frequency-modulation AFM, the sensor’s resonance frequency is tracked while its oscillation amplitude is controlled. The frequency shift provides a sensitive measure of the conservative interaction, while the drive required to sustain oscillation supplies information about dissipation. The method is especially effective in ultrahigh vacuum, where low damping produces a high mechanical quality factor. True noncontact operation can resolve atomic structure without maintaining mechanical contact, although the label “noncontact” does not imply that all interactions are weak or purely long-range.
Other imaging strategies repeatedly approach and retract the tip at each pixel or along a rapidly modulated trajectory. These can estimate peak interaction force and collect mechanical information together with topography. No single mode is best for every specimen: stiffness, adhesion, roughness, desired speed, and environmental conditions determine the useful operating regime. Mode names also conceal substantial variation between instruments, so reproducible reporting requires actual interaction settings, probe properties, and feedback conditions rather than a mode label alone.
Force spectroscopy and nanomechanical measurements
AFM can function as a local force-measuring instrument without continuously scanning an image. In force spectroscopy, the probe approaches and retracts from a selected location while cantilever response is recorded against scanner displacement. A force curve may show long-range attraction, jump to contact, elastic deformation, adhesion during withdrawal, and final detachment. Repeating the measurement over a grid produces force maps that associate mechanical or adhesive behavior with position.
Scanner displacement is not identical to either tip–sample separation or specimen indentation. Part of the imposed motion bends the cantilever; another part may deform the sample, the substrate, or a compliant surface layer. Analysis therefore requires a suitable reference and corrections for the instrument geometry. On soft specimens, uncertainty in the point of first contact can substantially affect the fitted deformation and inferred material properties.
Elastic moduli are commonly estimated using contact mechanics models. For an ideal rigid spherical indenter pressing into a homogeneous, isotropic, linearly elastic half-space without adhesion, the Hertz model gives
where is the indenter radius, is indentation depth, and is the reduced elastic modulus. The reduced modulus accounts for elastic deformation of both contacting bodies. Adhesive contacts may instead require models such as Johnson–Kendall–Roberts or Derjaguin–Muller–Toporov, selected according to the physical contact regime rather than convenience.
Real specimens frequently violate these assumptions. Thin films respond partly through their underlying substrate, cells are heterogeneous and time-dependent, and polymers may show substantial viscoelasticity. A sharp tip can probe a different effective response from a larger spherical probe because the strain field and sampled volume differ. Loading rate, indentation depth, tip geometry, and the assumed Poisson ratio must therefore accompany any reported modulus. An AFM-derived modulus is often a model-dependent local estimate, not an intrinsic material constant established independently of the experiment.
Single-molecule force spectroscopy uses the same mechanical principle to stretch or separate molecular structures. A molecule can be attached between the tip and a substrate, allowing force–extension curves to reveal unfolding events, bond rupture, or polymer elasticity. Work on titin in the 1990s showed repeated unfolding of protein domains as characteristic force peaks. Measurements on proteins, DNA, and receptor–ligand systems subsequently connected molecular structure to mechanical stability.
Molecular rupture forces depend on loading history and pulling geometry as well as the interaction itself. A measured rupture force is therefore not equivalent to a bond’s equilibrium binding energy. Multiple attachments, nonspecific adhesion, linker elasticity, and uncertain attachment location can complicate interpretation. Chemical functionalization and control measurements help distinguish specific interactions from background adhesion, while statistical analysis of repeated events is essential because thermally activated rupture produces distributions rather than a single fixed force.
Resolution, calibration, and image artifacts
AFM resolution has separate vertical and lateral components. Vertical sensitivity can be well below a nanometre under suitable conditions, but this does not imply equivalent lateral resolution. Lateral detail depends strongly on the tip apex, interaction range, sampling density, and specimen stability. Increasing the number of pixels cannot recover structures that the probe cannot physically distinguish. Likewise, scanner travel determines field of view, not the smallest measurable feature.
A major limitation is tip-shape convolution, more precisely a geometric interaction between tip and sample shapes. A finite-width tip broadens isolated particles and narrow ridges, while deep trenches may be inaccessible because the probe sides contact their edges. A spherical particle can consequently appear much wider than it is tall. Multiple asperities on a damaged or contaminated tip can create repeated copies of every object, a “double-tip” or multiple-tip artifact. Image processing cannot reliably restore information never acquired by the probe.
Accurate dimensional measurements require calibration of the lateral and vertical axes using appropriate reference structures. Step-height standards assess vertical displacement, while pitch standards assess lateral scale. Closed-loop sensors improve positioning but do not remove the need for calibration or establish metrological traceability by themselves. Scanner orthogonality, environmental drift, probe–sample alignment, and reference-standard uncertainty can all contribute to the final measurement uncertainty.
Force measurements require additional calibration of cantilever stiffness and detector sensitivity. In a simple single-mode description, thermal fluctuations satisfy an equipartition relationship of the form
where is the Boltzmann constant and is absolute temperature. Practical thermal-noise calibration must account for mode shape, detector response, bandwidth, and the difference between static and dynamic deflection sensitivity. Hydrodynamic methods, including the Sader method, infer stiffness from resonance behavior, cantilever dimensions, and fluid properties under specified assumptions.
Data processing can introduce distortions comparable to instrumental artifacts. Plane subtraction removes sample tilt, but aggressive line-by-line flattening can erase broad structures or create apparent depressions beside tall objects. Filtering changes roughness values and can suppress physically meaningful features. Quantitative metrology therefore depends on stating scan size, pixel spacing, leveling procedures, and the spatial bandwidth over which a roughness parameter was calculated.
Atomic resolution requires an additional distinction between reproducible atomic contrast and a literal representation of atomic boundaries. The recorded signal reflects the spatial variation of the interaction, which depends on the tip’s atomic termination and electronic structure. In 2009, Leo Gross and colleagues resolved intramolecular structure in pentacene using a carbon-monoxide-functionalized tip at low temperature. Such images provide detailed structural information, but their contrast involves force gradients and probe relaxation; bright lines cannot be read universally as direct pictures of chemical bonds.
Materials characterization and functional mapping
AFM is widely used to characterize surface morphology in thin films, coatings, crystals, polymers, and electronic materials. It measures step heights, grain structure, particle distributions, and roughness over scan areas that commonly extend from nanometres to tens of micrometres. Unlike techniques that infer average structure from a large illuminated region, it provides spatially localized information. Its small field of view, however, means that a single attractive image may not represent a heterogeneous sample.
In semiconductor research and manufacturing, AFM measures features such as deposited-film roughness, etched structures, and surface steps. Specialized critical-dimension instruments use probe geometries designed to access sidewalls and other structures that ordinary sharp tips cannot adequately follow. Dimensional interpretation still depends on probe shape and accessibility: a narrow trench cannot be characterized simply by assuming that a conventional tip reaches its bottom.
Conductive AFM combines topographic measurement with local electrical current detection through a biased conductive tip. It is used to investigate leakage paths, insulating films, conductive domains, and local transport behavior. Current contrast depends on contact area, pressure, contamination, tip coating, and the electrical circuit as well as intrinsic specimen conductivity. Local current–voltage curves can be informative, but a nanoscale contact is not automatically equivalent to a macroscopic electrode with known geometry.
Kelvin probe force microscopy measures an electrostatic response and applies a compensating voltage to determine a local contact-potential difference. With suitable calibration and interpretation, it can investigate variations related to work function, surface charge, or electronic band alignment. The signal is influenced not only by the apex but also by the tip cone and cantilever, which can average over a larger area. Ambient adsorbates, illumination, and trapped charges can further change the measured potential.
Magnetic force microscopy uses a magnetized probe to detect spatial variations in magnetic interaction, often imaging magnetic domains through their stray fields. Measurements frequently separate topographic acquisition from magnetic sensing by rescanning above the surface. Even so, electrostatic forces can mimic magnetic contrast, and the probe’s own magnetic field can disturb sufficiently susceptible structures. Piezoresponse force microscopy instead applies an alternating voltage and detects a mechanically modulated response; electrostatic and electrochemical contributions mean that contrast alone does not establish ferroelectric switching.
Other combinations extend the platform beyond conventional force imaging. AFM-based infrared methods detect local mechanical effects associated with optical absorption, while tip-enhanced Raman methods combine a scanning probe with Raman spectroscopy. These approaches associate nanoscale morphology with chemical information, although their spatial resolution and signal-generation mechanisms differ. AFM also supports nanolithography, local oxidation, and controlled manipulation of particles or molecules, where the probe deliberately modifies the specimen rather than acting solely as a passive sensor.
Biological imaging, high-speed AFM, and emerging limits
AFM has a distinctive role in biology because it can examine specimens immersed in aqueous solutions without requiring the vacuum conditions of many electron-microscopy methods. Samples include membranes, nucleic acids, protein assemblies, microbial surfaces, and living cells. Preparation remains crucial: the specimen must be immobilized strongly enough to resist probe forces while retaining the structure or activity being studied. Attachment to mica, supported lipid layers, or chemically modified substrates can alter orientation, conformation, or mobility.
For membrane proteins and ordered molecular arrays, AFM can resolve surface organization and conformational differences under appropriate conditions. It can also measure the height and morphology of DNA, protein aggregates, and membrane-associated complexes. The accessible information is primarily about exposed surfaces. Structures buried inside a cell or beneath an overlying layer are not directly imaged in the same manner as the uppermost accessible interface, and lateral dimensions remain subject to tip broadening.
Mechanical mapping of cells probes a composite system involving the cell membrane, cortex, cytoskeleton, internal structures, and sometimes the underlying substrate. Results vary with indentation depth, loading rate, location, and probe size. A large spherical indenter averages over a wider area than a sharp tip and generally generates a different strain distribution. Consequently, differences between reported cellular stiffness values need not indicate a biological contradiction; they may reflect different measurement conditions and mechanical models.
Conventional scanning can be too slow to follow molecular motion, because completing an image may require many seconds or minutes. High-speed AFM addresses this limitation through small, rapidly responding cantilevers, high-bandwidth scanners, fast detectors, and carefully designed feedback control. A prominent demonstration reported in 2011 by Noriyuki Kodera and colleagues directly visualized the stepping motion of myosin V along actin. Such experiments showed that AFM could record sequences of molecular structural changes rather than only static end states.
High-speed operation involves trade-offs among frame rate, scan area, spatial resolution, and interaction force. Liquid damping changes cantilever dynamics, and reducing the time available at each pixel increases demands on signal quality and control bandwidth. Immobilization can constrain molecular activity, while the probe itself may perturb the process being recorded. Biological interpretation therefore requires distinguishing motion intrinsic to the specimen from motion induced or restricted by the measurement.
Current development includes multifrequency sensing, simultaneous acquisition of several interaction channels, automated probe characterization, and data-driven control. Machine learning can assist feature recognition, experimental navigation, and artifact detection, but inferred structures remain dependent on training data and the information contained in the measurement. More observables do not automatically make an inverse problem unique: different combinations of tip shape, surface chemistry, and mechanical response can generate similar signals. Establishing quantitative biological or chemical specificity continues to require independent constraints, calibrated probes, and explicit models of the interaction.