afm (atomic force microscopy)

Atomic force microscopy profiles a surface by rastering a sharp tip on a flexible cantilever and using a feedback-controlled z scanner to follow the tip-sample interaction. The result is a quantitative height map rather than an edge inferred from electron yield or an optical model, but it is not an artifact-free copy of the surface: scanner calibration, feedback dynamics, vibration, drift, sample deformation, and especially the probe shape all contribute uncertainty. That balance explains AFM's semiconductor role. A calibrated instrument can provide subnanometer vertical resolution and traceable reference measurements, while its physical probe and minutes-per-site acquisition make it slower and more geometry-dependent than production CD-SEM or optical metrology. **The AFM cantilever senses the interaction, but the calibrated z motion commanded by the feedback loop—not Hooke's law alone—is what becomes the recorded height channel.** For a calibrated cantilever with spring constant $k$ and quasistatic deflection $\delta$, the corresponding force is approximated by $$ F = k \, \delta, $$ while the height value comes from the scanner's calibrated z displacement as the controller maintains its selected interaction setpoint. In amplitude-modulation, or tapping, mode the cantilever oscillates near resonance and the controller commonly holds an amplitude-related setpoint; in contact mode it holds a deflection-related setpoint. Tapping mode generally reduces lateral shear relative to continuous contact and is therefore useful for photoresist and other damage-sensitive films, although poor setpoint and gain choices can still deform the sample, excite feedback artifacts, or mix mechanical contrast into the apparent topography. **Tip convolution—the common shorthand for geometric broadening by a finite probe—is more precisely a nonlinear morphological dilation, and probe geometry is a major systematic uncertainty in AFM dimensional metrology.** A real apex can range from a few nanometers to tens of nanometers depending on probe design and wear. If it cannot enter a trench or follow a steep wall, the image is the set of positions accessible to that probe rather than the untouched surface itself. A protruding line therefore appears laterally wider, and an inaccessible trench may appear narrower and shallower. Height on an isolated, accessible object can be much less sensitive to lateral probe radius, which is why uncertainty must be assigned to the particular measurand instead of treating one lateral-resolution number as a universal AFM specification. **Specialized high-aspect-ratio and CD-AFM probes extend sidewall access, but accurate linewidth still depends on calibrating the probe width and flare against traceable reference structures.** Boot-shaped or flared probes and two-axis scanning let CD-AFM interrogate sidewall angle, depth, width, and some re-entrant shapes that a conventional top-down cone cannot follow. They do not remove the probe effect: tip width is subtracted or reconstructed from the apparent profile, and wear or contamination changes that correction over time. Probe qualification therefore belongs inside the measurement recipe, with periodic scans of a known characterizer and control limits that trigger recharacterization or replacement. | AFM mode / probe | Measurement strength | Semiconductor use | Dominant control | |---|---|---|---| | Amplitude-modulation / tapping, standard probe | Low-shear topography on delicate films | CMP roughness, residues, photoresist morphology | Setpoint, feedback bandwidth, apex radius | | Contact mode, standard probe | Direct deflection setpoint and compatible electrical contact | Conductive AFM and robust-surface profiling | Lateral force, wear, sample damage | | CD-AFM, flared probe with two-axis scan | Sidewall-sensitive dimensional profile | Width, sidewall angle, depth, line roughness | Traceable tip-width and flare calibration | | Kelvin probe force microscopy | Contact-potential-difference contrast alongside topography | Work-function and charge mapping | Electrical model, lift height, environment | | Scanning capacitance microscopy | Differential capacitance contrast | Qualitative or calibrated carrier-profile mapping | Oxide condition, tip contact, electrical calibration | **Surface roughness is a bandwidth-defined measurement, so $R_a$ and $R_q$ are meaningful only with the scan size, sampling pitch, leveling or filtering operation, probe, and environment that produced them.** For $N$ leveled height samples $z_i$ with mean height $\bar z$, the common discrete forms are $$ R_a=\frac{1}{N}\sum_{i=1}^{N}\left|z_i-\bar z\right|, \qquad R_q=\sqrt{\frac{1}{N}\sum_{i=1}^{N}\left(z_i-\bar z\right)^2}. $$ A small field emphasizes shorter spatial wavelengths; a larger field can include waviness and rare defects. Pixel spacing sets a high-spatial-frequency sampling limit, while flattening and filters can suppress long wavelengths. Production specifications must therefore lock the acquisition and processing recipe as well as the numerical threshold, and should use repeated sites or a designed sampling plan when wafer-level uniformity—not one local patch—is the actual process question. ```flowchart Select the probe: standard tapping tip for general roughness, CD-AFM boot tip for sidewall or narrow-feature work → Calibrate cantilever spring constant and tip radius against a reference standard → Load wafer and navigate to the target measurement site → Engage tip and establish stable feedback (constant amplitude for tapping, constant force for contact mode) → Scan the defined area at the qualified scan size and resolution → Extract topographic data and compute Ra, Rq, or feature-specific dimensions (depth, sidewall angle, CD) → Correct for known tip-shape convolution where the geometry and tip model allow → Compare results against the process specification, including its fixed scan-size and tip-type conditions → Cross-check periodically against SEM cross-section or optical reference measurements → Track tip wear and requalify or replace the probe when convolution artifacts drift beyond tolerance → Feed roughness or CD trend data back into the upstream deposition, etch, or CMP process ``` **AFM is most valuable as a traceable, local reference and failure-analysis technique rather than a universal high-volume monitor.** Surface roughness after CMP, etch sidewall validation, step height, and correlative calibration of SEM or optical models exploit its quantitative z axis and flexible probe interactions. Its small field, serial scan, navigation overhead, and tip-management burden constrain sampling, so routine fab control generally pairs sparse AFM reference measurements with faster CD-SEM or optical methods. Kelvin probe force microscopy and scanning capacitance microscopy add useful electrical contrast, but those channels require their own interaction models and calibrations and should not be interpreted as direct topography or direct dopant concentration without qualification. Read AFM through a probe-geometry lens: the recorded surface is shaped jointly by the sample, a finite physical probe, the interaction setpoint, and the feedback bandwidth, so reference-grade results come from calibrating those elements and reporting an uncertainty for the specific height, width, sidewall, or roughness measurand—not from assuming that a sharp-looking image is automatically an accurate one.