Atom Probe Tomography in Metallurgy: Nanoscale Composition, Clustering, and Grain Boundary Segregation

Atom probe tomography (APT) is the only characterisation technique capable of generating a three-dimensional, atom-by-atom map of a material volume with near-atomic spatial resolution and parts-per-million elemental sensitivity simultaneously across the entire periodic table. For metallurgists, it has transformed the study of solute clustering, precipitate chemistry, grain boundary segregation, and carbon redistribution in steel microstructures at length scales inaccessible to any other analytical method. This article covers the full technical picture: field evaporation physics, LEAP instrumentation, specimen preparation routes, 3D reconstruction algorithms, data analysis methods, and the critical metallurgical applications that have changed how engineers understand alloying and microstructure.

History and Development of the Atom Probe

The atom probe originated from Erwin Müller’s invention of the field ion microscope (FIM) at Pennsylvania State University in 1951 — the first instrument to image individual atoms on a metal surface. Müller extended this in 1967 by adding a time-of-flight mass spectrometer aligned with a probe hole in the FIM screen, creating the first atom probe: a single-atom mass spectrometer capable of identifying atoms selected from specific surface sites. Through the 1970s and 1980s, this evolved into the position-sensitive atom probe (PoSAP) and the three-dimensional atom probe (3DAP), which used a two-dimensional position-sensitive detector to collect many ions simultaneously, enabling 3D reconstruction.

The transformative step came in 1999 with the introduction of the local electrode atom probe (LEAP) by Thomas Kelly and colleagues at the University of Wisconsin. The local electrode — a micrometre-scale aperture positioned within 50 μm of the specimen apex — greatly reduced the specimen voltage required for evaporation (from 30+ kV to 3–15 kV) and permitted loading of multiple specimens on a carousel for sequential analysis. The CAMECA LEAP 4000 and LEAP 5000 series (current generation) dominate commercial APT globally, with laser pulsing at 355 nm, 10 ps pulse duration, achieving analysis of non-conductive oxides, geological materials, and semiconductor devices in addition to metals.

Field Evaporation: The Physics of Atom-by-Atom Removal

The entire APT measurement rests on the controlled, predictable process of field evaporation. Understanding its physics is essential for interpreting APT data correctly and avoiding compositional artefacts.

The Evaporation Field

When a sharp conductive needle is biased to a high positive voltage V, the electric field at the apex is concentrated by the geometry:

At these field strengths (~10–50 V/nm), the potential energy barrier for surface atom desorption is reduced to near zero by the image force and the applied field potential. The Müller-Schottky model describes the energy barrier:

For a metallic tip in the LEAP, this barrier approaches zero at the evaporation field, and atoms leave the surface as positively charged ions (mostly singly or doubly charged) with near-100% ionisation efficiency — meaning virtually every evaporated atom is detected as an ion, giving APT its unmatched detection sensitivity.

Pulsed Evaporation and Time-of-Flight Mass Spectrometry

Applying a DC voltage alone causes continuous field evaporation at an uncontrolled rate. To measure each ion’s flight time, evaporation must be triggered one-at-a-time by superimposing a short pulse onto the base voltage. Two pulse modes are used in the LEAP:

Voltage pulsing uses sub-nanosecond HV pulses at 15–20% of the DC standing voltage. It provides excellent mass resolution (Δm/m < 1/200) and is preferred for metals and conductive materials. Pulse repetition rate is typically 100–500 kHz, giving collection rates of tens of millions of ions per hour.

Laser pulsing uses 355 nm UV pulses of 10 ps duration focused to a ~5 μm spot on the specimen apex. The laser causes localised surface heating, briefly reducing the evaporation field. Laser mode enables analysis of poorly conductive materials (oxides, carbides, nitrides) and has lower thermal tailing artefacts in the latest LEAP 5000 instruments. The energy per pulse is typically 0.01–0.5 nJ; too high a laser energy causes surface migration and compositional redistribution artefacts.

Ion identity is determined by the time-of-flight between the pulse and the detector hit:

Specimen Preparation

APT specimen preparation demands extremely sharp needles with apex radii below 50 nm — a requirement that has historically been the primary barrier to wider adoption. Two routes dominate: electropolishing for bulk materials and FIB milling for site-specific targets.

Electropolishing

Electropolishing is the classical route for homogeneous metallic specimens (wires, rods, sheet strip). A 0.2–0.3 mm diameter wire or matchstick-shaped blank is mounted in a two-zone electrolyte bath. DC current dissolves the metal preferentially at the liquid–air meniscus, producing a double-cone shape that eventually fractures to leave two sharp tips, both usable for APT. Common electrolytes include:

Electropolishing is fast (minutes per tip), avoids Ga ion beam damage, and can produce excellent specimens from homogeneous materials. It cannot, however, target specific microstructural features such as a grain boundary of known misorientation, a crack tip, or a specific precipitate identified by EBSD or TEM.

FIB Lift-Out (Site-Specific Preparation)

Focused ion beam (FIB) preparation in a dual-beam FIB-SEM allows the metallurgist to place the APT analysis volume at any user-defined location identified in the SEM image — a grain boundary mapped by EBSD, a specific inclusion, a HAZ region adjacent to a weld fusion boundary, or an oxide layer on a corroded surface. The standard annular milling lift-out sequence is:

Step 1 — Protective cap deposition
  Deposit Pt or W layer (1–2 µm thick, ~20 µm long) over ROI by e-beam EBID
  then by ion-beam IBID; protects surface from Ga damage.

Step 2 — Trench milling
  Mill two trenches flanking the capped region at 30 kV Ga+.
  Remove material to leave a freestanding lamella ~2 µm thick.

Step 3 — Undercut and lift-out
  Cut one side and bottom; attach omniprobe micromanipulator with Pt weld.
  Extract lamella; attach to Si microtip coupon array with Pt weld.

Step 4 — Annular milling (sharpening)
  Mill annular patterns concentrically around lamella attachment point
  at 30 kV → 16 kV → 10 kV progressively, reducing outer diameter
  from ~3 µm to a needle-shaped tip with R < 100 nm at apex.

Step 5 — Low-energy cleaning
  Final annular mill at 2–5 kV to remove Ga-damaged amorphous surface
  layer (typically 5–10 nm) and reduce Ga implantation artefact.

Final tip geometry: apex radius < 50 nm, cone half-angle ~ 4–6°

Three-Dimensional Reconstruction: The Bas-Geiser Algorithm

Converting the recorded (x_det, y_det, ToF) dataset into a 3D atom map requires a reconstruction algorithm that back-projects each ion's detector impact coordinates to its original position on the specimen surface. The dominant method is the point projection model formalised by Bas et al. (1995) and refined by Geiser et al. (2009), known as the Bas-Geiser algorithm.

Point Projection Geometry

The algorithm assumes a sphere-on-cone emitter geometry: the tip apex is a hemisphere of radius R mounted on a cone. Ions are treated as launching from the surface at angles determined by the local field geometry and then projecting linearly onto the flat detector. The lateral position on the detector maps to the launch position on the sphere via:

The tip radius R increases progressively during evaporation as atoms are removed layer by layer; it is tracked in real time from the evolving specimen voltage (since V = F × k_f × R, and F is approximately constant at the evaporation field for a given material). This progressive radius update is critical for accurate depth calibration.

Calibration and Reconstruction Parameters

Reconstruction quality depends on careful calibration of three parameters: the image compression factor m_c, the field reduction factor k_f, and the evaporation field F_evap. In practice, these are optimised by:

Requiring that known crystallographic plane spacings (measured from the visible {hkl} atomic planes in the reconstructed dataset) match the known lattice parameter. For example, in a steel specimen, {002} planes in body-centred cubic iron should appear at 1.43 Å spacing; deviations indicate incorrect depth calibration. Commercial software (CAMECA IVAS, AP Suite) includes tools for automated plane-spacing fitting and parameter optimisation.

Data Analysis Methods

Isoconcentration Surfaces (Isosurfaces)

An isoconcentration surface is a 3D envelope drawn through the reconstructed dataset connecting all voxels at the same solute concentration. It delineates precipitate or cluster boundaries and grain boundary planes within the APT dataset. The threshold concentration must be chosen deliberately:

Proxigram (Proximity Histogram)

The proxigram averages composition as a function of perpendicular distance from an isoconcentration surface, replacing the simpler one-dimensional line profile which is sensitive to the measurement direction relative to an irregularly shaped feature. For each atom in the dataset, its signed distance to the nearest point on the isosurface is calculated; atoms are binned by distance in typically 0.1–0.5 nm increments; and the average composition in each bin is plotted. This provides the solute depletion zone width in the matrix, the interfacial width, and the equilibrium precipitate composition all from a single analysis.

Cluster Analysis Algorithms

For detecting solute clusters before they are visible as distinct isoconcentration features — the pre-precipitation state in aluminium alloys or vacancy-solute clusters in irradiated steels — dedicated cluster detection algorithms are applied to the raw point-cloud dataset. The two most widely used are:

Maximum separation method (Hyde et al.): Two solute atoms are assigned to the same cluster if their separation distance is below a threshold d_max. Clusters are then defined as connected sets of atoms satisfying this criterion. Additional parameters include N_min (minimum cluster size to be counted as significant) and L_erase (maximum distance between cluster and surrounding solute atoms to include in the cluster envelope). Parameter selection is material-specific and must be validated against simulated random datasets.

Density-based spatial clustering (DBSCAN): A more robust approach imported from data science that identifies high-density regions in the point cloud without requiring a pre-specified number of clusters. Each atom with at least N_pts neighbours within radius ε is classified as a core point; core points and their reachable neighbours form a cluster; all other atoms are matrix. DBSCAN is less sensitive to parameter choice than maximum separation and handles elongated or non-spherical clusters better.

Grain Boundary Segregation Analysis

When a grain boundary (GB) is captured in the analysis volume — visible as a planar discontinuity in crystallographic pole figure patterns within the reconstruction, or targeted by FIB after EBSD mapping — the enrichment factor and the Gibbsian interfacial excess can be quantified. The Gibbsian interfacial excess Γ_i (atoms/m²) integrates the excess solute above the matrix background across the boundary:

Key Metallurgical Applications

GP Zones and Early-Stage Precipitation in Aluminium Alloys

Guinier-Preston (GP) zones are coherent, solute-enriched planar or spherical regions formed during the earliest stages of age hardening in aluminium alloys, before any distinct precipitate phase has nucleated. They are too small (typically 1–5 nm) and have too little diffraction contrast for reliable TEM detection at very early aging times, but they produce strong hardening increments that are disproportionate to their size owing to the coherency strain field they impose on the surrounding matrix.

APT uniquely resolves GP zones in three dimensions because the composition contrast between the solute-enriched zone and the Al matrix is detectable even at sub-nanometre scale. In the 7xxx series (Al-Zn-Mg-Cu), APT has demonstrated that the aging sequence proceeds as:

The grain boundary precipitate-free zone (PFZ) detected by APT — a region of solute depletion flanking the boundary — is a key indicator of intergranular corrosion susceptibility. The PFZ width and the grain boundary enrichment factor together determine whether the alloy is susceptible to stress corrosion cracking under sensitising conditions.

Carbon and Carbide Analysis in Steel

Carbon is the most technically important solute in engineering steels, yet its spatial distribution at the nanometre scale is extremely difficult to measure by any technique other than APT. EDXS has poor sensitivity for carbon; EELS is sensitive but surface-projected and poorly quantitative for dilute carbon; SIMS provides depth profiles but with poor lateral resolution. APT directly maps carbon in steels with ~0.01 at.% sensitivity and sub-nm spatial resolution, enabling:

Martensite lath chemistry: In freshly quenched high-carbon martensite, carbon is in a supersaturated solid solution with a composition close to the nominal steel carbon content. APT has confirmed that carbon is not uniformly distributed even in as-quenched martensite — it segregates to lath boundaries and dislocations within hours at room temperature (auto-tempering), forming C-rich clusters of 0.5–2 nm with 3–10 at.% C detectable by APT. This has direct implications for delayed cracking and hydrogen embrittlement resistance.

Tempering carbide evolution: As martensite is tempered, transition carbides (epsilon-carbide, Fe_2.4C at 100–200°C; eta-carbide, Fe₂C at higher temperatures) nucleate on dislocations before converting to cementite (Fe₃C) above ~300°C. APT detects these transient carbides at sizes of 1–5 nm and measures their composition — cementite yields ~25 at.% C in the carbide and sub-0.1 at.% C in the tempered martensite matrix, providing the partitioning data required to validate CALPHAD steel databases at this temperature range.

Retained austenite carbon enrichment: The carbon content of bainite-stabilised retained austenite in TRIP steels and quench-and-partition (Q&P) steels is a critical parameter controlling austenite stability. APT has measured carbon concentrations of 0.8–1.5 at.% in interlath austenite films of 5–20 nm thickness in Q&P steel — values impossible to obtain by any other technique with sub-film spatial resolution.

Gamma/Gamma-Prime Elemental Partitioning in Nickel Superalloys

The high-temperature strength of Ni-base superalloys depends on coherent L1₂-ordered gamma-prime (γ') precipitates in an FCC gamma (γ) matrix. The mechanical properties are governed not only by the γ' volume fraction and size, but by the partitioning of alloying elements between the two phases — since γ' strengthening arises from the anti-phase boundary (APB) energy, which is controlled by the composition of the γ' phase.

APT is the definitive technique for measuring partition coefficients K_i = C_i^γ'/C_i^γ across the γ/γ' interface with sub-nm resolution. Elements partition as follows in typical Ni superalloys:

APT characterisation of the γ/γ' interface has also revealed that the interfacial region is not compositionally sharp — there is a 1–3 nm transition zone where composition changes progressively. This interfacial diffuseness is related to the lattice misfit between γ and γ' and has implications for the coarsening (Ostwald ripening) kinetics via the Lifshitz-Slyozov-Wagner (LSW) theory.

Radiation-Induced Segregation and Embrittlement in Reactor Pressure Vessel Steels

Reactor pressure vessel (RPV) steels are subject to neutron irradiation that displaces atoms from their lattice sites, creating vacancies and interstitials. These defects migrate and either recombine or form vacancy clusters, voids, and dislocation loops — but they also drive radiation-induced segregation (RIS), carrying solute atoms to grain boundaries and forming solute clusters in the matrix that pin dislocations and raise the ductile-to-brittle transition temperature (DBTT).

APT has been central to understanding RPV embrittlement because it can detect the nm-scale Cu, Ni, Mn, and Si clusters responsible for hardening at concentrations as low as 0.1 at.% Cu in the ferritic matrix. Key findings from APT studies of irradiated RPV steels include:

Cu clusters of 2–4 nm diameter form at doses as low as 0.01 dpa; their core composition is 60–90 at.% Cu despite the nominal matrix Cu content being 0.1–0.3 wt.%. Mn and Ni co-segregate to these clusters, increasing cluster density and hardening beyond what Cu alone would predict. The cluster number density is approximately 10²²–10²³ m⁻³ at design EOL dose (0.1–0.2 dpa for Western PWR vessels), each contributing an obstacle spacing of approximately 10–20 nm for dislocation glide — consistent with the measured yield strength increase of 50–200 MPa.

Hydrogen Trapping by Cryo-APT

Hydrogen embrittlement of high-strength steels and hydrogen storage materials requires knowledge of where hydrogen atoms reside within the microstructure — at dislocations, grain boundaries, carbide interfaces, or in the lattice. APT of hydrogen is technically challenging because H (1 Da) overlaps with residual gas in the UHV chamber and is highly mobile at room temperature. The cryo-APT approach addresses both issues:

Protocol for cryo-APT hydrogen trapping analysis:
  1. Cathodically charge specimen with D₂O electrolyte (deuterium = 2 Da, unambiguous)
  2. Remove specimen within 30 s; cryo-cool to <100 K immediately
  3. Transfer to LEAP cryo-stage at ≤ 50 K without atmospheric exposure
  4. Analyse: D signal (2 Da) is unambiguous vs residual H (1 Da)
  5. Map D distribution: clusters at ε-carbide interfaces, grain boundaries,
     dislocations identified as primary trapping sites in high-strength steel

Trapping site hierarchy (binding energies):
  Elastic strain field of dislocation: −20 to −30 kJ/mol
  ε-carbide/matrix interface:          −30 to −55 kJ/mol
  Grain boundary:                      −20 to −45 kJ/mol (depends on GB character)
  Martensite lath boundary:            −15 to −25 kJ/mol

APT in Context: Comparison with Complementary Techniques

Practical Considerations: What APT Can and Cannot Do

APT is a powerful but expensive, low-throughput, small-volume technique with specific practical constraints every metallurgist should understand before designing an experimental programme around it.

Analysis Volume

A typical LEAP dataset contains 5–50 million atoms from a volume of approximately 100 nm × 100 nm × 200 nm (2 × 10⁶ nm³ = 0.002 μm³). This is orders of magnitude smaller than the volume probed by SEM/EBSD (~10 μm³ typical scan area), SAXS (mm³), or even a single TEM thin foil (~0.3 μm³). Conclusions about volume fraction or number density of features from a single APT dataset must be treated with appropriate statistical caution; multiple datasets from different regions are required for statistically representative measurements of rare features.

Detection Efficiency

The microchannel plate (MCP) detector in the LEAP has a detection efficiency of approximately 37% in older instruments and up to ~80% in the LEAP 5000XR with an improved MCP assembly. This means 20–63% of all evaporated ions are never recorded. For dilute solutes (0.1–1 at.%), this translates to a significant Poisson statistical uncertainty in cluster compositions. The effect must be accounted for in any cluster analysis algorithm; simulations of randomised datasets with the same detection efficiency are required to establish significance thresholds for cluster detection.

Mass Spectral Overlaps

APT measures mass-to-charge ratio; it does not directly identify the element. Overlaps between isotopes, molecular ions, and charge states can create ambiguity. Common problematic overlaps in steel include: ⁵⁴Fe²⁺ (27 Da) overlapping with ⁵⁴Cr²⁺ (27 Da); ¹⁴N₂⁺ (28 Da) overlapping with ²⁸Si⁺ (28 Da) and ⁵⁶Fe²⁺ (28 Da). Natural isotope abundance ratios are used to deconvolute these overlaps, but at very low concentrations the deconvolution uncertainty becomes significant.

Software and Data Processing Standards

Commercial APT data is processed in CAMECA IVAS (Integrated Visualization and Analysis Software) or the newer AP Suite 6.x, which provides reconstruction, mass spectrum ranging, cluster detection, proxigram calculation, and isoconcentration surface tools. Open-source toolkits including IVAS-Plus, atom-probe-experimental (MPIE/Raabe group), and apt-cluster provide Python-based workflows for advanced cluster analysis and uncertainty quantification.

The community has also developed a standardised data exchange format, *.apt (CAMECA binary) and the open *.pos / *.epos formats, as well as the HDF5-based *.h5 format promoted by the atom probe community for FAIR (Findable, Accessible, Interoperable, Reusable) data sharing. ASTM E2989-15 provides a standard guide for specifying APT specimens and data reporting, including required metadata such as specimen preparation route, analysis parameters, number of detected ions, and detection efficiency.

Recommended Reference Books

Disclosure: MetallurgyZone participates in the Amazon Associates programme. If you purchase through these links, we may earn a small commission at no extra cost to you. This helps support free technical content on this site.