A nanocrystalline copper line, a reacted silicide contact, or a thin ferroelectric film may contain grains too small for conventional surface EBSD yet too numerous for a few selected-area TEM patterns to describe. Transmission Kikuchi diffraction moves orientation mapping into that gap. A focused SEM probe rasters an electron-transparent lamella while a detector below or beside the specimen records forward-scattered diffraction patterns. The resulting colored map is useful only when its physical origin remains visible: foil thickness and damage, exit-surface weighting, probe spreading, detector geometry, pattern calibration, phase competition, indexing residuals, drift, and projection overlap all stand between one scan coordinate and one crystallographic claim.

TKD is transmission orientation mapping in an SEM, not merely finer-step EBSD. In conventional EBSD, useful backscattered electrons escape a steeply tilted bulk surface and form Kikuchi bands on a side-mounted detector. In transmission Kikuchi diffraction, also called transmission EBSD or t-EBSD, the beam passes through an electron-transparent specimen and transmitted or forward-scattered electrons form diffraction features. Off-axis TKD commonly reuses a conventional EBSD detector with a modestly tilted foil; on-axis TKD places a detector beneath the specimen near the incident-beam axis. Both differ from precession electron diffraction and four-dimensional STEM, which form and record transmission diffraction in a TEM or STEM with different optics, detectors, scattering conditions, and reconstruction choices.

Kikuchi geometry retains the same crystallographic foundation as EBSD. For lattice planes with spacing $d_{hkl}$, diffraction order $m$, electron wavelength $\lambda$, and Bragg angle $\theta_B$,

but a real TKD pattern is not a simple kinematic drawing of that equation. Electrons scatter elastically and inelastically through the foil, the crystal dynamically redistributes intensity, and the detector integrates an angular and energy distribution. Depending on thickness, beam energy, atomic number, scattering angle, and detector geometry, on-axis data can contain spots, Kikuchi lines, and bright or dark bands. Indexing geometry may therefore transfer from an EBSD framework, while intensity interpretation and calibration must respect the transmission experiment.

The useful TKD signal is depth weighted, and foil thickness controls lateral mixing. Transmitted patterns can be generated through the foil, but absorption and scattering weight different depths unequally; experiments and simulations have shown strong sensitivity near the exit surface under common conditions. That directionality matters in multilayers and overlapping grains. Reversing a lamella can change which crystal dominates even though the same projected volume remains under the beam. A TKD map is therefore a depth-weighted two-dimensional projection, not an infinitesimally thin crystallographic section.

The lateral response widens as electrons spread through a thicker foil. A defensible blur budget can be written as an engineering approximation,

where the terms represent probe size, foil-dependent scattering, motion during acquisition, and the effective localization penalty from mixed or weak patterns. These terms need not be Gaussian or statistically independent, so the expression is a diagnostic budget rather than a universal physical law. A raster step below $r_{\mathrm{eff}}$ oversamples the response and increases dose; it does not prove resolution equal to the step. Resolution claims should specify whether they mean physical boundary response, effective indexed-map response, smallest visible feature, or nominal sampling pitch.

Recent systematic measurements reinforce why a single advertised number is unsafe. Lateral resolution changes strongly with foil thickness, and depth sensitivity changes with material. Voltage, back-tilt, detector settings, probe current, stability, and the resolution metric also matter. Boundary scans across a known interface, thickness series, and TEM comparison provide stronger evidence than one isolated map pixel.

Specimen preparation defines the structure that TKD is allowed to report. Site-specific semiconductor lamellae are commonly extracted by gallium or xenon focused-ion-beam milling, attached to a grid, thinned, and finished at lower energy. Bulk alloys may also be electropolished or broad-ion milled. Each route has a transfer function: implantation, amorphization, redeposition, curtaining, preferential sputtering, surface oxidation, contamination, bending, thickness gradients, and local heating can suppress patterns or create a microstructure that was absent before preparation.

A protective cap can preserve device topography but may shadow the region of interest. Progressively lower-current thinning limits curtaining and overshoot; a low-energy finish can reduce damaged layers but also removes material. Final thickness should be measured or bounded rather than inferred only from a nominal FIB recipe. EELS, convergent-beam methods, calibrated STEM intensity, tomography, and geometric estimates each carry assumptions; thickness should be mapped when a gradient crosses the orientation map.

Bending changes the local projection geometry and can create gradual orientation shifts unrelated to the original device. Charging deflects the beam or changes landing conditions; hydrocarbon deposition causes time-dependent pattern loss; scan heating and mechanical drift distort long maps. Before a high-dose acquisition, a low-dose survey and repeated reference patterns should test whether the lamella is stable. Fiducials, scan rotation or reversal, fast repeated frames, and post-map imaging help separate microstructure from time and scan direction.

Define the phase, orientation, boundary, texture, or failure question
  -> Decide whether surface EBSD, TKD, PED, 4D-STEM, or TEM is the right evidence scale
  -> Choose a representative site and preserve wafer or device coordinates
  -> Prepare an electron-transparent foil and document cap, ion species, energy, and finish
  -> Measure or bound thickness, bending, damage, contamination, and charging
  -> Select off-axis or on-axis geometry and calibrate detector projection
  -> Set beam energy, current, step, exposure, binning, and map size from resolution and dose tests
  -> Acquire backgrounds, standards, raw patterns, drift references, and contextual images
  -> Index all plausible phases while retaining alternatives, residuals, and unindexed pixels
  -> Compare raw and cleaned maps and test boundary and grain thresholds
  -> Validate critical grains, interfaces, and phases with TEM, EDS, or repeat measurements
  -> Archive foil orientation, thickness, geometry, patterns, software, processing, and uncertainty

On-axis and off-axis geometries exchange practical advantages rather than creating a universal winner. Off-axis TKD can reuse a conventional EBSD screen and software, making it accessible, but the tilted geometry can produce stronger gnomonic distortion and sensitivity to working-distance or detector-position changes. An on-axis detector collects around the transmitted-beam direction and often provides greater pattern intensity. That intensity can be spent on lower current, shorter exposure, larger maps, or better signal-to-noise. Faster maps reduce drift exposure, which can improve effective performance even when the intrinsic boundary response is similar.

On-axis patterns also challenge the acquisition chain. A bright central beam or spot-like features may consume detector dynamic range, and pattern appearance can change with thickness, beam energy, atomic number, and collection angle. Calibration must describe the actual screen or pixelated detector, camera length, projection center, distortion, detector tilt, and any beam stop or saturation handling. A geometry that yields a high band count on one material is not automatically optimal for a different thickness or composition.

Dose should be managed as a measured trade rather than a fixed recipe. The incident electron count during an exposure is

for beam current $I$, dwell time $t$, and elementary charge $e$. Dividing by the illuminated or sampling area gives an areal-dose convention, but the reported area definition must be stated. More electrons can improve pattern statistics while accelerating contamination, charging, radiolysis, heating, or structural change. Sensitive oxides, halide materials, organics, two-dimensional crystals, and battery compounds require dose-series qualification; even a robust metal lamella can bend or accumulate carbon during a long scan.

Calibration and phase competition determine whether a sharp pattern becomes the right answer. The pattern center, detector distance, distortion, specimen tilt, beam position, and scan coordinate transform map Kikuchi features to crystallographic directions. An error can create a systematic orientation bias, a false gradient across the scan, or reduced discrimination between pseudosymmetric solutions. Known-orientation standards, geometric calibration, pattern matching, detector-shadow methods, and multi-position tests probe different parts of that model. Calibration should be repeated after changing working distance, detector head, specimen height, accelerating voltage, or projection geometry.

Hough indexing detects band geometry efficiently, while dictionary, spherical, dynamical, and correlation-based methods can use more intensity information. Richer matching does not eliminate the candidate-library boundary. If an intermetallic, oxide, ordered variant, or reaction phase is missing, an algorithm may confidently select the nearest allowed structure. Retaining the best and runner-up scores, number of detected bands, reprojection or correlation residual, pattern quality, and raw frame makes ambiguity inspectable. A confidence index produced by one vendor or algorithm is a ranking under that model, not a calibrated probability of truth.

Mixed patterns are especially important in TKD. The beam can encounter multiple grains along the foil thickness or straddle a lateral boundary after scattering. Dominant-phase indexing may hide a minority contribution; cleanup can then spread the dominant answer into neighboring pixels. Raw-pattern inspection, multi-template decomposition, foil reversal, thinner regions, or correlative TEM imaging can expose overlap. Unindexed pixels should remain visible because they may identify a real interface, damage layer, unknown phase, excessive thickness, or detector failure.

Grain and boundary statistics inherit every acquisition and processing threshold. An orientation is defined relative to declared sample and crystal coordinate frames and reduced by the correct phase symmetry. Inverse-pole-figure colors require a stated sample direction and color key. Grain reconstruction then adds a connectivity rule, misorientation threshold, minimum size, treatment of unindexed pixels, and often a cleanup sequence. Changing those choices changes grain count, equivalent diameter, boundary fractions, and local orientation-spread metrics.

For two indexed orientations $g_1$ and $g_2$, a symmetry-reduced disorientation angle can be expressed as

where $\mathcal{G}$ is the relevant crystal symmetry group under the adopted convention. A boundary drawn from that result has a misorientation and a trace in the map plane. It does not, by itself, reveal the full three-dimensional boundary-plane normal. Because TKD also integrates through finite depth, apparent junctions and grain shapes can be projections of overlapping structures rather than true planar intersections.

Wild-spike removal, nearest-neighbor fill, grain dilation, smoothing, and minimum-grain filters can make a map legible but can also erase nanoscale phases, bridge a real boundary, or manufacture grains from noise. Cleanup must remain a reversible derivative. Quantities used for a process decision should be reported before and after reasonable parameter variation, with a resolution-based lower cutoff. Kernel average misorientation and orientation spread depend on step size, neighbor kernel, noise, foil bending, and cleanup; they are not universal strain or dislocation-density meters.

Correlative validation turns a TKD map into process evidence. Bright-field or annular-dark-field TEM/STEM imaging can test grain shapes, interfaces, foil thickness, and overlap. EDS or EELS constrains composition and candidate phases. High-resolution imaging or nanobeam diffraction can resolve a critical interface, while XRD or conventional EBSD samples a wider population. Repeating a small region at another scan direction, exposure, voltage, or foil orientation can reveal drift, dose change, depth weighting, and unstable indexing.

In semiconductor work, TKD can connect nanoscale texture and boundaries to mechanisms: interconnect grains to resistivity and electromigration; barrier phases to continuity; silicide orientation to contact variability; intermetallic grains to cracks; GaN or SiC variants to epitaxial defects; and functional-film grains to switching. One lamella is not representative of a wafer or lot, so statistical conclusions require an explicit sampling plan.

A production-ready result preserves site coordinates, foil normal, thickness evidence, preparation history, beam conditions, step size, scan order, detector geometry, calibration, phase library, raw patterns, indexing alternatives, cleanup, and validation. It distinguishes sampling pitch from measured response, precision from accuracy, and a depth-weighted projection from a three-dimensional structure. Read every TKD map through the foil-thickness-exit-surface-calibration-indexing-and-projection-overlap lens.