channeling rbs
An implanted SiC wafer can return the expected Si and C composition in a conventional Rutherford backscattering spectrum while retaining a buried band of lattice disorder after annealing. A second spectrum acquired with the beam aligned to a major crystal axis suppresses scattering from ordered rows but exposes displaced atoms and the dechanneling they cause. The difference contains depth information, yet it is not a direct photograph of damage: mass kinematics, stopping, detector response, entrance disorder, progressive dechanneling, surface structure, alignment, and analysis-beam dose all shape the result.
**Channeling RBS couples elemental depth sensitivity to crystallographic shadowing.** Rutherford backscattering spectrometry records the energy and yield of incident ions elastically scattered toward a detector, commonly using light MeV ions. In random geometry, elemental mass sets the surface-edge energy, energy loss maps scattering to depth, and corrected yield constrains areal density. In aligned geometry, ordered atomic rows or planes steer many trajectories away from close nuclear encounters, reducing their RBS yield. Atoms displaced into exposed positions scatter directly, while defects and interfaces also transfer ions from channeled to random-like trajectories.
For projectile mass $M_1$, target mass $M_2$, laboratory scattering angle $\theta$, and surface incident energy $E_0$, the elastic kinematic factor for the physically allowed branch is
$$
K=\left[\frac{\sqrt{M_2^2-M_1^2\sin^2\theta}+M_1\cos\theta}{M_1+M_2}\right]^2
$$
so a surface collision is detected near $K E_0$. Heavier target elements generally produce higher-energy edges for fixed projectile and angle. A collision at depth occurs after energy loss on the incoming path and is followed by loss on the outgoing path. In a local constant-stopping approximation,
$$
E_d \approx K(E_0-S_{in}x)-S_{out}x
$$
where $x$ represents an appropriate path-depth coordinate and $S_{in}$ and $S_{out}$ include geometry. Real reconstruction uses composition- and energy-dependent stopping, path lengths, straggling, detector response, and possibly non-Rutherford cross sections. Detected energy is therefore a calibrated depth coordinate, not depth itself.
| Analysis target | Random-spectrum contribution | Aligned-spectrum contribution | Main ambiguity | Strongest control |
|---|---|---|---|---|
| Implant damage and recovery | Composition, implant distribution and depth scale | Direct displaced-atom yield plus accumulated dechanneling | Deep excess caused by shallower defects | Virgin reference, dose series and forward model |
| Homoepitaxial film | Thickness and contamination | Film minimum, interface excess and substrate tail | Surface disorder versus film defects | Separate surface, film and substrate windows |
| Heteroepitaxial stack | Elemental edges and areal densities | Film/substrate registry, relaxation and domain response | Different optimum axes and stopping matrices | Angular scans for each elemental depth window |
| Impurity lattice location | Impurity amount and overlap | Impurity response relative to host across directions | Flux peaking and mixed sites | Several axes or planes plus site simulation |
| Compound semiconductor | Sublattice-weighted composition | Order and damage of RBS-visible sublattices | Light sublattice weak or overlapped | Add channeling NRA or another selective signal |
| Nominally amorphous layer | Layer composition and thickness | Entrance scattering and any residual order | Random-like yield is structurally non-unique | TEM, diffraction or Raman corroboration |
**The random spectrum establishes the quantitative reference before channeling is interpreted.** An RBS yield depends on incident charge, detector solid angle, number of target atoms, scattering cross section, stopping, energy bin width, detector efficiency, dead time, and geometry. For a sufficiently thin uniform slice, its scaling can be summarized as
$$
Y \propto Q\,\Omega\,N_t\,\frac{d\sigma}{d\Omega}
$$
where $Q$ is collected incident charge, $\Omega$ detector solid angle, $N_t$ areal density, and $d\sigma/d\Omega$ the applicable differential cross section. “Rutherford” behavior must be checked for the projectile, target, energy, and angle; nuclear resonances or screening corrections can invalidate an unqualified Coulomb cross section.
A useful random orientation is deliberately away from major axes and planes while preserving detector geometry and comparable path length. Merely rotating a few degrees can land on another planar feature. A two-axis map or established off-axis rotation is safer, and the acquisition record must preserve tilt and azimuth. Random and aligned spectra require consistent charge integration, dead-time correction, pileup rejection, beam spot, detector calibration, and background treatment before a ratio is meaningful.
The high-energy edge identifies surface scattering for an element, but finite detector resolution, beam-energy spread, straggling, roughness, isotope distribution, and overlapping elements broaden it. The width of a film feature is related to areal density through stopping rather than a universal nanometer-per-channel conversion. Density or stoichiometry assumptions are needed to convert areal density to geometric thickness. Channeling does not repair an incorrect random-spectrum model.
```flowchart
Define the film, substrate, implant, defect, or impurity-site decision
-> Select projectile, energy, scattering angle, detector, and safe fluence
-> Record composition hypotheses, crystal axes, surface normal, and film stack
-> Calibrate beam energy, charge, detector energy, resolution, solid angle, and goniometer
-> Acquire a verified random spectrum and fit composition plus areal density
-> Map tilt and azimuth using declared host and film energy windows
-> Refine the selected axis or plane and record full angular scans
-> Acquire aligned spectra in dose increments with identical normalization
-> Separate surface peak, film, interface, implant, and substrate windows
-> Build a forward model with stopping, cross sections, resolution, direct disorder, and dechanneling
-> Fit random and aligned spectra jointly rather than subtracting smoothed curves alone
-> Test virgin, damaged, annealed, and process-reference samples
-> Test alternative disorder depths, interface widths, and dechanneling models
-> Add multiple axes, NRA, TEM, diffraction, Raman, or electrical evidence as needed
-> Propagate counting, calibration, normalization, alignment, stopping, and model uncertainty
-> Archive raw spectra, angular maps, dose history, fit inputs, residuals, and provenance
```
**Aligned-to-random yield is conditional on energy window and alignment.** For corrected yields in a declared interval,
$$
\chi(E_1,E_2)=\frac{\int_{E_1}^{E_2}Y_{aligned}(E)dE}{\int_{E_1}^{E_2}Y_{random}(E)dE}
$$
and the reported minimum $\chi_{min}$ is the lowest qualified value reached during an angular scan, not simply the smallest noisy bin or hand-selected spectrum. A surface window, film window, interface window, and deeper substrate window can produce different ratios on the same sample. Ion species, energy, axis or plane, temperature, divergence, detector, integration interval, and reference geometry must accompany the number.
Well-ordered crystals can show low axial yields, but no universal percentage defines perfection. The unavoidable surface peak, thermal vibration, entrance scattering, finite divergence, normal dechanneling, surface oxide, roughness, crystal basis, planar versus axial geometry, and depth interval all contribute. A higher minimum indicates reduced channeling performance under the measured condition; it does not uniquely identify vacancies, interstitials, dislocations, strain, mosaicity, or amorphous material.
Angular-scan width and shape provide evidence separate from the minimum. Beam divergence and mosaic spread broaden a dip, wafer miscut shifts its center, multiple domains can split or shoulder it, and film/substrate tilt can place their minima at different angles. A spectrum recorded only at the apparent minimum cannot reveal these failure modes. Energy-windowed angular scans allow the film, interface, impurity, and substrate to be followed independently.
If aligned and random counts are $A$ and $R$ after corrections, simple Poisson statistics give the approximate ratio uncertainty
$$
\left(\frac{\sigma_{\chi}}{\chi}\right)^2 \approx \frac{1}{A}+\frac{1}{R}
$$
before systematic contributions from charge, background, dead time, calibration, alignment drift, and window selection. Smoothing does not create independent counts. Fits across energy bins must consider shared calibration and normalization parameters rather than reporting only diagonal statistical errors.
**A channeling excess contains direct disorder and inherited dechanneling.** An ion can backscatter from a displaced atom at the depth of interest, or a shallower defect can redirect it so that it later scatters from an otherwise ordered atom. The second pathway raises the deeper yield and makes simple point-by-point conversion nonlocal. Normal electronic scattering, thermal vibrations, curvature, strain gradients, interfaces, extended defects, and composition changes also dechannel trajectories.
For a shallow region where prior dechanneling is negligible, a reference-normalized estimate is sometimes written
$$
f_D \approx \frac{\chi_D-\chi_V}{1-\chi_V}
$$
with damaged and virgin normalized yields $\chi_D$ and $\chi_V$. This surface approximation is not a general depth inversion. Applying it independently to every energy channel can assign a near-surface dechanneling tail to fictitious deep disorder. A quantitative profile requires forward calculation or an iterative model coupling direct scattering, dechanneling, stopping, energy straggling, detector resolution, and the crystal’s depth-dependent structure.
Model identifiability should be tested. A thin highly disordered layer can resemble a thicker moderately disordered layer after resolution broadening; interface roughness can resemble intermixing; distributed dislocations can produce a rising tail; and an incorrect stopping matrix can shift all depths. Reporting residuals, alternative fits, parameter covariance, depth resolution, regularization, and reference sensitivity is more informative than a single smooth profile.
A random-like aligned yield means channeling suppression has been lost in that measurement volume, not that every atom occupies a homogeneous amorphous arrangement. Nanocrystalline material, severe mosaicity, dense extended defects, mixed domains, surface roughness, and misalignment can approach the random reference. Diffraction, cross-sectional TEM, Raman spectroscopy, or other structural evidence is required when amorphization is the actual claim.
**Film and interface analysis requires separate crystallographic and spectral coordinates.** In homoepitaxy, the film and substrate may share an axis yet differ in surface damage, defect density, and interface dechanneling. In heteroepitaxy, lattice mismatch, relaxation, tilt, twist, coincidence relationships, and domains can give film and substrate different optimum orientations. The elemental energy window used for alignment determines which lattice is being optimized.
An interface peak in an aligned spectrum can mark atoms displaced near the boundary or ions dechanneled by misfit dislocations and strain, but its area is not automatically an interface-defect density. Composition discontinuity changes stopping and RBS yield; roughness and interdiffusion broaden the random spectrum; threading defects distribute dechanneling through the film. Jointly fitting random and aligned spectra and measuring multiple axes constrains these alternatives.
Compound semiconductors add sublattice sensitivity limits. Conventional He RBS is often more sensitive to heavier constituents, so disorder on a light sublattice can be weak or spectrally obscured. Channeling nuclear-reaction analysis, elastic recoil detection, PIXE, or isotope-selective reactions may provide complementary sensitivity. A low yield from the heavy sublattice does not prove that every sublattice is equally ordered.
Across a wafer, a narrow beam samples a local region. Radial sites, dies, patterned environments, wafer edges, growth sectors, and process splits should be sampled without preview-based selection. Replicate scans quantify mounting and alignment repeatability. A showcase minimum from one site cannot establish wafer-level epitaxial quality or implant recovery.
**Impurity-site and activation claims require evidence beyond one suppressed peak.** If an impurity occupies substitutional sites and shares the host’s depth and channeling response, its aligned signal may be suppressed with the host. Under a simplified substitutional-plus-random model,
$$
f_s \approx \frac{1-\chi_I}{1-\chi_H}
$$
where $\chi_I$ and $\chi_H$ are impurity and host yields normalized consistently. Flux peaking, displaced substitutional sites, mixed interstitial sites, different depth distributions, overlapping edges, compound sublattices, and host damage can bias this estimate.
Angular scans across multiple axes and planes carry the lattice-site fingerprint. An impurity at a channel-center or other exposed site may show a peak or shoulder where the host shows a dip. Candidate-site simulations must include the nonuniform channeled flux, thermal vibration, displacement distributions, detector acceptance, and depth. Several crystallographically independent directions are needed to reject degenerate site models.
Lattice occupancy does not equal electrical activity. A substitutional dopant may be compensated, passivated, clustered, or in an inactive charge state; a defect complex can affect carriers despite a simple site label. Hall, capacitance, spreading-resistance, optical, or atom-probe measurements answer complementary questions. Channeling RBS supports the structural part of an activation argument but cannot close it alone.
**Analytical dose and model provenance determine whether the conclusion is trustworthy.** RBS/channeling is frequently low-consumption compared with destructive depth profiling, but an ion beam can create or anneal defects, charge insulators, heat a small spot, drive hydrogen, mix interfaces, contaminate a surface, or sputter material. Ion species, energy, current density, fluence, raster, dwell, temperature, and the sample’s initial state control risk. “Nondestructive” must be demonstrated for the material and decision threshold.
Dose-fractionated acquisition can reveal change: compare successive low-charge aligned spectra, their angular minima, surface peaks, and complementary signals. Total charge should be reported with illuminated area and raster history. If the signal evolves, use a lower dose, fresh sites, extrapolation toward zero fluence, or an explicitly beam-modified interpretation. A stable random spectrum alone cannot prove crystallographic stability because aligned yield can amplify subtle displacements.
A defensible deliverable preserves projectile and charge state, beam energy and spread, current, area, fluence, crystal temperature, random and aligned orientations, axis or plane, angular scan path, divergence, surface preparation, detector angle and solid angle, energy calibration and resolution, charge and dead-time corrections, raw spectra, energy windows, stopping and cross-section sources, simulation software and version, fit parameters and bounds, residuals, alternative models, reference samples, uncertainty, and corroborating data. It separates composition from crystallinity, detected energy from modeled depth, direct disorder from inherited dechanneling, loss of channeling from proof of amorphization, and substitutional occupancy from electrical activation. Read Channeling RBS through the random-reference-energy-depth-direct-disorder-dechanneling-and-model-validation lens.