Hard X-ray GI Ptychography

• Hard X-ray grazing-incidence ptychography is a lensless coherent diffraction imaging method that combines surface-selective sensitivity with high-resolution ptychographic reconstruction.
• It utilizes precise beam control and multislice algorithms to quantitatively map surface morphology, achieving transverse resolutions of 15–50 nm and depth resolutions as low as 2–5 nm.
• The technique finds applications in surface defect metrology, nanofabrication, and in situ studies, while facing challenges such as anisotropic resolution and computational complexity.

Hard X-ray grazing-incidence ptychography is a lensless coherent diffraction imaging modality that synergistically combines the surface-selective sensitivity of grazing-incidence X-ray techniques with the high-resolution, large field-of-view capabilities of ptychographic reconstruction. This approach enables nanostructure imaging at or near planar surfaces over statistically significant regions, providing quantitative real-space maps of surface morphology, local height, and in advanced implementations, three-dimensional (3D) refractive index distributions. The methodology exploits the unique interaction of X-rays incident at angles near or below the material’s critical angle for total external reflection, optimizing both surface specificity and spatial resolution, with particular utility in materials science, nanofabrication, metrology, and in situ investigations (Jørgensen et al., 2023, Myint et al., 2024, Besley et al., 8 Oct 2025).

1. Physical Principles and Scattering Models

In grazing-incidence ptychography, a coherent hard X-ray beam impinges on a planar or quasi-planar sample surface at a shallow incidence angle $\alpha_i$, typically near or just below the critical angle $\alpha_c$ for total external reflection. Under these conditions, the following physical regimes and models become relevant:

• Kinematic (Born) Approximation: At low coverage or for high-contrast/high-absorption structures, the scattered amplitude can be modeled by the single-scattering integral:

$$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$

with $\delta\rho(\mathbf{r})$ the local electron-density contrast and $\mathbf{q} = \mathbf{k}_f - \mathbf{k}_i$ is the momentum transfer vector (Jørgensen et al., 2023).

• Fresnel Reflectivity and Evanescent Fields: For $\alpha_i < \alpha_c$, the incident field excites an evanescent wave penetrating only nanometers into the surface, yielding almost total reflectivity and ultra-high surface sensitivity. The reflectivity is determined by the Fresnel coefficients:

$$R(\alpha_i) = \left| \frac{k_{z,1} - k_{z,2}}{k_{z,1} + k_{z,2}} \right|^2$$

where $k_{z,j}$ encodes the complex refractive index (Jørgensen et al., 2023, Myint et al., 2024).

• Multiple Scattering and Multislice/DWBA Formalism: At grazing incidence, the probe's footprint is elongated ($\sim 1/\sin\alpha_i$), leading to enhanced multiple scattering. The projection approximation (collapsing the object to a 2D transmission function) can become invalid, especially for low-absorption substrates or layered/rough morphologies. The distorted-wave Born approximation (DWBA) incorporates the substrate, but for general structure recovery, a multislice formalism is employed:

$$\psi_j(x, y) = \mathcal{P}(\Delta z)\left[O_j(x,y) \psi_{j-1}(x, y)\right]$$

where $\alpha_c$0 is the $\alpha_c$1th slice transmission and $\alpha_c$2 a Fresnel propagator (Besley et al., 8 Oct 2025).

This theoretical foundation allows modeling both amplitude and phase of the exit wave for realistic, arbitrarily complex stratified and structured surfaces in hard X-ray ptychography (Besley et al., 8 Oct 2025).

2. Experimental Configuration and Data Acquisition

Implementation is characterized by precise control over X-ray beam properties, sample geometry, and scanning trajectory:

• Probe Generation and Footprint: A focused zone plate or refractive lens generates a probe of $\alpha_c$31–4 μm transverse width, which diverges at the grazing incidence to cover a longitudinal footprint of hundreds of micrometers (Jørgensen et al., 2023). Coherence is maintained via synchrotron sources, pinholes, or lenses, delivering high photon fluxes (Myint et al., 2024).
• Sample Positioning and Scanning: The sample is positioned with a combination of piezoelectric stages (for fine 2D raster or Fermat-spiral scanning) and precision hexapod interferometric alignment (to maintain surface planarity, typically sub-10 nm over millimeter scales). Step sizes and scanning overlaps (60%–90%) are selected to account for footprint elongation and optimize phase retrieval redundancy (Jørgensen et al., 2023, Myint et al., 2024).
• Detector Arrangement: Far-field coherent diffraction patterns are collected on high-pixel-count detectors (e.g., Pilatus 2M) at distances of 3–7.4 m, enabling collection of both low- and high-angle scattering information. Data is geometrically remapped ("tilted-plane correction") to align Ewald sphere sampling with the physical detector plane (Jørgensen et al., 2023).
• Multimodal Data Collection: By varying $\alpha_c$4 across and above $\alpha_c$5, or rotating the sample azimuthally (ptychographic laminography), datasets suitable for depth-resolved and 3D reconstructions can be obtained (Myint et al., 2024, Besley et al., 8 Oct 2025).

3. Ptychographic Reconstruction Algorithms

Grazing-incidence ptychography reconstruction involves solving a nonlinear inverse problem to recover the spatially resolved complex object and (optionally) probe wavefields:

• Projection Approximation Models: For strong scatterers with moderate multiple scattering, the exit wave at scan position $\alpha_c$6 and angle $\alpha_c$7 is

$\alpha_c$8

where $\alpha_c$9 is the probe, $$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$0 the reflectivity, $$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$1 the surface topography (Jørgensen et al., 2023).

• Iterative Phase Retrieval: The reconstruction is typically initialized by difference map (DM) algorithms and refined by maximum-likelihood approaches, such as LSQ-MLc or extended ePIE/rPIE. The consistency constraint with measured intensities is

$$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$2

where $$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$3 is the measured diffraction intensity at scan $$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$4 (Jørgensen et al., 2023, Myint et al., 2024).

• Multislice/PDE-based Inversion: For samples requiring full multiple-scattering treatment, the multislice formalism underlies a computational graph implemented in frameworks such as PyTorch, enabling automatic differentiation and GPU-accelerated optimization. The cost function penalizes the discrepancy between simulated and measured Fourier magnitudes, with regularizers enforcing expected material constraints or smoothness:

$$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$5

where $$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$6 is the multislice-exit field, $$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$7 is the scattering potential (related to refractive index), and $$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$8 the regularizer (Besley et al., 8 Oct 2025).

• Depth Profiling and 3D Retrieval: Depth-dependent parameter maps (e.g., $$A(\mathbf{q}) = \int \! d^3 r \; \delta \rho(\mathbf{r})\, e^{i\mathbf{q}\cdot\mathbf{r}}$$9 for electron density, interfacial roughness $\delta\rho(\mathbf{r})$0) are extracted by fitting the local amplitude response across $\delta\rho(\mathbf{r})$1 to Parratt's recursive formalism or via voxelized model-free inversion. The phase encoding of $\delta\rho(\mathbf{r})$2 extracts surface height with nanometer precision (Myint et al., 2024).

4. Spatial Resolution, Sensitivity, and Quantitative Performance

The spatial resolution and quantitative performance of hard X-ray grazing-incidence ptychography are defined by intrinsic optical constraints, geometrical anisotropy, and algorithmic implementation:

Dimension	Typical Achievable Resolution	Governing Factors
Transverse ($\delta\rho(\mathbf{r})$3 beam)	15–50 nm	Diffraction limit, detector geometry (Jørgensen et al., 2023, Myint et al., 2024)
Longitudinal ($\delta\rho(\mathbf{r})$4 beam)	80 nm–2 μm (angular dependent)	$\delta\rho(\mathbf{r})$5 footprint anisotropy; improved by azimuthal rotation (Jørgensen et al., 2023, Myint et al., 2024)
Depth (vertical/$\delta\rho(\mathbf{r})$6)	2–5 nm (multi-angle)	$\delta\rho(\mathbf{r})$7, Parratt modeling (Myint et al., 2024)

• Phase Sensitivity: Height sensitivity down to 10 pm, with practical repeatability of 0.1–1 nm after noise filtering (Jørgensen et al., 2023).
• Field of View: By stitching multiple Fermat-spiral subscans, millimeter-scale fields can be mapped, enabling statistically significant surface area analysis (Jørgensen et al., 2023).
• Surface and Interface Sensitivity: The technique captures nanostructure morphology, interfacial roughness, and low-$\delta\rho(\mathbf{r})$8 contaminant layers, and quantifies local height variations against independent topography (e.g., AFM) to sub-nm precision (Jørgensen et al., 2023, Myint et al., 2024).
• Quantitative 3D Maps: By exploiting the combined amplitude and phase information at multiple angles, local layer thicknesses and electron density profiles can be reconstructed with nanometer-scale vertical discrimination (Myint et al., 2024).

5. Application Domains and Demonstrated Use Cases

Hard X-ray grazing-incidence ptychography is applied across surface and interface sciences, providing capabilities including:

• Nanostructure and Patterned Surface Imaging: High-resolution imaging of lithographically defined test objects (e.g., Siemens stars, pseudo-random metasurfaces) on gold or silicon substrates, resolving 20–50 nm features with nanometer height discrimination (Jørgensen et al., 2023, Besley et al., 8 Oct 2025).
• Surface Defect and Roughness Metrology: Mapping of roughness and defect distributions on semiconductor photomasks and integrated circuit substrates to detect nm-scale nonuniformities, interface disorder, and layer thickness deviations (Myint et al., 2024).
• In Situ and Depth-Profiling Studies: Tracking morphological evolution, film deposition, or dynamic surface reactions by monitoring phase and amplitude contrasts across varying $\delta\rho(\mathbf{r})$9 (Jørgensen et al., 2023, Myint et al., 2024).
• 3D and Tomographic Surface Reconstructions: Through multislice modeling and the inclusion of multiple incidence and rotation angles, volumetric refractive index maps and layered interface structures are quantitatively reconstructed, validated against ground-truth synthetic and experimental datasets (Besley et al., 8 Oct 2025).
• Comparative Metrology: Direct validation of ptychographically retrieved height and density profiles with atomic force microscopy and conventional X-ray reflectivity, observing sub-nanometer absolute agreement (Jørgensen et al., 2023).

6. Limitations, Challenges, and Directions for Development

Despite demonstrated performance and versatility, several physical and methodological limitations persist:

• Anisotropic Resolution: The footprint elongation at low $\mathbf{q} = \mathbf{k}_f - \mathbf{k}_i$0 and the consequent poor longitudinal resolution restrict true 3D isotropic reconstruction. Ptychographic laminography or multi-angle acquisition relaxes these constraints at the cost of increased data volume and computational complexity (Jørgensen et al., 2023, Besley et al., 8 Oct 2025).
• Multiple Scattering and Modeling Accuracy: For low-absorption materials, rough stratified substrates, or high-density heterogeneous samples, the projection approximation and even DWBA can fail. Full multislice forward modeling, while computationally demanding, is necessary for quantitative accuracy (Besley et al., 8 Oct 2025).
• Surface Placement and Flatness: Height noise is amplified by the $\mathbf{q} = \mathbf{k}_f - \mathbf{k}_i$1 factor; sub-10 nm planarity and nanometric stage precision are essential for artifact-free reconstructions (Jørgensen et al., 2023).
• Radiation Dose and Imaging Speed: Optimization is needed to mitigate radiation damage in sensitive samples while maintaining high spatial resolution and adequate SNR. Approaches include dose fractionation, compressed-sensing scan trajectories, and advanced regularization in phase retrieval (Jørgensen et al., 2023).
• Material and Contrast Limitations: For samples with strong chemical and topographic inhomogeneities, model extension to allow simultaneous retrieval of refractive index maps and explicit surface topography may be required, blending multislice ptychography with reflectometric depth profiling (Jørgensen et al., 2023, Besley et al., 8 Oct 2025).

7. Outlook and Synthesis

Hard X-ray grazing-incidence ptychography integrates the statistical power and surface sensitivity of GI-SAXS and reflectometry with the direct, lensless real-space mapping of ptychography. By leveraging advanced multislice algorithms, GPU-accelerated optimization, and optimized experimental design, this class of techniques delivers non-destructive, quantitative nanostructure characterization of extended mesoscopic surfaces. Ongoing developments focus on genuine 3D tomographic reconstructions, in situ time-resolved and chemically specific imaging, and further methodological convergence with metrological standards for semiconductor and quantum-device testing (Jørgensen et al., 2023, Myint et al., 2024, Besley et al., 8 Oct 2025).