Thin Identification Techniques

• Thin identification is a suite of techniques that characterizes ultra-thin interfaces, films, and defects using mathematical models, optical contrast, and spectroscopic methods.
• Advancements in neural network automation and regression analysis allow for high-throughput, precise detection of subwavelength and atomic-scale features.
• Applications range from non-destructive imaging in electrical impedance tomography to secure metamaterial barcodes and automated layer-counting in 2D materials.

Thin identification refers to the suite of mathematical, experimental, and computational techniques for detecting, localizing, and/or characterizing extremely thin objects—interfaces, inclusions, films, wires, or defects—embedded in host materials or constructed as functional materials themselves. This encompasses direct structural identification in thin films and layered materials, quantitative thickness analysis of atomically thin crystals, and inverse problems in mathematical imaging, especially for highly subwavelength or interface features that are nearly invisible to volumetric probes. The challenge is amplified by the minimal volume fraction, strong anisotropy, or severe electromagnetic contrast of these features, necessitating sensitive and non-destructive approaches spanning optical, electrical, and data-driven methods.

1. Mathematical Theory of Thin Inhomogeneity Identification

Rigorous mathematical frameworks underpin much of thin identification, especially for detecting insulating or conducting anomalies in materials via indirect measurements. A paradigmatic case is the identification problem for thin insulating cracks and small conducting inclusions in electrical impedance tomography (EIT), as formulated by Ammari, Seo, and Zhang (Ammari et al., 2015). Here, the forward model is the complex admittivity PDE

$$\nabla \cdot (\gamma^\omega\nabla u^\omega) = 0 \quad \text{in}~\Omega,$$

where $\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$, with $\omega$ the frequency.

The thin inhomogeneities ("cracks") are modeled as tubular neighborhoods $$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$ around $C^3$ curves $\mathcal{L}_k$, with small $\delta_k \ll 1$. For these, the key asymptotics show that the voltage jump $[u^\omega]_k(x)$ across the thin layer is governed by

$$[u^\omega]_k(x) = 2\delta_k \lambda_c(\omega)\;\partial_\nu u^\omega(x-\delta_k\nu_x)|_+ + O(\delta_k^2\log\delta_k),$$

with the contrast factor $$\lambda_c(\omega) = \frac{\sigma_c + i\omega\epsilon_c}{\sigma_b + i\omega\epsilon_b}$$ encapsulating pronounced frequency dependence.

Boundary observations are then rationalized via a far-field expansion, where the perturbations induced by thin and small inclusions manifest as singularities in complex-analytic functions on the boundary, with distinct residues encoding positions and types of features. These expansions enable not only identification but also discrimination between families of thin and small inclusions via multi-frequency data.

2. Optical and Spectroscopic Protocols for Atomically Thin Material Identification

In the material sciences, thin identification concerns thickness determination and structural assignment for 2D van der Waals crystals, such as h-BN, MoS₂, TaSe₂, and InSe. The primary methods are:

• Optical contrast models: Quantitative reflectance/transmittance models based on multilayer Fresnel equations,

$\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$0

where $\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$1 is the flake/substrate reflectance as a function of wavelength $\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$2 and thickness $\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$3. Fitting to such contrasts, especially on well-defined SiO₂/Si substrates, provides accurate thickness estimates, with monolayer contrast as low as $\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$4 for h-BN (Li et al., 2016) and up to $\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$5 (negative) for MoS₂ (Castellanos-Gomez et al., 2010).

• Raman spectroscopy: Layer-sensitive shifts and intensity changes in characteristic phonon modes such as the $\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$6 of h-BN (shift: $\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$74 cm⁻¹ over monolayer–bulk (Li et al., 2016)), $\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$8 of TaSe₂ (linear fit $\gamma^\omega(x) = \sigma(x) + i\omega\epsilon(x)$90.036 cm⁻¹/layer (Castellanos-Gomez et al., 2013)), and similar for TMDs. Only specific modes are present in certain materials (e.g., h-BN: only $\omega$0, TaSe₂: $\omega$1 and $\omega$2).
• AFM and TEM confirmation: AFM step heights ($\omega$3 nm/monolayer for h-BN), and HRTEM/lattice-fringe imaging, often combined for unambiguous monolayer confirmation (Li et al., 2016).
• Photoluminescence: In InSe and related materials, PL peak energy is a monotonic function of thickness, $\omega$4 (fit $\omega$5 eV·nm², $\omega$6 eV (Zhao et al., 2020)), allowing layer quantification independent of substrate.
• Empirical color charts: Relate RGB values of samples on known substrates to calibrated thickness-color relationships (InSe, TaSe₂, MoS₂).

3. Computational and Automated Thin Identification

Recent developments emphasize automating thin identification, both to minimize human error and enable high-throughput sample production.

• Neural network–driven optical classification: Systems based on convolutional neural networks, such as the ensemble approach demonstrated for 2D material sample picking (Greplova et al., 2019), segment raw micrographs into candidate flakes via pre-processing, and classify thickness and quality in a pipeline with $\omega$7 precision and $\omega$8 reduction in manual screening. Extensions to multiclass thickness estimation via regression on color features are suggested.
• Semantic segmentation and regression for complex architectures: In twisted bilayer materials, two-step CNNs (segmentation for layer counting; regression for twist angle) can extract both spatial and orientational features (Yang et al., 17 Apr 2026), achieving $\omega$9 twist-angle error (validated by SHG and Raman). This enables structural identification in complex CVD-grown heterostructures at rates of thousands of flakes per hour.
• Photoluminescence-based, high-throughput identification: Automated PL-microscopy routines with staged thresholding and connected-component analysis can process entire wafers, classifying monolayer/bilayer regions quantitatively and producing statistical datasets for device-yield optimization (Crimmann et al., 2024).

4. Thin Feature Identification in Functional and Mathematical Imaging Contexts

Beyond direct material science, thin identification arises in several advanced fields:

• Multi-frequency electrical impedance tomography (mfEIT): Frequency-resolved EIT can distinctly image thin insulating cracks (prominent at low $$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$0) and small conductive inhomogeneities (dominant at high $$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$1). Spectroscopic data fused via PCA yields integrated images where both feature types are simultaneously visible (Ammari et al., 2015). This approach is empirically validated by phantom experiments in large-scale electrolytic tanks.
• Secure identification via thin metamaterial-encoded barcodes: Subwavelength-patterned films designed for strong, narrowband mm-wave or terahertz absorption can be fabricated as “invisible” security barcodes readable only in the appropriate spectral window (Skirlo et al., 2016). Here, thinness ($$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$230 μm, $$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$3) is critical for covert marking and spectral selectivity. The spatial pattern encodes information retrievable only by spectrally resolved imaging, with high data density ($$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$4 bits/cm²/layer) and physical unclonability.
• Atomic-scale thin defect identification in thin films: Planar intergrowths or stacking faults (e.g., BaO between CuO chains, or fluorite-like Y–O₂–Y quadruple layers in YBa₂Cu₃O₇ (Gauquelin et al., 2017)) are imaged by HAADF-STEM and EELS. These atomic-scale defects are both structurally thin and require nanoscale spatial and compositional sensitivity to be unambiguously identified.

5. Guideline Protocols and Quantitative Performance

Thin identification protocols are highly method- and system-dependent. Across domains, validated protocols and key quantitative limits include:

Method (Material/Class)	Single-layer Detectability	Accuracy/Resolution
Optical contrast (h-BN, MoS₂, TaSe₂)	$$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$52.5% (h-BN), 60% (MoS₂)	$$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$6 nm (optical); $$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$7 nm (AFM); PL $$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$810% (InSe)
Raman spectroscopy (layer counting)	$$\mathcal{C}_k = \{ x + h \nu_x : x \in \mathcal{L}_k,~|h| < \delta_k \}$$9 shift $C^3$0 cm⁻¹ (h-BN)	$C^3$1 cm⁻¹ (TaSe₂), repeatable across devices
Neural-network optical classification	Automated, $C^3$2 pixel accuracy	End-to-end $C^3$3 manual reduction, $C^3$4 (twist angle)
PL-microscopy (TMDs)	$C^3$5–2 flakes/min, $C^3$6 s operator/flake	4 orders magnitude higher throughput than manual
mfEIT spectroscopic imaging	Cracks (visible at $C^3$7 kHz), Disks (at $C^3$8 kHz)	Integrated PCA image resolves both feature classes

In each approach, layer number, structural type, or geometric feature is inferred quantitatively, with cross-validation (e.g., AFM, TEM, SHG, Raman, PCA) as appropriate.

6. Broader Impact and Significance

Thin identification methods are foundational in both fundamental and applied science. Inverse-problem approaches yield mathematical guarantees for detectability thresholds and support efficient numerical imaging algorithms in EIT and related modalities (Ammari et al., 2015). In 2D materials research, reliable thin identification underpins device fabrication, structural-property correlation, and the scalable discovery of functional nanostructures. Optical, spectroscopic, and computational automation strategies are rapidly reversing throughput bottlenecks. In secure ID and metamaterials, thin encoding enables unforgeable markings, leveraging the interplay between thickness, invisibility, and narrowband addressability.

The domain thus encompasses a wide span—from zero-thickness mathematical models with frequency-dependent jump conditions, to atomic-scale identification of planar defects in complex oxides. Across techniques, thin identification operates at the intersection of precision measurement, advanced modeling, and cutting-edge automation, cementing its centrality in materials physics, applied mathematics, and robust device engineering.