I. Abstract of the Paper’s Core Research

This study focuses on the preparation of 1D/2D nanofilm networks via aerosol–water‑mist deposition and establishes an FIB‑SEM nano‑tomography system for quantitative three‑dimensional characterization. By addressing the limitations of conventional methods—namely, BET, X‑ray CT, and single‑section SEM—the research achieves three‑dimensional visualization and quantitative analysis of various types of nanofilms. The key findings are as follows:

Standardized Workflow for High-Resolution 3D Tomographic Imaging

Using gallium ion beam milling layer by layer combined with SEM frame-by-frame imaging, a single sample can yield 800 consecutive cross-sectional layers, with voxel dimensions as small as 5 nm × 5 nm × 15 nm. Paired with WEKA machine-learning–based image segmentation, this approach automatically distinguishes between the nanoplate and pore phases, enabling the complete reconstruction of the three-dimensional microscale thin-film structure while balancing high resolution with an effective characterization volume.

Full Quantitative Extraction of Multidimensional Microscopic Parameters

Based on 3D reconstruction models, we precisely quantify bulk and layered porosity, pore tortuosity, specific surface area, pore‑size roundness, the stacking‑aggregation coefficient of lamellae, and the Hermans orientation factor, thereby quantitatively elucidating how nanoplatelet size governs the internal microstructure of the film. Our results demonstrate that films composed of small‑sized nanoplates exhibit high density and elevated pore tortuosity, whereas those formed from large‑sized plates develop highly interconnected macroporous channels.

Universal characterization capabilities for multi‑system nanomaterials

It covers five mainstream aerosol‑printing materials: liquid‑phase exfoliated graphene, electrochemically exfoliated graphene, WS₂, silver nanosheets, and silver nanowires. A unified analytical framework has been established for one‑dimensional and two‑dimensional nanonetworks, enabling the characterization of interfacial penetration dynamics within single‑layer films and multilayer heterostructured devices.

Microstructure – Quantitative Correlation with Electrical Properties

A theoretical model of thin-film resistivity was developed based on 3D characterization data, demonstrating that nanoplatelet size, pore tortuosity, and stacking density directly govern current‑carrying transport efficiency. Electrochemically exfoliated graphene flakes are thinner and exhibit denser stacking, resulting in electrical conductivity that is six times higher than that of liquid‑phase exfoliated graphene.

AI Auxiliary Resolution Enhancement Scheme

By incorporating the DAIN video interpolation neural network to reconstruct intermediate slices, the longitudinal resolution is doubled, thereby mitigating structural distortion caused by the large slice spacing in FIB milling and further enhancing the accuracy of the 3D model.

II. Complete Experimental Workflow: Aerosol Printing + FIB-SEM

Ink preparation: Liquid-phase/electrochemical exfoliation of graphene, WS₂, silver nanosheets, and silver nanowires, followed by size‑selective centrifugation to isolate nanomaterials with distinct lateral dimensions.

Aerosol Jet Printing: Nitrogen‑atomization deposition, followed by drying on a 80°C substrate, was used to fabricate single‑layer films and multilayer devices comprising graphene–silver nanoparticle heterostructures.

Sample pre-conductivity treatment: A thin conductive layer is deposited on the surface to eliminate charging effects in SEM.

FIB-SEM nanotomography: 30 kV gallium ion beam milling is performed layer by layer, with 2 kV SEM used to acquire cross-sectional images; for each sample set, hundreds of slices are acquired.

Image AI Segmentation and 3D Reconstruction: FIJI+Dragonfly software performs two-phase segmentation via machine learning, reconstructing the three-dimensional network of thin films.

Quantitative calculation: TauFactor and morphological analysis plugins extract parameters such as porosity, orientation, and specific surface area, while a four‑probe electrical measurement system is used in parallel to establish structure–property relationships.

III. Five Core Advantages of Aerosol Printing Coupled with FIB‑SEM Characterization

Nanoscale three-dimensional full-field visualization fills the gaps in conventional characterization.

BET measures only the macroscopic specific surface area; X-ray CT has a resolution of less than 50 nm; and conventional SEM can only image the surface layer. In contrast, FIB‑SEM can resolve pores and lamellae down to a few nanometers, providing a complete visualization of the film’s internal stacking and the true three-dimensional connectivity of its pore channels.

Automatic quantitative output of all-dimensional microscopic parameters

It simultaneously measures more than ten key parameters, including porosity, tortuosity, lamellar orientation, and aggregation degree, eliminating the need for repeated testing on multiple instruments and significantly shortening the R&D cycle for aerosol inks.

Compatible with single-layer and multi-layer heterogeneous aerosol devices

Resolvable aerosol-based stepwise printing enables the visualization of interfacial penetration and interlayer defects in multilayer composite films, allowing direct observation of the depth to which silver nanoparticles infiltrate into the graphene network and guiding the optimization of multi-material aerosol co‑printing processes.

Compatible with all mainstream aerosol nano-inks.

Graphene, transition-metal dichalcogenides, and silver-based one-dimensional/ two-dimensional materials are universally applicable; a unified standard characterization protocol has been established, making it an indispensable tool for standardizing laboratory aerosol‑based fabrication processes.

AI-assisted optimization enhances imaging accuracy and reduces testing errors.

Neural network interpolation fills in missing data in tomographic images, addressing structural losses caused by slice spacing, thereby enhancing the fidelity of 3D model reconstruction and improving the reliability of process‑comparison data.

IV. Five Core Technological Innovations of This Study

Innovation 1: A Standardized FIB‑SEM Three‑Dimensional Characterization Paradigm for Aerosol‑Printed Nanofilms

A comprehensive workflow comprising tomographic scanning, image segmentation, and quantitative analysis has been established for AJP‑based additive nanonetworks, thereby establishing a reusable characterization standard.

Innovation 2: A Theoretical Framework for Multi-Parameter Coupled Structure–Property Relationships

Quantitatively elucidates how nanoplatelet size and printing–stacking density jointly regulate pore structure and electrical conductivity, providing direct guidance for optimizing aerosol ink particle size selection and print‑layer count.

Innovative Three-Dimensional Analysis Method for Heterogeneous Multilayer Aerosol Device Interfaces

This enables the visualization of internal permeation behavior in stepwise aerosol‑printed composite membranes, addressing a longstanding industry challenge: the difficulty of characterizing interfacial defects in multilayer electronic devices.

Innovation: A Unified Three-Dimensional Quantitative Analysis Framework for One- and Two-Dimensional Nanonetworks

It simultaneously supports both silver nanowires and two-dimensional nanosheets—two mainstream aerosol materials—enabling cross‑system comparisons of thin films.

Innovative 5‑D DAIN neural network enhances the vertical resolution of tomographic imaging.

AI‑based interpolation completes FIB slices, cost‑effectively enhancing the accuracy of 3D reconstruction without increasing the number of milling slices, thereby reducing test duration.

V. Current Industry Challenges and Corresponding Solutions

Challenge 1: Conventional characterization methods can only provide surface‑level or macroscopic average data, failing to reveal the true internal pore structure and layer stacking.

✅FIB lamella polishing combined with 3D reconstruction provides a complete visualization of the nanoscale three-dimensional network within micron-scale thin films, leaving no microscopic detail overlooked.

Challenge 2: The interfacial penetration and internal defects of multilayer-printed composite films made from aerosols are difficult to observe.

✅ 3D reconstruction enables layered delamination of different printing functional layers, allowing for intuitive visualization of filler penetration depth and interlayer void‑short‑circuit defects.

Challenge 3: The lack of quantitative methods limits the ability to describe film density qualitatively, making it impossible to precisely optimize ink formulations and printing parameters.

✅Automated extraction of quantitative metrics such as porosity and tortuosity, providing data-driven guidance for iterative aerosol process optimization.

Challenge 4: X-ray CT suffers from insufficient resolution, making it impossible to distinguish tiny nanopores and thin-layered nanosheets.

✅FIB voxel sizes as low as 5 nm enable precise characterization of sub‑10‑nm microstructures, making it well suited for ultra‑thin aerosol‑printed films.

Challenge 5: Large slice spacing in FIB, leading to distortion of the vertical structure in the 3D model and high data errors.

✅ Introduces the DAIN neural network to generate intermediate frames, doubling the vertical resolution and enhancing the reliability of quantitative analysis.

VI. Future Diversified Industrial Application Scenarios

1 Aerosol Flexible Electronics R&D Quality Control Platform

Batch characterization of graphene and MXene conductive traces enables rapid assessment of how the number of aerosol‑printed layers and the ink‑film thickness affect electrical conductivity, thereby shortening device iteration cycles.

2 Process Optimization of Micro-Scale Energy Storage Electrodes

Quantitative analysis of the three-dimensional porosity in aerosol‑printed microsupercapacitors and battery electrodes, guiding the development of high‑capacity porous electrode printing strategies.

3 MEMS Micro-Sensing Functional Thin-Film Testing

Characterization of the internal pore connectivity in gas-sensitive, electrochemical biosensing aerosol thin films to enhance sensor response speed and sensitivity.

4 Development of Multi-Material Heterogeneous Integrated Aerosol Devices

Layer-by-layer printing of composite optoelectronic devices and microstructural analysis of catalytic array interfaces address issues of layer‑by‑layer printing and uncontrolled penetration.

5 High-Throughput Screening of Two-Dimensional Nanoprinting Inks

A rapid three-dimensional structural comparison of aerosol inks with varying delamination processes and particle sizes is conducted to screen for the optimal printing‑material formulation.

Image and Text Guide

Figure 1. FIB‑SEM nanotomography (FIB‑SEM‑NT) imaging workflow: (a) Photograph of an LPE graphene dispersion and a representative TEM image; (b) SEM images of the surface and cross-section of a printed multilayer graphene network (nanosheet length lNS = 695 nm); (c) Principle of slice-and-scan: the FIB mills approximately 800 consecutive slices, and the resulting SEM images are stacked to reconstruct a 15‑nm‑thick section; (d) Image segmentation process (WEKA random forest classification: manual annotation → classification → binarization; orange = nanosheets, black = pores); (e) 3D reconstruction of the printed LPE graphene network; (f) Magnified view of (e), showing nanosheets and pores.

• The voxel size is 5 nm × 5 nm × 15 nm (with a volume of approximately 375 nm³), which is 10–1000 times smaller than that of conventional X-ray CT, while simultaneously covering representative sample volumes ranging from 10² to 10⁴ μm³.
• Typical reconstructed volume: 20 μm × 15 μm × 2 μm, containing approximately 1.6×10⁹ voxels; a 2 kV accelerating voltage combined with an SE2 detector mitigates charging and contrast‑enhancement effects.
• Mercury intrusion/BET requires thick samples, AFM/SEM examines only the surface, X-ray CT suffers from insufficient resolution, and 3D TEM offers too small a volume; FIB‑SEM‑NT precisely fills these gaps.

Figure 2. Quantitative analysis of the reconstructed LPE graphene network (lNS = 238 nm): (a) 3D reconstruction of the network, partitioned into pore and nanoplate volumes; (b) Pore‑by‑slice porosity measurements (P = 41 ± 1%; inset: histogram); (c) Nanoplate–pore tortuosity factors κ in the out-of‑plane (y) and in‑plane (x, z) directions (inset: diffusion flux within pore volume, with darker regions indicating constrictions); (d) Heat map showing the distribution of pore circularity C as a function of pore cross‑sectional area A; (e) Aggregation factors for nanoplate length and thickness (χl ≈ 1.5, χt ≈ 5.6); (f) Distribution of the angle φ between the nanoplate normal and the out-of‑plane direction (Cauchy–Lorentz fit: φC = −0.6°; Hermans factor S = 0.61 ± 0.07).

• κ = τ² (where τ is the tortuosity); κ = 1 indicates an unobstructed path; for nanosheets, κ_OOP/κ_IP > 1 signifies predominant in-plane orientation.
• >99% of the pore volume is contained within a single through-pore, with a connectivity length scale of approximately 251 nm; the large pores exhibit an elongated morphology, consistent with the BET data.
• The full width at half maximum (FWHM) is (29±1)°, comparable to the 21° value for inkjet-printed graphene; S = 0.61 indicates partial in-plane orientation (S = 1 corresponds to complete in-plane orientation).

• As the lNS decreases from 1087 nm to 298 nm, the porosity declines from 51% to 39%, and the characteristic pore size decreases accordingly.
• The small nanosheet network is more compact, with smaller and rounder pores, which facilitates charge transport; the large nanosheet network is more open and porous, promoting electrolyte wetting and mass transport.
• By selecting the nanoplate size, the network porosity can be tailored to suit the target application; FIB‑SEM‑NT enables pore‑size measurements ranging from a few nanometers to the micrometer scale.

• Correlation between in-plane resistivity and morphological parameters: ρ_IP ≈ (ρ_NS + R_J·lNS/kNS) / ((1-P)/κ_IP).
• The out-of-plane κ is significantly larger than the in-plane κ, reflecting the predominant in-plane orientation of the nanosheets; the orientation factor increases with increasing lNS.
• The Bruggeman index (1.9–5.6) exceeds the theoretical value of 1.5, consistent with the experimental expectations for highly anisotropic particles (β = 2–5).

• AgNWs have a diameter of approximately 55 nm, which is smaller than the pixel size of most X-ray CT systems, highlighting the resolution advantage of FIB-SEM-NT.
• The AgNW network exhibits a porosity as high as 82–91%, with (κ_OOP/κ_IP) ≈ 12, significantly exceeding that of 2D nanosheet networks (e.g., graphene, which is approximately 2.9).
• As the length increases, the porosity of each material rises while the pore tortuosity decreases—consistent with the trend observed in graphene—but the magnitude of these changes varies among materials (with AgNS exhibiting only minor variations).

• It covers three categories of nanomaterials—0D, 1D, and 2D—with a characteristic size range spanning from a few nanometers to several tens of micrometers.
• Corresponding reconstruction network locations for each material: AgNPs → Fig. 6b, AgNWs → Fig. 5c,f, LPE graphene → Figs. 1–4, EE graphene → Fig. 6a, WS₂ → Figs. 5a,d, AgNSs → Figs. 5b,e.

• The EE network exhibits approximately sixfold higher conductivity than the LPE, primarily due to its thinner nanosheets (3.7 nm vs. 20 nm) rather than morphological factors.
• In the large nanosheet network, AgNPs penetrate to approximately 1.3 μm (with a connectivity path of about 725 nm), whereas in the small nanosheet network, penetration is almost negligible.
• Surface roughness R_RMS: EE ≈ 122 nm, LPE ≈ 182 nm, directly affecting the interface quality and electrode continuity of vertically stacked devices.

• Increase the slice thickness from 15 nm to 30 nm, then replace the removed frames with intermediate frames generated by DAIN, thereby restoring the original 15‑nm slice resolution.
• The generated image closely matches the ground truth (as evidenced by the large white regions in the difference map), demonstrating that the neural network can further enhance the 3D resolution of FIB‑SEM‑NT.
• The color scale (a, b) represents normalized pixel intensity and offers a new approach to generating isotropic cubic voxels.