3-Channel ADF-EDS-EELS Tomography

• The paper introduces a simultaneous acquisition scheme that integrates ADF, EDS, and EELS signals in a single low-dose STEM tilt series to achieve co-registered 3D reconstructions of structure, composition, and valence.
• It employs advanced reconstruction techniques such as GENFIRE and fused joint inversion to correlate elemental volumes and valence maps, offering nanoscale resolution with significant dose reduction.
• Applications on battery cathodes (e.g., NCM111) and nanoparticle systems reveal depth-dependent chemical segregation and enhanced 3D spectroscopic insights for materials research.

Simultaneous ADF-EDS-EELS tomography is a correlative scanning transmission electron microscopy methodology in which an annular dark-field image, an energy-dispersive X-ray spectrum image, and an electron energy-loss spectrum image are acquired from the same tilt series and reconstructed into matched three-dimensional descriptions of structure, composition, and electronic state. In the strict sense, the three signals are recorded within a single low-dose STEM tilt-series acquisition, so that alignment, masking, and depth-resolved analysis are performed in a common coordinate frame; this was demonstrated for LiNi$_{1/3}$Co$_{1/3}$Mn$_{1/3}$O$_2$ cathode particles, where ADF provided the structural scaffold, EDS provided O/Mn/Co/Ni composition, and EELS provided Mn and Co valence-state maps (Oh et al., 27 Sep 2025). The broader field also includes fused bimodal tomography, foundational EELS-tomography theory, and integrated multimodal acquisition frameworks that do not yet realize all three channels in a single joint experiment (Schwartz et al., 2023, Hörl et al., 2013, Uusimaeki et al., 28 Apr 2025).

1. Scope of the term

The literature uses closely related but not identical acquisition and reconstruction regimes. In much of the precursor work, the elastic imaging channel is specifically HAADF rather than generic ADF, although the HAADF formulation is explicitly described as mapping onto the ADF/HAADF class (Schwartz et al., 2022). The distinction matters because some papers are simultaneously acquired but only bimodal, some are multimodal but sequential at each tilt, and some are EELS-tomography papers without ADF or EDS at all.

Regime	Operational meaning	Representative paper
Direct simultaneous three-channel tomography	One low-dose STEM tilt series records ADF, EDS, and EELS at each tilt	(Oh et al., 27 Sep 2025)
Fused bimodal tomography	Dense HAADF tilt series guides sparse EELS or EDX chemical tomography	(Schwartz et al., 2023)
Integrated multimodal tomography workflow	Coordinated STEM/4D-STEM/EELS/EDS acquisition, but sequential at each tilt	(Uusimaeki et al., 28 Apr 2025)

This distinction resolves a common misconception. “Multimodal tomography” is broader than “simultaneous ADF-EDS-EELS tomography,” and “simultaneous” in the strictest sense refers to one tilt-series acquisition with intrinsically co-registered channels rather than separate experiments later registered numerically (Oh et al., 27 Sep 2025).

2. Acquisition architectures and what is actually simultaneous

The clearest direct realization acquires, at each tilt angle, an ADF projection image, an EDS spectrum image, and an EELS spectrum image during the same STEM scan. In the NCM111 implementation, the microscope was a 300 kV Titan double Cs-corrected TEM, the tilt range was $-70^\circ$ to $+70^\circ$ in $10^\circ$ steps, the scan grid was $64 \times 64$ with 4.63 nm step size, the dwell time was 0.05 s per pixel, EDS was collected with Super-X, and EELS was collected with a GIF Quantum 965 in dual-EELS mode. The reported total dose was approximately $6.56 \times 10^5\ \mathrm{e^- \AA^{-2}}$, and zero-degree controls at the beginning, middle, and end of the tilt series were used to verify that no significant structural change or noticeable change in zero-degree Mn $L_3$ and Co $_{1/3}$0 intensity maps occurred during acquisition (Oh et al., 27 Sep 2025).

Earlier fused multimodal electron tomography used a different architecture. Dense HAADF projections were collected at fine angular spacing, while EELS or EDX maps were acquired only every few tilts. For Au-Fe$_{1/3}$1O$_{1/3}$2 and Co$_{1/3}$3O$_{1/3}$4-Mn$_{1/3}$5O$_{1/3}$6, the chemical maps were collected every $_{1/3}$7, while HAADF was collected at $_{1/3}$8 spacing. The measurements were still simultaneously acquired at those tilts where spectroscopy was recorded, but the reconstruction was bimodal rather than three-channel and the chemical angular sampling was intentionally sparse (Schwartz et al., 2023).

Integrated acquisition software has pushed the instrumentation further without necessarily achieving full hardware-synchronous three-channel simultaneity. St4DeM, implemented in Digital Micrograph, supports automatic tilt-series acquisition combined with 4D-STEM and/or EELS/EDS spectrum imaging, and defines a 7-dimensional dataset as $_{1/3}$9. Its validated tomography workflow is HAADF tilt series plus 4D-STEM plus EELS SI, with EELS acquired automatically after the 4D-STEM acquisition at each tilt; this is coordinated multimodal tomography, but not experimentally validated simultaneous ADF+EDS+EELS per probe dwell (Uusimaeki et al., 28 Apr 2025).

At the spectroscopic level, coincidence EELS-EDX has shown that EELS and EDX events can be timestamped and correlated at the single-event level in a TEM/STEM-compatible instrument, producing background-suppressed EELS and rejecting spurious EDX fluorescence. This is directly relevant to simultaneous chemistry acquisition, but the published demonstration did not include tomography, scan-position encoding for 3D reconstruction, or ADF integration (Jannis et al., 2019).

3. Reconstruction paradigms: correlative versus fused

Two reconstruction paradigms dominate the field. The first is fused joint inversion, in which all elemental volumes are estimated from a single objective that couples elastic and spectroscopic channels. The second is correlative multimodal reconstruction, in which each modality is reconstructed by its own pipeline but all channels share the same tilt geometry, masks, and alignment.

The clearest joint formulation is the fused multimodal tomography objective

$_{1/3}$0

Here $_{1/3}$1 are the elemental volumes, $_{1/3}$2 are HAADF projections, $_{1/3}$3 are EELS or EDX chemical maps, $_{1/3}$4 and $_{1/3}$5 are projection operators, $_{1/3}$6 encodes HAADF Z-contrast, the chemical term is a Poisson negative log-likelihood, and the prior is channel-wise isotropic total variation. This formulation explicitly links morphology and chemistry through a shared 3D object (Schwartz et al., 2023).

That tomography model was preceded by a 2D fused multi-modal electron microscopy framework with the same three ingredients: HAADF consistency, a Poisson-counting term for low-count EELS or EDX elemental maps, and TV regularization. The tomography extension is therefore not an unrelated development but a projection-operator generalization of the same inverse-problem design (Schwartz et al., 2022).

By contrast, the direct simultaneous ADF-EDS-EELS battery workflow is correlative rather than fully fused. ADF, EDS, and EELS are acquired simultaneously and share a common coordinate frame, but the reconstructions remain modality-specific. ADF tilt series are reconstructed with GENFIRE and provide the mask, surface definition, and shell geometry; EDS elemental tilt series are corrected for tilt-dependent intensity variation and reconstructed independently; EELS data are background-subtracted, deconvolved, decomposed into valence-state components, and then reconstructed valence-by-valence (Oh et al., 27 Sep 2025). The methodological gain is intrinsic spatial registration rather than a single all-channel objective.

4. The EELS channel and its distinct tomographic roles

EELS contributes at least two different types of tomographic observable. In plasmonics, Hörl, Trügler, and Hohenester showed that, in the quasistatic limit and for a dominant plasmon eigenmode, optical-frequency STEM-EELS maps over a rotation series can be reformulated as a tomography problem. For non-penetrating trajectories and $_{1/3}$7, the mode-resolved loss probability reduces to

$_{1/3}$8

so the target of reconstruction is the three-dimensional plasmonic potential $_{1/3}$9, or equivalently the surface charge distribution $_2$0. That work is not a simultaneous ADF-EDS-EELS paper, but it provides the mathematically explicit forward and inverse model for the EELS channel in a tomographic setting (Hörl et al., 2013).

In chemistry and valence tomography, EELS is used differently. In the NCM111 study, low-dose dual-EELS was drift-corrected using the zero-loss peak, spatially denoised, background-subtracted with a Hartree-Slater-constrained iterative scheme, Fourier-ratio deconvolved, and then decomposed into fixed reference spectra for Mn$_2$1, Mn$_2$2, Mn$_2$3, Co$_2$4, and Co$_2$5. The data cube was expressed conceptually as $_2$6, with $_2$7 fixed to measured reference spectra and $_2$8 containing the spatial weights that define valence-resolved tilt series for GENFIRE reconstruction (Oh et al., 27 Sep 2025).

A different low-dose route avoids structural priors entirely. “Soft” core-loss EELS tomography for FeO/Fe$_2$9O$-70^\circ$0 nanocubes uses the Fe-$-70^\circ$1 edge at $-70^\circ$2 eV rather than the conventional Fe-$-70^\circ$3 edge at $-70^\circ$4 eV, reporting $-70^\circ$5 higher dose efficiency. Its multichannel deep image prior with TV regularization reconstructs all channels jointly through

$-70^\circ$6

without HAADF-STEM signal or symmetry constraints. This is still EELS-only tomography, but it is directly relevant where HAADF contrast is not sufficiently discriminative to act as the dominant prior (Pelaez-Fernandez et al., 9 Jun 2026).

5. Demonstrated materials systems and scientific outputs

Fused bimodal tomography established that high-resolution 3D chemistry could be recovered from sparse spectroscopic sampling when the HAADF backbone is dense. In Au-Fe$-70^\circ$7O$-70^\circ$8, Co$-70^\circ$9O$+70^\circ$0-Mn$+70^\circ$1O$+70^\circ$2, and ZnS-Cu$+70^\circ$3S$+70^\circ$4, the method achieved near or below one nanometer 3D chemical resolution, often with 99% less dose or one-hundred-fold lower dose than conventional chemistry-only tomography of similar quality. For the Au-Fe$+70^\circ$5O$+70^\circ$6 case, real-space half-pitch resolutions of 0.8 nm along $+70^\circ$7, 0.8 nm along $+70^\circ$8, and 1.1 nm along $+70^\circ$9 were reported from only 9 chemical maps, and the simulated studies showed a 3–5 fold improvement in average NRMSE relative to conventional chemical tomography (Schwartz et al., 2023).

The direct simultaneous three-channel realization moved from bimodal chemistry to fully correlative structure-composition-valence mapping. In NCM111 cathodes, ADF tomography supplied the 3D particle morphology and shell partition, EDS tomography supplied voxelwise O/Mn/Co/Ni composition and inhomogeneity metrics, and EELS tomography supplied Mn and Co valence-state volumes. The central scientific result was that overall composition evolved relatively uniformly through the entire primary particle, whereas valence-state changes and transition-metal segregation were strongly depth-dependent and concentrated near the surface (Oh et al., 27 Sep 2025).

The battery dataset also provided explicit valence trends. From 100 to 200 cycles, Mn$10^\circ$0 changed from 80.7% to 53.4%, Mn$10^\circ$1 from 12.1% to 42.0%, Co$10^\circ$2 from 77.5% to 11.0%, and Co$10^\circ$3 from 22.5% to 89.0%. Surface-region segregation was about twice as high as the whole-particle average at 100 and 200 cycles. These results were interpreted as evidence that structural phase transitions, transition-metal migration, and dissolution-driven segregation coexist rather than reduce to a simple one-step phase-transition model (Oh et al., 27 Sep 2025).

EELS-only low-dose bonding tomography has extended the domain further. For FeO/Fe$10^\circ$4O$10^\circ$5 core-shell nanocubes, nine low-loss EELS projections over $10^\circ$6 to $10^\circ$7 combined with DIPm-TV yielded $10^\circ$8 nm isotropic resolution oxidation-state volumes, recovered a thin outer FeO shell, and revealed a small internal void. The significance here is methodological: useful 3D spectroscopic tomography can be obtained from sparse-view, low-dose data without HAADF priors when the spectroscopic channels themselves are jointly reconstructed (Pelaez-Fernandez et al., 9 Jun 2026).

6. Limitations, misconceptions, and future directions

The first limitation is definitional. Most precursor papers in this area are not simultaneous ADF-EDS-EELS tomography in the strict sense. The fused multimodal tomography literature validates HAADF+EELS or HAADF+EDX, not a single optimization over HAADF, EDS, and EELS together (Schwartz et al., 2023). The 2022 fused multi-modal microscopy framework is 2D rather than tomographic (Schwartz et al., 2022). St4DeM validates sequential but integrated acquisition at each tilt rather than per-dwell simultaneous three-channel readout (Uusimaeki et al., 28 Apr 2025). Foundational EELS-tomography work reconstructs plasmonic eigenfields rather than composition or valence (Hörl et al., 2013).

The second limitation is channel-specific physics. EDS tomography remains sensitive to tilt-dependent absorption, detector shadowing, and geometric anisotropy, so the simultaneous NCM111 workflow applied tilt-dependent normalization before reconstruction (Oh et al., 27 Sep 2025). Low-dose EELS tomography remains the most fragile channel: the battery study required iterative background subtraction constrained by pre-edge flatness and Hartree-Slater continuum behavior, Fourier-ratio deconvolution, low-pass filtering, and fixed-reference decomposition; Co$10^\circ$9 was not analyzed because no stable reference sample was available, and Ni valence was not determined because the Ni $64 \times 64$0 edge SNR was too low (Oh et al., 27 Sep 2025).

A further misconception is that HAADF or ADF must always be the dominant prior. The FeO/Fe$64 \times 64$1O$64 \times 64$2 soft-EELS study explicitly argues the opposite for systems where HAADF contrast is weakly discriminative; FeO and Fe$64 \times 64$3O$64 \times 64$4 have nearly identical HAADF signals because densities differ by only about 10%, so a spectroscopy-first reconstruction can be preferable (Pelaez-Fernandez et al., 9 Jun 2026). This suggests that future simultaneous ADF-EDS-EELS tomography will not be methodologically uniform: some specimens favor ADF-guided fusion, whereas others favor joint spectroscopic inversion with ADF retained mainly for alignment.

Two enabling directions are especially clear. One is structural-channel replacement or augmentation. MultiSlice Electron Tomography reconstructs a 3D electrostatic potential directly from 4D-STEM tilt series by a single integrated optimization over all diffraction patterns and tilts, addressing nonlinear contrast, light-element sensitivity, and dose efficiency in structural tomography (Lee et al., 2022). A plausible implication is that future chemistry tomography could be anchored not only by conventional ADF but also by multislice-aware structural backbones. The second is detector and acquisition geometry. Multislice hollow ptychography was introduced to make ptychographic structural imaging compatible with simultaneous EELS by leaving a central angular channel open for the spectrometer, with up to 70% of total electrons available for spectroscopy in simulation; however, that work did not yet demonstrate simultaneous EELS acquisition, simultaneous ADF, simultaneous EDS, or tilt-series tomography (Lei et al., 27 Jun 2025).

A final prospective direction is spectroscopic correlation at the event level. Coincidence EELS-EDX demonstrates that time-correlated EELS and EDX can yield background-suppressed EELS and suppress spurious fluorescence contributions, but it remains a spectroscopy advance rather than a tomography method (Jannis et al., 2019). This suggests that future simultaneous ADF-EDS-EELS tomography may evolve along two axes at once: more strongly coupled 3D inverse problems, and more discriminative per-tilt chemistry derived from temporally or physically correlated spectroscopic channels.