Research

We developed a compression workflow for large 4D-STEM datasets that preserves the quantities needed for subsequent analysis. It combines MGARD error bounds with a correction that preserves selected moments of the diffraction patterns, including those used for center-of-mass measurements. The work also introduces a frequency-based reliability check, connecting storage savings to the accuracy of the resulting microscopy measurements.

Sources: Abstract / article

We use electron microscopy to investigate how strain, chemistry, and electronic states vary across high entropy oxide multilayers. Combining 4D-STEM, imaging, EELS, and EDS reveals changes in cation valence and diffraction reflections absent from an ideal rock salt structure. These measurements help connect growth conditions and chemical substitutions to local structural distortions in functional oxide heterostructures.

Sources: Conference contribution

We compare Mn-containing and Cr-containing high entropy oxide films to understand why their magnetic behavior differs. Microscopy reveals mixed rock salt and spinel-like regions in the Mn-containing film, while spectroscopy and calculations connect these regions to cation site preferences and mixed valence. The results show how local chemistry and structure can change magnetic frustration and exchange bias even within closely related oxide compositions.

Sources: Open manuscript

We investigate how growth temperature changes nanoscale strain and electronic structure in high entropy oxide films with the same nominal composition. 4D-STEM strain mapping, EELS, and EDS connect local strain variations with cobalt valence and subtle chemical differences. This preprint identifies synthesis conditions as a way to tune strain and defect states without changing the nominal composition.

Sources: Abstract / article

We investigate how the number of atomic layers controls the anomalous Hall response of MnBi₂Te₄ films. Careful molecular beam epitaxy and structural characterization separate the effects of layer parity from defects and unintended doping. Odd-layer and even-layer films show markedly different magnetic responses, helping establish the conditions needed to study intrinsic topological magnetism.

Sources: OSTI abstract

We investigate how substrate steps suppress antiphase twins during Bi₂Se₃ thin-film growth. Experiments on sapphire with a high miscut angle, supported by first-principles calculations, connect preferred nucleation at steps to the energetic cost of twin formation. The work identifies a route to more uniform films while showing how that selectivity changes as the growing film covers the steps.

Sources: OSTI abstract

We introduce CryoBlob for finding features in noisy cryo-EM images without requiring labeled training data or templates. The preprint describes a JAX implementation that combines image filtering and feature detection with GPU execution, microscopy file support, and batching for large datasets. The current software documentation identifies compact particles as the validated use case and distinguishes its differentiable image response from the discrete particle-selection step.

Sources: Abstract / article · Current software scope

I am a co-inventor on this granted patent for learning relationships between chemical imaging and conventional tissue images. The method registers image features with spectral information during training, then uses those relationships to predict chemical information and cancer regions from new tissue images. It provides a concrete example of transferring information between measurement modalities through machine learning, with the grant issued as US 12,675,875 B2 in July 2026.

Sources: Granted patent record

We recast electron ptychography as a differentiable inverse problem in the Ptyrodactyl codebase. A JAX forward model predicts diffraction measurements, while automatic differentiation and complex-valued optimizers update the specimen and microscope parameters to fit the data. This makes effects such as position errors, partial coherence, and aberrations additional model parameters within a common reconstruction workflow.

Sources: OSTI manuscript

We describe a modular approach to optical ptychography in which lenses, propagation, apertures, and other optical elements form a differentiable model of an experiment. Automatic differentiation supplies the gradients needed to recover the sample and optical parameters from measured diffraction intensities. The conference contribution presents this approach in the early Ptyrodactyl codebase, allowing optical arrangements to be changed without deriving a new reconstruction algorithm for each setup.

Sources: OSTI manuscript

We use in-situ 4D-STEM to follow electron-beam-induced crystallization of amorphous Bi–Se films. Time-resolved diffraction records which crystalline domains form, their orientations, and their growth, while RHEED measurements and simulated diffraction connect those observations to synthesis conditions. The work explores how these experimental and simulated trajectories could inform decisions during subsequent growth experiments.

Sources: Conference contribution

In this commentary, we discuss evidence for how copper nanocatalysts restructure during electrochemical CO₂ conversion. The study discussed combines operando electron microscopy, X-ray spectroscopy, and first-principles modeling to connect structural changes with the reaction products themselves. It illustrates why understanding a catalyst requires observing its evolving structure under reaction conditions.

Sources: OSTI abstract

We combine complementary mass spectrometry imaging methods to recover spatial and spectral information that is difficult to obtain with a single instrument. The approach brings together FT-ICR, MALDI-ToF, and ToF-SIMS data using canonical correlation analysis and constraints informed by the measurement physics. The resulting fusion supports submicron molecular mapping while retaining detailed spectral information, and the published article is linked below alongside the preprint record.

Sources: Published article at OSTI

We developed a platform for moving microscope-control and image-reconstruction software from a virtual test environment into operation. The virtual infrastructure twin runs on the computing resources of the experimental system itself, reducing the need to rewrite code when moving to GPU-based production workflows. We demonstrate this transition with remote Nion microscope steering and image reconstruction on a connected multi-GPU system.

Sources: OSTI abstract

I built Janssen to compose optical microscopy simulations and phase-retrieval workflows in JAX. Optical elements act as functions on wavefronts, with PyTree data structures connecting propagation, lenses, coherence models, and reconstruction. This design lets the same physical model support forward simulation and gradient-based inverse problems, with compilation and GPU execution available through JAX.

Sources: Documentation

I built CryoBlob to detect compact particles in noisy cryo-EM and materials micrographs without templates or training labels. Its Laplacian-of-Gaussian detector supports GPU execution and memory-aware batches, with an optional watershed step for touching particles. The image response is differentiable with respect to the image and scale, while particle selection remains discrete and elongated-object detection is experimental.

Sources: Repository and scope

We investigate whether deep learning can reconstruct detailed Raman maps from measurements collected at lower spatial resolution. The models learn relationships between low- and high-resolution hyperspectral data, including settings where only limited training data are available. This offers a way to reduce the acquisition burden of Raman imaging while studying the practical limits of reconstruction from sparse measurements.

Sources: OSTI abstract

We follow the transition from amorphous to crystalline Bi₂Se₃ under an electron beam. By recording diffraction patterns at short time intervals from films grown on silicon nitride, we observe crystalline diffraction features emerge and strengthen with continued irradiation. The experiment demonstrates a way to study beam-driven structural transformations as they happen.

Sources: OSTI manuscript

We describe prototype workflows that connect electron microscopy images with atomistic simulations and high-performance computing. Deep learning extracts atomic features, which are converted into simulation inputs to investigate defects and energetic stability in materials such as transition metal dichalcogenides. The presentation examines the latency, data transfer, and orchestration challenges involved in using those calculations to guide subsequent measurements.

Sources: OSTI manuscript

This conference contribution brings together our work on resolving the atomic structure of polar materials and their domain walls. It discusses polar entropy in LiNbO₃, small polar distortions in layered titanates, and domain walls in the polar metal Ca₃Ru₂O₇. Together, these examples show how advances in electron microscopy and quantitative analysis make subtle interfacial structures experimentally accessible.

Sources: Conference contribution

We describe the practical work of creating a research-ready radiology collection linked to Veterans Affairs electronic health records. The challenges include differences in imaging modalities, annotations, patient identifiers, and the computational demands of processing large image collections. The preprint documents transfer procedures and bottlenecks that must be addressed before reliable automation and downstream medical-imaging research are possible.

Sources: Abstract / article

I built Ptyrodactyl to express electron-scattering experiments as composable, differentiable JAX models. Its physical models keep microscope and specimen parameters explicit so inverse methods can optimize them against measured data. The current package focuses on electron scattering, while the original software release and early conference work also document its development toward flexible ptychographic reconstruction.

Sources: Repository

We examine what is needed to make machine learning useful during scanning transmission electron microscopy experiments. The paper connects real-time analysis and experimental decisions with practical questions about distribution shifts, computing infrastructure, data handling, and the role of physical models. It provides a framework for building automated workflows in which measurements and analysis inform the next experimental step.

Sources: OSTI abstract

We discuss how focused interest groups help the microscopy community develop new instrumentation and analysis capabilities. The article uses the history of aberration-corrected electron microscopy to show the value of exchanging ideas, advocating for technical development, and training researchers. It places emerging data-driven approaches in that broader context of community-led progress.

Sources: OSTI abstract

We developed a framework for remotely steering scientific instruments and collecting their measurements across connected computing systems. Separate network paths for data and control are combined with Python wrappers and Pyro client–server interfaces that expose instrument operations remotely. A Nion microscopy demonstration shows automated transfers and steering, providing infrastructure for experiments that coordinate instruments with external computation.

Sources: OSTI abstract

We combine whole-slide tissue images with mass spectrometry imaging to learn relationships between tissue appearance and chemical composition. Deep image features are spatially registered with the chemical measurements, allowing trained models to predict cancer regions from conventional stained images. The study demonstrates how paired measurements during training can make chemical information useful when only optical images are available at prediction time.

Sources: OSTI abstract

We propose two ways to acquire images progressively on interlaced hexagonal sampling grids. The patterns provide an early view of the full field and then increase resolution as additional samples are collected. This supports live visualization and drift correction while making efficient use of sampling points in slowly scanned or dose-sensitive experiments.

Sources: Abstract / article

We developed an image-gradient decomposition method that distributes large ptychographic reconstructions across GPUs. Each device works on a spatial region and exchanges the overlapping gradient information needed to keep neighboring reconstructions consistent. The approach reduces memory requirements and scales to large computing systems, making reconstruction of extensive electron microscopy datasets more practical.

Sources: OSTI abstract

We outline a computing architecture that connects electron microscopes with nearby analysis resources and larger high-performance systems. Fast processing near the instrument can provide timely feedback, while more demanding reconstruction and modeling tasks use remote computing resources. The roadmap discusses demonstrations and the networking, storage, and software requirements for making these workflows routine in microscopy.

Sources: OSTI abstract

We demonstrate remote steering and automated data transfer across a network of electron microscopes and computing systems. The architecture separates control traffic from measurement data and uses a virtual infrastructure twin to develop and test the software without occupying the physical instruments. This provides a practical foundation for coordinating automated experiments across facilities.

Sources: OSTI abstract

We introduce virtual infrastructure twins for developing workflows that connect scientific instruments, networks, and computing resources. The twin emulates the operational environment and includes instrument simulators so orchestration software can be tested before access to the real system is available. Examples involving microscope steering and GPU access show how this approach supports design choices and reduces disruption during development.

Sources: OSTI abstract

We directly observe double-bilayer polar nanoregions inside a nonpolar matrix in lightly strontium-doped Ca₃Mn₂O₇. Atomic-resolution microscopy, EELS, and first-principles calculations connect their stability to manganese occupying calcium sites and changing oxidation state. In-situ heating shows that these regions can persist to high temperatures, revealing how defects can stabilize local polar structures.

Sources: OSTI abstract

We compare lossless compression strategies for the large diffraction datasets generated by 4D-STEM. A method based on pixel-specific intensity distributions exploits differences between the diffraction center and background, outperforming the generic codecs tested on the measured dataset. The study demonstrates substantial storage and transfer savings while preserving the recorded data exactly.

Sources: OSTI manuscript

We describe an integrated workflow that connects microscope acquisition, shared storage, and remote GPU analysis. Python remote procedure calls expose Nion Swift controls to a Jupyter session, while distinct control, data, and management paths coordinate the work. This allows an operator to run acquisition and processing across networked systems with fewer manual file transfers and handoffs.

Sources: OSTI manuscript

We address the memory bottleneck in large electron ptychography reconstructions by splitting the object and probe positions across GPUs. Overlapping regions, or halos, carry the gradient information needed to reconcile updates at tile boundaries. The conference study demonstrates reduced memory use and strong parallel scaling on simulated datasets while preserving consistency with an undecomposed reconstruction.

Sources: OSTI manuscript

We explore whether chemical information from mass spectrometry can help predict prostate cancer regions in conventional tissue images. Features from a pretrained image network are registered with mass-spectrometry components and used in regression and classification models. The conference results demonstrate cross-modal prediction while identifying the need to validate additional regions highlighted by the models.

Sources: OSTI manuscript

We combine electron microscopy and first-principles calculations to investigate polar and nonpolar phase coexistence in Ca₂.₉Sr₀.₁Mn₂O₇. The measurements reveal polar nanoscale regions inside a nonpolar matrix, while calculations connect their stabilization to Mn²⁺ substituting for calcium. The presentation identifies local defect chemistry as a route to understanding and potentially controlling that phase coexistence.

Sources: APS abstract

We developed DistGANs for distributed training of conditional generative adversarial networks on labeled image datasets. The package partitions training data by class and trains multiple generators concurrently, with each generator focused on one label. This organization makes the class structure of the data part of the parallelization strategy.

Sources: Abstract / article

This conference contribution focuses on mapping how surface strain evolves in PtCo core–shell catalysts using 4D-STEM. In our related methods work, diffraction-disk positions provide measurements of projected lattice expansion and contraction within individual particles. That measurement approach connects nanoscale structure with the surface strain that influences catalyst behavior.

The full conference abstract was not accessible; this overview uses the conference record and the related methods paper linked here.

Sources: Conference record · Related methods manuscript

We show how to measure projected unit-cell size variations in PtCo nanoparticles that are not aligned to a low-index zone axis. The workflow locates noncollinear diffraction disks with subpixel precision and uses their relative positions to map local expansion or contraction. Implemented in stemtool, it supports analysis of multiple differently oriented particles within one dataset and reduces the need to tilt each particle individually.

Sources: OSTI manuscript

We describe a prototype architecture for using microscopy data, machine learning, and simulations in a feedback loop. Neural networks turn diffraction patterns into structural descriptors, while high-performance calculations and Gaussian-process models estimate properties and uncertainty. The proposed workflow uses that uncertainty to choose subsequent scan positions, with the aim of bringing physically informed decisions closer to acquisition time.

Sources: Publisher first-page preview

We present a practical workflow for extracting precise atomic positions from STEM images. It covers acquisition, denoising, drift correction, atom indexing, and single- or multiple-peak fitting before calculating strain and structural distortions. A graphical analysis tool accompanies the methodology, making quantitative measurements of defects and interfaces accessible without writing a complete analysis package.

Sources: OSTI abstract

We combine automated microscopy acquisition with Python analysis to improve sampling of alloy nanoparticle catalysts. The workflow measures particle sizes across fuel-cell electrodes and uses spectroscopy with radial averaging to examine changes in platinum-shell thickness after stress testing. These methods make it easier to compare catalyst degradation across many particles and locations rather than relying on a few representative images.

Sources: Publisher first-page preview

We show that hydrogenation can transform nonmagnetic PdCoO₂ films into a palladium–cobalt alloy with strong out-of-plane ferromagnetism. The treatment also makes the sign of the anomalous Hall effect tunable through processing conditions. This connects a chemical transformation of a conductive oxide with a route to magnetic materials and possible spintronic heterostructures.

Sources: OSTI abstract

We present the emergence of ferromagnetism when PdCoO₂ films are treated with hydrogen. The resulting material has a strong out-of-plane magnetic response and an anomalous Hall signal whose sign depends on processing. The conference presentation connects controlled chemical modification with changes in magnetic and electronic transport behavior.

Sources: Abstract / article

We investigate how magnetic interfaces affect quantum transport in Bi₂Se₃ thin films. Weak antilocalization is strongly suppressed at interfaces with paramagnetic Co₇Se₈, while cobalt incorporated in a nonmagnetic charge state has a much smaller effect. The comparison points to local magnetic moments at the interface as an important source of spin dephasing.

Sources: OSTI abstract

We explore whether substantial chemical disorder can be introduced into a layered Ruddlesden–Popper oxide. Epitaxial synthesis and complementary structural and spectroscopic measurements show that several rare-earth species can share the same sublattice in a layered cuprate. The results also reveal the role of substrate strain in selecting the resulting phase, linking compositional complexity with growth constraints.

Sources: OSTI abstract

We use electron microscopy to examine how oxygen annealing changes molecular-beam-epitaxy-grown PdCoO₂ films. Annealing improves crystallinity but also changes the film–substrate interface, where an interfacial cobalt oxide layer disappears and voids are observed. The images reveal remaining mirror twins, helping explain why improved overall crystallinity does not eliminate every structural limitation on transport.

Sources: Publisher first-page preview

We examine when the phase-object approximation fails in interpreting 4D-STEM center-of-mass measurements. Simulations of an asymmetric ZnS structure show that increasing specimen thickness can change the contrast qualitatively, including a reversal between atomic columns. The contribution highlights the need to account for electron scattering beyond a thin-specimen approximation when interpreting fields, interfaces, and polar structures.

Sources: Publisher first-page preview

I introduced stemtool as an open source Python toolkit for quantitative electron microscopy analysis. The conference contribution describes workflows for atomic-resolution images and 4D-STEM datasets, including precise atomic-position measurements and diffraction-based analysis. The aim is to turn complex microscopy measurements into reusable analysis procedures for strain, distortions, and related material properties.

Sources: OSTI manuscript

We study how defects in a titania support change the activity of ligand-stabilized single platinum atoms. Spectroscopy establishes the isolated metal environment, while hydrosilylation measurements show higher activity on more defective supports. The results connect local coordination and metal–support interactions with catalytic performance and stability.

Sources: OSTI abstract

We compare atomic-resolution imaging and 4D-STEM for measuring strain in rhodium-core, platinum-shell nanocubes. Diffraction-data processing provides subpicometer precision, and multivariate analysis separates core and shell contributions to follow lattice changes across their interface. The work establishes a quantitative route to evaluating strain engineering in individual catalyst nanoparticles.

Sources: OSTI abstract

We developed mpfit to locate atomic columns when neighboring image peaks overlap. The method fits multiple two-dimensional Gaussian components and refines the relevant subset, reducing the bias that can arise from fitting each peak independently. Tests with simulated and experimental images show why accounting for overlap matters for precise measurements of atomic displacements.

Sources: OSTI abstract

We establish a molecular beam epitaxy route for growing the highly conductive layered oxide PdCoO₂. Low-temperature growth with reactive oxygen, followed by annealing, addresses the tendency of palladium to reduce and segregate during synthesis. Thickness-dependent transport and structural measurements identify surface scattering and twins as important limits on film conductivity.

Sources: OSTI abstract

We use precise atomic-displacement measurements to quantify polar configurational entropy in LiNbO₃. STEM measurements of niobium and oxygen positions reveal variations in the local polar direction, which become more pronounced near domain walls. Combined with first-principles and effective-Hamiltonian calculations, the analysis connects local symmetry breaking with the statistical distribution of atomic configurations.

Sources: Author manuscript

We investigate how polar and nonpolar structures in Ca₃Mn₂O₇ evolve during heating inside the electron microscope. Atomic-resolution imaging of a lightly strontium-doped crystal follows changes in local domain morphology and the polar regions trapped within the material. The experiment connects temperature-driven structural changes with the phase coexistence that complicates polarization switching in this layered oxide.

Sources: Publisher first-page preview

We compare strain measurements from atomic-resolution images and scanning nanodiffraction of the same Pd@Pt nanoparticles. The analysis evaluates atomic-column fitting and geometric phase analysis alongside diffraction-disk localization with edge detection and cross-correlation. This comparison helps assess measurement sensitivity and map lattice changes across the shell of an electrocatalyst particle.

Sources: Author manuscript

We use high-resolution electron microscopy to examine densely stacked domains and coexisting polar and nonpolar structures in Ca₃Mn₂O₇. Imaging along selected crystallographic directions resolves domain boundaries and the in-plane components of polar displacements. The study investigates the local structural complexity underlying the unusual domain behavior of this layered perovskite.

Sources: Publisher first-page preview

We use nanodiffraction-based differential phase contrast to investigate potential variations across domain walls in titanium-doped Ca₃Ru₂O₇. A larger probe keeps diffraction disks separated, allowing small momentum shifts to be measured alongside atomic-resolution images of the domain structure. The observed momentum reversal at a wall demonstrates how 4D-STEM can probe local electrostatic variations at interfaces.

Sources: Author manuscript

My PhD thesis develops quantitative STEM methods for understanding ferroelectric domain walls. It combines precise atomic-position measurements with theory to investigate polar distortions and entropy in LiNbO₃, then extends the analysis to diffraction patterns recorded at every probe position. Together, imaging and differential phase contrast provide evidence for charged domain walls in polar metals and connect interfacial structure with electrostatic behavior.

Sources: Thesis abstract

We develop a statistical approach to measuring small polar displacements in complex oxides despite scan distortion and image noise. Relative measurements between neighboring atomic species and averaging over unit cells improve sensitivity to displacements below ten picometers. Examples in brownmillerite films and LiNbO₃ connect these measurements with theoretical predictions and the structure of ferroelectric domain walls.

Sources: Publisher first-page preview

We use aberration-corrected STEM to study domain walls and junctions in titanium-doped Ca₃Ru₂O₇. Atomic imaging resolves both 90° and 180° walls, while EELS compares electronic hybridization in the metallic and insulating states. The observed head-to-head, head-to-tail, and tail-to-tail arrangements motivate questions about how local polar distortions coexist with metallic screening.

Sources: APS abstract

We demonstrate a materials-design approach that combines ordered oxygen vacancies with ordered cations to produce a polar oxide. Alternating strontium- and calcium-containing brownmillerite layers creates a superlattice whose structure is confirmed by diffraction and atomic-resolution microscopy. Symmetry analysis and first-principles calculations show how chemical ordering removes inversion symmetry even though the constituent compounds are centrosymmetric.

Sources: Article abstract

We image ferroelectric domain walls in LiNbO₃ with simultaneous annular bright-field and dark-field STEM. Combining the two signals locates both oxygen and niobium columns, allowing their relative displacements across the wall to be measured. The work connects atomic distortions and wall width with theoretical descriptions of polarization at these interfaces.

Sources: Author manuscript

We investigate the atomic structure of alternating calcium- and strontium-containing brownmillerite layers. Aberration-corrected STEM resolves the chemistry and atomic displacements associated with oxygen-vacancy ordering and the superlattice interfaces. The presentation examines how these structural ingredients can produce polar behavior in deliberately assembled oxide layers.

Sources: Abstract / article

We demonstrate an electrically triggered RF switch using epitaxial VO₂ on sapphire. The device uses the insulator-to-metal transition to combine a short turn-on delay with low insertion loss and strong isolation at microwave frequencies. The reported 26.5 THz figure is the switching cutoff figure of merit, connecting oxide-film quality with electronic device performance.

Sources: Institutional abstract

We develop a combinatorial growth approach for finding the conditions needed to produce high-quality VO₂ films across a wafer. An oxygen-activity gradient creates a continuous set of growth environments, allowing the desired vanadium valence state to be identified and controlled. The resulting single-crystalline films show a large resistivity change at the metal–insulator transition, demonstrating a route toward scalable oxide electronics.

Sources: Abstract / article

We grow freestanding graphene–transition-metal-dichalcogenide heterostructures that can be examined directly by electron microscopy. Imaging and diffraction reveal different orientation relationships for WSe₂ and MoS₂ on graphene and show how defects influence nucleation and growth. The approach makes the atomic structure of these interfaces accessible without the handling and transfer steps normally required for supported heterostructures.

Sources: Article abstract

My master's thesis investigates semiconductor fiber structures for transmitting and controlling mid-infrared light. I used modified chemical vapor deposition to deposit silicon and germanium inside glass tubes, then studied preform formation and fiber drawing with microscopy, diffraction, and modeling. The work connects deposition conditions and processing with the fabrication of refractive-index structures in semiconductor preforms.

Sources: Thesis abstract

We investigate how iron ions affect the wet-chemical formation of platinum nanoparticle films on glass. Comparing ions supplied in situ with direct addition reveals competing pathways between particle precipitation and film formation at room temperature. Microscopy, spectroscopy, and adhesion measurements characterize the resulting films and establish a route for depositing metal nanoparticles directly on glass substrates.

Sources: Article abstract