Scientific user facilities generate X-ray scattering data faster than traditional workflows can process them. We address this challenge across two settings, offline dataset exploration and live on-the-fly analysis. We train a domain-specific attention-based Convolutional Variational Autoencoder (C-VAE) on 1.5 million X-ray scattering images to learn low-dimensional representations capturing structural variation across diverse experimental conditions. The learned latent space reveals well-organized clusters and smooth trajectories reflecting experimental progression. It further supports controlled synthetic scattering image generation across diverse structural states. When deployed without retraining, the model organizes time-resolved film formation experiments at two synchrotron facilities into interpretable latent structures. Benchmarking against DINOv3 (ViT-7B), a general-purpose vision foundation model, demonstrates that domain-specific training yields more interpretable latent organization for scattering data. Both workflows are integrated within Latent Space Explorer, a component of the MLExchange platform, supporting interactive structural exploration across archived datasets and live experiments.

This research presents the development of a versatile metrology tool for characterizing 3D nanostructures in semiconductor architectures using Resonant Soft X-ray Scattering (RSoXS). Our approach has been applied to various application areas, including Complementary Field-Effect Transistors (CFETs), demonstrating its adaptability for a range of semiconductor devices. Our models are developed using a cross-section of Transmission Electron Microscopy (TEM) images featuring intricate shapes and morphologies as the baseline. We then leverage Resonant Soft X-ray Scattering (RSoXS) as a non-destructive metrology tool to analyze the optical properties of complex 3D structures. This approach leverages the relationships between nanostructure morphology and diffractive scattering patterns, allowing for efficient and accurate characterization of complex semiconductor architectures. Understanding these architectures not only enables discovery but also paves the way for next-generation nanoFETs. The structural, spectroscopical, and interfacial properties of these 3D devices reveals how smaller transistors can increase computational efficiency, further pushing the boundaries of computational capabilities. By revolutionizing the characterization and optimization of complex semiconductor architectures, our methodology has the potential to drive significant advancements in the field, enabling the creation of more efficient, sustainable, and powerful electronic devices.

This study presents a machine learning approach to derive the film formation of biopolymer-templated titania nanostructures during spray deposition, in combination with in situ grazing-incidence small-angle X-ray scattering (GISAXS). A neural network trained on synthetic GISAXS data directly predicts domain-size distributions from experimental two-dimensional scattering patterns, capturing the full kinetics of nanostructure evolution with high temporal resolution. The predictions reveal hierarchical size distributions and periodic growth features, consistent with layer-by-layer spray deposition and validated by complementary scanning electron microscopy (SEM) imaging. Quantitative comparison with conventional parametric GISAXS fits shows good qualitative agreement, with systematic differences explained by domain-shape assumptions and resolved by applying a geometric scaling factor. Simulated SEM-like surfaces derived from neural network outputs reproduce the porous, foam-like nanoscale morphology observed experimentally, reinforcing the method's credibility. This integrated approach enables real-time, nondestructive, statistically averaged monitoring of bulk nanostructure development in functional coatings, offering a scalable methodology to accelerate the characterization and process control of sustainably manufactured nanostructured titania films for energy-related applications such as photocatalysis and photovoltaics.

Unveiling interfaces at sub-nanometer scales is essential for advancing the understanding of complex chemical transformations. However, characterizing solid-liquid interfaces with high dimensional sensitivity and temporal resolution remains challenging, due to their dynamic nature and inaccessibility by conventional probes. Here we present an approach, Pattern-enhanced Resonant Soft X-ray Scattering, to overcome the challenges. Rooted in a "sample-as-optics" philosophy, this technique utilizes precisely engineered line-grating nanopatterns to modulate near-field X-ray illumination, coherently enhancing scattering signals from the line-gratings. We implement the method using Ni line-grating nanopatterns in electrochemical water oxidation. The periodic nanostructures serve as diffractive optical elements to reveal the Ni oxidation gradients and structural dynamics at the electrode-electrolyte interfaces. Finite-element simulations corroborate the observed trends by modeling variations in compositions and structures during electrocatalysis. Through integrating advanced sample design with coherent wave nature of soft X-rays, our approach opens accessible pathways to operando exploring chemical evolution and sub-nanometer dimensional variations simultaneously in electrochemical systems. This non-destructive method is efficient and element-specific, making it valuable for probing chemical and dimensional dynamics with appropriate modeling.

Ptychography neural networks suffer from scaling inconsistencies when generalizing out of distribution, limiting their real world viability. We address this scaling mismatch using a factorization strategy which decouples the learned object texture from measurement scaling, enabling a single trained network to produce measurement-consistent reconstructions across varying illumination conditions. This requires predicting the learned object in real and imaginary units instead of the canonical amplitude and phase representation. We additionally introduce a synthetic object sampling strategy that minimizes phase distribution mismatch between synthetic training data and experimental targets. These improvements yield up to a 5x reduction in Fourier error over the previous PtychoPINN-torch baseline across 5 experimental datasets spanning multiple beamlines and facilities.

Modern synchrotron X-ray facilities generate X-ray diffraction (XRD) image data at rates that far exceed the capacity of manual analysis, yet critical processing steps such as artifact removal rely on human intervention. Rule-based algorithms in standard beamline software cannot reliably distinguish harmful artifacts like single-crystal diffraction spots from desirable features like preferred orientation, which leads to incorrectly processed data. We demonstrate that a U-Net convolutional neural network trained on expert-labeled experimental data can help automate artifact segmentation in time-resolved XRD measurements of battery materials. To address overfitting from redundant time-series frames, we introduce a mutual information-based pruning algorithm that selects maximally diverse training images. Our optimized model achieves an 85.1% true positive rate on a fully held-out battery dataset while reducing false positives by 34% compared with GSAS-II . Critically, the model preserves preferred orientation features that GSAS-II incorrectly removes, preventing systematic underestimation of phase intensities in downstream analysis.

Synchrotron beamlines generate high-resolution volumetric datasets that reveal detailed sample microstructure, but converting raw scans into scientific insight remains computationally demanding. We present an automated AI-driven high-performance computing (HPC) pipeline for near real-time segmentation and analysis of micro-CT scans collected at Beamline 8.3.2 at the Advanced Light Source (ALS). After acquisition, data are automatically transferred to HPC systems at the National Energy Research Scientific Computing Center (NERSC) and the Argonne Leadership Computing Facility (ALCF) for distributed tomographic reconstruction, followed by large-scale segmentation using two fine-tuned state-of-the-art foundation models, Segment Anything Model 3 (SAM3) and a Self-Distillation with No Labels (DINO)-based segmentation model, DINOv3-Seg. Their results are aggregated to produce final segmentation masks. Orchestrated with Prefect, it achieves a scan-to-discovery turnaround time of approximately 20 minutes. The pipeline generalizes across diverse micro-CT experiments; here, we applied it to examine xylem vessels in grapevine petiole datasets, revealing structural changes for in situ plant physiology studies.

X-ray ptychography is a data-intensive imaging technique expected to become ubiquitous at next-generation light sources delivering many-fold increases in coherent flux. The need for real-time feedback under accelerated acquisition rates motivates surrogate reconstruction models like deep neural networks, which offer orders-of-magnitude speedup over conventional methods. However, existing deep learning approaches lack robustness across diverse experimental conditions. We propose an unsupervised training workflow emphasizing probe conditioning by combining experimentally-measured probes with synthetic, procedurally generated objects. This probe-centric approach enables a single physics-informed neural network to reconstruct unseen experiments across multiple beamlines—among the first demonstrations of multi-probe generalization. We find probe conditioning is equally important as in-distribution training; models trained using this synthetic workflow achieve reconstruction fidelity comparable to those trained exclusively on experimental data, even when changing the type of synthetic training object. The proposed approach enables training of experiment-steering models that provide real-time feedback under dynamic experimental conditions.

Iterative ptychographic reconstruction algorithms are widely used for coherent diffractive imaging but can exhibit slow convergence under realistic experimental conditions. We propose a machine learning-augmented approach that accelerates iterative ptychographic reconstruction by introducing a learned fast-forward operator applied during reconstruction. Following an initial warm-up using standard iterations, the fast-forward operator advances the reconstruction toward a more converged state, after which conventional iterative updates are resumed. This strategy preserves the physical consistency and flexibility of established ptychographic solvers while reducing the number of iterations required for convergence. The model is trained on diverse ptychographic datasets and evaluated on experimental data acquired in a different year, demonstrating robustness and temporal generalization. Compared with conventional iterative solvers, the machine learning-augmented method achieves comparable reconstruction quality while converging faster in terms of Poisson negative log-likelihood, yielding over a two-fold reduction in wall-clock time. The approach has been integrated into an existing reconstruction pipeline and deployed in production at a synchrotron beamline, demonstrating practicality for real-time experimental operation.

Synchrotron light sources support a wide array of techniques to investigate materials, often producing complex, high-volume data that challenge traditional workflows. At the Advanced Light Source (ALS), we developed infrastructure to move microtomography data over ESnet to ALCF and NERSC, where CPU- and GPU-based algorithms generate 3D reconstructed volumes of experimental samples. We employ two data movement and reconstruction models: real-time processing as data streams directly to NERSC compute nodes, and automated file transfer to NERSC and ALCF file systems. The streaming pipeline provides users with feedback in under ten seconds, while the file-based workflow produces high-quality reconstructions suitable for deeper analysis in 20-30 minutes. This infrastructure enables users to utilize HPC resources without direct access to backend systems. We plan to extend this architecture to more endstations, supporting our beamline scientists and users.

The thermoelectric properties of fully sprayed thin films of poly(3-hexylthiophen-2,5-diyl) (P3HT) doped with chloroauric acid are investigated for different film thicknesses. The film thickness increases logarithmically with increasing amount of deposited material on the surfaces. Both the electrical conductivity and measured Seebeck coefficients of the doped thin films show an optimal polymer layer thickness between 275 and 310 nm and yield a maximum power factor of (1.77 ± 0.22) μW/m·K^2 . The optimum layer thickness results from the optimal amount of dopant molecules per monomer between 1.1 and 1.3 at these ratios of P3HT and HAuCl _4 for the thin film fabrication.

DLSIA (Deep Learning for Scientific Image Analysis) is a Python-based machine learning library that empowers scientists and researchers across diverse scientific domains with a range of customizable convolutional neural network (CNN) architectures for a wide variety of tasks in image analysis to be used in downstream data processing. DLSIA features easy-to-use architectures, such as autoencoders, tunable U-Nets and parameter-lean mixed-scale dense networks (MSDNets). Additionally, this article introduces sparse mixed-scale networks (SMSNets), generated using random graphs, sparse connections and dilated convolutions connecting different length scales. For verification, several DLSIA-instantiated networks and training scripts are employed in multiple applications, including inpainting for X-ray scattering data using U-Nets and MSDNets, segmenting 3D fibers in X-ray tomographic reconstructions of concrete using an ensemble of SMSNets, and leveraging autoencoder latent spaces for data compression and clustering. As experimental data continue to grow in scale and complexity, DLSIA provides accessible CNN construction and abstracts CNN complexities, allowing scientists to tailor their machine learning approaches, accelerate discoveries, foster interdisciplinary collaboration and advance research in scientific image analysis.

Metrology plays a crucial role in semiconductor manufacturing by providing accurate and precise measurement and characterization of critical parameters. With the development of high-resolution extreme ultraviolet lithography (EUVL) processes, critical dimensions are shrinking to sub-10 nm. Resist materials encounter the challenge of providing heightened sensitivity and a handle on exacerbating stochastic variations. A comprehensive understanding of the chemical profile of the latent image is pivotal for mitigating stochastic effects and optimizing pattern quality. However, the subtle differences in chemistry between the exposed and unexposed regions of the resists make it extremely challenging to characterize the latent images with sub-nanometer precision. Here, we develop the metrology with critical-dimension resonant soft X-ray scattering (CD-RSoXS) to probe the chemical profiles of latent images stored in resist after exposure. The combination of absorption spectroscopy and enhanced scattering contrast makes it possible to characterize the subtle structural and chemical variations in the latent image. Moreover, the results of the measurements are compared with the simulations with a finite element method-based Maxwell solver to extract a detailed profile of the latent and developed images. We demonstrate that the CD-RSoXS technique can provide valuable insights into the high spatial resolution and local chemical sensitivity simultaneously, which is crucial to understanding the resolution limits and stochastic effects in EUVL processes.

The twist-bend nematic liquid crystal phase is a three-dimensional fluid in which achiral bent molecules spontaneously form an orientationally ordered, macroscopically chiral, heliconical winding of a ten nanometer-scale pitch in the absence of positional ordering. Here, the structure of the twist-bend phase of the bent dimer CB7CB and its mixtures with 5CB is characterized, revealing a hidden invariance of the self-assembly of the twist-bend structure of CB7CB, such that over a wide range of concentrations and temperatures, the helix pitch and cone angle change as if the ground state for a pitch of the TB helix is an inextensible heliconical ribbon along the contour formed by following the local molecular long axis (the director). Remarkably, the distance along the length for a single turn of this helix is given by 2πRmol, where Rmol is the radius of bend curvature of a single all-trans CB7CB molecule. This relationship emerges from frustrated steric packing due to the bent molecular shape: space in the fluid that is hard to fill attracts the most flexible molecular subcomponents, a theme of nanosegregation that generates self-assembled, oligomer-like correlations of interlocking bent molecules in the form of a brickwork-like tiling of pairs of molecular strands into duplex double-helical chains. At higher temperatures in the twist-bend phase, the cone angle is small, the director contour is nearly along the helix axis z, and the duplex chains are sequences of biaxial elements formed by overlapping half-molecule pairs, with an approximately 45° rotation of the biaxis between each such element along the chain.