Search for: All records

Substrate crystal facet-driven self-assembly of porphyrin derivatives on metal substrates provides a versatile platform for understanding structure-property relationships at molecule-surface interfaces. In this work, we investigate the adsorption and supramolecular organization of a porphyrin derivative on Ag(100) and Ag(110) single-crystal surfaces using ultrahigh vacuum (UHV) scanning tunneling microscopy, and tip-enhanced Raman spectroscopy (TERS). On Ag(100), the molecules form nearly isotropic, close-packed two-dimensional assemblies characterized by a quasi-square unit cell, consistent with weak surface corrugation and intermolecular-interaction-dominated ordering. In contrast, adsorption on the anisotropic Ag(110) surface leads to pronounced directional self-assembly, where molecules align along the close-packed atomic rows to form rectangular unit cells and quasi-one-dimensional domains. Quantitative analysis reveals that this transition from isotropic to anisotropic packing originates from enhanced molecule-substrate interactions and direction-dependent diffusion barriers on Ag(110), which impose strong surface-templating effects. Furthermore, UHV-TERS measurements reveal distinct variations in spectral response between the two surfaces, reflecting differences in local adsorption environments. These experimental observations are further supported by phonon calculations, which capture the specific modifications of vibrational modes and provide insight into the origin of the observed spectral variations. These findings highlight the fundamental role of surface symmetry and corrugation in directing molecular organization and provide insight into the interplay between structural anisotropy and spectroscopic response at metal-organic interfaces. 

In this work, we design and fabricate tunable pixelated infrared emitter arrays based on graphene transistors. We use finite element analysis to simulate a 5-by-3 pixelated emitter array, where localized heating area can be confined within a size of 25 μm and reaches more than 10 K temperature rise within 3 μs. By applying fast sweeping scheme with electrostatic voltage inputs, tunable surface patterns assembling dynamic alphabetical letters of microscale sizes are demonstrated and captured via thermal mapping with its thermal emission variation caused by localized temperature rise. Our results show that thermal emission can be highly localized and reconfigurable both spatially and temporally, which paves the way for developing and boosting the performance of microscale and nanoscale thermally driven devices for communication, characterization, sensing and actuation. 

This work demonstrates out-of-plane quadrupolar characteristic of excitons in two-dimensional Ruddlesden–Popper perovskites, which is attributed to the hydrogen bonding between organic spacer cations and apical I ions of the inorganic octahedron. 

Resonant interactions between electromagnetic whistler-mode waves and energetic electrons play a key role in controlling electron flux dynamics in Earth's radiation belts and driving electron precipitation into the upper atmosphere. Although this process is well investigated and modeled under the assumption of a dipole magnetic field, the dipole approximation often breaks down in plasma injection regions, where strong currents of hot ions significantly deform the magnetic field configuration. In these regions, spacecraft often detect intense whistler-mode waves, whereas ground-based observations suggest enhanced electron precipitation. In this study, we combine quasi-linear theory, extended to account for the non-dipole magnetic field configuration, and spacecraft observations of whistler-mode waves to quantify wave–particle interactions. We demonstrate that electron scattering by these waves is largely affected by non-dipole magnetic fields. We also provide a simple empirical fit for realistic electron scattering rates, which can be readily incorporated into existing global models of electron dynamics. 

Mixture of Experts (MoE) have shown remarkable success in leveraging specialized expert networks for complex machine learning tasks. However, their susceptibility to adversarial attacks presents a critical challenge for deployment in robust applications. This paper addresses the critical question of how to incorporate robustness into MoEs while maintaining high natural accuracy. We begin by analyzing the vulnerability of MoE components, finding that expert networks are notably more susceptible to adversarial attacks than the router. Based on this insight, we propose a targeted robust training technique that integrates a novel loss function to enhance the adversarial robustness of MoE, requiring only the robustification of one additional expert without compromising training or inference efficiency. Building on this, we introduce a dual-model strategy that linearly combines a standard MoE model with our robustified MoE model using a smoothing parameter. This approach allows for flexible control over the robustness-accuracy trade-off. We further provide theoretical foundations by deriving certified robustness bounds for both the single MoE and the dual-model. To push the boundaries of robustness and accuracy, we propose a novel joint training strategy JTDMoE for the dual-model. This joint training enhances both robustness and accuracy beyond what is achievable with separate models. Experimental results on CIFAR-10 and TinyImageNet datasets using ResNet18 and Vision Transformer (ViT) architectures demonstrate the effectiveness of our proposed methods.