Search

How well can deep learning models trained on human-generated sounds distinguish between another species' vocalization types? We analyze the encoding of bat song syllables in several self-supervised audio encoders, and find that models pre-trained on human speech generate the most distinctive representations of different syllable types. These findings form first steps towards the application of cross-species transfer learning in bat bioacoustics, as well as an improved understanding of out-of-distribution signal processing in audio encoder models., Comment: Presented at VIHAR-2024; see https://vihar-2024.vihar.org/

Hexagonal boron nitride encapsulation is the method of choice for protecting graphene from environmental doping and impurity scattering. It was previously demonstrated that metal-organic vapor phase epitaxy (MOVPE) grows epitaxially ordered, uniform BN layers on epigraphene (graphene grown on SiC). Due to graphene non-wetting properties, h-BN growth starts preferentially from the graphene ledges. We use this fact here to selectively promote growth of high-quality flat h-BN on epigraphene by patterning epigraphene microstructures prior to BN growth. Thin h-BN films (down to 6 nm) grown by MOVPE show smooth and pleated surface morphology on epigraphene, while crumpled BN is observed on the SiC. Cross-sectional high-resolution transmission electron microscopy images and fluorescence imaging confirm the higher BN quality grown on the epigraphene. Transport measurements reveal p-doping as expected from hydrogen intercalation of epigraphene and regions of high and low mobility. This method can be used to produce structurally uniform high-quality h-BN/epigraphene micro/nano scale heterostructure.

Representations from deep neural networks (DNNs) have proven remarkably predictive of neural activity involved in both visual and linguistic processing. Despite these successes, most studies to date concern unimodal DNNs, encoding either visual or textual input but not both. Yet, there is growing evidence that human meaning representations integrate linguistic and sensory-motor information. Here we investigate whether the integration of multimodal information operated by current vision-and-language DNN models (VLMs) leads to representations that are more aligned with human brain activity than those obtained by language-only and vision-only DNNs. We focus on fMRI responses recorded while participants read concept words in the context of either a full sentence or an accompanying picture. Our results reveal that VLM representations correlate more strongly than language- and vision-only DNNs with activations in brain areas functionally related to language processing. A comparison between different types of visuo-linguistic architectures shows that recent generative VLMs tend to be less brain-aligned than previous architectures with lower performance on downstream applications. Moreover, through an additional analysis comparing brain vs. behavioural alignment across multiple VLMs, we show that -- with one remarkable exception -- representations that strongly align with behavioural judgments do not correlate highly with brain responses. This indicates that brain similarity does not go hand in hand with behavioural similarity, and vice versa.

What do deep neural speech models know about phonology? Existing work has examined the encoding of individual linguistic units such as phonemes in these models. Here we investigate interactions between units. Inspired by classic experiments on human speech perception, we study how Wav2Vec2 resolves phonotactic constraints. We synthesize sounds on an acoustic continuum between /l/ and /r/ and embed them in controlled contexts where only /l/, only /r/, or neither occur in English. Like humans, Wav2Vec2 models show a bias towards the phonotactically admissable category in processing such ambiguous sounds. Using simple measures to analyze model internals on the level of individual stimuli, we find that this bias emerges in early layers of the model's Transformer module. This effect is amplified by ASR finetuning but also present in fully self-supervised models. Our approach demonstrates how controlled stimulus designs can help localize specific linguistic knowledge in neural speech models., Comment: Accepted to Interspeech 2024. For code and materials, see https://github.com/mdhk/phonotactic-sensitivity

Human listeners effortlessly compensate for phonological changes during speech perception, often unconsciously inferring the intended sounds. For example, listeners infer the underlying /n/ when hearing an utterance such as "clea[m] pan", where [m] arises from place assimilation to the following labial [p]. This article explores how the neural speech recognition model Wav2Vec2 perceives assimilated sounds, and identifies the linguistic knowledge that is implemented by the model to compensate for assimilation during Automatic Speech Recognition (ASR). Using psycholinguistic stimuli, we systematically analyze how various linguistic context cues influence compensation patterns in the model's output. Complementing these behavioral experiments, our probing experiments indicate that the model shifts its interpretation of assimilated sounds from their acoustic form to their underlying form in its final layers. Finally, our causal intervention experiments suggest that the model relies on minimal phonological context cues to accomplish this shift. These findings represent a step towards better understanding the similarities and differences in phonological processing between neural ASR models and humans., Comment: Accepted for publication in Computational Linguistics (Special Issue on Language Learning, Representation, and Processing in Humans and Machines)

We present the submission of the ILLC at the University of Amsterdam to the BabyLM challenge (Warstadt et al., 2023), in the strict-small track. Our final model, ChapGTP, is a masked language model that was trained for 200 epochs, aided by a novel data augmentation technique called Automatic Task Formation. We discuss in detail the performance of this model on the three evaluation suites: BLiMP, (Super)GLUE, and MSGS. Furthermore, we present a wide range of methods that were ultimately not included in the model, but may serve as inspiration for training LMs in low-resource settings., Comment: Part of the BabyLM challenge at CoNLL

Graphene nanoelectronics potential was limited by the lack of an intrinsic bandgap[1] and attempts to tailor a bandgap either by quantum confinement or by chemical functionalization failed to produce a semiconductor with a large enough band gap and a sufficient mobility. It is well known that by evaporating silicon from commercial electronics grade silicon carbide crystals an epitaxial graphene layer forms on the surfaces [2]. The first epigraphene layer to form on the silicon terminated face, known as the buffer layer, is insulating. It is chemically bonded to the SiC and spectroscopic measurements [3] have identified semiconducting signatures on the microscopic domains. However, the bonding to the SiC is disordered and the mobilities are small. Here we demonstrate a quasi-equilibrium annealing method that produces macroscopic atomically flat terraces covered with a well ordered epigraphene buffer layer that has a 0.6 eV bandgap. Room temperature mobilities exceed 5000 cm2/Vs which is much larger than silicon and 20 times larger than the phonon scattering imposed limit of current 2D semiconductors. Critical for nanotechnology, its lattice is aligned with the SiC substrate, it is chemically, mechanically, and thermally robust, and it can be conventionally patterned and seamlessly connected to semimetallic epigraphene making semiconducting epigraphene ideally suited for nanoelectronics., Comment: Text and figures + supplemental information

Tur\'{a}n number $T(n,5,3)$ is the minimum size of a system of triples out of a base set $X$ of $n$ elements such that every quintuple in $X$ contains a triple from the system. The exact values of $T(n,5,3)$ are known for $n \leq 17$. Tur\'{a}n conjectured that $T(2m,5,3) = 2\binom{m}{3}$, and no counterexamples have been found so far. If this conjecture is true, then $T(2m+1,5,3) \geq \lceil m(m-2)(2m+1)/6\rceil$. We prove the matching upper bound for all $n = 2m+1 > 17$ except $n=27$.

Realizing high-performance nanoelectronics requires control of materials at the nanoscale. Methods to produce high quality epitaxial graphene (EG) nanostructures on silicon carbide are known. The next step is to grow Van der Waals semiconductors on top of EG nanostructures. Hexagonal boron nitride (h-BN) is a wide bandgap semiconductor with a honeycomb lattice structure that matches that of graphene, making it ideally suited for graphene-based nanoelectronics. Here, we describe the preparation and characterization of multilayer h-BN grown epitaxially on EG using a migration-enhanced metalorganic vapor phase epitaxy process. As a result of the lateral epitaxial deposition (LED) mechanism, the grown h-BN/EG heterostructures have highly ordered epitaxial interfaces, as desired in order to preserve the transport properties of pristine graphene. Atomic scale structural and energetic details of the observed row-by-row, growth mechanism of the 2D epitaxial h-BN film are analyzed through first-principles simulations, demonstrating one-dimensional nucleation-free-energy-barrierless growth. This industrially relevant LED process can be applied to a wide variety of van der Waals materials.

Large-scale atomically thin metals can be stabilized through confinement epitaxy at graphene / SiC interface, which exhibit a gradient bonding type and are air stable, providing a compelling platform for quantum and optoelectronic technologies.

The recent observation of non-classical electron transport regimes in two-dimensional materials has called for new high-resolution non-invasive techniques to locally probe electronic properties. We introduce a novel hybrid scanning probe technique to map the local resistance and electrochemical potential with nm- and $\mu$V resolution, and we apply it to study epigraphene nanoribbons grown on the sidewalls of SiC substrate steps. Remarkably, the potential drop is non uniform along the ribbons, and $\mu$m-long segments show no potential variation with distance. The potential maps are in excellent agreement with measurements of the local resistance. This reveals ballistic transport in ambient condition, compatible with micrometer-long room-temperature electronic mean free paths.

Graphene's original promise to succeed silicon faltered due to pervasive edge disorder in lithographically patterned deposited graphene and the lack of a new electronics paradigm. Here we demonstrate that the annealed edges in conventionally patterned graphene epitaxially grown on a silicon carbide substrate (epigraphene) are stabilized by the substrate and support a protected edge state. The edge state has a mean free path that is greater than 50 microns, 5000 times greater that the bulk states and involves a theoretically unexpected Majorana-like zero-energy non-degenerate quasiparticle that does not produce a Hall voltage. In seamless integrated structures, the edge state forms a zero-energy one-dimensional ballistic network with essentially dissipationless nodes at ribbon-ribbon junctions. Seamless device structures offer a variety of switching possibilities including quantum coherent devices at low temperatures. This makes epigraphene a technologically viable graphene nanoelectronics platform that has the potential to succeed silicon nanoelectronics., Comment: File contains article and supplementary. Nature Communications (in print)

Circular polarization resolved magneto-infrared studies of multilayer epitaxial graphene (MEG) are performed using tunable quantum cascade lasers in high magnetic fields up to 17.5 T. Landau level (LL) transitions in the monolayer and bilayer graphene inclusions of MEG are resolved, and considerable electron-hole asymmetry is observed in the extracted electronic band structure. For monolayer graphene, a four-fold splitting of the $n=0$ to $n=1$ LL transition is evidenced and attributed to the lifting of the valley and spin degeneracy of the zeroth LL and the broken electron-hole symmetry. The magnetic field dependence of the splitting further reveals its possible mechanisms. The best fit to experimental data yields effective $g$-factors, $g^*_{VS}=6.7$ and $g^*_{ZS}=4.8$, for the valley and Zeeman splitting, respectively., Comment: One reference added in this arxiv version

The study of the nanomechanics of graphene $-$ and other 2D materials $-$ has led to the discovery of exciting new properties in 2D crystals, such as their remarkable in-plane stiffness and out of plane flexibility, as well as their unique frictional and wear properties at the nanoscale. Recently, nanomechanics of graphene has generated renovated interest for new findings on the pressure-induced chemical transformation of a few-layer thick epitaxial graphene into a new ultra-hard carbon phase, named diamene. In this work, by means of a machine learning technique, we provide a fast and efficient tool for identification of graphene domains (areas with a defined number of layers) in epitaxial and exfoliated films, by combining data from Atomic Force Microscopy (AFM) topography and friction force microscopy (FFM). Through the analysis of the number of graphene layers and detailed \r{A}-indentation experiments, we demonstrate that the formation of ultra-stiff diamene is exclusively found in 1-layer plus buffer layer epitaxial graphene on silicon carbide (SiC) and that an ultra-stiff phase is not observed in neither thicker epitaxial graphene (2-layer or more) nor exfoliated graphene films of any thickness on silicon oxide (SiO$_{2}$)., Comment: 30 pages, 7 figures

The Coulomb scattering dynamics in graphene in energetic proximity to the Dirac point is investigated by polarization resolved pump-probe spectroscopy and microscopic theory. Collinear Coulomb scattering rapidly thermalizes the carrier distribution in k-directions pointing radially away from the Dirac point. Our study reveals, however, that in almost intrinsic graphene full thermalization in all directions relying on noncollinear scattering is much slower. For low photon energies, carrier-optical-phonon processes are strongly suppressed and Coulomb mediated noncollinear scattering is remarkably slow, namely on a ps timescale. This effect is very promising for infrared and THz devices based on hot carrier effects.

Recent pump-probe experiments performed on graphene in a perpendicular magnetic field have revealed carrier relaxation times ranging from picoseconds to nanoseconds depending on the quality of the sample. To explain this surprising behavior, we propose a novel symmetry-breaking defect-assisted relaxation channel. This enables scattering of electrons with single out-of-plane phonons, which drastically accelerate the carrier scattering time in low-quality samples. The gained insights provide a strategy for tuning the carrier relaxation time in graphene and related materials by orders of magnitude.

Atomically thin graphene exhibits fascinating mechanical properties, although its hardness and transverse stiffness are inferior to those of diamond. To date, there hasn't been any practical demonstration of the transformation of multi-layer graphene into diamond-like ultra-hard structures. Here we show that at room temperature and after nano-indentation, two-layer graphene on SiC(0001) exhibits a transverse stiffness and hardness comparable to diamond, resisting to perforation with a diamond indenter, and showing a reversible drop in electrical conductivity upon indentation. Density functional theory calculations suggest that upon compression, the two-layer graphene film transforms into a diamond-like film, producing both elastic deformations and sp2-to-sp3 chemical changes. Experiments and calculations show that this reversible phase change is not observed for a single buffer layer on SiC or graphene films thicker than 3 to 5 layers. Indeed, calculations show that whereas in two-layer graphene layer-stacking configuration controls the conformation of the diamond-like film, in a multilayer film it hinders the phase transformation., Comment: Published online on Nature Nanotechnology on December 18, 2017

This article presents a review of epitaxial graphene on silicon carbide, from fabrication to properties, put in the context of other forms of graphene., Comment: 46 pages, 322 references, 35 figures. Submitted December 2015

Epitaxial graphene grown on SiC by the confinement controlled sublimation method is reviewed, with an emphasis on multilayer and monolayer epitaxial graphene on the carbon face of 4H-SiC and on directed and selectively grown structures under growth-arresting or growth-enhancing masks. Recent developments in the growth of templated graphene nanostructures are also presented, as exemplified by tens of micron long very well confined and isolated 20-40nm wide graphene ribbons. Scheme for large scale integration of ribbon arrays with Si wafer is also presented., Comment: 19 pages, 9 figures

Here we follow, both experimentally and theoretically, the development of magnetism in Tb clusters from the atomic limit, adding one atom at a time. The exchange interaction is, surprisingly, observed to drastically increase compared to that of bulk, and to exhibit irregular oscillations as a function of the interatomic distance. From electronic structure theory we find that the theoretical magnetic moments oscillate with cluster size in exact agreement with experimental data. Unlike the bulk, the oscillation is not caused by the RKKY mechanism. Instead, the inter-atomic exchange is shown to be driven by a competition between wave-function overlap of the 5d shell and the on-site exchange interaction, which leads to a competition between ferromagnetic double-exchange and antiferromagnetic super-exchange. This understanding opens up new ways to tune the magnetic properties of rare-earth based magnets with nano-sized building blocks.